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/fs.h>
  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/module.h>
  47 #include <linux/reboot.h>
  48 #include <linux/file.h>
  49 #include <linux/compat.h>
  50 #include <linux/delay.h>
  51 #include <linux/raid/md_p.h>
  52 #include <linux/raid/md_u.h>
  53 #include <linux/slab.h>
  54 #include "md.h"
  55 #include "bitmap.h"
  56
  57 #ifndef MODULE
  58 static void autostart_arrays(int part);
  59 #endif
  60
  61 /* pers_list is a list of registered personalities protected
  62  * by pers_lock.
  63  * pers_lock does extra service to protect accesses to
  64  * mddev->thread when the mutex cannot be held.
  65  */
  66 static LIST_HEAD(pers_list);
  67 static DEFINE_SPINLOCK(pers_lock);
  68
  69 static void md_print_devices(void);
  70
  71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  72 static struct workqueue_struct *md_wq;
  73 static struct workqueue_struct *md_misc_wq;
  74
  75 static int remove_and_add_spares(struct mddev *mddev,
  76                                  struct md_rdev *this);
  77
  78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  79
  80 /*
  81  * Default number of read corrections we'll attempt on an rdev
  82  * before ejecting it from the array. We divide the read error
  83  * count by 2 for every hour elapsed between read errors.
  84  */
  85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  86 /*
  87  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  88  * is 1000 KB/sec, so the extra system load does not show up that much.
  89  * Increase it if you want to have more _guaranteed_ speed. Note that
  90  * the RAID driver will use the maximum available bandwidth if the IO
  91  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  92  * speed limit - in case reconstruction slows down your system despite
  93  * idle IO detection.
  94  *
  95  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  96  * or /sys/block/mdX/md/sync_speed_{min,max}
  97  */
  98
  99 static int sysctl_speed_limit_min = 1000;
 100 static int sysctl_speed_limit_max = 200000;
 101 static inline int speed_min(struct mddev *mddev)
 102 {
 103         return mddev->sync_speed_min ?
 104                 mddev->sync_speed_min : sysctl_speed_limit_min;
 105 }
 106
 107 static inline int speed_max(struct mddev *mddev)
 108 {
 109         return mddev->sync_speed_max ?
 110                 mddev->sync_speed_max : sysctl_speed_limit_max;
 111 }
 112
 113 static struct ctl_table_header *raid_table_header;
 114
 115 static ctl_table raid_table[] = {
 116         {
 117                 .procname       = "speed_limit_min",
 118                 .data           = &sysctl_speed_limit_min,
 119                 .maxlen         = sizeof(int),
 120                 .mode           = S_IRUGO|S_IWUSR,
 121                 .proc_handler   = proc_dointvec,
 122         },
 123         {
 124                 .procname       = "speed_limit_max",
 125                 .data           = &sysctl_speed_limit_max,
 126                 .maxlen         = sizeof(int),
 127                 .mode           = S_IRUGO|S_IWUSR,
 128                 .proc_handler   = proc_dointvec,
 129         },
 130         { }
 131 };
 132
 133 static ctl_table raid_dir_table[] = {
 134         {
 135                 .procname       = "raid",
 136                 .maxlen         = 0,
 137                 .mode           = S_IRUGO|S_IXUGO,
 138                 .child          = raid_table,
 139         },
 140         { }
 141 };
 142
 143 static ctl_table raid_root_table[] = {
 144         {
 145                 .procname       = "dev",
 146                 .maxlen         = 0,
 147                 .mode           = 0555,
 148                 .child          = raid_dir_table,
 149         },
 150         {  }
 151 };
 152
 153 static const struct block_device_operations md_fops;
 154
 155 static int start_readonly;
 156
 157 /* bio_clone_mddev
 158  * like bio_clone, but with a local bio set
 159  */
 160
 161 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 162                             struct mddev *mddev)
 163 {
 164         struct bio *b;
 165
 166         if (!mddev || !mddev->bio_set)
 167                 return bio_alloc(gfp_mask, nr_iovecs);
 168
 169         b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
 170         if (!b)
 171                 return NULL;
 172         return b;
 173 }
 174 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 175
 176 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
 177                             struct mddev *mddev)
 178 {
 179         if (!mddev || !mddev->bio_set)
 180                 return bio_clone(bio, gfp_mask);
 181
 182         return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
 183 }
 184 EXPORT_SYMBOL_GPL(bio_clone_mddev);
 185
 186 /*
 187  * We have a system wide 'event count' that is incremented
 188  * on any 'interesting' event, and readers of /proc/mdstat
 189  * can use 'poll' or 'select' to find out when the event
 190  * count increases.
 191  *
 192  * Events are:
 193  *  start array, stop array, error, add device, remove device,
 194  *  start build, activate spare
 195  */
 196 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 197 static atomic_t md_event_count;
 198 void md_new_event(struct mddev *mddev)
 199 {
 200         atomic_inc(&md_event_count);
 201         wake_up(&md_event_waiters);
 202 }
 203 EXPORT_SYMBOL_GPL(md_new_event);
 204
 205 /* Alternate version that can be called from interrupts
 206  * when calling sysfs_notify isn't needed.
 207  */
 208 static void md_new_event_inintr(struct mddev *mddev)
 209 {
 210         atomic_inc(&md_event_count);
 211         wake_up(&md_event_waiters);
 212 }
 213
 214 /*
 215  * Enables to iterate over all existing md arrays
 216  * all_mddevs_lock protects this list.
 217  */
 218 static LIST_HEAD(all_mddevs);
 219 static DEFINE_SPINLOCK(all_mddevs_lock);
 220
 221
 222 /*
 223  * iterates through all used mddevs in the system.
 224  * We take care to grab the all_mddevs_lock whenever navigating
 225  * the list, and to always hold a refcount when unlocked.
 226  * Any code which breaks out of this loop while own
 227  * a reference to the current mddev and must mddev_put it.
 228  */
 229 #define for_each_mddev(_mddev,_tmp)                                     \
 230                                                                         \
 231         for (({ spin_lock(&all_mddevs_lock);                            \
 232                 _tmp = all_mddevs.next;                                 \
 233                 _mddev = NULL;});                                       \
 234              ({ if (_tmp != &all_mddevs)                                \
 235                         mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
 236                 spin_unlock(&all_mddevs_lock);                          \
 237                 if (_mddev) mddev_put(_mddev);                          \
 238                 _mddev = list_entry(_tmp, struct mddev, all_mddevs);    \
 239                 _tmp != &all_mddevs;});                                 \
 240              ({ spin_lock(&all_mddevs_lock);                            \
 241                 _tmp = _tmp->next;})                                    \
 242                 )
 243
 244
 245 /* Rather than calling directly into the personality make_request function,
 246  * IO requests come here first so that we can check if the device is
 247  * being suspended pending a reconfiguration.
 248  * We hold a refcount over the call to ->make_request.  By the time that
 249  * call has finished, the bio has been linked into some internal structure
 250  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 251  */
 252 static void md_make_request(struct request_queue *q, struct bio *bio)
 253 {
 254         const int rw = bio_data_dir(bio);
 255         struct mddev *mddev = q->queuedata;
 256         int cpu;
 257         unsigned int sectors;
 258
 259         if (mddev == NULL || mddev->pers == NULL
 260             || !mddev->ready) {
 261                 bio_io_error(bio);
 262                 return;
 263         }
 264         if (mddev->ro == 1 && unlikely(rw == WRITE)) {
 265                 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
 266                 return;
 267         }
 268         smp_rmb(); /* Ensure implications of  'active' are visible */
 269         rcu_read_lock();
 270         if (mddev->suspended) {
 271                 DEFINE_WAIT(__wait);
 272                 for (;;) {
 273                         prepare_to_wait(&mddev->sb_wait, &__wait,
 274                                         TASK_UNINTERRUPTIBLE);
 275                         if (!mddev->suspended)
 276                                 break;
 277                         rcu_read_unlock();
 278                         schedule();
 279                         rcu_read_lock();
 280                 }
 281                 finish_wait(&mddev->sb_wait, &__wait);
 282         }
 283         atomic_inc(&mddev->active_io);
 284         rcu_read_unlock();
 285
 286         /*
 287          * save the sectors now since our bio can
 288          * go away inside make_request
 289          */
 290         sectors = bio_sectors(bio);
 291         mddev->pers->make_request(mddev, bio);
 292
 293         cpu = part_stat_lock();
 294         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 295         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
 296         part_stat_unlock();
 297
 298         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 299                 wake_up(&mddev->sb_wait);
 300 }
 301
 302 /* mddev_suspend makes sure no new requests are submitted
 303  * to the device, and that any requests that have been submitted
 304  * are completely handled.
 305  * Once ->stop is called and completes, the module will be completely
 306  * unused.
 307  */
 308 void mddev_suspend(struct mddev *mddev)
 309 {
 310         BUG_ON(mddev->suspended);
 311         mddev->suspended = 1;
 312         synchronize_rcu();
 313         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 314         mddev->pers->quiesce(mddev, 1);
 315
 316         del_timer_sync(&mddev->safemode_timer);
 317 }
 318 EXPORT_SYMBOL_GPL(mddev_suspend);
 319
 320 void mddev_resume(struct mddev *mddev)
 321 {
 322         mddev->suspended = 0;
 323         wake_up(&mddev->sb_wait);
 324         mddev->pers->quiesce(mddev, 0);
 325
 326         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 327         md_wakeup_thread(mddev->thread);
 328         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 329 }
 330 EXPORT_SYMBOL_GPL(mddev_resume);
 331
 332 int mddev_congested(struct mddev *mddev, int bits)
 333 {
 334         return mddev->suspended;
 335 }
 336 EXPORT_SYMBOL(mddev_congested);
 337
 338 /*
 339  * Generic flush handling for md
 340  */
 341
 342 static void md_end_flush(struct bio *bio, int err)
 343 {
 344         struct md_rdev *rdev = bio->bi_private;
 345         struct mddev *mddev = rdev->mddev;
 346
 347         rdev_dec_pending(rdev, mddev);
 348
 349         if (atomic_dec_and_test(&mddev->flush_pending)) {
 350                 /* The pre-request flush has finished */
 351                 queue_work(md_wq, &mddev->flush_work);
 352         }
 353         bio_put(bio);
 354 }
 355
 356 static void md_submit_flush_data(struct work_struct *ws);
 357
 358 static void submit_flushes(struct work_struct *ws)
 359 {
 360         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 361         struct md_rdev *rdev;
 362
 363         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 364         atomic_set(&mddev->flush_pending, 1);
 365         rcu_read_lock();
 366         rdev_for_each_rcu(rdev, mddev)
 367                 if (rdev->raid_disk >= 0 &&
 368                     !test_bit(Faulty, &rdev->flags)) {
 369                         /* Take two references, one is dropped
 370                          * when request finishes, one after
 371                          * we reclaim rcu_read_lock
 372                          */
 373                         struct bio *bi;
 374                         atomic_inc(&rdev->nr_pending);
 375                         atomic_inc(&rdev->nr_pending);
 376                         rcu_read_unlock();
 377                         bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
 378                         bi->bi_end_io = md_end_flush;
 379                         bi->bi_private = rdev;
 380                         bi->bi_bdev = rdev->bdev;
 381                         atomic_inc(&mddev->flush_pending);
 382                         submit_bio(WRITE_FLUSH, bi);
 383                         rcu_read_lock();
 384                         rdev_dec_pending(rdev, mddev);
 385                 }
 386         rcu_read_unlock();
 387         if (atomic_dec_and_test(&mddev->flush_pending))
 388                 queue_work(md_wq, &mddev->flush_work);
 389 }
 390
 391 static void md_submit_flush_data(struct work_struct *ws)
 392 {
 393         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 394         struct bio *bio = mddev->flush_bio;
 395
 396         if (bio->bi_size == 0)
 397                 /* an empty barrier - all done */
 398                 bio_endio(bio, 0);
 399         else {
 400                 bio->bi_rw &= ~REQ_FLUSH;
 401                 mddev->pers->make_request(mddev, bio);
 402         }
 403
 404         mddev->flush_bio = NULL;
 405         wake_up(&mddev->sb_wait);
 406 }
 407
 408 void md_flush_request(struct mddev *mddev, struct bio *bio)
 409 {
 410         spin_lock_irq(&mddev->write_lock);
 411         wait_event_lock_irq(mddev->sb_wait,
 412                             !mddev->flush_bio,
 413                             mddev->write_lock);
 414         mddev->flush_bio = bio;
 415         spin_unlock_irq(&mddev->write_lock);
 416
 417         INIT_WORK(&mddev->flush_work, submit_flushes);
 418         queue_work(md_wq, &mddev->flush_work);
 419 }
 420 EXPORT_SYMBOL(md_flush_request);
 421
 422 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
 423 {
 424         struct mddev *mddev = cb->data;
 425         md_wakeup_thread(mddev->thread);
 426         kfree(cb);
 427 }
 428 EXPORT_SYMBOL(md_unplug);
 429
 430 static inline struct mddev *mddev_get(struct mddev *mddev)
 431 {
 432         atomic_inc(&mddev->active);
 433         return mddev;
 434 }
 435
 436 static void mddev_delayed_delete(struct work_struct *ws);
 437
 438 static void mddev_put(struct mddev *mddev)
 439 {
 440         struct bio_set *bs = NULL;
 441
 442         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 443                 return;
 444         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 445             mddev->ctime == 0 && !mddev->hold_active) {
 446                 /* Array is not configured at all, and not held active,
 447                  * so destroy it */
 448                 list_del_init(&mddev->all_mddevs);
 449                 bs = mddev->bio_set;
 450                 mddev->bio_set = NULL;
 451                 if (mddev->gendisk) {
 452                         /* We did a probe so need to clean up.  Call
 453                          * queue_work inside the spinlock so that
 454                          * flush_workqueue() after mddev_find will
 455                          * succeed in waiting for the work to be done.
 456                          */
 457                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 458                         queue_work(md_misc_wq, &mddev->del_work);
 459                 } else
 460                         kfree(mddev);
 461         }
 462         spin_unlock(&all_mddevs_lock);
 463         if (bs)
 464                 bioset_free(bs);
 465 }
 466
 467 void mddev_init(struct mddev *mddev)
 468 {
 469         mutex_init(&mddev->open_mutex);
 470         mutex_init(&mddev->reconfig_mutex);
 471         mutex_init(&mddev->bitmap_info.mutex);
 472         INIT_LIST_HEAD(&mddev->disks);
 473         INIT_LIST_HEAD(&mddev->all_mddevs);
 474         init_timer(&mddev->safemode_timer);
 475         atomic_set(&mddev->active, 1);
 476         atomic_set(&mddev->openers, 0);
 477         atomic_set(&mddev->active_io, 0);
 478         spin_lock_init(&mddev->write_lock);
 479         atomic_set(&mddev->flush_pending, 0);
 480         init_waitqueue_head(&mddev->sb_wait);
 481         init_waitqueue_head(&mddev->recovery_wait);
 482         mddev->reshape_position = MaxSector;
 483         mddev->reshape_backwards = 0;
 484         mddev->last_sync_action = "none";
 485         mddev->resync_min = 0;
 486         mddev->resync_max = MaxSector;
 487         mddev->level = LEVEL_NONE;
 488 }
 489 EXPORT_SYMBOL_GPL(mddev_init);
 490
 491 static struct mddev * mddev_find(dev_t unit)
 492 {
 493         struct mddev *mddev, *new = NULL;
 494
 495         if (unit && MAJOR(unit) != MD_MAJOR)
 496                 unit &= ~((1<<MdpMinorShift)-1);
 497
 498  retry:
 499         spin_lock(&all_mddevs_lock);
 500
 501         if (unit) {
 502                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 503                         if (mddev->unit == unit) {
 504                                 mddev_get(mddev);
 505                                 spin_unlock(&all_mddevs_lock);
 506                                 kfree(new);
 507                                 return mddev;
 508                         }
 509
 510                 if (new) {
 511                         list_add(&new->all_mddevs, &all_mddevs);
 512                         spin_unlock(&all_mddevs_lock);
 513                         new->hold_active = UNTIL_IOCTL;
 514                         return new;
 515                 }
 516         } else if (new) {
 517                 /* find an unused unit number */
 518                 static int next_minor = 512;
 519                 int start = next_minor;
 520                 int is_free = 0;
 521                 int dev = 0;
 522                 while (!is_free) {
 523                         dev = MKDEV(MD_MAJOR, next_minor);
 524                         next_minor++;
 525                         if (next_minor > MINORMASK)
 526                                 next_minor = 0;
 527                         if (next_minor == start) {
 528                                 /* Oh dear, all in use. */
 529                                 spin_unlock(&all_mddevs_lock);
 530                                 kfree(new);
 531                                 return NULL;
 532                         }
 533
 534                         is_free = 1;
 535                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 536                                 if (mddev->unit == dev) {
 537                                         is_free = 0;
 538                                         break;
 539                                 }
 540                 }
 541                 new->unit = dev;
 542                 new->md_minor = MINOR(dev);
 543                 new->hold_active = UNTIL_STOP;
 544                 list_add(&new->all_mddevs, &all_mddevs);
 545                 spin_unlock(&all_mddevs_lock);
 546                 return new;
 547         }
 548         spin_unlock(&all_mddevs_lock);
 549
 550         new = kzalloc(sizeof(*new), GFP_KERNEL);
 551         if (!new)
 552                 return NULL;
 553
 554         new->unit = unit;
 555         if (MAJOR(unit) == MD_MAJOR)
 556                 new->md_minor = MINOR(unit);
 557         else
 558                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 559
 560         mddev_init(new);
 561
 562         goto retry;
 563 }
 564
 565 static inline int mddev_lock(struct mddev * mddev)
 566 {
 567         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 568 }
 569
 570 static inline int mddev_is_locked(struct mddev *mddev)
 571 {
 572         return mutex_is_locked(&mddev->reconfig_mutex);
 573 }
 574
 575 static inline int mddev_trylock(struct mddev * mddev)
 576 {
 577         return mutex_trylock(&mddev->reconfig_mutex);
 578 }
 579
 580 static struct attribute_group md_redundancy_group;
 581
 582 static void mddev_unlock(struct mddev * mddev)
 583 {
 584         if (mddev->to_remove) {
 585                 /* These cannot be removed under reconfig_mutex as
 586                  * an access to the files will try to take reconfig_mutex
 587                  * while holding the file unremovable, which leads to
 588                  * a deadlock.
 589                  * So hold set sysfs_active while the remove in happeing,
 590                  * and anything else which might set ->to_remove or my
 591                  * otherwise change the sysfs namespace will fail with
 592                  * -EBUSY if sysfs_active is still set.
 593                  * We set sysfs_active under reconfig_mutex and elsewhere
 594                  * test it under the same mutex to ensure its correct value
 595                  * is seen.
 596                  */
 597                 struct attribute_group *to_remove = mddev->to_remove;
 598                 mddev->to_remove = NULL;
 599                 mddev->sysfs_active = 1;
 600                 mutex_unlock(&mddev->reconfig_mutex);
 601
 602                 if (mddev->kobj.sd) {
 603                         if (to_remove != &md_redundancy_group)
 604                                 sysfs_remove_group(&mddev->kobj, to_remove);
 605                         if (mddev->pers == NULL ||
 606                             mddev->pers->sync_request == NULL) {
 607                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 608                                 if (mddev->sysfs_action)
 609                                         sysfs_put(mddev->sysfs_action);
 610                                 mddev->sysfs_action = NULL;
 611                         }
 612                 }
 613                 mddev->sysfs_active = 0;
 614         } else
 615                 mutex_unlock(&mddev->reconfig_mutex);
 616
 617         /* As we've dropped the mutex we need a spinlock to
 618          * make sure the thread doesn't disappear
 619          */
 620         spin_lock(&pers_lock);
 621         md_wakeup_thread(mddev->thread);
 622         spin_unlock(&pers_lock);
 623 }
 624
 625 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
 626 {
 627         struct md_rdev *rdev;
 628
 629         rdev_for_each(rdev, mddev)
 630                 if (rdev->desc_nr == nr)
 631                         return rdev;
 632
 633         return NULL;
 634 }
 635
 636 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
 637 {
 638         struct md_rdev *rdev;
 639
 640         rdev_for_each_rcu(rdev, mddev)
 641                 if (rdev->desc_nr == nr)
 642                         return rdev;
 643
 644         return NULL;
 645 }
 646
 647 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 648 {
 649         struct md_rdev *rdev;
 650
 651         rdev_for_each(rdev, mddev)
 652                 if (rdev->bdev->bd_dev == dev)
 653                         return rdev;
 654
 655         return NULL;
 656 }
 657
 658 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
 659 {
 660         struct md_rdev *rdev;
 661
 662         rdev_for_each_rcu(rdev, mddev)
 663                 if (rdev->bdev->bd_dev == dev)
 664                         return rdev;
 665
 666         return NULL;
 667 }
 668
 669 static struct md_personality *find_pers(int level, char *clevel)
 670 {
 671         struct md_personality *pers;
 672         list_for_each_entry(pers, &pers_list, list) {
 673                 if (level != LEVEL_NONE && pers->level == level)
 674                         return pers;
 675                 if (strcmp(pers->name, clevel)==0)
 676                         return pers;
 677         }
 678         return NULL;
 679 }
 680
 681 /* return the offset of the super block in 512byte sectors */
 682 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 683 {
 684         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 685         return MD_NEW_SIZE_SECTORS(num_sectors);
 686 }
 687
 688 static int alloc_disk_sb(struct md_rdev * rdev)
 689 {
 690         if (rdev->sb_page)
 691                 MD_BUG();
 692
 693         rdev->sb_page = alloc_page(GFP_KERNEL);
 694         if (!rdev->sb_page) {
 695                 printk(KERN_ALERT "md: out of memory.\n");
 696                 return -ENOMEM;
 697         }
 698
 699         return 0;
 700 }
 701
 702 void md_rdev_clear(struct md_rdev *rdev)
 703 {
 704         if (rdev->sb_page) {
 705                 put_page(rdev->sb_page);
 706                 rdev->sb_loaded = 0;
 707                 rdev->sb_page = NULL;
 708                 rdev->sb_start = 0;
 709                 rdev->sectors = 0;
 710         }
 711         if (rdev->bb_page) {
 712                 put_page(rdev->bb_page);
 713                 rdev->bb_page = NULL;
 714         }
 715         kfree(rdev->badblocks.page);
 716         rdev->badblocks.page = NULL;
 717 }
 718 EXPORT_SYMBOL_GPL(md_rdev_clear);
 719
 720 static void super_written(struct bio *bio, int error)
 721 {
 722         struct md_rdev *rdev = bio->bi_private;
 723         struct mddev *mddev = rdev->mddev;
 724
 725         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 726                 printk("md: super_written gets error=%d, uptodate=%d\n",
 727                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 728                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 729                 md_error(mddev, rdev);
 730         }
 731
 732         if (atomic_dec_and_test(&mddev->pending_writes))
 733                 wake_up(&mddev->sb_wait);
 734         bio_put(bio);
 735 }
 736
 737 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 738                    sector_t sector, int size, struct page *page)
 739 {
 740         /* write first size bytes of page to sector of rdev
 741          * Increment mddev->pending_writes before returning
 742          * and decrement it on completion, waking up sb_wait
 743          * if zero is reached.
 744          * If an error occurred, call md_error
 745          */
 746         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 747
 748         bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
 749         bio->bi_sector = sector;
 750         bio_add_page(bio, page, size, 0);
 751         bio->bi_private = rdev;
 752         bio->bi_end_io = super_written;
 753
 754         atomic_inc(&mddev->pending_writes);
 755         submit_bio(WRITE_FLUSH_FUA, bio);
 756 }
 757
 758 void md_super_wait(struct mddev *mddev)
 759 {
 760         /* wait for all superblock writes that were scheduled to complete */
 761         DEFINE_WAIT(wq);
 762         for(;;) {
 763                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 764                 if (atomic_read(&mddev->pending_writes)==0)
 765                         break;
 766                 schedule();
 767         }
 768         finish_wait(&mddev->sb_wait, &wq);
 769 }
 770
 771 static void bi_complete(struct bio *bio, int error)
 772 {
 773         complete((struct completion*)bio->bi_private);
 774 }
 775
 776 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
 777                  struct page *page, int rw, bool metadata_op)
 778 {
 779         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 780         struct completion event;
 781         int ret;
 782
 783         rw |= REQ_SYNC;
 784
 785         bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
 786                 rdev->meta_bdev : rdev->bdev;
 787         if (metadata_op)
 788                 bio->bi_sector = sector + rdev->sb_start;
 789         else if (rdev->mddev->reshape_position != MaxSector &&
 790                  (rdev->mddev->reshape_backwards ==
 791                   (sector >= rdev->mddev->reshape_position)))
 792                 bio->bi_sector = sector + rdev->new_data_offset;
 793         else
 794                 bio->bi_sector = sector + rdev->data_offset;
 795         bio_add_page(bio, page, size, 0);
 796         init_completion(&event);
 797         bio->bi_private = &event;
 798         bio->bi_end_io = bi_complete;
 799         submit_bio(rw, bio);
 800         wait_for_completion(&event);
 801
 802         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 803         bio_put(bio);
 804         return ret;
 805 }
 806 EXPORT_SYMBOL_GPL(sync_page_io);
 807
 808 static int read_disk_sb(struct md_rdev * rdev, int size)
 809 {
 810         char b[BDEVNAME_SIZE];
 811         if (!rdev->sb_page) {
 812                 MD_BUG();
 813                 return -EINVAL;
 814         }
 815         if (rdev->sb_loaded)
 816                 return 0;
 817
 818
 819         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
 820                 goto fail;
 821         rdev->sb_loaded = 1;
 822         return 0;
 823
 824 fail:
 825         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 826                 bdevname(rdev->bdev,b));
 827         return -EINVAL;
 828 }
 829
 830 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 831 {
 832         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 833                 sb1->set_uuid1 == sb2->set_uuid1 &&
 834                 sb1->set_uuid2 == sb2->set_uuid2 &&
 835                 sb1->set_uuid3 == sb2->set_uuid3;
 836 }
 837
 838 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 839 {
 840         int ret;
 841         mdp_super_t *tmp1, *tmp2;
 842
 843         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 844         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 845
 846         if (!tmp1 || !tmp2) {
 847                 ret = 0;
 848                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 849                 goto abort;
 850         }
 851
 852         *tmp1 = *sb1;
 853         *tmp2 = *sb2;
 854
 855         /*
 856          * nr_disks is not constant
 857          */
 858         tmp1->nr_disks = 0;
 859         tmp2->nr_disks = 0;
 860
 861         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 862 abort:
 863         kfree(tmp1);
 864         kfree(tmp2);
 865         return ret;
 866 }
 867
 868
 869 static u32 md_csum_fold(u32 csum)
 870 {
 871         csum = (csum & 0xffff) + (csum >> 16);
 872         return (csum & 0xffff) + (csum >> 16);
 873 }
 874
 875 static unsigned int calc_sb_csum(mdp_super_t * sb)
 876 {
 877         u64 newcsum = 0;
 878         u32 *sb32 = (u32*)sb;
 879         int i;
 880         unsigned int disk_csum, csum;
 881
 882         disk_csum = sb->sb_csum;
 883         sb->sb_csum = 0;
 884
 885         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 886                 newcsum += sb32[i];
 887         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 888
 889
 890 #ifdef CONFIG_ALPHA
 891         /* This used to use csum_partial, which was wrong for several
 892          * reasons including that different results are returned on
 893          * different architectures.  It isn't critical that we get exactly
 894          * the same return value as before (we always csum_fold before
 895          * testing, and that removes any differences).  However as we
 896          * know that csum_partial always returned a 16bit value on
 897          * alphas, do a fold to maximise conformity to previous behaviour.
 898          */
 899         sb->sb_csum = md_csum_fold(disk_csum);
 900 #else
 901         sb->sb_csum = disk_csum;
 902 #endif
 903         return csum;
 904 }
 905
 906
 907 /*
 908  * Handle superblock details.
 909  * We want to be able to handle multiple superblock formats
 910  * so we have a common interface to them all, and an array of
 911  * different handlers.
 912  * We rely on user-space to write the initial superblock, and support
 913  * reading and updating of superblocks.
 914  * Interface methods are:
 915  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
 916  *      loads and validates a superblock on dev.
 917  *      if refdev != NULL, compare superblocks on both devices
 918  *    Return:
 919  *      0 - dev has a superblock that is compatible with refdev
 920  *      1 - dev has a superblock that is compatible and newer than refdev
 921  *          so dev should be used as the refdev in future
 922  *     -EINVAL superblock incompatible or invalid
 923  *     -othererror e.g. -EIO
 924  *
 925  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
 926  *      Verify that dev is acceptable into mddev.
 927  *       The first time, mddev->raid_disks will be 0, and data from
 928  *       dev should be merged in.  Subsequent calls check that dev
 929  *       is new enough.  Return 0 or -EINVAL
 930  *
 931  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
 932  *     Update the superblock for rdev with data in mddev
 933  *     This does not write to disc.
 934  *
 935  */
 936
 937 struct super_type  {
 938         char                *name;
 939         struct module       *owner;
 940         int                 (*load_super)(struct md_rdev *rdev,
 941                                           struct md_rdev *refdev,
 942                                           int minor_version);
 943         int                 (*validate_super)(struct mddev *mddev,
 944                                               struct md_rdev *rdev);
 945         void                (*sync_super)(struct mddev *mddev,
 946                                           struct md_rdev *rdev);
 947         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
 948                                                 sector_t num_sectors);
 949         int                 (*allow_new_offset)(struct md_rdev *rdev,
 950                                                 unsigned long long new_offset);
 951 };
 952
 953 /*
 954  * Check that the given mddev has no bitmap.
 955  *
 956  * This function is called from the run method of all personalities that do not
 957  * support bitmaps. It prints an error message and returns non-zero if mddev
 958  * has a bitmap. Otherwise, it returns 0.
 959  *
 960  */
 961 int md_check_no_bitmap(struct mddev *mddev)
 962 {
 963         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
 964                 return 0;
 965         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
 966                 mdname(mddev), mddev->pers->name);
 967         return 1;
 968 }
 969 EXPORT_SYMBOL(md_check_no_bitmap);
 970
 971 /*
 972  * load_super for 0.90.0
 973  */
 974 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
 975 {
 976         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 977         mdp_super_t *sb;
 978         int ret;
 979
 980         /*
 981          * Calculate the position of the superblock (512byte sectors),
 982          * it's at the end of the disk.
 983          *
 984          * It also happens to be a multiple of 4Kb.
 985          */
 986         rdev->sb_start = calc_dev_sboffset(rdev);
 987
 988         ret = read_disk_sb(rdev, MD_SB_BYTES);
 989         if (ret) return ret;
 990
 991         ret = -EINVAL;
 992
 993         bdevname(rdev->bdev, b);
 994         sb = page_address(rdev->sb_page);
 995
 996         if (sb->md_magic != MD_SB_MAGIC) {
 997                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 998                        b);
 999                 goto abort;
1000         }
1001
1002         if (sb->major_version != 0 ||
1003             sb->minor_version < 90 ||
1004             sb->minor_version > 91) {
1005                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1006                         sb->major_version, sb->minor_version,
1007                         b);
1008                 goto abort;
1009         }
1010
1011         if (sb->raid_disks <= 0)
1012                 goto abort;
1013
1014         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1015                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1016                         b);
1017                 goto abort;
1018         }
1019
1020         rdev->preferred_minor = sb->md_minor;
1021         rdev->data_offset = 0;
1022         rdev->new_data_offset = 0;
1023         rdev->sb_size = MD_SB_BYTES;
1024         rdev->badblocks.shift = -1;
1025
1026         if (sb->level == LEVEL_MULTIPATH)
1027                 rdev->desc_nr = -1;
1028         else
1029                 rdev->desc_nr = sb->this_disk.number;
1030
1031         if (!refdev) {
1032                 ret = 1;
1033         } else {
1034                 __u64 ev1, ev2;
1035                 mdp_super_t *refsb = page_address(refdev->sb_page);
1036                 if (!uuid_equal(refsb, sb)) {
1037                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
1038                                 b, bdevname(refdev->bdev,b2));
1039                         goto abort;
1040                 }
1041                 if (!sb_equal(refsb, sb)) {
1042                         printk(KERN_WARNING "md: %s has same UUID"
1043                                " but different superblock to %s\n",
1044                                b, bdevname(refdev->bdev, b2));
1045                         goto abort;
1046                 }
1047                 ev1 = md_event(sb);
1048                 ev2 = md_event(refsb);
1049                 if (ev1 > ev2)
1050                         ret = 1;
1051                 else
1052                         ret = 0;
1053         }
1054         rdev->sectors = rdev->sb_start;
1055         /* Limit to 4TB as metadata cannot record more than that.
1056          * (not needed for Linear and RAID0 as metadata doesn't
1057          * record this size)
1058          */
1059         if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1060                 rdev->sectors = (2ULL << 32) - 2;
1061
1062         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1063                 /* "this cannot possibly happen" ... */
1064                 ret = -EINVAL;
1065
1066  abort:
1067         return ret;
1068 }
1069
1070 /*
1071  * validate_super for 0.90.0
1072  */
1073 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1074 {
1075         mdp_disk_t *desc;
1076         mdp_super_t *sb = page_address(rdev->sb_page);
1077         __u64 ev1 = md_event(sb);
1078
1079         rdev->raid_disk = -1;
1080         clear_bit(Faulty, &rdev->flags);
1081         clear_bit(In_sync, &rdev->flags);
1082         clear_bit(WriteMostly, &rdev->flags);
1083
1084         if (mddev->raid_disks == 0) {
1085                 mddev->major_version = 0;
1086                 mddev->minor_version = sb->minor_version;
1087                 mddev->patch_version = sb->patch_version;
1088                 mddev->external = 0;
1089                 mddev->chunk_sectors = sb->chunk_size >> 9;
1090                 mddev->ctime = sb->ctime;
1091                 mddev->utime = sb->utime;
1092                 mddev->level = sb->level;
1093                 mddev->clevel[0] = 0;
1094                 mddev->layout = sb->layout;
1095                 mddev->raid_disks = sb->raid_disks;
1096                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1097                 mddev->events = ev1;
1098                 mddev->bitmap_info.offset = 0;
1099                 mddev->bitmap_info.space = 0;
1100                 /* bitmap can use 60 K after the 4K superblocks */
1101                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1102                 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1103                 mddev->reshape_backwards = 0;
1104
1105                 if (mddev->minor_version >= 91) {
1106                         mddev->reshape_position = sb->reshape_position;
1107                         mddev->delta_disks = sb->delta_disks;
1108                         mddev->new_level = sb->new_level;
1109                         mddev->new_layout = sb->new_layout;
1110                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1111                         if (mddev->delta_disks < 0)
1112                                 mddev->reshape_backwards = 1;
1113                 } else {
1114                         mddev->reshape_position = MaxSector;
1115                         mddev->delta_disks = 0;
1116                         mddev->new_level = mddev->level;
1117                         mddev->new_layout = mddev->layout;
1118                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1119                 }
1120
1121                 if (sb->state & (1<<MD_SB_CLEAN))
1122                         mddev->recovery_cp = MaxSector;
1123                 else {
1124                         if (sb->events_hi == sb->cp_events_hi &&
1125                                 sb->events_lo == sb->cp_events_lo) {
1126                                 mddev->recovery_cp = sb->recovery_cp;
1127                         } else
1128                                 mddev->recovery_cp = 0;
1129                 }
1130
1131                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1132                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1133                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1134                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1135
1136                 mddev->max_disks = MD_SB_DISKS;
1137
1138                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1139                     mddev->bitmap_info.file == NULL) {
1140                         mddev->bitmap_info.offset =
1141                                 mddev->bitmap_info.default_offset;
1142                         mddev->bitmap_info.space =
1143                                 mddev->bitmap_info.default_space;
1144                 }
1145
1146         } else if (mddev->pers == NULL) {
1147                 /* Insist on good event counter while assembling, except
1148                  * for spares (which don't need an event count) */
1149                 ++ev1;
1150                 if (sb->disks[rdev->desc_nr].state & (
1151                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1152                         if (ev1 < mddev->events)
1153                                 return -EINVAL;
1154         } else if (mddev->bitmap) {
1155                 /* if adding to array with a bitmap, then we can accept an
1156                  * older device ... but not too old.
