drivers/md/md.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3    md.c : Multiple Devices driver for Linux
   4      Copyright (C) 1998, 1999, 2000 Ingo Molnar
   5
   6      completely rewritten, based on the MD driver code from Marc Zyngier
   7
   8    Changes:
   9
  10    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  11    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  12    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  13    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  14    - kmod support by: Cyrus Durgin
  15    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  16    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  17
  18    - lots of fixes and improvements to the RAID1/RAID5 and generic
  19      RAID code (such as request based resynchronization):
  20
  21      Neil Brown <neilb@cse.unsw.edu.au>.
  22
  23    - persistent bitmap code
  24      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  25
  26
  27    Errors, Warnings, etc.
  28    Please use:
  29      pr_crit() for error conditions that risk data loss
  30      pr_err() for error conditions that are unexpected, like an IO error
  31          or internal inconsistency
  32      pr_warn() for error conditions that could have been predicated, like
  33          adding a device to an array when it has incompatible metadata
  34      pr_info() for every interesting, very rare events, like an array starting
  35          or stopping, or resync starting or stopping
  36      pr_debug() for everything else.
  37
  38 */
  39
  40 #include <linux/sched/mm.h>
  41 #include <linux/sched/signal.h>
  42 #include <linux/kthread.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/badblocks.h>
  45 #include <linux/sysctl.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/fs.h>
  48 #include <linux/poll.h>
  49 #include <linux/ctype.h>
  50 #include <linux/string.h>
  51 #include <linux/hdreg.h>
  52 #include <linux/proc_fs.h>
  53 #include <linux/random.h>
  54 #include <linux/module.h>
  55 #include <linux/reboot.h>
  56 #include <linux/file.h>
  57 #include <linux/compat.h>
  58 #include <linux/delay.h>
  59 #include <linux/raid/md_p.h>
  60 #include <linux/raid/md_u.h>
  61 #include <linux/raid/detect.h>
  62 #include <linux/slab.h>
  63 #include <linux/percpu-refcount.h>
  64 #include <linux/part_stat.h>
  65
  66 #include <trace/events/block.h>
  67 #include "md.h"
  68 #include "md-bitmap.h"
  69 #include "md-cluster.h"
  70
  71 /* pers_list is a list of registered personalities protected
  72  * by pers_lock.
  73  * pers_lock does extra service to protect accesses to
  74  * mddev->thread when the mutex cannot be held.
  75  */
  76 static LIST_HEAD(pers_list);
  77 static DEFINE_SPINLOCK(pers_lock);
  78
  79 static struct kobj_type md_ktype;
  80
  81 struct md_cluster_operations *md_cluster_ops;
  82 EXPORT_SYMBOL(md_cluster_ops);
  83 static struct module *md_cluster_mod;
  84
  85 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  86 static struct workqueue_struct *md_wq;
  87 static struct workqueue_struct *md_misc_wq;
  88 static struct workqueue_struct *md_rdev_misc_wq;
  89
  90 static int remove_and_add_spares(struct mddev *mddev,
  91                                  struct md_rdev *this);
  92 static void mddev_detach(struct mddev *mddev);
  93
  94 /*
  95  * Default number of read corrections we'll attempt on an rdev
  96  * before ejecting it from the array. We divide the read error
  97  * count by 2 for every hour elapsed between read errors.
  98  */
  99 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
 100 /* Default safemode delay: 200 msec */
 101 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
 102 /*
 103  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
 104  * is 1000 KB/sec, so the extra system load does not show up that much.
 105  * Increase it if you want to have more _guaranteed_ speed. Note that
 106  * the RAID driver will use the maximum available bandwidth if the IO
 107  * subsystem is idle. There is also an 'absolute maximum' reconstruction
 108  * speed limit - in case reconstruction slows down your system despite
 109  * idle IO detection.
 110  *
 111  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
 112  * or /sys/block/mdX/md/sync_speed_{min,max}
 113  */
 114
 115 static int sysctl_speed_limit_min = 1000;
 116 static int sysctl_speed_limit_max = 200000;
 117 static inline int speed_min(struct mddev *mddev)
 118 {
 119         return mddev->sync_speed_min ?
 120                 mddev->sync_speed_min : sysctl_speed_limit_min;
 121 }
 122
 123 static inline int speed_max(struct mddev *mddev)
 124 {
 125         return mddev->sync_speed_max ?
 126                 mddev->sync_speed_max : sysctl_speed_limit_max;
 127 }
 128
 129 static void rdev_uninit_serial(struct md_rdev *rdev)
 130 {
 131         if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
 132                 return;
 133
 134         kvfree(rdev->serial);
 135         rdev->serial = NULL;
 136 }
 137
 138 static void rdevs_uninit_serial(struct mddev *mddev)
 139 {
 140         struct md_rdev *rdev;
 141
 142         rdev_for_each(rdev, mddev)
 143                 rdev_uninit_serial(rdev);
 144 }
 145
 146 static int rdev_init_serial(struct md_rdev *rdev)
 147 {
 148         /* serial_nums equals with BARRIER_BUCKETS_NR */
 149         int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
 150         struct serial_in_rdev *serial = NULL;
 151
 152         if (test_bit(CollisionCheck, &rdev->flags))
 153                 return 0;
 154
 155         serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
 156                           GFP_KERNEL);
 157         if (!serial)
 158                 return -ENOMEM;
 159
 160         for (i = 0; i < serial_nums; i++) {
 161                 struct serial_in_rdev *serial_tmp = &serial[i];
 162
 163                 spin_lock_init(&serial_tmp->serial_lock);
 164                 serial_tmp->serial_rb = RB_ROOT_CACHED;
 165                 init_waitqueue_head(&serial_tmp->serial_io_wait);
 166         }
 167
 168         rdev->serial = serial;
 169         set_bit(CollisionCheck, &rdev->flags);
 170
 171         return 0;
 172 }
 173
 174 static int rdevs_init_serial(struct mddev *mddev)
 175 {
 176         struct md_rdev *rdev;
 177         int ret = 0;
 178
 179         rdev_for_each(rdev, mddev) {
 180                 ret = rdev_init_serial(rdev);
 181                 if (ret)
 182                         break;
 183         }
 184
 185         /* Free all resources if pool is not existed */
 186         if (ret && !mddev->serial_info_pool)
 187                 rdevs_uninit_serial(mddev);
 188
 189         return ret;
 190 }
 191
 192 /*
 193  * rdev needs to enable serial stuffs if it meets the conditions:
 194  * 1. it is multi-queue device flaged with writemostly.
 195  * 2. the write-behind mode is enabled.
 196  */
 197 static int rdev_need_serial(struct md_rdev *rdev)
 198 {
 199         return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
 200                 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
 201                 test_bit(WriteMostly, &rdev->flags));
 202 }
 203
 204 /*
 205  * Init resource for rdev(s), then create serial_info_pool if:
 206  * 1. rdev is the first device which return true from rdev_enable_serial.
 207  * 2. rdev is NULL, means we want to enable serialization for all rdevs.
 208  */
 209 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
 210                               bool is_suspend)
 211 {
 212         int ret = 0;
 213
 214         if (rdev && !rdev_need_serial(rdev) &&
 215             !test_bit(CollisionCheck, &rdev->flags))
 216                 return;
 217
 218         if (!is_suspend)
 219                 mddev_suspend(mddev);
 220
 221         if (!rdev)
 222                 ret = rdevs_init_serial(mddev);
 223         else
 224                 ret = rdev_init_serial(rdev);
 225         if (ret)
 226                 goto abort;
 227
 228         if (mddev->serial_info_pool == NULL) {
 229                 /*
 230                  * already in memalloc noio context by
 231                  * mddev_suspend()
 232                  */
 233                 mddev->serial_info_pool =
 234                         mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
 235                                                 sizeof(struct serial_info));
 236                 if (!mddev->serial_info_pool) {
 237                         rdevs_uninit_serial(mddev);
 238                         pr_err("can't alloc memory pool for serialization\n");
 239                 }
 240         }
 241
 242 abort:
 243         if (!is_suspend)
 244                 mddev_resume(mddev);
 245 }
 246
 247 /*
 248  * Free resource from rdev(s), and destroy serial_info_pool under conditions:
 249  * 1. rdev is the last device flaged with CollisionCheck.
 250  * 2. when bitmap is destroyed while policy is not enabled.
 251  * 3. for disable policy, the pool is destroyed only when no rdev needs it.
 252  */
 253 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
 254                                bool is_suspend)
 255 {
 256         if (rdev && !test_bit(CollisionCheck, &rdev->flags))
 257                 return;
 258
 259         if (mddev->serial_info_pool) {
 260                 struct md_rdev *temp;
 261                 int num = 0; /* used to track if other rdevs need the pool */
 262
 263                 if (!is_suspend)
 264                         mddev_suspend(mddev);
 265                 rdev_for_each(temp, mddev) {
 266                         if (!rdev) {
 267                                 if (!mddev->serialize_policy ||
 268                                     !rdev_need_serial(temp))
 269                                         rdev_uninit_serial(temp);
 270                                 else
 271                                         num++;
 272                         } else if (temp != rdev &&
 273                                    test_bit(CollisionCheck, &temp->flags))
 274                                 num++;
 275                 }
 276
 277                 if (rdev)
 278                         rdev_uninit_serial(rdev);
 279
 280                 if (num)
 281                         pr_info("The mempool could be used by other devices\n");
 282                 else {
 283                         mempool_destroy(mddev->serial_info_pool);
 284                         mddev->serial_info_pool = NULL;
 285                 }
 286                 if (!is_suspend)
 287                         mddev_resume(mddev);
 288         }
 289 }
 290
 291 static struct ctl_table_header *raid_table_header;
 292
 293 static struct ctl_table raid_table[] = {
 294         {
 295                 .procname       = "speed_limit_min",
 296                 .data           = &sysctl_speed_limit_min,
 297                 .maxlen         = sizeof(int),
 298                 .mode           = S_IRUGO|S_IWUSR,
 299                 .proc_handler   = proc_dointvec,
 300         },
 301         {
 302                 .procname       = "speed_limit_max",
 303                 .data           = &sysctl_speed_limit_max,
 304                 .maxlen         = sizeof(int),
 305                 .mode           = S_IRUGO|S_IWUSR,
 306                 .proc_handler   = proc_dointvec,
 307         },
 308         { }
 309 };
 310
 311 static struct ctl_table raid_dir_table[] = {
 312         {
 313                 .procname       = "raid",
 314                 .maxlen         = 0,
 315                 .mode           = S_IRUGO|S_IXUGO,
 316                 .child          = raid_table,
 317         },
 318         { }
 319 };
 320
 321 static struct ctl_table raid_root_table[] = {
 322         {
 323                 .procname       = "dev",
 324                 .maxlen         = 0,
 325                 .mode           = 0555,
 326                 .child          = raid_dir_table,
 327         },
 328         {  }
 329 };
 330
 331 static int start_readonly;
 332
 333 /*
 334  * The original mechanism for creating an md device is to create
 335  * a device node in /dev and to open it.  This causes races with device-close.
 336  * The preferred method is to write to the "new_array" module parameter.
 337  * This can avoid races.
 338  * Setting create_on_open to false disables the original mechanism
 339  * so all the races disappear.
 340  */
 341 static bool create_on_open = true;
 342
 343 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 344                             struct mddev *mddev)
 345 {
 346         if (!mddev || !bioset_initialized(&mddev->bio_set))
 347                 return bio_alloc(gfp_mask, nr_iovecs);
 348
 349         return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
 350 }
 351 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 352
 353 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
 354 {
 355         if (!mddev || !bioset_initialized(&mddev->sync_set))
 356                 return bio_alloc(GFP_NOIO, 1);
 357
 358         return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
 359 }
 360
 361 /*
 362  * We have a system wide 'event count' that is incremented
 363  * on any 'interesting' event, and readers of /proc/mdstat
 364  * can use 'poll' or 'select' to find out when the event
 365  * count increases.
 366  *
 367  * Events are:
 368  *  start array, stop array, error, add device, remove device,
 369  *  start build, activate spare
 370  */
 371 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 372 static atomic_t md_event_count;
 373 void md_new_event(struct mddev *mddev)
 374 {
 375         atomic_inc(&md_event_count);
 376         wake_up(&md_event_waiters);
 377 }
 378 EXPORT_SYMBOL_GPL(md_new_event);
 379
 380 /*
 381  * Enables to iterate over all existing md arrays
 382  * all_mddevs_lock protects this list.
 383  */
 384 static LIST_HEAD(all_mddevs);
 385 static DEFINE_SPINLOCK(all_mddevs_lock);
 386
 387 /*
 388  * iterates through all used mddevs in the system.
 389  * We take care to grab the all_mddevs_lock whenever navigating
 390  * the list, and to always hold a refcount when unlocked.
 391  * Any code which breaks out of this loop while own
 392  * a reference to the current mddev and must mddev_put it.
 393  */
 394 #define for_each_mddev(_mddev,_tmp)                                     \
 395                                                                         \
 396         for (({ spin_lock(&all_mddevs_lock);                            \
 397                 _tmp = all_mddevs.next;                                 \
 398                 _mddev = NULL;});                                       \
 399              ({ if (_tmp != &all_mddevs)                                \
 400                         mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
 401                 spin_unlock(&all_mddevs_lock);                          \
 402                 if (_mddev) mddev_put(_mddev);                          \
 403                 _mddev = list_entry(_tmp, struct mddev, all_mddevs);    \
 404                 _tmp != &all_mddevs;});                                 \
 405              ({ spin_lock(&all_mddevs_lock);                            \
 406                 _tmp = _tmp->next;})                                    \
 407                 )
 408
 409 /* Rather than calling directly into the personality make_request function,
 410  * IO requests come here first so that we can check if the device is
 411  * being suspended pending a reconfiguration.
 412  * We hold a refcount over the call to ->make_request.  By the time that
 413  * call has finished, the bio has been linked into some internal structure
 414  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 415  */
 416 static bool is_suspended(struct mddev *mddev, struct bio *bio)
 417 {
 418         if (mddev->suspended)
 419                 return true;
 420         if (bio_data_dir(bio) != WRITE)
 421                 return false;
 422         if (mddev->suspend_lo >= mddev->suspend_hi)
 423                 return false;
 424         if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
 425                 return false;
 426         if (bio_end_sector(bio) < mddev->suspend_lo)
 427                 return false;
 428         return true;
 429 }
 430
 431 void md_handle_request(struct mddev *mddev, struct bio *bio)
 432 {
 433 check_suspended:
 434         rcu_read_lock();
 435         if (is_suspended(mddev, bio)) {
 436                 DEFINE_WAIT(__wait);
 437                 for (;;) {
 438                         prepare_to_wait(&mddev->sb_wait, &__wait,
 439                                         TASK_UNINTERRUPTIBLE);
 440                         if (!is_suspended(mddev, bio))
 441                                 break;
 442                         rcu_read_unlock();
 443                         schedule();
 444                         rcu_read_lock();
 445                 }
 446                 finish_wait(&mddev->sb_wait, &__wait);
 447         }
 448         atomic_inc(&mddev->active_io);
 449         rcu_read_unlock();
 450
 451         if (!mddev->pers->make_request(mddev, bio)) {
 452                 atomic_dec(&mddev->active_io);
 453                 wake_up(&mddev->sb_wait);
 454                 goto check_suspended;
 455         }
 456
 457         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 458                 wake_up(&mddev->sb_wait);
 459 }
 460 EXPORT_SYMBOL(md_handle_request);
 461
 462 struct md_io {
 463         struct mddev *mddev;
 464         bio_end_io_t *orig_bi_end_io;
 465         void *orig_bi_private;
 466         unsigned long start_time;
 467         struct hd_struct *part;
 468 };
 469
 470 static void md_end_io(struct bio *bio)
 471 {
 472         struct md_io *md_io = bio->bi_private;
 473         struct mddev *mddev = md_io->mddev;
 474
 475         part_end_io_acct(md_io->part, bio, md_io->start_time);
 476
 477         bio->bi_end_io = md_io->orig_bi_end_io;
 478         bio->bi_private = md_io->orig_bi_private;
 479
 480         mempool_free(md_io, &mddev->md_io_pool);
 481
 482         if (bio->bi_end_io)
 483                 bio->bi_end_io(bio);
 484 }
 485
 486 static blk_qc_t md_submit_bio(struct bio *bio)
 487 {
 488         const int rw = bio_data_dir(bio);
 489         struct mddev *mddev = bio->bi_disk->private_data;
 490
 491         if (mddev == NULL || mddev->pers == NULL) {
 492                 bio_io_error(bio);
 493                 return BLK_QC_T_NONE;
 494         }
 495
 496         if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
 497                 bio_io_error(bio);
 498                 return BLK_QC_T_NONE;
 499         }
 500
 501         blk_queue_split(&bio);
 502
 503         if (mddev->ro == 1 && unlikely(rw == WRITE)) {
 504                 if (bio_sectors(bio) != 0)
 505                         bio->bi_status = BLK_STS_IOERR;
 506                 bio_endio(bio);
 507                 return BLK_QC_T_NONE;
 508         }
 509
 510         if (bio->bi_end_io != md_end_io) {
 511                 struct md_io *md_io;
 512
 513                 md_io = mempool_alloc(&mddev->md_io_pool, GFP_NOIO);
 514                 md_io->mddev = mddev;
 515                 md_io->orig_bi_end_io = bio->bi_end_io;
 516                 md_io->orig_bi_private = bio->bi_private;
 517
 518                 bio->bi_end_io = md_end_io;
 519                 bio->bi_private = md_io;
 520
 521                 md_io->start_time = part_start_io_acct(mddev->gendisk,
 522                                                        &md_io->part, bio);
 523         }
 524
 525         /* bio could be mergeable after passing to underlayer */
 526         bio->bi_opf &= ~REQ_NOMERGE;
 527
 528         md_handle_request(mddev, bio);
 529
 530         return BLK_QC_T_NONE;
 531 }
 532
 533 /* mddev_suspend makes sure no new requests are submitted
 534  * to the device, and that any requests that have been submitted
 535  * are completely handled.
 536  * Once mddev_detach() is called and completes, the module will be
 537  * completely unused.
 538  */
 539 void mddev_suspend(struct mddev *mddev)
 540 {
 541         WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
 542         lockdep_assert_held(&mddev->reconfig_mutex);
 543         if (mddev->suspended++)
 544                 return;
 545         synchronize_rcu();
 546         wake_up(&mddev->sb_wait);
 547         set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
 548         smp_mb__after_atomic();
 549         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 550         mddev->pers->quiesce(mddev, 1);
 551         clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
 552         wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
 553
 554         del_timer_sync(&mddev->safemode_timer);
 555         /* restrict memory reclaim I/O during raid array is suspend */
 556         mddev->noio_flag = memalloc_noio_save();
 557 }
 558 EXPORT_SYMBOL_GPL(mddev_suspend);
 559
 560 void mddev_resume(struct mddev *mddev)
 561 {
 562         /* entred the memalloc scope from mddev_suspend() */
 563         memalloc_noio_restore(mddev->noio_flag);
 564         lockdep_assert_held(&mddev->reconfig_mutex);
 565         if (--mddev->suspended)
 566                 return;
 567         wake_up(&mddev->sb_wait);
 568         mddev->pers->quiesce(mddev, 0);
 569
 570         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 571         md_wakeup_thread(mddev->thread);
 572         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 573 }
 574 EXPORT_SYMBOL_GPL(mddev_resume);
 575
 576 /*
 577  * Generic flush handling for md
 578  */
 579
 580 static void md_end_flush(struct bio *bio)
 581 {
 582         struct md_rdev *rdev = bio->bi_private;
 583         struct mddev *mddev = rdev->mddev;
 584
 585         rdev_dec_pending(rdev, mddev);
 586
 587         if (atomic_dec_and_test(&mddev->flush_pending)) {
 588                 /* The pre-request flush has finished */
 589                 queue_work(md_wq, &mddev->flush_work);
 590         }
 591         bio_put(bio);
 592 }
 593
 594 static void md_submit_flush_data(struct work_struct *ws);
 595
 596 static void submit_flushes(struct work_struct *ws)
 597 {
 598         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 599         struct md_rdev *rdev;
 600
 601         mddev->start_flush = ktime_get_boottime();
 602         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 603         atomic_set(&mddev->flush_pending, 1);
 604         rcu_read_lock();
 605         rdev_for_each_rcu(rdev, mddev)
 606                 if (rdev->raid_disk >= 0 &&
 607                     !test_bit(Faulty, &rdev->flags)) {
 608                         /* Take two references, one is dropped
 609                          * when request finishes, one after
 610                          * we reclaim rcu_read_lock
 611                          */
 612                         struct bio *bi;
 613                         atomic_inc(&rdev->nr_pending);
 614                         atomic_inc(&rdev->nr_pending);
 615                         rcu_read_unlock();
 616                         bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
 617                         bi->bi_end_io = md_end_flush;
 618                         bi->bi_private = rdev;
 619                         bio_set_dev(bi, rdev->bdev);
 620                         bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
 621                         atomic_inc(&mddev->flush_pending);
 622                         submit_bio(bi);
 623                         rcu_read_lock();
 624                         rdev_dec_pending(rdev, mddev);
 625                 }
 626         rcu_read_unlock();
 627         if (atomic_dec_and_test(&mddev->flush_pending))
 628                 queue_work(md_wq, &mddev->flush_work);
 629 }
 630
 631 static void md_submit_flush_data(struct work_struct *ws)
 632 {
 633         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 634         struct bio *bio = mddev->flush_bio;
 635
 636         /*
 637          * must reset flush_bio before calling into md_handle_request to avoid a
 638          * deadlock, because other bios passed md_handle_request suspend check
 639          * could wait for this and below md_handle_request could wait for those
 640          * bios because of suspend check
 641          */
 642         mddev->last_flush = mddev->start_flush;
 643         mddev->flush_bio = NULL;
 644         wake_up(&mddev->sb_wait);
 645
 646         if (bio->bi_iter.bi_size == 0) {
 647                 /* an empty barrier - all done */
 648                 bio_endio(bio);
 649         } else {
 650                 bio->bi_opf &= ~REQ_PREFLUSH;
 651                 md_handle_request(mddev, bio);
 652         }
 653 }
 654
 655 /*
 656  * Manages consolidation of flushes and submitting any flushes needed for
 657  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
 658  * being finished in another context.  Returns false if the flushing is
 659  * complete but still needs the I/O portion of the bio to be processed.
 660  */
 661 bool md_flush_request(struct mddev *mddev, struct bio *bio)
 662 {
 663         ktime_t start = ktime_get_boottime();
 664         spin_lock_irq(&mddev->lock);
 665         wait_event_lock_irq(mddev->sb_wait,
 666                             !mddev->flush_bio ||
 667                             ktime_after(mddev->last_flush, start),
 668                             mddev->lock);
 669         if (!ktime_after(mddev->last_flush, start)) {
 670                 WARN_ON(mddev->flush_bio);
 671                 mddev->flush_bio = bio;
 672                 bio = NULL;
 673         }
 674         spin_unlock_irq(&mddev->lock);
 675
 676         if (!bio) {
 677                 INIT_WORK(&mddev->flush_work, submit_flushes);
 678                 queue_work(md_wq, &mddev->flush_work);
 679         } else {
 680                 /* flush was performed for some other bio while we waited. */
 681                 if (bio->bi_iter.bi_size == 0)
 682                         /* an empty barrier - all done */
 683                         bio_endio(bio);
 684                 else {
 685                         bio->bi_opf &= ~REQ_PREFLUSH;
 686                         return false;
 687                 }
 688         }
 689         return true;
 690 }
 691 EXPORT_SYMBOL(md_flush_request);
 692
 693 static inline struct mddev *mddev_get(struct mddev *mddev)
 694 {
 695         atomic_inc(&mddev->active);
 696         return mddev;
 697 }
 698
 699 static void mddev_delayed_delete(struct work_struct *ws);
 700
 701 static void mddev_put(struct mddev *mddev)
 702 {
 703         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 704                 return;
 705         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 706             mddev->ctime == 0 && !mddev->hold_active) {
 707                 /* Array is not configured at all, and not held active,
 708                  * so destroy it */
 709                 list_del_init(&mddev->all_mddevs);
 710
 711                 /*
 712                  * Call queue_work inside the spinlock so that
 713                  * flush_workqueue() after mddev_find will succeed in waiting
 714                  * for the work to be done.
 715                  */
 716                 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 717                 queue_work(md_misc_wq, &mddev->del_work);
 718         }
 719         spin_unlock(&all_mddevs_lock);
 720 }
 721
 722 static void md_safemode_timeout(struct timer_list *t);
 723
 724 void mddev_init(struct mddev *mddev)
 725 {
 726         kobject_init(&mddev->kobj, &md_ktype);
 727         mutex_init(&mddev->open_mutex);
 728         mutex_init(&mddev->reconfig_mutex);
 729         mutex_init(&mddev->bitmap_info.mutex);
 730         INIT_LIST_HEAD(&mddev->disks);
 731         INIT_LIST_HEAD(&mddev->all_mddevs);
 732         timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
 733         atomic_set(&mddev->active, 1);
 734         atomic_set(&mddev->openers, 0);
 735         atomic_set(&mddev->active_io, 0);
 736         spin_lock_init(&mddev->lock);
 737         atomic_set(&mddev->flush_pending, 0);
 738         init_waitqueue_head(&mddev->sb_wait);
 739         init_waitqueue_head(&mddev->recovery_wait);
 740         mddev->reshape_position = MaxSector;
 741         mddev->reshape_backwards = 0;
 742         mddev->last_sync_action = "none";
 743         mddev->resync_min = 0;
 744         mddev->resync_max = MaxSector;
 745         mddev->level = LEVEL_NONE;
 746 }
 747 EXPORT_SYMBOL_GPL(mddev_init);
 748
 749 static struct mddev *mddev_find(dev_t unit)
 750 {
 751         struct mddev *mddev, *new = NULL;
 752
 753         if (unit && MAJOR(unit) != MD_MAJOR)
 754                 unit &= ~((1<<MdpMinorShift)-1);
 755
 756  retry:
 757         spin_lock(&all_mddevs_lock);
 758
 759         if (unit) {
 760                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 761                         if (mddev->unit == unit) {
 762                                 mddev_get(mddev);
 763                                 spin_unlock(&all_mddevs_lock);
 764                                 kfree(new);
 765                                 return mddev;
 766                         }
 767
 768                 if (new) {
 769                         list_add(&new->all_mddevs, &all_mddevs);
 770                         spin_unlock(&all_mddevs_lock);
 771                         new->hold_active = UNTIL_IOCTL;
 772                         return new;
 773                 }
 774         } else if (new) {
 775                 /* find an unused unit number */
 776                 static int next_minor = 512;
 777                 int start = next_minor;
 778                 int is_free = 0;
 779                 int dev = 0;
 780                 while (!is_free) {
 781                         dev = MKDEV(MD_MAJOR, next_minor);
 782                         next_minor++;
 783                         if (next_minor > MINORMASK)
 784                                 next_minor = 0;
 785                         if (next_minor == start) {
 786                                 /* Oh dear, all in use. */
 787                                 spin_unlock(&all_mddevs_lock);
 788                                 kfree(new);
 789                                 return NULL;
 790                         }
 791
 792                         is_free = 1;
 793                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 794                                 if (mddev->unit == dev) {
 795                                         is_free = 0;
 796                                         break;
 797                                 }
 798                 }
 799                 new->unit = dev;
 800                 new->md_minor = MINOR(dev);
 801                 new->hold_active = UNTIL_STOP;
 802                 list_add(&new->all_mddevs, &all_mddevs);
 803                 spin_unlock(&all_mddevs_lock);
 804                 return new;
 805         }
 806         spin_unlock(&all_mddevs_lock);
 807
 808         new = kzalloc(sizeof(*new), GFP_KERNEL);
 809         if (!new)
 810                 return NULL;
 811
 812         new->unit = unit;
 813         if (MAJOR(unit) == MD_MAJOR)
 814                 new->md_minor = MINOR(unit);
 815         else
 816                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 817
 818         mddev_init(new);
 819
 820         goto retry;
 821 }
 822
 823 static struct attribute_group md_redundancy_group;
 824
 825 void mddev_unlock(struct mddev *mddev)
 826 {
 827         if (mddev->to_remove) {
 828                 /* These cannot be removed under reconfig_mutex as
 829                  * an access to the files will try to take reconfig_mutex
 830                  * while holding the file unremovable, which leads to
 831                  * a deadlock.
 832                  * So hold set sysfs_active while the remove in happeing,
 833                  * and anything else which might set ->to_remove or my
 834                  * otherwise change the sysfs namespace will fail with
 835                  * -EBUSY if sysfs_active is still set.
 836                  * We set sysfs_active under reconfig_mutex and elsewhere
 837                  * test it under the same mutex to ensure its correct value
 838                  * is seen.
 839                  */
 840                 struct attribute_group *to_remove = mddev->to_remove;
 841                 mddev->to_remove = NULL;
 842                 mddev->sysfs_active = 1;
 843                 mutex_unlock(&mddev->reconfig_mutex);
 844
 845                 if (mddev->kobj.sd) {
 846                         if (to_remove != &md_redundancy_group)
 847                                 sysfs_remove_group(&mddev->kobj, to_remove);
 848                         if (mddev->pers == NULL ||
 849                             mddev->pers->sync_request == NULL) {
 850                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 851                                 if (mddev->sysfs_action)
 852                                         sysfs_put(mddev->sysfs_action);
 853                                 if (mddev->sysfs_completed)
 854                                         sysfs_put(mddev->sysfs_completed);
 855                                 if (mddev->sysfs_degraded)
 856                                         sysfs_put(mddev->sysfs_degraded);
 857                                 mddev->sysfs_action = NULL;
 858                                 mddev->sysfs_completed = NULL;
 859                                 mddev->sysfs_degraded = NULL;
 860                         }
 861                 }
 862                 mddev->sysfs_active = 0;
 863         } else
 864                 mutex_unlock(&mddev->reconfig_mutex);
 865
 866         /* As we've dropped the mutex we need a spinlock to
 867          * make sure the thread doesn't disappear
 868          */
 869         spin_lock(&pers_lock);
 870         md_wakeup_thread(mddev->thread);
 871         wake_up(&mddev->sb_wait);
 872         spin_unlock(&pers_lock);
 873 }
 874 EXPORT_SYMBOL_GPL(mddev_unlock);
 875
 876 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
 877 {
 878         struct md_rdev *rdev;
 879
 880         rdev_for_each_rcu(rdev, mddev)
 881                 if (rdev->desc_nr == nr)
 882                         return rdev;
 883
 884         return NULL;
 885 }
 886 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
 887
 888 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 889 {
 890         struct md_rdev *rdev;
 891
 892         rdev_for_each(rdev, mddev)
 893                 if (rdev->bdev->bd_dev == dev)
 894                         return rdev;
 895
 896         return NULL;
 897 }
 898
 899 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
 900 {
 901         struct md_rdev *rdev;
 902
 903         rdev_for_each_rcu(rdev, mddev)
 904                 if (rdev->bdev->bd_dev == dev)
 905                         return rdev;
 906
 907         return NULL;
 908 }
 909 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
 910
 911 static struct md_personality *find_pers(int level, char *clevel)
 912 {
 913         struct md_personality *pers;
 914         list_for_each_entry(pers, &pers_list, list) {
 915                 if (level != LEVEL_NONE && pers->level == level)
 916                         return pers;
 917                 if (strcmp(pers->name, clevel)==0)
 918                         return pers;
 919         }
 920         return NULL;
 921 }
 922
 923 /* return the offset of the super block in 512byte sectors */
 924 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 925 {
 926         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 927         return MD_NEW_SIZE_SECTORS(num_sectors);
 928 }
 929
 930 static int alloc_disk_sb(struct md_rdev *rdev)
 931 {
 932         rdev->sb_page = alloc_page(GFP_KERNEL);
 933         if (!rdev->sb_page)
 934                 return -ENOMEM;
 935         return 0;
 936 }
 937
 938 void md_rdev_clear(struct md_rdev *rdev)
 939 {
 940         if (rdev->sb_page) {
 941                 put_page(rdev->sb_page);
 942                 rdev->sb_loaded = 0;
 943                 rdev->sb_page = NULL;
 944                 rdev->sb_start = 0;
 945                 rdev->sectors = 0;
 946         }
 947         if (rdev->bb_page) {
 948                 put_page(rdev->bb_page);
 949                 rdev->bb_page = NULL;
 950         }
 951         badblocks_exit(&rdev->badblocks);
 952 }
 953 EXPORT_SYMBOL_GPL(md_rdev_clear);
 954
 955 static void super_written(struct bio *bio)
 956 {
 957         struct md_rdev *rdev = bio->bi_private;
 958         struct mddev *mddev = rdev->mddev;
 959
 960         if (bio->bi_status) {
 961                 pr_err("md: %s gets error=%d\n", __func__,
 962                        blk_status_to_errno(bio->bi_status));
 963                 md_error(mddev, rdev);
 964                 if (!test_bit(Faulty, &rdev->flags)
 965                     && (bio->bi_opf & MD_FAILFAST)) {
 966                         set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
 967                         set_bit(LastDev, &rdev->flags);
 968                 }
 969         } else
 970                 clear_bit(LastDev, &rdev->flags);
 971
 972         if (atomic_dec_and_test(&mddev->pending_writes))
 973                 wake_up(&mddev->sb_wait);
 974         rdev_dec_pending(rdev, mddev);
 975         bio_put(bio);
 976 }
 977
 978 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 979                    sector_t sector, int size, struct page *page)
 980 {
 981         /* write first size bytes of page to sector of rdev
 982          * Increment mddev->pending_writes before returning
 983          * and decrement it on completion, waking up sb_wait
 984          * if zero is reached.
 985          * If an error occurred, call md_error
 986          */
 987         struct bio *bio;
 988         int ff = 0;
 989
 990         if (!page)
 991                 return;
 992
 993         if (test_bit(Faulty, &rdev->flags))
 994                 return;
 995
 996         bio = md_bio_alloc_sync(mddev);
 997
 998         atomic_inc(&rdev->nr_pending);
 999
1000         bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
1001         bio->bi_iter.bi_sector = sector;
1002         bio_add_page(bio, page, size, 0);
1003         bio->bi_private = rdev;
1004         bio->bi_end_io = super_written;
1005
1006         if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1007             test_bit(FailFast, &rdev->flags) &&
1008             !test_bit(LastDev, &rdev->flags))
1009                 ff = MD_FAILFAST;
1010         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
1011
1012         atomic_inc(&mddev->pending_writes);
1013         submit_bio(bio);
1014 }
1015
1016 int md_super_wait(struct mddev *mddev)
1017 {
1018         /* wait for all superblock writes that were scheduled to complete */
1019         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1020         if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1021                 return -EAGAIN;
1022         return 0;
1023 }
1024
1025 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1026                  struct page *page, int op, int op_flags, bool metadata_op)
1027 {
1028         struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1029         int ret;
1030
1031         if (metadata_op && rdev->meta_bdev)
1032                 bio_set_dev(bio, rdev->meta_bdev);
1033         else
1034                 bio_set_dev(bio, rdev->bdev);
1035         bio_set_op_attrs(bio, op, op_flags);
1036         if (metadata_op)
1037                 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1038         else if (rdev->mddev->reshape_position != MaxSector &&
1039                  (rdev->mddev->reshape_backwards ==
1040                   (sector >= rdev->mddev->reshape_position)))
1041                 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
1042         else
1043                 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1044         bio_add_page(bio, page, size, 0);
1045
1046         submit_bio_wait(bio);
1047
1048         ret = !bio->bi_status;
1049         bio_put(bio);
1050         return ret;
1051 }
1052 EXPORT_SYMBOL_GPL(sync_page_io);
1053
1054 static int read_disk_sb(struct md_rdev *rdev, int size)
1055 {
1056         char b[BDEVNAME_SIZE];
1057
1058         if (rdev->sb_loaded)
1059                 return 0;
1060
1061         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1062                 goto fail;
1063         rdev->sb_loaded = 1;
1064         return 0;
1065
1066 fail:
1067         pr_err("md: disabled device %s, could not read superblock.\n",
1068                bdevname(rdev->bdev,b));
1069         return -EINVAL;
1070 }
1071
1072 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1073 {
1074         return  sb1->set_uuid0 == sb2->set_uuid0 &&
1075                 sb1->set_uuid1 == sb2->set_uuid1 &&
1076                 sb1->set_uuid2 == sb2->set_uuid2 &&
1077                 sb1->set_uuid3 == sb2->set_uuid3;
1078 }
1079
1080 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1081 {
1082         int ret;
1083         mdp_super_t *tmp1, *tmp2;
1084
1085         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1086         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1087
1088         if (!tmp1 || !tmp2) {
1089                 ret = 0;
1090                 goto abort;
1091         }
1092
1093         *tmp1 = *sb1;
1094         *tmp2 = *sb2;
1095
1096         /*
1097          * nr_disks is not constant
1098          */
1099         tmp1->nr_disks = 0;
1100         tmp2->nr_disks = 0;
1101
1102         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1103 abort:
1104         kfree(tmp1);
1105         kfree(tmp2);
1106         return ret;
1107 }
1108
1109 static u32 md_csum_fold(u32 csum)
1110 {
1111         csum = (csum & 0xffff) + (csum >> 16);
1112         return (csum & 0xffff) + (csum >> 16);
1113 }
1114
1115 static unsigned int calc_sb_csum(mdp_super_t *sb)
1116 {
1117         u64 newcsum = 0;
1118         u32 *sb32 = (u32*)sb;
1119         int i;
1120         unsigned int disk_csum, csum;
1121
1122         disk_csum = sb->sb_csum;
1123         sb->sb_csum = 0;
1124
1125         for (i = 0; i < MD_SB_BYTES/4 ; i++)
1126                 newcsum += sb32[i];
1127         csum = (newcsum & 0xffffffff) + (newcsum>>32);
1128
1129 #ifdef CONFIG_ALPHA
1130         /* This used to use csum_partial, which was wrong for several
1131          * reasons including that different results are returned on
1132          * different architectures.  It isn't critical that we get exactly
1133          * the same return value as before (we always csum_fold before
1134          * testing, and that removes any differences).  However as we
1135          * know that csum_partial always returned a 16bit value on
1136          * alphas, do a fold to maximise conformity to previous behaviour.
1137          */
1138         sb->sb_csum = md_csum_fold(disk_csum);
1139 #else
1140         sb->sb_csum = disk_csum;
1141 #endif
1142         return csum;
1143 }
1144
1145 /*
1146  * Handle superblock details.
1147  * We want to be able to handle multiple superblock formats
1148  * so we have a common interface to them all, and an array of
1149  * different handlers.
1150  * We rely on user-space to write the initial superblock, and support
1151  * reading and updating of superblocks.
1152  * Interface methods are:
1153  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1154  *      loads and validates a superblock on dev.
1155  *      if refdev != NULL, compare superblocks on both devices
1156  *    Return:
1157  *      0 - dev has a superblock that is compatible with refdev
1158  *      1 - dev has a superblock that is compatible and newer than refdev
1159  *          so dev should be used as the refdev in future
1160  *     -EINVAL superblock incompatible or invalid
1161  *     -othererror e.g. -EIO
1162  *
1163  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1164  *      Verify that dev is acceptable into mddev.