1157                  */
1158                 if (ev1 < mddev->bitmap->events_cleared)
1159                         return 0;
1160         } else {
1161                 if (ev1 < mddev->events)
1162                         /* just a hot-add of a new device, leave raid_disk at -1 */
1163                         return 0;
1164         }
1165
1166         if (mddev->level != LEVEL_MULTIPATH) {
1167                 desc = sb->disks + rdev->desc_nr;
1168
1169                 if (desc->state & (1<<MD_DISK_FAULTY))
1170                         set_bit(Faulty, &rdev->flags);
1171                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1172                             desc->raid_disk < mddev->raid_disks */) {
1173                         set_bit(In_sync, &rdev->flags);
1174                         rdev->raid_disk = desc->raid_disk;
1175                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1176                         /* active but not in sync implies recovery up to
1177                          * reshape position.  We don't know exactly where
1178                          * that is, so set to zero for now */
1179                         if (mddev->minor_version >= 91) {
1180                                 rdev->recovery_offset = 0;
1181                                 rdev->raid_disk = desc->raid_disk;
1182                         }
1183                 }
1184                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1185                         set_bit(WriteMostly, &rdev->flags);
1186         } else /* MULTIPATH are always insync */
1187                 set_bit(In_sync, &rdev->flags);
1188         return 0;
1189 }
1190
1191 /*
1192  * sync_super for 0.90.0
1193  */
1194 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1195 {
1196         mdp_super_t *sb;
1197         struct md_rdev *rdev2;
1198         int next_spare = mddev->raid_disks;
1199
1200
1201         /* make rdev->sb match mddev data..
1202          *
1203          * 1/ zero out disks
1204          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1205          * 3/ any empty disks < next_spare become removed
1206          *
1207          * disks[0] gets initialised to REMOVED because
1208          * we cannot be sure from other fields if it has
1209          * been initialised or not.
1210          */
1211         int i;
1212         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1213
1214         rdev->sb_size = MD_SB_BYTES;
1215
1216         sb = page_address(rdev->sb_page);
1217
1218         memset(sb, 0, sizeof(*sb));
1219
1220         sb->md_magic = MD_SB_MAGIC;
1221         sb->major_version = mddev->major_version;
1222         sb->patch_version = mddev->patch_version;
1223         sb->gvalid_words  = 0; /* ignored */
1224         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1225         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1226         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1227         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1228
1229         sb->ctime = mddev->ctime;
1230         sb->level = mddev->level;
1231         sb->size = mddev->dev_sectors / 2;
1232         sb->raid_disks = mddev->raid_disks;
1233         sb->md_minor = mddev->md_minor;
1234         sb->not_persistent = 0;
1235         sb->utime = mddev->utime;
1236         sb->state = 0;
1237         sb->events_hi = (mddev->events>>32);
1238         sb->events_lo = (u32)mddev->events;
1239
1240         if (mddev->reshape_position == MaxSector)
1241                 sb->minor_version = 90;
1242         else {
1243                 sb->minor_version = 91;
1244                 sb->reshape_position = mddev->reshape_position;
1245                 sb->new_level = mddev->new_level;
1246                 sb->delta_disks = mddev->delta_disks;
1247                 sb->new_layout = mddev->new_layout;
1248                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1249         }
1250         mddev->minor_version = sb->minor_version;
1251         if (mddev->in_sync)
1252         {
1253                 sb->recovery_cp = mddev->recovery_cp;
1254                 sb->cp_events_hi = (mddev->events>>32);
1255                 sb->cp_events_lo = (u32)mddev->events;
1256                 if (mddev->recovery_cp == MaxSector)
1257                         sb->state = (1<< MD_SB_CLEAN);
1258         } else
1259                 sb->recovery_cp = 0;
1260
1261         sb->layout = mddev->layout;
1262         sb->chunk_size = mddev->chunk_sectors << 9;
1263
1264         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1265                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1266
1267         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1268         rdev_for_each(rdev2, mddev) {
1269                 mdp_disk_t *d;
1270                 int desc_nr;
1271                 int is_active = test_bit(In_sync, &rdev2->flags);
1272
1273                 if (rdev2->raid_disk >= 0 &&
1274                     sb->minor_version >= 91)
1275                         /* we have nowhere to store the recovery_offset,
1276                          * but if it is not below the reshape_position,
1277                          * we can piggy-back on that.
1278                          */
1279                         is_active = 1;
1280                 if (rdev2->raid_disk < 0 ||
1281                     test_bit(Faulty, &rdev2->flags))
1282                         is_active = 0;
1283                 if (is_active)
1284                         desc_nr = rdev2->raid_disk;
1285                 else
1286                         desc_nr = next_spare++;
1287                 rdev2->desc_nr = desc_nr;
1288                 d = &sb->disks[rdev2->desc_nr];
1289                 nr_disks++;
1290                 d->number = rdev2->desc_nr;
1291                 d->major = MAJOR(rdev2->bdev->bd_dev);
1292                 d->minor = MINOR(rdev2->bdev->bd_dev);
1293                 if (is_active)
1294                         d->raid_disk = rdev2->raid_disk;
1295                 else
1296                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1297                 if (test_bit(Faulty, &rdev2->flags))
1298                         d->state = (1<<MD_DISK_FAULTY);
1299                 else if (is_active) {
1300                         d->state = (1<<MD_DISK_ACTIVE);
1301                         if (test_bit(In_sync, &rdev2->flags))
1302                                 d->state |= (1<<MD_DISK_SYNC);
1303                         active++;
1304                         working++;
1305                 } else {
1306                         d->state = 0;
1307                         spare++;
1308                         working++;
1309                 }
1310                 if (test_bit(WriteMostly, &rdev2->flags))
1311                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1312         }
1313         /* now set the "removed" and "faulty" bits on any missing devices */
1314         for (i=0 ; i < mddev->raid_disks ; i++) {
1315                 mdp_disk_t *d = &sb->disks[i];
1316                 if (d->state == 0 && d->number == 0) {
1317                         d->number = i;
1318                         d->raid_disk = i;
1319                         d->state = (1<<MD_DISK_REMOVED);
1320                         d->state |= (1<<MD_DISK_FAULTY);
1321                         failed++;
1322                 }
1323         }
1324         sb->nr_disks = nr_disks;
1325         sb->active_disks = active;
1326         sb->working_disks = working;
1327         sb->failed_disks = failed;
1328         sb->spare_disks = spare;
1329
1330         sb->this_disk = sb->disks[rdev->desc_nr];
1331         sb->sb_csum = calc_sb_csum(sb);
1332 }
1333
1334 /*
1335  * rdev_size_change for 0.90.0
1336  */
1337 static unsigned long long
1338 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1339 {
1340         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1341                 return 0; /* component must fit device */
1342         if (rdev->mddev->bitmap_info.offset)
1343                 return 0; /* can't move bitmap */
1344         rdev->sb_start = calc_dev_sboffset(rdev);
1345         if (!num_sectors || num_sectors > rdev->sb_start)
1346                 num_sectors = rdev->sb_start;
1347         /* Limit to 4TB as metadata cannot record more than that.
1348          * 4TB == 2^32 KB, or 2*2^32 sectors.
1349          */
1350         if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1351                 num_sectors = (2ULL << 32) - 2;
1352         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1353                        rdev->sb_page);
1354         md_super_wait(rdev->mddev);
1355         return num_sectors;
1356 }
1357
1358 static int
1359 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1360 {
1361         /* non-zero offset changes not possible with v0.90 */
1362         return new_offset == 0;
1363 }
1364
1365 /*
1366  * version 1 superblock
1367  */
1368
1369 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1370 {
1371         __le32 disk_csum;
1372         u32 csum;
1373         unsigned long long newcsum;
1374         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1375         __le32 *isuper = (__le32*)sb;
1376
1377         disk_csum = sb->sb_csum;
1378         sb->sb_csum = 0;
1379         newcsum = 0;
1380         for (; size >= 4; size -= 4)
1381                 newcsum += le32_to_cpu(*isuper++);
1382
1383         if (size == 2)
1384                 newcsum += le16_to_cpu(*(__le16*) isuper);
1385
1386         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1387         sb->sb_csum = disk_csum;
1388         return cpu_to_le32(csum);
1389 }
1390
1391 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1392                             int acknowledged);
1393 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1394 {
1395         struct mdp_superblock_1 *sb;
1396         int ret;
1397         sector_t sb_start;
1398         sector_t sectors;
1399         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1400         int bmask;
1401
1402         /*
1403          * Calculate the position of the superblock in 512byte sectors.
1404          * It is always aligned to a 4K boundary and
1405          * depeding on minor_version, it can be:
1406          * 0: At least 8K, but less than 12K, from end of device
1407          * 1: At start of device
1408          * 2: 4K from start of device.
1409          */
1410         switch(minor_version) {
1411         case 0:
1412                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1413                 sb_start -= 8*2;
1414                 sb_start &= ~(sector_t)(4*2-1);
1415                 break;
1416         case 1:
1417                 sb_start = 0;
1418                 break;
1419         case 2:
1420                 sb_start = 8;
1421                 break;
1422         default:
1423                 return -EINVAL;
1424         }
1425         rdev->sb_start = sb_start;
1426
1427         /* superblock is rarely larger than 1K, but it can be larger,
1428          * and it is safe to read 4k, so we do that
1429          */
1430         ret = read_disk_sb(rdev, 4096);
1431         if (ret) return ret;
1432
1433
1434         sb = page_address(rdev->sb_page);
1435
1436         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1437             sb->major_version != cpu_to_le32(1) ||
1438             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1439             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1440             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1441                 return -EINVAL;
1442
1443         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1444                 printk("md: invalid superblock checksum on %s\n",
1445                         bdevname(rdev->bdev,b));
1446                 return -EINVAL;
1447         }
1448         if (le64_to_cpu(sb->data_size) < 10) {
1449                 printk("md: data_size too small on %s\n",
1450                        bdevname(rdev->bdev,b));
1451                 return -EINVAL;
1452         }
1453         if (sb->pad0 ||
1454             sb->pad3[0] ||
1455             memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1456                 /* Some padding is non-zero, might be a new feature */
1457                 return -EINVAL;
1458
1459         rdev->preferred_minor = 0xffff;
1460         rdev->data_offset = le64_to_cpu(sb->data_offset);
1461         rdev->new_data_offset = rdev->data_offset;
1462         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1463             (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1464                 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1465         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1466
1467         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1468         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1469         if (rdev->sb_size & bmask)
1470                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1471
1472         if (minor_version
1473             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1474                 return -EINVAL;
1475         if (minor_version
1476             && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1477                 return -EINVAL;
1478
1479         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1480                 rdev->desc_nr = -1;
1481         else
1482                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1483
1484         if (!rdev->bb_page) {
1485                 rdev->bb_page = alloc_page(GFP_KERNEL);
1486                 if (!rdev->bb_page)
1487                         return -ENOMEM;
1488         }
1489         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1490             rdev->badblocks.count == 0) {
1491                 /* need to load the bad block list.
1492                  * Currently we limit it to one page.
1493                  */
1494                 s32 offset;
1495                 sector_t bb_sector;
1496                 u64 *bbp;
1497                 int i;
1498                 int sectors = le16_to_cpu(sb->bblog_size);
1499                 if (sectors > (PAGE_SIZE / 512))
1500                         return -EINVAL;
1501                 offset = le32_to_cpu(sb->bblog_offset);
1502                 if (offset == 0)
1503                         return -EINVAL;
1504                 bb_sector = (long long)offset;
1505                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1506                                   rdev->bb_page, READ, true))
1507                         return -EIO;
1508                 bbp = (u64 *)page_address(rdev->bb_page);
1509                 rdev->badblocks.shift = sb->bblog_shift;
1510                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1511                         u64 bb = le64_to_cpu(*bbp);
1512                         int count = bb & (0x3ff);
1513                         u64 sector = bb >> 10;
1514                         sector <<= sb->bblog_shift;
1515                         count <<= sb->bblog_shift;
1516                         if (bb + 1 == 0)
1517                                 break;
1518                         if (md_set_badblocks(&rdev->badblocks,
1519                                              sector, count, 1) == 0)
1520                                 return -EINVAL;
1521                 }
1522         } else if (sb->bblog_offset != 0)
1523                 rdev->badblocks.shift = 0;
1524
1525         if (!refdev) {
1526                 ret = 1;
1527         } else {
1528                 __u64 ev1, ev2;
1529                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1530
1531                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1532                     sb->level != refsb->level ||
1533                     sb->layout != refsb->layout ||
1534                     sb->chunksize != refsb->chunksize) {
1535                         printk(KERN_WARNING "md: %s has strangely different"
1536                                 " superblock to %s\n",
1537                                 bdevname(rdev->bdev,b),
1538                                 bdevname(refdev->bdev,b2));
1539                         return -EINVAL;
1540                 }
1541                 ev1 = le64_to_cpu(sb->events);
1542                 ev2 = le64_to_cpu(refsb->events);
1543
1544                 if (ev1 > ev2)
1545                         ret = 1;
1546                 else
1547                         ret = 0;
1548         }
1549         if (minor_version) {
1550                 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1551                 sectors -= rdev->data_offset;
1552         } else
1553                 sectors = rdev->sb_start;
1554         if (sectors < le64_to_cpu(sb->data_size))
1555                 return -EINVAL;
1556         rdev->sectors = le64_to_cpu(sb->data_size);
1557         return ret;
1558 }
1559
1560 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1561 {
1562         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1563         __u64 ev1 = le64_to_cpu(sb->events);
1564
1565         rdev->raid_disk = -1;
1566         clear_bit(Faulty, &rdev->flags);
1567         clear_bit(In_sync, &rdev->flags);
1568         clear_bit(WriteMostly, &rdev->flags);
1569
1570         if (mddev->raid_disks == 0) {
1571                 mddev->major_version = 1;
1572                 mddev->patch_version = 0;
1573                 mddev->external = 0;
1574                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1575                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1576                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1577                 mddev->level = le32_to_cpu(sb->level);
1578                 mddev->clevel[0] = 0;
1579                 mddev->layout = le32_to_cpu(sb->layout);
1580                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1581                 mddev->dev_sectors = le64_to_cpu(sb->size);
1582                 mddev->events = ev1;
1583                 mddev->bitmap_info.offset = 0;
1584                 mddev->bitmap_info.space = 0;
1585                 /* Default location for bitmap is 1K after superblock
1586                  * using 3K - total of 4K
1587                  */
1588                 mddev->bitmap_info.default_offset = 1024 >> 9;
1589                 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1590                 mddev->reshape_backwards = 0;
1591
1592                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1593                 memcpy(mddev->uuid, sb->set_uuid, 16);
1594
1595                 mddev->max_disks =  (4096-256)/2;
1596
1597                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1598                     mddev->bitmap_info.file == NULL) {
1599                         mddev->bitmap_info.offset =
1600                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1601                         /* Metadata doesn't record how much space is available.
1602                          * For 1.0, we assume we can use up to the superblock
1603                          * if before, else to 4K beyond superblock.
1604                          * For others, assume no change is possible.
1605                          */
1606                         if (mddev->minor_version > 0)
1607                                 mddev->bitmap_info.space = 0;
1608                         else if (mddev->bitmap_info.offset > 0)
1609                                 mddev->bitmap_info.space =
1610                                         8 - mddev->bitmap_info.offset;
1611                         else
1612                                 mddev->bitmap_info.space =
1613                                         -mddev->bitmap_info.offset;
1614                 }
1615
1616                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1617                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1618                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1619                         mddev->new_level = le32_to_cpu(sb->new_level);
1620                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1621                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1622                         if (mddev->delta_disks < 0 ||
1623                             (mddev->delta_disks == 0 &&
1624                              (le32_to_cpu(sb->feature_map)
1625                               & MD_FEATURE_RESHAPE_BACKWARDS)))
1626                                 mddev->reshape_backwards = 1;
1627                 } else {
1628                         mddev->reshape_position = MaxSector;
1629                         mddev->delta_disks = 0;
1630                         mddev->new_level = mddev->level;
1631                         mddev->new_layout = mddev->layout;
1632                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1633                 }
1634
1635         } else if (mddev->pers == NULL) {
1636                 /* Insist of good event counter while assembling, except for
1637                  * spares (which don't need an event count) */
1638                 ++ev1;
1639                 if (rdev->desc_nr >= 0 &&
1640                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1641                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1642                         if (ev1 < mddev->events)
1643                                 return -EINVAL;
1644         } else if (mddev->bitmap) {
1645                 /* If adding to array with a bitmap, then we can accept an
1646                  * older device, but not too old.
1647                  */
1648                 if (ev1 < mddev->bitmap->events_cleared)
1649                         return 0;
1650         } else {
1651                 if (ev1 < mddev->events)
1652                         /* just a hot-add of a new device, leave raid_disk at -1 */
1653                         return 0;
1654         }
1655         if (mddev->level != LEVEL_MULTIPATH) {
1656                 int role;
1657                 if (rdev->desc_nr < 0 ||
1658                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1659                         role = 0xffff;
1660                         rdev->desc_nr = -1;
1661                 } else
1662                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1663                 switch(role) {
1664                 case 0xffff: /* spare */
1665                         break;
1666                 case 0xfffe: /* faulty */
1667                         set_bit(Faulty, &rdev->flags);
1668                         break;
1669                 default:
1670                         if ((le32_to_cpu(sb->feature_map) &
1671                              MD_FEATURE_RECOVERY_OFFSET))
1672                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1673                         else
1674                                 set_bit(In_sync, &rdev->flags);
1675                         rdev->raid_disk = role;
1676                         break;
1677                 }
1678                 if (sb->devflags & WriteMostly1)
1679                         set_bit(WriteMostly, &rdev->flags);
1680                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1681                         set_bit(Replacement, &rdev->flags);
1682         } else /* MULTIPATH are always insync */
1683                 set_bit(In_sync, &rdev->flags);
1684
1685         return 0;
1686 }
1687
1688 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1689 {
1690         struct mdp_superblock_1 *sb;
1691         struct md_rdev *rdev2;
1692         int max_dev, i;
1693         /* make rdev->sb match mddev and rdev data. */
1694
1695         sb = page_address(rdev->sb_page);
1696
1697         sb->feature_map = 0;
1698         sb->pad0 = 0;
1699         sb->recovery_offset = cpu_to_le64(0);
1700         memset(sb->pad3, 0, sizeof(sb->pad3));
1701
1702         sb->utime = cpu_to_le64((__u64)mddev->utime);
1703         sb->events = cpu_to_le64(mddev->events);
1704         if (mddev->in_sync)
1705                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1706         else
1707                 sb->resync_offset = cpu_to_le64(0);
1708
1709         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1710
1711         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1712         sb->size = cpu_to_le64(mddev->dev_sectors);
1713         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1714         sb->level = cpu_to_le32(mddev->level);
1715         sb->layout = cpu_to_le32(mddev->layout);
1716
1717         if (test_bit(WriteMostly, &rdev->flags))
1718                 sb->devflags |= WriteMostly1;
1719         else
1720                 sb->devflags &= ~WriteMostly1;
1721         sb->data_offset = cpu_to_le64(rdev->data_offset);
1722         sb->data_size = cpu_to_le64(rdev->sectors);
1723
1724         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1725                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1726                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1727         }
1728
1729         if (rdev->raid_disk >= 0 &&
1730             !test_bit(In_sync, &rdev->flags)) {
1731                 sb->feature_map |=
1732                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1733                 sb->recovery_offset =
1734                         cpu_to_le64(rdev->recovery_offset);
1735         }
1736         if (test_bit(Replacement, &rdev->flags))
1737                 sb->feature_map |=
1738                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
1739
1740         if (mddev->reshape_position != MaxSector) {
1741                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1742                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1743                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1744                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1745                 sb->new_level = cpu_to_le32(mddev->new_level);
1746                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1747                 if (mddev->delta_disks == 0 &&
1748                     mddev->reshape_backwards)
1749                         sb->feature_map
1750                                 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1751                 if (rdev->new_data_offset != rdev->data_offset) {
1752                         sb->feature_map
1753                                 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1754                         sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1755                                                              - rdev->data_offset));
1756                 }
1757         }
1758
1759         if (rdev->badblocks.count == 0)
1760                 /* Nothing to do for bad blocks*/ ;
1761         else if (sb->bblog_offset == 0)
1762                 /* Cannot record bad blocks on this device */
1763                 md_error(mddev, rdev);
1764         else {
1765                 struct badblocks *bb = &rdev->badblocks;
1766                 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1767                 u64 *p = bb->page;
1768                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1769                 if (bb->changed) {
1770                         unsigned seq;
1771
1772 retry:
1773                         seq = read_seqbegin(&bb->lock);
1774
1775                         memset(bbp, 0xff, PAGE_SIZE);
1776
1777                         for (i = 0 ; i < bb->count ; i++) {
1778                                 u64 internal_bb = p[i];
1779                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1780                                                 | BB_LEN(internal_bb));
1781                                 bbp[i] = cpu_to_le64(store_bb);
1782                         }
1783                         bb->changed = 0;
1784                         if (read_seqretry(&bb->lock, seq))
1785                                 goto retry;
1786
1787                         bb->sector = (rdev->sb_start +
1788                                       (int)le32_to_cpu(sb->bblog_offset));
1789                         bb->size = le16_to_cpu(sb->bblog_size);
1790                 }
1791         }
1792
1793         max_dev = 0;
1794         rdev_for_each(rdev2, mddev)
1795                 if (rdev2->desc_nr+1 > max_dev)
1796                         max_dev = rdev2->desc_nr+1;
1797
1798         if (max_dev > le32_to_cpu(sb->max_dev)) {
1799                 int bmask;
1800                 sb->max_dev = cpu_to_le32(max_dev);
1801                 rdev->sb_size = max_dev * 2 + 256;
1802                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1803                 if (rdev->sb_size & bmask)
1804                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1805         } else
1806                 max_dev = le32_to_cpu(sb->max_dev);
1807
1808         for (i=0; i<max_dev;i++)
1809                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1810
1811         rdev_for_each(rdev2, mddev) {
1812                 i = rdev2->desc_nr;
1813                 if (test_bit(Faulty, &rdev2->flags))
1814                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1815                 else if (test_bit(In_sync, &rdev2->flags))
1816                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1817                 else if (rdev2->raid_disk >= 0)
1818                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1819                 else
1820                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1821         }
1822
1823         sb->sb_csum = calc_sb_1_csum(sb);
1824 }
1825
1826 static unsigned long long
1827 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1828 {
1829         struct mdp_superblock_1 *sb;
1830         sector_t max_sectors;
1831         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1832                 return 0; /* component must fit device */
1833         if (rdev->data_offset != rdev->new_data_offset)
1834                 return 0; /* too confusing */
1835         if (rdev->sb_start < rdev->data_offset) {
1836                 /* minor versions 1 and 2; superblock before data */
1837                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1838                 max_sectors -= rdev->data_offset;
1839                 if (!num_sectors || num_sectors > max_sectors)
1840                         num_sectors = max_sectors;
1841         } else if (rdev->mddev->bitmap_info.offset) {
1842                 /* minor version 0 with bitmap we can't move */
1843                 return 0;
1844         } else {
1845                 /* minor version 0; superblock after data */
1846                 sector_t sb_start;
1847                 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1848                 sb_start &= ~(sector_t)(4*2 - 1);
1849                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1850                 if (!num_sectors || num_sectors > max_sectors)
1851                         num_sectors = max_sectors;
1852                 rdev->sb_start = sb_start;
1853         }
1854         sb = page_address(rdev->sb_page);
1855         sb->data_size = cpu_to_le64(num_sectors);
1856         sb->super_offset = rdev->sb_start;
1857         sb->sb_csum = calc_sb_1_csum(sb);
1858         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1859                        rdev->sb_page);
1860         md_super_wait(rdev->mddev);
1861         return num_sectors;
1862
1863 }
1864
1865 static int
1866 super_1_allow_new_offset(struct md_rdev *rdev,
1867                          unsigned long long new_offset)
1868 {
1869         /* All necessary checks on new >= old have been done */
1870         struct bitmap *bitmap;
1871         if (new_offset >= rdev->data_offset)
1872                 return 1;
1873
1874         /* with 1.0 metadata, there is no metadata to tread on
1875          * so we can always move back */
1876         if (rdev->mddev->minor_version == 0)
1877                 return 1;
1878
1879         /* otherwise we must be sure not to step on
1880          * any metadata, so stay:
1881          * 36K beyond start of superblock
1882          * beyond end of badblocks
1883          * beyond write-intent bitmap
1884          */
1885         if (rdev->sb_start + (32+4)*2 > new_offset)
1886                 return 0;
1887         bitmap = rdev->mddev->bitmap;
1888         if (bitmap && !rdev->mddev->bitmap_info.file &&
1889             rdev->sb_start + rdev->mddev->bitmap_info.offset +
1890             bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1891                 return 0;
1892         if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1893                 return 0;
1894
1895         return 1;
1896 }
1897
1898 static struct super_type super_types[] = {
1899         [0] = {
1900                 .name   = "0.90.0",
1901                 .owner  = THIS_MODULE,
1902                 .load_super         = super_90_load,
1903                 .validate_super     = super_90_validate,
1904                 .sync_super         = super_90_sync,
1905                 .rdev_size_change   = super_90_rdev_size_change,
1906                 .allow_new_offset   = super_90_allow_new_offset,
1907         },
1908         [1] = {
1909                 .name   = "md-1",
1910                 .owner  = THIS_MODULE,
1911                 .load_super         = super_1_load,
1912                 .validate_super     = super_1_validate,
1913                 .sync_super         = super_1_sync,
1914                 .rdev_size_change   = super_1_rdev_size_change,
1915                 .allow_new_offset   = super_1_allow_new_offset,
1916         },
1917 };
1918
1919 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1920 {
1921         if (mddev->sync_super) {
1922                 mddev->sync_super(mddev, rdev);
1923                 return;
1924         }
1925
1926         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1927
1928         super_types[mddev->major_version].sync_super(mddev, rdev);
1929 }
1930
1931 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1932 {
1933         struct md_rdev *rdev, *rdev2;
1934
1935         rcu_read_lock();
1936         rdev_for_each_rcu(rdev, mddev1)
1937                 rdev_for_each_rcu(rdev2, mddev2)
1938                         if (rdev->bdev->bd_contains ==
1939                             rdev2->bdev->bd_contains) {
1940                                 rcu_read_unlock();
1941                                 return 1;
1942                         }
1943         rcu_read_unlock();
1944         return 0;
1945 }
1946
1947 static LIST_HEAD(pending_raid_disks);
1948
1949 /*
1950  * Try to register data integrity profile for an mddev
1951  *
1952  * This is called when an array is started and after a disk has been kicked
1953  * from the array. It only succeeds if all working and active component devices
1954  * are integrity capable with matching profiles.
1955  */
1956 int md_integrity_register(struct mddev *mddev)
1957 {
1958         struct md_rdev *rdev, *reference = NULL;
1959
1960         if (list_empty(&mddev->disks))
1961                 return 0; /* nothing to do */
1962         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1963                 return 0; /* shouldn't register, or already is */
1964         rdev_for_each(rdev, mddev) {
1965                 /* skip spares and non-functional disks */
1966                 if (test_bit(Faulty, &rdev->flags))
1967                         continue;
1968                 if (rdev->raid_disk < 0)
1969                         continue;
1970                 if (!reference) {
1971                         /* Use the first rdev as the reference */
1972                         reference = rdev;
1973                         continue;
1974                 }
1975                 /* does this rdev's profile match the reference profile? */
1976                 if (blk_integrity_compare(reference->bdev->bd_disk,
1977                                 rdev->bdev->bd_disk) < 0)
1978                         return -EINVAL;
1979         }
1980         if (!reference || !bdev_get_integrity(reference->bdev))
1981                 return 0;
1982         /*
1983          * All component devices are integrity capable and have matching
1984          * profiles, register the common profile for the md device.
1985          */
1986         if (blk_integrity_register(mddev->gendisk,
1987                         bdev_get_integrity(reference->bdev)) != 0) {
1988                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1989                         mdname(mddev));
1990                 return -EINVAL;
1991         }
1992         printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1993         if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1994                 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1995                        mdname(mddev));
1996                 return -EINVAL;
1997         }
1998         return 0;
1999 }
2000 EXPORT_SYMBOL(md_integrity_register);
2001
2002 /* Disable data integrity if non-capable/non-matching disk is being added */
2003 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2004 {
2005         struct blk_integrity *bi_rdev;
2006         struct blk_integrity *bi_mddev;
2007
2008         if (!mddev->gendisk)
2009                 return;
2010
2011         bi_rdev = bdev_get_integrity(rdev->bdev);
2012         bi_mddev = blk_get_integrity(mddev->gendisk);
2013
2014         if (!bi_mddev) /* nothing to do */
2015                 return;
2016         if (rdev->raid_disk < 0) /* skip spares */
2017                 return;
2018         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2019                                              rdev->bdev->bd_disk) >= 0)
2020                 return;
2021         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2022         blk_integrity_unregister(mddev->gendisk);
2023 }
2024 EXPORT_SYMBOL(md_integrity_add_rdev);
2025
2026 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2027 {
2028         char b[BDEVNAME_SIZE];
2029         struct kobject *ko;
2030         char *s;
2031         int err;
2032
2033         if (rdev->mddev) {
2034                 MD_BUG();
2035                 return -EINVAL;
2036         }
2037
2038         /* prevent duplicates */
2039         if (find_rdev(mddev, rdev->bdev->bd_dev))
2040                 return -EEXIST;
2041
2042         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2043         if (rdev->sectors && (mddev->dev_sectors == 0 ||
2044                         rdev->sectors < mddev->dev_sectors)) {
2045                 if (mddev->pers) {
2046                         /* Cannot change size, so fail
2047                          * If mddev->level <= 0, then we don't care
2048                          * about aligning sizes (e.g. linear)
2049                          */
2050                         if (mddev->level > 0)
2051                                 return -ENOSPC;
2052                 } else
2053                         mddev->dev_sectors = rdev->sectors;
2054         }
2055
2056         /* Verify rdev->desc_nr is unique.
2057          * If it is -1, assign a free number, else
2058          * check number is not in use
2059          */
2060         if (rdev->desc_nr < 0) {
2061                 int choice = 0;
2062                 if (mddev->pers) choice = mddev->raid_disks;
2063                 while (find_rdev_nr(mddev, choice))
2064                         choice++;
2065                 rdev->desc_nr = choice;
2066         } else {
2067                 if (find_rdev_nr(mddev, rdev->desc_nr))
2068                         return -EBUSY;
2069         }
2070         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2071                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2072                        mdname(mddev), mddev->max_disks);
2073                 return -EBUSY;
2074         }
2075         bdevname(rdev->bdev,b);
2076         while ( (s=strchr(b, '/')) != NULL)
2077                 *s = '!';
2078
2079         rdev->mddev = mddev;
2080         printk(KERN_INFO "md: bind<%s>\n", b);
2081
2082         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2083                 goto fail;
2084
2085         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2086         if (sysfs_create_link(&rdev->kobj, ko, "block"))
2087                 /* failure here is OK */;
2088         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2089
2090         list_add_rcu(&rdev->same_set, &mddev->disks);
2091         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2092
2093         /* May as well allow recovery to be retried once */
2094         mddev->recovery_disabled++;
2095
2096         return 0;
2097
2098  fail:
2099         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2100                b, mdname(mddev));
2101         return err;
2102 }
2103
2104 static void md_delayed_delete(struct work_struct *ws)
2105 {
2106         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2107         kobject_del(&rdev->kobj);
2108         kobject_put(&rdev->kobj);
2109 }
2110
2111 static void unbind_rdev_from_array(struct md_rdev * rdev)
2112 {
2113         char b[BDEVNAME_SIZE];
2114         if (!rdev->mddev) {
2115                 MD_BUG();
2116                 return;
2117         }
2118         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2119         list_del_rcu(&rdev->same_set);
2120         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2121         rdev->mddev = NULL;
2122         sysfs_remove_link(&rdev->kobj, "block");
2123         sysfs_put(rdev->sysfs_state);
2124         rdev->sysfs_state = NULL;
2125         rdev->badblocks.count = 0;
2126         /* We need to delay this, otherwise we can deadlock when
2127          * writing to 'remove' to "dev/state".  We also need
2128          * to delay it due to rcu usage.
2129          */
2130         synchronize_rcu();
2131         INIT_WORK(&rdev->del_work, md_delayed_delete);
2132         kobject_get(&rdev->kobj);
2133         queue_work(md_misc_wq, &rdev->del_work);
2134 }
2135
2136 /*
2137  * prevent the device from being mounted, repartitioned or
2138  * otherwise reused by a RAID array (or any other kernel
2139  * subsystem), by bd_claiming the device.
2140  */
2141 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2142 {
2143         int err = 0;
2144         struct block_device *bdev;
2145         char b[BDEVNAME_SIZE];
2146
2147         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2148                                  shared ? (struct md_rdev *)lock_rdev : rdev);
2149         if (IS_ERR(bdev)) {
2150                 printk(KERN_ERR "md: could not open %s.\n",
2151                         __bdevname(dev, b));
2152                 return PTR_ERR(bdev);
2153         }
2154         rdev->bdev = bdev;
2155         return err;
2156 }
2157
2158 static void unlock_rdev(struct md_rdev *rdev)
2159 {
2160         struct block_device *bdev = rdev->bdev;
2161         rdev->bdev = NULL;
2162         if (!bdev)
2163                 MD_BUG();
2164         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2165 }
2166
2167 void md_autodetect_dev(dev_t dev);
2168
2169 static void export_rdev(struct md_rdev * rdev)
2170 {
2171         char b[BDEVNAME_SIZE];
2172         printk(KERN_INFO "md: export_rdev(%s)\n",
2173                 bdevname(rdev->bdev,b));
2174         if (rdev->mddev)
2175                 MD_BUG();
2176         md_rdev_clear(rdev);
2177 #ifndef MODULE
2178         if (test_bit(AutoDetected, &rdev->flags))
2179                 md_autodetect_dev(rdev->bdev->bd_dev);
2180 #endif
2181         unlock_rdev(rdev);
2182         kobject_put(&rdev->kobj);
2183 }
2184
2185 static void kick_rdev_from_array(struct md_rdev * rdev)
2186 {
2187         unbind_rdev_from_array(rdev);
2188         export_rdev(rdev);
2189 }
2190
2191 static void export_array(struct mddev *mddev)
2192 {
2193         struct md_rdev *rdev, *tmp;
2194
2195         rdev_for_each_safe(rdev, tmp, mddev) {
2196                 if (!rdev->mddev) {
2197                         MD_BUG();
2198                         continue;
2199                 }
2200                 kick_rdev_from_array(rdev);
2201         }
2202         if (!list_empty(&mddev->disks))
2203                 MD_BUG();
2204         mddev->raid_disks = 0;
2205         mddev->major_version = 0;
2206 }
2207
2208 static void print_desc(mdp_disk_t *desc)
2209 {
2210         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2211                 desc->major,desc->minor,desc->raid_disk,desc->state);
2212 }
2213
2214 static void print_sb_90(mdp_super_t *sb)
2215 {
2216         int i;
2217
2218         printk(KERN_INFO
2219                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2220                 sb->major_version, sb->minor_version, sb->patch_version,
2221                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2222                 sb->ctime);
2223         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2224                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2225                 sb->md_minor, sb->layout, sb->chunk_size);
2226         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2227                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2228                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2229                 sb->failed_disks, sb->spare_disks,
2230                 sb->sb_csum, (unsigned long)sb->events_lo);
2231
2232         printk(KERN_INFO);
2233         for (i = 0; i < MD_SB_DISKS; i++) {
2234                 mdp_disk_t *desc;
2235
2236                 desc = sb->disks + i;
2237                 if (desc->number || desc->major || desc->minor ||
2238                     desc->raid_disk || (desc->state && (desc->state != 4))) {
2239                         printk("     D %2d: ", i);
2240                         print_desc(desc);
2241                 }
2242         }
2243         printk(KERN_INFO "md:     THIS: ");
2244         print_desc(&sb->this_disk);
2245 }
2246
2247 static void print_sb_1(struct mdp_superblock_1 *sb)
2248 {
2249         __u8 *uuid;
2250
2251         uuid = sb->set_uuid;
2252         printk(KERN_INFO
2253                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2254                "md:    Name: \"%s\" CT:%llu\n",
2255                 le32_to_cpu(sb->major_version),
2256                 le32_to_cpu(sb->feature_map),
2257                 uuid,
2258                 sb->set_name,
2259                 (unsigned long long)le64_to_cpu(sb->ctime)
2260                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2261
2262         uuid = sb->device_uuid;
2263         printk(KERN_INFO
2264                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2265                         " RO:%llu\n"
2266                "md:     Dev:%08x UUID: %pU\n"
2267                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2268                "md:         (MaxDev:%u) \n",
2269                 le32_to_cpu(sb->level),
2270                 (unsigned long long)le64_to_cpu(sb->size),
2271                 le32_to_cpu(sb->raid_disks),
2272                 le32_to_cpu(sb->layout),
2273                 le32_to_cpu(sb->chunksize),
2274                 (unsigned long long)le64_to_cpu(sb->data_offset),
2275                 (unsigned long long)le64_to_cpu(sb->data_size),
2276                 (unsigned long long)le64_to_cpu(sb->super_offset),
2277                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2278                 le32_to_cpu(sb->dev_number),
2279                 uuid,
2280                 sb->devflags,
2281                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2282                 (unsigned long long)le64_to_cpu(sb->events),
2283                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2284                 le32_to_cpu(sb->sb_csum),
2285                 le32_to_cpu(sb->max_dev)
2286                 );
2287 }
2288
2289 static void print_rdev(struct md_rdev *rdev, int major_version)
2290 {
2291         char b[BDEVNAME_SIZE];
2292         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2293                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2294                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2295                 rdev->desc_nr);
2296         if (rdev->sb_loaded) {
2297                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2298                 switch (major_version) {
2299                 case 0:
2300                         print_sb_90(page_address(rdev->sb_page));
2301                         break;
2302                 case 1:
2303                         print_sb_1(page_address(rdev->sb_page));
2304                         break;
2305                 }
2306         } else
2307                 printk(KERN_INFO "md: no rdev superblock!\n");
2308 }
2309
2310 static void md_print_devices(void)
2311 {
2312         struct list_head *tmp;
2313         struct md_rdev *rdev;
2314         struct mddev *mddev;
2315         char b[BDEVNAME_SIZE];
2316
2317         printk("\n");
2318         printk("md:     **********************************\n");
2319         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2320         printk("md:     **********************************\n");
2321         for_each_mddev(mddev, tmp) {
2322
2323                 if (mddev->bitmap)
2324                         bitmap_print_sb(mddev->bitmap);
2325                 else
2326                         printk("%s: ", mdname(mddev));
2327                 rdev_for_each(rdev, mddev)
2328                         printk("<%s>", bdevname(rdev->bdev,b));
2329                 printk("\n");
2330
2331                 rdev_for_each(rdev, mddev)
2332                         print_rdev(rdev, mddev->major_version);
2333         }
2334         printk("md:     **********************************\n");
2335         printk("\n");
2336 }
2337
2338
2339 static void sync_sbs(struct mddev * mddev, int nospares)
2340 {
2341         /* Update each superblock (in-memory image), but
2342          * if we are allowed to, skip spares which already
2343          * have the right event counter, or have one earlier
2344          * (which would mean they aren't being marked as dirty
2345          * with the rest of the array)
2346          */
2347         struct md_rdev *rdev;
2348         rdev_for_each(rdev, mddev) {
2349                 if (rdev->sb_events == mddev->events ||
2350                     (nospares &&
2351                      rdev->raid_disk < 0 &&
2352                      rdev->sb_events+1 == mddev->events)) {
2353                         /* Don't update this superblock */
2354                         rdev->sb_loaded = 2;
2355                 } else {
2356                         sync_super(mddev, rdev);
2357                         rdev->sb_loaded = 1;
2358                 }
2359         }
2360 }
2361
2362 static void md_update_sb(struct mddev * mddev, int force_change)
2363 {
2364         struct md_rdev *rdev;
2365         int sync_req;
2366         int nospares = 0;
2367         int any_badblocks_changed = 0;
2368
2369         if (mddev->ro) {
2370                 if (force_change)
2371                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
2372                 return;
2373         }
2374 repeat:
2375         /* First make sure individual recovery_offsets are correct */
2376         rdev_for_each(rdev, mddev) {
2377                 if (rdev->raid_disk >= 0 &&
2378                     mddev->delta_disks >= 0 &&
2379                     !test_bit(In_sync, &rdev->flags) &&
2380                     mddev->curr_resync_completed > rdev->recovery_offset)
2381                                 rdev->recovery_offset = mddev->curr_resync_completed;
2382
2383         }
2384         if (!mddev->persistent) {
2385                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2386                 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2387                 if (!mddev->external) {
2388                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2389                         rdev_for_each(rdev, mddev) {
2390                                 if (rdev->badblocks.changed) {
2391                                         rdev->badblocks.changed = 0;
2392                                         md_ack_all_badblocks(&rdev->badblocks);
2393                                         md_error(mddev, rdev);
2394                                 }
2395                                 clear_bit(Blocked, &rdev->flags);
2396                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2397                                 wake_up(&rdev->blocked_wait);
2398                         }
2399                 }
2400                 wake_up(&mddev->sb_wait);
2401                 return;
2402         }
2403
2404         spin_lock_irq(&mddev->write_lock);
2405
2406         mddev->utime = get_seconds();
2407
2408         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2409                 force_change = 1;
2410         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2411                 /* just a clean<-> dirty transition, possibly leave spares alone,
2412                  * though if events isn't the right even/odd, we will have to do
2413                  * spares after all
2414                  */
2415                 nospares = 1;
2416         if (force_change)
2417                 nospares = 0;
2418         if (mddev->degraded)
2419                 /* If the array is degraded, then skipping spares is both
2420                  * dangerous and fairly pointless.
2421                  * Dangerous because a device that was removed from the array
2422                  * might have a event_count that still looks up-to-date,
2423                  * so it can be re-added without a resync.
2424                  * Pointless because if there are any spares to skip,
2425                  * then a recovery will happen and soon that array won't
2426                  * be degraded any more and the spare can go back to sleep then.
2427                  */
2428                 nospares = 0;
2429
2430         sync_req = mddev->in_sync;
2431
2432         /* If this is just a dirty<->clean transition, and the array is clean
2433          * and 'events' is odd, we can roll back to the previous clean state */
2434         if (nospares
2435             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2436             && mddev->can_decrease_events
2437             && mddev->events != 1) {
2438                 mddev->events--;
2439                 mddev->can_decrease_events = 0;
2440         } else {
2441                 /* otherwise we have to go forward and ... */
2442                 mddev->events ++;
2443                 mddev->can_decrease_events = nospares;
2444         }
2445
2446         if (!mddev->events) {
2447                 /*
2448                  * oops, this 64-bit counter should never wrap.
2449                  * Either we are in around ~1 trillion A.C., assuming
2450                  * 1 reboot per second, or we have a bug:
2451                  */
2452                 MD_BUG();
2453                 mddev->events --;
2454         }
2455
2456         rdev_for_each(rdev, mddev) {
2457                 if (rdev->badblocks.changed)
2458                         any_badblocks_changed++;
2459                 if (test_bit(Faulty, &rdev->flags))
2460                         set_bit(FaultRecorded, &rdev->flags);
2461         }
2462
2463         sync_sbs(mddev, nospares);
2464         spin_unlock_irq(&mddev->write_lock);
2465
2466         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2467                  mdname(mddev), mddev->in_sync);
2468
2469         bitmap_update_sb(mddev->bitmap);
2470         rdev_for_each(rdev, mddev) {
2471                 char b[BDEVNAME_SIZE];
2472
2473                 if (rdev->sb_loaded != 1)
2474                         continue; /* no noise on spare devices */
2475
2476                 if (!test_bit(Faulty, &rdev->flags) &&
2477                     rdev->saved_raid_disk == -1) {
2478                         md_super_write(mddev,rdev,
2479                                        rdev->sb_start, rdev->sb_size,
2480                                        rdev->sb_page);
2481                         pr_debug("md: (write) %s's sb offset: %llu\n",
2482                                  bdevname(rdev->bdev, b),
2483                                  (unsigned long long)rdev->sb_start);
2484                         rdev->sb_events = mddev->events;
2485                         if (rdev->badblocks.size) {
2486                                 md_super_write(mddev, rdev,
2487                                                rdev->badblocks.sector,
2488                                                rdev->badblocks.size << 9,
2489                                                rdev->bb_page);
2490                                 rdev->badblocks.size = 0;
2491                         }
2492
2493                 } else if (test_bit(Faulty, &rdev->flags))
2494                         pr_debug("md: %s (skipping faulty)\n",
2495                                  bdevname(rdev->bdev, b));
2496                 else
2497                         pr_debug("(skipping incremental s/r ");
2498
2499                 if (mddev->level == LEVEL_MULTIPATH)
2500                         /* only need to write one superblock... */
2501                         break;
2502         }
2503         md_super_wait(mddev);
2504         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2505
2506         spin_lock_irq(&mddev->write_lock);
2507         if (mddev->in_sync != sync_req ||
2508             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2509                 /* have to write it out again */
2510                 spin_unlock_irq(&mddev->write_lock);
2511                 goto repeat;
2512         }
2513         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2514         spin_unlock_irq(&mddev->write_lock);
2515         wake_up(&mddev->sb_wait);
2516         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2517                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2518
2519         rdev_for_each(rdev, mddev) {
2520                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2521                         clear_bit(Blocked, &rdev->flags);
2522
2523                 if (any_badblocks_changed)
2524                         md_ack_all_badblocks(&rdev->badblocks);
2525                 clear_bit(BlockedBadBlocks, &rdev->flags);
2526                 wake_up(&rdev->blocked_wait);
2527         }
2528 }
2529
2530 /* words written to sysfs files may, or may not, be \n terminated.
2531  * We want to accept with case. For this we use cmd_match.
2532  */
2533 static int cmd_match(const char *cmd, const char *str)
2534 {
2535         /* See if cmd, written into a sysfs file, matches
2536          * str.  They must either be the same, or cmd can
2537          * have a trailing newline
2538          */
2539         while (*cmd && *str && *cmd == *str) {
2540                 cmd++;
2541                 str++;
2542         }
2543         if (*cmd == '\n')
2544                 cmd++;
2545         if (*str || *cmd)
2546                 return 0;
2547         return 1;
2548 }
2549
2550 struct rdev_sysfs_entry {
2551         struct attribute attr;
2552         ssize_t (*show)(struct md_rdev *, char *);
2553         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2554 };
2555
2556 static ssize_t
2557 state_show(struct md_rdev *rdev, char *page)
2558 {
2559         char *sep = "";
2560         size_t len = 0;
2561
2562         if (test_bit(Faulty, &rdev->flags) ||
2563             rdev->badblocks.unacked_exist) {
2564                 len+= sprintf(page+len, "%sfaulty",sep);
2565                 sep = ",";
2566         }
2567         if (test_bit(In_sync, &rdev->flags)) {
2568                 len += sprintf(page+len, "%sin_sync",sep);
2569                 sep = ",";
2570         }
2571         if (test_bit(WriteMostly, &rdev->flags)) {
2572                 len += sprintf(page+len, "%swrite_mostly",sep);
2573                 sep = ",";
2574         }
2575         if (test_bit(Blocked, &rdev->flags) ||
2576             (rdev->badblocks.unacked_exist
2577              && !test_bit(Faulty, &rdev->flags))) {
2578                 len += sprintf(page+len, "%sblocked", sep);
2579                 sep = ",";
2580         }
2581         if (!test_bit(Faulty, &rdev->flags) &&
2582             !test_bit(In_sync, &rdev->flags)) {
2583                 len += sprintf(page+len, "%sspare", sep);
2584                 sep = ",";
2585         }
2586         if (test_bit(WriteErrorSeen, &rdev->flags)) {
2587                 len += sprintf(page+len, "%swrite_error", sep);
2588                 sep = ",";
2589         }
2590         if (test_bit(WantReplacement, &rdev->flags)) {
2591                 len += sprintf(page+len, "%swant_replacement", sep);
2592                 sep = ",";
2593         }
2594         if (test_bit(Replacement, &rdev->flags)) {
2595                 len += sprintf(page+len, "%sreplacement", sep);
2596                 sep = ",";
2597         }
2598
2599         return len+sprintf(page+len, "\n");
2600 }
2601
2602 static ssize_t
2603 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2604 {
2605         /* can write
2606          *  faulty  - simulates an error
2607          *  remove  - disconnects the device
2608          *  writemostly - sets write_mostly
2609          *  -writemostly - clears write_mostly
2610          *  blocked - sets the Blocked flags
2611          *  -blocked - clears the Blocked and possibly simulates an error
2612          *  insync - sets Insync providing device isn't active
2613          *  write_error - sets WriteErrorSeen
2614          *  -write_error - clears WriteErrorSeen
2615          */
2616         int err = -EINVAL;
2617         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2618                 md_error(rdev->mddev, rdev);
2619                 if (test_bit(Faulty, &rdev->flags))
2620                         err = 0;
2621                 else
2622                         err = -EBUSY;
2623         } else if (cmd_match(buf, "remove")) {
2624                 if (rdev->raid_disk >= 0)
2625                         err = -EBUSY;
2626                 else {
2627                         struct mddev *mddev = rdev->mddev;
2628                         kick_rdev_from_array(rdev);
2629                         if (mddev->pers)
2630                                 md_update_sb(mddev, 1);
2631                         md_new_event(mddev);
2632                         err = 0;
2633                 }
2634         } else if (cmd_match(buf, "writemostly")) {
2635                 set_bit(WriteMostly, &rdev->flags);
2636                 err = 0;
2637         } else if (cmd_match(buf, "-writemostly")) {
2638                 clear_bit(WriteMostly, &rdev->flags);
2639                 err = 0;
2640         } else if (cmd_match(buf, "blocked")) {
2641                 set_bit(Blocked, &rdev->flags);
2642                 err = 0;
2643         } else if (cmd_match(buf, "-blocked")) {
2644                 if (!test_bit(Faulty, &rdev->flags) &&
2645                     rdev->badblocks.unacked_exist) {
2646                         /* metadata handler doesn't understand badblocks,
2647                          * so we need to fail the device
2648                          */
2649                         md_error(rdev->mddev, rdev);
2650                 }
2651                 clear_bit(Blocked, &rdev->flags);
2652                 clear_bit(BlockedBadBlocks, &rdev->flags);
2653                 wake_up(&rdev->blocked_wait);
2654                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2655                 md_wakeup_thread(rdev->mddev->thread);
2656
2657                 err = 0;
2658         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2659                 set_bit(In_sync, &rdev->flags);
2660                 err = 0;
2661         } else if (cmd_match(buf, "write_error")) {
2662                 set_bit(WriteErrorSeen, &rdev->flags);
2663                 err = 0;
2664         } else if (cmd_match(buf, "-write_error")) {
2665                 clear_bit(WriteErrorSeen, &rdev->flags);
2666                 err = 0;
2667         } else if (cmd_match(buf, "want_replacement")) {
2668                 /* Any non-spare device that is not a replacement can
2669                  * become want_replacement at any time, but we then need to
2670                  * check if recovery is needed.
2671                  */
2672                 if (rdev->raid_disk >= 0 &&
2673                     !test_bit(Replacement, &rdev->flags))
2674                         set_bit(WantReplacement, &rdev->flags);
2675                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2676                 md_wakeup_thread(rdev->mddev->thread);
2677                 err = 0;
2678         } else if (cmd_match(buf, "-want_replacement")) {
2679                 /* Clearing 'want_replacement' is always allowed.
2680                  * Once replacements starts it is too late though.
2681                  */
2682                 err = 0;
2683                 clear_bit(WantReplacement, &rdev->flags);
2684         } else if (cmd_match(buf, "replacement")) {
2685                 /* Can only set a device as a replacement when array has not
2686                  * yet been started.  Once running, replacement is automatic
2687                  * from spares, or by assigning 'slot'.
2688                  */
2689                 if (rdev->mddev->pers)
2690                         err = -EBUSY;
2691                 else {
2692                         set_bit(Replacement, &rdev->flags);
2693                         err = 0;
2694                 }
2695         } else if (cmd_match(buf, "-replacement")) {
2696                 /* Similarly, can only clear Replacement before start */
2697                 if (rdev->mddev->pers)
2698                         err = -EBUSY;
2699                 else {
2700                         clear_bit(Replacement, &rdev->flags);
2701                         err = 0;
2702                 }
2703         }
2704         if (!err)
2705                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2706         return err ? err : len;
2707 }
2708 static struct rdev_sysfs_entry rdev_state =
2709 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2710
2711 static ssize_t
2712 errors_show(struct md_rdev *rdev, char *page)
2713 {
2714         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2715 }
2716
2717 static ssize_t
2718 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2719 {
2720         char *e;
2721         unsigned long n = simple_strtoul(buf, &e, 10);
2722         if (*buf && (*e == 0 || *e == '\n')) {
2723                 atomic_set(&rdev->corrected_errors, n);
2724                 return len;
2725         }
2726         return -EINVAL;
2727 }
2728 static struct rdev_sysfs_entry rdev_errors =
2729 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2730
2731 static ssize_t
2732 slot_show(struct md_rdev *rdev, char *page)
2733 {
2734         if (rdev->raid_disk < 0)
2735                 return sprintf(page, "none\n");
2736         else
2737                 return sprintf(page, "%d\n", rdev->raid_disk);
2738 }
2739
2740 static ssize_t
2741 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2742 {
2743         char *e;
2744         int err;
2745         int slot = simple_strtoul(buf, &e, 10);
2746         if (strncmp(buf, "none", 4)==0)
2747                 slot = -1;
2748         else if (e==buf || (*e && *e!= '\n'))
2749                 return -EINVAL;
2750         if (rdev->mddev->pers && slot == -1) {
2751                 /* Setting 'slot' on an active array requires also
2752                  * updating the 'rd%d' link, and communicating
2753                  * with the personality with ->hot_*_disk.
2754                  * For now we only support removing
2755                  * failed/spare devices.  This normally happens automatically,
2756                  * but not when the metadata is externally managed.
2757                  */
2758                 if (rdev->raid_disk == -1)
2759                         return -EEXIST;
2760                 /* personality does all needed checks */
2761                 if (rdev->mddev->pers->hot_remove_disk == NULL)
2762                         return -EINVAL;
2763                 clear_bit(Blocked, &rdev->flags);
2764                 remove_and_add_spares(rdev->mddev, rdev);
2765                 if (rdev->raid_disk >= 0)
2766                         return -EBUSY;
2767                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2768                 md_wakeup_thread(rdev->mddev->thread);
2769         } else if (rdev->mddev->pers) {
2770                 /* Activating a spare .. or possibly reactivating
2771                  * if we ever get bitmaps working here.
2772                  */
2773
2774                 if (rdev->raid_disk != -1)
2775                         return -EBUSY;
2776
2777                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2778                         return -EBUSY;
2779
2780                 if (rdev->mddev->pers->hot_add_disk == NULL)
2781                         return -EINVAL;
2782
2783                 if (slot >= rdev->mddev->raid_disks &&
2784                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2785                         return -ENOSPC;
2786
2787                 rdev->raid_disk = slot;
2788                 if (test_bit(In_sync, &rdev->flags))
2789                         rdev->saved_raid_disk = slot;
2790                 else
2791                         rdev->saved_raid_disk = -1;
2792                 clear_bit(In_sync, &rdev->flags);
2793                 err = rdev->mddev->pers->
2794                         hot_add_disk(rdev->mddev, rdev);
2795                 if (err) {
2796                         rdev->raid_disk = -1;
2797                         return err;
2798                 } else
2799                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2800                 if (sysfs_link_rdev(rdev->mddev, rdev))
2801                         /* failure here is OK */;
2802                 /* don't wakeup anyone, leave that to userspace. */
2803         } else {
2804                 if (slot >= rdev->mddev->raid_disks &&
2805                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2806                         return -ENOSPC;
2807                 rdev->raid_disk = slot;
2808                 /* assume it is working */
2809                 clear_bit(Faulty, &rdev->flags);
2810                 clear_bit(WriteMostly, &rdev->flags);
2811                 set_bit(In_sync, &rdev->flags);
2812                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2813         }
2814         return len;
2815 }
2816
2817
2818 static struct rdev_sysfs_entry rdev_slot =
2819 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2820
2821 static ssize_t
2822 offset_show(struct md_rdev *rdev, char *page)
2823 {
2824         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2825 }
2826
2827 static ssize_t
2828 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2829 {
2830         unsigned long long offset;
2831         if (kstrtoull(buf, 10, &offset) < 0)
2832                 return -EINVAL;
2833         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2834                 return -EBUSY;
2835         if (rdev->sectors && rdev->mddev->external)
2836                 /* Must set offset before size, so overlap checks
2837                  * can be sane */
2838                 return -EBUSY;
2839         rdev->data_offset = offset;
2840         rdev->new_data_offset = offset;
2841         return len;
2842 }
2843
2844 static struct rdev_sysfs_entry rdev_offset =
2845 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2846
2847 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2848 {
2849         return sprintf(page, "%llu\n",
2850                        (unsigned long long)rdev->new_data_offset);
2851 }
2852
2853 static ssize_t new_offset_store(struct md_rdev *rdev,
2854                                 const char *buf, size_t len)
2855 {
2856         unsigned long long new_offset;
2857         struct mddev *mddev = rdev->mddev;
2858
2859         if (kstrtoull(buf, 10, &new_offset) < 0)
2860                 return -EINVAL;
2861
2862         if (mddev->sync_thread)
2863                 return -EBUSY;
2864         if (new_offset == rdev->data_offset)
2865                 /* reset is always permitted */
2866                 ;
2867         else if (new_offset > rdev->data_offset) {
2868                 /* must not push array size beyond rdev_sectors */
2869                 if (new_offset - rdev->data_offset
2870                     + mddev->dev_sectors > rdev->sectors)
2871                                 return -E2BIG;
2872         }
2873         /* Metadata worries about other space details. */
2874
2875         /* decreasing the offset is inconsistent with a backwards
2876          * reshape.
2877          */
2878         if (new_offset < rdev->data_offset &&
2879             mddev->reshape_backwards)
2880                 return -EINVAL;
2881         /* Increasing offset is inconsistent with forwards
2882          * reshape.  reshape_direction should be set to
2883          * 'backwards' first.
2884          */
2885         if (new_offset > rdev->data_offset &&
2886             !mddev->reshape_backwards)
2887                 return -EINVAL;
2888
2889         if (mddev->pers && mddev->persistent &&
2890             !super_types[mddev->major_version]
2891             .allow_new_offset(rdev, new_offset))
2892                 return -E2BIG;
2893         rdev->new_data_offset = new_offset;
2894         if (new_offset > rdev->data_offset)
2895                 mddev->reshape_backwards = 1;
2896         else if (new_offset < rdev->data_offset)
2897                 mddev->reshape_backwards = 0;
2898
2899         return len;
2900 }
2901 static struct rdev_sysfs_entry rdev_new_offset =
2902 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2903
2904 static ssize_t
2905 rdev_size_show(struct md_rdev *rdev, char *page)
2906 {
2907         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2908 }
2909
2910 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2911 {
2912         /* check if two start/length pairs overlap */
2913         if (s1+l1 <= s2)
2914                 return 0;
2915         if (s2+l2 <= s1)
2916                 return 0;
2917         return 1;
2918 }
2919
2920 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2921 {
2922         unsigned long long blocks;
2923         sector_t new;
2924
2925         if (kstrtoull(buf, 10, &blocks) < 0)
2926                 return -EINVAL;
2927
2928         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2929                 return -EINVAL; /* sector conversion overflow */
2930
2931         new = blocks * 2;
2932         if (new != blocks * 2)
2933                 return -EINVAL; /* unsigned long long to sector_t overflow */
2934
2935         *sectors = new;
2936         return 0;
2937 }
2938
2939 static ssize_t
2940 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2941 {
2942         struct mddev *my_mddev = rdev->mddev;
2943         sector_t oldsectors = rdev->sectors;
2944         sector_t sectors;
2945
2946         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2947                 return -EINVAL;
2948         if (rdev->data_offset != rdev->new_data_offset)
2949                 return -EINVAL; /* too confusing */
2950         if (my_mddev->pers && rdev->raid_disk >= 0) {
2951                 if (my_mddev->persistent) {
2952                         sectors = super_types[my_mddev->major_version].
2953                                 rdev_size_change(rdev, sectors);
2954                         if (!sectors)
2955                                 return -EBUSY;
2956                 } else if (!sectors)
2957                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2958                                 rdev->data_offset;
2959                 if (!my_mddev->pers->resize)
2960                         /* Cannot change size for RAID0 or Linear etc */
2961                         return -EINVAL;
2962         }
2963         if (sectors < my_mddev->dev_sectors)
2964                 return -EINVAL; /* component must fit device */
2965
2966         rdev->sectors = sectors;
2967         if (sectors > oldsectors && my_mddev->external) {
2968                 /* need to check that all other rdevs with the same ->bdev
2969                  * do not overlap.  We need to unlock the mddev to avoid
2970                  * a deadlock.  We have already changed rdev->sectors, and if
2971                  * we have to change it back, we will have the lock again.
2972                  */
2973                 struct mddev *mddev;
2974                 int overlap = 0;
2975                 struct list_head *tmp;
2976
2977                 mddev_unlock(my_mddev);
2978                 for_each_mddev(mddev, tmp) {
2979                         struct md_rdev *rdev2;
2980
2981                         mddev_lock(mddev);
2982                         rdev_for_each(rdev2, mddev)
2983                                 if (rdev->bdev == rdev2->bdev &&
2984                                     rdev != rdev2 &&
2985                                     overlaps(rdev->data_offset, rdev->sectors,
2986                                              rdev2->data_offset,
2987                                              rdev2->sectors)) {
2988                                         overlap = 1;
2989                                         break;
2990                                 }
2991                         mddev_unlock(mddev);
2992                         if (overlap) {
2993                                 mddev_put(mddev);
2994                                 break;
2995                         }
2996                 }
2997                 mddev_lock(my_mddev);
2998                 if (overlap) {
2999                         /* Someone else could have slipped in a size
3000                          * change here, but doing so is just silly.
3001                          * We put oldsectors back because we *know* it is
3002                          * safe, and trust userspace not to race with
3003                          * itself
3004                          */
3005                         rdev->sectors = oldsectors;
3006                         return -EBUSY;
3007                 }
3008         }
3009         return len;
3010 }
3011
3012 static struct rdev_sysfs_entry rdev_size =
3013 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3014
3015
3016 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3017 {
3018         unsigned long long recovery_start = rdev->recovery_offset;
3019
3020         if (test_bit(In_sync, &rdev->flags) ||
3021             recovery_start == MaxSector)
3022                 return sprintf(page, "none\n");
3023
3024         return sprintf(page, "%llu\n", recovery_start);
3025 }
3026
3027 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3028 {
3029         unsigned long long recovery_start;
3030
3031         if (cmd_match(buf, "none"))
3032                 recovery_start = MaxSector;
3033         else if (kstrtoull(buf, 10, &recovery_start))
3034                 return -EINVAL;
3035
3036         if (rdev->mddev->pers &&
3037             rdev->raid_disk >= 0)
3038                 return -EBUSY;
3039
3040         rdev->recovery_offset = recovery_start;
3041         if (recovery_start == MaxSector)
3042                 set_bit(In_sync, &rdev->flags);
3043         else
3044                 clear_bit(In_sync, &rdev->flags);
3045         return len;
3046 }
3047
3048 static struct rdev_sysfs_entry rdev_recovery_start =
3049 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3050
3051
3052 static ssize_t
3053 badblocks_show(struct badblocks *bb, char *page, int unack);
3054 static ssize_t
3055 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3056
3057 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3058 {
3059         return badblocks_show(&rdev->badblocks, page, 0);
3060 }
3061 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3062 {
3063         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3064         /* Maybe that ack was all we needed */
3065         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3066                 wake_up(&rdev->blocked_wait);
3067         return rv;
3068 }
3069 static struct rdev_sysfs_entry rdev_bad_blocks =
3070 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3071
3072
3073 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3074 {
3075         return badblocks_show(&rdev->badblocks, page, 1);
3076 }
3077 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3078 {
3079         return badblocks_store(&rdev->badblocks, page, len, 1);
3080 }
3081 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3082 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3083
3084 static struct attribute *rdev_default_attrs[] = {
3085         &rdev_state.attr,
3086         &rdev_errors.attr,
3087         &rdev_slot.attr,
3088         &rdev_offset.attr,
3089         &rdev_new_offset.attr,
3090         &rdev_size.attr,
3091         &rdev_recovery_start.attr,
3092         &rdev_bad_blocks.attr,
3093         &rdev_unack_bad_blocks.attr,
3094         NULL,
3095 };
3096 static ssize_t
3097 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3098 {
3099         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3100         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3101         struct mddev *mddev = rdev->mddev;
3102         ssize_t rv;
3103
3104         if (!entry->show)
3105                 return -EIO;
3106
3107         rv = mddev ? mddev_lock(mddev) : -EBUSY;
3108         if (!rv) {
3109                 if (rdev->mddev == NULL)
3110                         rv = -EBUSY;
3111                 else
3112                         rv = entry->show(rdev, page);
3113                 mddev_unlock(mddev);
3114         }
3115         return rv;
3116 }
3117
3118 static ssize_t
3119 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3120               const char *page, size_t length)
3121 {
3122         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3123         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3124         ssize_t rv;
3125         struct mddev *mddev = rdev->mddev;
3126
3127         if (!entry->store)
3128                 return -EIO;
3129         if (!capable(CAP_SYS_ADMIN))
3130                 return -EACCES;
3131         rv = mddev ? mddev_lock(mddev): -EBUSY;
3132         if (!rv) {
3133                 if (rdev->mddev == NULL)
3134                         rv = -EBUSY;
3135                 else
3136                         rv = entry->store(rdev, page, length);
3137                 mddev_unlock(mddev);
3138         }
3139         return rv;
3140 }
3141
3142 static void rdev_free(struct kobject *ko)
3143 {
3144         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3145         kfree(rdev);
3146 }
3147 static const struct sysfs_ops rdev_sysfs_ops = {
3148         .show           = rdev_attr_show,
3149         .store          = rdev_attr_store,
3150 };
3151 static struct kobj_type rdev_ktype = {
3152         .release        = rdev_free,
3153         .sysfs_ops      = &rdev_sysfs_ops,
3154         .default_attrs  = rdev_default_attrs,
3155 };
3156
3157 int md_rdev_init(struct md_rdev *rdev)
3158 {
3159         rdev->desc_nr = -1;
3160         rdev->saved_raid_disk = -1;
3161         rdev->raid_disk = -1;
3162         rdev->flags = 0;
3163         rdev->data_offset = 0;
3164         rdev->new_data_offset = 0;
3165         rdev->sb_events = 0;
3166         rdev->last_read_error.tv_sec  = 0;
3167         rdev->last_read_error.tv_nsec = 0;
3168         rdev->sb_loaded = 0;
3169         rdev->bb_page = NULL;
3170         atomic_set(&rdev->nr_pending, 0);
3171         atomic_set(&rdev->read_errors, 0);
3172         atomic_set(&rdev->corrected_errors, 0);
3173
3174         INIT_LIST_HEAD(&rdev->same_set);
3175         init_waitqueue_head(&rdev->blocked_wait);
3176
3177         /* Add space to store bad block list.
3178          * This reserves the space even on arrays where it cannot
3179          * be used - I wonder if that matters
3180          */
3181         rdev->badblocks.count = 0;
3182         rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3183         rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3184         seqlock_init(&rdev->badblocks.lock);
3185         if (rdev->badblocks.page == NULL)
3186                 return -ENOMEM;
3187
3188         return 0;
3189 }
3190 EXPORT_SYMBOL_GPL(md_rdev_init);
3191 /*
3192  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3193  *
3194  * mark the device faulty if:
3195  *
3196  *   - the device is nonexistent (zero size)
3197  *   - the device has no valid superblock
3198  *
3199  * a faulty rdev _never_ has rdev->sb set.
3200  */
3201 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3202 {
3203         char b[BDEVNAME_SIZE];
3204         int err;
3205         struct md_rdev *rdev;
3206         sector_t size;
3207
3208         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3209         if (!rdev) {
3210                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3211                 return ERR_PTR(-ENOMEM);
3212         }
3213
3214         err = md_rdev_init(rdev);
3215         if (err)
3216                 goto abort_free;
3217         err = alloc_disk_sb(rdev);
3218         if (err)
3219                 goto abort_free;
3220
3221         err = lock_rdev(rdev, newdev, super_format == -2);
3222         if (err)
3223                 goto abort_free;
3224
3225         kobject_init(&rdev->kobj, &rdev_ktype);
3226
3227         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3228         if (!size) {
3229                 printk(KERN_WARNING
3230                         "md: %s has zero or unknown size, marking faulty!\n",
3231                         bdevname(rdev->bdev,b));
3232                 err = -EINVAL;
3233                 goto abort_free;
3234         }
3235
3236         if (super_format >= 0) {
3237                 err = super_types[super_format].