1165  *       The first time, mddev->raid_disks will be 0, and data from
1166  *       dev should be merged in.  Subsequent calls check that dev
1167  *       is new enough.  Return 0 or -EINVAL
1168  *
1169  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1170  *     Update the superblock for rdev with data in mddev
1171  *     This does not write to disc.
1172  *
1173  */
1174
1175 struct super_type  {
1176         char                *name;
1177         struct module       *owner;
1178         int                 (*load_super)(struct md_rdev *rdev,
1179                                           struct md_rdev *refdev,
1180                                           int minor_version);
1181         int                 (*validate_super)(struct mddev *mddev,
1182                                               struct md_rdev *rdev);
1183         void                (*sync_super)(struct mddev *mddev,
1184                                           struct md_rdev *rdev);
1185         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1186                                                 sector_t num_sectors);
1187         int                 (*allow_new_offset)(struct md_rdev *rdev,
1188                                                 unsigned long long new_offset);
1189 };
1190
1191 /*
1192  * Check that the given mddev has no bitmap.
1193  *
1194  * This function is called from the run method of all personalities that do not
1195  * support bitmaps. It prints an error message and returns non-zero if mddev
1196  * has a bitmap. Otherwise, it returns 0.
1197  *
1198  */
1199 int md_check_no_bitmap(struct mddev *mddev)
1200 {
1201         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1202                 return 0;
1203         pr_warn("%s: bitmaps are not supported for %s\n",
1204                 mdname(mddev), mddev->pers->name);
1205         return 1;
1206 }
1207 EXPORT_SYMBOL(md_check_no_bitmap);
1208
1209 /*
1210  * load_super for 0.90.0
1211  */
1212 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1213 {
1214         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1215         mdp_super_t *sb;
1216         int ret;
1217         bool spare_disk = true;
1218
1219         /*
1220          * Calculate the position of the superblock (512byte sectors),
1221          * it's at the end of the disk.
1222          *
1223          * It also happens to be a multiple of 4Kb.
1224          */
1225         rdev->sb_start = calc_dev_sboffset(rdev);
1226
1227         ret = read_disk_sb(rdev, MD_SB_BYTES);
1228         if (ret)
1229                 return ret;
1230
1231         ret = -EINVAL;
1232
1233         bdevname(rdev->bdev, b);
1234         sb = page_address(rdev->sb_page);
1235
1236         if (sb->md_magic != MD_SB_MAGIC) {
1237                 pr_warn("md: invalid raid superblock magic on %s\n", b);
1238                 goto abort;
1239         }
1240
1241         if (sb->major_version != 0 ||
1242             sb->minor_version < 90 ||
1243             sb->minor_version > 91) {
1244                 pr_warn("Bad version number %d.%d on %s\n",
1245                         sb->major_version, sb->minor_version, b);
1246                 goto abort;
1247         }
1248
1249         if (sb->raid_disks <= 0)
1250                 goto abort;
1251
1252         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1253                 pr_warn("md: invalid superblock checksum on %s\n", b);
1254                 goto abort;
1255         }
1256
1257         rdev->preferred_minor = sb->md_minor;
1258         rdev->data_offset = 0;
1259         rdev->new_data_offset = 0;
1260         rdev->sb_size = MD_SB_BYTES;
1261         rdev->badblocks.shift = -1;
1262
1263         if (sb->level == LEVEL_MULTIPATH)
1264                 rdev->desc_nr = -1;
1265         else
1266                 rdev->desc_nr = sb->this_disk.number;
1267
1268         /* not spare disk, or LEVEL_MULTIPATH */
1269         if (sb->level == LEVEL_MULTIPATH ||
1270                 (rdev->desc_nr >= 0 &&
1271                  rdev->desc_nr < MD_SB_DISKS &&
1272                  sb->disks[rdev->desc_nr].state &
1273                  ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1274                 spare_disk = false;
1275
1276         if (!refdev) {
1277                 if (!spare_disk)
1278                         ret = 1;
1279                 else
1280                         ret = 0;
1281         } else {
1282                 __u64 ev1, ev2;
1283                 mdp_super_t *refsb = page_address(refdev->sb_page);
1284                 if (!md_uuid_equal(refsb, sb)) {
1285                         pr_warn("md: %s has different UUID to %s\n",
1286                                 b, bdevname(refdev->bdev,b2));
1287                         goto abort;
1288                 }
1289                 if (!md_sb_equal(refsb, sb)) {
1290                         pr_warn("md: %s has same UUID but different superblock to %s\n",
1291                                 b, bdevname(refdev->bdev, b2));
1292                         goto abort;
1293                 }
1294                 ev1 = md_event(sb);
1295                 ev2 = md_event(refsb);
1296
1297                 if (!spare_disk && ev1 > ev2)
1298                         ret = 1;
1299                 else
1300                         ret = 0;
1301         }
1302         rdev->sectors = rdev->sb_start;
1303         /* Limit to 4TB as metadata cannot record more than that.
1304          * (not needed for Linear and RAID0 as metadata doesn't
1305          * record this size)
1306          */
1307         if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1308                 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1309
1310         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1311                 /* "this cannot possibly happen" ... */
1312                 ret = -EINVAL;
1313
1314  abort:
1315         return ret;
1316 }
1317
1318 /*
1319  * validate_super for 0.90.0
1320  */
1321 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1322 {
1323         mdp_disk_t *desc;
1324         mdp_super_t *sb = page_address(rdev->sb_page);
1325         __u64 ev1 = md_event(sb);
1326
1327         rdev->raid_disk = -1;
1328         clear_bit(Faulty, &rdev->flags);
1329         clear_bit(In_sync, &rdev->flags);
1330         clear_bit(Bitmap_sync, &rdev->flags);
1331         clear_bit(WriteMostly, &rdev->flags);
1332
1333         if (mddev->raid_disks == 0) {
1334                 mddev->major_version = 0;
1335                 mddev->minor_version = sb->minor_version;
1336                 mddev->patch_version = sb->patch_version;
1337                 mddev->external = 0;
1338                 mddev->chunk_sectors = sb->chunk_size >> 9;
1339                 mddev->ctime = sb->ctime;
1340                 mddev->utime = sb->utime;
1341                 mddev->level = sb->level;
1342                 mddev->clevel[0] = 0;
1343                 mddev->layout = sb->layout;
1344                 mddev->raid_disks = sb->raid_disks;
1345                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1346                 mddev->events = ev1;
1347                 mddev->bitmap_info.offset = 0;
1348                 mddev->bitmap_info.space = 0;
1349                 /* bitmap can use 60 K after the 4K superblocks */
1350                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1351                 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1352                 mddev->reshape_backwards = 0;
1353
1354                 if (mddev->minor_version >= 91) {
1355                         mddev->reshape_position = sb->reshape_position;
1356                         mddev->delta_disks = sb->delta_disks;
1357                         mddev->new_level = sb->new_level;
1358                         mddev->new_layout = sb->new_layout;
1359                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1360                         if (mddev->delta_disks < 0)
1361                                 mddev->reshape_backwards = 1;
1362                 } else {
1363                         mddev->reshape_position = MaxSector;
1364                         mddev->delta_disks = 0;
1365                         mddev->new_level = mddev->level;
1366                         mddev->new_layout = mddev->layout;
1367                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1368                 }
1369                 if (mddev->level == 0)
1370                         mddev->layout = -1;
1371
1372                 if (sb->state & (1<<MD_SB_CLEAN))
1373                         mddev->recovery_cp = MaxSector;
1374                 else {
1375                         if (sb->events_hi == sb->cp_events_hi &&
1376                                 sb->events_lo == sb->cp_events_lo) {
1377                                 mddev->recovery_cp = sb->recovery_cp;
1378                         } else
1379                                 mddev->recovery_cp = 0;
1380                 }
1381
1382                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1383                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1384                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1385                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1386
1387                 mddev->max_disks = MD_SB_DISKS;
1388
1389                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1390                     mddev->bitmap_info.file == NULL) {
1391                         mddev->bitmap_info.offset =
1392                                 mddev->bitmap_info.default_offset;
1393                         mddev->bitmap_info.space =
1394                                 mddev->bitmap_info.default_space;
1395                 }
1396
1397         } else if (mddev->pers == NULL) {
1398                 /* Insist on good event counter while assembling, except
1399                  * for spares (which don't need an event count) */
1400                 ++ev1;
1401                 if (sb->disks[rdev->desc_nr].state & (
1402                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1403                         if (ev1 < mddev->events)
1404                                 return -EINVAL;
1405         } else if (mddev->bitmap) {
1406                 /* if adding to array with a bitmap, then we can accept an
1407                  * older device ... but not too old.
1408                  */
1409                 if (ev1 < mddev->bitmap->events_cleared)
1410                         return 0;
1411                 if (ev1 < mddev->events)
1412                         set_bit(Bitmap_sync, &rdev->flags);
1413         } else {
1414                 if (ev1 < mddev->events)
1415                         /* just a hot-add of a new device, leave raid_disk at -1 */
1416                         return 0;
1417         }
1418
1419         if (mddev->level != LEVEL_MULTIPATH) {
1420                 desc = sb->disks + rdev->desc_nr;
1421
1422                 if (desc->state & (1<<MD_DISK_FAULTY))
1423                         set_bit(Faulty, &rdev->flags);
1424                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1425                             desc->raid_disk < mddev->raid_disks */) {
1426                         set_bit(In_sync, &rdev->flags);
1427                         rdev->raid_disk = desc->raid_disk;
1428                         rdev->saved_raid_disk = desc->raid_disk;
1429                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1430                         /* active but not in sync implies recovery up to
1431                          * reshape position.  We don't know exactly where
1432                          * that is, so set to zero for now */
1433                         if (mddev->minor_version >= 91) {
1434                                 rdev->recovery_offset = 0;
1435                                 rdev->raid_disk = desc->raid_disk;
1436                         }
1437                 }
1438                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1439                         set_bit(WriteMostly, &rdev->flags);
1440                 if (desc->state & (1<<MD_DISK_FAILFAST))
1441                         set_bit(FailFast, &rdev->flags);
1442         } else /* MULTIPATH are always insync */
1443                 set_bit(In_sync, &rdev->flags);
1444         return 0;
1445 }
1446
1447 /*
1448  * sync_super for 0.90.0
1449  */
1450 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1451 {
1452         mdp_super_t *sb;
1453         struct md_rdev *rdev2;
1454         int next_spare = mddev->raid_disks;
1455
1456         /* make rdev->sb match mddev data..
1457          *
1458          * 1/ zero out disks
1459          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1460          * 3/ any empty disks < next_spare become removed
1461          *
1462          * disks[0] gets initialised to REMOVED because
1463          * we cannot be sure from other fields if it has
1464          * been initialised or not.
1465          */
1466         int i;
1467         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1468
1469         rdev->sb_size = MD_SB_BYTES;
1470
1471         sb = page_address(rdev->sb_page);
1472
1473         memset(sb, 0, sizeof(*sb));
1474
1475         sb->md_magic = MD_SB_MAGIC;
1476         sb->major_version = mddev->major_version;
1477         sb->patch_version = mddev->patch_version;
1478         sb->gvalid_words  = 0; /* ignored */
1479         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1480         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1481         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1482         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1483
1484         sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1485         sb->level = mddev->level;
1486         sb->size = mddev->dev_sectors / 2;
1487         sb->raid_disks = mddev->raid_disks;
1488         sb->md_minor = mddev->md_minor;
1489         sb->not_persistent = 0;
1490         sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1491         sb->state = 0;
1492         sb->events_hi = (mddev->events>>32);
1493         sb->events_lo = (u32)mddev->events;
1494
1495         if (mddev->reshape_position == MaxSector)
1496                 sb->minor_version = 90;
1497         else {
1498                 sb->minor_version = 91;
1499                 sb->reshape_position = mddev->reshape_position;
1500                 sb->new_level = mddev->new_level;
1501                 sb->delta_disks = mddev->delta_disks;
1502                 sb->new_layout = mddev->new_layout;
1503                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1504         }
1505         mddev->minor_version = sb->minor_version;
1506         if (mddev->in_sync)
1507         {
1508                 sb->recovery_cp = mddev->recovery_cp;
1509                 sb->cp_events_hi = (mddev->events>>32);
1510                 sb->cp_events_lo = (u32)mddev->events;
1511                 if (mddev->recovery_cp == MaxSector)
1512                         sb->state = (1<< MD_SB_CLEAN);
1513         } else
1514                 sb->recovery_cp = 0;
1515
1516         sb->layout = mddev->layout;
1517         sb->chunk_size = mddev->chunk_sectors << 9;
1518
1519         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1520                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1521
1522         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1523         rdev_for_each(rdev2, mddev) {
1524                 mdp_disk_t *d;
1525                 int desc_nr;
1526                 int is_active = test_bit(In_sync, &rdev2->flags);
1527
1528                 if (rdev2->raid_disk >= 0 &&
1529                     sb->minor_version >= 91)
1530                         /* we have nowhere to store the recovery_offset,
1531                          * but if it is not below the reshape_position,
1532                          * we can piggy-back on that.
1533                          */
1534                         is_active = 1;
1535                 if (rdev2->raid_disk < 0 ||
1536                     test_bit(Faulty, &rdev2->flags))
1537                         is_active = 0;
1538                 if (is_active)
1539                         desc_nr = rdev2->raid_disk;
1540                 else
1541                         desc_nr = next_spare++;
1542                 rdev2->desc_nr = desc_nr;
1543                 d = &sb->disks[rdev2->desc_nr];
1544                 nr_disks++;
1545                 d->number = rdev2->desc_nr;
1546                 d->major = MAJOR(rdev2->bdev->bd_dev);
1547                 d->minor = MINOR(rdev2->bdev->bd_dev);
1548                 if (is_active)
1549                         d->raid_disk = rdev2->raid_disk;
1550                 else
1551                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1552                 if (test_bit(Faulty, &rdev2->flags))
1553                         d->state = (1<<MD_DISK_FAULTY);
1554                 else if (is_active) {
1555                         d->state = (1<<MD_DISK_ACTIVE);
1556                         if (test_bit(In_sync, &rdev2->flags))
1557                                 d->state |= (1<<MD_DISK_SYNC);
1558                         active++;
1559                         working++;
1560                 } else {
1561                         d->state = 0;
1562                         spare++;
1563                         working++;
1564                 }
1565                 if (test_bit(WriteMostly, &rdev2->flags))
1566                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1567                 if (test_bit(FailFast, &rdev2->flags))
1568                         d->state |= (1<<MD_DISK_FAILFAST);
1569         }
1570         /* now set the "removed" and "faulty" bits on any missing devices */
1571         for (i=0 ; i < mddev->raid_disks ; i++) {
1572                 mdp_disk_t *d = &sb->disks[i];
1573                 if (d->state == 0 && d->number == 0) {
1574                         d->number = i;
1575                         d->raid_disk = i;
1576                         d->state = (1<<MD_DISK_REMOVED);
1577                         d->state |= (1<<MD_DISK_FAULTY);
1578                         failed++;
1579                 }
1580         }
1581         sb->nr_disks = nr_disks;
1582         sb->active_disks = active;
1583         sb->working_disks = working;
1584         sb->failed_disks = failed;
1585         sb->spare_disks = spare;
1586
1587         sb->this_disk = sb->disks[rdev->desc_nr];
1588         sb->sb_csum = calc_sb_csum(sb);
1589 }
1590
1591 /*
1592  * rdev_size_change for 0.90.0
1593  */
1594 static unsigned long long
1595 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1596 {
1597         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1598                 return 0; /* component must fit device */
1599         if (rdev->mddev->bitmap_info.offset)
1600                 return 0; /* can't move bitmap */
1601         rdev->sb_start = calc_dev_sboffset(rdev);
1602         if (!num_sectors || num_sectors > rdev->sb_start)
1603                 num_sectors = rdev->sb_start;
1604         /* Limit to 4TB as metadata cannot record more than that.
1605          * 4TB == 2^32 KB, or 2*2^32 sectors.
1606          */
1607         if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1608                 num_sectors = (sector_t)(2ULL << 32) - 2;
1609         do {
1610                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1611                        rdev->sb_page);
1612         } while (md_super_wait(rdev->mddev) < 0);
1613         return num_sectors;
1614 }
1615
1616 static int
1617 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1618 {
1619         /* non-zero offset changes not possible with v0.90 */
1620         return new_offset == 0;
1621 }
1622
1623 /*
1624  * version 1 superblock
1625  */
1626
1627 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1628 {
1629         __le32 disk_csum;
1630         u32 csum;
1631         unsigned long long newcsum;
1632         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1633         __le32 *isuper = (__le32*)sb;
1634
1635         disk_csum = sb->sb_csum;
1636         sb->sb_csum = 0;
1637         newcsum = 0;
1638         for (; size >= 4; size -= 4)
1639                 newcsum += le32_to_cpu(*isuper++);
1640
1641         if (size == 2)
1642                 newcsum += le16_to_cpu(*(__le16*) isuper);
1643
1644         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1645         sb->sb_csum = disk_csum;
1646         return cpu_to_le32(csum);
1647 }
1648
1649 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1650 {
1651         struct mdp_superblock_1 *sb;
1652         int ret;
1653         sector_t sb_start;
1654         sector_t sectors;
1655         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1656         int bmask;
1657         bool spare_disk = true;
1658
1659         /*
1660          * Calculate the position of the superblock in 512byte sectors.
1661          * It is always aligned to a 4K boundary and
1662          * depeding on minor_version, it can be:
1663          * 0: At least 8K, but less than 12K, from end of device
1664          * 1: At start of device
1665          * 2: 4K from start of device.
1666          */
1667         switch(minor_version) {
1668         case 0:
1669                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1670                 sb_start -= 8*2;
1671                 sb_start &= ~(sector_t)(4*2-1);
1672                 break;
1673         case 1:
1674                 sb_start = 0;
1675                 break;
1676         case 2:
1677                 sb_start = 8;
1678                 break;
1679         default:
1680                 return -EINVAL;
1681         }
1682         rdev->sb_start = sb_start;
1683
1684         /* superblock is rarely larger than 1K, but it can be larger,
1685          * and it is safe to read 4k, so we do that
1686          */
1687         ret = read_disk_sb(rdev, 4096);
1688         if (ret) return ret;
1689
1690         sb = page_address(rdev->sb_page);
1691
1692         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1693             sb->major_version != cpu_to_le32(1) ||
1694             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1695             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1696             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1697                 return -EINVAL;
1698
1699         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1700                 pr_warn("md: invalid superblock checksum on %s\n",
1701                         bdevname(rdev->bdev,b));
1702                 return -EINVAL;
1703         }
1704         if (le64_to_cpu(sb->data_size) < 10) {
1705                 pr_warn("md: data_size too small on %s\n",
1706                         bdevname(rdev->bdev,b));
1707                 return -EINVAL;
1708         }
1709         if (sb->pad0 ||
1710             sb->pad3[0] ||
1711             memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1712                 /* Some padding is non-zero, might be a new feature */
1713                 return -EINVAL;
1714
1715         rdev->preferred_minor = 0xffff;
1716         rdev->data_offset = le64_to_cpu(sb->data_offset);
1717         rdev->new_data_offset = rdev->data_offset;
1718         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1719             (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1720                 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1721         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1722
1723         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1724         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1725         if (rdev->sb_size & bmask)
1726                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1727
1728         if (minor_version
1729             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1730                 return -EINVAL;
1731         if (minor_version
1732             && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1733                 return -EINVAL;
1734
1735         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1736                 rdev->desc_nr = -1;
1737         else
1738                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1739
1740         if (!rdev->bb_page) {
1741                 rdev->bb_page = alloc_page(GFP_KERNEL);
1742                 if (!rdev->bb_page)
1743                         return -ENOMEM;
1744         }
1745         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1746             rdev->badblocks.count == 0) {
1747                 /* need to load the bad block list.
1748                  * Currently we limit it to one page.
1749                  */
1750                 s32 offset;
1751                 sector_t bb_sector;
1752                 __le64 *bbp;
1753                 int i;
1754                 int sectors = le16_to_cpu(sb->bblog_size);
1755                 if (sectors > (PAGE_SIZE / 512))
1756                         return -EINVAL;
1757                 offset = le32_to_cpu(sb->bblog_offset);
1758                 if (offset == 0)
1759                         return -EINVAL;
1760                 bb_sector = (long long)offset;
1761                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1762                                   rdev->bb_page, REQ_OP_READ, 0, true))
1763                         return -EIO;
1764                 bbp = (__le64 *)page_address(rdev->bb_page);
1765                 rdev->badblocks.shift = sb->bblog_shift;
1766                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1767                         u64 bb = le64_to_cpu(*bbp);
1768                         int count = bb & (0x3ff);
1769                         u64 sector = bb >> 10;
1770                         sector <<= sb->bblog_shift;
1771                         count <<= sb->bblog_shift;
1772                         if (bb + 1 == 0)
1773                                 break;
1774                         if (badblocks_set(&rdev->badblocks, sector, count, 1))
1775                                 return -EINVAL;
1776                 }
1777         } else if (sb->bblog_offset != 0)
1778                 rdev->badblocks.shift = 0;
1779
1780         if ((le32_to_cpu(sb->feature_map) &
1781             (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1782                 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1783                 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1784                 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1785         }
1786
1787         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1788             sb->level != 0)
1789                 return -EINVAL;
1790
1791         /* not spare disk, or LEVEL_MULTIPATH */
1792         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1793                 (rdev->desc_nr >= 0 &&
1794                 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1795                 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1796                  le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1797                 spare_disk = false;
1798
1799         if (!refdev) {
1800                 if (!spare_disk)
1801                         ret = 1;
1802                 else
1803                         ret = 0;
1804         } else {
1805                 __u64 ev1, ev2;
1806                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1807
1808                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1809                     sb->level != refsb->level ||
1810                     sb->layout != refsb->layout ||
1811                     sb->chunksize != refsb->chunksize) {
1812                         pr_warn("md: %s has strangely different superblock to %s\n",
1813                                 bdevname(rdev->bdev,b),
1814                                 bdevname(refdev->bdev,b2));
1815                         return -EINVAL;
1816                 }
1817                 ev1 = le64_to_cpu(sb->events);
1818                 ev2 = le64_to_cpu(refsb->events);
1819
1820                 if (!spare_disk && ev1 > ev2)
1821                         ret = 1;
1822                 else
1823                         ret = 0;
1824         }
1825         if (minor_version) {
1826                 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1827                 sectors -= rdev->data_offset;
1828         } else
1829                 sectors = rdev->sb_start;
1830         if (sectors < le64_to_cpu(sb->data_size))
1831                 return -EINVAL;
1832         rdev->sectors = le64_to_cpu(sb->data_size);
1833         return ret;
1834 }
1835
1836 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1837 {
1838         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1839         __u64 ev1 = le64_to_cpu(sb->events);
1840
1841         rdev->raid_disk = -1;
1842         clear_bit(Faulty, &rdev->flags);
1843         clear_bit(In_sync, &rdev->flags);
1844         clear_bit(Bitmap_sync, &rdev->flags);
1845         clear_bit(WriteMostly, &rdev->flags);
1846
1847         if (mddev->raid_disks == 0) {
1848                 mddev->major_version = 1;
1849                 mddev->patch_version = 0;
1850                 mddev->external = 0;
1851                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1852                 mddev->ctime = le64_to_cpu(sb->ctime);
1853                 mddev->utime = le64_to_cpu(sb->utime);
1854                 mddev->level = le32_to_cpu(sb->level);
1855                 mddev->clevel[0] = 0;
1856                 mddev->layout = le32_to_cpu(sb->layout);
1857                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1858                 mddev->dev_sectors = le64_to_cpu(sb->size);
1859                 mddev->events = ev1;
1860                 mddev->bitmap_info.offset = 0;
1861                 mddev->bitmap_info.space = 0;
1862                 /* Default location for bitmap is 1K after superblock
1863                  * using 3K - total of 4K
1864                  */
1865                 mddev->bitmap_info.default_offset = 1024 >> 9;
1866                 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1867                 mddev->reshape_backwards = 0;
1868
1869                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1870                 memcpy(mddev->uuid, sb->set_uuid, 16);
1871
1872                 mddev->max_disks =  (4096-256)/2;
1873
1874                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1875                     mddev->bitmap_info.file == NULL) {
1876                         mddev->bitmap_info.offset =
1877                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1878                         /* Metadata doesn't record how much space is available.
1879                          * For 1.0, we assume we can use up to the superblock
1880                          * if before, else to 4K beyond superblock.
1881                          * For others, assume no change is possible.
1882                          */
1883                         if (mddev->minor_version > 0)
1884                                 mddev->bitmap_info.space = 0;
1885                         else if (mddev->bitmap_info.offset > 0)
1886                                 mddev->bitmap_info.space =
1887                                         8 - mddev->bitmap_info.offset;
1888                         else
1889                                 mddev->bitmap_info.space =
1890                                         -mddev->bitmap_info.offset;
1891                 }
1892
1893                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1894                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1895                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1896                         mddev->new_level = le32_to_cpu(sb->new_level);
1897                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1898                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1899                         if (mddev->delta_disks < 0 ||
1900                             (mddev->delta_disks == 0 &&
1901                              (le32_to_cpu(sb->feature_map)
1902                               & MD_FEATURE_RESHAPE_BACKWARDS)))
1903                                 mddev->reshape_backwards = 1;
1904                 } else {
1905                         mddev->reshape_position = MaxSector;
1906                         mddev->delta_disks = 0;
1907                         mddev->new_level = mddev->level;
1908                         mddev->new_layout = mddev->layout;
1909                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1910                 }
1911
1912                 if (mddev->level == 0 &&
1913                     !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1914                         mddev->layout = -1;
1915
1916                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1917                         set_bit(MD_HAS_JOURNAL, &mddev->flags);
1918
1919                 if (le32_to_cpu(sb->feature_map) &
1920                     (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1921                         if (le32_to_cpu(sb->feature_map) &
1922                             (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1923                                 return -EINVAL;
1924                         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1925                             (le32_to_cpu(sb->feature_map) &
1926                                             MD_FEATURE_MULTIPLE_PPLS))
1927                                 return -EINVAL;
1928                         set_bit(MD_HAS_PPL, &mddev->flags);
1929                 }
1930         } else if (mddev->pers == NULL) {
1931                 /* Insist of good event counter while assembling, except for
1932                  * spares (which don't need an event count) */
1933                 ++ev1;
1934                 if (rdev->desc_nr >= 0 &&
1935                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1936                     (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1937                      le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1938                         if (ev1 < mddev->events)
1939                                 return -EINVAL;
1940         } else if (mddev->bitmap) {
1941                 /* If adding to array with a bitmap, then we can accept an
1942                  * older device, but not too old.
1943                  */
1944                 if (ev1 < mddev->bitmap->events_cleared)
1945                         return 0;
1946                 if (ev1 < mddev->events)
1947                         set_bit(Bitmap_sync, &rdev->flags);
1948         } else {
1949                 if (ev1 < mddev->events)
1950                         /* just a hot-add of a new device, leave raid_disk at -1 */
1951                         return 0;
1952         }
1953         if (mddev->level != LEVEL_MULTIPATH) {
1954                 int role;
1955                 if (rdev->desc_nr < 0 ||
1956                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1957                         role = MD_DISK_ROLE_SPARE;
1958                         rdev->desc_nr = -1;
1959                 } else
1960                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1961                 switch(role) {
1962                 case MD_DISK_ROLE_SPARE: /* spare */
1963                         break;
1964                 case MD_DISK_ROLE_FAULTY: /* faulty */
1965                         set_bit(Faulty, &rdev->flags);
1966                         break;
1967                 case MD_DISK_ROLE_JOURNAL: /* journal device */
1968                         if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1969                                 /* journal device without journal feature */
1970                                 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1971                                 return -EINVAL;
1972                         }
1973                         set_bit(Journal, &rdev->flags);
1974                         rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1975                         rdev->raid_disk = 0;
1976                         break;
1977                 default:
1978                         rdev->saved_raid_disk = role;
1979                         if ((le32_to_cpu(sb->feature_map) &
1980                              MD_FEATURE_RECOVERY_OFFSET)) {
1981                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1982                                 if (!(le32_to_cpu(sb->feature_map) &
1983                                       MD_FEATURE_RECOVERY_BITMAP))
1984                                         rdev->saved_raid_disk = -1;
1985                         } else {
1986                                 /*
1987                                  * If the array is FROZEN, then the device can't
1988                                  * be in_sync with rest of array.
1989                                  */
1990                                 if (!test_bit(MD_RECOVERY_FROZEN,
1991                                               &mddev->recovery))
1992                                         set_bit(In_sync, &rdev->flags);
1993                         }
1994                         rdev->raid_disk = role;
1995                         break;
1996                 }
1997                 if (sb->devflags & WriteMostly1)
1998                         set_bit(WriteMostly, &rdev->flags);
1999                 if (sb->devflags & FailFast1)
2000                         set_bit(FailFast, &rdev->flags);
2001                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2002                         set_bit(Replacement, &rdev->flags);
2003         } else /* MULTIPATH are always insync */
2004                 set_bit(In_sync, &rdev->flags);
2005
2006         return 0;
2007 }
2008
2009 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2010 {
2011         struct mdp_superblock_1 *sb;
2012         struct md_rdev *rdev2;
2013         int max_dev, i;
2014         /* make rdev->sb match mddev and rdev data. */
2015
2016         sb = page_address(rdev->sb_page);
2017
2018         sb->feature_map = 0;
2019         sb->pad0 = 0;
2020         sb->recovery_offset = cpu_to_le64(0);
2021         memset(sb->pad3, 0, sizeof(sb->pad3));
2022
2023         sb->utime = cpu_to_le64((__u64)mddev->utime);
2024         sb->events = cpu_to_le64(mddev->events);
2025         if (mddev->in_sync)
2026                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2027         else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2028                 sb->resync_offset = cpu_to_le64(MaxSector);
2029         else
2030                 sb->resync_offset = cpu_to_le64(0);
2031
2032         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2033
2034         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2035         sb->size = cpu_to_le64(mddev->dev_sectors);
2036         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2037         sb->level = cpu_to_le32(mddev->level);
2038         sb->layout = cpu_to_le32(mddev->layout);
2039         if (test_bit(FailFast, &rdev->flags))
2040                 sb->devflags |= FailFast1;
2041         else
2042                 sb->devflags &= ~FailFast1;
2043
2044         if (test_bit(WriteMostly, &rdev->flags))
2045                 sb->devflags |= WriteMostly1;
2046         else
2047                 sb->devflags &= ~WriteMostly1;
2048         sb->data_offset = cpu_to_le64(rdev->data_offset);
2049         sb->data_size = cpu_to_le64(rdev->sectors);
2050
2051         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2052                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2053                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2054         }
2055
2056         if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2057             !test_bit(In_sync, &rdev->flags)) {
2058                 sb->feature_map |=
2059                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2060                 sb->recovery_offset =
2061                         cpu_to_le64(rdev->recovery_offset);
2062                 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2063                         sb->feature_map |=
2064                                 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2065         }
2066         /* Note: recovery_offset and journal_tail share space  */
2067         if (test_bit(Journal, &rdev->flags))
2068                 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2069         if (test_bit(Replacement, &rdev->flags))
2070                 sb->feature_map |=
2071                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
2072
2073         if (mddev->reshape_position != MaxSector) {
2074                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2075                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2076                 sb->new_layout = cpu_to_le32(mddev->new_layout);
2077                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2078                 sb->new_level = cpu_to_le32(mddev->new_level);
2079                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2080                 if (mddev->delta_disks == 0 &&
2081                     mddev->reshape_backwards)
2082                         sb->feature_map
2083                                 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2084                 if (rdev->new_data_offset != rdev->data_offset) {
2085                         sb->feature_map
2086                                 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2087                         sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2088                                                              - rdev->data_offset));
2089                 }
2090         }
2091
2092         if (mddev_is_clustered(mddev))
2093                 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2094
2095         if (rdev->badblocks.count == 0)
2096                 /* Nothing to do for bad blocks*/ ;
2097         else if (sb->bblog_offset == 0)
2098                 /* Cannot record bad blocks on this device */
2099                 md_error(mddev, rdev);
2100         else {
2101                 struct badblocks *bb = &rdev->badblocks;
2102                 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2103                 u64 *p = bb->page;
2104                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2105                 if (bb->changed) {
2106                         unsigned seq;
2107
2108 retry:
2109                         seq = read_seqbegin(&bb->lock);
2110
2111                         memset(bbp, 0xff, PAGE_SIZE);
2112
2113                         for (i = 0 ; i < bb->count ; i++) {
2114                                 u64 internal_bb = p[i];
2115                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2116                                                 | BB_LEN(internal_bb));
2117                                 bbp[i] = cpu_to_le64(store_bb);
2118                         }
2119                         bb->changed = 0;
2120                         if (read_seqretry(&bb->lock, seq))
2121                                 goto retry;
2122
2123                         bb->sector = (rdev->sb_start +
2124                                       (int)le32_to_cpu(sb->bblog_offset));
2125                         bb->size = le16_to_cpu(sb->bblog_size);
2126                 }
2127         }
2128
2129         max_dev = 0;
2130         rdev_for_each(rdev2, mddev)
2131                 if (rdev2->desc_nr+1 > max_dev)
2132                         max_dev = rdev2->desc_nr+1;
2133
2134         if (max_dev > le32_to_cpu(sb->max_dev)) {
2135                 int bmask;
2136                 sb->max_dev = cpu_to_le32(max_dev);
2137                 rdev->sb_size = max_dev * 2 + 256;
2138                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2139                 if (rdev->sb_size & bmask)
2140                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
2141         } else
2142                 max_dev = le32_to_cpu(sb->max_dev);
2143
2144         for (i=0; i<max_dev;i++)
2145                 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2146
2147         if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2148                 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2149
2150         if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2151                 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2152                         sb->feature_map |=
2153                             cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2154                 else
2155                         sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2156                 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2157                 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2158         }
2159
2160         rdev_for_each(rdev2, mddev) {
2161                 i = rdev2->desc_nr;
2162                 if (test_bit(Faulty, &rdev2->flags))
2163                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2164                 else if (test_bit(In_sync, &rdev2->flags))
2165                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2166                 else if (test_bit(Journal, &rdev2->flags))
2167                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2168                 else if (rdev2->raid_disk >= 0)
2169                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2170                 else
2171                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2172         }
2173
2174         sb->sb_csum = calc_sb_1_csum(sb);
2175 }
2176
2177 static sector_t super_1_choose_bm_space(sector_t dev_size)
2178 {
2179         sector_t bm_space;
2180
2181         /* if the device is bigger than 8Gig, save 64k for bitmap
2182          * usage, if bigger than 200Gig, save 128k
2183          */
2184         if (dev_size < 64*2)
2185                 bm_space = 0;
2186         else if (dev_size - 64*2 >= 200*1024*1024*2)
2187                 bm_space = 128*2;
2188         else if (dev_size - 4*2 > 8*1024*1024*2)
2189                 bm_space = 64*2;
2190         else
2191                 bm_space = 4*2;
2192         return bm_space;
2193 }
2194
2195 static unsigned long long
2196 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2197 {
2198         struct mdp_superblock_1 *sb;
2199         sector_t max_sectors;
2200         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2201                 return 0; /* component must fit device */
2202         if (rdev->data_offset != rdev->new_data_offset)
2203                 return 0; /* too confusing */
2204         if (rdev->sb_start < rdev->data_offset) {
2205                 /* minor versions 1 and 2; superblock before data */
2206                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2207                 max_sectors -= rdev->data_offset;
2208                 if (!num_sectors || num_sectors > max_sectors)
2209                         num_sectors = max_sectors;
2210         } else if (rdev->mddev->bitmap_info.offset) {
2211                 /* minor version 0 with bitmap we can't move */
2212                 return 0;
2213         } else {
2214                 /* minor version 0; superblock after data */
2215                 sector_t sb_start, bm_space;
2216                 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2217
2218                 /* 8K is for superblock */
2219                 sb_start = dev_size - 8*2;
2220                 sb_start &= ~(sector_t)(4*2 - 1);
2221
2222                 bm_space = super_1_choose_bm_space(dev_size);
2223
2224                 /* Space that can be used to store date needs to decrease
2225                  * superblock bitmap space and bad block space(4K)
2226                  */
2227                 max_sectors = sb_start - bm_space - 4*2;
2228
2229                 if (!num_sectors || num_sectors > max_sectors)
2230                         num_sectors = max_sectors;
2231         }
2232         sb = page_address(rdev->sb_page);
2233         sb->data_size = cpu_to_le64(num_sectors);
2234         sb->super_offset = cpu_to_le64(rdev->sb_start);
2235         sb->sb_csum = calc_sb_1_csum(sb);
2236         do {
2237                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2238                                rdev->sb_page);
2239         } while (md_super_wait(rdev->mddev) < 0);
2240         return num_sectors;
2241
2242 }
2243
2244 static int
2245 super_1_allow_new_offset(struct md_rdev *rdev,
2246                          unsigned long long new_offset)
2247 {
2248         /* All necessary checks on new >= old have been done */
2249         struct bitmap *bitmap;
2250         if (new_offset >= rdev->data_offset)
2251                 return 1;
2252
2253         /* with 1.0 metadata, there is no metadata to tread on
2254          * so we can always move back */
2255         if (rdev->mddev->minor_version == 0)
2256                 return 1;
2257
2258         /* otherwise we must be sure not to step on
2259          * any metadata, so stay:
2260          * 36K beyond start of superblock
2261          * beyond end of badblocks
2262          * beyond write-intent bitmap
2263          */
2264         if (rdev->sb_start + (32+4)*2 > new_offset)
2265                 return 0;
2266         bitmap = rdev->mddev->bitmap;
2267         if (bitmap && !rdev->mddev->bitmap_info.file &&
2268             rdev->sb_start + rdev->mddev->bitmap_info.offset +
2269             bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2270                 return 0;
2271         if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2272                 return 0;
2273
2274         return 1;
2275 }
2276
2277 static struct super_type super_types[] = {
2278         [0] = {
2279                 .name   = "0.90.0",
2280                 .owner  = THIS_MODULE,
2281                 .load_super         = super_90_load,
2282                 .validate_super     = super_90_validate,
2283                 .sync_super         = super_90_sync,
2284                 .rdev_size_change   = super_90_rdev_size_change,
2285                 .allow_new_offset   = super_90_allow_new_offset,
2286         },
2287         [1] = {
2288                 .name   = "md-1",
2289                 .owner  = THIS_MODULE,
2290                 .load_super         = super_1_load,
2291                 .validate_super     = super_1_validate,
2292                 .sync_super         = super_1_sync,
2293                 .rdev_size_change   = super_1_rdev_size_change,
2294                 .allow_new_offset   = super_1_allow_new_offset,
2295         },
2296 };
2297
2298 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2299 {
2300         if (mddev->sync_super) {
2301                 mddev->sync_super(mddev, rdev);
2302                 return;
2303         }
2304
2305         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2306
2307         super_types[mddev->major_version].sync_super(mddev, rdev);
2308 }
2309
2310 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2311 {
2312         struct md_rdev *rdev, *rdev2;
2313
2314         rcu_read_lock();
2315         rdev_for_each_rcu(rdev, mddev1) {
2316                 if (test_bit(Faulty, &rdev->flags) ||
2317                     test_bit(Journal, &rdev->flags) ||
2318                     rdev->raid_disk == -1)
2319                         continue;
2320                 rdev_for_each_rcu(rdev2, mddev2) {
2321                         if (test_bit(Faulty, &rdev2->flags) ||
2322                             test_bit(Journal, &rdev2->flags) ||
2323                             rdev2->raid_disk == -1)
2324                                 continue;
2325                         if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2326                                 rcu_read_unlock();
2327                                 return 1;
2328                         }
2329                 }
2330         }
2331         rcu_read_unlock();
2332         return 0;
2333 }
2334
2335 static LIST_HEAD(pending_raid_disks);
2336
2337 /*
2338  * Try to register data integrity profile for an mddev
2339  *
2340  * This is called when an array is started and after a disk has been kicked
2341  * from the array. It only succeeds if all working and active component devices
2342  * are integrity capable with matching profiles.