3238                         load_super(rdev, NULL, super_minor);
3239                 if (err == -EINVAL) {
3240                         printk(KERN_WARNING
3241                                 "md: %s does not have a valid v%d.%d "
3242                                "superblock, not importing!\n",
3243                                 bdevname(rdev->bdev,b),
3244                                super_format, super_minor);
3245                         goto abort_free;
3246                 }
3247                 if (err < 0) {
3248                         printk(KERN_WARNING
3249                                 "md: could not read %s's sb, not importing!\n",
3250                                 bdevname(rdev->bdev,b));
3251                         goto abort_free;
3252                 }
3253         }
3254
3255         return rdev;
3256
3257 abort_free:
3258         if (rdev->bdev)
3259                 unlock_rdev(rdev);
3260         md_rdev_clear(rdev);
3261         kfree(rdev);
3262         return ERR_PTR(err);
3263 }
3264
3265 /*
3266  * Check a full RAID array for plausibility
3267  */
3268
3269
3270 static void analyze_sbs(struct mddev * mddev)
3271 {
3272         int i;
3273         struct md_rdev *rdev, *freshest, *tmp;
3274         char b[BDEVNAME_SIZE];
3275
3276         freshest = NULL;
3277         rdev_for_each_safe(rdev, tmp, mddev)
3278                 switch (super_types[mddev->major_version].
3279                         load_super(rdev, freshest, mddev->minor_version)) {
3280                 case 1:
3281                         freshest = rdev;
3282                         break;
3283                 case 0:
3284                         break;
3285                 default:
3286                         printk( KERN_ERR \
3287                                 "md: fatal superblock inconsistency in %s"
3288                                 " -- removing from array\n",
3289                                 bdevname(rdev->bdev,b));
3290                         kick_rdev_from_array(rdev);
3291                 }
3292
3293
3294         super_types[mddev->major_version].
3295                 validate_super(mddev, freshest);
3296
3297         i = 0;
3298         rdev_for_each_safe(rdev, tmp, mddev) {
3299                 if (mddev->max_disks &&
3300                     (rdev->desc_nr >= mddev->max_disks ||
3301                      i > mddev->max_disks)) {
3302                         printk(KERN_WARNING
3303                                "md: %s: %s: only %d devices permitted\n",
3304                                mdname(mddev), bdevname(rdev->bdev, b),
3305                                mddev->max_disks);
3306                         kick_rdev_from_array(rdev);
3307                         continue;
3308                 }
3309                 if (rdev != freshest)
3310                         if (super_types[mddev->major_version].
3311                             validate_super(mddev, rdev)) {
3312                                 printk(KERN_WARNING "md: kicking non-fresh %s"
3313                                         " from array!\n",
3314                                         bdevname(rdev->bdev,b));
3315                                 kick_rdev_from_array(rdev);
3316                                 continue;
3317                         }
3318                 if (mddev->level == LEVEL_MULTIPATH) {
3319                         rdev->desc_nr = i++;
3320                         rdev->raid_disk = rdev->desc_nr;
3321                         set_bit(In_sync, &rdev->flags);
3322                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3323                         rdev->raid_disk = -1;
3324                         clear_bit(In_sync, &rdev->flags);
3325                 }
3326         }
3327 }
3328
3329 /* Read a fixed-point number.
3330  * Numbers in sysfs attributes should be in "standard" units where
3331  * possible, so time should be in seconds.
3332  * However we internally use a a much smaller unit such as
3333  * milliseconds or jiffies.
3334  * This function takes a decimal number with a possible fractional
3335  * component, and produces an integer which is the result of
3336  * multiplying that number by 10^'scale'.
3337  * all without any floating-point arithmetic.
3338  */
3339 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3340 {
3341         unsigned long result = 0;
3342         long decimals = -1;
3343         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3344                 if (*cp == '.')
3345                         decimals = 0;
3346                 else if (decimals < scale) {
3347                         unsigned int value;
3348                         value = *cp - '0';
3349                         result = result * 10 + value;
3350                         if (decimals >= 0)
3351                                 decimals++;
3352                 }
3353                 cp++;
3354         }
3355         if (*cp == '\n')
3356                 cp++;
3357         if (*cp)
3358                 return -EINVAL;
3359         if (decimals < 0)
3360                 decimals = 0;
3361         while (decimals < scale) {
3362                 result *= 10;
3363                 decimals ++;
3364         }
3365         *res = result;
3366         return 0;
3367 }
3368
3369
3370 static void md_safemode_timeout(unsigned long data);
3371
3372 static ssize_t
3373 safe_delay_show(struct mddev *mddev, char *page)
3374 {
3375         int msec = (mddev->safemode_delay*1000)/HZ;
3376         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3377 }
3378 static ssize_t
3379 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3380 {
3381         unsigned long msec;
3382
3383         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3384                 return -EINVAL;
3385         if (msec == 0)
3386                 mddev->safemode_delay = 0;
3387         else {
3388                 unsigned long old_delay = mddev->safemode_delay;
3389                 mddev->safemode_delay = (msec*HZ)/1000;
3390                 if (mddev->safemode_delay == 0)
3391                         mddev->safemode_delay = 1;
3392                 if (mddev->safemode_delay < old_delay || old_delay == 0)
3393                         md_safemode_timeout((unsigned long)mddev);
3394         }
3395         return len;
3396 }
3397 static struct md_sysfs_entry md_safe_delay =
3398 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3399
3400 static ssize_t
3401 level_show(struct mddev *mddev, char *page)
3402 {
3403         struct md_personality *p = mddev->pers;
3404         if (p)
3405                 return sprintf(page, "%s\n", p->name);
3406         else if (mddev->clevel[0])
3407                 return sprintf(page, "%s\n", mddev->clevel);
3408         else if (mddev->level != LEVEL_NONE)
3409                 return sprintf(page, "%d\n", mddev->level);
3410         else
3411                 return 0;
3412 }
3413
3414 static ssize_t
3415 level_store(struct mddev *mddev, const char *buf, size_t len)
3416 {
3417         char clevel[16];
3418         ssize_t rv = len;
3419         struct md_personality *pers;
3420         long level;
3421         void *priv;
3422         struct md_rdev *rdev;
3423
3424         if (mddev->pers == NULL) {
3425                 if (len == 0)
3426                         return 0;
3427                 if (len >= sizeof(mddev->clevel))
3428                         return -ENOSPC;
3429                 strncpy(mddev->clevel, buf, len);
3430                 if (mddev->clevel[len-1] == '\n')
3431                         len--;
3432                 mddev->clevel[len] = 0;
3433                 mddev->level = LEVEL_NONE;
3434                 return rv;
3435         }
3436
3437         /* request to change the personality.  Need to ensure:
3438          *  - array is not engaged in resync/recovery/reshape
3439          *  - old personality can be suspended
3440          *  - new personality will access other array.
3441          */
3442
3443         if (mddev->sync_thread ||
3444             mddev->reshape_position != MaxSector ||
3445             mddev->sysfs_active)
3446                 return -EBUSY;
3447
3448         if (!mddev->pers->quiesce) {
3449                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3450                        mdname(mddev), mddev->pers->name);
3451                 return -EINVAL;
3452         }
3453
3454         /* Now find the new personality */
3455         if (len == 0 || len >= sizeof(clevel))
3456                 return -EINVAL;
3457         strncpy(clevel, buf, len);
3458         if (clevel[len-1] == '\n')
3459                 len--;
3460         clevel[len] = 0;
3461         if (kstrtol(clevel, 10, &level))
3462                 level = LEVEL_NONE;
3463
3464         if (request_module("md-%s", clevel) != 0)
3465                 request_module("md-level-%s", clevel);
3466         spin_lock(&pers_lock);
3467         pers = find_pers(level, clevel);
3468         if (!pers || !try_module_get(pers->owner)) {
3469                 spin_unlock(&pers_lock);
3470                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3471                 return -EINVAL;
3472         }
3473         spin_unlock(&pers_lock);
3474
3475         if (pers == mddev->pers) {
3476                 /* Nothing to do! */
3477                 module_put(pers->owner);
3478                 return rv;
3479         }
3480         if (!pers->takeover) {
3481                 module_put(pers->owner);
3482                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3483                        mdname(mddev), clevel);
3484                 return -EINVAL;
3485         }
3486
3487         rdev_for_each(rdev, mddev)
3488                 rdev->new_raid_disk = rdev->raid_disk;
3489
3490         /* ->takeover must set new_* and/or delta_disks
3491          * if it succeeds, and may set them when it fails.
3492          */
3493         priv = pers->takeover(mddev);
3494         if (IS_ERR(priv)) {
3495                 mddev->new_level = mddev->level;
3496                 mddev->new_layout = mddev->layout;
3497                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3498                 mddev->raid_disks -= mddev->delta_disks;
3499                 mddev->delta_disks = 0;
3500                 mddev->reshape_backwards = 0;
3501                 module_put(pers->owner);
3502                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3503                        mdname(mddev), clevel);
3504                 return PTR_ERR(priv);
3505         }
3506
3507         /* Looks like we have a winner */
3508         mddev_suspend(mddev);
3509         mddev->pers->stop(mddev);
3510
3511         if (mddev->pers->sync_request == NULL &&
3512             pers->sync_request != NULL) {
3513                 /* need to add the md_redundancy_group */
3514                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3515                         printk(KERN_WARNING
3516                                "md: cannot register extra attributes for %s\n",
3517                                mdname(mddev));
3518                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3519         }
3520         if (mddev->pers->sync_request != NULL &&
3521             pers->sync_request == NULL) {
3522                 /* need to remove the md_redundancy_group */
3523                 if (mddev->to_remove == NULL)
3524                         mddev->to_remove = &md_redundancy_group;
3525         }
3526
3527         if (mddev->pers->sync_request == NULL &&
3528             mddev->external) {
3529                 /* We are converting from a no-redundancy array
3530                  * to a redundancy array and metadata is managed
3531                  * externally so we need to be sure that writes
3532                  * won't block due to a need to transition
3533                  *      clean->dirty
3534                  * until external management is started.
3535                  */
3536                 mddev->in_sync = 0;
3537                 mddev->safemode_delay = 0;
3538                 mddev->safemode = 0;
3539         }
3540
3541         rdev_for_each(rdev, mddev) {
3542                 if (rdev->raid_disk < 0)
3543                         continue;
3544                 if (rdev->new_raid_disk >= mddev->raid_disks)
3545                         rdev->new_raid_disk = -1;
3546                 if (rdev->new_raid_disk == rdev->raid_disk)
3547                         continue;
3548                 sysfs_unlink_rdev(mddev, rdev);
3549         }
3550         rdev_for_each(rdev, mddev) {
3551                 if (rdev->raid_disk < 0)
3552                         continue;
3553                 if (rdev->new_raid_disk == rdev->raid_disk)
3554                         continue;
3555                 rdev->raid_disk = rdev->new_raid_disk;
3556                 if (rdev->raid_disk < 0)
3557                         clear_bit(In_sync, &rdev->flags);
3558                 else {
3559                         if (sysfs_link_rdev(mddev, rdev))
3560                                 printk(KERN_WARNING "md: cannot register rd%d"
3561                                        " for %s after level change\n",
3562                                        rdev->raid_disk, mdname(mddev));
3563                 }
3564         }
3565
3566         module_put(mddev->pers->owner);
3567         mddev->pers = pers;
3568         mddev->private = priv;
3569         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3570         mddev->level = mddev->new_level;
3571         mddev->layout = mddev->new_layout;
3572         mddev->chunk_sectors = mddev->new_chunk_sectors;
3573         mddev->delta_disks = 0;
3574         mddev->reshape_backwards = 0;
3575         mddev->degraded = 0;
3576         if (mddev->pers->sync_request == NULL) {
3577                 /* this is now an array without redundancy, so
3578                  * it must always be in_sync
3579                  */
3580                 mddev->in_sync = 1;
3581                 del_timer_sync(&mddev->safemode_timer);
3582         }
3583         pers->run(mddev);
3584         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3585         mddev_resume(mddev);
3586         sysfs_notify(&mddev->kobj, NULL, "level");
3587         md_new_event(mddev);
3588         return rv;
3589 }
3590
3591 static struct md_sysfs_entry md_level =
3592 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3593
3594
3595 static ssize_t
3596 layout_show(struct mddev *mddev, char *page)
3597 {
3598         /* just a number, not meaningful for all levels */
3599         if (mddev->reshape_position != MaxSector &&
3600             mddev->layout != mddev->new_layout)
3601                 return sprintf(page, "%d (%d)\n",
3602                                mddev->new_layout, mddev->layout);
3603         return sprintf(page, "%d\n", mddev->layout);
3604 }
3605
3606 static ssize_t
3607 layout_store(struct mddev *mddev, const char *buf, size_t len)
3608 {
3609         char *e;
3610         unsigned long n = simple_strtoul(buf, &e, 10);
3611
3612         if (!*buf || (*e && *e != '\n'))
3613                 return -EINVAL;
3614
3615         if (mddev->pers) {
3616                 int err;
3617                 if (mddev->pers->check_reshape == NULL)
3618                         return -EBUSY;
3619                 mddev->new_layout = n;
3620                 err = mddev->pers->check_reshape(mddev);
3621                 if (err) {
3622                         mddev->new_layout = mddev->layout;
3623                         return err;
3624                 }
3625         } else {
3626                 mddev->new_layout = n;
3627                 if (mddev->reshape_position == MaxSector)
3628                         mddev->layout = n;
3629         }
3630         return len;
3631 }
3632 static struct md_sysfs_entry md_layout =
3633 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3634
3635
3636 static ssize_t
3637 raid_disks_show(struct mddev *mddev, char *page)
3638 {
3639         if (mddev->raid_disks == 0)
3640                 return 0;
3641         if (mddev->reshape_position != MaxSector &&
3642             mddev->delta_disks != 0)
3643                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3644                                mddev->raid_disks - mddev->delta_disks);
3645         return sprintf(page, "%d\n", mddev->raid_disks);
3646 }
3647
3648 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3649
3650 static ssize_t
3651 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3652 {
3653         char *e;
3654         int rv = 0;
3655         unsigned long n = simple_strtoul(buf, &e, 10);
3656
3657         if (!*buf || (*e && *e != '\n'))
3658                 return -EINVAL;
3659
3660         if (mddev->pers)
3661                 rv = update_raid_disks(mddev, n);
3662         else if (mddev->reshape_position != MaxSector) {
3663                 struct md_rdev *rdev;
3664                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3665
3666                 rdev_for_each(rdev, mddev) {
3667                         if (olddisks < n &&
3668                             rdev->data_offset < rdev->new_data_offset)
3669                                 return -EINVAL;
3670                         if (olddisks > n &&
3671                             rdev->data_offset > rdev->new_data_offset)
3672                                 return -EINVAL;
3673                 }
3674                 mddev->delta_disks = n - olddisks;
3675                 mddev->raid_disks = n;
3676                 mddev->reshape_backwards = (mddev->delta_disks < 0);
3677         } else
3678                 mddev->raid_disks = n;
3679         return rv ? rv : len;
3680 }
3681 static struct md_sysfs_entry md_raid_disks =
3682 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3683
3684 static ssize_t
3685 chunk_size_show(struct mddev *mddev, char *page)
3686 {
3687         if (mddev->reshape_position != MaxSector &&
3688             mddev->chunk_sectors != mddev->new_chunk_sectors)
3689                 return sprintf(page, "%d (%d)\n",
3690                                mddev->new_chunk_sectors << 9,
3691                                mddev->chunk_sectors << 9);
3692         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3693 }
3694
3695 static ssize_t
3696 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3697 {
3698         char *e;
3699         unsigned long n = simple_strtoul(buf, &e, 10);
3700
3701         if (!*buf || (*e && *e != '\n'))
3702                 return -EINVAL;
3703
3704         if (mddev->pers) {
3705                 int err;
3706                 if (mddev->pers->check_reshape == NULL)
3707                         return -EBUSY;
3708                 mddev->new_chunk_sectors = n >> 9;
3709                 err = mddev->pers->check_reshape(mddev);
3710                 if (err) {
3711                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3712                         return err;
3713                 }
3714         } else {
3715                 mddev->new_chunk_sectors = n >> 9;
3716                 if (mddev->reshape_position == MaxSector)
3717                         mddev->chunk_sectors = n >> 9;
3718         }
3719         return len;
3720 }
3721 static struct md_sysfs_entry md_chunk_size =
3722 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3723
3724 static ssize_t
3725 resync_start_show(struct mddev *mddev, char *page)
3726 {
3727         if (mddev->recovery_cp == MaxSector)
3728                 return sprintf(page, "none\n");
3729         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3730 }
3731
3732 static ssize_t
3733 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3734 {
3735         char *e;
3736         unsigned long long n = simple_strtoull(buf, &e, 10);
3737
3738         if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3739                 return -EBUSY;
3740         if (cmd_match(buf, "none"))
3741                 n = MaxSector;
3742         else if (!*buf || (*e && *e != '\n'))
3743                 return -EINVAL;
3744
3745         mddev->recovery_cp = n;
3746         if (mddev->pers)
3747                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3748         return len;
3749 }
3750 static struct md_sysfs_entry md_resync_start =
3751 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3752
3753 /*
3754  * The array state can be:
3755  *
3756  * clear
3757  *     No devices, no size, no level
3758  *     Equivalent to STOP_ARRAY ioctl
3759  * inactive
3760  *     May have some settings, but array is not active
3761  *        all IO results in error
3762  *     When written, doesn't tear down array, but just stops it
3763  * suspended (not supported yet)
3764  *     All IO requests will block. The array can be reconfigured.
3765  *     Writing this, if accepted, will block until array is quiescent
3766  * readonly
3767  *     no resync can happen.  no superblocks get written.
3768  *     write requests fail
3769  * read-auto
3770  *     like readonly, but behaves like 'clean' on a write request.
3771  *
3772  * clean - no pending writes, but otherwise active.
3773  *     When written to inactive array, starts without resync
3774  *     If a write request arrives then
3775  *       if metadata is known, mark 'dirty' and switch to 'active'.
3776  *       if not known, block and switch to write-pending
3777  *     If written to an active array that has pending writes, then fails.
3778  * active
3779  *     fully active: IO and resync can be happening.
3780  *     When written to inactive array, starts with resync
3781  *
3782  * write-pending
3783  *     clean, but writes are blocked waiting for 'active' to be written.
3784  *
3785  * active-idle
3786  *     like active, but no writes have been seen for a while (100msec).
3787  *
3788  */
3789 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3790                    write_pending, active_idle, bad_word};
3791 static char *array_states[] = {
3792         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3793         "write-pending", "active-idle", NULL };
3794
3795 static int match_word(const char *word, char **list)
3796 {
3797         int n;
3798         for (n=0; list[n]; n++)
3799                 if (cmd_match(word, list[n]))
3800                         break;
3801         return n;
3802 }
3803
3804 static ssize_t
3805 array_state_show(struct mddev *mddev, char *page)
3806 {
3807         enum array_state st = inactive;
3808
3809         if (mddev->pers)
3810                 switch(mddev->ro) {
3811                 case 1:
3812                         st = readonly;
3813                         break;
3814                 case 2:
3815                         st = read_auto;
3816                         break;
3817                 case 0:
3818                         if (mddev->in_sync)
3819                                 st = clean;
3820                         else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3821                                 st = write_pending;
3822                         else if (mddev->safemode)
3823                                 st = active_idle;
3824                         else
3825                                 st = active;
3826                 }
3827         else {
3828                 if (list_empty(&mddev->disks) &&
3829                     mddev->raid_disks == 0 &&
3830                     mddev->dev_sectors == 0)
3831                         st = clear;
3832                 else
3833                         st = inactive;
3834         }
3835         return sprintf(page, "%s\n", array_states[st]);
3836 }
3837
3838 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3839 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3840 static int do_md_run(struct mddev * mddev);
3841 static int restart_array(struct mddev *mddev);
3842
3843 static ssize_t
3844 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3845 {
3846         int err = -EINVAL;
3847         enum array_state st = match_word(buf, array_states);
3848         switch(st) {
3849         case bad_word:
3850                 break;
3851         case clear:
3852                 /* stopping an active array */
3853                 err = do_md_stop(mddev, 0, NULL);
3854                 break;
3855         case inactive:
3856                 /* stopping an active array */
3857                 if (mddev->pers)
3858                         err = do_md_stop(mddev, 2, NULL);
3859                 else
3860                         err = 0; /* already inactive */
3861                 break;
3862         case suspended:
3863                 break; /* not supported yet */
3864         case readonly:
3865                 if (mddev->pers)
3866                         err = md_set_readonly(mddev, NULL);
3867                 else {
3868                         mddev->ro = 1;
3869                         set_disk_ro(mddev->gendisk, 1);
3870                         err = do_md_run(mddev);
3871                 }
3872                 break;
3873         case read_auto:
3874                 if (mddev->pers) {
3875                         if (mddev->ro == 0)
3876                                 err = md_set_readonly(mddev, NULL);
3877                         else if (mddev->ro == 1)
3878                                 err = restart_array(mddev);
3879                         if (err == 0) {
3880                                 mddev->ro = 2;
3881                                 set_disk_ro(mddev->gendisk, 0);
3882                         }
3883                 } else {
3884                         mddev->ro = 2;
3885                         err = do_md_run(mddev);
3886                 }
3887                 break;
3888         case clean:
3889                 if (mddev->pers) {
3890                         restart_array(mddev);
3891                         spin_lock_irq(&mddev->write_lock);
3892                         if (atomic_read(&mddev->writes_pending) == 0) {
3893                                 if (mddev->in_sync == 0) {
3894                                         mddev->in_sync = 1;
3895                                         if (mddev->safemode == 1)
3896                                                 mddev->safemode = 0;
3897                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3898                                 }
3899                                 err = 0;
3900                         } else
3901                                 err = -EBUSY;
3902                         spin_unlock_irq(&mddev->write_lock);
3903                 } else
3904                         err = -EINVAL;
3905                 break;
3906         case active:
3907                 if (mddev->pers) {
3908                         restart_array(mddev);
3909                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3910                         wake_up(&mddev->sb_wait);
3911                         err = 0;
3912                 } else {
3913                         mddev->ro = 0;
3914                         set_disk_ro(mddev->gendisk, 0);
3915                         err = do_md_run(mddev);
3916                 }
3917                 break;
3918         case write_pending:
3919         case active_idle:
3920                 /* these cannot be set */
3921                 break;
3922         }
3923         if (err)
3924                 return err;
3925         else {
3926                 if (mddev->hold_active == UNTIL_IOCTL)
3927                         mddev->hold_active = 0;
3928                 sysfs_notify_dirent_safe(mddev->sysfs_state);
3929                 return len;
3930         }
3931 }
3932 static struct md_sysfs_entry md_array_state =
3933 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3934
3935 static ssize_t
3936 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3937         return sprintf(page, "%d\n",
3938                        atomic_read(&mddev->max_corr_read_errors));
3939 }
3940
3941 static ssize_t
3942 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3943 {
3944         char *e;
3945         unsigned long n = simple_strtoul(buf, &e, 10);
3946
3947         if (*buf && (*e == 0 || *e == '\n')) {
3948                 atomic_set(&mddev->max_corr_read_errors, n);
3949                 return len;
3950         }
3951         return -EINVAL;
3952 }
3953
3954 static struct md_sysfs_entry max_corr_read_errors =
3955 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3956         max_corrected_read_errors_store);
3957
3958 static ssize_t
3959 null_show(struct mddev *mddev, char *page)
3960 {
3961         return -EINVAL;
3962 }
3963
3964 static ssize_t
3965 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3966 {
3967         /* buf must be %d:%d\n? giving major and minor numbers */
3968         /* The new device is added to the array.
3969          * If the array has a persistent superblock, we read the
3970          * superblock to initialise info and check validity.
3971          * Otherwise, only checking done is that in bind_rdev_to_array,
3972          * which mainly checks size.
3973          */
3974         char *e;
3975         int major = simple_strtoul(buf, &e, 10);
3976         int minor;
3977         dev_t dev;
3978         struct md_rdev *rdev;
3979         int err;
3980
3981         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3982                 return -EINVAL;
3983         minor = simple_strtoul(e+1, &e, 10);
3984         if (*e && *e != '\n')
3985                 return -EINVAL;
3986         dev = MKDEV(major, minor);
3987         if (major != MAJOR(dev) ||
3988             minor != MINOR(dev))
3989                 return -EOVERFLOW;
3990
3991
3992         if (mddev->persistent) {
3993                 rdev = md_import_device(dev, mddev->major_version,
3994                                         mddev->minor_version);
3995                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3996                         struct md_rdev *rdev0
3997                                 = list_entry(mddev->disks.next,
3998                                              struct md_rdev, same_set);
3999                         err = super_types[mddev->major_version]
4000                                 .load_super(rdev, rdev0, mddev->minor_version);
4001                         if (err < 0)
4002                                 goto out;
4003                 }
4004         } else if (mddev->external)
4005                 rdev = md_import_device(dev, -2, -1);
4006         else
4007                 rdev = md_import_device(dev, -1, -1);
4008
4009         if (IS_ERR(rdev))
4010                 return PTR_ERR(rdev);
4011         err = bind_rdev_to_array(rdev, mddev);
4012  out:
4013         if (err)
4014                 export_rdev(rdev);
4015         return err ? err : len;
4016 }
4017
4018 static struct md_sysfs_entry md_new_device =
4019 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4020
4021 static ssize_t
4022 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4023 {
4024         char *end;
4025         unsigned long chunk, end_chunk;
4026
4027         if (!mddev->bitmap)
4028                 goto out;
4029         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4030         while (*buf) {
4031                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4032                 if (buf == end) break;
4033                 if (*end == '-') { /* range */
4034                         buf = end + 1;
4035                         end_chunk = simple_strtoul(buf, &end, 0);
4036                         if (buf == end) break;
4037                 }
4038                 if (*end && !isspace(*end)) break;
4039                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4040                 buf = skip_spaces(end);
4041         }
4042         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4043 out:
4044         return len;
4045 }
4046
4047 static struct md_sysfs_entry md_bitmap =
4048 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4049
4050 static ssize_t
4051 size_show(struct mddev *mddev, char *page)
4052 {
4053         return sprintf(page, "%llu\n",
4054                 (unsigned long long)mddev->dev_sectors / 2);
4055 }
4056
4057 static int update_size(struct mddev *mddev, sector_t num_sectors);
4058
4059 static ssize_t
4060 size_store(struct mddev *mddev, const char *buf, size_t len)
4061 {
4062         /* If array is inactive, we can reduce the component size, but
4063          * not increase it (except from 0).
4064          * If array is active, we can try an on-line resize
4065          */
4066         sector_t sectors;
4067         int err = strict_blocks_to_sectors(buf, &sectors);
4068
4069         if (err < 0)
4070                 return err;
4071         if (mddev->pers) {
4072                 err = update_size(mddev, sectors);
4073                 md_update_sb(mddev, 1);
4074         } else {
4075                 if (mddev->dev_sectors == 0 ||
4076                     mddev->dev_sectors > sectors)
4077                         mddev->dev_sectors = sectors;
4078                 else
4079                         err = -ENOSPC;
4080         }
4081         return err ? err : len;
4082 }
4083
4084 static struct md_sysfs_entry md_size =
4085 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4086
4087
4088 /* Metadata version.
4089  * This is one of
4090  *   'none' for arrays with no metadata (good luck...)
4091  *   'external' for arrays with externally managed metadata,
4092  * or N.M for internally known formats
4093  */
4094 static ssize_t
4095 metadata_show(struct mddev *mddev, char *page)
4096 {
4097         if (mddev->persistent)
4098                 return sprintf(page, "%d.%d\n",
4099                                mddev->major_version, mddev->minor_version);
4100         else if (mddev->external)
4101                 return sprintf(page, "external:%s\n", mddev->metadata_type);
4102         else
4103                 return sprintf(page, "none\n");
4104 }
4105
4106 static ssize_t
4107 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4108 {
4109         int major, minor;
4110         char *e;
4111         /* Changing the details of 'external' metadata is
4112          * always permitted.  Otherwise there must be
4113          * no devices attached to the array.
4114          */
4115         if (mddev->external && strncmp(buf, "external:", 9) == 0)
4116                 ;
4117         else if (!list_empty(&mddev->disks))
4118                 return -EBUSY;
4119
4120         if (cmd_match(buf, "none")) {
4121                 mddev->persistent = 0;
4122                 mddev->external = 0;
4123                 mddev->major_version = 0;
4124                 mddev->minor_version = 90;
4125                 return len;
4126         }
4127         if (strncmp(buf, "external:", 9) == 0) {
4128                 size_t namelen = len-9;
4129                 if (namelen >= sizeof(mddev->metadata_type))
4130                         namelen = sizeof(mddev->metadata_type)-1;
4131                 strncpy(mddev->metadata_type, buf+9, namelen);
4132                 mddev->metadata_type[namelen] = 0;
4133                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4134                         mddev->metadata_type[--namelen] = 0;
4135                 mddev->persistent = 0;
4136                 mddev->external = 1;
4137                 mddev->major_version = 0;
4138                 mddev->minor_version = 90;
4139                 return len;
4140         }
4141         major = simple_strtoul(buf, &e, 10);
4142         if (e==buf || *e != '.')
4143                 return -EINVAL;
4144         buf = e+1;
4145         minor = simple_strtoul(buf, &e, 10);
4146         if (e==buf || (*e && *e != '\n') )
4147                 return -EINVAL;
4148         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4149                 return -ENOENT;
4150         mddev->major_version = major;
4151         mddev->minor_version = minor;
4152         mddev->persistent = 1;
4153         mddev->external = 0;
4154         return len;
4155 }
4156
4157 static struct md_sysfs_entry md_metadata =
4158 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4159
4160 static ssize_t
4161 action_show(struct mddev *mddev, char *page)
4162 {
4163         char *type = "idle";
4164         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4165                 type = "frozen";
4166         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4167             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4168                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4169                         type = "reshape";
4170                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4171                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4172                                 type = "resync";
4173                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4174                                 type = "check";
4175                         else
4176                                 type = "repair";
4177                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4178                         type = "recover";
4179         }
4180         return sprintf(page, "%s\n", type);
4181 }
4182
4183 static ssize_t
4184 action_store(struct mddev *mddev, const char *page, size_t len)
4185 {
4186         if (!mddev->pers || !mddev->pers->sync_request)
4187                 return -EINVAL;
4188
4189         if (cmd_match(page, "frozen"))
4190                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4191         else
4192                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4193
4194         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4195                 if (mddev->sync_thread) {
4196                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4197                         md_reap_sync_thread(mddev);
4198                 }
4199         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4200                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4201                 return -EBUSY;
4202         else if (cmd_match(page, "resync"))
4203                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4204         else if (cmd_match(page, "recover")) {
4205                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4206                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4207         } else if (cmd_match(page, "reshape")) {
4208                 int err;
4209                 if (mddev->pers->start_reshape == NULL)
4210                         return -EINVAL;
4211                 err = mddev->pers->start_reshape(mddev);
4212                 if (err)
4213                         return err;
4214                 sysfs_notify(&mddev->kobj, NULL, "degraded");
4215         } else {
4216                 if (cmd_match(page, "check"))
4217                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4218                 else if (!cmd_match(page, "repair"))
4219                         return -EINVAL;
4220                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4221                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4222         }
4223         if (mddev->ro == 2) {
4224                 /* A write to sync_action is enough to justify
4225                  * canceling read-auto mode
4226                  */
4227                 mddev->ro = 0;
4228                 md_wakeup_thread(mddev->sync_thread);
4229         }
4230         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4231         md_wakeup_thread(mddev->thread);
4232         sysfs_notify_dirent_safe(mddev->sysfs_action);
4233         return len;
4234 }
4235
4236 static struct md_sysfs_entry md_scan_mode =
4237 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4238
4239 static ssize_t
4240 last_sync_action_show(struct mddev *mddev, char *page)
4241 {
4242         return sprintf(page, "%s\n", mddev->last_sync_action);
4243 }
4244
4245 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4246
4247 static ssize_t
4248 mismatch_cnt_show(struct mddev *mddev, char *page)
4249 {
4250         return sprintf(page, "%llu\n",
4251                        (unsigned long long)
4252                        atomic64_read(&mddev->resync_mismatches));
4253 }
4254
4255 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4256
4257 static ssize_t
4258 sync_min_show(struct mddev *mddev, char *page)
4259 {
4260         return sprintf(page, "%d (%s)\n", speed_min(mddev),
4261                        mddev->sync_speed_min ? "local": "system");
4262 }
4263
4264 static ssize_t
4265 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4266 {
4267         int min;
4268         char *e;
4269         if (strncmp(buf, "system", 6)==0) {
4270                 mddev->sync_speed_min = 0;
4271                 return len;
4272         }
4273         min = simple_strtoul(buf, &e, 10);
4274         if (buf == e || (*e && *e != '\n') || min <= 0)
4275                 return -EINVAL;
4276         mddev->sync_speed_min = min;
4277         return len;
4278 }
4279
4280 static struct md_sysfs_entry md_sync_min =
4281 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4282
4283 static ssize_t
4284 sync_max_show(struct mddev *mddev, char *page)
4285 {
4286         return sprintf(page, "%d (%s)\n", speed_max(mddev),
4287                        mddev->sync_speed_max ? "local": "system");
4288 }
4289
4290 static ssize_t
4291 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4292 {
4293         int max;
4294         char *e;
4295         if (strncmp(buf, "system", 6)==0) {
4296                 mddev->sync_speed_max = 0;
4297                 return len;
4298         }
4299         max = simple_strtoul(buf, &e, 10);
4300         if (buf == e || (*e && *e != '\n') || max <= 0)
4301                 return -EINVAL;
4302         mddev->sync_speed_max = max;
4303         return len;
4304 }
4305
4306 static struct md_sysfs_entry md_sync_max =
4307 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4308
4309 static ssize_t
4310 degraded_show(struct mddev *mddev, char *page)
4311 {
4312         return sprintf(page, "%d\n", mddev->degraded);
4313 }
4314 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4315
4316 static ssize_t
4317 sync_force_parallel_show(struct mddev *mddev, char *page)
4318 {
4319         return sprintf(page, "%d\n", mddev->parallel_resync);
4320 }
4321
4322 static ssize_t
4323 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4324 {
4325         long n;
4326
4327         if (kstrtol(buf, 10, &n))
4328                 return -EINVAL;
4329
4330         if (n != 0 && n != 1)
4331                 return -EINVAL;
4332
4333         mddev->parallel_resync = n;
4334
4335         if (mddev->sync_thread)
4336                 wake_up(&resync_wait);
4337
4338         return len;
4339 }
4340
4341 /* force parallel resync, even with shared block devices */
4342 static struct md_sysfs_entry md_sync_force_parallel =
4343 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4344        sync_force_parallel_show, sync_force_parallel_store);
4345
4346 static ssize_t
4347 sync_speed_show(struct mddev *mddev, char *page)
4348 {
4349         unsigned long resync, dt, db;
4350         if (mddev->curr_resync == 0)
4351                 return sprintf(page, "none\n");
4352         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4353         dt = (jiffies - mddev->resync_mark) / HZ;
4354         if (!dt) dt++;
4355         db = resync - mddev->resync_mark_cnt;
4356         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4357 }
4358
4359 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4360
4361 static ssize_t
4362 sync_completed_show(struct mddev *mddev, char *page)
4363 {
4364         unsigned long long max_sectors, resync;
4365
4366         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4367                 return sprintf(page, "none\n");
4368
4369         if (mddev->curr_resync == 1 ||
4370             mddev->curr_resync == 2)
4371                 return sprintf(page, "delayed\n");
4372
4373         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4374             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4375                 max_sectors = mddev->resync_max_sectors;
4376         else
4377                 max_sectors = mddev->dev_sectors;
4378
4379         resync = mddev->curr_resync_completed;
4380         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4381 }
4382
4383 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4384
4385 static ssize_t
4386 min_sync_show(struct mddev *mddev, char *page)
4387 {
4388         return sprintf(page, "%llu\n",
4389                        (unsigned long long)mddev->resync_min);
4390 }
4391 static ssize_t
4392 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4393 {
4394         unsigned long long min;
4395         if (kstrtoull(buf, 10, &min))
4396                 return -EINVAL;
4397         if (min > mddev->resync_max)
4398                 return -EINVAL;
4399         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4400                 return -EBUSY;
4401
4402         /* Must be a multiple of chunk_size */
4403         if (mddev->chunk_sectors) {
4404                 sector_t temp = min;
4405                 if (sector_div(temp, mddev->chunk_sectors))
4406                         return -EINVAL;
4407         }
4408         mddev->resync_min = min;
4409
4410         return len;
4411 }
4412
4413 static struct md_sysfs_entry md_min_sync =
4414 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4415
4416 static ssize_t
4417 max_sync_show(struct mddev *mddev, char *page)
4418 {
4419         if (mddev->resync_max == MaxSector)
4420                 return sprintf(page, "max\n");
4421         else
4422                 return sprintf(page, "%llu\n",
4423                                (unsigned long long)mddev->resync_max);
4424 }
4425 static ssize_t
4426 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4427 {
4428         if (strncmp(buf, "max", 3) == 0)
4429                 mddev->resync_max = MaxSector;
4430         else {
4431                 unsigned long long max;
4432                 if (kstrtoull(buf, 10, &max))
4433                         return -EINVAL;
4434                 if (max < mddev->resync_min)
4435                         return -EINVAL;
4436                 if (max < mddev->resync_max &&
4437                     mddev->ro == 0 &&
4438                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4439                         return -EBUSY;
4440
4441                 /* Must be a multiple of chunk_size */
4442                 if (mddev->chunk_sectors) {
4443                         sector_t temp = max;
4444                         if (sector_div(temp, mddev->chunk_sectors))
4445                                 return -EINVAL;
4446                 }
4447                 mddev->resync_max = max;
4448         }
4449         wake_up(&mddev->recovery_wait);
4450         return len;
4451 }
4452
4453 static struct md_sysfs_entry md_max_sync =
4454 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4455
4456 static ssize_t
4457 suspend_lo_show(struct mddev *mddev, char *page)
4458 {
4459         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4460 }
4461
4462 static ssize_t
4463 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4464 {
4465         char *e;
4466         unsigned long long new = simple_strtoull(buf, &e, 10);
4467         unsigned long long old = mddev->suspend_lo;
4468
4469         if (mddev->pers == NULL ||
4470             mddev->pers->quiesce == NULL)
4471                 return -EINVAL;
4472         if (buf == e || (*e && *e != '\n'))
4473                 return -EINVAL;
4474
4475         mddev->suspend_lo = new;
4476         if (new >= old)
4477                 /* Shrinking suspended region */
4478                 mddev->pers->quiesce(mddev, 2);
4479         else {
4480                 /* Expanding suspended region - need to wait */
4481                 mddev->pers->quiesce(mddev, 1);
4482                 mddev->pers->quiesce(mddev, 0);
4483         }
4484         return len;
4485 }
4486 static struct md_sysfs_entry md_suspend_lo =
4487 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4488
4489
4490 static ssize_t
4491 suspend_hi_show(struct mddev *mddev, char *page)
4492 {
4493         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4494 }
4495
4496 static ssize_t
4497 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4498 {
4499         char *e;
4500         unsigned long long new = simple_strtoull(buf, &e, 10);
4501         unsigned long long old = mddev->suspend_hi;
4502
4503         if (mddev->pers == NULL ||
4504             mddev->pers->quiesce == NULL)
4505                 return -EINVAL;
4506         if (buf == e || (*e && *e != '\n'))
4507                 return -EINVAL;
4508
4509         mddev->suspend_hi = new;
4510         if (new <= old)
4511                 /* Shrinking suspended region */
4512                 mddev->pers->quiesce(mddev, 2);
4513         else {
4514                 /* Expanding suspended region - need to wait */
4515                 mddev->pers->quiesce(mddev, 1);
4516                 mddev->pers->quiesce(mddev, 0);
4517         }
4518         return len;
4519 }
4520 static struct md_sysfs_entry md_suspend_hi =
4521 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4522
4523 static ssize_t
4524 reshape_position_show(struct mddev *mddev, char *page)
4525 {
4526         if (mddev->reshape_position != MaxSector)
4527                 return sprintf(page, "%llu\n",
4528                                (unsigned long long)mddev->reshape_position);
4529         strcpy(page, "none\n");
4530         return 5;
4531 }
4532
4533 static ssize_t
4534 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4535 {
4536         struct md_rdev *rdev;
4537         char *e;
4538         unsigned long long new = simple_strtoull(buf, &e, 10);
4539         if (mddev->pers)
4540                 return -EBUSY;
4541         if (buf == e || (*e && *e != '\n'))
4542                 return -EINVAL;
4543         mddev->reshape_position = new;
4544         mddev->delta_disks = 0;
4545         mddev->reshape_backwards = 0;
4546         mddev->new_level = mddev->level;
4547         mddev->new_layout = mddev->layout;
4548         mddev->new_chunk_sectors = mddev->chunk_sectors;
4549         rdev_for_each(rdev, mddev)
4550                 rdev->new_data_offset = rdev->data_offset;
4551         return len;
4552 }
4553
4554 static struct md_sysfs_entry md_reshape_position =
4555 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4556        reshape_position_store);
4557
4558 static ssize_t
4559 reshape_direction_show(struct mddev *mddev, char *page)
4560 {
4561         return sprintf(page, "%s\n",
4562                        mddev->reshape_backwards ? "backwards" : "forwards");
4563 }
4564
4565 static ssize_t
4566 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4567 {
4568         int backwards = 0;
4569         if (cmd_match(buf, "forwards"))
4570                 backwards = 0;
4571         else if (cmd_match(buf, "backwards"))
4572                 backwards = 1;
4573         else
4574                 return -EINVAL;
4575         if (mddev->reshape_backwards == backwards)
4576                 return len;
4577
4578         /* check if we are allowed to change */
4579         if (mddev->delta_disks)
4580                 return -EBUSY;
4581
4582         if (mddev->persistent &&
4583             mddev->major_version == 0)
4584                 return -EINVAL;
4585
4586         mddev->reshape_backwards = backwards;
4587         return len;
4588 }
4589
4590 static struct md_sysfs_entry md_reshape_direction =
4591 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4592        reshape_direction_store);
4593
4594 static ssize_t
4595 array_size_show(struct mddev *mddev, char *page)
4596 {
4597         if (mddev->external_size)
4598                 return sprintf(page, "%llu\n",
4599                                (unsigned long long)mddev->array_sectors/2);
4600         else
4601                 return sprintf(page, "default\n");
4602 }
4603
4604 static ssize_t
4605 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4606 {
4607         sector_t sectors;
4608
4609         if (strncmp(buf, "default", 7) == 0) {
4610                 if (mddev->pers)
4611                         sectors = mddev->pers->size(mddev, 0, 0);
4612                 else
4613                         sectors = mddev->array_sectors;
4614
4615                 mddev->external_size = 0;
4616         } else {
4617                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4618                         return -EINVAL;
4619                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4620                         return -E2BIG;
4621
4622                 mddev->external_size = 1;
4623         }
4624
4625         mddev->array_sectors = sectors;
4626         if (mddev->pers) {
4627                 set_capacity(mddev->gendisk, mddev->array_sectors);
4628                 revalidate_disk(mddev->gendisk);
4629         }
4630         return len;
4631 }
4632
4633 static struct md_sysfs_entry md_array_size =
4634 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4635        array_size_store);
4636
4637 static struct attribute *md_default_attrs[] = {
4638         &md_level.attr,
4639         &md_layout.attr,
4640         &md_raid_disks.attr,
4641         &md_chunk_size.attr,
4642         &md_size.attr,
4643         &md_resync_start.attr,
4644         &md_metadata.attr,
4645         &md_new_device.attr,
4646         &md_safe_delay.attr,
4647         &md_array_state.attr,
4648         &md_reshape_position.attr,
4649         &md_reshape_direction.attr,
4650         &md_array_size.attr,
4651         &max_corr_read_errors.attr,
4652         NULL,
4653 };
4654
4655 static struct attribute *md_redundancy_attrs[] = {
4656         &md_scan_mode.attr,
4657         &md_last_scan_mode.attr,
4658         &md_mismatches.attr,
4659         &md_sync_min.attr,
4660         &md_sync_max.attr,
4661         &md_sync_speed.attr,
4662         &md_sync_force_parallel.attr,
4663         &md_sync_completed.attr,
4664         &md_min_sync.attr,
4665         &md_max_sync.attr,
4666         &md_suspend_lo.attr,
4667         &md_suspend_hi.attr,
4668         &md_bitmap.attr,
4669         &md_degraded.attr,
4670         NULL,
4671 };
4672 static struct attribute_group md_redundancy_group = {
4673         .name = NULL,
4674         .attrs = md_redundancy_attrs,
4675 };
4676
4677
4678 static ssize_t
4679 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4680 {
4681         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4682         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4683         ssize_t rv;
4684
4685         if (!entry->show)
4686                 return -EIO;
4687         spin_lock(&all_mddevs_lock);
4688         if (list_empty(&mddev->all_mddevs)) {
4689                 spin_unlock(&all_mddevs_lock);
4690                 return -EBUSY;
4691         }
4692         mddev_get(mddev);
4693         spin_unlock(&all_mddevs_lock);
4694
4695         rv = mddev_lock(mddev);
4696         if (!rv) {
4697                 rv = entry->show(mddev, page);
4698                 mddev_unlock(mddev);
4699         }
4700         mddev_put(mddev);
4701         return rv;
4702 }
4703
4704 static ssize_t
4705 md_attr_store(struct kobject *kobj, struct attribute *attr,
4706               const char *page, size_t length)
4707 {
4708         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4709         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4710         ssize_t rv;
4711
4712         if (!entry->store)
4713                 return -EIO;
4714         if (!capable(CAP_SYS_ADMIN))
4715                 return -EACCES;
4716         spin_lock(&all_mddevs_lock);
4717         if (list_empty(&mddev->all_mddevs)) {
4718                 spin_unlock(&all_mddevs_lock);
4719                 return -EBUSY;
4720         }
4721         mddev_get(mddev);
4722         spin_unlock(&all_mddevs_lock);
4723         if (entry->store == new_dev_store)
4724                 flush_workqueue(md_misc_wq);
4725         rv = mddev_lock(mddev);
4726         if (!rv) {
4727                 rv = entry->store(mddev, page, length);
4728                 mddev_unlock(mddev);
4729         }
4730         mddev_put(mddev);
4731         return rv;
4732 }
4733
4734 static void md_free(struct kobject *ko)
4735 {
4736         struct mddev *mddev = container_of(ko, struct mddev, kobj);
4737
4738         if (mddev->sysfs_state)
4739                 sysfs_put(mddev->sysfs_state);
4740
4741         if (mddev->gendisk) {
4742                 del_gendisk(mddev->gendisk);
4743                 put_disk(mddev->gendisk);
4744         }
4745         if (mddev->queue)
4746                 blk_cleanup_queue(mddev->queue);
4747
4748         kfree(mddev);
4749 }
4750
4751 static const struct sysfs_ops md_sysfs_ops = {
4752         .show   = md_attr_show,
4753         .store  = md_attr_store,
4754 };
4755 static struct kobj_type md_ktype = {
4756         .release        = md_free,
4757         .sysfs_ops      = &md_sysfs_ops,
4758         .default_attrs  = md_default_attrs,
4759 };
4760
4761 int mdp_major = 0;
4762
4763 static void mddev_delayed_delete(struct work_struct *ws)
4764 {
4765         struct mddev *mddev = container_of(ws, struct mddev, del_work);
4766
4767         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4768         kobject_del(&mddev->kobj);
4769         kobject_put(&mddev->kobj);
4770 }
4771
4772 static int md_alloc(dev_t dev, char *name)
4773 {
4774         static DEFINE_MUTEX(disks_mutex);
4775         struct mddev *mddev = mddev_find(dev);
4776         struct gendisk *disk;
4777         int partitioned;
4778         int shift;
4779         int unit;
4780         int error;
4781
4782         if (!mddev)
4783                 return -ENODEV;
4784
4785         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4786         shift = partitioned ? MdpMinorShift : 0;
4787         unit = MINOR(mddev->unit) >> shift;
4788
4789         /* wait for any previous instance of this device to be
4790          * completely removed (mddev_delayed_delete).
4791          */
4792         flush_workqueue(md_misc_wq);
4793
4794         mutex_lock(&disks_mutex);
4795         error = -EEXIST;
4796         if (mddev->gendisk)
4797                 goto abort;
4798
4799         if (name) {
4800                 /* Need to ensure that 'name' is not a duplicate.
4801                  */
4802                 struct mddev *mddev2;
4803                 spin_lock(&all_mddevs_lock);
4804
4805                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4806                         if (mddev2->gendisk &&
4807                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4808                                 spin_unlock(&all_mddevs_lock);
4809                                 goto abort;
4810                         }
4811                 spin_unlock(&all_mddevs_lock);
4812         }
4813
4814         error = -ENOMEM;
4815         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4816         if (!mddev->queue)
4817                 goto abort;
4818         mddev->queue->queuedata = mddev;
4819
4820         blk_queue_make_request(mddev->queue, md_make_request);
4821         blk_set_stacking_limits(&mddev->queue->limits);
4822
4823         disk = alloc_disk(1 << shift);
4824         if (!disk) {
4825                 blk_cleanup_queue(mddev->queue);
4826                 mddev->queue = NULL;
4827                 goto abort;
4828         }
4829         disk->major = MAJOR(mddev->unit);
4830         disk->first_minor = unit << shift;
4831         if (name)
4832                 strcpy(disk->disk_name, name);
4833         else if (partitioned)
4834                 sprintf(disk->disk_name, "md_d%d", unit);
4835         else
4836                 sprintf(disk->disk_name, "md%d", unit);
4837         disk->fops = &md_fops;
4838         disk->private_data = mddev;
4839         disk->queue = mddev->queue;
4840         blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4841         /* Allow extended partitions.  This makes the
4842          * 'mdp' device redundant, but we can't really
4843          * remove it now.
4844          */
4845         disk->flags |= GENHD_FL_EXT_DEVT;
4846         mddev->gendisk = disk;
4847         /* As soon as we call add_disk(), another thread could get
4848          * through to md_open, so make sure it doesn't get too far
4849          */
4850         mutex_lock(&mddev->open_mutex);
4851         add_disk(disk);
4852
4853         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4854                                      &disk_to_dev(disk)->kobj, "%s", "md");
4855         if (error) {
4856                 /* This isn't possible, but as kobject_init_and_add is marked
4857                  * __must_check, we must do something with the result
4858                  */
4859                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4860                        disk->disk_name);
4861                 error = 0;
4862         }
4863         if (mddev->kobj.sd &&
4864             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4865                 printk(KERN_DEBUG "pointless warning\n");
4866         mutex_unlock(&mddev->open_mutex);
4867  abort:
4868         mutex_unlock(&disks_mutex);
4869         if (!error && mddev->kobj.sd) {
4870                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4871                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4872         }
4873         mddev_put(mddev);
4874         return error;
4875 }
4876
4877 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4878 {
4879         md_alloc(dev, NULL);
4880         return NULL;
4881 }
4882
4883 static int add_named_array(const char *val, struct kernel_param *kp)
4884 {
4885         /* val must be "md_*" where * is not all digits.
4886          * We allocate an array with a large free minor number, and
4887          * set the name to val.  val must not already be an active name.
4888          */
4889         int len = strlen(val);
4890         char buf[DISK_NAME_LEN];
4891
4892         while (len && val[len-1] == '\n')
4893                 len--;
4894         if (len >= DISK_NAME_LEN)
4895                 return -E2BIG;
4896         strlcpy(buf, val, len+1);
4897         if (strncmp(buf, "md_", 3) != 0)
4898                 return -EINVAL;
4899         return md_alloc(0, buf);
4900 }
4901
4902 static void md_safemode_timeout(unsigned long data)
4903 {
4904         struct mddev *mddev = (struct mddev *) data;
4905
4906         if (!atomic_read(&mddev->writes_pending)) {
4907                 mddev->safemode = 1;
4908                 if (mddev->external)
4909                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4910         }
4911         md_wakeup_thread(mddev->thread);
4912 }
4913
4914 static int start_dirty_degraded;
4915
4916 int md_run(struct mddev *mddev)
4917 {
4918         int err;
4919         struct md_rdev *rdev;
4920         struct md_personality *pers;
4921
4922         if (list_empty(&mddev->disks))
4923                 /* cannot run an array with no devices.. */
4924                 return -EINVAL;
4925
4926         if (mddev->pers)
4927                 return -EBUSY;
4928         /* Cannot run until previous stop completes properly */
4929         if (mddev->sysfs_active)
4930                 return -EBUSY;
4931
4932         /*
4933          * Analyze all RAID superblock(s)
4934          */
4935         if (!mddev->raid_disks) {
4936                 if (!mddev->persistent)
4937                         return -EINVAL;
4938                 analyze_sbs(mddev);
4939         }
4940
4941         if (mddev->level != LEVEL_NONE)
4942                 request_module("md-level-%d", mddev->level);
4943         else if (mddev->clevel[0])
4944                 request_module("md-%s", mddev->clevel);
4945
4946         /*
4947          * Drop all container device buffers, from now on
4948          * the only valid external interface is through the md
4949          * device.
4950          */
4951         rdev_for_each(rdev, mddev) {
4952                 if (test_bit(Faulty, &rdev->flags))
4953                         continue;
4954                 sync_blockdev(rdev->bdev);
4955                 invalidate_bdev(rdev->bdev);
4956
4957                 /* perform some consistency tests on the device.
4958                  * We don't want the data to overlap the metadata,
4959                  * Internal Bitmap issues have been handled elsewhere.
4960                  */
4961                 if (rdev->meta_bdev) {
4962                         /* Nothing to check */;
4963                 } else if (rdev->data_offset < rdev->sb_start) {
4964                         if (mddev->dev_sectors &&
4965                             rdev->data_offset + mddev->dev_sectors
4966                             > rdev->sb_start) {
4967                                 printk("md: %s: data overlaps metadata\n",
4968                                        mdname(mddev));
4969                                 return -EINVAL;
4970                         }
4971                 } else {
4972                         if (rdev->sb_start + rdev->sb_size/512
4973                             > rdev->data_offset) {
4974                                 printk("md: %s: metadata overlaps data\n",
4975                                        mdname(mddev));
4976                                 return -EINVAL;
4977                         }
4978                 }
4979                 sysfs_notify_dirent_safe(rdev->sysfs_state);
4980         }
4981
4982         if (mddev->bio_set == NULL)
4983                 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4984
4985         spin_lock(&pers_lock);
4986         pers = find_pers(mddev->level, mddev->clevel);
4987         if (!pers || !try_module_get(pers->owner)) {
4988                 spin_unlock(&pers_lock);
4989                 if (mddev->level != LEVEL_NONE)
4990                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4991                                mddev->level);
4992                 else
4993                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4994                                mddev->clevel);
4995                 return -EINVAL;
4996         }
4997         mddev->pers = pers;
4998         spin_unlock(&pers_lock);
4999         if (mddev->level != pers->level) {
5000                 mddev->level = pers->level;
5001                 mddev->new_level = pers->level;
5002         }
5003         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5004
5005         if (mddev->reshape_position != MaxSector &&
5006             pers->start_reshape == NULL) {
5007                 /* This personality cannot handle reshaping... */
5008                 mddev->pers = NULL;
5009                 module_put(pers->owner);
5010                 return -EINVAL;
5011         }
5012
5013         if (pers->sync_request) {
5014                 /* Warn if this is a potentially silly
5015                  * configuration.
5016                  */
5017                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5018                 struct md_rdev *rdev2;
5019                 int warned = 0;
5020
5021                 rdev_for_each(rdev, mddev)
5022                         rdev_for_each(rdev2, mddev) {
5023                                 if (rdev < rdev2 &&
5024                                     rdev->bdev->bd_contains ==
5025                                     rdev2->bdev->bd_contains) {
5026                                         printk(KERN_WARNING
5027                                                "%s: WARNING: %s appears to be"
5028                                                " on the same physical disk as"
5029                                                " %s.\n",
5030                                                mdname(mddev),
5031                                                bdevname(rdev->bdev,b),
5032                                                bdevname(rdev2->bdev,b2));
5033                                         warned = 1;
5034                                 }
5035                         }
5036
5037                 if (warned)
5038                         printk(KERN_WARNING
5039                                "True protection against single-disk"
5040                                " failure might be compromised.\n");
5041         }
5042
5043         mddev->recovery = 0;
5044         /* may be over-ridden by personality */
5045         mddev->resync_max_sectors = mddev->dev_sectors;
5046
5047         mddev->ok_start_degraded = start_dirty_degraded;
5048
5049         if (start_readonly && mddev->ro == 0)
5050                 mddev->ro = 2; /* read-only, but switch on first write */
5051
5052         err = mddev->pers->run(mddev);
5053         if (err)
5054                 printk(KERN_ERR "md: pers->run() failed ...\n");
5055         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5056                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5057                           " but 'external_size' not in effect?\n", __func__);
5058                 printk(KERN_ERR
5059                        "md: invalid array_size %llu > default size %llu\n",
5060                        (unsigned long long)mddev->array_sectors / 2,
5061                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5062                 err = -EINVAL;
5063                 mddev->pers->stop(mddev);
5064         }
5065         if (err == 0 && mddev->pers->sync_request &&
5066             (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5067                 err = bitmap_create(mddev);
5068                 if (err) {
5069                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5070                                mdname(mddev), err);
5071                         mddev->pers->stop(mddev);
5072                 }
5073         }
5074         if (err) {
5075                 module_put(mddev->pers->owner);
5076                 mddev->pers = NULL;
5077                 bitmap_destroy(mddev);
5078                 return err;
5079         }
5080         if (mddev->pers->sync_request) {
5081                 if (mddev->kobj.sd &&
5082                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5083                         printk(KERN_WARNING
5084                                "md: cannot register extra attributes for %s\n",
5085                                mdname(mddev));
5086                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5087         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5088                 mddev->ro = 0;
5089
5090         atomic_set(&mddev->writes_pending,0);
5091         atomic_set(&mddev->max_corr_read_errors,
5092                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5093         mddev->safemode = 0;
5094         mddev->safemode_timer.function = md_safemode_timeout;
5095         mddev->safemode_timer.data = (unsigned long) mddev;
5096         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5097         mddev->in_sync = 1;
5098         smp_wmb();
5099         mddev->ready = 1;
5100         rdev_for_each(rdev, mddev)
5101                 if (rdev->raid_disk >= 0)
5102                         if (sysfs_link_rdev(mddev, rdev))
5103                                 /* failure here is OK */;
5104
5105         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5106
5107         if (mddev->flags & MD_UPDATE_SB_FLAGS)
5108                 md_update_sb(mddev, 0);
5109
5110         md_new_event(mddev);
5111         sysfs_notify_dirent_safe(mddev->sysfs_state);
5112         sysfs_notify_dirent_safe(mddev->sysfs_action);
5113         sysfs_notify(&mddev->kobj, NULL, "degraded");
5114         return 0;
5115 }
5116 EXPORT_SYMBOL_GPL(md_run);
5117
5118 static int do_md_run(struct mddev *mddev)
5119 {
5120         int err;
5121
5122         err = md_run(mddev);
5123         if (err)
5124                 goto out;
5125         err = bitmap_load(mddev);
5126         if (err) {
5127                 bitmap_destroy(mddev);
5128                 goto out;
5129         }
5130
5131         md_wakeup_thread(mddev->thread);
5132         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5133
5134         set_capacity(mddev->gendisk, mddev->array_sectors);
5135         revalidate_disk(mddev->gendisk);
5136         mddev->changed = 1;
5137         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5138 out:
5139         return err;
5140 }
5141
5142 static int restart_array(struct mddev *mddev)
5143 {
5144         struct gendisk *disk = mddev->gendisk;
5145
5146         /* Complain if it has no devices */
5147         if (list_empty(&mddev->disks))
5148                 return -ENXIO;
5149         if (!mddev->pers)
5150                 return -EINVAL;
5151         if (!mddev->ro)
5152                 return -EBUSY;
5153         mddev->safemode = 0;
5154         mddev->ro = 0;
5155         set_disk_ro(disk, 0);
5156         printk(KERN_INFO "md: %s switched to read-write mode.\n",
5157                 mdname(mddev));
5158         /* Kick recovery or resync if necessary */
5159         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5160         md_wakeup_thread(mddev->thread);
5161         md_wakeup_thread(mddev->sync_thread);
5162         sysfs_notify_dirent_safe(mddev->sysfs_state);
5163         return 0;
5164 }
5165
5166 /* similar to deny_write_access, but accounts for our holding a reference
5167  * to the file ourselves */
5168 static int deny_bitmap_write_access(struct file * file)
5169 {
5170         struct inode *inode = file->f_mapping->host;
5171
5172         spin_lock(&inode->i_lock);
5173         if (atomic_read(&inode->i_writecount) > 1) {
5174                 spin_unlock(&inode->i_lock);
5175                 return -ETXTBSY;
5176         }
5177         atomic_set(&inode->i_writecount, -1);
5178         spin_unlock(&inode->i_lock);
5179
5180         return 0;
5181 }
5182
5183 void restore_bitmap_write_access(struct file *file)
5184 {
5185         struct inode *inode = file->f_mapping->host;
5186
5187         spin_lock(&inode->i_lock);
5188         atomic_set(&inode->i_writecount, 1);
5189         spin_unlock(&inode->i_lock);
5190 }
5191
5192 static void md_clean(struct mddev *mddev)
5193 {
5194         mddev->array_sectors = 0;
5195         mddev->external_size = 0;
5196         mddev->dev_sectors = 0;
5197         mddev->raid_disks = 0;
5198         mddev->recovery_cp = 0;
5199         mddev->resync_min = 0;
5200         mddev->resync_max = MaxSector;
5201         mddev->reshape_position = MaxSector;
5202         mddev->external = 0;
5203         mddev->persistent = 0;
5204         mddev->level = LEVEL_NONE;
5205         mddev->clevel[0] = 0;
5206         mddev->flags = 0;
5207         mddev->ro = 0;
5208         mddev->metadata_type[0] = 0;
5209         mddev->chunk_sectors = 0;
5210         mddev->ctime = mddev->utime = 0;
5211         mddev->layout = 0;
5212         mddev->max_disks = 0;
5213         mddev->events = 0;
5214         mddev->can_decrease_events = 0;
5215         mddev->delta_disks = 0;
5216         mddev->reshape_backwards = 0;
5217         mddev->new_level = LEVEL_NONE;
5218         mddev->new_layout = 0;
5219         mddev->new_chunk_sectors = 0;
5220         mddev->curr_resync = 0;
5221         atomic64_set(&mddev->resync_mismatches, 0);
5222         mddev->suspend_lo = mddev->suspend_hi = 0;
5223         mddev->sync_speed_min = mddev->sync_speed_max = 0;
5224         mddev->recovery = 0;
5225         mddev->in_sync = 0;
5226         mddev->changed = 0;
5227         mddev->degraded = 0;
5228         mddev->safemode = 0;
5229         mddev->merge_check_needed = 0;
5230         mddev->bitmap_info.offset = 0;
5231         mddev->bitmap_info.default_offset = 0;
5232         mddev->bitmap_info.default_space = 0;
5233         mddev->bitmap_info.chunksize = 0;
5234         mddev->bitmap_info.daemon_sleep = 0;
5235         mddev->bitmap_info.max_write_behind = 0;
5236 }
5237
5238 static void __md_stop_writes(struct mddev *mddev)
5239 {
5240         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5241         if (mddev->sync_thread) {
5242                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5243                 md_reap_sync_thread(mddev);
5244         }
5245
5246         del_timer_sync(&mddev->safemode_timer);
5247
5248         bitmap_flush(mddev);
5249         md_super_wait(mddev);
5250
5251         if (mddev->ro == 0 &&
5252             (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5253                 /* mark array as shutdown cleanly */
5254                 mddev->in_sync = 1;
5255                 md_update_sb(mddev, 1);
5256         }
5257 }
5258
5259 void md_stop_writes(struct mddev *mddev)
5260 {
5261         mddev_lock(mddev);
5262         __md_stop_writes(mddev);
5263         mddev_unlock(mddev);
5264 }
5265 EXPORT_SYMBOL_GPL(md_stop_writes);
5266
5267 static void __md_stop(struct mddev *mddev)
5268 {
5269         mddev->ready = 0;
5270         mddev->pers->stop(mddev);
5271         if (mddev->pers->sync_request && mddev->to_remove == NULL)
5272                 mddev->to_remove = &md_redundancy_group;
5273         module_put(mddev->pers->owner);
5274         mddev->pers = NULL;
5275         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5276 }
5277
5278 void md_stop(struct mddev *mddev)
5279 {
5280         /* stop the array and free an attached data structures.
5281          * This is called from dm-raid
5282          */
5283         __md_stop(mddev);
5284         bitmap_destroy(mddev);
5285         if (mddev->bio_set)
5286                 bioset_free(mddev->bio_set);
5287 }
5288
5289 EXPORT_SYMBOL_GPL(md_stop);
5290
5291 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5292 {
5293         int err = 0;
5294         mutex_lock(&mddev->open_mutex);
5295         if (atomic_read(&mddev->openers) > !!bdev) {
5296                 printk("md: %s still in use.\n",mdname(mddev));
5297                 err = -EBUSY;
5298                 goto out;
5299         }
5300         if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
5301                 /* Someone opened the device since we flushed it
5302                  * so page cache could be dirty and it is too late
5303                  * to flush.  So abort
5304                  */
5305                 mutex_unlock(&mddev->open_mutex);
5306                 return -EBUSY;
5307         }
5308         if (mddev->pers) {
5309                 __md_stop_writes(mddev);
5310
5311                 err  = -ENXIO;
5312                 if (mddev->ro==1)
5313                         goto out;
5314                 mddev->ro = 1;
5315                 set_disk_ro(mddev->gendisk, 1);
5316                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5317                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5318                 err = 0;
5319         }
5320 out:
5321         mutex_unlock(&mddev->open_mutex);
5322         return err;
5323 }
5324
5325 /* mode:
5326  *   0 - completely stop and dis-assemble array
5327  *   2 - stop but do not disassemble array
5328  */
5329 static int do_md_stop(struct mddev * mddev, int mode,
5330                       struct block_device *bdev)
5331 {
5332         struct gendisk *disk = mddev->gendisk;
5333         struct md_rdev *rdev;
5334
5335         mutex_lock(&mddev->open_mutex);
5336         if (atomic_read(&mddev->openers) > !!bdev ||
5337             mddev->sysfs_active) {
5338                 printk("md: %s still in use.\n",mdname(mddev));
5339                 mutex_unlock(&mddev->open_mutex);
5340                 return -EBUSY;
5341         }
5342         if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
5343                 /* Someone opened the device since we flushed it
5344                  * so page cache could be dirty and it is too late
5345                  * to flush.  So abort
5346                  */
5347                 mutex_unlock(&mddev->open_mutex);
5348                 return -EBUSY;
5349         }
5350         if (mddev->pers) {
5351                 if (mddev->ro)
5352                         set_disk_ro(disk, 0);
5353
5354                 __md_stop_writes(mddev);
5355                 __md_stop(mddev);
5356                 mddev->queue->merge_bvec_fn = NULL;
5357                 mddev->queue->backing_dev_info.congested_fn = NULL;
5358
5359                 /* tell userspace to handle 'inactive' */
5360                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5361
5362                 rdev_for_each(rdev, mddev)
5363                         if (rdev->raid_disk >= 0)
5364                                 sysfs_unlink_rdev(mddev, rdev);
5365
5366                 set_capacity(disk, 0);
5367                 mutex_unlock(&mddev->open_mutex);
5368                 mddev->changed = 1;
5369                 revalidate_disk(disk);
5370
5371                 if (mddev->ro)
5372                         mddev->ro = 0;
5373         } else
5374                 mutex_unlock(&mddev->open_mutex);
5375         /*
5376          * Free resources if final stop
5377          */
5378         if (mode == 0) {
5379                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5380
5381                 bitmap_destroy(mddev);
5382                 if (mddev->bitmap_info.file) {
5383                         restore_bitmap_write_access(mddev->bitmap_info.file);
5384                         fput(mddev->bitmap_info.file);
5385                         mddev->bitmap_info.file = NULL;
5386                 }
5387                 mddev->bitmap_info.offset = 0;
5388
5389                 export_array(mddev);
5390
5391                 md_clean(mddev);
5392                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5393                 if (mddev->hold_active == UNTIL_STOP)
5394                         mddev->hold_active = 0;
5395         }
5396         blk_integrity_unregister(disk);
5397         md_new_event(mddev);
5398         sysfs_notify_dirent_safe(mddev->sysfs_state);
5399         return 0;
5400 }
5401
5402 #ifndef MODULE
5403 static void autorun_array(struct mddev *mddev)
5404 {
5405         struct md_rdev *rdev;
5406         int err;
5407
5408         if (list_empty(&mddev->disks))
5409                 return;
5410
5411         printk(KERN_INFO "md: running: ");
5412
5413         rdev_for_each(rdev, mddev) {
5414                 char b[BDEVNAME_SIZE];
5415                 printk("<%s>", bdevname(rdev->bdev,b));
5416         }
5417         printk("\n");
5418
5419         err = do_md_run(mddev);
5420         if (err) {
5421                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5422                 do_md_stop(mddev, 0, NULL);
5423         }
5424 }
5425
5426 /*
5427  * lets try to run arrays based on all disks that have arrived
5428  * until now. (those are in pending_raid_disks)
5429  *
5430  * the method: pick the first pending disk, collect all disks with
5431  * the same UUID, remove all from the pending list and put them into
5432  * the 'same_array' list. Then order this list based on superblock
5433  * update time (freshest comes first), kick out 'old' disks and
5434  * compare superblocks. If everything's fine then run it.