2343  */
2344 int md_integrity_register(struct mddev *mddev)
2345 {
2346         struct md_rdev *rdev, *reference = NULL;
2347
2348         if (list_empty(&mddev->disks))
2349                 return 0; /* nothing to do */
2350         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2351                 return 0; /* shouldn't register, or already is */
2352         rdev_for_each(rdev, mddev) {
2353                 /* skip spares and non-functional disks */
2354                 if (test_bit(Faulty, &rdev->flags))
2355                         continue;
2356                 if (rdev->raid_disk < 0)
2357                         continue;
2358                 if (!reference) {
2359                         /* Use the first rdev as the reference */
2360                         reference = rdev;
2361                         continue;
2362                 }
2363                 /* does this rdev's profile match the reference profile? */
2364                 if (blk_integrity_compare(reference->bdev->bd_disk,
2365                                 rdev->bdev->bd_disk) < 0)
2366                         return -EINVAL;
2367         }
2368         if (!reference || !bdev_get_integrity(reference->bdev))
2369                 return 0;
2370         /*
2371          * All component devices are integrity capable and have matching
2372          * profiles, register the common profile for the md device.
2373          */
2374         blk_integrity_register(mddev->gendisk,
2375                                bdev_get_integrity(reference->bdev));
2376
2377         pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2378         if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2379                 pr_err("md: failed to create integrity pool for %s\n",
2380                        mdname(mddev));
2381                 return -EINVAL;
2382         }
2383         return 0;
2384 }
2385 EXPORT_SYMBOL(md_integrity_register);
2386
2387 /*
2388  * Attempt to add an rdev, but only if it is consistent with the current
2389  * integrity profile
2390  */
2391 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2392 {
2393         struct blk_integrity *bi_mddev;
2394         char name[BDEVNAME_SIZE];
2395
2396         if (!mddev->gendisk)
2397                 return 0;
2398
2399         bi_mddev = blk_get_integrity(mddev->gendisk);
2400
2401         if (!bi_mddev) /* nothing to do */
2402                 return 0;
2403
2404         if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2405                 pr_err("%s: incompatible integrity profile for %s\n",
2406                        mdname(mddev), bdevname(rdev->bdev, name));
2407                 return -ENXIO;
2408         }
2409
2410         return 0;
2411 }
2412 EXPORT_SYMBOL(md_integrity_add_rdev);
2413
2414 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2415 {
2416         char b[BDEVNAME_SIZE];
2417         struct kobject *ko;
2418         int err;
2419
2420         /* prevent duplicates */
2421         if (find_rdev(mddev, rdev->bdev->bd_dev))
2422                 return -EEXIST;
2423
2424         if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2425             mddev->pers)
2426                 return -EROFS;
2427
2428         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2429         if (!test_bit(Journal, &rdev->flags) &&
2430             rdev->sectors &&
2431             (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2432                 if (mddev->pers) {
2433                         /* Cannot change size, so fail
2434                          * If mddev->level <= 0, then we don't care
2435                          * about aligning sizes (e.g. linear)
2436                          */
2437                         if (mddev->level > 0)
2438                                 return -ENOSPC;
2439                 } else
2440                         mddev->dev_sectors = rdev->sectors;
2441         }
2442
2443         /* Verify rdev->desc_nr is unique.
2444          * If it is -1, assign a free number, else
2445          * check number is not in use
2446          */
2447         rcu_read_lock();
2448         if (rdev->desc_nr < 0) {
2449                 int choice = 0;
2450                 if (mddev->pers)
2451                         choice = mddev->raid_disks;
2452                 while (md_find_rdev_nr_rcu(mddev, choice))
2453                         choice++;
2454                 rdev->desc_nr = choice;
2455         } else {
2456                 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2457                         rcu_read_unlock();
2458                         return -EBUSY;
2459                 }
2460         }
2461         rcu_read_unlock();
2462         if (!test_bit(Journal, &rdev->flags) &&
2463             mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2464                 pr_warn("md: %s: array is limited to %d devices\n",
2465                         mdname(mddev), mddev->max_disks);
2466                 return -EBUSY;
2467         }
2468         bdevname(rdev->bdev,b);
2469         strreplace(b, '/', '!');
2470
2471         rdev->mddev = mddev;
2472         pr_debug("md: bind<%s>\n", b);
2473
2474         if (mddev->raid_disks)
2475                 mddev_create_serial_pool(mddev, rdev, false);
2476
2477         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2478                 goto fail;
2479
2480         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2481         /* failure here is OK */
2482         err = sysfs_create_link(&rdev->kobj, ko, "block");
2483         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2484         rdev->sysfs_unack_badblocks =
2485                 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2486         rdev->sysfs_badblocks =
2487                 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2488
2489         list_add_rcu(&rdev->same_set, &mddev->disks);
2490         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2491
2492         /* May as well allow recovery to be retried once */
2493         mddev->recovery_disabled++;
2494
2495         return 0;
2496
2497  fail:
2498         pr_warn("md: failed to register dev-%s for %s\n",
2499                 b, mdname(mddev));
2500         return err;
2501 }
2502
2503 static void rdev_delayed_delete(struct work_struct *ws)
2504 {
2505         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2506         kobject_del(&rdev->kobj);
2507         kobject_put(&rdev->kobj);
2508 }
2509
2510 static void unbind_rdev_from_array(struct md_rdev *rdev)
2511 {
2512         char b[BDEVNAME_SIZE];
2513
2514         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2515         list_del_rcu(&rdev->same_set);
2516         pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2517         mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2518         rdev->mddev = NULL;
2519         sysfs_remove_link(&rdev->kobj, "block");
2520         sysfs_put(rdev->sysfs_state);
2521         sysfs_put(rdev->sysfs_unack_badblocks);
2522         sysfs_put(rdev->sysfs_badblocks);
2523         rdev->sysfs_state = NULL;
2524         rdev->sysfs_unack_badblocks = NULL;
2525         rdev->sysfs_badblocks = NULL;
2526         rdev->badblocks.count = 0;
2527         /* We need to delay this, otherwise we can deadlock when
2528          * writing to 'remove' to "dev/state".  We also need
2529          * to delay it due to rcu usage.
2530          */
2531         synchronize_rcu();
2532         INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2533         kobject_get(&rdev->kobj);
2534         queue_work(md_rdev_misc_wq, &rdev->del_work);
2535 }
2536
2537 /*
2538  * prevent the device from being mounted, repartitioned or
2539  * otherwise reused by a RAID array (or any other kernel
2540  * subsystem), by bd_claiming the device.
2541  */
2542 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2543 {
2544         int err = 0;
2545         struct block_device *bdev;
2546
2547         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2548                                  shared ? (struct md_rdev *)lock_rdev : rdev);
2549         if (IS_ERR(bdev)) {
2550                 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2551                         MAJOR(dev), MINOR(dev));
2552                 return PTR_ERR(bdev);
2553         }
2554         rdev->bdev = bdev;
2555         return err;
2556 }
2557
2558 static void unlock_rdev(struct md_rdev *rdev)
2559 {
2560         struct block_device *bdev = rdev->bdev;
2561         rdev->bdev = NULL;
2562         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2563 }
2564
2565 void md_autodetect_dev(dev_t dev);
2566
2567 static void export_rdev(struct md_rdev *rdev)
2568 {
2569         char b[BDEVNAME_SIZE];
2570
2571         pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2572         md_rdev_clear(rdev);
2573 #ifndef MODULE
2574         if (test_bit(AutoDetected, &rdev->flags))
2575                 md_autodetect_dev(rdev->bdev->bd_dev);
2576 #endif
2577         unlock_rdev(rdev);
2578         kobject_put(&rdev->kobj);
2579 }
2580
2581 void md_kick_rdev_from_array(struct md_rdev *rdev)
2582 {
2583         unbind_rdev_from_array(rdev);
2584         export_rdev(rdev);
2585 }
2586 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2587
2588 static void export_array(struct mddev *mddev)
2589 {
2590         struct md_rdev *rdev;
2591
2592         while (!list_empty(&mddev->disks)) {
2593                 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2594                                         same_set);
2595                 md_kick_rdev_from_array(rdev);
2596         }
2597         mddev->raid_disks = 0;
2598         mddev->major_version = 0;
2599 }
2600
2601 static bool set_in_sync(struct mddev *mddev)
2602 {
2603         lockdep_assert_held(&mddev->lock);
2604         if (!mddev->in_sync) {
2605                 mddev->sync_checkers++;
2606                 spin_unlock(&mddev->lock);
2607                 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2608                 spin_lock(&mddev->lock);
2609                 if (!mddev->in_sync &&
2610                     percpu_ref_is_zero(&mddev->writes_pending)) {
2611                         mddev->in_sync = 1;
2612                         /*
2613                          * Ensure ->in_sync is visible before we clear
2614                          * ->sync_checkers.
2615                          */
2616                         smp_mb();
2617                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2618                         sysfs_notify_dirent_safe(mddev->sysfs_state);
2619                 }
2620                 if (--mddev->sync_checkers == 0)
2621                         percpu_ref_switch_to_percpu(&mddev->writes_pending);
2622         }
2623         if (mddev->safemode == 1)
2624                 mddev->safemode = 0;
2625         return mddev->in_sync;
2626 }
2627
2628 static void sync_sbs(struct mddev *mddev, int nospares)
2629 {
2630         /* Update each superblock (in-memory image), but
2631          * if we are allowed to, skip spares which already
2632          * have the right event counter, or have one earlier
2633          * (which would mean they aren't being marked as dirty
2634          * with the rest of the array)
2635          */
2636         struct md_rdev *rdev;
2637         rdev_for_each(rdev, mddev) {
2638                 if (rdev->sb_events == mddev->events ||
2639                     (nospares &&
2640                      rdev->raid_disk < 0 &&
2641                      rdev->sb_events+1 == mddev->events)) {
2642                         /* Don't update this superblock */
2643                         rdev->sb_loaded = 2;
2644                 } else {
2645                         sync_super(mddev, rdev);
2646                         rdev->sb_loaded = 1;
2647                 }
2648         }
2649 }
2650
2651 static bool does_sb_need_changing(struct mddev *mddev)
2652 {
2653         struct md_rdev *rdev;
2654         struct mdp_superblock_1 *sb;
2655         int role;
2656
2657         /* Find a good rdev */
2658         rdev_for_each(rdev, mddev)
2659                 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2660                         break;
2661
2662         /* No good device found. */
2663         if (!rdev)
2664                 return false;
2665
2666         sb = page_address(rdev->sb_page);
2667         /* Check if a device has become faulty or a spare become active */
2668         rdev_for_each(rdev, mddev) {
2669                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2670                 /* Device activated? */
2671                 if (role == 0xffff && rdev->raid_disk >=0 &&
2672                     !test_bit(Faulty, &rdev->flags))
2673                         return true;
2674                 /* Device turned faulty? */
2675                 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2676                         return true;
2677         }
2678
2679         /* Check if any mddev parameters have changed */
2680         if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2681             (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2682             (mddev->layout != le32_to_cpu(sb->layout)) ||
2683             (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2684             (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2685                 return true;
2686
2687         return false;
2688 }
2689
2690 void md_update_sb(struct mddev *mddev, int force_change)
2691 {
2692         struct md_rdev *rdev;
2693         int sync_req;
2694         int nospares = 0;
2695         int any_badblocks_changed = 0;
2696         int ret = -1;
2697
2698         if (mddev->ro) {
2699                 if (force_change)
2700                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2701                 return;
2702         }
2703
2704 repeat:
2705         if (mddev_is_clustered(mddev)) {
2706                 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2707                         force_change = 1;
2708                 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2709                         nospares = 1;
2710                 ret = md_cluster_ops->metadata_update_start(mddev);
2711                 /* Has someone else has updated the sb */
2712                 if (!does_sb_need_changing(mddev)) {
2713                         if (ret == 0)
2714                                 md_cluster_ops->metadata_update_cancel(mddev);
2715                         bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2716                                                          BIT(MD_SB_CHANGE_DEVS) |
2717                                                          BIT(MD_SB_CHANGE_CLEAN));
2718                         return;
2719                 }
2720         }
2721
2722         /*
2723          * First make sure individual recovery_offsets are correct
2724          * curr_resync_completed can only be used during recovery.
2725          * During reshape/resync it might use array-addresses rather
2726          * that device addresses.
2727          */
2728         rdev_for_each(rdev, mddev) {
2729                 if (rdev->raid_disk >= 0 &&
2730                     mddev->delta_disks >= 0 &&
2731                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2732                     test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2733                     !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2734                     !test_bit(Journal, &rdev->flags) &&
2735                     !test_bit(In_sync, &rdev->flags) &&
2736                     mddev->curr_resync_completed > rdev->recovery_offset)
2737                                 rdev->recovery_offset = mddev->curr_resync_completed;
2738
2739         }
2740         if (!mddev->persistent) {
2741                 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2742                 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2743                 if (!mddev->external) {
2744                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2745                         rdev_for_each(rdev, mddev) {
2746                                 if (rdev->badblocks.changed) {
2747                                         rdev->badblocks.changed = 0;
2748                                         ack_all_badblocks(&rdev->badblocks);
2749                                         md_error(mddev, rdev);
2750                                 }
2751                                 clear_bit(Blocked, &rdev->flags);
2752                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2753                                 wake_up(&rdev->blocked_wait);
2754                         }
2755                 }
2756                 wake_up(&mddev->sb_wait);
2757                 return;
2758         }
2759
2760         spin_lock(&mddev->lock);
2761
2762         mddev->utime = ktime_get_real_seconds();
2763
2764         if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2765                 force_change = 1;
2766         if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2767                 /* just a clean<-> dirty transition, possibly leave spares alone,
2768                  * though if events isn't the right even/odd, we will have to do
2769                  * spares after all
2770                  */
2771                 nospares = 1;
2772         if (force_change)
2773                 nospares = 0;
2774         if (mddev->degraded)
2775                 /* If the array is degraded, then skipping spares is both
2776                  * dangerous and fairly pointless.
2777                  * Dangerous because a device that was removed from the array
2778                  * might have a event_count that still looks up-to-date,
2779                  * so it can be re-added without a resync.
2780                  * Pointless because if there are any spares to skip,
2781                  * then a recovery will happen and soon that array won't
2782                  * be degraded any more and the spare can go back to sleep then.
2783                  */
2784                 nospares = 0;
2785
2786         sync_req = mddev->in_sync;
2787
2788         /* If this is just a dirty<->clean transition, and the array is clean
2789          * and 'events' is odd, we can roll back to the previous clean state */
2790         if (nospares
2791             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2792             && mddev->can_decrease_events
2793             && mddev->events != 1) {
2794                 mddev->events--;
2795                 mddev->can_decrease_events = 0;
2796         } else {
2797                 /* otherwise we have to go forward and ... */
2798                 mddev->events ++;
2799                 mddev->can_decrease_events = nospares;
2800         }
2801
2802         /*
2803          * This 64-bit counter should never wrap.
2804          * Either we are in around ~1 trillion A.C., assuming
2805          * 1 reboot per second, or we have a bug...
2806          */
2807         WARN_ON(mddev->events == 0);
2808
2809         rdev_for_each(rdev, mddev) {
2810                 if (rdev->badblocks.changed)
2811                         any_badblocks_changed++;
2812                 if (test_bit(Faulty, &rdev->flags))
2813                         set_bit(FaultRecorded, &rdev->flags);
2814         }
2815
2816         sync_sbs(mddev, nospares);
2817         spin_unlock(&mddev->lock);
2818
2819         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2820                  mdname(mddev), mddev->in_sync);
2821
2822         if (mddev->queue)
2823                 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2824 rewrite:
2825         md_bitmap_update_sb(mddev->bitmap);
2826         rdev_for_each(rdev, mddev) {
2827                 char b[BDEVNAME_SIZE];
2828
2829                 if (rdev->sb_loaded != 1)
2830                         continue; /* no noise on spare devices */
2831
2832                 if (!test_bit(Faulty, &rdev->flags)) {
2833                         md_super_write(mddev,rdev,
2834                                        rdev->sb_start, rdev->sb_size,
2835                                        rdev->sb_page);
2836                         pr_debug("md: (write) %s's sb offset: %llu\n",
2837                                  bdevname(rdev->bdev, b),
2838                                  (unsigned long long)rdev->sb_start);
2839                         rdev->sb_events = mddev->events;
2840                         if (rdev->badblocks.size) {
2841                                 md_super_write(mddev, rdev,
2842                                                rdev->badblocks.sector,
2843                                                rdev->badblocks.size << 9,
2844                                                rdev->bb_page);
2845                                 rdev->badblocks.size = 0;
2846                         }
2847
2848                 } else
2849                         pr_debug("md: %s (skipping faulty)\n",
2850                                  bdevname(rdev->bdev, b));
2851
2852                 if (mddev->level == LEVEL_MULTIPATH)
2853                         /* only need to write one superblock... */
2854                         break;
2855         }
2856         if (md_super_wait(mddev) < 0)
2857                 goto rewrite;
2858         /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2859
2860         if (mddev_is_clustered(mddev) && ret == 0)
2861                 md_cluster_ops->metadata_update_finish(mddev);
2862
2863         if (mddev->in_sync != sync_req ||
2864             !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2865                                BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2866                 /* have to write it out again */
2867                 goto repeat;
2868         wake_up(&mddev->sb_wait);
2869         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2870                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2871
2872         rdev_for_each(rdev, mddev) {
2873                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2874                         clear_bit(Blocked, &rdev->flags);
2875
2876                 if (any_badblocks_changed)
2877                         ack_all_badblocks(&rdev->badblocks);
2878                 clear_bit(BlockedBadBlocks, &rdev->flags);
2879                 wake_up(&rdev->blocked_wait);
2880         }
2881 }
2882 EXPORT_SYMBOL(md_update_sb);
2883
2884 static int add_bound_rdev(struct md_rdev *rdev)
2885 {
2886         struct mddev *mddev = rdev->mddev;
2887         int err = 0;
2888         bool add_journal = test_bit(Journal, &rdev->flags);
2889
2890         if (!mddev->pers->hot_remove_disk || add_journal) {
2891                 /* If there is hot_add_disk but no hot_remove_disk
2892                  * then added disks for geometry changes,
2893                  * and should be added immediately.
2894                  */
2895                 super_types[mddev->major_version].
2896                         validate_super(mddev, rdev);
2897                 if (add_journal)
2898                         mddev_suspend(mddev);
2899                 err = mddev->pers->hot_add_disk(mddev, rdev);
2900                 if (add_journal)
2901                         mddev_resume(mddev);
2902                 if (err) {
2903                         md_kick_rdev_from_array(rdev);
2904                         return err;
2905                 }
2906         }
2907         sysfs_notify_dirent_safe(rdev->sysfs_state);
2908
2909         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2910         if (mddev->degraded)
2911                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2912         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2913         md_new_event(mddev);
2914         md_wakeup_thread(mddev->thread);
2915         return 0;
2916 }
2917
2918 /* words written to sysfs files may, or may not, be \n terminated.
2919  * We want to accept with case. For this we use cmd_match.
2920  */
2921 static int cmd_match(const char *cmd, const char *str)
2922 {
2923         /* See if cmd, written into a sysfs file, matches
2924          * str.  They must either be the same, or cmd can
2925          * have a trailing newline
2926          */
2927         while (*cmd && *str && *cmd == *str) {
2928                 cmd++;
2929                 str++;
2930         }
2931         if (*cmd == '\n')
2932                 cmd++;
2933         if (*str || *cmd)
2934                 return 0;
2935         return 1;
2936 }
2937
2938 struct rdev_sysfs_entry {
2939         struct attribute attr;
2940         ssize_t (*show)(struct md_rdev *, char *);
2941         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2942 };
2943
2944 static ssize_t
2945 state_show(struct md_rdev *rdev, char *page)
2946 {
2947         char *sep = ",";
2948         size_t len = 0;
2949         unsigned long flags = READ_ONCE(rdev->flags);
2950
2951         if (test_bit(Faulty, &flags) ||
2952             (!test_bit(ExternalBbl, &flags) &&
2953             rdev->badblocks.unacked_exist))
2954                 len += sprintf(page+len, "faulty%s", sep);
2955         if (test_bit(In_sync, &flags))
2956                 len += sprintf(page+len, "in_sync%s", sep);
2957         if (test_bit(Journal, &flags))
2958                 len += sprintf(page+len, "journal%s", sep);
2959         if (test_bit(WriteMostly, &flags))
2960                 len += sprintf(page+len, "write_mostly%s", sep);
2961         if (test_bit(Blocked, &flags) ||
2962             (rdev->badblocks.unacked_exist
2963              && !test_bit(Faulty, &flags)))
2964                 len += sprintf(page+len, "blocked%s", sep);
2965         if (!test_bit(Faulty, &flags) &&
2966             !test_bit(Journal, &flags) &&
2967             !test_bit(In_sync, &flags))
2968                 len += sprintf(page+len, "spare%s", sep);
2969         if (test_bit(WriteErrorSeen, &flags))
2970                 len += sprintf(page+len, "write_error%s", sep);
2971         if (test_bit(WantReplacement, &flags))
2972                 len += sprintf(page+len, "want_replacement%s", sep);
2973         if (test_bit(Replacement, &flags))
2974                 len += sprintf(page+len, "replacement%s", sep);
2975         if (test_bit(ExternalBbl, &flags))
2976                 len += sprintf(page+len, "external_bbl%s", sep);
2977         if (test_bit(FailFast, &flags))
2978                 len += sprintf(page+len, "failfast%s", sep);
2979
2980         if (len)
2981                 len -= strlen(sep);
2982
2983         return len+sprintf(page+len, "\n");
2984 }
2985
2986 static ssize_t
2987 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2988 {
2989         /* can write
2990          *  faulty  - simulates an error
2991          *  remove  - disconnects the device
2992          *  writemostly - sets write_mostly
2993          *  -writemostly - clears write_mostly
2994          *  blocked - sets the Blocked flags
2995          *  -blocked - clears the Blocked and possibly simulates an error
2996          *  insync - sets Insync providing device isn't active
2997          *  -insync - clear Insync for a device with a slot assigned,
2998          *            so that it gets rebuilt based on bitmap
2999          *  write_error - sets WriteErrorSeen
3000          *  -write_error - clears WriteErrorSeen
3001          *  {,-}failfast - set/clear FailFast
3002          */
3003         int err = -EINVAL;
3004         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
3005                 md_error(rdev->mddev, rdev);
3006                 if (test_bit(Faulty, &rdev->flags))
3007                         err = 0;
3008                 else
3009                         err = -EBUSY;
3010         } else if (cmd_match(buf, "remove")) {
3011                 if (rdev->mddev->pers) {
3012                         clear_bit(Blocked, &rdev->flags);
3013                         remove_and_add_spares(rdev->mddev, rdev);
3014                 }
3015                 if (rdev->raid_disk >= 0)
3016                         err = -EBUSY;
3017                 else {
3018                         struct mddev *mddev = rdev->mddev;
3019                         err = 0;
3020                         if (mddev_is_clustered(mddev))
3021                                 err = md_cluster_ops->remove_disk(mddev, rdev);
3022
3023                         if (err == 0) {
3024                                 md_kick_rdev_from_array(rdev);
3025                                 if (mddev->pers) {
3026                                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3027                                         md_wakeup_thread(mddev->thread);
3028                                 }
3029                                 md_new_event(mddev);
3030                         }
3031                 }
3032         } else if (cmd_match(buf, "writemostly")) {
3033                 set_bit(WriteMostly, &rdev->flags);
3034                 mddev_create_serial_pool(rdev->mddev, rdev, false);
3035                 err = 0;
3036         } else if (cmd_match(buf, "-writemostly")) {
3037                 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3038                 clear_bit(WriteMostly, &rdev->flags);
3039                 err = 0;
3040         } else if (cmd_match(buf, "blocked")) {
3041                 set_bit(Blocked, &rdev->flags);
3042                 err = 0;
3043         } else if (cmd_match(buf, "-blocked")) {
3044                 if (!test_bit(Faulty, &rdev->flags) &&
3045                     !test_bit(ExternalBbl, &rdev->flags) &&
3046                     rdev->badblocks.unacked_exist) {
3047                         /* metadata handler doesn't understand badblocks,
3048                          * so we need to fail the device
3049                          */
3050                         md_error(rdev->mddev, rdev);
3051                 }
3052                 clear_bit(Blocked, &rdev->flags);
3053                 clear_bit(BlockedBadBlocks, &rdev->flags);
3054                 wake_up(&rdev->blocked_wait);
3055                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3056                 md_wakeup_thread(rdev->mddev->thread);
3057
3058                 err = 0;
3059         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3060                 set_bit(In_sync, &rdev->flags);
3061                 err = 0;
3062         } else if (cmd_match(buf, "failfast")) {
3063                 set_bit(FailFast, &rdev->flags);
3064                 err = 0;
3065         } else if (cmd_match(buf, "-failfast")) {
3066                 clear_bit(FailFast, &rdev->flags);
3067                 err = 0;
3068         } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3069                    !test_bit(Journal, &rdev->flags)) {
3070                 if (rdev->mddev->pers == NULL) {
3071                         clear_bit(In_sync, &rdev->flags);
3072                         rdev->saved_raid_disk = rdev->raid_disk;
3073                         rdev->raid_disk = -1;
3074                         err = 0;
3075                 }
3076         } else if (cmd_match(buf, "write_error")) {
3077                 set_bit(WriteErrorSeen, &rdev->flags);
3078                 err = 0;
3079         } else if (cmd_match(buf, "-write_error")) {
3080                 clear_bit(WriteErrorSeen, &rdev->flags);
3081                 err = 0;
3082         } else if (cmd_match(buf, "want_replacement")) {
3083                 /* Any non-spare device that is not a replacement can
3084                  * become want_replacement at any time, but we then need to
3085                  * check if recovery is needed.
3086                  */
3087                 if (rdev->raid_disk >= 0 &&
3088                     !test_bit(Journal, &rdev->flags) &&
3089                     !test_bit(Replacement, &rdev->flags))
3090                         set_bit(WantReplacement, &rdev->flags);
3091                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3092                 md_wakeup_thread(rdev->mddev->thread);
3093                 err = 0;
3094         } else if (cmd_match(buf, "-want_replacement")) {
3095                 /* Clearing 'want_replacement' is always allowed.
3096                  * Once replacements starts it is too late though.
3097                  */
3098                 err = 0;
3099                 clear_bit(WantReplacement, &rdev->flags);
3100         } else if (cmd_match(buf, "replacement")) {
3101                 /* Can only set a device as a replacement when array has not
3102                  * yet been started.  Once running, replacement is automatic
3103                  * from spares, or by assigning 'slot'.
3104                  */
3105                 if (rdev->mddev->pers)
3106                         err = -EBUSY;
3107                 else {
3108                         set_bit(Replacement, &rdev->flags);
3109                         err = 0;
3110                 }
3111         } else if (cmd_match(buf, "-replacement")) {
3112                 /* Similarly, can only clear Replacement before start */
3113                 if (rdev->mddev->pers)
3114                         err = -EBUSY;
3115                 else {
3116                         clear_bit(Replacement, &rdev->flags);
3117                         err = 0;
3118                 }
3119         } else if (cmd_match(buf, "re-add")) {
3120                 if (!rdev->mddev->pers)
3121                         err = -EINVAL;
3122                 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3123                                 rdev->saved_raid_disk >= 0) {
3124                         /* clear_bit is performed _after_ all the devices
3125                          * have their local Faulty bit cleared. If any writes
3126                          * happen in the meantime in the local node, they
3127                          * will land in the local bitmap, which will be synced
3128                          * by this node eventually
3129                          */
3130                         if (!mddev_is_clustered(rdev->mddev) ||
3131                             (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3132                                 clear_bit(Faulty, &rdev->flags);
3133                                 err = add_bound_rdev(rdev);
3134                         }
3135                 } else
3136                         err = -EBUSY;
3137         } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3138                 set_bit(ExternalBbl, &rdev->flags);
3139                 rdev->badblocks.shift = 0;
3140                 err = 0;
3141         } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3142                 clear_bit(ExternalBbl, &rdev->flags);
3143                 err = 0;
3144         }
3145         if (!err)
3146                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3147         return err ? err : len;
3148 }
3149 static struct rdev_sysfs_entry rdev_state =
3150 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3151
3152 static ssize_t
3153 errors_show(struct md_rdev *rdev, char *page)
3154 {
3155         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3156 }
3157
3158 static ssize_t
3159 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3160 {
3161         unsigned int n;
3162         int rv;
3163
3164         rv = kstrtouint(buf, 10, &n);
3165         if (rv < 0)
3166                 return rv;
3167         atomic_set(&rdev->corrected_errors, n);
3168         return len;
3169 }
3170 static struct rdev_sysfs_entry rdev_errors =
3171 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3172
3173 static ssize_t
3174 slot_show(struct md_rdev *rdev, char *page)
3175 {
3176         if (test_bit(Journal, &rdev->flags))
3177                 return sprintf(page, "journal\n");
3178         else if (rdev->raid_disk < 0)
3179                 return sprintf(page, "none\n");
3180         else
3181                 return sprintf(page, "%d\n", rdev->raid_disk);
3182 }
3183
3184 static ssize_t
3185 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3186 {
3187         int slot;
3188         int err;
3189
3190         if (test_bit(Journal, &rdev->flags))
3191                 return -EBUSY;
3192         if (strncmp(buf, "none", 4)==0)
3193                 slot = -1;
3194         else {
3195                 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3196                 if (err < 0)
3197                         return err;
3198         }
3199         if (rdev->mddev->pers && slot == -1) {
3200                 /* Setting 'slot' on an active array requires also
3201                  * updating the 'rd%d' link, and communicating
3202                  * with the personality with ->hot_*_disk.
3203                  * For now we only support removing
3204                  * failed/spare devices.  This normally happens automatically,
3205                  * but not when the metadata is externally managed.
3206                  */
3207                 if (rdev->raid_disk == -1)
3208                         return -EEXIST;
3209                 /* personality does all needed checks */
3210                 if (rdev->mddev->pers->hot_remove_disk == NULL)
3211                         return -EINVAL;
3212                 clear_bit(Blocked, &rdev->flags);
3213                 remove_and_add_spares(rdev->mddev, rdev);
3214                 if (rdev->raid_disk >= 0)
3215                         return -EBUSY;
3216                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3217                 md_wakeup_thread(rdev->mddev->thread);
3218         } else if (rdev->mddev->pers) {
3219                 /* Activating a spare .. or possibly reactivating
3220                  * if we ever get bitmaps working here.
3221                  */
3222                 int err;
3223
3224                 if (rdev->raid_disk != -1)
3225                         return -EBUSY;
3226
3227                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3228                         return -EBUSY;
3229
3230                 if (rdev->mddev->pers->hot_add_disk == NULL)
3231                         return -EINVAL;
3232
3233                 if (slot >= rdev->mddev->raid_disks &&
3234                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3235                         return -ENOSPC;
3236
3237                 rdev->raid_disk = slot;
3238                 if (test_bit(In_sync, &rdev->flags))
3239                         rdev->saved_raid_disk = slot;
3240                 else
3241                         rdev->saved_raid_disk = -1;
3242                 clear_bit(In_sync, &rdev->flags);
3243                 clear_bit(Bitmap_sync, &rdev->flags);
3244                 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3245                 if (err) {
3246                         rdev->raid_disk = -1;
3247                         return err;
3248                 } else
3249                         sysfs_notify_dirent_safe(rdev->sysfs_state);
3250                 /* failure here is OK */;
3251                 sysfs_link_rdev(rdev->mddev, rdev);
3252                 /* don't wakeup anyone, leave that to userspace. */
3253         } else {
3254                 if (slot >= rdev->mddev->raid_disks &&
3255                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3256                         return -ENOSPC;
3257                 rdev->raid_disk = slot;
3258                 /* assume it is working */
3259                 clear_bit(Faulty, &rdev->flags);
3260                 clear_bit(WriteMostly, &rdev->flags);
3261                 set_bit(In_sync, &rdev->flags);
3262                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3263         }
3264         return len;
3265 }
3266
3267 static struct rdev_sysfs_entry rdev_slot =
3268 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3269
3270 static ssize_t
3271 offset_show(struct md_rdev *rdev, char *page)
3272 {
3273         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3274 }
3275
3276 static ssize_t
3277 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3278 {
3279         unsigned long long offset;
3280         if (kstrtoull(buf, 10, &offset) < 0)
3281                 return -EINVAL;
3282         if (rdev->mddev->pers && rdev->raid_disk >= 0)
3283                 return -EBUSY;
3284         if (rdev->sectors && rdev->mddev->external)
3285                 /* Must set offset before size, so overlap checks
3286                  * can be sane */
3287                 return -EBUSY;
3288         rdev->data_offset = offset;
3289         rdev->new_data_offset = offset;
3290         return len;
3291 }
3292
3293 static struct rdev_sysfs_entry rdev_offset =
3294 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3295
3296 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3297 {
3298         return sprintf(page, "%llu\n",
3299                        (unsigned long long)rdev->new_data_offset);
3300 }
3301
3302 static ssize_t new_offset_store(struct md_rdev *rdev,
3303                                 const char *buf, size_t len)
3304 {
3305         unsigned long long new_offset;
3306         struct mddev *mddev = rdev->mddev;
3307
3308         if (kstrtoull(buf, 10, &new_offset) < 0)
3309                 return -EINVAL;
3310
3311         if (mddev->sync_thread ||
3312             test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3313                 return -EBUSY;
3314         if (new_offset == rdev->data_offset)
3315                 /* reset is always permitted */
3316                 ;
3317         else if (new_offset > rdev->data_offset) {
3318                 /* must not push array size beyond rdev_sectors */
3319                 if (new_offset - rdev->data_offset
3320                     + mddev->dev_sectors > rdev->sectors)
3321                                 return -E2BIG;
3322         }
3323         /* Metadata worries about other space details. */
3324
3325         /* decreasing the offset is inconsistent with a backwards
3326          * reshape.
3327          */
3328         if (new_offset < rdev->data_offset &&
3329             mddev->reshape_backwards)
3330                 return -EINVAL;
3331         /* Increasing offset is inconsistent with forwards
3332          * reshape.  reshape_direction should be set to
3333          * 'backwards' first.
3334          */
3335         if (new_offset > rdev->data_offset &&
3336             !mddev->reshape_backwards)
3337                 return -EINVAL;
3338
3339         if (mddev->pers && mddev->persistent &&
3340             !super_types[mddev->major_version]
3341             .allow_new_offset(rdev, new_offset))
3342                 return -E2BIG;
3343         rdev->new_data_offset = new_offset;
3344         if (new_offset > rdev->data_offset)
3345                 mddev->reshape_backwards = 1;
3346         else if (new_offset < rdev->data_offset)
3347                 mddev->reshape_backwards = 0;
3348
3349         return len;
3350 }
3351 static struct rdev_sysfs_entry rdev_new_offset =
3352 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3353
3354 static ssize_t
3355 rdev_size_show(struct md_rdev *rdev, char *page)
3356 {
3357         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3358 }
3359
3360 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3361 {
3362         /* check if two start/length pairs overlap */
3363         if (s1+l1 <= s2)
3364                 return 0;
3365         if (s2+l2 <= s1)
3366                 return 0;
3367         return 1;
3368 }
3369
3370 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3371 {
3372         unsigned long long blocks;
3373         sector_t new;
3374
3375         if (kstrtoull(buf, 10, &blocks) < 0)
3376                 return -EINVAL;
3377
3378         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3379                 return -EINVAL; /* sector conversion overflow */
3380
3381         new = blocks * 2;
3382         if (new != blocks * 2)
3383                 return -EINVAL; /* unsigned long long to sector_t overflow */
3384
3385         *sectors = new;
3386         return 0;
3387 }
3388
3389 static ssize_t
3390 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3391 {
3392         struct mddev *my_mddev = rdev->mddev;
3393         sector_t oldsectors = rdev->sectors;
3394         sector_t sectors;
3395
3396         if (test_bit(Journal, &rdev->flags))
3397                 return -EBUSY;
3398         if (strict_blocks_to_sectors(buf, &sectors) < 0)
3399                 return -EINVAL;
3400         if (rdev->data_offset != rdev->new_data_offset)
3401                 return -EINVAL; /* too confusing */
3402         if (my_mddev->pers && rdev->raid_disk >= 0) {
3403                 if (my_mddev->persistent) {
3404                         sectors = super_types[my_mddev->major_version].
3405                                 rdev_size_change(rdev, sectors);
3406                         if (!sectors)
3407                                 return -EBUSY;
3408                 } else if (!sectors)
3409                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3410                                 rdev->data_offset;
3411                 if (!my_mddev->pers->resize)
3412                         /* Cannot change size for RAID0 or Linear etc */
3413                         return -EINVAL;
3414         }
3415         if (sectors < my_mddev->dev_sectors)
3416                 return -EINVAL; /* component must fit device */
3417
3418         rdev->sectors = sectors;
3419         if (sectors > oldsectors && my_mddev->external) {
3420                 /* Need to check that all other rdevs with the same
3421                  * ->bdev do not overlap.  'rcu' is sufficient to walk
3422                  * the rdev lists safely.
3423                  * This check does not provide a hard guarantee, it
3424                  * just helps avoid dangerous mistakes.
3425                  */
3426                 struct mddev *mddev;
3427                 int overlap = 0;
3428                 struct list_head *tmp;
3429
3430                 rcu_read_lock();
3431                 for_each_mddev(mddev, tmp) {
3432                         struct md_rdev *rdev2;
3433
3434                         rdev_for_each(rdev2, mddev)
3435                                 if (rdev->bdev == rdev2->bdev &&
3436                                     rdev != rdev2 &&
3437                                     overlaps(rdev->data_offset, rdev->sectors,
3438                                              rdev2->data_offset,
3439                                              rdev2->sectors)) {
3440                                         overlap = 1;
3441                                         break;
3442                                 }
3443                         if (overlap) {
3444                                 mddev_put(mddev);
3445                                 break;
3446                         }
3447                 }
3448                 rcu_read_unlock();
3449                 if (overlap) {
3450                         /* Someone else could have slipped in a size
3451                          * change here, but doing so is just silly.