5435  *
5436  * If "unit" is allocated, then bump its reference count
5437  */
5438 static void autorun_devices(int part)
5439 {
5440         struct md_rdev *rdev0, *rdev, *tmp;
5441         struct mddev *mddev;
5442         char b[BDEVNAME_SIZE];
5443
5444         printk(KERN_INFO "md: autorun ...\n");
5445         while (!list_empty(&pending_raid_disks)) {
5446                 int unit;
5447                 dev_t dev;
5448                 LIST_HEAD(candidates);
5449                 rdev0 = list_entry(pending_raid_disks.next,
5450                                          struct md_rdev, same_set);
5451
5452                 printk(KERN_INFO "md: considering %s ...\n",
5453                         bdevname(rdev0->bdev,b));
5454                 INIT_LIST_HEAD(&candidates);
5455                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5456                         if (super_90_load(rdev, rdev0, 0) >= 0) {
5457                                 printk(KERN_INFO "md:  adding %s ...\n",
5458                                         bdevname(rdev->bdev,b));
5459                                 list_move(&rdev->same_set, &candidates);
5460                         }
5461                 /*
5462                  * now we have a set of devices, with all of them having
5463                  * mostly sane superblocks. It's time to allocate the
5464                  * mddev.
5465                  */
5466                 if (part) {
5467                         dev = MKDEV(mdp_major,
5468                                     rdev0->preferred_minor << MdpMinorShift);
5469                         unit = MINOR(dev) >> MdpMinorShift;
5470                 } else {
5471                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5472                         unit = MINOR(dev);
5473                 }
5474                 if (rdev0->preferred_minor != unit) {
5475                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5476                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5477                         break;
5478                 }
5479
5480                 md_probe(dev, NULL, NULL);
5481                 mddev = mddev_find(dev);
5482                 if (!mddev || !mddev->gendisk) {
5483                         if (mddev)
5484                                 mddev_put(mddev);
5485                         printk(KERN_ERR
5486                                 "md: cannot allocate memory for md drive.\n");
5487                         break;
5488                 }
5489                 if (mddev_lock(mddev))
5490                         printk(KERN_WARNING "md: %s locked, cannot run\n",
5491                                mdname(mddev));
5492                 else if (mddev->raid_disks || mddev->major_version
5493                          || !list_empty(&mddev->disks)) {
5494                         printk(KERN_WARNING
5495                                 "md: %s already running, cannot run %s\n",
5496                                 mdname(mddev), bdevname(rdev0->bdev,b));
5497                         mddev_unlock(mddev);
5498                 } else {
5499                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
5500                         mddev->persistent = 1;
5501                         rdev_for_each_list(rdev, tmp, &candidates) {
5502                                 list_del_init(&rdev->same_set);
5503                                 if (bind_rdev_to_array(rdev, mddev))
5504                                         export_rdev(rdev);
5505                         }
5506                         autorun_array(mddev);
5507                         mddev_unlock(mddev);
5508                 }
5509                 /* on success, candidates will be empty, on error
5510                  * it won't...
5511                  */
5512                 rdev_for_each_list(rdev, tmp, &candidates) {
5513                         list_del_init(&rdev->same_set);
5514                         export_rdev(rdev);
5515                 }
5516                 mddev_put(mddev);
5517         }
5518         printk(KERN_INFO "md: ... autorun DONE.\n");
5519 }
5520 #endif /* !MODULE */
5521
5522 static int get_version(void __user * arg)
5523 {
5524         mdu_version_t ver;
5525
5526         ver.major = MD_MAJOR_VERSION;
5527         ver.minor = MD_MINOR_VERSION;
5528         ver.patchlevel = MD_PATCHLEVEL_VERSION;
5529
5530         if (copy_to_user(arg, &ver, sizeof(ver)))
5531                 return -EFAULT;
5532
5533         return 0;
5534 }
5535
5536 static int get_array_info(struct mddev * mddev, void __user * arg)
5537 {
5538         mdu_array_info_t info;
5539         int nr,working,insync,failed,spare;
5540         struct md_rdev *rdev;
5541
5542         nr = working = insync = failed = spare = 0;
5543         rcu_read_lock();
5544         rdev_for_each_rcu(rdev, mddev) {
5545                 nr++;
5546                 if (test_bit(Faulty, &rdev->flags))
5547                         failed++;
5548                 else {
5549                         working++;
5550                         if (test_bit(In_sync, &rdev->flags))
5551                                 insync++;
5552                         else
5553                                 spare++;
5554                 }
5555         }
5556         rcu_read_unlock();
5557
5558         info.major_version = mddev->major_version;
5559         info.minor_version = mddev->minor_version;
5560         info.patch_version = MD_PATCHLEVEL_VERSION;
5561         info.ctime         = mddev->ctime;
5562         info.level         = mddev->level;
5563         info.size          = mddev->dev_sectors / 2;
5564         if (info.size != mddev->dev_sectors / 2) /* overflow */
5565                 info.size = -1;
5566         info.nr_disks      = nr;
5567         info.raid_disks    = mddev->raid_disks;
5568         info.md_minor      = mddev->md_minor;
5569         info.not_persistent= !mddev->persistent;
5570
5571         info.utime         = mddev->utime;
5572         info.state         = 0;
5573         if (mddev->in_sync)
5574                 info.state = (1<<MD_SB_CLEAN);
5575         if (mddev->bitmap && mddev->bitmap_info.offset)
5576                 info.state = (1<<MD_SB_BITMAP_PRESENT);
5577         info.active_disks  = insync;
5578         info.working_disks = working;
5579         info.failed_disks  = failed;
5580         info.spare_disks   = spare;
5581
5582         info.layout        = mddev->layout;
5583         info.chunk_size    = mddev->chunk_sectors << 9;
5584
5585         if (copy_to_user(arg, &info, sizeof(info)))
5586                 return -EFAULT;
5587
5588         return 0;
5589 }
5590
5591 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5592 {
5593         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5594         char *ptr, *buf = NULL;
5595         int err = -ENOMEM;
5596
5597         file = kmalloc(sizeof(*file), GFP_NOIO);
5598
5599         if (!file)
5600                 goto out;
5601
5602         /* bitmap disabled, zero the first byte and copy out */
5603         if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5604                 file->pathname[0] = '\0';
5605                 goto copy_out;
5606         }
5607
5608         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5609         if (!buf)
5610                 goto out;
5611
5612         ptr = d_path(&mddev->bitmap->storage.file->f_path,
5613                      buf, sizeof(file->pathname));
5614         if (IS_ERR(ptr))
5615                 goto out;
5616
5617         strcpy(file->pathname, ptr);
5618
5619 copy_out:
5620         err = 0;
5621         if (copy_to_user(arg, file, sizeof(*file)))
5622                 err = -EFAULT;
5623 out:
5624         kfree(buf);
5625         kfree(file);
5626         return err;
5627 }
5628
5629 static int get_disk_info(struct mddev * mddev, void __user * arg)
5630 {
5631         mdu_disk_info_t info;
5632         struct md_rdev *rdev;
5633
5634         if (copy_from_user(&info, arg, sizeof(info)))
5635                 return -EFAULT;
5636
5637         rcu_read_lock();
5638         rdev = find_rdev_nr_rcu(mddev, info.number);
5639         if (rdev) {
5640                 info.major = MAJOR(rdev->bdev->bd_dev);
5641                 info.minor = MINOR(rdev->bdev->bd_dev);
5642                 info.raid_disk = rdev->raid_disk;
5643                 info.state = 0;
5644                 if (test_bit(Faulty, &rdev->flags))
5645                         info.state |= (1<<MD_DISK_FAULTY);
5646                 else if (test_bit(In_sync, &rdev->flags)) {
5647                         info.state |= (1<<MD_DISK_ACTIVE);
5648                         info.state |= (1<<MD_DISK_SYNC);
5649                 }
5650                 if (test_bit(WriteMostly, &rdev->flags))
5651                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5652         } else {
5653                 info.major = info.minor = 0;
5654                 info.raid_disk = -1;
5655                 info.state = (1<<MD_DISK_REMOVED);
5656         }
5657         rcu_read_unlock();
5658
5659         if (copy_to_user(arg, &info, sizeof(info)))
5660                 return -EFAULT;
5661
5662         return 0;
5663 }
5664
5665 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5666 {
5667         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5668         struct md_rdev *rdev;
5669         dev_t dev = MKDEV(info->major,info->minor);
5670
5671         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5672                 return -EOVERFLOW;
5673
5674         if (!mddev->raid_disks) {
5675                 int err;
5676                 /* expecting a device which has a superblock */
5677                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5678                 if (IS_ERR(rdev)) {
5679                         printk(KERN_WARNING
5680                                 "md: md_import_device returned %ld\n",
5681                                 PTR_ERR(rdev));
5682                         return PTR_ERR(rdev);
5683                 }
5684                 if (!list_empty(&mddev->disks)) {
5685                         struct md_rdev *rdev0
5686                                 = list_entry(mddev->disks.next,
5687                                              struct md_rdev, same_set);
5688                         err = super_types[mddev->major_version]
5689                                 .load_super(rdev, rdev0, mddev->minor_version);
5690                         if (err < 0) {
5691                                 printk(KERN_WARNING
5692                                         "md: %s has different UUID to %s\n",
5693                                         bdevname(rdev->bdev,b),
5694                                         bdevname(rdev0->bdev,b2));
5695                                 export_rdev(rdev);
5696                                 return -EINVAL;
5697                         }
5698                 }
5699                 err = bind_rdev_to_array(rdev, mddev);
5700                 if (err)
5701                         export_rdev(rdev);
5702                 return err;
5703         }
5704
5705         /*
5706          * add_new_disk can be used once the array is assembled
5707          * to add "hot spares".  They must already have a superblock
5708          * written
5709          */
5710         if (mddev->pers) {
5711                 int err;
5712                 if (!mddev->pers->hot_add_disk) {
5713                         printk(KERN_WARNING
5714                                 "%s: personality does not support diskops!\n",
5715                                mdname(mddev));
5716                         return -EINVAL;
5717                 }
5718                 if (mddev->persistent)
5719                         rdev = md_import_device(dev, mddev->major_version,
5720                                                 mddev->minor_version);
5721                 else
5722                         rdev = md_import_device(dev, -1, -1);
5723                 if (IS_ERR(rdev)) {
5724                         printk(KERN_WARNING
5725                                 "md: md_import_device returned %ld\n",
5726                                 PTR_ERR(rdev));
5727                         return PTR_ERR(rdev);
5728                 }
5729                 /* set saved_raid_disk if appropriate */
5730                 if (!mddev->persistent) {
5731                         if (info->state & (1<<MD_DISK_SYNC)  &&
5732                             info->raid_disk < mddev->raid_disks) {
5733                                 rdev->raid_disk = info->raid_disk;
5734                                 set_bit(In_sync, &rdev->flags);
5735                         } else
5736                                 rdev->raid_disk = -1;
5737                 } else
5738                         super_types[mddev->major_version].
5739                                 validate_super(mddev, rdev);
5740                 if ((info->state & (1<<MD_DISK_SYNC)) &&
5741                      rdev->raid_disk != info->raid_disk) {
5742                         /* This was a hot-add request, but events doesn't
5743                          * match, so reject it.
5744                          */
5745                         export_rdev(rdev);
5746                         return -EINVAL;
5747                 }
5748
5749                 if (test_bit(In_sync, &rdev->flags))
5750                         rdev->saved_raid_disk = rdev->raid_disk;
5751                 else
5752                         rdev->saved_raid_disk = -1;
5753
5754                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5755                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5756                         set_bit(WriteMostly, &rdev->flags);
5757                 else
5758                         clear_bit(WriteMostly, &rdev->flags);
5759
5760                 rdev->raid_disk = -1;
5761                 err = bind_rdev_to_array(rdev, mddev);
5762                 if (!err && !mddev->pers->hot_remove_disk) {
5763                         /* If there is hot_add_disk but no hot_remove_disk
5764                          * then added disks for geometry changes,
5765                          * and should be added immediately.
5766                          */
5767                         super_types[mddev->major_version].
5768                                 validate_super(mddev, rdev);
5769                         err = mddev->pers->hot_add_disk(mddev, rdev);
5770                         if (err)
5771                                 unbind_rdev_from_array(rdev);
5772                 }
5773                 if (err)
5774                         export_rdev(rdev);
5775                 else
5776                         sysfs_notify_dirent_safe(rdev->sysfs_state);
5777
5778                 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5779                 if (mddev->degraded)
5780                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5781                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5782                 if (!err)
5783                         md_new_event(mddev);
5784                 md_wakeup_thread(mddev->thread);
5785                 return err;
5786         }
5787
5788         /* otherwise, add_new_disk is only allowed
5789          * for major_version==0 superblocks
5790          */
5791         if (mddev->major_version != 0) {
5792                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5793                        mdname(mddev));
5794                 return -EINVAL;
5795         }
5796
5797         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5798                 int err;
5799                 rdev = md_import_device(dev, -1, 0);
5800                 if (IS_ERR(rdev)) {
5801                         printk(KERN_WARNING
5802                                 "md: error, md_import_device() returned %ld\n",
5803                                 PTR_ERR(rdev));
5804                         return PTR_ERR(rdev);
5805                 }
5806                 rdev->desc_nr = info->number;
5807                 if (info->raid_disk < mddev->raid_disks)
5808                         rdev->raid_disk = info->raid_disk;
5809                 else
5810                         rdev->raid_disk = -1;
5811
5812                 if (rdev->raid_disk < mddev->raid_disks)
5813                         if (info->state & (1<<MD_DISK_SYNC))
5814                                 set_bit(In_sync, &rdev->flags);
5815
5816                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5817                         set_bit(WriteMostly, &rdev->flags);
5818
5819                 if (!mddev->persistent) {
5820                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5821                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5822                 } else
5823                         rdev->sb_start = calc_dev_sboffset(rdev);
5824                 rdev->sectors = rdev->sb_start;
5825
5826                 err = bind_rdev_to_array(rdev, mddev);
5827                 if (err) {
5828                         export_rdev(rdev);
5829                         return err;
5830                 }
5831         }
5832
5833         return 0;
5834 }
5835
5836 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5837 {
5838         char b[BDEVNAME_SIZE];
5839         struct md_rdev *rdev;
5840
5841         rdev = find_rdev(mddev, dev);
5842         if (!rdev)
5843                 return -ENXIO;
5844
5845         clear_bit(Blocked, &rdev->flags);
5846         remove_and_add_spares(mddev, rdev);
5847
5848         if (rdev->raid_disk >= 0)
5849                 goto busy;
5850
5851         kick_rdev_from_array(rdev);
5852         md_update_sb(mddev, 1);
5853         md_new_event(mddev);
5854
5855         return 0;
5856 busy:
5857         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5858                 bdevname(rdev->bdev,b), mdname(mddev));
5859         return -EBUSY;
5860 }
5861
5862 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5863 {
5864         char b[BDEVNAME_SIZE];
5865         int err;
5866         struct md_rdev *rdev;
5867
5868         if (!mddev->pers)
5869                 return -ENODEV;
5870
5871         if (mddev->major_version != 0) {
5872                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5873                         " version-0 superblocks.\n",
5874                         mdname(mddev));
5875                 return -EINVAL;
5876         }
5877         if (!mddev->pers->hot_add_disk) {
5878                 printk(KERN_WARNING
5879                         "%s: personality does not support diskops!\n",
5880                         mdname(mddev));
5881                 return -EINVAL;
5882         }
5883
5884         rdev = md_import_device(dev, -1, 0);
5885         if (IS_ERR(rdev)) {
5886                 printk(KERN_WARNING
5887                         "md: error, md_import_device() returned %ld\n",
5888                         PTR_ERR(rdev));
5889                 return -EINVAL;
5890         }
5891
5892         if (mddev->persistent)
5893                 rdev->sb_start = calc_dev_sboffset(rdev);
5894         else
5895                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5896
5897         rdev->sectors = rdev->sb_start;
5898
5899         if (test_bit(Faulty, &rdev->flags)) {
5900                 printk(KERN_WARNING
5901                         "md: can not hot-add faulty %s disk to %s!\n",
5902                         bdevname(rdev->bdev,b), mdname(mddev));
5903                 err = -EINVAL;
5904                 goto abort_export;
5905         }
5906         clear_bit(In_sync, &rdev->flags);
5907         rdev->desc_nr = -1;
5908         rdev->saved_raid_disk = -1;
5909         err = bind_rdev_to_array(rdev, mddev);
5910         if (err)
5911                 goto abort_export;
5912
5913         /*
5914          * The rest should better be atomic, we can have disk failures
5915          * noticed in interrupt contexts ...
5916          */
5917
5918         rdev->raid_disk = -1;
5919
5920         md_update_sb(mddev, 1);
5921
5922         /*
5923          * Kick recovery, maybe this spare has to be added to the
5924          * array immediately.
5925          */
5926         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5927         md_wakeup_thread(mddev->thread);
5928         md_new_event(mddev);
5929         return 0;
5930
5931 abort_export:
5932         export_rdev(rdev);
5933         return err;
5934 }
5935
5936 static int set_bitmap_file(struct mddev *mddev, int fd)
5937 {
5938         int err;
5939
5940         if (mddev->pers) {
5941                 if (!mddev->pers->quiesce)
5942                         return -EBUSY;
5943                 if (mddev->recovery || mddev->sync_thread)
5944                         return -EBUSY;
5945                 /* we should be able to change the bitmap.. */
5946         }
5947
5948
5949         if (fd >= 0) {
5950                 if (mddev->bitmap)
5951                         return -EEXIST; /* cannot add when bitmap is present */
5952                 mddev->bitmap_info.file = fget(fd);
5953
5954                 if (mddev->bitmap_info.file == NULL) {
5955                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5956                                mdname(mddev));
5957                         return -EBADF;
5958                 }
5959
5960                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5961                 if (err) {
5962                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5963                                mdname(mddev));
5964                         fput(mddev->bitmap_info.file);
5965                         mddev->bitmap_info.file = NULL;
5966                         return err;
5967                 }
5968                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5969         } else if (mddev->bitmap == NULL)
5970                 return -ENOENT; /* cannot remove what isn't there */
5971         err = 0;
5972         if (mddev->pers) {
5973                 mddev->pers->quiesce(mddev, 1);
5974                 if (fd >= 0) {
5975                         err = bitmap_create(mddev);
5976                         if (!err)
5977                                 err = bitmap_load(mddev);
5978                 }
5979                 if (fd < 0 || err) {
5980                         bitmap_destroy(mddev);
5981                         fd = -1; /* make sure to put the file */
5982                 }
5983                 mddev->pers->quiesce(mddev, 0);
5984         }
5985         if (fd < 0) {
5986                 if (mddev->bitmap_info.file) {
5987                         restore_bitmap_write_access(mddev->bitmap_info.file);
5988                         fput(mddev->bitmap_info.file);
5989                 }
5990                 mddev->bitmap_info.file = NULL;
5991         }
5992
5993         return err;
5994 }
5995
5996 /*
5997  * set_array_info is used two different ways
5998  * The original usage is when creating a new array.
5999  * In this usage, raid_disks is > 0 and it together with
6000  *  level, size, not_persistent,layout,chunksize determine the
6001  *  shape of the array.
6002  *  This will always create an array with a type-0.90.0 superblock.
6003  * The newer usage is when assembling an array.
6004  *  In this case raid_disks will be 0, and the major_version field is
6005  *  use to determine which style super-blocks are to be found on the devices.
6006  *  The minor and patch _version numbers are also kept incase the
6007  *  super_block handler wishes to interpret them.
6008  */
6009 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6010 {
6011
6012         if (info->raid_disks == 0) {
6013                 /* just setting version number for superblock loading */
6014                 if (info->major_version < 0 ||
6015                     info->major_version >= ARRAY_SIZE(super_types) ||
6016                     super_types[info->major_version].name == NULL) {
6017                         /* maybe try to auto-load a module? */
6018                         printk(KERN_INFO
6019                                 "md: superblock version %d not known\n",
6020                                 info->major_version);
6021                         return -EINVAL;
6022                 }
6023                 mddev->major_version = info->major_version;
6024                 mddev->minor_version = info->minor_version;
6025                 mddev->patch_version = info->patch_version;
6026                 mddev->persistent = !info->not_persistent;
6027                 /* ensure mddev_put doesn't delete this now that there
6028                  * is some minimal configuration.
6029                  */
6030                 mddev->ctime         = get_seconds();
6031                 return 0;
6032         }
6033         mddev->major_version = MD_MAJOR_VERSION;
6034         mddev->minor_version = MD_MINOR_VERSION;
6035         mddev->patch_version = MD_PATCHLEVEL_VERSION;
6036         mddev->ctime         = get_seconds();
6037
6038         mddev->level         = info->level;
6039         mddev->clevel[0]     = 0;
6040         mddev->dev_sectors   = 2 * (sector_t)info->size;
6041         mddev->raid_disks    = info->raid_disks;
6042         /* don't set md_minor, it is determined by which /dev/md* was
6043          * openned
6044          */
6045         if (info->state & (1<<MD_SB_CLEAN))
6046                 mddev->recovery_cp = MaxSector;
6047         else
6048                 mddev->recovery_cp = 0;
6049         mddev->persistent    = ! info->not_persistent;
6050         mddev->external      = 0;
6051
6052         mddev->layout        = info->layout;
6053         mddev->chunk_sectors = info->chunk_size >> 9;
6054
6055         mddev->max_disks     = MD_SB_DISKS;
6056
6057         if (mddev->persistent)
6058                 mddev->flags         = 0;
6059         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6060
6061         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6062         mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6063         mddev->bitmap_info.offset = 0;
6064
6065         mddev->reshape_position = MaxSector;
6066
6067         /*
6068          * Generate a 128 bit UUID
6069          */
6070         get_random_bytes(mddev->uuid, 16);
6071
6072         mddev->new_level = mddev->level;
6073         mddev->new_chunk_sectors = mddev->chunk_sectors;
6074         mddev->new_layout = mddev->layout;
6075         mddev->delta_disks = 0;
6076         mddev->reshape_backwards = 0;
6077
6078         return 0;
6079 }
6080
6081 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6082 {
6083         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6084
6085         if (mddev->external_size)
6086                 return;
6087
6088         mddev->array_sectors = array_sectors;
6089 }
6090 EXPORT_SYMBOL(md_set_array_sectors);
6091
6092 static int update_size(struct mddev *mddev, sector_t num_sectors)
6093 {
6094         struct md_rdev *rdev;
6095         int rv;
6096         int fit = (num_sectors == 0);
6097
6098         if (mddev->pers->resize == NULL)
6099                 return -EINVAL;
6100         /* The "num_sectors" is the number of sectors of each device that
6101          * is used.  This can only make sense for arrays with redundancy.
6102          * linear and raid0 always use whatever space is available. We can only
6103          * consider changing this number if no resync or reconstruction is
6104          * happening, and if the new size is acceptable. It must fit before the
6105          * sb_start or, if that is <data_offset, it must fit before the size
6106          * of each device.  If num_sectors is zero, we find the largest size
6107          * that fits.
6108          */
6109         if (mddev->sync_thread)
6110                 return -EBUSY;
6111
6112         rdev_for_each(rdev, mddev) {
6113                 sector_t avail = rdev->sectors;
6114
6115                 if (fit && (num_sectors == 0 || num_sectors > avail))
6116                         num_sectors = avail;
6117                 if (avail < num_sectors)
6118                         return -ENOSPC;
6119         }
6120         rv = mddev->pers->resize(mddev, num_sectors);
6121         if (!rv)
6122                 revalidate_disk(mddev->gendisk);
6123         return rv;
6124 }
6125
6126 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6127 {
6128         int rv;
6129         struct md_rdev *rdev;
6130         /* change the number of raid disks */
6131         if (mddev->pers->check_reshape == NULL)
6132                 return -EINVAL;
6133         if (raid_disks <= 0 ||
6134             (mddev->max_disks && raid_disks >= mddev->max_disks))
6135                 return -EINVAL;
6136         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6137                 return -EBUSY;
6138
6139         rdev_for_each(rdev, mddev) {
6140                 if (mddev->raid_disks < raid_disks &&
6141                     rdev->data_offset < rdev->new_data_offset)
6142                         return -EINVAL;
6143                 if (mddev->raid_disks > raid_disks &&
6144                     rdev->data_offset > rdev->new_data_offset)
6145                         return -EINVAL;
6146         }
6147
6148         mddev->delta_disks = raid_disks - mddev->raid_disks;
6149         if (mddev->delta_disks < 0)
6150                 mddev->reshape_backwards = 1;
6151         else if (mddev->delta_disks > 0)
6152                 mddev->reshape_backwards = 0;
6153
6154         rv = mddev->pers->check_reshape(mddev);
6155         if (rv < 0) {
6156                 mddev->delta_disks = 0;
6157                 mddev->reshape_backwards = 0;
6158         }
6159         return rv;
6160 }
6161
6162
6163 /*
6164  * update_array_info is used to change the configuration of an
6165  * on-line array.
6166  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6167  * fields in the info are checked against the array.
6168  * Any differences that cannot be handled will cause an error.
6169  * Normally, only one change can be managed at a time.
6170  */
6171 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6172 {
6173         int rv = 0;
6174         int cnt = 0;
6175         int state = 0;
6176
6177         /* calculate expected state,ignoring low bits */
6178         if (mddev->bitmap && mddev->bitmap_info.offset)
6179                 state |= (1 << MD_SB_BITMAP_PRESENT);
6180
6181         if (mddev->major_version != info->major_version ||
6182             mddev->minor_version != info->minor_version ||
6183 /*          mddev->patch_version != info->patch_version || */
6184             mddev->ctime         != info->ctime         ||
6185             mddev->level         != info->level         ||
6186 /*          mddev->layout        != info->layout        || */
6187             !mddev->persistent   != info->not_persistent||
6188             mddev->chunk_sectors != info->chunk_size >> 9 ||
6189             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6190             ((state^info->state) & 0xfffffe00)
6191                 )
6192                 return -EINVAL;
6193         /* Check there is only one change */
6194         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6195                 cnt++;
6196         if (mddev->raid_disks != info->raid_disks)
6197                 cnt++;
6198         if (mddev->layout != info->layout)
6199                 cnt++;
6200         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6201                 cnt++;
6202         if (cnt == 0)
6203                 return 0;
6204         if (cnt > 1)
6205                 return -EINVAL;
6206
6207         if (mddev->layout != info->layout) {
6208                 /* Change layout
6209                  * we don't need to do anything at the md level, the
6210                  * personality will take care of it all.
6211                  */
6212                 if (mddev->pers->check_reshape == NULL)
6213                         return -EINVAL;
6214                 else {
6215                         mddev->new_layout = info->layout;
6216                         rv = mddev->pers->check_reshape(mddev);
6217                         if (rv)
6218                                 mddev->new_layout = mddev->layout;
6219                         return rv;
6220                 }
6221         }
6222         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6223                 rv = update_size(mddev, (sector_t)info->size * 2);
6224
6225         if (mddev->raid_disks    != info->raid_disks)
6226                 rv = update_raid_disks(mddev, info->raid_disks);
6227
6228         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6229                 if (mddev->pers->quiesce == NULL)
6230                         return -EINVAL;
6231                 if (mddev->recovery || mddev->sync_thread)
6232                         return -EBUSY;
6233                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6234                         /* add the bitmap */
6235                         if (mddev->bitmap)
6236                                 return -EEXIST;
6237                         if (mddev->bitmap_info.default_offset == 0)
6238                                 return -EINVAL;
6239                         mddev->bitmap_info.offset =
6240                                 mddev->bitmap_info.default_offset;
6241                         mddev->bitmap_info.space =
6242                                 mddev->bitmap_info.default_space;
6243                         mddev->pers->quiesce(mddev, 1);
6244                         rv = bitmap_create(mddev);
6245                         if (!rv)
6246                                 rv = bitmap_load(mddev);
6247                         if (rv)
6248                                 bitmap_destroy(mddev);
6249                         mddev->pers->quiesce(mddev, 0);
6250                 } else {
6251                         /* remove the bitmap */
6252                         if (!mddev->bitmap)
6253                                 return -ENOENT;
6254                         if (mddev->bitmap->storage.file)
6255                                 return -EINVAL;
6256                         mddev->pers->quiesce(mddev, 1);
6257                         bitmap_destroy(mddev);
6258                         mddev->pers->quiesce(mddev, 0);
6259                         mddev->bitmap_info.offset = 0;
6260                 }
6261         }
6262         md_update_sb(mddev, 1);
6263         return rv;
6264 }
6265
6266 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6267 {
6268         struct md_rdev *rdev;
6269         int err = 0;
6270
6271         if (mddev->pers == NULL)
6272                 return -ENODEV;
6273
6274         rcu_read_lock();
6275         rdev = find_rdev_rcu(mddev, dev);
6276         if (!rdev)
6277                 err =  -ENODEV;
6278         else {
6279                 md_error(mddev, rdev);
6280                 if (!test_bit(Faulty, &rdev->flags))
6281                         err = -EBUSY;
6282         }
6283         rcu_read_unlock();
6284         return err;
6285 }
6286
6287 /*
6288  * We have a problem here : there is no easy way to give a CHS
6289  * virtual geometry. We currently pretend that we have a 2 heads
6290  * 4 sectors (with a BIG number of cylinders...). This drives
6291  * dosfs just mad... ;-)
6292  */
6293 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6294 {
6295         struct mddev *mddev = bdev->bd_disk->private_data;
6296
6297         geo->heads = 2;
6298         geo->sectors = 4;
6299         geo->cylinders = mddev->array_sectors / 8;
6300         return 0;
6301 }
6302
6303 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6304                         unsigned int cmd, unsigned long arg)
6305 {
6306         int err = 0;
6307         void __user *argp = (void __user *)arg;
6308         struct mddev *mddev = NULL;
6309         int ro;
6310
6311         switch (cmd) {
6312         case RAID_VERSION:
6313         case GET_ARRAY_INFO:
6314         case GET_DISK_INFO:
6315                 break;
6316         default:
6317                 if (!capable(CAP_SYS_ADMIN))
6318                         return -EACCES;
6319         }
6320
6321         /*
6322          * Commands dealing with the RAID driver but not any
6323          * particular array:
6324          */
6325         switch (cmd) {
6326         case RAID_VERSION:
6327                 err = get_version(argp);
6328                 goto done;
6329
6330         case PRINT_RAID_DEBUG:
6331                 err = 0;
6332                 md_print_devices();
6333                 goto done;
6334
6335 #ifndef MODULE
6336         case RAID_AUTORUN:
6337                 err = 0;
6338                 autostart_arrays(arg);
6339                 goto done;
6340 #endif
6341         default:;
6342         }
6343
6344         /*
6345          * Commands creating/starting a new array:
6346          */
6347
6348         mddev = bdev->bd_disk->private_data;
6349
6350         if (!mddev) {
6351                 BUG();
6352                 goto abort;
6353         }
6354
6355         /* Some actions do not requires the mutex */
6356         switch (cmd) {
6357         case GET_ARRAY_INFO:
6358                 if (!mddev->raid_disks && !mddev->external)
6359                         err = -ENODEV;
6360                 else
6361                         err = get_array_info(mddev, argp);
6362                 goto abort;
6363
6364         case GET_DISK_INFO:
6365                 if (!mddev->raid_disks && !mddev->external)
6366                         err = -ENODEV;
6367                 else
6368                         err = get_disk_info(mddev, argp);
6369                 goto abort;
6370
6371         case SET_DISK_FAULTY:
6372                 err = set_disk_faulty(mddev, new_decode_dev(arg));
6373                 goto abort;
6374         }
6375
6376         if (cmd == ADD_NEW_DISK)
6377                 /* need to ensure md_delayed_delete() has completed */
6378                 flush_workqueue(md_misc_wq);
6379
6380         if (cmd == HOT_REMOVE_DISK)
6381                 /* need to ensure recovery thread has run */
6382                 wait_event_interruptible_timeout(mddev->sb_wait,
6383                                                  !test_bit(MD_RECOVERY_NEEDED,
6384                                                            &mddev->flags),
6385                                                  msecs_to_jiffies(5000));
6386         if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6387                 /* Need to flush page cache, and ensure no-one else opens
6388                  * and writes
6389                  */
6390                 mutex_lock(&mddev->open_mutex);
6391                 if (atomic_read(&mddev->openers) > 1) {
6392                         mutex_unlock(&mddev->open_mutex);
6393                         err = -EBUSY;
6394                         goto abort;
6395                 }
6396                 set_bit(MD_STILL_CLOSED, &mddev->flags);
6397                 mutex_unlock(&mddev->open_mutex);
6398                 sync_blockdev(bdev);
6399         }
6400         err = mddev_lock(mddev);
6401         if (err) {
6402                 printk(KERN_INFO
6403                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
6404                         err, cmd);
6405                 goto abort;
6406         }
6407
6408         if (cmd == SET_ARRAY_INFO) {
6409                 mdu_array_info_t info;
6410                 if (!arg)
6411                         memset(&info, 0, sizeof(info));
6412                 else if (copy_from_user(&info, argp, sizeof(info))) {
6413                         err = -EFAULT;
6414                         goto abort_unlock;
6415                 }
6416                 if (mddev->pers) {
6417                         err = update_array_info(mddev, &info);
6418                         if (err) {
6419                                 printk(KERN_WARNING "md: couldn't update"
6420                                        " array info. %d\n", err);
6421                                 goto abort_unlock;
6422                         }
6423                         goto done_unlock;
6424                 }
6425                 if (!list_empty(&mddev->disks)) {
6426                         printk(KERN_WARNING
6427                                "md: array %s already has disks!\n",
6428                                mdname(mddev));
6429                         err = -EBUSY;
6430                         goto abort_unlock;
6431                 }
6432                 if (mddev->raid_disks) {
6433                         printk(KERN_WARNING
6434                                "md: array %s already initialised!\n",
6435                                mdname(mddev));
6436                         err = -EBUSY;
6437                         goto abort_unlock;
6438                 }
6439                 err = set_array_info(mddev, &info);
6440                 if (err) {
6441                         printk(KERN_WARNING "md: couldn't set"
6442                                " array info. %d\n", err);
6443                         goto abort_unlock;
6444                 }
6445                 goto done_unlock;
6446         }
6447
6448         /*
6449          * Commands querying/configuring an existing array:
6450          */
6451         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6452          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6453         if ((!mddev->raid_disks && !mddev->external)
6454             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6455             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6456             && cmd != GET_BITMAP_FILE) {
6457                 err = -ENODEV;
6458                 goto abort_unlock;
6459         }
6460
6461         /*
6462          * Commands even a read-only array can execute:
6463          */
6464         switch (cmd) {
6465         case GET_BITMAP_FILE:
6466                 err = get_bitmap_file(mddev, argp);
6467                 goto done_unlock;
6468
6469         case RESTART_ARRAY_RW:
6470                 err = restart_array(mddev);
6471                 goto done_unlock;
6472
6473         case STOP_ARRAY:
6474                 err = do_md_stop(mddev, 0, bdev);
6475                 goto done_unlock;
6476
6477         case STOP_ARRAY_RO:
6478                 err = md_set_readonly(mddev, bdev);
6479                 goto done_unlock;
6480
6481         case HOT_REMOVE_DISK:
6482                 err = hot_remove_disk(mddev, new_decode_dev(arg));
6483                 goto done_unlock;
6484
6485         case ADD_NEW_DISK:
6486                 /* We can support ADD_NEW_DISK on read-only arrays
6487                  * on if we are re-adding a preexisting device.