3452                          * We put oldsectors back because we *know* it is
3453                          * safe, and trust userspace not to race with
3454                          * itself
3455                          */
3456                         rdev->sectors = oldsectors;
3457                         return -EBUSY;
3458                 }
3459         }
3460         return len;
3461 }
3462
3463 static struct rdev_sysfs_entry rdev_size =
3464 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3465
3466 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3467 {
3468         unsigned long long recovery_start = rdev->recovery_offset;
3469
3470         if (test_bit(In_sync, &rdev->flags) ||
3471             recovery_start == MaxSector)
3472                 return sprintf(page, "none\n");
3473
3474         return sprintf(page, "%llu\n", recovery_start);
3475 }
3476
3477 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3478 {
3479         unsigned long long recovery_start;
3480
3481         if (cmd_match(buf, "none"))
3482                 recovery_start = MaxSector;
3483         else if (kstrtoull(buf, 10, &recovery_start))
3484                 return -EINVAL;
3485
3486         if (rdev->mddev->pers &&
3487             rdev->raid_disk >= 0)
3488                 return -EBUSY;
3489
3490         rdev->recovery_offset = recovery_start;
3491         if (recovery_start == MaxSector)
3492                 set_bit(In_sync, &rdev->flags);
3493         else
3494                 clear_bit(In_sync, &rdev->flags);
3495         return len;
3496 }
3497
3498 static struct rdev_sysfs_entry rdev_recovery_start =
3499 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3500
3501 /* sysfs access to bad-blocks list.
3502  * We present two files.
3503  * 'bad-blocks' lists sector numbers and lengths of ranges that
3504  *    are recorded as bad.  The list is truncated to fit within
3505  *    the one-page limit of sysfs.
3506  *    Writing "sector length" to this file adds an acknowledged
3507  *    bad block list.
3508  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3509  *    been acknowledged.  Writing to this file adds bad blocks
3510  *    without acknowledging them.  This is largely for testing.
3511  */
3512 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3513 {
3514         return badblocks_show(&rdev->badblocks, page, 0);
3515 }
3516 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3517 {
3518         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3519         /* Maybe that ack was all we needed */
3520         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3521                 wake_up(&rdev->blocked_wait);
3522         return rv;
3523 }
3524 static struct rdev_sysfs_entry rdev_bad_blocks =
3525 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3526
3527 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3528 {
3529         return badblocks_show(&rdev->badblocks, page, 1);
3530 }
3531 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3532 {
3533         return badblocks_store(&rdev->badblocks, page, len, 1);
3534 }
3535 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3536 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3537
3538 static ssize_t
3539 ppl_sector_show(struct md_rdev *rdev, char *page)
3540 {
3541         return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3542 }
3543
3544 static ssize_t
3545 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3546 {
3547         unsigned long long sector;
3548
3549         if (kstrtoull(buf, 10, &sector) < 0)
3550                 return -EINVAL;
3551         if (sector != (sector_t)sector)
3552                 return -EINVAL;
3553
3554         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3555             rdev->raid_disk >= 0)
3556                 return -EBUSY;
3557
3558         if (rdev->mddev->persistent) {
3559                 if (rdev->mddev->major_version == 0)
3560                         return -EINVAL;
3561                 if ((sector > rdev->sb_start &&
3562                      sector - rdev->sb_start > S16_MAX) ||
3563                     (sector < rdev->sb_start &&
3564                      rdev->sb_start - sector > -S16_MIN))
3565                         return -EINVAL;
3566                 rdev->ppl.offset = sector - rdev->sb_start;
3567         } else if (!rdev->mddev->external) {
3568                 return -EBUSY;
3569         }
3570         rdev->ppl.sector = sector;
3571         return len;
3572 }
3573
3574 static struct rdev_sysfs_entry rdev_ppl_sector =
3575 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3576
3577 static ssize_t
3578 ppl_size_show(struct md_rdev *rdev, char *page)
3579 {
3580         return sprintf(page, "%u\n", rdev->ppl.size);
3581 }
3582
3583 static ssize_t
3584 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3585 {
3586         unsigned int size;
3587
3588         if (kstrtouint(buf, 10, &size) < 0)
3589                 return -EINVAL;
3590
3591         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3592             rdev->raid_disk >= 0)
3593                 return -EBUSY;
3594
3595         if (rdev->mddev->persistent) {
3596                 if (rdev->mddev->major_version == 0)
3597                         return -EINVAL;
3598                 if (size > U16_MAX)
3599                         return -EINVAL;
3600         } else if (!rdev->mddev->external) {
3601                 return -EBUSY;
3602         }
3603         rdev->ppl.size = size;
3604         return len;
3605 }
3606
3607 static struct rdev_sysfs_entry rdev_ppl_size =
3608 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3609
3610 static struct attribute *rdev_default_attrs[] = {
3611         &rdev_state.attr,
3612         &rdev_errors.attr,
3613         &rdev_slot.attr,
3614         &rdev_offset.attr,
3615         &rdev_new_offset.attr,
3616         &rdev_size.attr,
3617         &rdev_recovery_start.attr,
3618         &rdev_bad_blocks.attr,
3619         &rdev_unack_bad_blocks.attr,
3620         &rdev_ppl_sector.attr,
3621         &rdev_ppl_size.attr,
3622         NULL,
3623 };
3624 static ssize_t
3625 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3626 {
3627         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3628         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3629
3630         if (!entry->show)
3631                 return -EIO;
3632         if (!rdev->mddev)
3633                 return -ENODEV;
3634         return entry->show(rdev, page);
3635 }
3636
3637 static ssize_t
3638 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3639               const char *page, size_t length)
3640 {
3641         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3642         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3643         ssize_t rv;
3644         struct mddev *mddev = rdev->mddev;
3645
3646         if (!entry->store)
3647                 return -EIO;
3648         if (!capable(CAP_SYS_ADMIN))
3649                 return -EACCES;
3650         rv = mddev ? mddev_lock(mddev) : -ENODEV;
3651         if (!rv) {
3652                 if (rdev->mddev == NULL)
3653                         rv = -ENODEV;
3654                 else
3655                         rv = entry->store(rdev, page, length);
3656                 mddev_unlock(mddev);
3657         }
3658         return rv;
3659 }
3660
3661 static void rdev_free(struct kobject *ko)
3662 {
3663         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3664         kfree(rdev);
3665 }
3666 static const struct sysfs_ops rdev_sysfs_ops = {
3667         .show           = rdev_attr_show,
3668         .store          = rdev_attr_store,
3669 };
3670 static struct kobj_type rdev_ktype = {
3671         .release        = rdev_free,
3672         .sysfs_ops      = &rdev_sysfs_ops,
3673         .default_attrs  = rdev_default_attrs,
3674 };
3675
3676 int md_rdev_init(struct md_rdev *rdev)
3677 {
3678         rdev->desc_nr = -1;
3679         rdev->saved_raid_disk = -1;
3680         rdev->raid_disk = -1;
3681         rdev->flags = 0;
3682         rdev->data_offset = 0;
3683         rdev->new_data_offset = 0;
3684         rdev->sb_events = 0;
3685         rdev->last_read_error = 0;
3686         rdev->sb_loaded = 0;
3687         rdev->bb_page = NULL;
3688         atomic_set(&rdev->nr_pending, 0);
3689         atomic_set(&rdev->read_errors, 0);
3690         atomic_set(&rdev->corrected_errors, 0);
3691
3692         INIT_LIST_HEAD(&rdev->same_set);
3693         init_waitqueue_head(&rdev->blocked_wait);
3694
3695         /* Add space to store bad block list.
3696          * This reserves the space even on arrays where it cannot
3697          * be used - I wonder if that matters
3698          */
3699         return badblocks_init(&rdev->badblocks, 0);
3700 }
3701 EXPORT_SYMBOL_GPL(md_rdev_init);
3702 /*
3703  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3704  *
3705  * mark the device faulty if:
3706  *
3707  *   - the device is nonexistent (zero size)
3708  *   - the device has no valid superblock
3709  *
3710  * a faulty rdev _never_ has rdev->sb set.
3711  */
3712 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3713 {
3714         char b[BDEVNAME_SIZE];
3715         int err;
3716         struct md_rdev *rdev;
3717         sector_t size;
3718
3719         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3720         if (!rdev)
3721                 return ERR_PTR(-ENOMEM);
3722
3723         err = md_rdev_init(rdev);
3724         if (err)
3725                 goto abort_free;
3726         err = alloc_disk_sb(rdev);
3727         if (err)
3728                 goto abort_free;
3729
3730         err = lock_rdev(rdev, newdev, super_format == -2);
3731         if (err)
3732                 goto abort_free;
3733
3734         kobject_init(&rdev->kobj, &rdev_ktype);
3735
3736         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3737         if (!size) {
3738                 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3739                         bdevname(rdev->bdev,b));
3740                 err = -EINVAL;
3741                 goto abort_free;
3742         }
3743
3744         if (super_format >= 0) {
3745                 err = super_types[super_format].
3746                         load_super(rdev, NULL, super_minor);
3747                 if (err == -EINVAL) {
3748                         pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3749                                 bdevname(rdev->bdev,b),
3750                                 super_format, super_minor);
3751                         goto abort_free;
3752                 }
3753                 if (err < 0) {
3754                         pr_warn("md: could not read %s's sb, not importing!\n",
3755                                 bdevname(rdev->bdev,b));
3756                         goto abort_free;
3757                 }
3758         }
3759
3760         return rdev;
3761
3762 abort_free:
3763         if (rdev->bdev)
3764                 unlock_rdev(rdev);
3765         md_rdev_clear(rdev);
3766         kfree(rdev);
3767         return ERR_PTR(err);
3768 }
3769
3770 /*
3771  * Check a full RAID array for plausibility
3772  */
3773
3774 static int analyze_sbs(struct mddev *mddev)
3775 {
3776         int i;
3777         struct md_rdev *rdev, *freshest, *tmp;
3778         char b[BDEVNAME_SIZE];
3779
3780         freshest = NULL;
3781         rdev_for_each_safe(rdev, tmp, mddev)
3782                 switch (super_types[mddev->major_version].
3783                         load_super(rdev, freshest, mddev->minor_version)) {
3784                 case 1:
3785                         freshest = rdev;
3786                         break;
3787                 case 0:
3788                         break;
3789                 default:
3790                         pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3791                                 bdevname(rdev->bdev,b));
3792                         md_kick_rdev_from_array(rdev);
3793                 }
3794
3795         /* Cannot find a valid fresh disk */
3796         if (!freshest) {
3797                 pr_warn("md: cannot find a valid disk\n");
3798                 return -EINVAL;
3799         }
3800
3801         super_types[mddev->major_version].
3802                 validate_super(mddev, freshest);
3803
3804         i = 0;
3805         rdev_for_each_safe(rdev, tmp, mddev) {
3806                 if (mddev->max_disks &&
3807                     (rdev->desc_nr >= mddev->max_disks ||
3808                      i > mddev->max_disks)) {
3809                         pr_warn("md: %s: %s: only %d devices permitted\n",
3810                                 mdname(mddev), bdevname(rdev->bdev, b),
3811                                 mddev->max_disks);
3812                         md_kick_rdev_from_array(rdev);
3813                         continue;
3814                 }
3815                 if (rdev != freshest) {
3816                         if (super_types[mddev->major_version].
3817                             validate_super(mddev, rdev)) {
3818                                 pr_warn("md: kicking non-fresh %s from array!\n",
3819                                         bdevname(rdev->bdev,b));
3820                                 md_kick_rdev_from_array(rdev);
3821                                 continue;
3822                         }
3823                 }
3824                 if (mddev->level == LEVEL_MULTIPATH) {
3825                         rdev->desc_nr = i++;
3826                         rdev->raid_disk = rdev->desc_nr;
3827                         set_bit(In_sync, &rdev->flags);
3828                 } else if (rdev->raid_disk >=
3829                             (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3830                            !test_bit(Journal, &rdev->flags)) {
3831                         rdev->raid_disk = -1;
3832                         clear_bit(In_sync, &rdev->flags);
3833                 }
3834         }
3835
3836         return 0;
3837 }
3838
3839 /* Read a fixed-point number.
3840  * Numbers in sysfs attributes should be in "standard" units where
3841  * possible, so time should be in seconds.
3842  * However we internally use a a much smaller unit such as
3843  * milliseconds or jiffies.
3844  * This function takes a decimal number with a possible fractional
3845  * component, and produces an integer which is the result of
3846  * multiplying that number by 10^'scale'.
3847  * all without any floating-point arithmetic.
3848  */
3849 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3850 {
3851         unsigned long result = 0;
3852         long decimals = -1;
3853         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3854                 if (*cp == '.')
3855                         decimals = 0;
3856                 else if (decimals < scale) {
3857                         unsigned int value;
3858                         value = *cp - '0';
3859                         result = result * 10 + value;
3860                         if (decimals >= 0)
3861                                 decimals++;
3862                 }
3863                 cp++;
3864         }
3865         if (*cp == '\n')
3866                 cp++;
3867         if (*cp)
3868                 return -EINVAL;
3869         if (decimals < 0)
3870                 decimals = 0;
3871         *res = result * int_pow(10, scale - decimals);
3872         return 0;
3873 }
3874
3875 static ssize_t
3876 safe_delay_show(struct mddev *mddev, char *page)
3877 {
3878         int msec = (mddev->safemode_delay*1000)/HZ;
3879         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3880 }
3881 static ssize_t
3882 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3883 {
3884         unsigned long msec;
3885
3886         if (mddev_is_clustered(mddev)) {
3887                 pr_warn("md: Safemode is disabled for clustered mode\n");
3888                 return -EINVAL;
3889         }
3890
3891         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3892                 return -EINVAL;
3893         if (msec == 0)
3894                 mddev->safemode_delay = 0;
3895         else {
3896                 unsigned long old_delay = mddev->safemode_delay;
3897                 unsigned long new_delay = (msec*HZ)/1000;
3898
3899                 if (new_delay == 0)
3900                         new_delay = 1;
3901                 mddev->safemode_delay = new_delay;
3902                 if (new_delay < old_delay || old_delay == 0)
3903                         mod_timer(&mddev->safemode_timer, jiffies+1);
3904         }
3905         return len;
3906 }
3907 static struct md_sysfs_entry md_safe_delay =
3908 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3909
3910 static ssize_t
3911 level_show(struct mddev *mddev, char *page)
3912 {
3913         struct md_personality *p;
3914         int ret;
3915         spin_lock(&mddev->lock);
3916         p = mddev->pers;
3917         if (p)
3918                 ret = sprintf(page, "%s\n", p->name);
3919         else if (mddev->clevel[0])
3920                 ret = sprintf(page, "%s\n", mddev->clevel);
3921         else if (mddev->level != LEVEL_NONE)
3922                 ret = sprintf(page, "%d\n", mddev->level);
3923         else
3924                 ret = 0;
3925         spin_unlock(&mddev->lock);
3926         return ret;
3927 }
3928
3929 static ssize_t
3930 level_store(struct mddev *mddev, const char *buf, size_t len)
3931 {
3932         char clevel[16];
3933         ssize_t rv;
3934         size_t slen = len;
3935         struct md_personality *pers, *oldpers;
3936         long level;
3937         void *priv, *oldpriv;
3938         struct md_rdev *rdev;
3939
3940         if (slen == 0 || slen >= sizeof(clevel))
3941                 return -EINVAL;
3942
3943         rv = mddev_lock(mddev);
3944         if (rv)
3945                 return rv;
3946
3947         if (mddev->pers == NULL) {
3948                 strncpy(mddev->clevel, buf, slen);
3949                 if (mddev->clevel[slen-1] == '\n')
3950                         slen--;
3951                 mddev->clevel[slen] = 0;
3952                 mddev->level = LEVEL_NONE;
3953                 rv = len;
3954                 goto out_unlock;
3955         }
3956         rv = -EROFS;
3957         if (mddev->ro)
3958                 goto out_unlock;
3959
3960         /* request to change the personality.  Need to ensure:
3961          *  - array is not engaged in resync/recovery/reshape
3962          *  - old personality can be suspended
3963          *  - new personality will access other array.
3964          */
3965
3966         rv = -EBUSY;
3967         if (mddev->sync_thread ||
3968             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3969             mddev->reshape_position != MaxSector ||
3970             mddev->sysfs_active)
3971                 goto out_unlock;
3972
3973         rv = -EINVAL;
3974         if (!mddev->pers->quiesce) {
3975                 pr_warn("md: %s: %s does not support online personality change\n",
3976                         mdname(mddev), mddev->pers->name);
3977                 goto out_unlock;
3978         }
3979
3980         /* Now find the new personality */
3981         strncpy(clevel, buf, slen);
3982         if (clevel[slen-1] == '\n')
3983                 slen--;
3984         clevel[slen] = 0;
3985         if (kstrtol(clevel, 10, &level))
3986                 level = LEVEL_NONE;
3987
3988         if (request_module("md-%s", clevel) != 0)
3989                 request_module("md-level-%s", clevel);
3990         spin_lock(&pers_lock);
3991         pers = find_pers(level, clevel);
3992         if (!pers || !try_module_get(pers->owner)) {
3993                 spin_unlock(&pers_lock);
3994                 pr_warn("md: personality %s not loaded\n", clevel);
3995                 rv = -EINVAL;
3996                 goto out_unlock;
3997         }
3998         spin_unlock(&pers_lock);
3999
4000         if (pers == mddev->pers) {
4001                 /* Nothing to do! */
4002                 module_put(pers->owner);
4003                 rv = len;
4004                 goto out_unlock;
4005         }
4006         if (!pers->takeover) {
4007                 module_put(pers->owner);
4008                 pr_warn("md: %s: %s does not support personality takeover\n",
4009                         mdname(mddev), clevel);
4010                 rv = -EINVAL;
4011                 goto out_unlock;
4012         }
4013
4014         rdev_for_each(rdev, mddev)
4015                 rdev->new_raid_disk = rdev->raid_disk;
4016
4017         /* ->takeover must set new_* and/or delta_disks
4018          * if it succeeds, and may set them when it fails.
4019          */
4020         priv = pers->takeover(mddev);
4021         if (IS_ERR(priv)) {
4022                 mddev->new_level = mddev->level;
4023                 mddev->new_layout = mddev->layout;
4024                 mddev->new_chunk_sectors = mddev->chunk_sectors;
4025                 mddev->raid_disks -= mddev->delta_disks;
4026                 mddev->delta_disks = 0;
4027                 mddev->reshape_backwards = 0;
4028                 module_put(pers->owner);
4029                 pr_warn("md: %s: %s would not accept array\n",
4030                         mdname(mddev), clevel);
4031                 rv = PTR_ERR(priv);
4032                 goto out_unlock;
4033         }
4034
4035         /* Looks like we have a winner */
4036         mddev_suspend(mddev);
4037         mddev_detach(mddev);
4038
4039         spin_lock(&mddev->lock);
4040         oldpers = mddev->pers;
4041         oldpriv = mddev->private;
4042         mddev->pers = pers;
4043         mddev->private = priv;
4044         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4045         mddev->level = mddev->new_level;
4046         mddev->layout = mddev->new_layout;
4047         mddev->chunk_sectors = mddev->new_chunk_sectors;
4048         mddev->delta_disks = 0;
4049         mddev->reshape_backwards = 0;
4050         mddev->degraded = 0;
4051         spin_unlock(&mddev->lock);
4052
4053         if (oldpers->sync_request == NULL &&
4054             mddev->external) {
4055                 /* We are converting from a no-redundancy array
4056                  * to a redundancy array and metadata is managed
4057                  * externally so we need to be sure that writes
4058                  * won't block due to a need to transition
4059                  *      clean->dirty
4060                  * until external management is started.
4061                  */
4062                 mddev->in_sync = 0;
4063                 mddev->safemode_delay = 0;
4064                 mddev->safemode = 0;
4065         }
4066
4067         oldpers->free(mddev, oldpriv);
4068
4069         if (oldpers->sync_request == NULL &&
4070             pers->sync_request != NULL) {
4071                 /* need to add the md_redundancy_group */
4072                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4073                         pr_warn("md: cannot register extra attributes for %s\n",
4074                                 mdname(mddev));
4075                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4076                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4077                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4078         }
4079         if (oldpers->sync_request != NULL &&
4080             pers->sync_request == NULL) {
4081                 /* need to remove the md_redundancy_group */
4082                 if (mddev->to_remove == NULL)
4083                         mddev->to_remove = &md_redundancy_group;
4084         }
4085
4086         module_put(oldpers->owner);
4087
4088         rdev_for_each(rdev, mddev) {
4089                 if (rdev->raid_disk < 0)
4090                         continue;
4091                 if (rdev->new_raid_disk >= mddev->raid_disks)
4092                         rdev->new_raid_disk = -1;
4093                 if (rdev->new_raid_disk == rdev->raid_disk)
4094                         continue;
4095                 sysfs_unlink_rdev(mddev, rdev);
4096         }
4097         rdev_for_each(rdev, mddev) {
4098                 if (rdev->raid_disk < 0)
4099                         continue;
4100                 if (rdev->new_raid_disk == rdev->raid_disk)
4101                         continue;
4102                 rdev->raid_disk = rdev->new_raid_disk;
4103                 if (rdev->raid_disk < 0)
4104                         clear_bit(In_sync, &rdev->flags);
4105                 else {
4106                         if (sysfs_link_rdev(mddev, rdev))
4107                                 pr_warn("md: cannot register rd%d for %s after level change\n",
4108                                         rdev->raid_disk, mdname(mddev));
4109                 }
4110         }
4111
4112         if (pers->sync_request == NULL) {
4113                 /* this is now an array without redundancy, so
4114                  * it must always be in_sync
4115                  */
4116                 mddev->in_sync = 1;
4117                 del_timer_sync(&mddev->safemode_timer);
4118         }
4119         blk_set_stacking_limits(&mddev->queue->limits);
4120         pers->run(mddev);
4121         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4122         mddev_resume(mddev);
4123         if (!mddev->thread)
4124                 md_update_sb(mddev, 1);
4125         sysfs_notify_dirent_safe(mddev->sysfs_level);
4126         md_new_event(mddev);
4127         rv = len;
4128 out_unlock:
4129         mddev_unlock(mddev);
4130         return rv;
4131 }
4132
4133 static struct md_sysfs_entry md_level =
4134 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4135
4136 static ssize_t
4137 layout_show(struct mddev *mddev, char *page)
4138 {
4139         /* just a number, not meaningful for all levels */
4140         if (mddev->reshape_position != MaxSector &&
4141             mddev->layout != mddev->new_layout)
4142                 return sprintf(page, "%d (%d)\n",
4143                                mddev->new_layout, mddev->layout);
4144         return sprintf(page, "%d\n", mddev->layout);
4145 }
4146
4147 static ssize_t
4148 layout_store(struct mddev *mddev, const char *buf, size_t len)
4149 {
4150         unsigned int n;
4151         int err;
4152
4153         err = kstrtouint(buf, 10, &n);
4154         if (err < 0)
4155                 return err;
4156         err = mddev_lock(mddev);
4157         if (err)
4158                 return err;
4159
4160         if (mddev->pers) {
4161                 if (mddev->pers->check_reshape == NULL)
4162                         err = -EBUSY;
4163                 else if (mddev->ro)
4164                         err = -EROFS;
4165                 else {
4166                         mddev->new_layout = n;
4167                         err = mddev->pers->check_reshape(mddev);
4168                         if (err)
4169                                 mddev->new_layout = mddev->layout;
4170                 }
4171         } else {
4172                 mddev->new_layout = n;
4173                 if (mddev->reshape_position == MaxSector)
4174                         mddev->layout = n;
4175         }
4176         mddev_unlock(mddev);
4177         return err ?: len;
4178 }
4179 static struct md_sysfs_entry md_layout =
4180 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4181
4182 static ssize_t
4183 raid_disks_show(struct mddev *mddev, char *page)
4184 {
4185         if (mddev->raid_disks == 0)
4186                 return 0;
4187         if (mddev->reshape_position != MaxSector &&
4188             mddev->delta_disks != 0)
4189                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4190                                mddev->raid_disks - mddev->delta_disks);
4191         return sprintf(page, "%d\n", mddev->raid_disks);
4192 }
4193
4194 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4195
4196 static ssize_t
4197 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4198 {
4199         unsigned int n;
4200         int err;
4201
4202         err = kstrtouint(buf, 10, &n);
4203         if (err < 0)
4204                 return err;
4205
4206         err = mddev_lock(mddev);
4207         if (err)
4208                 return err;
4209         if (mddev->pers)
4210                 err = update_raid_disks(mddev, n);
4211         else if (mddev->reshape_position != MaxSector) {
4212                 struct md_rdev *rdev;
4213                 int olddisks = mddev->raid_disks - mddev->delta_disks;
4214
4215                 err = -EINVAL;
4216                 rdev_for_each(rdev, mddev) {
4217                         if (olddisks < n &&
4218                             rdev->data_offset < rdev->new_data_offset)
4219                                 goto out_unlock;
4220                         if (olddisks > n &&
4221                             rdev->data_offset > rdev->new_data_offset)
4222                                 goto out_unlock;
4223                 }
4224                 err = 0;
4225                 mddev->delta_disks = n - olddisks;
4226                 mddev->raid_disks = n;
4227                 mddev->reshape_backwards = (mddev->delta_disks < 0);
4228         } else
4229                 mddev->raid_disks = n;
4230 out_unlock:
4231         mddev_unlock(mddev);
4232         return err ? err : len;
4233 }
4234 static struct md_sysfs_entry md_raid_disks =
4235 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4236
4237 static ssize_t
4238 uuid_show(struct mddev *mddev, char *page)
4239 {
4240         return sprintf(page, "%pU\n", mddev->uuid);
4241 }
4242 static struct md_sysfs_entry md_uuid =
4243 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4244
4245 static ssize_t
4246 chunk_size_show(struct mddev *mddev, char *page)
4247 {
4248         if (mddev->reshape_position != MaxSector &&
4249             mddev->chunk_sectors != mddev->new_chunk_sectors)
4250                 return sprintf(page, "%d (%d)\n",
4251                                mddev->new_chunk_sectors << 9,
4252                                mddev->chunk_sectors << 9);
4253         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4254 }
4255
4256 static ssize_t
4257 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4258 {
4259         unsigned long n;
4260         int err;
4261
4262         err = kstrtoul(buf, 10, &n);
4263         if (err < 0)
4264                 return err;
4265
4266         err = mddev_lock(mddev);
4267         if (err)
4268                 return err;
4269         if (mddev->pers) {
4270                 if (mddev->pers->check_reshape == NULL)
4271                         err = -EBUSY;
4272                 else if (mddev->ro)
4273                         err = -EROFS;
4274                 else {
4275                         mddev->new_chunk_sectors = n >> 9;
4276                         err = mddev->pers->check_reshape(mddev);
4277                         if (err)
4278                                 mddev->new_chunk_sectors = mddev->chunk_sectors;
4279                 }
4280         } else {
4281                 mddev->new_chunk_sectors = n >> 9;
4282                 if (mddev->reshape_position == MaxSector)
4283                         mddev->chunk_sectors = n >> 9;
4284         }
4285         mddev_unlock(mddev);
4286         return err ?: len;
4287 }
4288 static struct md_sysfs_entry md_chunk_size =
4289 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4290
4291 static ssize_t
4292 resync_start_show(struct mddev *mddev, char *page)
4293 {
4294         if (mddev->recovery_cp == MaxSector)
4295                 return sprintf(page, "none\n");
4296         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4297 }
4298
4299 static ssize_t
4300 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4301 {
4302         unsigned long long n;
4303         int err;
4304
4305         if (cmd_match(buf, "none"))
4306                 n = MaxSector;
4307         else {
4308                 err = kstrtoull(buf, 10, &n);
4309                 if (err < 0)
4310                         return err;
4311                 if (n != (sector_t)n)
4312                         return -EINVAL;
4313         }
4314
4315         err = mddev_lock(mddev);
4316         if (err)
4317                 return err;
4318         if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4319                 err = -EBUSY;
4320
4321         if (!err) {
4322                 mddev->recovery_cp = n;
4323                 if (mddev->pers)
4324                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4325         }
4326         mddev_unlock(mddev);
4327         return err ?: len;
4328 }
4329 static struct md_sysfs_entry md_resync_start =
4330 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4331                 resync_start_show, resync_start_store);
4332
4333 /*
4334  * The array state can be:
4335  *
4336  * clear
4337  *     No devices, no size, no level
4338  *     Equivalent to STOP_ARRAY ioctl
4339  * inactive
4340  *     May have some settings, but array is not active
4341  *        all IO results in error
4342  *     When written, doesn't tear down array, but just stops it
4343  * suspended (not supported yet)
4344  *     All IO requests will block. The array can be reconfigured.
4345  *     Writing this, if accepted, will block until array is quiescent
4346  * readonly
4347  *     no resync can happen.  no superblocks get written.
4348  *     write requests fail
4349  * read-auto
4350  *     like readonly, but behaves like 'clean' on a write request.
4351  *
4352  * clean - no pending writes, but otherwise active.
4353  *     When written to inactive array, starts without resync
4354  *     If a write request arrives then
4355  *       if metadata is known, mark 'dirty' and switch to 'active'.
4356  *       if not known, block and switch to write-pending
4357  *     If written to an active array that has pending writes, then fails.
4358  * active
4359  *     fully active: IO and resync can be happening.
4360  *     When written to inactive array, starts with resync
4361  *
4362  * write-pending
4363  *     clean, but writes are blocked waiting for 'active' to be written.
4364  *
4365  * active-idle
4366  *     like active, but no writes have been seen for a while (100msec).
4367  *
4368  * broken
4369  *     RAID0/LINEAR-only: same as clean, but array is missing a member.
4370  *     It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4371  *     when a member is gone, so this state will at least alert the
4372  *     user that something is wrong.
4373  */
4374 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4375                    write_pending, active_idle, broken, bad_word};
4376 static char *array_states[] = {
4377         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4378         "write-pending", "active-idle", "broken", NULL };
4379
4380 static int match_word(const char *word, char **list)
4381 {
4382         int n;
4383         for (n=0; list[n]; n++)
4384                 if (cmd_match(word, list[n]))
4385                         break;
4386         return n;
4387 }
4388
4389 static ssize_t
4390 array_state_show(struct mddev *mddev, char *page)
4391 {
4392         enum array_state st = inactive;
4393
4394         if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4395                 switch(mddev->ro) {
4396                 case 1:
4397                         st = readonly;
4398                         break;
4399                 case 2:
4400                         st = read_auto;
4401                         break;
4402                 case 0:
4403                         spin_lock(&mddev->lock);
4404                         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4405                                 st = write_pending;
4406                         else if (mddev->in_sync)
4407                                 st = clean;
4408                         else if (mddev->safemode)
4409                                 st = active_idle;
4410                         else
4411                                 st = active;
4412                         spin_unlock(&mddev->lock);
4413                 }
4414
4415                 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4416                         st = broken;
4417         } else {
4418                 if (list_empty(&mddev->disks) &&
4419                     mddev->raid_disks == 0 &&
4420                     mddev->dev_sectors == 0)
4421                         st = clear;
4422                 else
4423                         st = inactive;
4424         }
4425         return sprintf(page, "%s\n", array_states[st]);
4426 }
4427
4428 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4429 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4430 static int restart_array(struct mddev *mddev);
4431
4432 static ssize_t
4433 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4434 {
4435         int err = 0;
4436         enum array_state st = match_word(buf, array_states);
4437
4438         if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4439                 /* don't take reconfig_mutex when toggling between
4440                  * clean and active
4441                  */
4442                 spin_lock(&mddev->lock);
4443                 if (st == active) {
4444                         restart_array(mddev);
4445                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4446                         md_wakeup_thread(mddev->thread);
4447                         wake_up(&mddev->sb_wait);
4448                 } else /* st == clean */ {
4449                         restart_array(mddev);
4450                         if (!set_in_sync(mddev))
4451                                 err = -EBUSY;
4452                 }
4453                 if (!err)
4454                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4455                 spin_unlock(&mddev->lock);
4456                 return err ?: len;
4457         }
4458         err = mddev_lock(mddev);
4459         if (err)
4460                 return err;
4461         err = -EINVAL;
4462         switch(st) {
4463         case bad_word:
4464                 break;
4465         case clear:
4466                 /* stopping an active array */
4467                 err = do_md_stop(mddev, 0, NULL);
4468                 break;
4469         case inactive:
4470                 /* stopping an active array */
4471                 if (mddev->pers)
4472                         err = do_md_stop(mddev, 2, NULL);
4473                 else
4474                         err = 0; /* already inactive */
4475                 break;
4476         case suspended:
4477                 break; /* not supported yet */
4478         case readonly:
4479                 if (mddev->pers)
4480                         err = md_set_readonly(mddev, NULL);
4481                 else {
4482                         mddev->ro = 1;
4483                         set_disk_ro(mddev->gendisk, 1);
4484                         err = do_md_run(mddev);
4485                 }
4486                 break;
4487         case read_auto:
4488                 if (mddev->pers) {
4489                         if (mddev->ro == 0)
4490                                 err = md_set_readonly(mddev, NULL);
4491                         else if (mddev->ro == 1)
4492                                 err = restart_array(mddev);
4493                         if (err == 0) {
4494                                 mddev->ro = 2;
4495                                 set_disk_ro(mddev->gendisk, 0);
4496                         }
4497                 } else {
4498                         mddev->ro = 2;
4499                         err = do_md_run(mddev);
4500                 }
4501                 break;
4502         case clean:
4503                 if (mddev->pers) {
4504                         err = restart_array(mddev);
4505                         if (err)
4506                                 break;
4507                         spin_lock(&mddev->lock);
4508                         if (!set_in_sync(mddev))
4509                                 err = -EBUSY;
4510                         spin_unlock(&mddev->lock);
4511                 } else
4512                         err = -EINVAL;
4513                 break;
4514         case active:
4515                 if (mddev->pers) {
4516                         err = restart_array(mddev);
4517                         if (err)
4518                                 break;
4519                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4520                         wake_up(&mddev->sb_wait);
4521                         err = 0;
4522                 } else {
4523                         mddev->ro = 0;
4524                         set_disk_ro(mddev->gendisk, 0);
4525                         err = do_md_run(mddev);
4526                 }
4527                 break;
4528         case write_pending:
4529         case active_idle:
4530         case broken:
4531                 /* these cannot be set */
4532                 break;
4533         }
4534
4535         if (!err) {
4536                 if (mddev->hold_active == UNTIL_IOCTL)
4537                         mddev->hold_active = 0;
4538                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4539         }
4540         mddev_unlock(mddev);
4541         return err ?: len;
4542 }
4543 static struct md_sysfs_entry md_array_state =
4544 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4545
4546 static ssize_t
4547 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4548         return sprintf(page, "%d\n",
4549                        atomic_read(&mddev->max_corr_read_errors));
4550 }
4551
4552 static ssize_t
4553 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4554 {
4555         unsigned int n;
4556         int rv;
4557
4558         rv = kstrtouint(buf, 10, &n);
4559         if (rv < 0)
4560                 return rv;
4561         atomic_set(&mddev->max_corr_read_errors, n);
4562         return len;
4563 }
4564
4565 static struct md_sysfs_entry max_corr_read_errors =
4566 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4567         max_corrected_read_errors_store);
4568
4569 static ssize_t
4570 null_show(struct mddev *mddev, char *page)
4571 {
4572         return -EINVAL;
4573 }
4574
4575 /* need to ensure rdev_delayed_delete() has completed */
4576 static void flush_rdev_wq(struct mddev *mddev)
4577 {
4578         struct md_rdev *rdev;
4579
4580         rcu_read_lock();
4581         rdev_for_each_rcu(rdev, mddev)
4582                 if (work_pending(&rdev->del_work)) {
4583                         flush_workqueue(md_rdev_misc_wq);
4584                         break;
4585                 }
4586         rcu_read_unlock();
4587 }
4588
4589 static ssize_t
4590 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4591 {
4592         /* buf must be %d:%d\n? giving major and minor numbers */
4593         /* The new device is added to the array.
4594          * If the array has a persistent superblock, we read the
4595          * superblock to initialise info and check validity.
4596          * Otherwise, only checking done is that in bind_rdev_to_array,
4597          * which mainly checks size.
4598          */
4599         char *e;
4600         int major = simple_strtoul(buf, &e, 10);
4601         int minor;
4602         dev_t dev;
4603         struct md_rdev *rdev;
4604         int err;
4605
4606         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4607                 return -EINVAL;
4608         minor = simple_strtoul(e+1, &e, 10);
4609         if (*e && *e != '\n')
4610                 return -EINVAL;
4611         dev = MKDEV(major, minor);
4612         if (major != MAJOR(dev) ||
4613             minor != MINOR(dev))
4614                 return -EOVERFLOW;
4615
4616         flush_rdev_wq(mddev);
4617         err = mddev_lock(mddev);
4618         if (err)
4619                 return err;
4620         if (mddev->persistent) {
4621                 rdev = md_import_device(dev, mddev->major_version,
4622                                         mddev->minor_version);
4623                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4624                         struct md_rdev *rdev0
4625                                 = list_entry(mddev->disks.next,
4626                                              struct md_rdev, same_set);
4627                         err = super_types[mddev->major_version]
4628                                 .load_super(rdev, rdev0, mddev->minor_version);
4629                         if (err < 0)
4630                                 goto out;
4631                 }
4632         } else if (mddev->external)
4633                 rdev = md_import_device(dev, -2, -1);
4634         else
4635                 rdev = md_import_device(dev, -1, -1);
4636
4637         if (IS_ERR(rdev)) {
4638                 mddev_unlock(mddev);
4639                 return PTR_ERR(rdev);
4640         }
4641         err = bind_rdev_to_array(rdev, mddev);
4642  out:
4643         if (err)
4644                 export_rdev(rdev);
4645         mddev_unlock(mddev);
4646         if (!err)
4647                 md_new_event(mddev);
4648         return err ? err : len;
4649 }
4650
4651 static struct md_sysfs_entry md_new_device =
4652 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4653
4654 static ssize_t
4655 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4656 {
4657         char *end;
4658         unsigned long chunk, end_chunk;
4659         int err;
4660
4661         err = mddev_lock(mddev);
4662         if (err)
4663                 return err;
4664         if (!mddev->bitmap)
4665                 goto out;
4666         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4667         while (*buf) {
4668                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4669                 if (buf == end) break;
4670                 if (*end == '-') { /* range */
4671                         buf = end + 1;
4672                         end_chunk = simple_strtoul(buf, &end, 0);
4673                         if (buf == end) break;
4674                 }
4675                 if (*end && !isspace(*end)) break;
4676                 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4677                 buf = skip_spaces(end);
4678         }
4679         md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4680 out:
4681         mddev_unlock(mddev);
4682         return len;
4683 }
4684
4685 static struct md_sysfs_entry md_bitmap =
4686 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4687
4688 static ssize_t
4689 size_show(struct mddev *mddev, char *page)
4690 {
4691         return sprintf(page, "%llu\n",
4692                 (unsigned long long)mddev->dev_sectors / 2);
4693 }
4694
4695 static int update_size(struct mddev *mddev, sector_t num_sectors);
4696
4697 static ssize_t
4698 size_store(struct mddev *mddev, const char *buf, size_t len)
4699 {
4700         /* If array is inactive, we can reduce the component size, but
4701          * not increase it (except from 0).