6488                  * So require mddev->pers and MD_DISK_SYNC.
6489                  */
6490                 if (mddev->pers) {
6491                         mdu_disk_info_t info;
6492                         if (copy_from_user(&info, argp, sizeof(info)))
6493                                 err = -EFAULT;
6494                         else if (!(info.state & (1<<MD_DISK_SYNC)))
6495                                 /* Need to clear read-only for this */
6496                                 break;
6497                         else
6498                                 err = add_new_disk(mddev, &info);
6499                         goto done_unlock;
6500                 }
6501                 break;
6502
6503         case BLKROSET:
6504                 if (get_user(ro, (int __user *)(arg))) {
6505                         err = -EFAULT;
6506                         goto done_unlock;
6507                 }
6508                 err = -EINVAL;
6509
6510                 /* if the bdev is going readonly the value of mddev->ro
6511                  * does not matter, no writes are coming
6512                  */
6513                 if (ro)
6514                         goto done_unlock;
6515
6516                 /* are we are already prepared for writes? */
6517                 if (mddev->ro != 1)
6518                         goto done_unlock;
6519
6520                 /* transitioning to readauto need only happen for
6521                  * arrays that call md_write_start
6522                  */
6523                 if (mddev->pers) {
6524                         err = restart_array(mddev);
6525                         if (err == 0) {
6526                                 mddev->ro = 2;
6527                                 set_disk_ro(mddev->gendisk, 0);
6528                         }
6529                 }
6530                 goto done_unlock;
6531         }
6532
6533         /*
6534          * The remaining ioctls are changing the state of the
6535          * superblock, so we do not allow them on read-only arrays.
6536          * However non-MD ioctls (e.g. get-size) will still come through
6537          * here and hit the 'default' below, so only disallow
6538          * 'md' ioctls, and switch to rw mode if started auto-readonly.
6539          */
6540         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6541                 if (mddev->ro == 2) {
6542                         mddev->ro = 0;
6543                         sysfs_notify_dirent_safe(mddev->sysfs_state);
6544                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6545                         /* mddev_unlock will wake thread */
6546                         /* If a device failed while we were read-only, we
6547                          * need to make sure the metadata is updated now.
6548                          */
6549                         if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6550                                 mddev_unlock(mddev);
6551                                 wait_event(mddev->sb_wait,
6552                                            !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6553                                            !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6554                                 mddev_lock(mddev);
6555                         }
6556                 } else {
6557                         err = -EROFS;
6558                         goto abort_unlock;
6559                 }
6560         }
6561
6562         switch (cmd) {
6563         case ADD_NEW_DISK:
6564         {
6565                 mdu_disk_info_t info;
6566                 if (copy_from_user(&info, argp, sizeof(info)))
6567                         err = -EFAULT;
6568                 else
6569                         err = add_new_disk(mddev, &info);
6570                 goto done_unlock;
6571         }
6572
6573         case HOT_ADD_DISK:
6574                 err = hot_add_disk(mddev, new_decode_dev(arg));
6575                 goto done_unlock;
6576
6577         case RUN_ARRAY:
6578                 err = do_md_run(mddev);
6579                 goto done_unlock;
6580
6581         case SET_BITMAP_FILE:
6582                 err = set_bitmap_file(mddev, (int)arg);
6583                 goto done_unlock;
6584
6585         default:
6586                 err = -EINVAL;
6587                 goto abort_unlock;
6588         }
6589
6590 done_unlock:
6591 abort_unlock:
6592         if (mddev->hold_active == UNTIL_IOCTL &&
6593             err != -EINVAL)
6594                 mddev->hold_active = 0;
6595         mddev_unlock(mddev);
6596
6597         return err;
6598 done:
6599         if (err)
6600                 MD_BUG();
6601 abort:
6602         return err;
6603 }
6604 #ifdef CONFIG_COMPAT
6605 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6606                     unsigned int cmd, unsigned long arg)
6607 {
6608         switch (cmd) {
6609         case HOT_REMOVE_DISK:
6610         case HOT_ADD_DISK:
6611         case SET_DISK_FAULTY:
6612         case SET_BITMAP_FILE:
6613                 /* These take in integer arg, do not convert */
6614                 break;
6615         default:
6616                 arg = (unsigned long)compat_ptr(arg);
6617                 break;
6618         }
6619
6620         return md_ioctl(bdev, mode, cmd, arg);
6621 }
6622 #endif /* CONFIG_COMPAT */
6623
6624 static int md_open(struct block_device *bdev, fmode_t mode)
6625 {
6626         /*
6627          * Succeed if we can lock the mddev, which confirms that
6628          * it isn't being stopped right now.
6629          */
6630         struct mddev *mddev = mddev_find(bdev->bd_dev);
6631         int err;
6632
6633         if (!mddev)
6634                 return -ENODEV;
6635
6636         if (mddev->gendisk != bdev->bd_disk) {
6637                 /* we are racing with mddev_put which is discarding this
6638                  * bd_disk.
6639                  */
6640                 mddev_put(mddev);
6641                 /* Wait until bdev->bd_disk is definitely gone */
6642                 flush_workqueue(md_misc_wq);
6643                 /* Then retry the open from the top */
6644                 return -ERESTARTSYS;
6645         }
6646         BUG_ON(mddev != bdev->bd_disk->private_data);
6647
6648         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6649                 goto out;
6650
6651         err = 0;
6652         atomic_inc(&mddev->openers);
6653         clear_bit(MD_STILL_CLOSED, &mddev->flags);
6654         mutex_unlock(&mddev->open_mutex);
6655
6656         check_disk_change(bdev);
6657  out:
6658         return err;
6659 }
6660
6661 static void md_release(struct gendisk *disk, fmode_t mode)
6662 {
6663         struct mddev *mddev = disk->private_data;
6664
6665         BUG_ON(!mddev);
6666         atomic_dec(&mddev->openers);
6667         mddev_put(mddev);
6668 }
6669
6670 static int md_media_changed(struct gendisk *disk)
6671 {
6672         struct mddev *mddev = disk->private_data;
6673
6674         return mddev->changed;
6675 }
6676
6677 static int md_revalidate(struct gendisk *disk)
6678 {
6679         struct mddev *mddev = disk->private_data;
6680
6681         mddev->changed = 0;
6682         return 0;
6683 }
6684 static const struct block_device_operations md_fops =
6685 {
6686         .owner          = THIS_MODULE,
6687         .open           = md_open,
6688         .release        = md_release,
6689         .ioctl          = md_ioctl,
6690 #ifdef CONFIG_COMPAT
6691         .compat_ioctl   = md_compat_ioctl,
6692 #endif
6693         .getgeo         = md_getgeo,
6694         .media_changed  = md_media_changed,
6695         .revalidate_disk= md_revalidate,
6696 };
6697
6698 static int md_thread(void * arg)
6699 {
6700         struct md_thread *thread = arg;
6701
6702         /*
6703          * md_thread is a 'system-thread', it's priority should be very
6704          * high. We avoid resource deadlocks individually in each
6705          * raid personality. (RAID5 does preallocation) We also use RR and
6706          * the very same RT priority as kswapd, thus we will never get
6707          * into a priority inversion deadlock.
6708          *
6709          * we definitely have to have equal or higher priority than
6710          * bdflush, otherwise bdflush will deadlock if there are too
6711          * many dirty RAID5 blocks.
6712          */
6713
6714         allow_signal(SIGKILL);
6715         while (!kthread_should_stop()) {
6716
6717                 /* We need to wait INTERRUPTIBLE so that
6718                  * we don't add to the load-average.
6719                  * That means we need to be sure no signals are
6720                  * pending
6721                  */
6722                 if (signal_pending(current))
6723                         flush_signals(current);
6724
6725                 wait_event_interruptible_timeout
6726                         (thread->wqueue,
6727                          test_bit(THREAD_WAKEUP, &thread->flags)
6728                          || kthread_should_stop(),
6729                          thread->timeout);
6730
6731                 clear_bit(THREAD_WAKEUP, &thread->flags);
6732                 if (!kthread_should_stop())
6733                         thread->run(thread);
6734         }
6735
6736         return 0;
6737 }
6738
6739 void md_wakeup_thread(struct md_thread *thread)
6740 {
6741         if (thread) {
6742                 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6743                 set_bit(THREAD_WAKEUP, &thread->flags);
6744                 wake_up(&thread->wqueue);
6745         }
6746 }
6747
6748 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6749                 struct mddev *mddev, const char *name)
6750 {
6751         struct md_thread *thread;
6752
6753         thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6754         if (!thread)
6755                 return NULL;
6756
6757         init_waitqueue_head(&thread->wqueue);
6758
6759         thread->run = run;
6760         thread->mddev = mddev;
6761         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6762         thread->tsk = kthread_run(md_thread, thread,
6763                                   "%s_%s",
6764                                   mdname(thread->mddev),
6765                                   name);
6766         if (IS_ERR(thread->tsk)) {
6767                 kfree(thread);
6768                 return NULL;
6769         }
6770         return thread;
6771 }
6772
6773 void md_unregister_thread(struct md_thread **threadp)
6774 {
6775         struct md_thread *thread = *threadp;
6776         if (!thread)
6777                 return;
6778         pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6779         /* Locking ensures that mddev_unlock does not wake_up a
6780          * non-existent thread
6781          */
6782         spin_lock(&pers_lock);
6783         *threadp = NULL;
6784         spin_unlock(&pers_lock);
6785
6786         kthread_stop(thread->tsk);
6787         kfree(thread);
6788 }
6789
6790 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6791 {
6792         if (!mddev) {
6793                 MD_BUG();
6794                 return;
6795         }
6796
6797         if (!rdev || test_bit(Faulty, &rdev->flags))
6798                 return;
6799
6800         if (!mddev->pers || !mddev->pers->error_handler)
6801                 return;
6802         mddev->pers->error_handler(mddev,rdev);
6803         if (mddev->degraded)
6804                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6805         sysfs_notify_dirent_safe(rdev->sysfs_state);
6806         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6807         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6808         md_wakeup_thread(mddev->thread);
6809         if (mddev->event_work.func)
6810                 queue_work(md_misc_wq, &mddev->event_work);
6811         md_new_event_inintr(mddev);
6812 }
6813
6814 /* seq_file implementation /proc/mdstat */
6815
6816 static void status_unused(struct seq_file *seq)
6817 {
6818         int i = 0;
6819         struct md_rdev *rdev;
6820
6821         seq_printf(seq, "unused devices: ");
6822
6823         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6824                 char b[BDEVNAME_SIZE];
6825                 i++;
6826                 seq_printf(seq, "%s ",
6827                               bdevname(rdev->bdev,b));
6828         }
6829         if (!i)
6830                 seq_printf(seq, "<none>");
6831
6832         seq_printf(seq, "\n");
6833 }
6834
6835
6836 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6837 {
6838         sector_t max_sectors, resync, res;
6839         unsigned long dt, db;
6840         sector_t rt;
6841         int scale;
6842         unsigned int per_milli;
6843
6844         if (mddev->curr_resync <= 3)
6845                 resync = 0;
6846         else
6847                 resync = mddev->curr_resync
6848                         - atomic_read(&mddev->recovery_active);
6849
6850         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6851             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6852                 max_sectors = mddev->resync_max_sectors;
6853         else
6854                 max_sectors = mddev->dev_sectors;
6855
6856         /*
6857          * Should not happen.
6858          */
6859         if (!max_sectors) {
6860                 MD_BUG();
6861                 return;
6862         }
6863         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6864          * in a sector_t, and (max_sectors>>scale) will fit in a
6865          * u32, as those are the requirements for sector_div.
6866          * Thus 'scale' must be at least 10
6867          */
6868         scale = 10;
6869         if (sizeof(sector_t) > sizeof(unsigned long)) {
6870                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6871                         scale++;
6872         }
6873         res = (resync>>scale)*1000;
6874         sector_div(res, (u32)((max_sectors>>scale)+1));
6875
6876         per_milli = res;
6877         {
6878                 int i, x = per_milli/50, y = 20-x;
6879                 seq_printf(seq, "[");
6880                 for (i = 0; i < x; i++)
6881                         seq_printf(seq, "=");
6882                 seq_printf(seq, ">");
6883                 for (i = 0; i < y; i++)
6884                         seq_printf(seq, ".");
6885                 seq_printf(seq, "] ");
6886         }
6887         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6888                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6889                     "reshape" :
6890                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6891                      "check" :
6892                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6893                       "resync" : "recovery"))),
6894                    per_milli/10, per_milli % 10,
6895                    (unsigned long long) resync/2,
6896                    (unsigned long long) max_sectors/2);
6897
6898         /*
6899          * dt: time from mark until now
6900          * db: blocks written from mark until now
6901          * rt: remaining time
6902          *
6903          * rt is a sector_t, so could be 32bit or 64bit.
6904          * So we divide before multiply in case it is 32bit and close
6905          * to the limit.
6906          * We scale the divisor (db) by 32 to avoid losing precision
6907          * near the end of resync when the number of remaining sectors
6908          * is close to 'db'.
6909          * We then divide rt by 32 after multiplying by db to compensate.
6910          * The '+1' avoids division by zero if db is very small.
6911          */
6912         dt = ((jiffies - mddev->resync_mark) / HZ);
6913         if (!dt) dt++;
6914         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6915                 - mddev->resync_mark_cnt;
6916
6917         rt = max_sectors - resync;    /* number of remaining sectors */
6918         sector_div(rt, db/32+1);
6919         rt *= dt;
6920         rt >>= 5;
6921
6922         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6923                    ((unsigned long)rt % 60)/6);
6924
6925         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6926 }
6927
6928 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6929 {
6930         struct list_head *tmp;
6931         loff_t l = *pos;
6932         struct mddev *mddev;
6933
6934         if (l >= 0x10000)
6935                 return NULL;
6936         if (!l--)
6937                 /* header */
6938                 return (void*)1;
6939
6940         spin_lock(&all_mddevs_lock);
6941         list_for_each(tmp,&all_mddevs)
6942                 if (!l--) {
6943                         mddev = list_entry(tmp, struct mddev, all_mddevs);
6944                         mddev_get(mddev);
6945                         spin_unlock(&all_mddevs_lock);
6946                         return mddev;
6947                 }
6948         spin_unlock(&all_mddevs_lock);
6949         if (!l--)
6950                 return (void*)2;/* tail */
6951         return NULL;
6952 }
6953
6954 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6955 {
6956         struct list_head *tmp;
6957         struct mddev *next_mddev, *mddev = v;
6958
6959         ++*pos;
6960         if (v == (void*)2)
6961                 return NULL;
6962
6963         spin_lock(&all_mddevs_lock);
6964         if (v == (void*)1)
6965                 tmp = all_mddevs.next;
6966         else
6967                 tmp = mddev->all_mddevs.next;
6968         if (tmp != &all_mddevs)
6969                 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6970         else {
6971                 next_mddev = (void*)2;
6972                 *pos = 0x10000;
6973         }
6974         spin_unlock(&all_mddevs_lock);
6975
6976         if (v != (void*)1)
6977                 mddev_put(mddev);
6978         return next_mddev;
6979
6980 }
6981
6982 static void md_seq_stop(struct seq_file *seq, void *v)
6983 {
6984         struct mddev *mddev = v;
6985
6986         if (mddev && v != (void*)1 && v != (void*)2)
6987                 mddev_put(mddev);
6988 }
6989
6990 static int md_seq_show(struct seq_file *seq, void *v)
6991 {
6992         struct mddev *mddev = v;
6993         sector_t sectors;
6994         struct md_rdev *rdev;
6995
6996         if (v == (void*)1) {
6997                 struct md_personality *pers;
6998                 seq_printf(seq, "Personalities : ");
6999                 spin_lock(&pers_lock);
7000                 list_for_each_entry(pers, &pers_list, list)
7001                         seq_printf(seq, "[%s] ", pers->name);
7002
7003                 spin_unlock(&pers_lock);
7004                 seq_printf(seq, "\n");
7005                 seq->poll_event = atomic_read(&md_event_count);
7006                 return 0;
7007         }
7008         if (v == (void*)2) {
7009                 status_unused(seq);
7010                 return 0;
7011         }
7012
7013         if (mddev_lock(mddev) < 0)
7014                 return -EINTR;
7015
7016         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7017                 seq_printf(seq, "%s : %sactive", mdname(mddev),
7018                                                 mddev->pers ? "" : "in");
7019                 if (mddev->pers) {
7020                         if (mddev->ro==1)
7021                                 seq_printf(seq, " (read-only)");
7022                         if (mddev->ro==2)
7023                                 seq_printf(seq, " (auto-read-only)");
7024                         seq_printf(seq, " %s", mddev->pers->name);
7025                 }
7026
7027                 sectors = 0;
7028                 rdev_for_each(rdev, mddev) {
7029                         char b[BDEVNAME_SIZE];
7030                         seq_printf(seq, " %s[%d]",
7031                                 bdevname(rdev->bdev,b), rdev->desc_nr);
7032                         if (test_bit(WriteMostly, &rdev->flags))
7033                                 seq_printf(seq, "(W)");
7034                         if (test_bit(Faulty, &rdev->flags)) {
7035                                 seq_printf(seq, "(F)");
7036                                 continue;
7037                         }
7038                         if (rdev->raid_disk < 0)
7039                                 seq_printf(seq, "(S)"); /* spare */
7040                         if (test_bit(Replacement, &rdev->flags))
7041                                 seq_printf(seq, "(R)");
7042                         sectors += rdev->sectors;
7043                 }
7044
7045                 if (!list_empty(&mddev->disks)) {
7046                         if (mddev->pers)
7047                                 seq_printf(seq, "\n      %llu blocks",
7048                                            (unsigned long long)
7049                                            mddev->array_sectors / 2);
7050                         else
7051                                 seq_printf(seq, "\n      %llu blocks",
7052                                            (unsigned long long)sectors / 2);
7053                 }
7054                 if (mddev->persistent) {
7055                         if (mddev->major_version != 0 ||
7056                             mddev->minor_version != 90) {
7057                                 seq_printf(seq," super %d.%d",
7058                                            mddev->major_version,
7059                                            mddev->minor_version);
7060                         }
7061                 } else if (mddev->external)
7062                         seq_printf(seq, " super external:%s",
7063                                    mddev->metadata_type);
7064                 else
7065                         seq_printf(seq, " super non-persistent");
7066
7067                 if (mddev->pers) {
7068                         mddev->pers->status(seq, mddev);
7069                         seq_printf(seq, "\n      ");
7070                         if (mddev->pers->sync_request) {
7071                                 if (mddev->curr_resync > 2) {
7072                                         status_resync(seq, mddev);
7073                                         seq_printf(seq, "\n      ");
7074                                 } else if (mddev->curr_resync >= 1)
7075                                         seq_printf(seq, "\tresync=DELAYED\n      ");
7076                                 else if (mddev->recovery_cp < MaxSector)
7077                                         seq_printf(seq, "\tresync=PENDING\n      ");
7078                         }
7079                 } else
7080                         seq_printf(seq, "\n       ");
7081
7082                 bitmap_status(seq, mddev->bitmap);
7083
7084                 seq_printf(seq, "\n");
7085         }
7086         mddev_unlock(mddev);
7087
7088         return 0;
7089 }
7090
7091 static const struct seq_operations md_seq_ops = {
7092         .start  = md_seq_start,
7093         .next   = md_seq_next,
7094         .stop   = md_seq_stop,
7095         .show   = md_seq_show,
7096 };
7097
7098 static int md_seq_open(struct inode *inode, struct file *file)
7099 {
7100         struct seq_file *seq;
7101         int error;
7102
7103         error = seq_open(file, &md_seq_ops);
7104         if (error)
7105                 return error;
7106
7107         seq = file->private_data;
7108         seq->poll_event = atomic_read(&md_event_count);
7109         return error;
7110 }
7111
7112 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7113 {
7114         struct seq_file *seq = filp->private_data;
7115         int mask;
7116
7117         poll_wait(filp, &md_event_waiters, wait);
7118
7119         /* always allow read */
7120         mask = POLLIN | POLLRDNORM;
7121
7122         if (seq->poll_event != atomic_read(&md_event_count))
7123                 mask |= POLLERR | POLLPRI;
7124         return mask;
7125 }
7126
7127 static const struct file_operations md_seq_fops = {
7128         .owner          = THIS_MODULE,
7129         .open           = md_seq_open,
7130         .read           = seq_read,
7131         .llseek         = seq_lseek,
7132         .release        = seq_release_private,
7133         .poll           = mdstat_poll,
7134 };
7135
7136 int register_md_personality(struct md_personality *p)
7137 {
7138         spin_lock(&pers_lock);
7139         list_add_tail(&p->list, &pers_list);
7140         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7141         spin_unlock(&pers_lock);
7142         return 0;
7143 }
7144
7145 int unregister_md_personality(struct md_personality *p)
7146 {
7147         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7148         spin_lock(&pers_lock);
7149         list_del_init(&p->list);
7150         spin_unlock(&pers_lock);
7151         return 0;
7152 }
7153
7154 static int is_mddev_idle(struct mddev *mddev, int init)
7155 {
7156         struct md_rdev * rdev;
7157         int idle;
7158         int curr_events;
7159
7160         idle = 1;
7161         rcu_read_lock();
7162         rdev_for_each_rcu(rdev, mddev) {
7163                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7164                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7165                               (int)part_stat_read(&disk->part0, sectors[1]) -
7166                               atomic_read(&disk->sync_io);
7167                 /* sync IO will cause sync_io to increase before the disk_stats
7168                  * as sync_io is counted when a request starts, and
7169                  * disk_stats is counted when it completes.
7170                  * So resync activity will cause curr_events to be smaller than
7171                  * when there was no such activity.
7172                  * non-sync IO will cause disk_stat to increase without
7173                  * increasing sync_io so curr_events will (eventually)
7174                  * be larger than it was before.  Once it becomes
7175                  * substantially larger, the test below will cause
7176                  * the array to appear non-idle, and resync will slow
7177                  * down.
7178                  * If there is a lot of outstanding resync activity when
7179                  * we set last_event to curr_events, then all that activity
7180                  * completing might cause the array to appear non-idle
7181                  * and resync will be slowed down even though there might
7182                  * not have been non-resync activity.  This will only
7183                  * happen once though.  'last_events' will soon reflect
7184                  * the state where there is little or no outstanding
7185                  * resync requests, and further resync activity will
7186                  * always make curr_events less than last_events.
7187                  *
7188                  */
7189                 if (init || curr_events - rdev->last_events > 64) {
7190                         rdev->last_events = curr_events;
7191                         idle = 0;
7192                 }
7193         }
7194         rcu_read_unlock();
7195         return idle;
7196 }
7197
7198 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7199 {
7200         /* another "blocks" (512byte) blocks have been synced */
7201         atomic_sub(blocks, &mddev->recovery_active);
7202         wake_up(&mddev->recovery_wait);
7203         if (!ok) {
7204                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7205                 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7206                 md_wakeup_thread(mddev->thread);
7207                 // stop recovery, signal do_sync ....
7208         }
7209 }
7210
7211
7212 /* md_write_start(mddev, bi)
7213  * If we need to update some array metadata (e.g. 'active' flag
7214  * in superblock) before writing, schedule a superblock update
7215  * and wait for it to complete.
7216  */
7217 void md_write_start(struct mddev *mddev, struct bio *bi)
7218 {
7219         int did_change = 0;
7220         if (bio_data_dir(bi) != WRITE)
7221                 return;
7222
7223         BUG_ON(mddev->ro == 1);
7224         if (mddev->ro == 2) {
7225                 /* need to switch to read/write */
7226                 mddev->ro = 0;
7227                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7228                 md_wakeup_thread(mddev->thread);
7229                 md_wakeup_thread(mddev->sync_thread);
7230                 did_change = 1;
7231         }
7232         atomic_inc(&mddev->writes_pending);
7233         if (mddev->safemode == 1)
7234                 mddev->safemode = 0;
7235         if (mddev->in_sync) {
7236                 spin_lock_irq(&mddev->write_lock);
7237                 if (mddev->in_sync) {
7238                         mddev->in_sync = 0;
7239                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7240                         set_bit(MD_CHANGE_PENDING, &mddev->flags);
7241                         md_wakeup_thread(mddev->thread);
7242                         did_change = 1;
7243                 }
7244                 spin_unlock_irq(&mddev->write_lock);
7245         }
7246         if (did_change)
7247                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7248         wait_event(mddev->sb_wait,
7249                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7250 }
7251
7252 void md_write_end(struct mddev *mddev)
7253 {
7254         if (atomic_dec_and_test(&mddev->writes_pending)) {
7255                 if (mddev->safemode == 2)
7256                         md_wakeup_thread(mddev->thread);
7257                 else if (mddev->safemode_delay)
7258                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7259         }
7260 }
7261
7262 /* md_allow_write(mddev)
7263  * Calling this ensures that the array is marked 'active' so that writes
7264  * may proceed without blocking.  It is important to call this before
7265  * attempting a GFP_KERNEL allocation while holding the mddev lock.
7266  * Must be called with mddev_lock held.
7267  *
7268  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7269  * is dropped, so return -EAGAIN after notifying userspace.
7270  */
7271 int md_allow_write(struct mddev *mddev)
7272 {
7273         if (!mddev->pers)
7274                 return 0;
7275         if (mddev->ro)
7276                 return 0;
7277         if (!mddev->pers->sync_request)
7278                 return 0;
7279
7280         spin_lock_irq(&mddev->write_lock);
7281         if (mddev->in_sync) {
7282                 mddev->in_sync = 0;
7283                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7284                 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7285                 if (mddev->safemode_delay &&
7286                     mddev->safemode == 0)
7287                         mddev->safemode = 1;
7288                 spin_unlock_irq(&mddev->write_lock);
7289                 md_update_sb(mddev, 0);
7290                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7291         } else
7292                 spin_unlock_irq(&mddev->write_lock);
7293
7294         if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7295                 return -EAGAIN;
7296         else
7297                 return 0;
7298 }
7299 EXPORT_SYMBOL_GPL(md_allow_write);
7300
7301 #define SYNC_MARKS      10
7302 #define SYNC_MARK_STEP  (3*HZ)
7303 #define UPDATE_FREQUENCY (5*60*HZ)
7304 void md_do_sync(struct md_thread *thread)
7305 {
7306         struct mddev *mddev = thread->mddev;
7307         struct mddev *mddev2;
7308         unsigned int currspeed = 0,
7309                  window;
7310         sector_t max_sectors,j, io_sectors;
7311         unsigned long mark[SYNC_MARKS];
7312         unsigned long update_time;
7313         sector_t mark_cnt[SYNC_MARKS];
7314         int last_mark,m;
7315         struct list_head *tmp;
7316         sector_t last_check;
7317         int skipped = 0;
7318         struct md_rdev *rdev;
7319         char *desc, *action = NULL;
7320         struct blk_plug plug;
7321
7322         /* just incase thread restarts... */
7323         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7324                 return;
7325         if (mddev->ro) /* never try to sync a read-only array */
7326                 return;
7327
7328         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7329                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7330                         desc = "data-check";
7331                         action = "check";
7332                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7333                         desc = "requested-resync";
7334                         action = "repair";
7335                 } else
7336                         desc = "resync";
7337         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7338                 desc = "reshape";
7339         else
7340                 desc = "recovery";
7341
7342         mddev->last_sync_action = action ?: desc;
7343
7344         /* we overload curr_resync somewhat here.
7345          * 0 == not engaged in resync at all
7346          * 2 == checking that there is no conflict with another sync
7347          * 1 == like 2, but have yielded to allow conflicting resync to
7348          *              commense
7349          * other == active in resync - this many blocks
7350          *
7351          * Before starting a resync we must have set curr_resync to
7352          * 2, and then checked that every "conflicting" array has curr_resync
7353          * less than ours.  When we find one that is the same or higher
7354          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
7355          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7356          * This will mean we have to start checking from the beginning again.
7357          *
7358          */
7359
7360         do {
7361                 mddev->curr_resync = 2;
7362
7363         try_again:
7364                 if (kthread_should_stop())
7365                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7366
7367                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7368                         goto skip;
7369                 for_each_mddev(mddev2, tmp) {
7370                         if (mddev2 == mddev)
7371                                 continue;
7372                         if (!mddev->parallel_resync
7373                         &&  mddev2->curr_resync
7374                         &&  match_mddev_units(mddev, mddev2)) {
7375                                 DEFINE_WAIT(wq);
7376                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
7377                                         /* arbitrarily yield */
7378                                         mddev->curr_resync = 1;
7379                                         wake_up(&resync_wait);
7380                                 }
7381                                 if (mddev > mddev2 && mddev->curr_resync == 1)
7382                                         /* no need to wait here, we can wait the next
7383                                          * time 'round when curr_resync == 2
7384                                          */
7385                                         continue;
7386                                 /* We need to wait 'interruptible' so as not to
7387                                  * contribute to the load average, and not to
7388                                  * be caught by 'softlockup'
7389                                  */
7390                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7391                                 if (!kthread_should_stop() &&
7392                                     mddev2->curr_resync >= mddev->curr_resync) {
7393                                         printk(KERN_INFO "md: delaying %s of %s"
7394                                                " until %s has finished (they"
7395                                                " share one or more physical units)\n",
7396                                                desc, mdname(mddev), mdname(mddev2));
7397                                         mddev_put(mddev2);
7398                                         if (signal_pending(current))
7399                                                 flush_signals(current);
7400                                         schedule();
7401                                         finish_wait(&resync_wait, &wq);
7402                                         goto try_again;
7403                                 }
7404                                 finish_wait(&resync_wait, &wq);
7405                         }
7406                 }
7407         } while (mddev->curr_resync < 2);
7408
7409         j = 0;
7410         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7411                 /* resync follows the size requested by the personality,
7412                  * which defaults to physical size, but can be virtual size
7413                  */
7414                 max_sectors = mddev->resync_max_sectors;
7415                 atomic64_set(&mddev->resync_mismatches, 0);
7416                 /* we don't use the checkpoint if there's a bitmap */
7417                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7418                         j = mddev->resync_min;
7419                 else if (!mddev->bitmap)
7420                         j = mddev->recovery_cp;
7421
7422         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7423                 max_sectors = mddev->resync_max_sectors;
7424         else {
7425                 /* recovery follows the physical size of devices */
7426                 max_sectors = mddev->dev_sectors;
7427                 j = MaxSector;
7428                 rcu_read_lock();
7429                 rdev_for_each_rcu(rdev, mddev)
7430                         if (rdev->raid_disk >= 0 &&
7431                             !test_bit(Faulty, &rdev->flags) &&
7432                             !test_bit(In_sync, &rdev->flags) &&
7433                             rdev->recovery_offset < j)
7434                                 j = rdev->recovery_offset;
7435                 rcu_read_unlock();
7436         }
7437
7438         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7439         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
7440                 " %d KB/sec/disk.\n", speed_min(mddev));
7441         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7442                "(but not more than %d KB/sec) for %s.\n",
7443                speed_max(mddev), desc);
7444
7445         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7446
7447         io_sectors = 0;
7448         for (m = 0; m < SYNC_MARKS; m++) {
7449                 mark[m] = jiffies;
7450                 mark_cnt[m] = io_sectors;
7451         }
7452         last_mark = 0;
7453         mddev->resync_mark = mark[last_mark];
7454         mddev->resync_mark_cnt = mark_cnt[last_mark];
7455
7456         /*
7457          * Tune reconstruction:
7458          */
7459         window = 32*(PAGE_SIZE/512);
7460         printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7461                 window/2, (unsigned long long)max_sectors/2);
7462
7463         atomic_set(&mddev->recovery_active, 0);
7464         last_check = 0;
7465
7466         if (j>2) {
7467                 printk(KERN_INFO
7468                        "md: resuming %s of %s from checkpoint.\n",
7469                        desc, mdname(mddev));
7470                 mddev->curr_resync = j;
7471         } else
7472                 mddev->curr_resync = 3; /* no longer delayed */
7473         mddev->curr_resync_completed = j;
7474         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7475         md_new_event(mddev);
7476         update_time = jiffies;
7477
7478         blk_start_plug(&plug);
7479         while (j < max_sectors) {
7480                 sector_t sectors;
7481
7482                 skipped = 0;
7483
7484                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7485                     ((mddev->curr_resync > mddev->curr_resync_completed &&
7486                       (mddev->curr_resync - mddev->curr_resync_completed)
7487                       > (max_sectors >> 4)) ||
7488                      time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7489                      (j - mddev->curr_resync_completed)*2
7490                      >= mddev->resync_max - mddev->curr_resync_completed
7491                             )) {
7492                         /* time to update curr_resync_completed */
7493                         wait_event(mddev->recovery_wait,
7494                                    atomic_read(&mddev->recovery_active) == 0);
7495                         mddev->curr_resync_completed = j;
7496                         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7497                             j > mddev->recovery_cp)
7498                                 mddev->recovery_cp = j;
7499                         update_time = jiffies;
7500                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7501                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7502                 }
7503
7504                 while (j >= mddev->resync_max && !kthread_should_stop()) {
7505                         /* As this condition is controlled by user-space,
7506                          * we can block indefinitely, so use '_interruptible'
7507                          * to avoid triggering warnings.
7508                          */
7509                         flush_signals(current); /* just in case */
7510                         wait_event_interruptible(mddev->recovery_wait,
7511                                                  mddev->resync_max > j
7512                                                  || kthread_should_stop());
7513                 }
7514
7515                 if (kthread_should_stop())
7516                         goto interrupted;
7517
7518                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7519                                                   currspeed < speed_min(mddev));
7520                 if (sectors == 0) {
7521                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7522                         goto out;
7523                 }
7524
7525                 if (!skipped) { /* actual IO requested */
7526                         io_sectors += sectors;
7527                         atomic_add(sectors, &mddev->recovery_active);
7528                 }
7529
7530                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7531                         break;
7532
7533                 j += sectors;
7534                 if (j > 2)
7535                         mddev->curr_resync = j;
7536                 mddev->curr_mark_cnt = io_sectors;
7537                 if (last_check == 0)
7538                         /* this is the earliest that rebuild will be
7539                          * visible in /proc/mdstat
7540                          */
7541                         md_new_event(mddev);
7542
7543                 if (last_check + window > io_sectors || j == max_sectors)
7544                         continue;
7545
7546                 last_check = io_sectors;
7547         repeat:
7548                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7549                         /* step marks */
7550                         int next = (last_mark+1) % SYNC_MARKS;
7551
7552                         mddev->resync_mark = mark[next];
7553                         mddev->resync_mark_cnt = mark_cnt[next];
7554                         mark[next] = jiffies;
7555                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7556                         last_mark = next;
7557                 }
7558
7559
7560                 if (kthread_should_stop())
7561                         goto interrupted;
7562
7563
7564                 /*
7565                  * this loop exits only if either when we are slower than
7566                  * the 'hard' speed limit, or the system was IO-idle for
7567                  * a jiffy.
7568                  * the system might be non-idle CPU-wise, but we only care
7569                  * about not overloading the IO subsystem. (things like an
7570                  * e2fsck being done on the RAID array should execute fast)
7571                  */
7572                 cond_resched();
7573
7574                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7575                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
7576
7577                 if (currspeed > speed_min(mddev)) {
7578                         if ((currspeed > speed_max(mddev)) ||
7579                                         !is_mddev_idle(mddev, 0)) {
7580                                 msleep(500);
7581                                 goto repeat;
7582                         }
7583                 }
7584         }
7585         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7586         /*
7587          * this also signals 'finished resyncing' to md_stop
7588          */
7589  out:
7590         blk_finish_plug(&plug);
7591         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7592
7593         /* tell personality that we are finished */
7594         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7595
7596         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7597             mddev->curr_resync > 2) {
7598                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7599                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7600                                 if (mddev->curr_resync >= mddev->recovery_cp) {
7601                                         printk(KERN_INFO
7602                                                "md: checkpointing %s of %s.\n",
7603                                                desc, mdname(mddev));
7604                                         if (test_bit(MD_RECOVERY_ERROR,
7605                                                 &mddev->recovery))
7606                                                 mddev->recovery_cp =
7607                                                         mddev->curr_resync_completed;
7608                                         else
7609                                                 mddev->recovery_cp =
7610                                                         mddev->curr_resync;
7611                                 }
7612                         } else
7613                                 mddev->recovery_cp = MaxSector;
7614                 } else {
7615                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7616                                 mddev->curr_resync = MaxSector;
7617                         rcu_read_lock();
7618                         rdev_for_each_rcu(rdev, mddev)
7619                                 if (rdev->raid_disk >= 0 &&
7620                                     mddev->delta_disks >= 0 &&
7621                                     !test_bit(Faulty, &rdev->flags) &&
7622                                     !test_bit(In_sync, &rdev->flags) &&
7623                                     rdev->recovery_offset < mddev->curr_resync)
7624                                         rdev->recovery_offset = mddev->curr_resync;
7625                         rcu_read_unlock();
7626                 }
7627         }
7628  skip:
7629         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7630
7631         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7632                 /* We completed so min/max setting can be forgotten if used. */
7633                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7634                         mddev->resync_min = 0;
7635                 mddev->resync_max = MaxSector;
7636         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7637                 mddev->resync_min = mddev->curr_resync_completed;
7638         mddev->curr_resync = 0;
7639         wake_up(&resync_wait);
7640         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7641         md_wakeup_thread(mddev->thread);
7642         return;
7643
7644  interrupted:
7645         /*
7646          * got a signal, exit.
7647          */
7648         printk(KERN_INFO
7649                "md: md_do_sync() got signal ... exiting\n");
7650         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7651         goto out;
7652
7653 }
7654 EXPORT_SYMBOL_GPL(md_do_sync);
7655
7656 static int remove_and_add_spares(struct mddev *mddev,
7657                                  struct md_rdev *this)
7658 {
7659         struct md_rdev *rdev;
7660         int spares = 0;
7661         int removed = 0;
7662
7663         rdev_for_each(rdev, mddev)
7664                 if ((this == NULL || rdev == this) &&
7665                     rdev->raid_disk >= 0 &&
7666                     !test_bit(Blocked, &rdev->flags) &&
7667                     (test_bit(Faulty, &rdev->flags) ||
7668                      ! test_bit(In_sync, &rdev->flags)) &&
7669                     atomic_read(&rdev->nr_pending)==0) {
7670                         if (mddev->pers->hot_remove_disk(
7671                                     mddev, rdev) == 0) {
7672                                 sysfs_unlink_rdev(mddev, rdev);
7673                                 rdev->raid_disk = -1;
7674                                 removed++;
7675                         }
7676                 }
7677         if (removed && mddev->kobj.sd)
7678                 sysfs_notify(&mddev->kobj, NULL, "degraded");
7679
7680         if (this)
7681                 goto no_add;
7682
7683         rdev_for_each(rdev, mddev) {
7684                 if (rdev->raid_disk >= 0 &&
7685                     !test_bit(In_sync, &rdev->flags) &&
7686                     !test_bit(Faulty, &rdev->flags))
7687                         spares++;
7688                 if (rdev->raid_disk >= 0)
7689                         continue;
7690                 if (test_bit(Faulty, &rdev->flags))
7691                         continue;
7692                 if (mddev->ro &&
7693                     rdev->saved_raid_disk < 0)
7694                         continue;
7695
7696                 rdev->recovery_offset = 0;
7697                 if (mddev->pers->
7698                     hot_add_disk(mddev, rdev) == 0) {
7699                         if (sysfs_link_rdev(mddev, rdev))
7700                                 /* failure here is OK */;
7701                         spares++;
7702                         md_new_event(mddev);
7703                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7704                 }
7705         }
7706 no_add:
7707         if (removed)
7708                 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7709         return spares;
7710 }
7711
7712 /*
7713  * This routine is regularly called by all per-raid-array threads to
7714  * deal with generic issues like resync and super-block update.
7715  * Raid personalities that don't have a thread (linear/raid0) do not
7716  * need this as they never do any recovery or update the superblock.
7717  *
7718  * It does not do any resync itself, but rather "forks" off other threads
7719  * to do that as needed.
7720  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7721  * "->recovery" and create a thread at ->sync_thread.
7722  * When the thread finishes it sets MD_RECOVERY_DONE
7723  * and wakeups up this thread which will reap the thread and finish up.
7724  * This thread also removes any faulty devices (with nr_pending == 0).
7725  *
7726  * The overall approach is:
7727  *  1/ if the superblock needs updating, update it.
7728  *  2/ If a recovery thread is running, don't do anything else.
7729  *  3/ If recovery has finished, clean up, possibly marking spares active.
7730  *  4/ If there are any faulty devices, remove them.
7731  *  5/ If array is degraded, try to add spares devices
7732  *  6/ If array has spares or is not in-sync, start a resync thread.
7733  */
7734 void md_check_recovery(struct mddev *mddev)
7735 {
7736         if (mddev->suspended)
7737                 return;
7738
7739         if (mddev->bitmap)
7740                 bitmap_daemon_work(mddev);
7741
7742         if (signal_pending(current)) {
7743                 if (mddev->pers->sync_request && !mddev->external) {
7744                         printk(KERN_INFO "md: %s in immediate safe mode\n",
7745                                mdname(mddev));
7746                         mddev->safemode = 2;
7747                 }
7748                 flush_signals(current);
7749         }
7750
7751         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7752                 return;
7753         if ( ! (
7754                 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7755                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7756                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7757                 (mddev->external == 0 && mddev->safemode == 1) ||
7758                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7759                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7760                 ))
7761                 return;
7762
7763         if (mddev_trylock(mddev)) {
7764                 int spares = 0;
7765
7766                 if (mddev->ro) {
7767                         /* On a read-only array we can:
7768                          * - remove failed devices
7769                          * - add already-in_sync devices if the array itself
7770                          *   is in-sync.
7771                          * As we only add devices that are already in-sync,
7772                          * we can activate the spares immediately.
7773                          */
7774                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7775                         remove_and_add_spares(mddev, NULL);
7776                         mddev->pers->spare_active(mddev);
7777                         goto unlock;
7778                 }
7779
7780                 if (!mddev->external) {
7781                         int did_change = 0;
7782                         spin_lock_irq(&mddev->write_lock);
7783                         if (mddev->safemode &&
7784                             !atomic_read(&mddev->writes_pending) &&
7785                             !mddev->in_sync &&
7786                             mddev->recovery_cp == MaxSector) {
7787                                 mddev->in_sync = 1;
7788                                 did_change = 1;
7789                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7790                         }
7791                         if (mddev->safemode == 1)
7792                                 mddev->safemode = 0;
7793                         spin_unlock_irq(&mddev->write_lock);
7794                         if (did_change)
7795                                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7796                 }
7797
7798                 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7799                         md_update_sb(mddev, 0);
7800
7801                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7802                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7803                         /* resync/recovery still happening */
7804                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7805                         goto unlock;
7806                 }
7807                 if (mddev->sync_thread) {
7808                         md_reap_sync_thread(mddev);
7809                         goto unlock;
7810                 }
7811                 /* Set RUNNING before clearing NEEDED to avoid
7812                  * any transients in the value of "sync_action".
7813                  */
7814                 mddev->curr_resync_completed = 0;
7815                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7816                 /* Clear some bits that don't mean anything, but
7817                  * might be left set
7818                  */
7819                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7820                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7821
7822                 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7823                     test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7824                         goto unlock;
7825                 /* no recovery is running.
7826                  * remove any failed drives, then
7827                  * add spares if possible.
7828                  * Spares are also removed and re-added, to allow
7829                  * the personality to fail the re-add.
7830                  */
7831
7832                 if (mddev->reshape_position != MaxSector) {
7833                         if (mddev->pers->check_reshape == NULL ||
7834                             mddev->pers->check_reshape(mddev) != 0)
7835                                 /* Cannot proceed */
7836                                 goto unlock;
7837                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7838                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7839                 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7840                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7841                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7842                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7843                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7844                 } else if (mddev->recovery_cp < MaxSector) {
7845                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7846                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7847                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7848                         /* nothing to be done ... */
7849                         goto unlock;
7850
7851                 if (mddev->pers->sync_request) {
7852                         if (spares) {
7853                                 /* We are adding a device or devices to an array
7854                                  * which has the bitmap stored on all devices.
7855                                  * So make sure all bitmap pages get written
7856                                  */
7857                                 bitmap_write_all(mddev->bitmap);
7858                         }
7859                         mddev->sync_thread = md_register_thread(md_do_sync,
7860                                                                 mddev,
7861                                                                 "resync");
7862                         if (!mddev->sync_thread) {
7863                                 printk(KERN_ERR "%s: could not start resync"
7864                                         " thread...\n",
7865                                         mdname(mddev));
7866                                 /* leave the spares where they are, it shouldn't hurt */
7867                                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7868                                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7869                                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7870                                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7871                                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7872                         } else
7873                                 md_wakeup_thread(mddev->sync_thread);
7874                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7875                         md_new_event(mddev);
7876                 }
7877         unlock:
7878                 wake_up(&mddev->sb_wait);
7879
7880                 if (!mddev->sync_thread) {
7881                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7882                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7883                                                &mddev->recovery))
7884                                 if (mddev->sysfs_action)
7885                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7886                 }
7887                 mddev_unlock(mddev);
7888         }
7889 }
7890
7891 void md_reap_sync_thread(struct mddev *mddev)
7892 {
7893         struct md_rdev *rdev;
7894
7895         /* resync has finished, collect result */
7896         md_unregister_thread(&mddev->sync_thread);
7897         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7898             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7899                 /* success...*/
7900                 /* activate any spares */
7901                 if (mddev->pers->spare_active(mddev)) {
7902                         sysfs_notify(&mddev->kobj, NULL,
7903                                      "degraded");
7904                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7905                 }
7906         }
7907         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7908             mddev->pers->finish_reshape)
7909                 mddev->pers->finish_reshape(mddev);
7910
7911         /* If array is no-longer degraded, then any saved_raid_disk
7912          * information must be scrapped.  Also if any device is now
7913          * In_sync we must scrape the saved_raid_disk for that device
7914          * do the superblock for an incrementally recovered device
7915          * written out.
7916          */
7917         rdev_for_each(rdev, mddev)
7918                 if (!mddev->degraded ||
7919                     test_bit(In_sync, &rdev->flags))
7920                         rdev->saved_raid_disk = -1;
7921
7922         md_update_sb(mddev, 1);
7923         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7924         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7925         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7926         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7927         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7928         /* flag recovery needed just to double check */
7929         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7930         sysfs_notify_dirent_safe(mddev->sysfs_action);
7931         md_new_event(mddev);
7932         if (mddev->event_work.func)
7933                 queue_work(md_misc_wq, &mddev->event_work);
7934 }
7935
7936 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7937 {
7938         sysfs_notify_dirent_safe(rdev->sysfs_state);
7939         wait_event_timeout(rdev->blocked_wait,
7940                            !test_bit(Blocked, &rdev->flags) &&
7941                            !test_bit(BlockedBadBlocks, &rdev->flags),
7942                            msecs_to_jiffies(5000));
7943         rdev_dec_pending(rdev, mddev);
7944 }
7945 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7946
7947 void md_finish_reshape(struct mddev *mddev)
7948 {
7949         /* called be personality module when reshape completes. */
7950         struct md_rdev *rdev;
7951
7952         rdev_for_each(rdev, mddev) {
7953                 if (rdev->data_offset > rdev->new_data_offset)
7954                         rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7955                 else
7956                         rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7957                 rdev->data_offset = rdev->new_data_offset;
7958         }
7959 }
7960 EXPORT_SYMBOL(md_finish_reshape);
7961
7962 /* Bad block management.
7963  * We can record which blocks on each device are 'bad' and so just
7964  * fail those blocks, or that stripe, rather than the whole device.
7965  * Entries in the bad-block table are 64bits wide.  This comprises:
7966  * Length of bad-range, in sectors: 0-511 for lengths 1-512
7967  * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7968  *  A 'shift' can be set so that larger blocks are tracked and
7969  *  consequently larger devices can be covered.
7970  * 'Acknowledged' flag - 1 bit. - the most significant bit.
7971  *
7972  * Locking of the bad-block table uses a seqlock so md_is_badblock
7973  * might need to retry if it is very unlucky.
7974  * We will sometimes want to check for bad blocks in a bi_end_io function,
7975  * so we use the write_seqlock_irq variant.
7976  *
7977  * When looking for a bad block we specify a range and want to
7978  * know if any block in the range is bad.  So we binary-search
7979  * to the last range that starts at-or-before the given endpoint,
7980  * (or "before the sector after the target range")
7981  * then see if it ends after the given start.
7982  * We return
7983  *  0 if there are no known bad blocks in the range
7984  *  1 if there are known bad block which are all acknowledged
7985  * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7986  * plus the start/length of the first bad section we overlap.
7987  */
7988 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7989                    sector_t *first_bad, int *bad_sectors)
7990 {
7991         int hi;
7992         int lo;
7993         u64 *p = bb->page;
7994         int rv;
7995         sector_t target = s + sectors;
7996         unsigned seq;
7997
7998         if (bb->shift > 0) {
7999                 /* round the start down, and the end up */
8000                 s >>= bb->shift;
8001                 target += (1<<bb->shift) - 1;
8002                 target >>= bb->shift;
8003                 sectors = target - s;
8004         }
8005         /* 'target' is now the first block after the bad range */
8006
8007 retry:
8008         seq = read_seqbegin(&bb->lock);
8009         lo = 0;
8010         rv = 0;
8011         hi = bb->count;
8012
8013         /* Binary search between lo and hi for 'target'
8014          * i.e. for the last range that starts before 'target'
8015          */
8016         /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8017          * are known not to be the last range before target.
8018          * VARIANT: hi-lo is the number of possible
8019          * ranges, and decreases until it reaches 1
8020          */
8021         while (hi - lo > 1) {
8022                 int mid = (lo + hi) / 2;
8023                 sector_t a = BB_OFFSET(p[mid]);
8024                 if (a < target)
8025                         /* This could still be the one, earlier ranges
8026                          * could not. */
8027                         lo = mid;
8028                 else
8029                         /* This and later ranges are definitely out. */
8030                         hi = mid;
8031         }
8032         /* 'lo' might be the last that started before target, but 'hi' isn't */
8033         if (hi > lo) {
8034                 /* need to check all range that end after 's' to see if
8035                  * any are unacknowledged.
8036                  */
8037                 while (lo >= 0 &&
8038                        BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8039                         if (BB_OFFSET(p[lo]) < target) {
8040                                 /* starts before the end, and finishes after
8041                                  * the start, so they must overlap
8042                                  */
8043                                 if (rv != -1 && BB_ACK(p[lo]))
8044                                         rv = 1;
8045                                 else
8046                                         rv = -1;
8047                                 *first_bad = BB_OFFSET(p[lo]);
8048                                 *bad_sectors = BB_LEN(p[lo]);
8049                         }
8050                         lo--;
8051                 }
8052         }
8053
8054         if (read_seqretry(&bb->lock, seq))
8055                 goto retry;
8056
8057         return rv;
8058 }
8059 EXPORT_SYMBOL_GPL(md_is_badblock);
8060
8061 /*
8062  * Add a range of bad blocks to the table.
8063  * This might extend the table, or might contract it
8064  * if two adjacent ranges can be merged.
8065  * We binary-search to find the 'insertion' point, then
8066  * decide how best to handle it.
8067  */
8068 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8069                             int acknowledged)
8070 {
8071         u64 *p;
8072         int lo, hi;
8073         int rv = 1;
8074         unsigned long flags;
8075
8076         if (bb->shift < 0)
8077                 /* badblocks are disabled */
8078                 return 0;
8079
8080         if (bb->shift) {
8081                 /* round the start down, and the end up */
8082                 sector_t next = s + sectors;
8083                 s >>= bb->shift;
8084                 next += (1<<bb->shift) - 1;
8085                 next >>= bb->shift;
8086                 sectors = next - s;
8087         }
8088
8089         write_seqlock_irqsave(&bb->lock, flags);
8090
8091         p = bb->page;
8092         lo = 0;
8093         hi = bb->count;
8094         /* Find the last range that starts at-or-before 's' */
8095         while (hi - lo > 1) {
8096                 int mid = (lo + hi) / 2;
8097                 sector_t a = BB_OFFSET(p[mid]);
8098                 if (a <= s)
8099                         lo = mid;
8100                 else
8101                         hi = mid;
8102         }
8103         if (hi > lo && BB_OFFSET(p[lo]) > s)
8104                 hi = lo;
8105
8106         if (hi > lo) {
8107                 /* we found a range that might merge with the start
8108                  * of our new range
8109                  */
8110                 sector_t a = BB_OFFSET(p[lo]);
8111                 sector_t e = a + BB_LEN(p[lo]);
8112                 int ack = BB_ACK(p[lo]);
8113                 if (e >= s) {
8114                         /* Yes, we can merge with a previous range */
8115                         if (s == a && s + sectors >= e)
8116                                 /* new range covers old */
8117                                 ack = acknowledged;
8118                         else
8119                                 ack = ack && acknowledged;
8120
8121                         if (e < s + sectors)
8122                                 e = s + sectors;
8123                         if (e - a <= BB_MAX_LEN) {
8124                                 p[lo] = BB_MAKE(a, e-a, ack);
8125                                 s = e;
8126                         } else {
8127                                 /* does not all fit in one range,
8128                                  * make p[lo] maximal
8129                                  */
8130                                 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8131                                         p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8132                                 s = a + BB_MAX_LEN;
8133                         }
8134                         sectors = e - s;
8135                 }
8136         }
8137         if (sectors && hi < bb->count) {
8138                 /* 'hi' points to the first range that starts after 's'.
8139                  * Maybe we can merge with the start of that range */
8140                 sector_t a = BB_OFFSET(p[hi]);
8141                 sector_t e = a + BB_LEN(p[hi]);
8142                 int ack = BB_ACK(p[hi]);
8143                 if (a <= s + sectors) {
8144                         /* merging is possible */
8145                         if (e <= s + sectors) {
8146                                 /* full overlap */
8147                                 e = s + sectors;
8148                                 ack = acknowledged;
8149                         } else
8150                                 ack = ack && acknowledged;
8151
8152                         a = s;
8153                         if (e - a <= BB_MAX_LEN) {
8154                                 p[hi] = BB_MAKE(a, e-a, ack);
8155                                 s = e;
8156                         } else {
8157                                 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8158                                 s = a + BB_MAX_LEN;
8159                         }
8160                         sectors = e - s;
8161                         lo = hi;
8162                         hi++;
8163                 }
8164         }
8165         if (sectors == 0 && hi < bb->count) {
8166                 /* we might be able to combine lo and hi */
8167                 /* Note: 's' is at the end of 'lo' */
8168                 sector_t a = BB_OFFSET(p[hi]);
8169                 int lolen = BB_LEN(p[lo]);
8170                 int hilen = BB_LEN(p[hi]);
8171                 int newlen = lolen + hilen - (s - a);
8172                 if (s >= a && newlen < BB_MAX_LEN) {
8173                         /* yes, we can combine them */
8174                         int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8175                         p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8176                         memmove(p + hi, p + hi + 1,
8177                                 (bb->count - hi - 1) * 8);
8178                         bb->count--;
8179                 }
8180         }
8181         while (sectors) {
8182                 /* didn't merge (it all).
8183                  * Need to add a range just before 'hi' */
8184                 if (bb->count >= MD_MAX_BADBLOCKS) {
8185                         /* No room for more */
8186                         rv = 0;
8187                         break;
8188                 } else {
8189                         int this_sectors = sectors;
8190                         memmove(p + hi + 1, p + hi,
8191                                 (bb->count - hi) * 8);
8192                         bb->count++;
8193
8194                         if (this_sectors > BB_MAX_LEN)
8195                                 this_sectors = BB_MAX_LEN;
8196                         p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8197                         sectors -= this_sectors;
8198                         s += this_sectors;
8199                 }
8200         }
8201
8202         bb->changed = 1;
8203         if (!acknowledged)
8204                 bb->unacked_exist = 1;
8205         write_sequnlock_irqrestore(&bb->lock, flags);
8206
8207         return rv;
8208 }
8209
8210 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8211                        int is_new)
8212 {
8213         int rv;
8214         if (is_new)
8215                 s += rdev->new_data_offset;
8216         else
8217                 s += rdev->data_offset;
8218         rv = md_set_badblocks(&rdev->badblocks,
8219                               s, sectors, 0);
8220         if (rv) {
8221                 /* Make sure they get written out promptly */
8222                 sysfs_notify_dirent_safe(rdev->sysfs_state);
8223                 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8224                 md_wakeup_thread(rdev->mddev->thread);
8225         }
8226         return rv;
8227 }
8228 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8229
8230 /*
8231  * Remove a range of bad blocks from the table.
8232  * This may involve extending the table if we spilt a region,
8233  * but it must not fail.  So if the table becomes full, we just
8234  * drop the remove request.
8235  */
8236 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8237 {
8238         u64 *p;
8239         int lo, hi;
8240         sector_t target = s + sectors;
8241         int rv = 0;
8242
8243         if (bb->shift > 0) {
8244                 /* When clearing we round the start up and the end down.
8245                  * This should not matter as the shift should align with
8246                  * the block size and no rounding should ever be needed.
8247                  * However it is better the think a block is bad when it
8248                  * isn't than to think a block is not bad when it is.
8249                  */
8250                 s += (1<<bb->shift) - 1;
8251                 s >>= bb->shift;
8252                 target >>= bb->shift;
8253                 sectors = target - s;
8254         }
8255
8256         write_seqlock_irq(&bb->lock);
8257
8258         p = bb->page;
8259         lo = 0;
8260         hi = bb->count;
8261         /* Find the last range that starts before 'target' */
8262         while (hi - lo > 1) {
8263                 int mid = (lo + hi) / 2;
8264                 sector_t a = BB_OFFSET(p[mid]);
8265                 if (a < target)
8266                         lo = mid;
8267                 else
8268                         hi = mid;
8269         }
8270         if (hi > lo) {
8271                 /* p[lo] is the last range that could overlap the
8272                  * current range.  Earlier ranges could also overlap,
8273                  * but only this one can overlap the end of the range.
8274                  */
8275                 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8276                         /* Partial overlap, leave the tail of this range */
8277                         int ack = BB_ACK(p[lo]);
8278                         sector_t a = BB_OFFSET(p[lo]);
8279                         sector_t end = a + BB_LEN(p[lo]);
8280
8281                         if (a < s) {
8282                                 /* we need to split this range */
8283                                 if (bb->count >= MD_MAX_BADBLOCKS) {
8284                                         rv = 0;
8285                                         goto out;
8286                                 }
8287                                 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8288                                 bb->count++;
8289                                 p[lo] = BB_MAKE(a, s-a, ack);
8290                                 lo++;
8291                         }
8292                         p[lo] = BB_MAKE(target, end - target, ack);
8293                         /* there is no longer an overlap */
8294                         hi = lo;
8295                         lo--;
8296                 }
8297                 while (lo >= 0 &&
8298                        BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8299                         /* This range does overlap */
8300                         if (BB_OFFSET(p[lo]) < s) {
8301                                 /* Keep the early parts of this range. */
8302                                 int ack = BB_ACK(p[lo]);
8303                                 sector_t start = BB_OFFSET(p[lo]);
8304                                 p[lo] = BB_MAKE(start, s - start, ack);
8305                                 /* now low doesn't overlap, so.. */
8306                                 break;
8307                         }
8308                         lo--;
8309                 }
8310                 /* 'lo' is strictly before, 'hi' is strictly after,
8311                  * anything between needs to be discarded
8312                  */
8313                 if (hi - lo > 1) {
8314                         memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8315                         bb->count -= (hi - lo - 1);
8316                 }
8317         }
8318
8319         bb->changed = 1;
8320 out:
8321         write_sequnlock_irq(&bb->lock);
8322         return rv;
8323 }
8324
8325 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8326                          int is_new)
8327 {
8328         if (is_new)
8329                 s += rdev->new_data_offset;
8330         else
8331                 s += rdev->data_offset;
8332         return md_clear_badblocks(&rdev->badblocks,
8333                                   s, sectors);
8334 }
8335 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8336
8337 /*
8338  * Acknowledge all bad blocks in a list.
8339  * This only succeeds if ->changed is clear.  It is used by
8340  * in-kernel metadata updates
8341  */
8342 void md_ack_all_badblocks(struct badblocks *bb)
8343 {
8344         if (bb->page == NULL || bb->changed)
8345                 /* no point even trying */
8346                 return;
8347         write_seqlock_irq(&bb->lock);
8348
8349         if (bb->changed == 0 && bb->unacked_exist) {
8350                 u64 *p = bb->page;
8351                 int i;
8352                 for (i = 0; i < bb->count ; i++) {
8353                         if (!BB_ACK(p[i])) {
8354                                 sector_t start = BB_OFFSET(p[i]);
8355                                 int len = BB_LEN(p[i]);
8356                                 p[i] = BB_MAKE(start, len, 1);
8357                         }
8358                 }
8359                 bb->unacked_exist = 0;
8360         }
8361         write_sequnlock_irq(&bb->lock);
8362 }
8363 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8364
8365 /* sysfs access to bad-blocks list.
8366  * We present two files.
8367  * 'bad-blocks' lists sector numbers and lengths of ranges that
8368  *    are recorded as bad.  The list is truncated to fit within
8369  *    the one-page limit of sysfs.
8370  *    Writing "sector length" to this file adds an acknowledged
8371  *    bad block list.
8372  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8373  *    been acknowledged.  Writing to this file adds bad blocks
8374  *    without acknowledging them.  This is largely for testing.
8375  */
8376
8377 static ssize_t
8378 badblocks_show(struct badblocks *bb, char *page, int unack)
8379 {
8380         size_t len;
8381         int i;
8382         u64 *p = bb->page;
8383         unsigned seq;
8384
8385         if (bb->shift < 0)
8386                 return 0;
8387
8388 retry:
8389         seq = read_seqbegin(&bb->lock);
8390
8391         len = 0;
8392         i = 0;
8393
8394         while (len < PAGE_SIZE && i < bb->count) {
8395                 sector_t s = BB_OFFSET(p[i]);
8396                 unsigned int length = BB_LEN(p[i]);
8397                 int ack = BB_ACK(p[i]);
8398                 i++;
8399
8400                 if (unack && ack)
8401                         continue;
8402
8403                 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8404                                 (unsigned long long)s << bb->shift,
8405                                 length << bb->shift);
8406         }
8407         if (unack && len == 0)
8408                 bb->unacked_exist = 0;
8409
8410         if (read_seqretry(&bb->lock, seq))
8411                 goto retry;
8412
8413         return len;
8414 }
8415
8416 #define DO_DEBUG 1
8417
8418 static ssize_t
8419 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8420 {
8421         unsigned long long sector;
8422         int length;
8423         char newline;
8424 #ifdef DO_DEBUG
8425         /* Allow clearing via sysfs *only* for testing/debugging.
8426          * Normally only a successful write may clear a badblock
8427          */
8428         int clear = 0;
8429         if (page[0] == '-') {
8430                 clear = 1;
8431                 page++;
8432         }
8433 #endif /* DO_DEBUG */
8434
8435         switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8436         case 3:
8437                 if (newline != '\n')
8438                         return -EINVAL;
8439         case 2:
8440                 if (length <= 0)
8441                         return -EINVAL;
8442                 break;
8443         default:
8444                 return -EINVAL;
8445         }
8446
8447 #ifdef DO_DEBUG
8448         if (clear) {
8449                 md_clear_badblocks(bb, sector, length);
8450                 return len;
8451         }
8452 #endif /* DO_DEBUG */
8453         if (md_set_badblocks(bb, sector, length, !unack))
8454                 return len;
8455         else
8456                 return -ENOSPC;
8457 }
8458
8459 static int md_notify_reboot(struct notifier_block *this,
8460                             unsigned long code, void *x)
8461 {
8462         struct list_head *tmp;
8463         struct mddev *mddev;
8464         int need_delay = 0;
8465
8466         for_each_mddev(mddev, tmp) {
8467                 if (mddev_trylock(mddev)) {
8468                         if (mddev->pers)
8469                                 __md_stop_writes(mddev);
8470                         mddev->safemode = 2;
8471                         mddev_unlock(mddev);
8472                 }
8473                 need_delay = 1;
8474         }
8475         /*
8476          * certain more exotic SCSI devices are known to be
8477          * volatile wrt too early system reboots. While the
8478          * right place to handle this issue is the given
8479          * driver, we do want to have a safe RAID driver ...
8480          */
8481         if (need_delay)
8482                 mdelay(1000*1);
8483
8484         return NOTIFY_DONE;
8485 }
8486
8487 static struct notifier_block md_notifier = {
8488         .notifier_call  = md_notify_reboot,
8489         .next           = NULL,
8490         .priority       = INT_MAX, /* before any real devices */
8491 };
8492
8493 static void md_geninit(void)
8494 {
8495         pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8496
8497         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8498 }
8499
8500 static int __init md_init(void)
8501 {
8502         int ret = -ENOMEM;
8503
8504         md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8505         if (!md_wq)
8506                 goto err_wq;
8507
8508         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8509         if (!md_misc_wq)
8510                 goto err_misc_wq;
8511
8512         if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8513                 goto err_md;
8514
8515         if ((ret = register_blkdev(0, "mdp")) < 0)
8516                 goto err_mdp;
8517         mdp_major = ret;
8518
8519         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8520                             md_probe, NULL, NULL);
8521         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8522                             md_probe, NULL, NULL);
8523
8524         register_reboot_notifier(&md_notifier);
8525         raid_table_header = register_sysctl_table(raid_root_table);
8526
8527         md_geninit();
8528         return 0;
8529
8530 err_mdp:
8531         unregister_blkdev(MD_MAJOR, "md");
8532 err_md:
8533         destroy_workqueue(md_misc_wq);
8534 err_misc_wq:
8535         destroy_workqueue(md_wq);
8536 err_wq:
8537         return ret;
8538 }
8539
8540 #ifndef MODULE
8541
8542 /*
8543  * Searches all registered partitions for autorun RAID arrays
8544  * at boot time.
8545  */
8546
8547 static LIST_HEAD(all_detected_devices);
8548 struct detected_devices_node {
8549         struct list_head list;
8550         dev_t dev;
8551 };
8552
8553 void md_autodetect_dev(dev_t dev)
8554 {
8555         struct detected_devices_node *node_detected_dev;
8556
8557         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8558         if (node_detected_dev) {
8559                 node_detected_dev->dev = dev;
8560                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8561         } else {
8562                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8563                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8564         }
8565 }
8566
8567
8568 static void autostart_arrays(int part)
8569 {
8570         struct md_rdev *rdev;
8571         struct detected_devices_node *node_detected_dev;
8572         dev_t dev;
8573         int i_scanned, i_passed;
8574
8575         i_scanned = 0;
8576         i_passed = 0;
8577
8578         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8579
8580         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8581                 i_scanned++;
8582                 node_detected_dev = list_entry(all_detected_devices.next,
8583                                         struct detected_devices_node, list);
8584                 list_del(&node_detected_dev->list);
8585                 dev = node_detected_dev->dev;
8586                 kfree(node_detected_dev);
8587                 rdev = md_import_device(dev,0, 90);
8588                 if (IS_ERR(rdev))
8589                         continue;
8590
8591                 if (test_bit(Faulty, &rdev->flags)) {
8592                         MD_BUG();
8593                         continue;
8594                 }
8595                 set_bit(AutoDetected, &rdev->flags);
8596                 list_add(&rdev->same_set, &pending_raid_disks);
8597                 i_passed++;
8598         }
8599
8600         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8601                                                 i_scanned, i_passed);
8602
8603         autorun_devices(part);
8604 }
8605
8606 #endif /* !MODULE */
8607
8608 static __exit void md_exit(void)
8609 {
8610         struct mddev *mddev;
8611         struct list_head *tmp;
8612
8613         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8614         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8615
8616         unregister_blkdev(MD_MAJOR,"md");
8617         unregister_blkdev(mdp_major, "mdp");
8618         unregister_reboot_notifier(&md_notifier);
8619         unregister_sysctl_table(raid_table_header);
8620         remove_proc_entry("mdstat", NULL);
8621         for_each_mddev(mddev, tmp) {
8622                 export_array(mddev);
8623                 mddev->hold_active = 0;
8624         }
8625         destroy_workqueue(md_misc_wq);
8626         destroy_workqueue(md_wq);
8627 }
8628
8629 subsys_initcall(md_init);
8630 module_exit(md_exit)
8631
8632 static int get_ro(char *buffer, struct kernel_param *kp)
8633 {
8634         return sprintf(buffer, "%d", start_readonly);
8635 }
8636 static int set_ro(const char *val, struct kernel_param *kp)
8637 {
8638         char *e;
8639         int num = simple_strtoul(val, &e, 10);
8640         if (*val && (*e == '\0' || *e == '\n')) {
8641                 start_readonly = num;
8642                 return 0;
8643         }
8644         return -EINVAL;
8645 }
8646
8647 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8648 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8649
8650 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8651
8652 EXPORT_SYMBOL(register_md_personality);
8653 EXPORT_SYMBOL(unregister_md_personality);
8654 EXPORT_SYMBOL(md_error);
8655 EXPORT_SYMBOL(md_done_sync);
8656 EXPORT_SYMBOL(md_write_start);
8657 EXPORT_SYMBOL(md_write_end);
8658 EXPORT_SYMBOL(md_register_thread);
8659 EXPORT_SYMBOL(md_unregister_thread);
8660 EXPORT_SYMBOL(md_wakeup_thread);
8661 EXPORT_SYMBOL(md_check_recovery);
8662 EXPORT_SYMBOL(md_reap_sync_thread);
8663 MODULE_LICENSE("GPL");
8664 MODULE_DESCRIPTION("MD RAID framework");
8665 MODULE_ALIAS("md");
8666 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);