4702          * If array is active, we can try an on-line resize
4703          */
4704         sector_t sectors;
4705         int err = strict_blocks_to_sectors(buf, &sectors);
4706
4707         if (err < 0)
4708                 return err;
4709         err = mddev_lock(mddev);
4710         if (err)
4711                 return err;
4712         if (mddev->pers) {
4713                 err = update_size(mddev, sectors);
4714                 if (err == 0)
4715                         md_update_sb(mddev, 1);
4716         } else {
4717                 if (mddev->dev_sectors == 0 ||
4718                     mddev->dev_sectors > sectors)
4719                         mddev->dev_sectors = sectors;
4720                 else
4721                         err = -ENOSPC;
4722         }
4723         mddev_unlock(mddev);
4724         return err ? err : len;
4725 }
4726
4727 static struct md_sysfs_entry md_size =
4728 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4729
4730 /* Metadata version.
4731  * This is one of
4732  *   'none' for arrays with no metadata (good luck...)
4733  *   'external' for arrays with externally managed metadata,
4734  * or N.M for internally known formats
4735  */
4736 static ssize_t
4737 metadata_show(struct mddev *mddev, char *page)
4738 {
4739         if (mddev->persistent)
4740                 return sprintf(page, "%d.%d\n",
4741                                mddev->major_version, mddev->minor_version);
4742         else if (mddev->external)
4743                 return sprintf(page, "external:%s\n", mddev->metadata_type);
4744         else
4745                 return sprintf(page, "none\n");
4746 }
4747
4748 static ssize_t
4749 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4750 {
4751         int major, minor;
4752         char *e;
4753         int err;
4754         /* Changing the details of 'external' metadata is
4755          * always permitted.  Otherwise there must be
4756          * no devices attached to the array.
4757          */
4758
4759         err = mddev_lock(mddev);
4760         if (err)
4761                 return err;
4762         err = -EBUSY;
4763         if (mddev->external && strncmp(buf, "external:", 9) == 0)
4764                 ;
4765         else if (!list_empty(&mddev->disks))
4766                 goto out_unlock;
4767
4768         err = 0;
4769         if (cmd_match(buf, "none")) {
4770                 mddev->persistent = 0;
4771                 mddev->external = 0;
4772                 mddev->major_version = 0;
4773                 mddev->minor_version = 90;
4774                 goto out_unlock;
4775         }
4776         if (strncmp(buf, "external:", 9) == 0) {
4777                 size_t namelen = len-9;
4778                 if (namelen >= sizeof(mddev->metadata_type))
4779                         namelen = sizeof(mddev->metadata_type)-1;
4780                 strncpy(mddev->metadata_type, buf+9, namelen);
4781                 mddev->metadata_type[namelen] = 0;
4782                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4783                         mddev->metadata_type[--namelen] = 0;
4784                 mddev->persistent = 0;
4785                 mddev->external = 1;
4786                 mddev->major_version = 0;
4787                 mddev->minor_version = 90;
4788                 goto out_unlock;
4789         }
4790         major = simple_strtoul(buf, &e, 10);
4791         err = -EINVAL;
4792         if (e==buf || *e != '.')
4793                 goto out_unlock;
4794         buf = e+1;
4795         minor = simple_strtoul(buf, &e, 10);
4796         if (e==buf || (*e && *e != '\n') )
4797                 goto out_unlock;
4798         err = -ENOENT;
4799         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4800                 goto out_unlock;
4801         mddev->major_version = major;
4802         mddev->minor_version = minor;
4803         mddev->persistent = 1;
4804         mddev->external = 0;
4805         err = 0;
4806 out_unlock:
4807         mddev_unlock(mddev);
4808         return err ?: len;
4809 }
4810
4811 static struct md_sysfs_entry md_metadata =
4812 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4813
4814 static ssize_t
4815 action_show(struct mddev *mddev, char *page)
4816 {
4817         char *type = "idle";
4818         unsigned long recovery = mddev->recovery;
4819         if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4820                 type = "frozen";
4821         else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4822             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4823                 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4824                         type = "reshape";
4825                 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4826                         if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4827                                 type = "resync";
4828                         else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4829                                 type = "check";
4830                         else
4831                                 type = "repair";
4832                 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4833                         type = "recover";
4834                 else if (mddev->reshape_position != MaxSector)
4835                         type = "reshape";
4836         }
4837         return sprintf(page, "%s\n", type);
4838 }
4839
4840 static ssize_t
4841 action_store(struct mddev *mddev, const char *page, size_t len)
4842 {
4843         if (!mddev->pers || !mddev->pers->sync_request)
4844                 return -EINVAL;
4845
4846
4847         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4848                 if (cmd_match(page, "frozen"))
4849                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4850                 else
4851                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4852                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4853                     mddev_lock(mddev) == 0) {
4854                         if (work_pending(&mddev->del_work))
4855                                 flush_workqueue(md_misc_wq);
4856                         if (mddev->sync_thread) {
4857                                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4858                                 md_reap_sync_thread(mddev);
4859                         }
4860                         mddev_unlock(mddev);
4861                 }
4862         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4863                 return -EBUSY;
4864         else if (cmd_match(page, "resync"))
4865                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4866         else if (cmd_match(page, "recover")) {
4867                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4868                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4869         } else if (cmd_match(page, "reshape")) {
4870                 int err;
4871                 if (mddev->pers->start_reshape == NULL)
4872                         return -EINVAL;
4873                 err = mddev_lock(mddev);
4874                 if (!err) {
4875                         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4876                                 err =  -EBUSY;
4877                         else {
4878                                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4879                                 err = mddev->pers->start_reshape(mddev);
4880                         }
4881                         mddev_unlock(mddev);
4882                 }
4883                 if (err)
4884                         return err;
4885                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4886         } else {
4887                 if (cmd_match(page, "check"))
4888                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4889                 else if (!cmd_match(page, "repair"))
4890                         return -EINVAL;
4891                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4892                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4893                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4894         }
4895         if (mddev->ro == 2) {
4896                 /* A write to sync_action is enough to justify
4897                  * canceling read-auto mode
4898                  */
4899                 mddev->ro = 0;
4900                 md_wakeup_thread(mddev->sync_thread);
4901         }
4902         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4903         md_wakeup_thread(mddev->thread);
4904         sysfs_notify_dirent_safe(mddev->sysfs_action);
4905         return len;
4906 }
4907
4908 static struct md_sysfs_entry md_scan_mode =
4909 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4910
4911 static ssize_t
4912 last_sync_action_show(struct mddev *mddev, char *page)
4913 {
4914         return sprintf(page, "%s\n", mddev->last_sync_action);
4915 }
4916
4917 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4918
4919 static ssize_t
4920 mismatch_cnt_show(struct mddev *mddev, char *page)
4921 {
4922         return sprintf(page, "%llu\n",
4923                        (unsigned long long)
4924                        atomic64_read(&mddev->resync_mismatches));
4925 }
4926
4927 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4928
4929 static ssize_t
4930 sync_min_show(struct mddev *mddev, char *page)
4931 {
4932         return sprintf(page, "%d (%s)\n", speed_min(mddev),
4933                        mddev->sync_speed_min ? "local": "system");
4934 }
4935
4936 static ssize_t
4937 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4938 {
4939         unsigned int min;
4940         int rv;
4941
4942         if (strncmp(buf, "system", 6)==0) {
4943                 min = 0;
4944         } else {
4945                 rv = kstrtouint(buf, 10, &min);
4946                 if (rv < 0)
4947                         return rv;
4948                 if (min == 0)
4949                         return -EINVAL;
4950         }
4951         mddev->sync_speed_min = min;
4952         return len;
4953 }
4954
4955 static struct md_sysfs_entry md_sync_min =
4956 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4957
4958 static ssize_t
4959 sync_max_show(struct mddev *mddev, char *page)
4960 {
4961         return sprintf(page, "%d (%s)\n", speed_max(mddev),
4962                        mddev->sync_speed_max ? "local": "system");
4963 }
4964
4965 static ssize_t
4966 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4967 {
4968         unsigned int max;
4969         int rv;
4970
4971         if (strncmp(buf, "system", 6)==0) {
4972                 max = 0;
4973         } else {
4974                 rv = kstrtouint(buf, 10, &max);
4975                 if (rv < 0)
4976                         return rv;
4977                 if (max == 0)
4978                         return -EINVAL;
4979         }
4980         mddev->sync_speed_max = max;
4981         return len;
4982 }
4983
4984 static struct md_sysfs_entry md_sync_max =
4985 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4986
4987 static ssize_t
4988 degraded_show(struct mddev *mddev, char *page)
4989 {
4990         return sprintf(page, "%d\n", mddev->degraded);
4991 }
4992 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4993
4994 static ssize_t
4995 sync_force_parallel_show(struct mddev *mddev, char *page)
4996 {
4997         return sprintf(page, "%d\n", mddev->parallel_resync);
4998 }
4999
5000 static ssize_t
5001 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5002 {
5003         long n;
5004
5005         if (kstrtol(buf, 10, &n))
5006                 return -EINVAL;
5007
5008         if (n != 0 && n != 1)
5009                 return -EINVAL;
5010
5011         mddev->parallel_resync = n;
5012
5013         if (mddev->sync_thread)
5014                 wake_up(&resync_wait);
5015
5016         return len;
5017 }
5018
5019 /* force parallel resync, even with shared block devices */
5020 static struct md_sysfs_entry md_sync_force_parallel =
5021 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5022        sync_force_parallel_show, sync_force_parallel_store);
5023
5024 static ssize_t
5025 sync_speed_show(struct mddev *mddev, char *page)
5026 {
5027         unsigned long resync, dt, db;
5028         if (mddev->curr_resync == 0)
5029                 return sprintf(page, "none\n");
5030         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5031         dt = (jiffies - mddev->resync_mark) / HZ;
5032         if (!dt) dt++;
5033         db = resync - mddev->resync_mark_cnt;
5034         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5035 }
5036
5037 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5038
5039 static ssize_t
5040 sync_completed_show(struct mddev *mddev, char *page)
5041 {
5042         unsigned long long max_sectors, resync;
5043
5044         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5045                 return sprintf(page, "none\n");
5046
5047         if (mddev->curr_resync == 1 ||
5048             mddev->curr_resync == 2)
5049                 return sprintf(page, "delayed\n");
5050
5051         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5052             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5053                 max_sectors = mddev->resync_max_sectors;
5054         else
5055                 max_sectors = mddev->dev_sectors;
5056
5057         resync = mddev->curr_resync_completed;
5058         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5059 }
5060
5061 static struct md_sysfs_entry md_sync_completed =
5062         __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5063
5064 static ssize_t
5065 min_sync_show(struct mddev *mddev, char *page)
5066 {
5067         return sprintf(page, "%llu\n",
5068                        (unsigned long long)mddev->resync_min);
5069 }
5070 static ssize_t
5071 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5072 {
5073         unsigned long long min;
5074         int err;
5075
5076         if (kstrtoull(buf, 10, &min))
5077                 return -EINVAL;
5078
5079         spin_lock(&mddev->lock);
5080         err = -EINVAL;
5081         if (min > mddev->resync_max)
5082                 goto out_unlock;
5083
5084         err = -EBUSY;
5085         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5086                 goto out_unlock;
5087
5088         /* Round down to multiple of 4K for safety */
5089         mddev->resync_min = round_down(min, 8);
5090         err = 0;
5091
5092 out_unlock:
5093         spin_unlock(&mddev->lock);
5094         return err ?: len;
5095 }
5096
5097 static struct md_sysfs_entry md_min_sync =
5098 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5099
5100 static ssize_t
5101 max_sync_show(struct mddev *mddev, char *page)
5102 {
5103         if (mddev->resync_max == MaxSector)
5104                 return sprintf(page, "max\n");
5105         else
5106                 return sprintf(page, "%llu\n",
5107                                (unsigned long long)mddev->resync_max);
5108 }
5109 static ssize_t
5110 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5111 {
5112         int err;
5113         spin_lock(&mddev->lock);
5114         if (strncmp(buf, "max", 3) == 0)
5115                 mddev->resync_max = MaxSector;
5116         else {
5117                 unsigned long long max;
5118                 int chunk;
5119
5120                 err = -EINVAL;
5121                 if (kstrtoull(buf, 10, &max))
5122                         goto out_unlock;
5123                 if (max < mddev->resync_min)
5124                         goto out_unlock;
5125
5126                 err = -EBUSY;
5127                 if (max < mddev->resync_max &&
5128                     mddev->ro == 0 &&
5129                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5130                         goto out_unlock;
5131
5132                 /* Must be a multiple of chunk_size */
5133                 chunk = mddev->chunk_sectors;
5134                 if (chunk) {
5135                         sector_t temp = max;
5136
5137                         err = -EINVAL;
5138                         if (sector_div(temp, chunk))
5139                                 goto out_unlock;
5140                 }
5141                 mddev->resync_max = max;
5142         }
5143         wake_up(&mddev->recovery_wait);
5144         err = 0;
5145 out_unlock:
5146         spin_unlock(&mddev->lock);
5147         return err ?: len;
5148 }
5149
5150 static struct md_sysfs_entry md_max_sync =
5151 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5152
5153 static ssize_t
5154 suspend_lo_show(struct mddev *mddev, char *page)
5155 {
5156         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5157 }
5158
5159 static ssize_t
5160 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5161 {
5162         unsigned long long new;
5163         int err;
5164
5165         err = kstrtoull(buf, 10, &new);
5166         if (err < 0)
5167                 return err;
5168         if (new != (sector_t)new)
5169                 return -EINVAL;
5170
5171         err = mddev_lock(mddev);
5172         if (err)
5173                 return err;
5174         err = -EINVAL;
5175         if (mddev->pers == NULL ||
5176             mddev->pers->quiesce == NULL)
5177                 goto unlock;
5178         mddev_suspend(mddev);
5179         mddev->suspend_lo = new;
5180         mddev_resume(mddev);
5181
5182         err = 0;
5183 unlock:
5184         mddev_unlock(mddev);
5185         return err ?: len;
5186 }
5187 static struct md_sysfs_entry md_suspend_lo =
5188 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5189
5190 static ssize_t
5191 suspend_hi_show(struct mddev *mddev, char *page)
5192 {
5193         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5194 }
5195
5196 static ssize_t
5197 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5198 {
5199         unsigned long long new;
5200         int err;
5201
5202         err = kstrtoull(buf, 10, &new);
5203         if (err < 0)
5204                 return err;
5205         if (new != (sector_t)new)
5206                 return -EINVAL;
5207
5208         err = mddev_lock(mddev);
5209         if (err)
5210                 return err;
5211         err = -EINVAL;
5212         if (mddev->pers == NULL)
5213                 goto unlock;
5214
5215         mddev_suspend(mddev);
5216         mddev->suspend_hi = new;
5217         mddev_resume(mddev);
5218
5219         err = 0;
5220 unlock:
5221         mddev_unlock(mddev);
5222         return err ?: len;
5223 }
5224 static struct md_sysfs_entry md_suspend_hi =
5225 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5226
5227 static ssize_t
5228 reshape_position_show(struct mddev *mddev, char *page)
5229 {
5230         if (mddev->reshape_position != MaxSector)
5231                 return sprintf(page, "%llu\n",
5232                                (unsigned long long)mddev->reshape_position);
5233         strcpy(page, "none\n");
5234         return 5;
5235 }
5236
5237 static ssize_t
5238 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5239 {
5240         struct md_rdev *rdev;
5241         unsigned long long new;
5242         int err;
5243
5244         err = kstrtoull(buf, 10, &new);
5245         if (err < 0)
5246                 return err;
5247         if (new != (sector_t)new)
5248                 return -EINVAL;
5249         err = mddev_lock(mddev);
5250         if (err)
5251                 return err;
5252         err = -EBUSY;
5253         if (mddev->pers)
5254                 goto unlock;
5255         mddev->reshape_position = new;
5256         mddev->delta_disks = 0;
5257         mddev->reshape_backwards = 0;
5258         mddev->new_level = mddev->level;
5259         mddev->new_layout = mddev->layout;
5260         mddev->new_chunk_sectors = mddev->chunk_sectors;
5261         rdev_for_each(rdev, mddev)
5262                 rdev->new_data_offset = rdev->data_offset;
5263         err = 0;
5264 unlock:
5265         mddev_unlock(mddev);
5266         return err ?: len;
5267 }
5268
5269 static struct md_sysfs_entry md_reshape_position =
5270 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5271        reshape_position_store);
5272
5273 static ssize_t
5274 reshape_direction_show(struct mddev *mddev, char *page)
5275 {
5276         return sprintf(page, "%s\n",
5277                        mddev->reshape_backwards ? "backwards" : "forwards");
5278 }
5279
5280 static ssize_t
5281 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5282 {
5283         int backwards = 0;
5284         int err;
5285
5286         if (cmd_match(buf, "forwards"))
5287                 backwards = 0;
5288         else if (cmd_match(buf, "backwards"))
5289                 backwards = 1;
5290         else
5291                 return -EINVAL;
5292         if (mddev->reshape_backwards == backwards)
5293                 return len;
5294
5295         err = mddev_lock(mddev);
5296         if (err)
5297                 return err;
5298         /* check if we are allowed to change */
5299         if (mddev->delta_disks)
5300                 err = -EBUSY;
5301         else if (mddev->persistent &&
5302             mddev->major_version == 0)
5303                 err =  -EINVAL;
5304         else
5305                 mddev->reshape_backwards = backwards;
5306         mddev_unlock(mddev);
5307         return err ?: len;
5308 }
5309
5310 static struct md_sysfs_entry md_reshape_direction =
5311 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5312        reshape_direction_store);
5313
5314 static ssize_t
5315 array_size_show(struct mddev *mddev, char *page)
5316 {
5317         if (mddev->external_size)
5318                 return sprintf(page, "%llu\n",
5319                                (unsigned long long)mddev->array_sectors/2);
5320         else
5321                 return sprintf(page, "default\n");
5322 }
5323
5324 static ssize_t
5325 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5326 {
5327         sector_t sectors;
5328         int err;
5329
5330         err = mddev_lock(mddev);
5331         if (err)
5332                 return err;
5333
5334         /* cluster raid doesn't support change array_sectors */
5335         if (mddev_is_clustered(mddev)) {
5336                 mddev_unlock(mddev);
5337                 return -EINVAL;
5338         }
5339
5340         if (strncmp(buf, "default", 7) == 0) {
5341                 if (mddev->pers)
5342                         sectors = mddev->pers->size(mddev, 0, 0);
5343                 else
5344                         sectors = mddev->array_sectors;
5345
5346                 mddev->external_size = 0;
5347         } else {
5348                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
5349                         err = -EINVAL;
5350                 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5351                         err = -E2BIG;
5352                 else
5353                         mddev->external_size = 1;
5354         }
5355
5356         if (!err) {
5357                 mddev->array_sectors = sectors;
5358                 if (mddev->pers) {
5359                         set_capacity(mddev->gendisk, mddev->array_sectors);
5360                         revalidate_disk_size(mddev->gendisk, true);
5361                 }
5362         }
5363         mddev_unlock(mddev);
5364         return err ?: len;
5365 }
5366
5367 static struct md_sysfs_entry md_array_size =
5368 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5369        array_size_store);
5370
5371 static ssize_t
5372 consistency_policy_show(struct mddev *mddev, char *page)
5373 {
5374         int ret;
5375
5376         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5377                 ret = sprintf(page, "journal\n");
5378         } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5379                 ret = sprintf(page, "ppl\n");
5380         } else if (mddev->bitmap) {
5381                 ret = sprintf(page, "bitmap\n");
5382         } else if (mddev->pers) {
5383                 if (mddev->pers->sync_request)
5384                         ret = sprintf(page, "resync\n");
5385                 else
5386                         ret = sprintf(page, "none\n");
5387         } else {
5388                 ret = sprintf(page, "unknown\n");
5389         }
5390
5391         return ret;
5392 }
5393
5394 static ssize_t
5395 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5396 {
5397         int err = 0;
5398
5399         if (mddev->pers) {
5400                 if (mddev->pers->change_consistency_policy)
5401                         err = mddev->pers->change_consistency_policy(mddev, buf);
5402                 else
5403                         err = -EBUSY;
5404         } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5405                 set_bit(MD_HAS_PPL, &mddev->flags);
5406         } else {
5407                 err = -EINVAL;
5408         }
5409
5410         return err ? err : len;
5411 }
5412
5413 static struct md_sysfs_entry md_consistency_policy =
5414 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5415        consistency_policy_store);
5416
5417 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5418 {
5419         return sprintf(page, "%d\n", mddev->fail_last_dev);
5420 }
5421
5422 /*
5423  * Setting fail_last_dev to true to allow last device to be forcibly removed
5424  * from RAID1/RAID10.
5425  */
5426 static ssize_t
5427 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5428 {
5429         int ret;
5430         bool value;
5431
5432         ret = kstrtobool(buf, &value);
5433         if (ret)
5434                 return ret;
5435
5436         if (value != mddev->fail_last_dev)
5437                 mddev->fail_last_dev = value;
5438
5439         return len;
5440 }
5441 static struct md_sysfs_entry md_fail_last_dev =
5442 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5443        fail_last_dev_store);
5444
5445 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5446 {
5447         if (mddev->pers == NULL || (mddev->pers->level != 1))
5448                 return sprintf(page, "n/a\n");
5449         else
5450                 return sprintf(page, "%d\n", mddev->serialize_policy);
5451 }
5452
5453 /*
5454  * Setting serialize_policy to true to enforce write IO is not reordered
5455  * for raid1.
5456  */
5457 static ssize_t
5458 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5459 {
5460         int err;
5461         bool value;
5462
5463         err = kstrtobool(buf, &value);
5464         if (err)
5465                 return err;
5466
5467         if (value == mddev->serialize_policy)
5468                 return len;
5469
5470         err = mddev_lock(mddev);
5471         if (err)
5472                 return err;
5473         if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5474                 pr_err("md: serialize_policy is only effective for raid1\n");
5475                 err = -EINVAL;
5476                 goto unlock;
5477         }
5478
5479         mddev_suspend(mddev);
5480         if (value)
5481                 mddev_create_serial_pool(mddev, NULL, true);
5482         else
5483                 mddev_destroy_serial_pool(mddev, NULL, true);
5484         mddev->serialize_policy = value;
5485         mddev_resume(mddev);
5486 unlock:
5487         mddev_unlock(mddev);
5488         return err ?: len;
5489 }
5490
5491 static struct md_sysfs_entry md_serialize_policy =
5492 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5493        serialize_policy_store);
5494
5495
5496 static struct attribute *md_default_attrs[] = {
5497         &md_level.attr,
5498         &md_layout.attr,
5499         &md_raid_disks.attr,
5500         &md_uuid.attr,
5501         &md_chunk_size.attr,
5502         &md_size.attr,
5503         &md_resync_start.attr,
5504         &md_metadata.attr,
5505         &md_new_device.attr,
5506         &md_safe_delay.attr,
5507         &md_array_state.attr,
5508         &md_reshape_position.attr,
5509         &md_reshape_direction.attr,
5510         &md_array_size.attr,
5511         &max_corr_read_errors.attr,
5512         &md_consistency_policy.attr,
5513         &md_fail_last_dev.attr,
5514         &md_serialize_policy.attr,
5515         NULL,
5516 };
5517
5518 static struct attribute *md_redundancy_attrs[] = {
5519         &md_scan_mode.attr,
5520         &md_last_scan_mode.attr,
5521         &md_mismatches.attr,
5522         &md_sync_min.attr,
5523         &md_sync_max.attr,
5524         &md_sync_speed.attr,
5525         &md_sync_force_parallel.attr,
5526         &md_sync_completed.attr,
5527         &md_min_sync.attr,
5528         &md_max_sync.attr,
5529         &md_suspend_lo.attr,
5530         &md_suspend_hi.attr,
5531         &md_bitmap.attr,
5532         &md_degraded.attr,
5533         NULL,
5534 };
5535 static struct attribute_group md_redundancy_group = {
5536         .name = NULL,
5537         .attrs = md_redundancy_attrs,
5538 };
5539
5540 static ssize_t
5541 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5542 {
5543         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5544         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5545         ssize_t rv;
5546
5547         if (!entry->show)
5548                 return -EIO;
5549         spin_lock(&all_mddevs_lock);
5550         if (list_empty(&mddev->all_mddevs)) {
5551                 spin_unlock(&all_mddevs_lock);
5552                 return -EBUSY;
5553         }
5554         mddev_get(mddev);
5555         spin_unlock(&all_mddevs_lock);
5556
5557         rv = entry->show(mddev, page);
5558         mddev_put(mddev);
5559         return rv;
5560 }
5561
5562 static ssize_t
5563 md_attr_store(struct kobject *kobj, struct attribute *attr,
5564               const char *page, size_t length)
5565 {
5566         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5567         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5568         ssize_t rv;
5569
5570         if (!entry->store)
5571                 return -EIO;
5572         if (!capable(CAP_SYS_ADMIN))
5573                 return -EACCES;
5574         spin_lock(&all_mddevs_lock);
5575         if (list_empty(&mddev->all_mddevs)) {
5576                 spin_unlock(&all_mddevs_lock);
5577                 return -EBUSY;
5578         }
5579         mddev_get(mddev);
5580         spin_unlock(&all_mddevs_lock);
5581         rv = entry->store(mddev, page, length);
5582         mddev_put(mddev);
5583         return rv;
5584 }
5585
5586 static void md_free(struct kobject *ko)
5587 {
5588         struct mddev *mddev = container_of(ko, struct mddev, kobj);
5589
5590         if (mddev->sysfs_state)
5591                 sysfs_put(mddev->sysfs_state);
5592         if (mddev->sysfs_level)
5593                 sysfs_put(mddev->sysfs_level);
5594
5595         if (mddev->gendisk)
5596                 del_gendisk(mddev->gendisk);
5597         if (mddev->queue)
5598                 blk_cleanup_queue(mddev->queue);
5599         if (mddev->gendisk)
5600                 put_disk(mddev->gendisk);
5601         percpu_ref_exit(&mddev->writes_pending);
5602
5603         bioset_exit(&mddev->bio_set);
5604         bioset_exit(&mddev->sync_set);
5605         mempool_exit(&mddev->md_io_pool);
5606         kfree(mddev);
5607 }
5608
5609 static const struct sysfs_ops md_sysfs_ops = {
5610         .show   = md_attr_show,
5611         .store  = md_attr_store,
5612 };
5613 static struct kobj_type md_ktype = {
5614         .release        = md_free,
5615         .sysfs_ops      = &md_sysfs_ops,
5616         .default_attrs  = md_default_attrs,
5617 };
5618
5619 int mdp_major = 0;
5620
5621 static void mddev_delayed_delete(struct work_struct *ws)
5622 {
5623         struct mddev *mddev = container_of(ws, struct mddev, del_work);
5624
5625         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5626         kobject_del(&mddev->kobj);
5627         kobject_put(&mddev->kobj);
5628 }
5629
5630 static void no_op(struct percpu_ref *r) {}
5631
5632 int mddev_init_writes_pending(struct mddev *mddev)
5633 {
5634         if (mddev->writes_pending.percpu_count_ptr)
5635                 return 0;
5636         if (percpu_ref_init(&mddev->writes_pending, no_op,
5637                             PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5638                 return -ENOMEM;
5639         /* We want to start with the refcount at zero */
5640         percpu_ref_put(&mddev->writes_pending);
5641         return 0;
5642 }
5643 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5644
5645 static int md_alloc(dev_t dev, char *name)
5646 {
5647         /*
5648          * If dev is zero, name is the name of a device to allocate with
5649          * an arbitrary minor number.  It will be "md_???"
5650          * If dev is non-zero it must be a device number with a MAJOR of
5651          * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5652          * the device is being created by opening a node in /dev.
5653          * If "name" is not NULL, the device is being created by
5654          * writing to /sys/module/md_mod/parameters/new_array.
5655          */
5656         static DEFINE_MUTEX(disks_mutex);
5657         struct mddev *mddev = mddev_find(dev);
5658         struct gendisk *disk;
5659         int partitioned;
5660         int shift;
5661         int unit;
5662         int error;
5663
5664         if (!mddev)
5665                 return -ENODEV;
5666
5667         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5668         shift = partitioned ? MdpMinorShift : 0;
5669         unit = MINOR(mddev->unit) >> shift;
5670
5671         /* wait for any previous instance of this device to be
5672          * completely removed (mddev_delayed_delete).
5673          */
5674         flush_workqueue(md_misc_wq);
5675
5676         mutex_lock(&disks_mutex);
5677         error = -EEXIST;
5678         if (mddev->gendisk)
5679                 goto abort;
5680
5681         if (name && !dev) {
5682                 /* Need to ensure that 'name' is not a duplicate.
5683                  */
5684                 struct mddev *mddev2;
5685                 spin_lock(&all_mddevs_lock);
5686
5687                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5688                         if (mddev2->gendisk &&
5689                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
5690                                 spin_unlock(&all_mddevs_lock);
5691                                 goto abort;
5692                         }
5693                 spin_unlock(&all_mddevs_lock);
5694         }
5695         if (name && dev)
5696                 /*
5697                  * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5698                  */
5699                 mddev->hold_active = UNTIL_STOP;
5700
5701         error = mempool_init_kmalloc_pool(&mddev->md_io_pool, BIO_POOL_SIZE,
5702                                           sizeof(struct md_io));
5703         if (error)
5704                 goto abort;
5705
5706         error = -ENOMEM;
5707         mddev->queue = blk_alloc_queue(NUMA_NO_NODE);
5708         if (!mddev->queue)
5709                 goto abort;
5710
5711         blk_set_stacking_limits(&mddev->queue->limits);
5712
5713         disk = alloc_disk(1 << shift);
5714         if (!disk) {
5715                 blk_cleanup_queue(mddev->queue);
5716                 mddev->queue = NULL;
5717                 goto abort;
5718         }
5719         disk->major = MAJOR(mddev->unit);
5720         disk->first_minor = unit << shift;
5721         if (name)
5722                 strcpy(disk->disk_name, name);
5723         else if (partitioned)
5724                 sprintf(disk->disk_name, "md_d%d", unit);
5725         else
5726                 sprintf(disk->disk_name, "md%d", unit);
5727         disk->fops = &md_fops;
5728         disk->private_data = mddev;
5729         disk->queue = mddev->queue;
5730         blk_queue_write_cache(mddev->queue, true, true);
5731         /* Allow extended partitions.  This makes the
5732          * 'mdp' device redundant, but we can't really
5733          * remove it now.
5734          */
5735         disk->flags |= GENHD_FL_EXT_DEVT;
5736         disk->events |= DISK_EVENT_MEDIA_CHANGE;
5737         mddev->gendisk = disk;
5738         /* As soon as we call add_disk(), another thread could get
5739          * through to md_open, so make sure it doesn't get too far
5740          */
5741         mutex_lock(&mddev->open_mutex);
5742         add_disk(disk);
5743
5744         error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5745         if (error) {
5746                 /* This isn't possible, but as kobject_init_and_add is marked
5747                  * __must_check, we must do something with the result
5748                  */
5749                 pr_debug("md: cannot register %s/md - name in use\n",
5750                          disk->disk_name);
5751                 error = 0;
5752         }
5753         if (mddev->kobj.sd &&
5754             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5755                 pr_debug("pointless warning\n");
5756         mutex_unlock(&mddev->open_mutex);
5757  abort:
5758         mutex_unlock(&disks_mutex);
5759         if (!error && mddev->kobj.sd) {
5760                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5761                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5762                 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5763         }
5764         mddev_put(mddev);
5765         return error;
5766 }
5767
5768 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5769 {
5770         if (create_on_open)
5771                 md_alloc(dev, NULL);
5772         return NULL;
5773 }
5774
5775 static int add_named_array(const char *val, const struct kernel_param *kp)
5776 {
5777         /*
5778          * val must be "md_*" or "mdNNN".
5779          * For "md_*" we allocate an array with a large free minor number, and
5780          * set the name to val.  val must not already be an active name.
5781          * For "mdNNN" we allocate an array with the minor number NNN
5782          * which must not already be in use.
5783          */
5784         int len = strlen(val);
5785         char buf[DISK_NAME_LEN];
5786         unsigned long devnum;
5787
5788         while (len && val[len-1] == '\n')
5789                 len--;
5790         if (len >= DISK_NAME_LEN)
5791                 return -E2BIG;
5792         strlcpy(buf, val, len+1);
5793         if (strncmp(buf, "md_", 3) == 0)
5794                 return md_alloc(0, buf);
5795         if (strncmp(buf, "md", 2) == 0 &&
5796             isdigit(buf[2]) &&
5797             kstrtoul(buf+2, 10, &devnum) == 0 &&
5798             devnum <= MINORMASK)
5799                 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5800
5801         return -EINVAL;
5802 }
5803
5804 static void md_safemode_timeout(struct timer_list *t)
5805 {
5806         struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5807
5808         mddev->safemode = 1;
5809         if (mddev->external)
5810                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5811
5812         md_wakeup_thread(mddev->thread);
5813 }
5814
5815 static int start_dirty_degraded;
5816
5817 int md_run(struct mddev *mddev)
5818 {
5819         int err;
5820         struct md_rdev *rdev;
5821         struct md_personality *pers;
5822
5823         if (list_empty(&mddev->disks))
5824                 /* cannot run an array with no devices.. */
5825                 return -EINVAL;
5826
5827         if (mddev->pers)
5828                 return -EBUSY;
5829         /* Cannot run until previous stop completes properly */
5830         if (mddev->sysfs_active)
5831                 return -EBUSY;
5832
5833         /*
5834          * Analyze all RAID superblock(s)
5835          */
5836         if (!mddev->raid_disks) {
5837                 if (!mddev->persistent)
5838                         return -EINVAL;
5839                 err = analyze_sbs(mddev);
5840                 if (err)
5841                         return -EINVAL;
5842         }
5843
5844         if (mddev->level != LEVEL_NONE)
5845                 request_module("md-level-%d", mddev->level);
5846         else if (mddev->clevel[0])
5847                 request_module("md-%s", mddev->clevel);
5848
5849         /*
5850          * Drop all container device buffers, from now on
5851          * the only valid external interface is through the md
5852          * device.
5853          */
5854         mddev->has_superblocks = false;
5855         rdev_for_each(rdev, mddev) {
5856                 if (test_bit(Faulty, &rdev->flags))
5857                         continue;
5858                 sync_blockdev(rdev->bdev);
5859                 invalidate_bdev(rdev->bdev);
5860                 if (mddev->ro != 1 &&
5861                     (bdev_read_only(rdev->bdev) ||
5862                      bdev_read_only(rdev->meta_bdev))) {
5863                         mddev->ro = 1;
5864                         if (mddev->gendisk)
5865                                 set_disk_ro(mddev->gendisk, 1);
5866                 }
5867
5868                 if (rdev->sb_page)
5869                         mddev->has_superblocks = true;
5870
5871                 /* perform some consistency tests on the device.
5872                  * We don't want the data to overlap the metadata,
5873                  * Internal Bitmap issues have been handled elsewhere.
5874                  */
5875                 if (rdev->meta_bdev) {
5876                         /* Nothing to check */;
5877                 } else if (rdev->data_offset < rdev->sb_start) {
5878                         if (mddev->dev_sectors &&
5879                             rdev->data_offset + mddev->dev_sectors
5880                             > rdev->sb_start) {
5881                                 pr_warn("md: %s: data overlaps metadata\n",
5882                                         mdname(mddev));
5883                                 return -EINVAL;
5884                         }
5885                 } else {
5886                         if (rdev->sb_start + rdev->sb_size/512
5887                             > rdev->data_offset) {
5888                                 pr_warn("md: %s: metadata overlaps data\n",
5889                                         mdname(mddev));
5890                                 return -EINVAL;
5891                         }
5892                 }
5893                 sysfs_notify_dirent_safe(rdev->sysfs_state);
5894         }
5895
5896         if (!bioset_initialized(&mddev->bio_set)) {
5897                 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5898                 if (err)
5899                         return err;
5900         }
5901         if (!bioset_initialized(&mddev->sync_set)) {
5902                 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5903                 if (err)
5904                         return err;
5905         }
5906
5907         spin_lock(&pers_lock);
5908         pers = find_pers(mddev->level, mddev->clevel);
5909         if (!pers || !try_module_get(pers->owner)) {
5910                 spin_unlock(&pers_lock);
5911                 if (mddev->level != LEVEL_NONE)
5912                         pr_warn("md: personality for level %d is not loaded!\n",
5913                                 mddev->level);
5914                 else
5915                         pr_warn("md: personality for level %s is not loaded!\n",
5916                                 mddev->clevel);
5917                 err = -EINVAL;
5918                 goto abort;
5919         }
5920         spin_unlock(&pers_lock);
5921         if (mddev->level != pers->level) {
5922                 mddev->level = pers->level;
5923                 mddev->new_level = pers->level;
5924         }
5925         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5926
5927         if (mddev->reshape_position != MaxSector &&
5928             pers->start_reshape == NULL) {
5929                 /* This personality cannot handle reshaping... */
5930                 module_put(pers->owner);
5931                 err = -EINVAL;
5932                 goto abort;
5933         }
5934
5935         if (pers->sync_request) {
5936                 /* Warn if this is a potentially silly
5937                  * configuration.
5938                  */
5939                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5940                 struct md_rdev *rdev2;
5941                 int warned = 0;
5942
5943                 rdev_for_each(rdev, mddev)
5944                         rdev_for_each(rdev2, mddev) {
5945                                 if (rdev < rdev2 &&
5946                                     rdev->bdev->bd_disk ==
5947                                     rdev2->bdev->bd_disk) {
5948                                         pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5949                                                 mdname(mddev),
5950                                                 bdevname(rdev->bdev,b),
5951                                                 bdevname(rdev2->bdev,b2));
5952                                         warned = 1;
5953                                 }
5954                         }
5955
5956                 if (warned)
5957                         pr_warn("True protection against single-disk failure might be compromised.\n");
5958         }
5959
5960         mddev->recovery = 0;
5961         /* may be over-ridden by personality */
5962         mddev->resync_max_sectors = mddev->dev_sectors;
5963
5964         mddev->ok_start_degraded = start_dirty_degraded;
5965
5966         if (start_readonly && mddev->ro == 0)
5967                 mddev->ro = 2; /* read-only, but switch on first write */
5968
5969         err = pers->run(mddev);
5970         if (err)
5971                 pr_warn("md: pers->run() failed ...\n");
5972         else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5973                 WARN_ONCE(!mddev->external_size,
5974                           "%s: default size too small, but 'external_size' not in effect?\n",
5975                           __func__);
5976                 pr_warn("md: invalid array_size %llu > default size %llu\n",
5977                         (unsigned long long)mddev->array_sectors / 2,
5978                         (unsigned long long)pers->size(mddev, 0, 0) / 2);
5979                 err = -EINVAL;
5980         }
5981         if (err == 0 && pers->sync_request &&
5982             (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5983                 struct bitmap *bitmap;
5984
5985                 bitmap = md_bitmap_create(mddev, -1);
5986                 if (IS_ERR(bitmap)) {
5987                         err = PTR_ERR(bitmap);
5988                         pr_warn("%s: failed to create bitmap (%d)\n",
5989                                 mdname(mddev), err);
5990                 } else
5991                         mddev->bitmap = bitmap;
5992
5993         }
5994         if (err)
5995                 goto bitmap_abort;
5996
5997         if (mddev->bitmap_info.max_write_behind > 0) {
5998                 bool create_pool = false;
5999
6000                 rdev_for_each(rdev, mddev) {
6001                         if (test_bit(WriteMostly, &rdev->flags) &&
6002                             rdev_init_serial(rdev))
6003                                 create_pool = true;
6004                 }
6005                 if (create_pool && mddev->serial_info_pool == NULL) {
6006                         mddev->serial_info_pool =
6007                                 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6008                                                     sizeof(struct serial_info));
6009                         if (!mddev->serial_info_pool) {
6010                                 err = -ENOMEM;
6011                                 goto bitmap_abort;
6012                         }
6013                 }
6014         }
6015
6016         if (mddev->queue) {
6017                 bool nonrot = true;
6018
6019                 rdev_for_each(rdev, mddev) {
6020                         if (rdev->raid_disk >= 0 &&
6021                             !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6022                                 nonrot = false;
6023                                 break;
6024                         }
6025                 }
6026                 if (mddev->degraded)
6027                         nonrot = false;
6028                 if (nonrot)
6029                         blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6030                 else
6031                         blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6032         }
6033         if (pers->sync_request) {
6034                 if (mddev->kobj.sd &&
6035                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6036                         pr_warn("md: cannot register extra attributes for %s\n",
6037                                 mdname(mddev));
6038                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6039                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6040                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6041         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6042                 mddev->ro = 0;
6043
6044         atomic_set(&mddev->max_corr_read_errors,
6045                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6046         mddev->safemode = 0;
6047         if (mddev_is_clustered(mddev))
6048                 mddev->safemode_delay = 0;
6049         else
6050                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6051         mddev->in_sync = 1;
6052         smp_wmb();
6053         spin_lock(&mddev->lock);
6054         mddev->pers = pers;
6055         spin_unlock(&mddev->lock);
6056         rdev_for_each(rdev, mddev)
6057                 if (rdev->raid_disk >= 0)
6058                         sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6059
6060         if (mddev->degraded && !mddev->ro)
6061                 /* This ensures that recovering status is reported immediately
6062                  * via sysfs - until a lack of spares is confirmed.
6063                  */
6064                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6065         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6066
6067         if (mddev->sb_flags)
6068                 md_update_sb(mddev, 0);
6069
6070         md_new_event(mddev);
6071         return 0;
6072
6073 bitmap_abort:
6074         mddev_detach(mddev);
6075         if (mddev->private)
6076                 pers->free(mddev, mddev->private);
6077         mddev->private = NULL;
6078         module_put(pers->owner);
6079         md_bitmap_destroy(mddev);
6080 abort:
6081         bioset_exit(&mddev->bio_set);
6082         bioset_exit(&mddev->sync_set);
6083         return err;
6084 }
6085 EXPORT_SYMBOL_GPL(md_run);
6086
6087 int do_md_run(struct mddev *mddev)
6088 {
6089         int err;
6090
6091         set_bit(MD_NOT_READY, &mddev->flags);
6092         err = md_run(mddev);
6093         if (err)
6094                 goto out;
6095         err = md_bitmap_load(mddev);
6096         if (err) {
6097                 md_bitmap_destroy(mddev);
6098                 goto out;
6099         }
6100
6101         if (mddev_is_clustered(mddev))
6102                 md_allow_write(mddev);
6103
6104         /* run start up tasks that require md_thread */
6105         md_start(mddev);
6106
6107         md_wakeup_thread(mddev->thread);
6108         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6109
6110         set_capacity(mddev->gendisk, mddev->array_sectors);
6111         revalidate_disk_size(mddev->gendisk, true);
6112         clear_bit(MD_NOT_READY, &mddev->flags);
6113         mddev->changed = 1;
6114         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6115         sysfs_notify_dirent_safe(mddev->sysfs_state);
6116         sysfs_notify_dirent_safe(mddev->sysfs_action);
6117         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6118 out:
6119         clear_bit(MD_NOT_READY, &mddev->flags);
6120         return err;
6121 }
6122
6123 int md_start(struct mddev *mddev)
6124 {
6125         int ret = 0;
6126
6127         if (mddev->pers->start) {
6128                 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6129                 md_wakeup_thread(mddev->thread);
6130                 ret = mddev->pers->start(mddev);
6131                 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6132                 md_wakeup_thread(mddev->sync_thread);
6133         }
6134         return ret;
6135 }
6136 EXPORT_SYMBOL_GPL(md_start);
6137
6138 static int restart_array(struct mddev *mddev)
6139 {
6140         struct gendisk *disk = mddev->gendisk;
6141         struct md_rdev *rdev;
6142         bool has_journal = false;
6143         bool has_readonly = false;
6144
6145         /* Complain if it has no devices */
6146         if (list_empty(&mddev->disks))
6147                 return -ENXIO;
6148         if (!mddev->pers)
6149                 return -EINVAL;
6150         if (!mddev->ro)
6151                 return -EBUSY;
6152
6153         rcu_read_lock();
6154         rdev_for_each_rcu(rdev, mddev) {
6155                 if (test_bit(Journal, &rdev->flags) &&
6156                     !test_bit(Faulty, &rdev->flags))
6157                         has_journal = true;
6158                 if (bdev_read_only(rdev->bdev))
6159                         has_readonly = true;
6160         }
6161         rcu_read_unlock();
6162         if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6163                 /* Don't restart rw with journal missing/faulty */
6164                         return -EINVAL;
6165         if (has_readonly)
6166                 return -EROFS;
6167
6168         mddev->safemode = 0;
6169         mddev->ro = 0;
6170         set_disk_ro(disk, 0);
6171         pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6172         /* Kick recovery or resync if necessary */
6173         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6174         md_wakeup_thread(mddev->thread);
6175         md_wakeup_thread(mddev->sync_thread);
6176         sysfs_notify_dirent_safe(mddev->sysfs_state);
6177         return 0;
6178 }
6179
6180 static void md_clean(struct mddev *mddev)
6181 {
6182         mddev->array_sectors = 0;
6183         mddev->external_size = 0;
6184         mddev->dev_sectors = 0;
6185         mddev->raid_disks = 0;
6186         mddev->recovery_cp = 0;
6187         mddev->resync_min = 0;
6188         mddev->resync_max = MaxSector;
6189         mddev->reshape_position = MaxSector;
6190         mddev->external = 0;
6191         mddev->persistent = 0;
6192         mddev->level = LEVEL_NONE;
6193         mddev->clevel[0] = 0;
6194         mddev->flags = 0;
6195         mddev->sb_flags = 0;
6196         mddev->ro = 0;
6197         mddev->metadata_type[0] = 0;
6198         mddev->chunk_sectors = 0;
6199         mddev->ctime = mddev->utime = 0;
6200         mddev->layout = 0;
6201         mddev->max_disks = 0;
6202         mddev->events = 0;
6203         mddev->can_decrease_events = 0;
6204         mddev->delta_disks = 0;
6205         mddev->reshape_backwards = 0;
6206         mddev->new_level = LEVEL_NONE;
6207         mddev->new_layout = 0;
6208         mddev->new_chunk_sectors = 0;
6209         mddev->curr_resync = 0;
6210         atomic64_set(&mddev->resync_mismatches, 0);
6211         mddev->suspend_lo = mddev->suspend_hi = 0;
6212         mddev->sync_speed_min = mddev->sync_speed_max = 0;
6213         mddev->recovery = 0;
6214         mddev->in_sync = 0;
6215         mddev->changed = 0;
6216         mddev->degraded = 0;
6217         mddev->safemode = 0;
6218         mddev->private = NULL;
6219         mddev->cluster_info = NULL;
6220         mddev->bitmap_info.offset = 0;
6221         mddev->bitmap_info.default_offset = 0;
6222         mddev->bitmap_info.default_space = 0;
6223         mddev->bitmap_info.chunksize = 0;
6224         mddev->bitmap_info.daemon_sleep = 0;
6225         mddev->bitmap_info.max_write_behind = 0;
6226         mddev->bitmap_info.nodes = 0;
6227 }
6228
6229 static void __md_stop_writes(struct mddev *mddev)
6230 {
6231         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6232         if (work_pending(&mddev->del_work))
6233                 flush_workqueue(md_misc_wq);
6234         if (mddev->sync_thread) {
6235                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6236                 md_reap_sync_thread(mddev);
6237         }
6238
6239         del_timer_sync(&mddev->safemode_timer);
6240
6241         if (mddev->pers && mddev->pers->quiesce) {
6242                 mddev->pers->quiesce(mddev, 1);
6243                 mddev->pers->quiesce(mddev, 0);
6244         }
6245         md_bitmap_flush(mddev);
6246
6247         if (mddev->ro == 0 &&
6248             ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6249              mddev->sb_flags)) {
6250                 /* mark array as shutdown cleanly */
6251                 if (!mddev_is_clustered(mddev))
6252                         mddev->in_sync = 1;
6253                 md_update_sb(mddev, 1);
6254         }
6255         /* disable policy to guarantee rdevs free resources for serialization */
6256         mddev->serialize_policy = 0;
6257         mddev_destroy_serial_pool(mddev, NULL, true);
6258 }
6259
6260 void md_stop_writes(struct mddev *mddev)
6261 {
6262         mddev_lock_nointr(mddev);
6263         __md_stop_writes(mddev);
6264         mddev_unlock(mddev);
6265 }
6266 EXPORT_SYMBOL_GPL(md_stop_writes);
6267
6268 static void mddev_detach(struct mddev *mddev)
6269 {
6270         md_bitmap_wait_behind_writes(mddev);
6271         if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6272                 mddev->pers->quiesce(mddev, 1);
6273                 mddev->pers->quiesce(mddev, 0);
6274         }
6275         md_unregister_thread(&mddev->thread);
6276         if (mddev->queue)
6277                 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6278 }
6279
6280 static void __md_stop(struct mddev *mddev)
6281 {
6282         struct md_personality *pers = mddev->pers;
6283         md_bitmap_destroy(mddev);
6284         mddev_detach(mddev);
6285         /* Ensure ->event_work is done */
6286         if (mddev->event_work.func)
6287                 flush_workqueue(md_misc_wq);
6288         spin_lock(&mddev->lock);
6289         mddev->pers = NULL;
6290         spin_unlock(&mddev->lock);
6291         pers->free(mddev, mddev->private);
6292         mddev->private = NULL;
6293         if (pers->sync_request && mddev->to_remove == NULL)
6294                 mddev->to_remove = &md_redundancy_group;
6295         module_put(pers->owner);
6296         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6297 }
6298
6299 void md_stop(struct mddev *mddev)
6300 {
6301         /* stop the array and free an attached data structures.
6302          * This is called from dm-raid
6303          */
6304         __md_stop(mddev);
6305         bioset_exit(&mddev->bio_set);
6306         bioset_exit(&mddev->sync_set);
6307 }
6308
6309 EXPORT_SYMBOL_GPL(md_stop);
6310
6311 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6312 {
6313         int err = 0;
6314         int did_freeze = 0;
6315
6316         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6317                 did_freeze = 1;
6318                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6319                 md_wakeup_thread(mddev->thread);
6320         }
6321         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6322                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6323         if (mddev->sync_thread)
6324                 /* Thread might be blocked waiting for metadata update
6325                  * which will now never happen */
6326                 wake_up_process(mddev->sync_thread->tsk);
6327
6328         if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6329                 return -EBUSY;
6330         mddev_unlock(mddev);
6331         wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6332                                           &mddev->recovery));
6333         wait_event(mddev->sb_wait,
6334                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6335         mddev_lock_nointr(mddev);
6336
6337         mutex_lock(&mddev->open_mutex);
6338         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6339             mddev->sync_thread ||
6340             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6341                 pr_warn("md: %s still in use.\n",mdname(mddev));
6342                 if (did_freeze) {
6343                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6344                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6345                         md_wakeup_thread(mddev->thread);
6346                 }
6347                 err = -EBUSY;
6348                 goto out;
6349         }
6350         if (mddev->pers) {
6351                 __md_stop_writes(mddev);
6352
6353                 err  = -ENXIO;
6354                 if (mddev->ro==1)
6355                         goto out;
6356                 mddev->ro = 1;
6357                 set_disk_ro(mddev->gendisk, 1);
6358                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6359                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6360                 md_wakeup_thread(mddev->thread);
6361                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6362                 err = 0;
6363         }
6364 out:
6365         mutex_unlock(&mddev->open_mutex);
6366         return err;
6367 }
6368
6369 /* mode:
6370  *   0 - completely stop and dis-assemble array
6371  *   2 - stop but do not disassemble array
6372  */
6373 static int do_md_stop(struct mddev *mddev, int mode,
6374                       struct block_device *bdev)
6375 {
6376         struct gendisk *disk = mddev->gendisk;
6377         struct md_rdev *rdev;
6378         int did_freeze = 0;
6379
6380         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6381                 did_freeze = 1;
6382                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6383                 md_wakeup_thread(mddev->thread);
6384         }
6385         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6386                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6387         if (mddev->sync_thread)
6388                 /* Thread might be blocked waiting for metadata update
6389                  * which will now never happen */
6390                 wake_up_process(mddev->sync_thread->tsk);
6391
6392         mddev_unlock(mddev);
6393         wait_event(resync_wait, (mddev->sync_thread == NULL &&
6394                                  !test_bit(MD_RECOVERY_RUNNING,
6395                                            &mddev->recovery)));
6396         mddev_lock_nointr(mddev);
6397
6398         mutex_lock(&mddev->open_mutex);
6399         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6400             mddev->sysfs_active ||
6401             mddev->sync_thread ||
6402             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6403                 pr_warn("md: %s still in use.\n",mdname(mddev));
6404                 mutex_unlock(&mddev->open_mutex);
6405                 if (did_freeze) {
6406                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6407                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6408                         md_wakeup_thread(mddev->thread);
6409                 }
6410                 return -EBUSY;
6411         }
6412         if (mddev->pers) {
6413                 if (mddev->ro)
6414                         set_disk_ro(disk, 0);
6415
6416                 __md_stop_writes(mddev);
6417                 __md_stop(mddev);
6418
6419                 /* tell userspace to handle 'inactive' */
6420                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6421
6422                 rdev_for_each(rdev, mddev)
6423                         if (rdev->raid_disk >= 0)
6424                                 sysfs_unlink_rdev(mddev, rdev);
6425
6426                 set_capacity(disk, 0);
6427                 mutex_unlock(&mddev->open_mutex);
6428                 mddev->changed = 1;
6429                 revalidate_disk_size(disk, true);
6430
6431                 if (mddev->ro)
6432                         mddev->ro = 0;
6433         } else
6434                 mutex_unlock(&mddev->open_mutex);
6435         /*
6436          * Free resources if final stop
6437          */
6438         if (mode == 0) {
6439                 pr_info("md: %s stopped.\n", mdname(mddev));
6440
6441                 if (mddev->bitmap_info.file) {
6442                         struct file *f = mddev->bitmap_info.file;
6443                         spin_lock(&mddev->lock);
6444                         mddev->bitmap_info.file = NULL;
6445                         spin_unlock(&mddev->lock);
6446                         fput(f);
6447                 }
6448                 mddev->bitmap_info.offset = 0;
6449
6450                 export_array(mddev);
6451
6452                 md_clean(mddev);
6453                 if (mddev->hold_active == UNTIL_STOP)
6454                         mddev->hold_active = 0;
6455         }
6456         md_new_event(mddev);
6457         sysfs_notify_dirent_safe(mddev->sysfs_state);
6458         return 0;
6459 }
6460
6461 #ifndef MODULE
6462 static void autorun_array(struct mddev *mddev)
6463 {
6464         struct md_rdev *rdev;
6465         int err;
6466
6467         if (list_empty(&mddev->disks))
6468                 return;
6469
6470         pr_info("md: running: ");
6471
6472         rdev_for_each(rdev, mddev) {
6473                 char b[BDEVNAME_SIZE];
6474                 pr_cont("<%s>", bdevname(rdev->bdev,b));
6475         }
6476         pr_cont("\n");
6477
6478         err = do_md_run(mddev);
6479         if (err) {
6480                 pr_warn("md: do_md_run() returned %d\n", err);
6481                 do_md_stop(mddev, 0, NULL);
6482         }
6483 }
6484
6485 /*
6486  * lets try to run arrays based on all disks that have arrived
6487  * until now. (those are in pending_raid_disks)
6488  *
6489  * the method: pick the first pending disk, collect all disks with
6490  * the same UUID, remove all from the pending list and put them into
6491  * the 'same_array' list. Then order this list based on superblock
6492  * update time (freshest comes first), kick out 'old' disks and
6493  * compare superblocks. If everything's fine then run it.
6494  *
6495  * If "unit" is allocated, then bump its reference count
6496  */
6497 static void autorun_devices(int part)
6498 {
6499         struct md_rdev *rdev0, *rdev, *tmp;
6500         struct mddev *mddev;
6501         char b[BDEVNAME_SIZE];
6502
6503         pr_info("md: autorun ...\n");
6504         while (!list_empty(&pending_raid_disks)) {
6505                 int unit;
6506                 dev_t dev;
6507                 LIST_HEAD(candidates);
6508                 rdev0 = list_entry(pending_raid_disks.next,
6509                                          struct md_rdev, same_set);
6510
6511                 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6512                 INIT_LIST_HEAD(&candidates);
6513                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6514                         if (super_90_load(rdev, rdev0, 0) >= 0) {
6515                                 pr_debug("md:  adding %s ...\n",
6516                                          bdevname(rdev->bdev,b));
6517                                 list_move(&rdev->same_set, &candidates);
6518                         }
6519                 /*
6520                  * now we have a set of devices, with all of them having
6521                  * mostly sane superblocks. It's time to allocate the
6522                  * mddev.
6523                  */
6524                 if (part) {
6525                         dev = MKDEV(mdp_major,
6526                                     rdev0->preferred_minor << MdpMinorShift);
6527                         unit = MINOR(dev) >> MdpMinorShift;
6528                 } else {
6529                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6530                         unit = MINOR(dev);
6531                 }
6532                 if (rdev0->preferred_minor != unit) {
6533                         pr_warn("md: unit number in %s is bad: %d\n",
6534                                 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6535                         break;
6536                 }
6537
6538                 md_probe(dev, NULL, NULL);
6539                 mddev = mddev_find(dev);
6540                 if (!mddev || !mddev->gendisk) {
6541                         if (mddev)
6542                                 mddev_put(mddev);
6543                         break;
6544                 }
6545                 if (mddev_lock(mddev))
6546                         pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6547                 else if (mddev->raid_disks || mddev->major_version
6548                          || !list_empty(&mddev->disks)) {
6549                         pr_warn("md: %s already running, cannot run %s\n",
6550                                 mdname(mddev), bdevname(rdev0->bdev,b));
6551                         mddev_unlock(mddev);
6552                 } else {
6553                         pr_debug("md: created %s\n", mdname(mddev));
6554                         mddev->persistent = 1;
6555                         rdev_for_each_list(rdev, tmp, &candidates) {
6556                                 list_del_init(&rdev->same_set);
6557                                 if (bind_rdev_to_array(rdev, mddev))
6558                                         export_rdev(rdev);
6559                         }
6560                         autorun_array(mddev);
6561                         mddev_unlock(mddev);
6562                 }
6563                 /* on success, candidates will be empty, on error
6564                  * it won't...
6565                  */
6566                 rdev_for_each_list(rdev, tmp, &candidates) {
6567                         list_del_init(&rdev->same_set);
6568                         export_rdev(rdev);
6569                 }
6570                 mddev_put(mddev);
6571         }
6572         pr_info("md: ... autorun DONE.\n");
6573 }
6574 #endif /* !MODULE */
6575
6576 static int get_version(void __user *arg)
6577 {
6578         mdu_version_t ver;
6579
6580         ver.major = MD_MAJOR_VERSION;
6581         ver.minor = MD_MINOR_VERSION;
6582         ver.patchlevel = MD_PATCHLEVEL_VERSION;
6583
6584         if (copy_to_user(arg, &ver, sizeof(ver)))
6585                 return -EFAULT;
6586
6587         return 0;
6588 }
6589
6590 static int get_array_info(struct mddev *mddev, void __user *arg)
6591 {
6592         mdu_array_info_t info;
6593         int nr,working,insync,failed,spare;
6594         struct md_rdev *rdev;
6595
6596         nr = working = insync = failed = spare = 0;
6597         rcu_read_lock();
6598         rdev_for_each_rcu(rdev, mddev) {
6599                 nr++;
6600                 if (test_bit(Faulty, &rdev->flags))
6601                         failed++;
6602                 else {
6603                         working++;
6604                         if (test_bit(In_sync, &rdev->flags))
6605                                 insync++;
6606                         else if (test_bit(Journal, &rdev->flags))
6607                                 /* TODO: add journal count to md_u.h */
6608                                 ;
6609                         else
6610                                 spare++;
6611                 }
6612         }
6613         rcu_read_unlock();
6614
6615         info.major_version = mddev->major_version;
6616         info.minor_version = mddev->minor_version;
6617         info.patch_version = MD_PATCHLEVEL_VERSION;
6618         info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6619         info.level         = mddev->level;
6620         info.size          = mddev->dev_sectors / 2;
6621         if (info.size != mddev->dev_sectors / 2) /* overflow */
6622                 info.size = -1;
6623         info.nr_disks      = nr;
6624         info.raid_disks    = mddev->raid_disks;
6625         info.md_minor      = mddev->md_minor;
6626         info.not_persistent= !mddev->persistent;
6627
6628         info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6629         info.state         = 0;
6630         if (mddev->in_sync)
6631                 info.state = (1<<MD_SB_CLEAN);
6632         if (mddev->bitmap && mddev->bitmap_info.offset)
6633                 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6634         if (mddev_is_clustered(mddev))
6635                 info.state |= (1<<MD_SB_CLUSTERED);
6636         info.active_disks  = insync;
6637         info.working_disks = working;
6638         info.failed_disks  = failed;
6639         info.spare_disks   = spare;
6640
6641         info.layout        = mddev->layout;
6642         info.chunk_size    = mddev->chunk_sectors << 9;
6643
6644         if (copy_to_user(arg, &info, sizeof(info)))
6645                 return -EFAULT;
6646
6647         return 0;
6648 }
6649
6650 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6651 {
6652         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6653         char *ptr;
6654         int err;
6655
6656         file = kzalloc(sizeof(*file), GFP_NOIO);
6657         if (!file)
6658                 return -ENOMEM;
6659
6660         err = 0;
6661         spin_lock(&mddev->lock);
6662         /* bitmap enabled */
6663         if (mddev->bitmap_info.file) {
6664                 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6665                                 sizeof(file->pathname));
6666                 if (IS_ERR(ptr))
6667                         err = PTR_ERR(ptr);
6668                 else
6669                         memmove(file->pathname, ptr,
6670                                 sizeof(file->pathname)-(ptr-file->pathname));
6671         }
6672         spin_unlock(&mddev->lock);
6673
6674         if (err == 0 &&
6675             copy_to_user(arg, file, sizeof(*file)))
6676                 err = -EFAULT;
6677
6678         kfree(file);
6679         return err;
6680 }
6681
6682 static int get_disk_info(struct mddev *mddev, void __user * arg)
6683 {
6684         mdu_disk_info_t info;
6685         struct md_rdev *rdev;
6686
6687         if (copy_from_user(&info, arg, sizeof(info)))
6688                 return -EFAULT;
6689
6690         rcu_read_lock();
6691         rdev = md_find_rdev_nr_rcu(mddev, info.number);
6692         if (rdev) {
6693                 info.major = MAJOR(rdev->bdev->bd_dev);
6694                 info.minor = MINOR(rdev->bdev->bd_dev);
6695                 info.raid_disk = rdev->raid_disk;
6696                 info.state = 0;
6697                 if (test_bit(Faulty, &rdev->flags))
6698                         info.state |= (1<<MD_DISK_FAULTY);
6699                 else if (test_bit(In_sync, &rdev->flags)) {
6700                         info.state |= (1<<MD_DISK_ACTIVE);
6701                         info.state |= (1<<MD_DISK_SYNC);
6702                 }
6703                 if (test_bit(Journal, &rdev->flags))
6704                         info.state |= (1<<MD_DISK_JOURNAL);
6705                 if (test_bit(WriteMostly, &rdev->flags))
6706                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
6707                 if (test_bit(FailFast, &rdev->flags))
6708                         info.state |= (1<<MD_DISK_FAILFAST);
6709         } else {
6710                 info.major = info.minor = 0;
6711                 info.raid_disk = -1;
6712                 info.state = (1<<MD_DISK_REMOVED);
6713         }
6714         rcu_read_unlock();
6715
6716         if (copy_to_user(arg, &info, sizeof(info)))
6717                 return -EFAULT;
6718
6719         return 0;
6720 }
6721
6722 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6723 {
6724         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6725         struct md_rdev *rdev;
6726         dev_t dev = MKDEV(info->major,info->minor);
6727
6728         if (mddev_is_clustered(mddev) &&
6729                 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6730                 pr_warn("%s: Cannot add to clustered mddev.\n",
6731                         mdname(mddev));
6732                 return -EINVAL;
6733         }
6734
6735         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6736                 return -EOVERFLOW;
6737
6738         if (!mddev->raid_disks) {
6739                 int err;
6740                 /* expecting a device which has a superblock */
6741                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6742                 if (IS_ERR(rdev)) {
6743                         pr_warn("md: md_import_device returned %ld\n",
6744                                 PTR_ERR(rdev));
6745                         return PTR_ERR(rdev);
6746                 }
6747                 if (!list_empty(&mddev->disks)) {
6748                         struct md_rdev *rdev0
6749                                 = list_entry(mddev->disks.next,
6750                                              struct md_rdev, same_set);
6751                         err = super_types[mddev->major_version]
6752                                 .load_super(rdev, rdev0, mddev->minor_version);
6753                         if (err < 0) {
6754                                 pr_warn("md: %s has different UUID to %s\n",
6755                                         bdevname(rdev->bdev,b),
6756                                         bdevname(rdev0->bdev,b2));
6757                                 export_rdev(rdev);
6758                                 return -EINVAL;
6759                         }
6760                 }
6761                 err = bind_rdev_to_array(rdev, mddev);
6762                 if (err)
6763                         export_rdev(rdev);
6764                 return err;
6765         }
6766
6767         /*
6768          * md_add_new_disk can be used once the array is assembled
6769          * to add "hot spares".  They must already have a superblock
6770          * written
6771          */
6772         if (mddev->pers) {
6773                 int err;
6774                 if (!mddev->pers->hot_add_disk) {
6775                         pr_warn("%s: personality does not support diskops!\n",
6776                                 mdname(mddev));
6777                         return -EINVAL;
6778                 }
6779                 if (mddev->persistent)
6780                         rdev = md_import_device(dev, mddev->major_version,
6781                                                 mddev->minor_version);
6782                 else
6783                         rdev = md_import_device(dev, -1, -1);
6784                 if (IS_ERR(rdev)) {
6785                         pr_warn("md: md_import_device returned %ld\n",
6786                                 PTR_ERR(rdev));
6787                         return PTR_ERR(rdev);
6788                 }
6789                 /* set saved_raid_disk if appropriate */
6790                 if (!mddev->persistent) {
6791                         if (info->state & (1<<MD_DISK_SYNC)  &&
6792                             info->raid_disk < mddev->raid_disks) {
6793                                 rdev->raid_disk = info->raid_disk;
6794                                 set_bit(In_sync, &rdev->flags);
6795                                 clear_bit(Bitmap_sync, &rdev->flags);
6796                         } else
6797                                 rdev->raid_disk = -1;
6798                         rdev->saved_raid_disk = rdev->raid_disk;
6799                 } else
6800                         super_types[mddev->major_version].
6801                                 validate_super(mddev, rdev);
6802                 if ((info->state & (1<<MD_DISK_SYNC)) &&
6803                      rdev->raid_disk != info->raid_disk) {
6804                         /* This was a hot-add request, but events doesn't
6805                          * match, so reject it.
6806                          */
6807                         export_rdev(rdev);
6808                         return -EINVAL;
6809                 }
6810
6811                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6812                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6813                         set_bit(WriteMostly, &rdev->flags);
6814                 else
6815                         clear_bit(WriteMostly, &rdev->flags);
6816                 if (info->state & (1<<MD_DISK_FAILFAST))
6817                         set_bit(FailFast, &rdev->flags);
6818                 else
6819                         clear_bit(FailFast, &rdev->flags);
6820
6821                 if (info->state & (1<<MD_DISK_JOURNAL)) {
6822                         struct md_rdev *rdev2;
6823                         bool has_journal = false;
6824
6825                         /* make sure no existing journal disk */
6826                         rdev_for_each(rdev2, mddev) {
6827                                 if (test_bit(Journal, &rdev2->flags)) {
6828                                         has_journal = true;
6829                                         break;
6830                                 }
6831                         }
6832                         if (has_journal || mddev->bitmap) {
6833                                 export_rdev(rdev);
6834                                 return -EBUSY;
6835                         }
6836                         set_bit(Journal, &rdev->flags);
6837                 }
6838                 /*
6839                  * check whether the device shows up in other nodes
6840                  */
6841                 if (mddev_is_clustered(mddev)) {
6842                         if (info->state & (1 << MD_DISK_CANDIDATE))
6843                                 set_bit(Candidate, &rdev->flags);
6844                         else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6845                                 /* --add initiated by this node */
6846                                 err = md_cluster_ops->add_new_disk(mddev, rdev);
6847                                 if (err) {
6848                                         export_rdev(rdev);
6849                                         return err;
6850                                 }
6851                         }
6852                 }
6853
6854                 rdev->raid_disk = -1;
6855                 err = bind_rdev_to_array(rdev, mddev);
6856
6857                 if (err)
6858                         export_rdev(rdev);
6859
6860                 if (mddev_is_clustered(mddev)) {
6861                         if (info->state & (1 << MD_DISK_CANDIDATE)) {
6862                                 if (!err) {
6863                                         err = md_cluster_ops->new_disk_ack(mddev,
6864                                                 err == 0);
6865                                         if (err)
6866                                                 md_kick_rdev_from_array(rdev);
6867                                 }
6868                         } else {
6869                                 if (err)
6870                                         md_cluster_ops->add_new_disk_cancel(mddev);
6871                                 else
6872                                         err = add_bound_rdev(rdev);
6873                         }
6874
6875                 } else if (!err)
6876                         err = add_bound_rdev(rdev);
6877
6878                 return err;
6879         }
6880
6881         /* otherwise, md_add_new_disk is only allowed
6882          * for major_version==0 superblocks
6883          */
6884         if (mddev->major_version != 0) {
6885                 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6886                 return -EINVAL;
6887         }
6888
6889         if (!(info->state & (1<<MD_DISK_FAULTY))) {
6890                 int err;
6891                 rdev = md_import_device(dev, -1, 0);
6892                 if (IS_ERR(rdev)) {
6893                         pr_warn("md: error, md_import_device() returned %ld\n",
6894                                 PTR_ERR(rdev));
6895                         return PTR_ERR(rdev);
6896                 }
6897                 rdev->desc_nr = info->number;
6898                 if (info->raid_disk < mddev->raid_disks)
6899                         rdev->raid_disk = info->raid_disk;
6900                 else
6901                         rdev->raid_disk = -1;
6902
6903                 if (rdev->raid_disk < mddev->raid_disks)
6904                         if (info->state & (1<<MD_DISK_SYNC))
6905                                 set_bit(In_sync, &rdev->flags);
6906
6907                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6908                         set_bit(WriteMostly, &rdev->flags);
6909                 if (info->state & (1<<MD_DISK_FAILFAST))
6910                         set_bit(FailFast, &rdev->flags);
6911
6912                 if (!mddev->persistent) {
6913                         pr_debug("md: nonpersistent superblock ...\n");
6914                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6915                 } else
6916                         rdev->sb_start = calc_dev_sboffset(rdev);
6917                 rdev->sectors = rdev->sb_start;
6918
6919                 err = bind_rdev_to_array(rdev, mddev);
6920                 if (err) {
6921                         export_rdev(rdev);
6922                         return err;
6923                 }
6924         }
6925
6926         return 0;
6927 }
6928
6929 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6930 {
6931         char b[BDEVNAME_SIZE];
6932         struct md_rdev *rdev;
6933
6934         if (!mddev->pers)
6935                 return -ENODEV;
6936
6937         rdev = find_rdev(mddev, dev);
6938         if (!rdev)
6939                 return -ENXIO;
6940
6941         if (rdev->raid_disk < 0)
6942                 goto kick_rdev;
6943
6944         clear_bit(Blocked, &rdev->flags);
6945         remove_and_add_spares(mddev, rdev);
6946
6947         if (rdev->raid_disk >= 0)
6948                 goto busy;
6949
6950 kick_rdev:
6951         if (mddev_is_clustered(mddev))
6952                 md_cluster_ops->remove_disk(mddev, rdev);
6953
6954         md_kick_rdev_from_array(rdev);
6955         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6956         if (mddev->thread)
6957                 md_wakeup_thread(mddev->thread);
6958         else
6959                 md_update_sb(mddev, 1);
6960         md_new_event(mddev);
6961
6962         return 0;
6963 busy:
6964         pr_debug("md: cannot remove active disk %s from %s ...\n",
6965                  bdevname(rdev->bdev,b), mdname(mddev));
6966         return -EBUSY;
6967 }
6968
6969 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6970 {
6971         char b[BDEVNAME_SIZE];
6972         int err;
6973         struct md_rdev *rdev;
6974
6975         if (!mddev->pers)
6976                 return -ENODEV;
6977
6978         if (mddev->major_version != 0) {
6979                 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6980                         mdname(mddev));
6981                 return -EINVAL;
6982         }
6983         if (!mddev->pers->hot_add_disk) {
6984                 pr_warn("%s: personality does not support diskops!\n",
6985                         mdname(mddev));
6986                 return -EINVAL;
6987         }
6988
6989         rdev = md_import_device(dev, -1, 0);
6990         if (IS_ERR(rdev)) {
6991                 pr_warn("md: error, md_import_device() returned %ld\n",
6992                         PTR_ERR(rdev));
6993                 return -EINVAL;
6994         }
6995
6996         if (mddev->persistent)
6997                 rdev->sb_start = calc_dev_sboffset(rdev);
6998         else
6999                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
7000
7001         rdev->sectors = rdev->sb_start;
7002
7003         if (test_bit(Faulty, &rdev->flags)) {
7004                 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
7005                         bdevname(rdev->bdev,b), mdname(mddev));
7006                 err = -EINVAL;
7007                 goto abort_export;
7008         }
7009
7010         clear_bit(In_sync, &rdev->flags);
7011         rdev->desc_nr = -1;
7012         rdev->saved_raid_disk = -1;
7013         err = bind_rdev_to_array(rdev, mddev);
7014         if (err)
7015                 goto abort_export;
7016
7017         /*
7018          * The rest should better be atomic, we can have disk failures
7019          * noticed in interrupt contexts ...
7020          */
7021
7022         rdev->raid_disk = -1;
7023
7024         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7025         if (!mddev->thread)
7026                 md_update_sb(mddev, 1);
7027         /*
7028          * Kick recovery, maybe this spare has to be added to the
7029          * array immediately.
7030          */
7031         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7032         md_wakeup_thread(mddev->thread);
7033         md_new_event(mddev);
7034         return 0;
7035
7036 abort_export:
7037         export_rdev(rdev);
7038         return err;
7039 }
7040
7041 static int set_bitmap_file(struct mddev *mddev, int fd)
7042 {
7043         int err = 0;
7044
7045         if (mddev->pers) {
7046                 if (!mddev->pers->quiesce || !mddev->thread)
7047                         return -EBUSY;
7048                 if (mddev->recovery || mddev->sync_thread)
7049                         return -EBUSY;
7050                 /* we should be able to change the bitmap.. */
7051         }
7052
7053         if (fd >= 0) {
7054                 struct inode *inode;
7055                 struct file *f;
7056
7057                 if (mddev->bitmap || mddev->bitmap_info.file)
7058                         return -EEXIST; /* cannot add when bitmap is present */
7059                 f = fget(fd);
7060
7061                 if (f == NULL) {
7062                         pr_warn("%s: error: failed to get bitmap file\n",
7063                                 mdname(mddev));
7064                         return -EBADF;
7065                 }
7066
7067                 inode = f->f_mapping->host;
7068                 if (!S_ISREG(inode->i_mode)) {
7069                         pr_warn("%s: error: bitmap file must be a regular file\n",
7070                                 mdname(mddev));
7071                         err = -EBADF;
7072                 } else if (!(f->f_mode & FMODE_WRITE)) {
7073                         pr_warn("%s: error: bitmap file must open for write\n",
7074                                 mdname(mddev));
7075                         err = -EBADF;
7076                 } else if (atomic_read(&inode->i_writecount) != 1) {
7077                         pr_warn("%s: error: bitmap file is already in use\n",
7078                                 mdname(mddev));
7079                         err = -EBUSY;
7080                 }
7081                 if (err) {
7082                         fput(f);
7083                         return err;
7084                 }
7085                 mddev->bitmap_info.file = f;
7086                 mddev->bitmap_info.offset = 0; /* file overrides offset */
7087         } else if (mddev->bitmap == NULL)
7088                 return -ENOENT; /* cannot remove what isn't there */
7089         err = 0;
7090         if (mddev->pers) {
7091                 if (fd >= 0) {
7092                         struct bitmap *bitmap;
7093
7094                         bitmap = md_bitmap_create(mddev, -1);
7095                         mddev_suspend(mddev);
7096                         if (!IS_ERR(bitmap)) {
7097                                 mddev->bitmap = bitmap;
7098                                 err = md_bitmap_load(mddev);
7099                         } else
7100                                 err = PTR_ERR(bitmap);
7101                         if (err) {
7102                                 md_bitmap_destroy(mddev);
7103                                 fd = -1;
7104                         }
7105                         mddev_resume(mddev);
7106                 } else if (fd < 0) {
7107                         mddev_suspend(mddev);
7108                         md_bitmap_destroy(mddev);
7109                         mddev_resume(mddev);
7110                 }
7111         }
7112         if (fd < 0) {
7113                 struct file *f = mddev->bitmap_info.file;
7114                 if (f) {
7115                         spin_lock(&mddev->lock);
7116                         mddev->bitmap_info.file = NULL;
7117                         spin_unlock(&mddev->lock);
7118                         fput(f);
7119                 }
7120         }
7121
7122         return err;
7123 }
7124
7125 /*
7126  * md_set_array_info is used two different ways
7127  * The original usage is when creating a new array.
7128  * In this usage, raid_disks is > 0 and it together with
7129  *  level, size, not_persistent,layout,chunksize determine the
7130  *  shape of the array.
7131  *  This will always create an array with a type-0.90.0 superblock.
7132  * The newer usage is when assembling an array.
7133  *  In this case raid_disks will be 0, and the major_version field is
7134  *  use to determine which style super-blocks are to be found on the devices.
7135  *  The minor and patch _version numbers are also kept incase the
7136  *  super_block handler wishes to interpret them.
7137  */
7138 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7139 {
7140         if (info->raid_disks == 0) {
7141                 /* just setting version number for superblock loading */
7142                 if (info->major_version < 0 ||
7143                     info->major_version >= ARRAY_SIZE(super_types) ||
7144                     super_types[info->major_version].name == NULL) {
7145                         /* maybe try to auto-load a module? */
7146                         pr_warn("md: superblock version %d not known\n",
7147                                 info->major_version);
7148                         return -EINVAL;
7149                 }
7150                 mddev->major_version = info->major_version;
7151                 mddev->minor_version = info->minor_version;
7152                 mddev->patch_version = info->patch_version;
7153                 mddev->persistent = !info->not_persistent;
7154                 /* ensure mddev_put doesn't delete this now that there
7155                  * is some minimal configuration.
7156                  */
7157                 mddev->ctime         = ktime_get_real_seconds();
7158                 return 0;
7159         }
7160         mddev->major_version = MD_MAJOR_VERSION;
7161         mddev->minor_version = MD_MINOR_VERSION;
7162         mddev->patch_version = MD_PATCHLEVEL_VERSION;
7163         mddev->ctime         = ktime_get_real_seconds();
7164
7165         mddev->level         = info->level;
7166         mddev->clevel[0]     = 0;
7167         mddev->dev_sectors   = 2 * (sector_t)info->size;
7168         mddev->raid_disks    = info->raid_disks;
7169         /* don't set md_minor, it is determined by which /dev/md* was
7170          * openned
7171          */
7172         if (info->state & (1<<MD_SB_CLEAN))
7173                 mddev->recovery_cp = MaxSector;
7174         else
7175                 mddev->recovery_cp = 0;
7176         mddev->persistent    = ! info->not_persistent;
7177         mddev->external      = 0;
7178
7179         mddev->layout        = info->layout;
7180         if (mddev->level == 0)
7181                 /* Cannot trust RAID0 layout info here */
7182                 mddev->layout = -1;
7183         mddev->chunk_sectors = info->chunk_size >> 9;
7184
7185         if (mddev->persistent) {
7186                 mddev->max_disks = MD_SB_DISKS;
7187                 mddev->flags = 0;
7188                 mddev->sb_flags = 0;
7189         }
7190         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7191
7192         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7193         mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7194         mddev->bitmap_info.offset = 0;
7195
7196         mddev->reshape_position = MaxSector;
7197
7198         /*
7199          * Generate a 128 bit UUID
7200          */
7201         get_random_bytes(mddev->uuid, 16);
7202
7203         mddev->new_level = mddev->level;
7204         mddev->new_chunk_sectors = mddev->chunk_sectors;
7205         mddev->new_layout = mddev->layout;
7206         mddev->delta_disks = 0;
7207         mddev->reshape_backwards = 0;
7208
7209         return 0;
7210 }
7211
7212 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7213 {
7214         lockdep_assert_held(&mddev->reconfig_mutex);
7215
7216         if (mddev->external_size)
7217                 return;
7218
7219         mddev->array_sectors = array_sectors;
7220 }
7221 EXPORT_SYMBOL(md_set_array_sectors);
7222
7223 static int update_size(struct mddev *mddev, sector_t num_sectors)
7224 {
7225         struct md_rdev *rdev;
7226         int rv;
7227         int fit = (num_sectors == 0);
7228         sector_t old_dev_sectors = mddev->dev_sectors;
7229
7230         if (mddev->pers->resize == NULL)
7231                 return -EINVAL;
7232         /* The "num_sectors" is the number of sectors of each device that
7233          * is used.  This can only make sense for arrays with redundancy.
7234          * linear and raid0 always use whatever space is available. We can only
7235          * consider changing this number if no resync or reconstruction is
7236          * happening, and if the new size is acceptable. It must fit before the
7237          * sb_start or, if that is <data_offset, it must fit before the size
7238          * of each device.  If num_sectors is zero, we find the largest size
7239          * that fits.
7240          */
7241         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7242             mddev->sync_thread)
7243                 return -EBUSY;
7244         if (mddev->ro)
7245                 return -EROFS;
7246
7247         rdev_for_each(rdev, mddev) {
7248                 sector_t avail = rdev->sectors;
7249
7250                 if (fit && (num_sectors == 0 || num_sectors > avail))
7251                         num_sectors = avail;
7252                 if (avail < num_sectors)
7253                         return -ENOSPC;
7254         }
7255         rv = mddev->pers->resize(mddev, num_sectors);
7256         if (!rv) {
7257                 if (mddev_is_clustered(mddev))
7258                         md_cluster_ops->update_size(mddev, old_dev_sectors);
7259                 else if (mddev->queue) {
7260                         set_capacity(mddev->gendisk, mddev->array_sectors);
7261                         revalidate_disk_size(mddev->gendisk, true);
7262                 }
7263         }
7264         return rv;
7265 }
7266
7267 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7268 {
7269         int rv;
7270         struct md_rdev *rdev;
7271         /* change the number of raid disks */
7272         if (mddev->pers->check_reshape == NULL)
7273                 return -EINVAL;
7274         if (mddev->ro)
7275                 return -EROFS;
7276         if (raid_disks <= 0 ||
7277             (mddev->max_disks && raid_disks >= mddev->max_disks))
7278                 return -EINVAL;
7279         if (mddev->sync_thread ||
7280             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7281             mddev->reshape_position != MaxSector)
7282                 return -EBUSY;
7283
7284         rdev_for_each(rdev, mddev) {
7285                 if (mddev->raid_disks < raid_disks &&
7286                     rdev->data_offset < rdev->new_data_offset)
7287                         return -EINVAL;
7288                 if (mddev->raid_disks > raid_disks &&
7289                     rdev->data_offset > rdev->new_data_offset)
7290                         return -EINVAL;
7291         }
7292
7293         mddev->delta_disks = raid_disks - mddev->raid_disks;
7294         if (mddev->delta_disks < 0)
7295                 mddev->reshape_backwards = 1;
7296         else if (mddev->delta_disks > 0)
7297                 mddev->reshape_backwards = 0;
7298
7299         rv = mddev->pers->check_reshape(mddev);
7300         if (rv < 0) {
7301                 mddev->delta_disks = 0;
7302                 mddev->reshape_backwards = 0;
7303         }
7304         return rv;
7305 }
7306
7307 /*
7308  * update_array_info is used to change the configuration of an
7309  * on-line array.
7310  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7311  * fields in the info are checked against the array.
7312  * Any differences that cannot be handled will cause an error.
7313  * Normally, only one change can be managed at a time.
7314  */
7315 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7316 {
7317         int rv = 0;
7318         int cnt = 0;
7319         int state = 0;
7320
7321         /* calculate expected state,ignoring low bits */
7322         if (mddev->bitmap && mddev->bitmap_info.offset)
7323                 state |= (1 << MD_SB_BITMAP_PRESENT);
7324
7325         if (mddev->major_version != info->major_version ||
7326             mddev->minor_version != info->minor_version ||
7327 /*          mddev->patch_version != info->patch_version || */
7328             mddev->ctime         != info->ctime         ||
7329             mddev->level         != info->level         ||
7330 /*          mddev->layout        != info->layout        || */
7331             mddev->persistent    != !info->not_persistent ||
7332             mddev->chunk_sectors != info->chunk_size >> 9 ||
7333             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7334             ((state^info->state) & 0xfffffe00)
7335                 )
7336                 return -EINVAL;
7337         /* Check there is only one change */
7338         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7339                 cnt++;
7340         if (mddev->raid_disks != info->raid_disks)
7341                 cnt++;
7342         if (mddev->layout != info->layout)
7343                 cnt++;
7344         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7345                 cnt++;
7346         if (cnt == 0)
7347                 return 0;
7348         if (cnt > 1)
7349                 return -EINVAL;
7350
7351         if (mddev->layout != info->layout) {
7352                 /* Change layout
7353                  * we don't need to do anything at the md level, the
7354                  * personality will take care of it all.
7355                  */
7356                 if (mddev->pers->check_reshape == NULL)
7357                         return -EINVAL;
7358                 else {
7359                         mddev->new_layout = info->layout;
7360                         rv = mddev->pers->check_reshape(mddev);
7361                         if (rv)
7362                                 mddev->new_layout = mddev->layout;
7363                         return rv;
7364                 }
7365         }
7366         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7367                 rv = update_size(mddev, (sector_t)info->size * 2);
7368
7369         if (mddev->raid_disks    != info->raid_disks)
7370                 rv = update_raid_disks(mddev, info->raid_disks);
7371
7372         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7373                 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7374                         rv = -EINVAL;
7375                         goto err;
7376                 }
7377                 if (mddev->recovery || mddev->sync_thread) {
7378                         rv = -EBUSY;
7379                         goto err;
7380                 }
7381                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7382                         struct bitmap *bitmap;
7383                         /* add the bitmap */
7384                         if (mddev->bitmap) {
7385                                 rv = -EEXIST;
7386                                 goto err;
7387                         }
7388                         if (mddev->bitmap_info.default_offset == 0) {
7389                                 rv = -EINVAL;
7390                                 goto err;
7391                         }
7392                         mddev->bitmap_info.offset =
7393                                 mddev->bitmap_info.default_offset;
7394                         mddev->bitmap_info.space =
7395                                 mddev->bitmap_info.default_space;
7396                         bitmap = md_bitmap_create(mddev, -1);
7397                         mddev_suspend(mddev);
7398                         if (!IS_ERR(bitmap)) {
7399                                 mddev->bitmap = bitmap;
7400                                 rv = md_bitmap_load(mddev);
7401                         } else
7402                                 rv = PTR_ERR(bitmap);
7403                         if (rv)
7404                                 md_bitmap_destroy(mddev);
7405                         mddev_resume(mddev);
7406                 } else {
7407                         /* remove the bitmap */
7408                         if (!mddev->bitmap) {
7409                                 rv = -ENOENT;
7410                                 goto err;
7411                         }
7412                         if (mddev->bitmap->storage.file) {
7413                                 rv = -EINVAL;
7414                                 goto err;
7415                         }
7416                         if (mddev->bitmap_info.nodes) {
7417                                 /* hold PW on all the bitmap lock */
7418                                 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7419                                         pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7420                                         rv = -EPERM;
7421                                         md_cluster_ops->unlock_all_bitmaps(mddev);
7422                                         goto err;
7423                                 }
7424
7425                                 mddev->bitmap_info.nodes = 0;
7426                                 md_cluster_ops->leave(mddev);
7427                                 module_put(md_cluster_mod);
7428                                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7429                         }
7430                         mddev_suspend(mddev);
7431                         md_bitmap_destroy(mddev);
7432                         mddev_resume(mddev);
7433                         mddev->bitmap_info.offset = 0;
7434                 }
7435         }
7436         md_update_sb(mddev, 1);
7437         return rv;
7438 err:
7439         return rv;
7440 }
7441
7442 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7443 {
7444         struct md_rdev *rdev;
7445         int err = 0;
7446
7447         if (mddev->pers == NULL)
7448                 return -ENODEV;
7449
7450         rcu_read_lock();
7451         rdev = md_find_rdev_rcu(mddev, dev);
7452         if (!rdev)
7453                 err =  -ENODEV;
7454         else {
7455                 md_error(mddev, rdev);
7456                 if (!test_bit(Faulty, &rdev->flags))
7457                         err = -EBUSY;
7458         }
7459         rcu_read_unlock();
7460         return err;
7461 }
7462
7463 /*
7464  * We have a problem here : there is no easy way to give a CHS
7465  * virtual geometry. We currently pretend that we have a 2 heads
7466  * 4 sectors (with a BIG number of cylinders...). This drives
7467  * dosfs just mad... ;-)
7468  */
7469 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7470 {
7471         struct mddev *mddev = bdev->bd_disk->private_data;
7472
7473         geo->heads = 2;
7474         geo->sectors = 4;
7475         geo->cylinders = mddev->array_sectors / 8;
7476         return 0;
7477 }
7478
7479 static inline bool md_ioctl_valid(unsigned int cmd)
7480 {
7481         switch (cmd) {
7482         case ADD_NEW_DISK:
7483         case BLKROSET:
7484         case GET_ARRAY_INFO:
7485         case GET_BITMAP_FILE:
7486         case GET_DISK_INFO:
7487         case HOT_ADD_DISK:
7488         case HOT_REMOVE_DISK:
7489         case RAID_VERSION:
7490         case RESTART_ARRAY_RW:
7491         case RUN_ARRAY:
7492         case SET_ARRAY_INFO:
7493         case SET_BITMAP_FILE:
7494         case SET_DISK_FAULTY:
7495         case STOP_ARRAY:
7496         case STOP_ARRAY_RO:
7497         case CLUSTERED_DISK_NACK:
7498                 return true;
7499         default:
7500                 return false;
7501         }
7502 }
7503
7504 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7505                         unsigned int cmd, unsigned long arg)
7506 {
7507         int err = 0;
7508         void __user *argp = (void __user *)arg;
7509         struct mddev *mddev = NULL;
7510         int ro;
7511         bool did_set_md_closing = false;
7512
7513         if (!md_ioctl_valid(cmd))
7514                 return -ENOTTY;
7515
7516         switch (cmd) {
7517         case RAID_VERSION:
7518         case GET_ARRAY_INFO:
7519         case GET_DISK_INFO:
7520                 break;
7521         default:
7522                 if (!capable(CAP_SYS_ADMIN))
7523                         return -EACCES;
7524         }
7525
7526         /*
7527          * Commands dealing with the RAID driver but not any
7528          * particular array:
7529          */
7530         switch (cmd) {
7531         case RAID_VERSION:
7532                 err = get_version(argp);
7533                 goto out;
7534         default:;
7535         }
7536
7537         /*
7538          * Commands creating/starting a new array:
7539          */
7540
7541         mddev = bdev->bd_disk->private_data;
7542
7543         if (!mddev) {
7544                 BUG();
7545                 goto out;
7546         }
7547
7548         /* Some actions do not requires the mutex */
7549         switch (cmd) {
7550         case GET_ARRAY_INFO:
7551                 if (!mddev->raid_disks && !mddev->external)
7552                         err = -ENODEV;
7553                 else
7554                         err = get_array_info(mddev, argp);
7555                 goto out;
7556
7557         case GET_DISK_INFO:
7558                 if (!mddev->raid_disks && !mddev->external)
7559                         err = -ENODEV;
7560                 else
7561                         err = get_disk_info(mddev, argp);
7562                 goto out;
7563
7564         case SET_DISK_FAULTY:
7565                 err = set_disk_faulty(mddev, new_decode_dev(arg));
7566                 goto out;
7567
7568         case GET_BITMAP_FILE:
7569                 err = get_bitmap_file(mddev, argp);
7570                 goto out;
7571
7572         }
7573
7574         if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7575                 flush_rdev_wq(mddev);
7576
7577         if (cmd == HOT_REMOVE_DISK)
7578                 /* need to ensure recovery thread has run */
7579                 wait_event_interruptible_timeout(mddev->sb_wait,
7580                                                  !test_bit(MD_RECOVERY_NEEDED,
7581                                                            &mddev->recovery),
7582                                                  msecs_to_jiffies(5000));
7583         if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7584                 /* Need to flush page cache, and ensure no-one else opens
7585                  * and writes
7586                  */
7587                 mutex_lock(&mddev->open_mutex);
7588                 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7589                         mutex_unlock(&mddev->open_mutex);
7590                         err = -EBUSY;
7591                         goto out;
7592                 }
7593                 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7594                 set_bit(MD_CLOSING, &mddev->flags);
7595                 did_set_md_closing = true;
7596                 mutex_unlock(&mddev->open_mutex);
7597                 sync_blockdev(bdev);
7598         }
7599         err = mddev_lock(mddev);
7600         if (err) {
7601                 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7602                          err, cmd);
7603                 goto out;
7604         }
7605
7606         if (cmd == SET_ARRAY_INFO) {
7607                 mdu_array_info_t info;
7608                 if (!arg)
7609                         memset(&info, 0, sizeof(info));
7610                 else if (copy_from_user(&info, argp, sizeof(info))) {
7611                         err = -EFAULT;
7612                         goto unlock;
7613                 }
7614                 if (mddev->pers) {
7615                         err = update_array_info(mddev, &info);
7616                         if (err) {
7617                                 pr_warn("md: couldn't update array info. %d\n", err);
7618                                 goto unlock;
7619                         }
7620                         goto unlock;
7621                 }
7622                 if (!list_empty(&mddev->disks)) {
7623                         pr_warn("md: array %s already has disks!\n", mdname(mddev));
7624                         err = -EBUSY;
7625                         goto unlock;
7626                 }
7627                 if (mddev->raid_disks) {
7628                         pr_warn("md: array %s already initialised!\n", mdname(mddev));
7629                         err = -EBUSY;
7630                         goto unlock;
7631                 }
7632                 err = md_set_array_info(mddev, &info);
7633                 if (err) {
7634                         pr_warn("md: couldn't set array info. %d\n", err);
7635                         goto unlock;
7636                 }
7637                 goto unlock;
7638         }
7639
7640         /*
7641          * Commands querying/configuring an existing array:
7642          */
7643         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7644          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7645         if ((!mddev->raid_disks && !mddev->external)
7646             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7647             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7648             && cmd != GET_BITMAP_FILE) {
7649                 err = -ENODEV;
7650                 goto unlock;
7651         }
7652
7653         /*
7654          * Commands even a read-only array can execute:
7655          */
7656         switch (cmd) {
7657         case RESTART_ARRAY_RW:
7658                 err = restart_array(mddev);
7659                 goto unlock;
7660
7661         case STOP_ARRAY:
7662                 err = do_md_stop(mddev, 0, bdev);
7663                 goto unlock;
7664
7665         case STOP_ARRAY_RO:
7666                 err = md_set_readonly(mddev, bdev);
7667                 goto unlock;
7668
7669         case HOT_REMOVE_DISK:
7670                 err = hot_remove_disk(mddev, new_decode_dev(arg));
7671                 goto unlock;
7672
7673         case ADD_NEW_DISK:
7674                 /* We can support ADD_NEW_DISK on read-only arrays
7675                  * only if we are re-adding a preexisting device.
7676                  * So require mddev->pers and MD_DISK_SYNC.
7677                  */
7678                 if (mddev->pers) {
7679                         mdu_disk_info_t info;
7680                         if (copy_from_user(&info, argp, sizeof(info)))
7681                                 err = -EFAULT;
7682                         else if (!(info.state & (1<<MD_DISK_SYNC)))
7683                                 /* Need to clear read-only for this */
7684                                 break;
7685                         else
7686                                 err = md_add_new_disk(mddev, &info);
7687                         goto unlock;
7688                 }
7689                 break;
7690
7691         case BLKROSET:
7692                 if (get_user(ro, (int __user *)(arg))) {
7693                         err = -EFAULT;
7694                         goto unlock;
7695                 }
7696                 err = -EINVAL;
7697
7698                 /* if the bdev is going readonly the value of mddev->ro
7699                  * does not matter, no writes are coming
7700                  */
7701                 if (ro)
7702                         goto unlock;
7703
7704                 /* are we are already prepared for writes? */
7705                 if (mddev->ro != 1)
7706                         goto unlock;
7707
7708                 /* transitioning to readauto need only happen for
7709                  * arrays that call md_write_start
7710                  */
7711                 if (mddev->pers) {
7712                         err = restart_array(mddev);
7713                         if (err == 0) {
7714                                 mddev->ro = 2;
7715                                 set_disk_ro(mddev->gendisk, 0);
7716                         }
7717                 }
7718                 goto unlock;
7719         }
7720
7721         /*
7722          * The remaining ioctls are changing the state of the
7723          * superblock, so we do not allow them on read-only arrays.
7724          */
7725         if (mddev->ro && mddev->pers) {
7726                 if (mddev->ro == 2) {
7727                         mddev->ro = 0;
7728                         sysfs_notify_dirent_safe(mddev->sysfs_state);
7729                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7730                         /* mddev_unlock will wake thread */
7731                         /* If a device failed while we were read-only, we
7732                          * need to make sure the metadata is updated now.
7733                          */
7734                         if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7735                                 mddev_unlock(mddev);
7736                                 wait_event(mddev->sb_wait,
7737                                            !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7738                                            !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7739                                 mddev_lock_nointr(mddev);
7740                         }
7741                 } else {
7742                         err = -EROFS;
7743                         goto unlock;
7744                 }
7745         }
7746
7747         switch (cmd) {
7748         case ADD_NEW_DISK:
7749         {
7750                 mdu_disk_info_t info;
7751                 if (copy_from_user(&info, argp, sizeof(info)))
7752                         err = -EFAULT;
7753                 else
7754                         err = md_add_new_disk(mddev, &info);
7755                 goto unlock;
7756         }
7757
7758         case CLUSTERED_DISK_NACK:
7759                 if (mddev_is_clustered(mddev))
7760                         md_cluster_ops->new_disk_ack(mddev, false);
7761                 else
7762                         err = -EINVAL;
7763                 goto unlock;
7764
7765         case HOT_ADD_DISK:
7766                 err = hot_add_disk(mddev, new_decode_dev(arg));
7767                 goto unlock;
7768
7769         case RUN_ARRAY:
7770                 err = do_md_run(mddev);
7771                 goto unlock;
7772
7773         case SET_BITMAP_FILE:
7774                 err = set_bitmap_file(mddev, (int)arg);
7775                 goto unlock;
7776
7777         default:
7778                 err = -EINVAL;
7779                 goto unlock;
7780         }
7781
7782 unlock:
7783         if (mddev->hold_active == UNTIL_IOCTL &&
7784             err != -EINVAL)
7785                 mddev->hold_active = 0;
7786         mddev_unlock(mddev);
7787 out:
7788         if(did_set_md_closing)
7789                 clear_bit(MD_CLOSING, &mddev->flags);
7790         return err;
7791 }
7792 #ifdef CONFIG_COMPAT
7793 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7794                     unsigned int cmd, unsigned long arg)
7795 {
7796         switch (cmd) {
7797         case HOT_REMOVE_DISK:
7798         case HOT_ADD_DISK:
7799         case SET_DISK_FAULTY:
7800         case SET_BITMAP_FILE:
7801                 /* These take in integer arg, do not convert */
7802                 break;
7803         default:
7804                 arg = (unsigned long)compat_ptr(arg);
7805                 break;
7806         }
7807
7808         return md_ioctl(bdev, mode, cmd, arg);
7809 }
7810 #endif /* CONFIG_COMPAT */
7811
7812 static int md_open(struct block_device *bdev, fmode_t mode)
7813 {
7814         /*
7815          * Succeed if we can lock the mddev, which confirms that
7816          * it isn't being stopped right now.
7817          */
7818         struct mddev *mddev = mddev_find(bdev->bd_dev);
7819         int err;
7820
7821         if (!mddev)
7822                 return -ENODEV;
7823
7824         if (mddev->gendisk != bdev->bd_disk) {
7825                 /* we are racing with mddev_put which is discarding this
7826                  * bd_disk.
7827                  */
7828                 mddev_put(mddev);
7829                 /* Wait until bdev->bd_disk is definitely gone */
7830                 if (work_pending(&mddev->del_work))
7831                         flush_workqueue(md_misc_wq);
7832                 /* Then retry the open from the top */
7833                 return -ERESTARTSYS;
7834         }
7835         BUG_ON(mddev != bdev->bd_disk->private_data);
7836
7837         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7838                 goto out;
7839
7840         if (test_bit(MD_CLOSING, &mddev->flags)) {
7841                 mutex_unlock(&mddev->open_mutex);
7842                 err = -ENODEV;
7843                 goto out;
7844         }
7845
7846         err = 0;
7847         atomic_inc(&mddev->openers);
7848         mutex_unlock(&mddev->open_mutex);
7849
7850         bdev_check_media_change(bdev);
7851  out:
7852         if (err)
7853                 mddev_put(mddev);
7854         return err;
7855 }
7856
7857 static void md_release(struct gendisk *disk, fmode_t mode)
7858 {
7859         struct mddev *mddev = disk->private_data;
7860
7861         BUG_ON(!mddev);
7862         atomic_dec(&mddev->openers);
7863         mddev_put(mddev);
7864 }
7865
7866 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7867 {
7868         struct mddev *mddev = disk->private_data;
7869         unsigned int ret = 0;
7870
7871         if (mddev->changed)
7872                 ret = DISK_EVENT_MEDIA_CHANGE;
7873         mddev->changed = 0;
7874         return ret;
7875 }
7876
7877 const struct block_device_operations md_fops =
7878 {
7879         .owner          = THIS_MODULE,
7880         .submit_bio     = md_submit_bio,
7881         .open           = md_open,
7882         .release        = md_release,
7883         .ioctl          = md_ioctl,
7884 #ifdef CONFIG_COMPAT
7885         .compat_ioctl   = md_compat_ioctl,
7886 #endif
7887         .getgeo         = md_getgeo,
7888         .check_events   = md_check_events,
7889 };
7890
7891 static int md_thread(void *arg)
7892 {
7893         struct md_thread *thread = arg;
7894
7895         /*
7896          * md_thread is a 'system-thread', it's priority should be very
7897          * high. We avoid resource deadlocks individually in each
7898          * raid personality. (RAID5 does preallocation) We also use RR and
7899          * the very same RT priority as kswapd, thus we will never get
7900          * into a priority inversion deadlock.
7901          *
7902          * we definitely have to have equal or higher priority than
7903          * bdflush, otherwise bdflush will deadlock if there are too
7904          * many dirty RAID5 blocks.
7905          */
7906
7907         allow_signal(SIGKILL);
7908         while (!kthread_should_stop()) {
7909
7910                 /* We need to wait INTERRUPTIBLE so that
7911                  * we don't add to the load-average.
7912                  * That means we need to be sure no signals are
7913                  * pending
7914                  */
7915                 if (signal_pending(current))
7916                         flush_signals(current);
7917
7918                 wait_event_interruptible_timeout
7919                         (thread->wqueue,
7920                          test_bit(THREAD_WAKEUP, &thread->flags)
7921                          || kthread_should_stop() || kthread_should_park(),
7922                          thread->timeout);
7923
7924                 clear_bit(THREAD_WAKEUP, &thread->flags);
7925                 if (kthread_should_park())
7926                         kthread_parkme();
7927                 if (!kthread_should_stop())
7928                         thread->run(thread);
7929         }
7930
7931         return 0;
7932 }
7933
7934 void md_wakeup_thread(struct md_thread *thread)
7935 {
7936         if (thread) {
7937                 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7938                 set_bit(THREAD_WAKEUP, &thread->flags);
7939                 wake_up(&thread->wqueue);
7940         }
7941 }
7942 EXPORT_SYMBOL(md_wakeup_thread);
7943
7944 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7945                 struct mddev *mddev, const char *name)
7946 {
7947         struct md_thread *thread;
7948
7949         thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7950         if (!thread)
7951                 return NULL;
7952
7953         init_waitqueue_head(&thread->wqueue);
7954
7955         thread->run = run;
7956         thread->mddev = mddev;
7957         thread->timeout = MAX_SCHEDULE_TIMEOUT;
7958         thread->tsk = kthread_run(md_thread, thread,
7959                                   "%s_%s",
7960                                   mdname(thread->mddev),
7961                                   name);
7962         if (IS_ERR(thread->tsk)) {
7963                 kfree(thread);
7964                 return NULL;
7965         }
7966         return thread;
7967 }
7968 EXPORT_SYMBOL(md_register_thread);
7969
7970 void md_unregister_thread(struct md_thread **threadp)
7971 {
7972         struct md_thread *thread = *threadp;
7973         if (!thread)
7974                 return;
7975         pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7976         /* Locking ensures that mddev_unlock does not wake_up a
7977          * non-existent thread
7978          */
7979         spin_lock(&pers_lock);
7980         *threadp = NULL;
7981         spin_unlock(&pers_lock);
7982
7983         kthread_stop(thread->tsk);
7984         kfree(thread);
7985 }
7986 EXPORT_SYMBOL(md_unregister_thread);
7987
7988 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7989 {
7990         if (!rdev || test_bit(Faulty, &rdev->flags))
7991                 return;
7992
7993         if (!mddev->pers || !mddev->pers->error_handler)
7994                 return;
7995         mddev->pers->error_handler(mddev,rdev);
7996         if (mddev->degraded)
7997                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7998         sysfs_notify_dirent_safe(rdev->sysfs_state);
7999         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8000         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8001         md_wakeup_thread(mddev->thread);
8002         if (mddev->event_work.func)
8003                 queue_work(md_misc_wq, &mddev->event_work);
8004         md_new_event(mddev);
8005 }
8006 EXPORT_SYMBOL(md_error);
8007
8008 /* seq_file implementation /proc/mdstat */
8009
8010 static void status_unused(struct seq_file *seq)
8011 {
8012         int i = 0;
8013         struct md_rdev *rdev;
8014
8015         seq_printf(seq, "unused devices: ");
8016
8017         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8018                 char b[BDEVNAME_SIZE];
8019                 i++;
8020                 seq_printf(seq, "%s ",
8021                               bdevname(rdev->bdev,b));
8022         }
8023         if (!i)
8024                 seq_printf(seq, "<none>");
8025
8026         seq_printf(seq, "\n");
8027 }
8028
8029 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8030 {
8031         sector_t max_sectors, resync, res;
8032         unsigned long dt, db = 0;
8033         sector_t rt, curr_mark_cnt, resync_mark_cnt;
8034         int scale, recovery_active;
8035         unsigned int per_milli;
8036
8037         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8038             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8039                 max_sectors = mddev->resync_max_sectors;
8040         else
8041                 max_sectors = mddev->dev_sectors;
8042
8043         resync = mddev->curr_resync;
8044         if (resync <= 3) {
8045                 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8046                         /* Still cleaning up */
8047                         resync = max_sectors;
8048         } else if (resync > max_sectors)
8049                 resync = max_sectors;
8050         else
8051                 resync -= atomic_read(&mddev->recovery_active);
8052
8053         if (resync == 0) {
8054                 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8055                         struct md_rdev *rdev;
8056
8057                         rdev_for_each(rdev, mddev)
8058                                 if (rdev->raid_disk >= 0 &&
8059                                     !test_bit(Faulty, &rdev->flags) &&
8060                                     rdev->recovery_offset != MaxSector &&
8061                                     rdev->recovery_offset) {
8062                                         seq_printf(seq, "\trecover=REMOTE");
8063                                         return 1;
8064                                 }
8065                         if (mddev->reshape_position != MaxSector)
8066                                 seq_printf(seq, "\treshape=REMOTE");
8067                         else
8068                                 seq_printf(seq, "\tresync=REMOTE");
8069                         return 1;
8070                 }
8071                 if (mddev->recovery_cp < MaxSector) {
8072                         seq_printf(seq, "\tresync=PENDING");
8073                         return 1;
8074                 }
8075                 return 0;
8076         }
8077         if (resync < 3) {
8078                 seq_printf(seq, "\tresync=DELAYED");
8079                 return 1;
8080         }
8081
8082         WARN_ON(max_sectors == 0);
8083         /* Pick 'scale' such that (resync>>scale)*1000 will fit
8084          * in a sector_t, and (max_sectors>>scale) will fit in a
8085          * u32, as those are the requirements for sector_div.
8086          * Thus 'scale' must be at least 10
8087          */
8088         scale = 10;
8089         if (sizeof(sector_t) > sizeof(unsigned long)) {
8090                 while ( max_sectors/2 > (1ULL<<(scale+32)))
8091                         scale++;
8092         }
8093         res = (resync>>scale)*1000;
8094         sector_div(res, (u32)((max_sectors>>scale)+1));
8095
8096         per_milli = res;
8097         {
8098                 int i, x = per_milli/50, y = 20-x;
8099                 seq_printf(seq, "[");
8100                 for (i = 0; i < x; i++)
8101                         seq_printf(seq, "=");
8102                 seq_printf(seq, ">");
8103                 for (i = 0; i < y; i++)
8104                         seq_printf(seq, ".");
8105                 seq_printf(seq, "] ");
8106         }
8107         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8108                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8109                     "reshape" :
8110                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8111                      "check" :
8112                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8113                       "resync" : "recovery"))),
8114                    per_milli/10, per_milli % 10,
8115                    (unsigned long long) resync/2,
8116                    (unsigned long long) max_sectors/2);
8117
8118         /*
8119          * dt: time from mark until now
8120          * db: blocks written from mark until now
8121          * rt: remaining time
8122          *
8123          * rt is a sector_t, which is always 64bit now. We are keeping
8124          * the original algorithm, but it is not really necessary.
8125          *
8126          * Original algorithm:
8127          *   So we divide before multiply in case it is 32bit and close
8128          *   to the limit.
8129          *   We scale the divisor (db) by 32 to avoid losing precision
8130          *   near the end of resync when the number of remaining sectors
8131          *   is close to 'db'.
8132          *   We then divide rt by 32 after multiplying by db to compensate.
8133          *   The '+1' avoids division by zero if db is very small.
8134          */
8135         dt = ((jiffies - mddev->resync_mark) / HZ);
8136         if (!dt) dt++;
8137
8138         curr_mark_cnt = mddev->curr_mark_cnt;
8139         recovery_active = atomic_read(&mddev->recovery_active);
8140         resync_mark_cnt = mddev->resync_mark_cnt;
8141
8142         if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8143                 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8144
8145         rt = max_sectors - resync;    /* number of remaining sectors */
8146         rt = div64_u64(rt, db/32+1);
8147         rt *= dt;
8148         rt >>= 5;
8149
8150         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8151                    ((unsigned long)rt % 60)/6);
8152
8153         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8154         return 1;
8155 }
8156
8157 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8158 {
8159         struct list_head *tmp;
8160         loff_t l = *pos;
8161         struct mddev *mddev;
8162
8163         if (l >= 0x10000)
8164                 return NULL;
8165         if (!l--)
8166                 /* header */
8167                 return (void*)1;
8168
8169         spin_lock(&all_mddevs_lock);
8170         list_for_each(tmp,&all_mddevs)
8171                 if (!l--) {
8172                         mddev = list_entry(tmp, struct mddev, all_mddevs);
8173                         mddev_get(mddev);
8174                         spin_unlock(&all_mddevs_lock);
8175                         return mddev;
8176                 }
8177         spin_unlock(&all_mddevs_lock);
8178         if (!l--)
8179                 return (void*)2;/* tail */
8180         return NULL;
8181 }
8182
8183 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8184 {
8185         struct list_head *tmp;
8186         struct mddev *next_mddev, *mddev = v;
8187
8188         ++*pos;
8189         if (v == (void*)2)
8190                 return NULL;
8191
8192         spin_lock(&all_mddevs_lock);
8193         if (v == (void*)1)
8194                 tmp = all_mddevs.next;
8195         else
8196                 tmp = mddev->all_mddevs.next;
8197         if (tmp != &all_mddevs)
8198                 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8199         else {
8200                 next_mddev = (void*)2;
8201                 *pos = 0x10000;
8202         }
8203         spin_unlock(&all_mddevs_lock);
8204
8205         if (v != (void*)1)
8206                 mddev_put(mddev);
8207         return next_mddev;
8208
8209 }
8210
8211 static void md_seq_stop(struct seq_file *seq, void *v)
8212 {
8213         struct mddev *mddev = v;
8214
8215         if (mddev && v != (void*)1 && v != (void*)2)
8216                 mddev_put(mddev);
8217 }
8218
8219 static int md_seq_show(struct seq_file *seq, void *v)
8220 {
8221         struct mddev *mddev = v;
8222         sector_t sectors;
8223         struct md_rdev *rdev;
8224
8225         if (v == (void*)1) {
8226                 struct md_personality *pers;
8227                 seq_printf(seq, "Personalities : ");
8228                 spin_lock(&pers_lock);
8229                 list_for_each_entry(pers, &pers_list, list)
8230                         seq_printf(seq, "[%s] ", pers->name);
8231
8232                 spin_unlock(&pers_lock);
8233                 seq_printf(seq, "\n");
8234                 seq->poll_event = atomic_read(&md_event_count);
8235                 return 0;
8236         }
8237         if (v == (void*)2) {
8238                 status_unused(seq);
8239                 return 0;
8240         }
8241
8242         spin_lock(&mddev->lock);
8243         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8244                 seq_printf(seq, "%s : %sactive", mdname(mddev),
8245                                                 mddev->pers ? "" : "in");
8246                 if (mddev->pers) {
8247                         if (mddev->ro==1)
8248                                 seq_printf(seq, " (read-only)");
8249                         if (mddev->ro==2)
8250                                 seq_printf(seq, " (auto-read-only)");
8251                         seq_printf(seq, " %s", mddev->pers->name);
8252                 }
8253
8254                 sectors = 0;
8255                 rcu_read_lock();
8256                 rdev_for_each_rcu(rdev, mddev) {
8257                         char b[BDEVNAME_SIZE];
8258                         seq_printf(seq, " %s[%d]",
8259                                 bdevname(rdev->bdev,b), rdev->desc_nr);
8260                         if (test_bit(WriteMostly, &rdev->flags))
8261                                 seq_printf(seq, "(W)");
8262                         if (test_bit(Journal, &rdev->flags))
8263                                 seq_printf(seq, "(J)");
8264                         if (test_bit(Faulty, &rdev->flags)) {
8265                                 seq_printf(seq, "(F)");
8266                                 continue;
8267                         }
8268                         if (rdev->raid_disk < 0)
8269                                 seq_printf(seq, "(S)"); /* spare */
8270                         if (test_bit(Replacement, &rdev->flags))
8271                                 seq_printf(seq, "(R)");
8272                         sectors += rdev->sectors;
8273                 }
8274                 rcu_read_unlock();
8275
8276                 if (!list_empty(&mddev->disks)) {
8277                         if (mddev->pers)
8278                                 seq_printf(seq, "\n      %llu blocks",
8279                                            (unsigned long long)
8280                                            mddev->array_sectors / 2);
8281                         else
8282                                 seq_printf(seq, "\n      %llu blocks",
8283                                            (unsigned long long)sectors / 2);
8284                 }
8285                 if (mddev->persistent) {
8286                         if (mddev->major_version != 0 ||
8287                             mddev->minor_version != 90) {
8288                                 seq_printf(seq," super %d.%d",
8289                                            mddev->major_version,
8290                                            mddev->minor_version);
8291                         }
8292                 } else if (mddev->external)
8293                         seq_printf(seq, " super external:%s",
8294                                    mddev->metadata_type);
8295                 else
8296                         seq_printf(seq, " super non-persistent");
8297
8298                 if (mddev->pers) {
8299                         mddev->pers->status(seq, mddev);
8300                         seq_printf(seq, "\n      ");
8301                         if (mddev->pers->sync_request) {
8302                                 if (status_resync(seq, mddev))
8303                                         seq_printf(seq, "\n      ");
8304                         }
8305                 } else
8306                         seq_printf(seq, "\n       ");
8307
8308                 md_bitmap_status(seq, mddev->bitmap);
8309
8310                 seq_printf(seq, "\n");
8311         }
8312         spin_unlock(&mddev->lock);
8313
8314         return 0;
8315 }
8316
8317 static const struct seq_operations md_seq_ops = {
8318         .start  = md_seq_start,
8319         .next   = md_seq_next,
8320         .stop   = md_seq_stop,
8321         .show   = md_seq_show,
8322 };
8323
8324 static int md_seq_open(struct inode *inode, struct file *file)
8325 {
8326         struct seq_file *seq;
8327         int error;
8328
8329         error = seq_open(file, &md_seq_ops);
8330         if (error)
8331                 return error;
8332
8333         seq = file->private_data;
8334         seq->poll_event = atomic_read(&md_event_count);
8335         return error;
8336 }
8337
8338 static int md_unloading;
8339 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8340 {
8341         struct seq_file *seq = filp->private_data;
8342         __poll_t mask;
8343
8344         if (md_unloading)
8345                 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8346         poll_wait(filp, &md_event_waiters, wait);
8347
8348         /* always allow read */
8349         mask = EPOLLIN | EPOLLRDNORM;
8350
8351         if (seq->poll_event != atomic_read(&md_event_count))
8352                 mask |= EPOLLERR | EPOLLPRI;
8353         return mask;
8354 }
8355
8356 static const struct proc_ops mdstat_proc_ops = {
8357         .proc_open      = md_seq_open,
8358         .proc_read      = seq_read,
8359         .proc_lseek     = seq_lseek,
8360         .proc_release   = seq_release,
8361         .proc_poll      = mdstat_poll,
8362 };
8363
8364 int register_md_personality(struct md_personality *p)
8365 {
8366         pr_debug("md: %s personality registered for level %d\n",
8367                  p->name, p->level);
8368         spin_lock(&pers_lock);
8369         list_add_tail(&p->list, &pers_list);
8370         spin_unlock(&pers_lock);
8371         return 0;
8372 }
8373 EXPORT_SYMBOL(register_md_personality);
8374
8375 int unregister_md_personality(struct md_personality *p)
8376 {
8377         pr_debug("md: %s personality unregistered\n", p->name);
8378         spin_lock(&pers_lock);
8379         list_del_init(&p->list);
8380         spin_unlock(&pers_lock);
8381         return 0;
8382 }
8383 EXPORT_SYMBOL(unregister_md_personality);
8384
8385 int register_md_cluster_operations(struct md_cluster_operations *ops,
8386                                    struct module *module)
8387 {
8388         int ret = 0;
8389         spin_lock(&pers_lock);
8390         if (md_cluster_ops != NULL)
8391                 ret = -EALREADY;
8392         else {
8393                 md_cluster_ops = ops;
8394                 md_cluster_mod = module;
8395         }
8396         spin_unlock(&pers_lock);
8397         return ret;
8398 }
8399 EXPORT_SYMBOL(register_md_cluster_operations);
8400
8401 int unregister_md_cluster_operations(void)
8402 {
8403         spin_lock(&pers_lock);
8404         md_cluster_ops = NULL;
8405         spin_unlock(&pers_lock);
8406         return 0;
8407 }
8408 EXPORT_SYMBOL(unregister_md_cluster_operations);
8409
8410 int md_setup_cluster(struct mddev *mddev, int nodes)
8411 {
8412         int ret;
8413         if (!md_cluster_ops)
8414                 request_module("md-cluster");
8415         spin_lock(&pers_lock);
8416         /* ensure module won't be unloaded */
8417         if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8418                 pr_warn("can't find md-cluster module or get it's reference.\n");
8419                 spin_unlock(&pers_lock);
8420                 return -ENOENT;
8421         }
8422         spin_unlock(&pers_lock);
8423
8424         ret = md_cluster_ops->join(mddev, nodes);
8425         if (!ret)
8426                 mddev->safemode_delay = 0;
8427         return ret;
8428 }
8429
8430 void md_cluster_stop(struct mddev *mddev)
8431 {
8432         if (!md_cluster_ops)
8433                 return;
8434         md_cluster_ops->leave(mddev);
8435         module_put(md_cluster_mod);
8436 }
8437
8438 static int is_mddev_idle(struct mddev *mddev, int init)
8439 {
8440         struct md_rdev *rdev;
8441         int idle;
8442         int curr_events;
8443
8444         idle = 1;
8445         rcu_read_lock();
8446         rdev_for_each_rcu(rdev, mddev) {
8447                 struct gendisk *disk = rdev->bdev->bd_disk;
8448                 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8449                               atomic_read(&disk->sync_io);
8450                 /* sync IO will cause sync_io to increase before the disk_stats
8451                  * as sync_io is counted when a request starts, and
8452                  * disk_stats is counted when it completes.
8453                  * So resync activity will cause curr_events to be smaller than
8454                  * when there was no such activity.
8455                  * non-sync IO will cause disk_stat to increase without
8456                  * increasing sync_io so curr_events will (eventually)
8457                  * be larger than it was before.  Once it becomes
8458                  * substantially larger, the test below will cause
8459                  * the array to appear non-idle, and resync will slow
8460                  * down.
8461                  * If there is a lot of outstanding resync activity when
8462                  * we set last_event to curr_events, then all that activity
8463                  * completing might cause the array to appear non-idle
8464                  * and resync will be slowed down even though there might
8465                  * not have been non-resync activity.  This will only
8466                  * happen once though.  'last_events' will soon reflect
8467                  * the state where there is little or no outstanding
8468                  * resync requests, and further resync activity will
8469                  * always make curr_events less than last_events.
8470                  *
8471                  */
8472                 if (init || curr_events - rdev->last_events > 64) {
8473                         rdev->last_events = curr_events;
8474                         idle = 0;
8475                 }
8476         }
8477         rcu_read_unlock();
8478         return idle;
8479 }
8480
8481 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8482 {
8483         /* another "blocks" (512byte) blocks have been synced */
8484         atomic_sub(blocks, &mddev->recovery_active);
8485         wake_up(&mddev->recovery_wait);
8486         if (!ok) {
8487                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8488                 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8489                 md_wakeup_thread(mddev->thread);
8490                 // stop recovery, signal do_sync ....
8491         }
8492 }
8493 EXPORT_SYMBOL(md_done_sync);
8494
8495 /* md_write_start(mddev, bi)
8496  * If we need to update some array metadata (e.g. 'active' flag
8497  * in superblock) before writing, schedule a superblock update
8498  * and wait for it to complete.
8499  * A return value of 'false' means that the write wasn't recorded
8500  * and cannot proceed as the array is being suspend.
8501  */
8502 bool md_write_start(struct mddev *mddev, struct bio *bi)
8503 {
8504         int did_change = 0;
8505
8506         if (bio_data_dir(bi) != WRITE)
8507                 return true;
8508
8509         BUG_ON(mddev->ro == 1);
8510         if (mddev->ro == 2) {
8511                 /* need to switch to read/write */
8512                 mddev->ro = 0;
8513                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8514                 md_wakeup_thread(mddev->thread);
8515                 md_wakeup_thread(mddev->sync_thread);
8516                 did_change = 1;
8517         }
8518         rcu_read_lock();
8519         percpu_ref_get(&mddev->writes_pending);
8520         smp_mb(); /* Match smp_mb in set_in_sync() */
8521         if (mddev->safemode == 1)
8522                 mddev->safemode = 0;
8523         /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8524         if (mddev->in_sync || mddev->sync_checkers) {
8525                 spin_lock(&mddev->lock);
8526                 if (mddev->in_sync) {
8527                         mddev->in_sync = 0;
8528                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8529                         set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8530                         md_wakeup_thread(mddev->thread);
8531                         did_change = 1;
8532                 }
8533                 spin_unlock(&mddev->lock);
8534         }
8535         rcu_read_unlock();
8536         if (did_change)
8537                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8538         if (!mddev->has_superblocks)
8539                 return true;
8540         wait_event(mddev->sb_wait,
8541                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8542                    mddev->suspended);
8543         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8544                 percpu_ref_put(&mddev->writes_pending);
8545                 return false;
8546         }
8547         return true;
8548 }
8549 EXPORT_SYMBOL(md_write_start);
8550
8551 /* md_write_inc can only be called when md_write_start() has
8552  * already been called at least once of the current request.
8553  * It increments the counter and is useful when a single request
8554  * is split into several parts.  Each part causes an increment and
8555  * so needs a matching md_write_end().
8556  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8557  * a spinlocked region.
8558  */
8559 void md_write_inc(struct mddev *mddev, struct bio *bi)
8560 {
8561         if (bio_data_dir(bi) != WRITE)
8562                 return;
8563         WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8564         percpu_ref_get(&mddev->writes_pending);
8565 }
8566 EXPORT_SYMBOL(md_write_inc);
8567
8568 void md_write_end(struct mddev *mddev)
8569 {
8570         percpu_ref_put(&mddev->writes_pending);
8571
8572         if (mddev->safemode == 2)
8573                 md_wakeup_thread(mddev->thread);
8574         else if (mddev->safemode_delay)
8575                 /* The roundup() ensures this only performs locking once
8576                  * every ->safemode_delay jiffies
8577                  */
8578                 mod_timer(&mddev->safemode_timer,
8579                           roundup(jiffies, mddev->safemode_delay) +
8580                           mddev->safemode_delay);
8581 }
8582
8583 EXPORT_SYMBOL(md_write_end);
8584
8585 /* This is used by raid0 and raid10 */
8586 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8587                         struct bio *bio, sector_t start, sector_t size)
8588 {
8589         struct bio *discard_bio = NULL;
8590
8591         if (__blkdev_issue_discard(rdev->bdev, start, size,
8592                 GFP_NOIO, 0, &discard_bio) || !discard_bio)
8593                 return;
8594
8595         bio_chain(discard_bio, bio);
8596         bio_clone_blkg_association(discard_bio, bio);
8597         if (mddev->gendisk)
8598                 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
8599                         discard_bio, disk_devt(mddev->gendisk),
8600                         bio->bi_iter.bi_sector);
8601         submit_bio_noacct(discard_bio);
8602 }
8603 EXPORT_SYMBOL(md_submit_discard_bio);
8604
8605 /* md_allow_write(mddev)
8606  * Calling this ensures that the array is marked 'active' so that writes
8607  * may proceed without blocking.  It is important to call this before
8608  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8609  * Must be called with mddev_lock held.
8610  */
8611 void md_allow_write(struct mddev *mddev)
8612 {
8613         if (!mddev->pers)
8614                 return;
8615         if (mddev->ro)
8616                 return;
8617         if (!mddev->pers->sync_request)
8618                 return;
8619
8620         spin_lock(&mddev->lock);
8621         if (mddev->in_sync) {
8622                 mddev->in_sync = 0;
8623                 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8624                 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8625                 if (mddev->safemode_delay &&
8626                     mddev->safemode == 0)
8627                         mddev->safemode = 1;
8628                 spin_unlock(&mddev->lock);
8629                 md_update_sb(mddev, 0);
8630                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8631                 /* wait for the dirty state to be recorded in the metadata */
8632                 wait_event(mddev->sb_wait,
8633                            !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8634         } else
8635                 spin_unlock(&mddev->lock);
8636 }
8637 EXPORT_SYMBOL_GPL(md_allow_write);
8638
8639 #define SYNC_MARKS      10
8640 #define SYNC_MARK_STEP  (3*HZ)
8641 #define UPDATE_FREQUENCY (5*60*HZ)
8642 void md_do_sync(struct md_thread *thread)
8643 {
8644         struct mddev *mddev = thread->mddev;
8645         struct mddev *mddev2;
8646         unsigned int currspeed = 0, window;
8647         sector_t max_sectors,j, io_sectors, recovery_done;
8648         unsigned long mark[SYNC_MARKS];
8649         unsigned long update_time;
8650         sector_t mark_cnt[SYNC_MARKS];
8651         int last_mark,m;
8652         struct list_head *tmp;
8653         sector_t last_check;
8654         int skipped = 0;
8655         struct md_rdev *rdev;
8656         char *desc, *action = NULL;
8657         struct blk_plug plug;
8658         int ret;
8659
8660         /* just incase thread restarts... */
8661         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8662             test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8663                 return;
8664         if (mddev->ro) {/* never try to sync a read-only array */
8665                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8666                 return;
8667         }
8668
8669         if (mddev_is_clustered(mddev)) {
8670                 ret = md_cluster_ops->resync_start(mddev);
8671                 if (ret)
8672                         goto skip;
8673
8674                 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8675                 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8676                         test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8677                         test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8678                      && ((unsigned long long)mddev->curr_resync_completed
8679                          < (unsigned long long)mddev->resync_max_sectors))
8680                         goto skip;
8681         }
8682
8683         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8684                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8685                         desc = "data-check";
8686                         action = "check";
8687                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8688                         desc = "requested-resync";
8689                         action = "repair";
8690                 } else
8691                         desc = "resync";
8692         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8693                 desc = "reshape";
8694         else
8695                 desc = "recovery";
8696
8697         mddev->last_sync_action = action ?: desc;
8698
8699         /* we overload curr_resync somewhat here.
8700          * 0 == not engaged in resync at all
8701          * 2 == checking that there is no conflict with another sync
8702          * 1 == like 2, but have yielded to allow conflicting resync to
8703          *              commence
8704          * other == active in resync - this many blocks
8705          *
8706          * Before starting a resync we must have set curr_resync to
8707          * 2, and then checked that every "conflicting" array has curr_resync
8708          * less than ours.  When we find one that is the same or higher
8709          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8710          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8711          * This will mean we have to start checking from the beginning again.
8712          *
8713          */
8714
8715         do {
8716                 int mddev2_minor = -1;
8717                 mddev->curr_resync = 2;
8718
8719         try_again:
8720                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8721                         goto skip;
8722                 for_each_mddev(mddev2, tmp) {
8723                         if (mddev2 == mddev)
8724                                 continue;
8725                         if (!mddev->parallel_resync
8726                         &&  mddev2->curr_resync
8727                         &&  match_mddev_units(mddev, mddev2)) {
8728                                 DEFINE_WAIT(wq);
8729                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
8730                                         /* arbitrarily yield */
8731                                         mddev->curr_resync = 1;
8732                                         wake_up(&resync_wait);
8733                                 }
8734                                 if (mddev > mddev2 && mddev->curr_resync == 1)
8735                                         /* no need to wait here, we can wait the next
8736                                          * time 'round when curr_resync == 2
8737                                          */
8738                                         continue;
8739                                 /* We need to wait 'interruptible' so as not to
8740                                  * contribute to the load average, and not to
8741                                  * be caught by 'softlockup'
8742                                  */
8743                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8744                                 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8745                                     mddev2->curr_resync >= mddev->curr_resync) {
8746                                         if (mddev2_minor != mddev2->md_minor) {
8747                                                 mddev2_minor = mddev2->md_minor;
8748                                                 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8749                                                         desc, mdname(mddev),
8750                                                         mdname(mddev2));
8751                                         }
8752                                         mddev_put(mddev2);
8753                                         if (signal_pending(current))
8754                                                 flush_signals(current);
8755                                         schedule();
8756                                         finish_wait(&resync_wait, &wq);
8757                                         goto try_again;
8758                                 }
8759                                 finish_wait(&resync_wait, &wq);
8760                         }
8761                 }
8762         } while (mddev->curr_resync < 2);
8763
8764         j = 0;
8765         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8766                 /* resync follows the size requested by the personality,
8767                  * which defaults to physical size, but can be virtual size
8768                  */
8769                 max_sectors = mddev->resync_max_sectors;
8770                 atomic64_set(&mddev->resync_mismatches, 0);
8771                 /* we don't use the checkpoint if there's a bitmap */
8772                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8773                         j = mddev->resync_min;
8774                 else if (!mddev->bitmap)
8775                         j = mddev->recovery_cp;
8776
8777         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8778                 max_sectors = mddev->resync_max_sectors;
8779                 /*
8780                  * If the original node aborts reshaping then we continue the
8781                  * reshaping, so set j again to avoid restart reshape from the
8782                  * first beginning
8783                  */
8784                 if (mddev_is_clustered(mddev) &&
8785                     mddev->reshape_position != MaxSector)
8786                         j = mddev->reshape_position;
8787         } else {
8788                 /* recovery follows the physical size of devices */
8789                 max_sectors = mddev->dev_sectors;
8790                 j = MaxSector;
8791                 rcu_read_lock();
8792                 rdev_for_each_rcu(rdev, mddev)
8793                         if (rdev->raid_disk >= 0 &&
8794                             !test_bit(Journal, &rdev->flags) &&
8795                             !test_bit(Faulty, &rdev->flags) &&
8796                             !test_bit(In_sync, &rdev->flags) &&
8797                             rdev->recovery_offset < j)
8798                                 j = rdev->recovery_offset;
8799                 rcu_read_unlock();
8800
8801                 /* If there is a bitmap, we need to make sure all
8802                  * writes that started before we added a spare
8803                  * complete before we start doing a recovery.
8804                  * Otherwise the write might complete and (via
8805                  * bitmap_endwrite) set a bit in the bitmap after the
8806                  * recovery has checked that bit and skipped that
8807                  * region.
8808                  */
8809                 if (mddev->bitmap) {
8810                         mddev->pers->quiesce(mddev, 1);
8811                         mddev->pers->quiesce(mddev, 0);
8812                 }
8813         }
8814
8815         pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8816         pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8817         pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8818                  speed_max(mddev), desc);
8819
8820         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8821
8822         io_sectors = 0;
8823         for (m = 0; m < SYNC_MARKS; m++) {
8824                 mark[m] = jiffies;
8825                 mark_cnt[m] = io_sectors;
8826         }
8827         last_mark = 0;
8828         mddev->resync_mark = mark[last_mark];
8829         mddev->resync_mark_cnt = mark_cnt[last_mark];
8830
8831         /*
8832          * Tune reconstruction:
8833          */
8834         window = 32 * (PAGE_SIZE / 512);
8835         pr_debug("md: using %dk window, over a total of %lluk.\n",
8836                  window/2, (unsigned long long)max_sectors/2);
8837
8838         atomic_set(&mddev->recovery_active, 0);
8839         last_check = 0;
8840
8841         if (j>2) {
8842                 pr_debug("md: resuming %s of %s from checkpoint.\n",
8843                          desc, mdname(mddev));
8844                 mddev->curr_resync = j;
8845         } else
8846                 mddev->curr_resync = 3; /* no longer delayed */
8847         mddev->curr_resync_completed = j;
8848         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8849         md_new_event(mddev);
8850         update_time = jiffies;
8851
8852         blk_start_plug(&plug);
8853         while (j < max_sectors) {
8854                 sector_t sectors;
8855
8856                 skipped = 0;
8857
8858                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8859                     ((mddev->curr_resync > mddev->curr_resync_completed &&
8860                       (mddev->curr_resync - mddev->curr_resync_completed)
8861                       > (max_sectors >> 4)) ||
8862                      time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8863                      (j - mddev->curr_resync_completed)*2
8864                      >= mddev->resync_max - mddev->curr_resync_completed ||
8865                      mddev->curr_resync_completed > mddev->resync_max
8866                             )) {
8867                         /* time to update curr_resync_completed */
8868                         wait_event(mddev->recovery_wait,
8869                                    atomic_read(&mddev->recovery_active) == 0);
8870                         mddev->curr_resync_completed = j;
8871                         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8872                             j > mddev->recovery_cp)
8873                                 mddev->recovery_cp = j;
8874                         update_time = jiffies;
8875                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8876                         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8877                 }
8878
8879                 while (j >= mddev->resync_max &&
8880                        !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8881                         /* As this condition is controlled by user-space,
8882                          * we can block indefinitely, so use '_interruptible'
8883                          * to avoid triggering warnings.
8884                          */
8885                         flush_signals(current); /* just in case */
8886                         wait_event_interruptible(mddev->recovery_wait,
8887                                                  mddev->resync_max > j
8888                                                  || test_bit(MD_RECOVERY_INTR,
8889                                                              &mddev->recovery));
8890                 }
8891
8892                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8893                         break;
8894
8895                 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8896                 if (sectors == 0) {
8897                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8898                         break;
8899                 }
8900
8901                 if (!skipped) { /* actual IO requested */
8902                         io_sectors += sectors;
8903                         atomic_add(sectors, &mddev->recovery_active);
8904                 }
8905
8906                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8907                         break;
8908
8909                 j += sectors;
8910                 if (j > max_sectors)
8911                         /* when skipping, extra large numbers can be returned. */
8912                         j = max_sectors;
8913                 if (j > 2)
8914                         mddev->curr_resync = j;
8915                 mddev->curr_mark_cnt = io_sectors;
8916                 if (last_check == 0)
8917                         /* this is the earliest that rebuild will be
8918                          * visible in /proc/mdstat
8919                          */
8920                         md_new_event(mddev);
8921
8922                 if (last_check + window > io_sectors || j == max_sectors)
8923                         continue;
8924
8925                 last_check = io_sectors;
8926         repeat:
8927                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8928                         /* step marks */
8929                         int next = (last_mark+1) % SYNC_MARKS;
8930
8931                         mddev->resync_mark = mark[next];
8932                         mddev->resync_mark_cnt = mark_cnt[next];
8933                         mark[next] = jiffies;
8934                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8935                         last_mark = next;
8936                 }
8937
8938                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8939                         break;
8940
8941                 /*
8942                  * this loop exits only if either when we are slower than
8943                  * the 'hard' speed limit, or the system was IO-idle for
8944                  * a jiffy.
8945                  * the system might be non-idle CPU-wise, but we only care
8946                  * about not overloading the IO subsystem. (things like an
8947                  * e2fsck being done on the RAID array should execute fast)
8948                  */
8949                 cond_resched();
8950
8951                 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8952                 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8953                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
8954
8955                 if (currspeed > speed_min(mddev)) {
8956                         if (currspeed > speed_max(mddev)) {
8957                                 msleep(500);
8958                                 goto repeat;
8959                         }
8960                         if (!is_mddev_idle(mddev, 0)) {
8961                                 /*
8962                                  * Give other IO more of a chance.
8963                                  * The faster the devices, the less we wait.
8964                                  */
8965                                 wait_event(mddev->recovery_wait,
8966                                            !atomic_read(&mddev->recovery_active));
8967                         }
8968                 }
8969         }
8970         pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8971                 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8972                 ? "interrupted" : "done");
8973         /*
8974          * this also signals 'finished resyncing' to md_stop
8975          */
8976         blk_finish_plug(&plug);
8977         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8978
8979         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8980             !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8981             mddev->curr_resync > 3) {
8982                 mddev->curr_resync_completed = mddev->curr_resync;
8983                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8984         }
8985         mddev->pers->sync_request(mddev, max_sectors, &skipped);
8986
8987         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8988             mddev->curr_resync > 3) {
8989                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8990                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8991                                 if (mddev->curr_resync >= mddev->recovery_cp) {
8992                                         pr_debug("md: checkpointing %s of %s.\n",
8993                                                  desc, mdname(mddev));
8994                                         if (test_bit(MD_RECOVERY_ERROR,
8995                                                 &mddev->recovery))
8996                                                 mddev->recovery_cp =
8997                                                         mddev->curr_resync_completed;
8998                                         else
8999                                                 mddev->recovery_cp =
9000                                                         mddev->curr_resync;
9001                                 }
9002                         } else
9003                                 mddev->recovery_cp = MaxSector;
9004                 } else {
9005                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9006                                 mddev->curr_resync = MaxSector;
9007                         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9008                             test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9009                                 rcu_read_lock();
9010                                 rdev_for_each_rcu(rdev, mddev)
9011                                         if (rdev->raid_disk >= 0 &&
9012                                             mddev->delta_disks >= 0 &&
9013                                             !test_bit(Journal, &rdev->flags) &&
9014                                             !test_bit(Faulty, &rdev->flags) &&
9015                                             !test_bit(In_sync, &rdev->flags) &&
9016                                             rdev->recovery_offset < mddev->curr_resync)
9017                                                 rdev->recovery_offset = mddev->curr_resync;
9018                                 rcu_read_unlock();
9019                         }
9020                 }
9021         }
9022  skip:
9023         /* set CHANGE_PENDING here since maybe another update is needed,
9024          * so other nodes are informed. It should be harmless for normal
9025          * raid */
9026         set_mask_bits(&mddev->sb_flags, 0,
9027                       BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9028
9029         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9030                         !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9031                         mddev->delta_disks > 0 &&
9032                         mddev->pers->finish_reshape &&
9033                         mddev->pers->size &&
9034                         mddev->queue) {
9035                 mddev_lock_nointr(mddev);
9036                 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9037                 mddev_unlock(mddev);
9038                 if (!mddev_is_clustered(mddev)) {
9039                         set_capacity(mddev->gendisk, mddev->array_sectors);
9040                         revalidate_disk_size(mddev->gendisk, true);
9041                 }
9042         }
9043
9044         spin_lock(&mddev->lock);
9045         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9046                 /* We completed so min/max setting can be forgotten if used. */
9047                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9048                         mddev->resync_min = 0;
9049                 mddev->resync_max = MaxSector;
9050         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9051                 mddev->resync_min = mddev->curr_resync_completed;
9052         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9053         mddev->curr_resync = 0;
9054         spin_unlock(&mddev->lock);
9055
9056         wake_up(&resync_wait);
9057         md_wakeup_thread(mddev->thread);
9058         return;
9059 }
9060 EXPORT_SYMBOL_GPL(md_do_sync);
9061
9062 static int remove_and_add_spares(struct mddev *mddev,
9063                                  struct md_rdev *this)
9064 {
9065         struct md_rdev *rdev;
9066         int spares = 0;
9067         int removed = 0;
9068         bool remove_some = false;
9069
9070         if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9071                 /* Mustn't remove devices when resync thread is running */
9072                 return 0;
9073
9074         rdev_for_each(rdev, mddev) {
9075                 if ((this == NULL || rdev == this) &&
9076                     rdev->raid_disk >= 0 &&
9077                     !test_bit(Blocked, &rdev->flags) &&
9078                     test_bit(Faulty, &rdev->flags) &&
9079                     atomic_read(&rdev->nr_pending)==0) {
9080                         /* Faulty non-Blocked devices with nr_pending == 0
9081                          * never get nr_pending incremented,
9082                          * never get Faulty cleared, and never get Blocked set.
9083                          * So we can synchronize_rcu now rather than once per device
9084                          */
9085                         remove_some = true;
9086                         set_bit(RemoveSynchronized, &rdev->flags);
9087                 }
9088         }
9089
9090         if (remove_some)
9091                 synchronize_rcu();
9092         rdev_for_each(rdev, mddev) {
9093                 if ((this == NULL || rdev == this) &&
9094                     rdev->raid_disk >= 0 &&
9095                     !test_bit(Blocked, &rdev->flags) &&
9096                     ((test_bit(RemoveSynchronized, &rdev->flags) ||
9097                      (!test_bit(In_sync, &rdev->flags) &&
9098                       !test_bit(Journal, &rdev->flags))) &&
9099                     atomic_read(&rdev->nr_pending)==0)) {
9100                         if (mddev->pers->hot_remove_disk(
9101                                     mddev, rdev) == 0) {
9102                                 sysfs_unlink_rdev(mddev, rdev);
9103                                 rdev->saved_raid_disk = rdev->raid_disk;
9104                                 rdev->raid_disk = -1;
9105                                 removed++;
9106                         }
9107                 }
9108                 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9109                         clear_bit(RemoveSynchronized, &rdev->flags);
9110         }
9111
9112         if (removed && mddev->kobj.sd)
9113                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9114
9115         if (this && removed)
9116                 goto no_add;
9117
9118         rdev_for_each(rdev, mddev) {
9119                 if (this && this != rdev)
9120                         continue;
9121                 if (test_bit(Candidate, &rdev->flags))
9122                         continue;
9123                 if (rdev->raid_disk >= 0 &&
9124                     !test_bit(In_sync, &rdev->flags) &&
9125                     !test_bit(Journal, &rdev->flags) &&
9126                     !test_bit(Faulty, &rdev->flags))
9127                         spares++;
9128                 if (rdev->raid_disk >= 0)
9129                         continue;
9130                 if (test_bit(Faulty, &rdev->flags))
9131                         continue;
9132                 if (!test_bit(Journal, &rdev->flags)) {
9133                         if (mddev->ro &&
9134                             ! (rdev->saved_raid_disk >= 0 &&
9135                                !test_bit(Bitmap_sync, &rdev->flags)))
9136                                 continue;
9137
9138                         rdev->recovery_offset = 0;
9139                 }
9140                 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9141                         /* failure here is OK */
9142                         sysfs_link_rdev(mddev, rdev);
9143                         if (!test_bit(Journal, &rdev->flags))
9144                                 spares++;
9145                         md_new_event(mddev);
9146                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9147                 }
9148         }
9149 no_add:
9150         if (removed)
9151                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9152         return spares;
9153 }
9154
9155 static void md_start_sync(struct work_struct *ws)
9156 {
9157         struct mddev *mddev = container_of(ws, struct mddev, del_work);
9158
9159         mddev->sync_thread = md_register_thread(md_do_sync,
9160                                                 mddev,
9161                                                 "resync");
9162         if (!mddev->sync_thread) {
9163                 pr_warn("%s: could not start resync thread...\n",
9164                         mdname(mddev));
9165                 /* leave the spares where they are, it shouldn't hurt */
9166                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9167                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9168                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9169                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9170                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9171                 wake_up(&resync_wait);
9172                 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9173                                        &mddev->recovery))
9174                         if (mddev->sysfs_action)
9175                                 sysfs_notify_dirent_safe(mddev->sysfs_action);
9176         } else
9177                 md_wakeup_thread(mddev->sync_thread);
9178         sysfs_notify_dirent_safe(mddev->sysfs_action);
9179         md_new_event(mddev);
9180 }
9181
9182 /*
9183  * This routine is regularly called by all per-raid-array threads to
9184  * deal with generic issues like resync and super-block update.
9185  * Raid personalities that don't have a thread (linear/raid0) do not
9186  * need this as they never do any recovery or update the superblock.
9187  *
9188  * It does not do any resync itself, but rather "forks" off other threads
9189  * to do that as needed.
9190  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9191  * "->recovery" and create a thread at ->sync_thread.
9192  * When the thread finishes it sets MD_RECOVERY_DONE
9193  * and wakeups up this thread which will reap the thread and finish up.
9194  * This thread also removes any faulty devices (with nr_pending == 0).
9195  *
9196  * The overall approach is:
9197  *  1/ if the superblock needs updating, update it.
9198  *  2/ If a recovery thread is running, don't do anything else.
9199  *  3/ If recovery has finished, clean up, possibly marking spares active.
9200  *  4/ If there are any faulty devices, remove them.
9201  *  5/ If array is degraded, try to add spares devices
9202  *  6/ If array has spares or is not in-sync, start a resync thread.
9203  */
9204 void md_check_recovery(struct mddev *mddev)
9205 {
9206         if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9207                 /* Write superblock - thread that called mddev_suspend()
9208                  * holds reconfig_mutex for us.
9209                  */
9210                 set_bit(MD_UPDATING_SB, &mddev->flags);
9211                 smp_mb__after_atomic();
9212                 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9213                         md_update_sb(mddev, 0);
9214                 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9215                 wake_up(&mddev->sb_wait);
9216         }
9217
9218         if (mddev->suspended)
9219                 return;
9220
9221         if (mddev->bitmap)
9222                 md_bitmap_daemon_work(mddev);
9223
9224         if (signal_pending(current)) {
9225                 if (mddev->pers->sync_request && !mddev->external) {
9226                         pr_debug("md: %s in immediate safe mode\n",
9227                                  mdname(mddev));
9228                         mddev->safemode = 2;
9229                 }
9230                 flush_signals(current);
9231         }
9232
9233         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9234                 return;
9235         if ( ! (
9236                 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9237                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9238                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9239                 (mddev->external == 0 && mddev->safemode == 1) ||
9240                 (mddev->safemode == 2
9241                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9242                 ))
9243                 return;
9244
9245         if (mddev_trylock(mddev)) {
9246                 int spares = 0;
9247                 bool try_set_sync = mddev->safemode != 0;
9248
9249                 if (!mddev->external && mddev->safemode == 1)
9250                         mddev->safemode = 0;
9251
9252                 if (mddev->ro) {
9253                         struct md_rdev *rdev;
9254                         if (!mddev->external && mddev->in_sync)
9255                                 /* 'Blocked' flag not needed as failed devices
9256                                  * will be recorded if array switched to read/write.
9257                                  * Leaving it set will prevent the device
9258                                  * from being removed.
9259                                  */
9260                                 rdev_for_each(rdev, mddev)
9261                                         clear_bit(Blocked, &rdev->flags);
9262                         /* On a read-only array we can:
9263                          * - remove failed devices
9264                          * - add already-in_sync devices if the array itself
9265                          *   is in-sync.
9266                          * As we only add devices that are already in-sync,
9267                          * we can activate the spares immediately.
9268                          */
9269                         remove_and_add_spares(mddev, NULL);
9270                         /* There is no thread, but we need to call
9271                          * ->spare_active and clear saved_raid_disk
9272                          */
9273                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9274                         md_reap_sync_thread(mddev);
9275                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9276                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9277                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9278                         goto unlock;
9279                 }
9280
9281                 if (mddev_is_clustered(mddev)) {
9282                         struct md_rdev *rdev;
9283                         /* kick the device if another node issued a
9284                          * remove disk.
9285                          */
9286                         rdev_for_each(rdev, mddev) {
9287                                 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9288                                                 rdev->raid_disk < 0)
9289                                         md_kick_rdev_from_array(rdev);
9290                         }
9291                 }
9292
9293                 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9294                         spin_lock(&mddev->lock);
9295                         set_in_sync(mddev);
9296                         spin_unlock(&mddev->lock);
9297                 }
9298
9299                 if (mddev->sb_flags)
9300                         md_update_sb(mddev, 0);
9301
9302                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9303                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9304                         /* resync/recovery still happening */
9305                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9306                         goto unlock;
9307                 }
9308                 if (mddev->sync_thread) {
9309                         md_reap_sync_thread(mddev);
9310                         goto unlock;
9311                 }
9312                 /* Set RUNNING before clearing NEEDED to avoid
9313                  * any transients in the value of "sync_action".
9314                  */
9315                 mddev->curr_resync_completed = 0;
9316                 spin_lock(&mddev->lock);
9317                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9318                 spin_unlock(&mddev->lock);
9319                 /* Clear some bits that don't mean anything, but
9320                  * might be left set
9321                  */
9322                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9323                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9324
9325                 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9326                     test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9327                         goto not_running;
9328                 /* no recovery is running.
9329                  * remove any failed drives, then
9330                  * add spares if possible.
9331                  * Spares are also removed and re-added, to allow
9332                  * the personality to fail the re-add.
9333                  */
9334
9335                 if (mddev->reshape_position != MaxSector) {
9336                         if (mddev->pers->check_reshape == NULL ||
9337                             mddev->pers->check_reshape(mddev) != 0)
9338                                 /* Cannot proceed */
9339                                 goto not_running;
9340                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9341                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9342                 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9343                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9344                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9345                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9346                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9347                 } else if (mddev->recovery_cp < MaxSector) {
9348                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9349                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9350                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9351                         /* nothing to be done ... */
9352                         goto not_running;
9353
9354                 if (mddev->pers->sync_request) {
9355                         if (spares) {
9356                                 /* We are adding a device or devices to an array
9357                                  * which has the bitmap stored on all devices.
9358                                  * So make sure all bitmap pages get written
9359                                  */
9360                                 md_bitmap_write_all(mddev->bitmap);
9361                         }
9362                         INIT_WORK(&mddev->del_work, md_start_sync);
9363                         queue_work(md_misc_wq, &mddev->del_work);
9364                         goto unlock;
9365                 }
9366         not_running:
9367                 if (!mddev->sync_thread) {
9368                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9369                         wake_up(&resync_wait);
9370                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9371                                                &mddev->recovery))
9372                                 if (mddev->sysfs_action)
9373                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
9374                 }
9375         unlock:
9376                 wake_up(&mddev->sb_wait);
9377                 mddev_unlock(mddev);
9378         }
9379 }
9380 EXPORT_SYMBOL(md_check_recovery);
9381
9382 void md_reap_sync_thread(struct mddev *mddev)
9383 {
9384         struct md_rdev *rdev;
9385         sector_t old_dev_sectors = mddev->dev_sectors;
9386         bool is_reshaped = false;
9387
9388         /* resync has finished, collect result */
9389         md_unregister_thread(&mddev->sync_thread);
9390         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9391             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9392             mddev->degraded != mddev->raid_disks) {
9393                 /* success...*/
9394                 /* activate any spares */
9395                 if (mddev->pers->spare_active(mddev)) {
9396                         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9397                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9398                 }
9399         }
9400         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9401             mddev->pers->finish_reshape) {
9402                 mddev->pers->finish_reshape(mddev);
9403                 if (mddev_is_clustered(mddev))
9404                         is_reshaped = true;
9405         }
9406
9407         /* If array is no-longer degraded, then any saved_raid_disk
9408          * information must be scrapped.
9409          */
9410         if (!mddev->degraded)
9411                 rdev_for_each(rdev, mddev)
9412                         rdev->saved_raid_disk = -1;
9413
9414         md_update_sb(mddev, 1);
9415         /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9416          * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9417          * clustered raid */
9418         if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9419                 md_cluster_ops->resync_finish(mddev);
9420         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9421         clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9422         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9423         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9424         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9425         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9426         /*
9427          * We call md_cluster_ops->update_size here because sync_size could
9428          * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9429          * so it is time to update size across cluster.
9430          */
9431         if (mddev_is_clustered(mddev) && is_reshaped
9432                                       && !test_bit(MD_CLOSING, &mddev->flags))
9433                 md_cluster_ops->update_size(mddev, old_dev_sectors);
9434         wake_up(&resync_wait);
9435         /* flag recovery needed just to double check */
9436         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9437         sysfs_notify_dirent_safe(mddev->sysfs_action);
9438         md_new_event(mddev);
9439         if (mddev->event_work.func)
9440                 queue_work(md_misc_wq, &mddev->event_work);
9441 }
9442 EXPORT_SYMBOL(md_reap_sync_thread);
9443
9444 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9445 {
9446         sysfs_notify_dirent_safe(rdev->sysfs_state);
9447         wait_event_timeout(rdev->blocked_wait,
9448                            !test_bit(Blocked, &rdev->flags) &&
9449                            !test_bit(BlockedBadBlocks, &rdev->flags),
9450                            msecs_to_jiffies(5000));
9451         rdev_dec_pending(rdev, mddev);
9452 }
9453 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9454
9455 void md_finish_reshape(struct mddev *mddev)
9456 {
9457         /* called be personality module when reshape completes. */
9458         struct md_rdev *rdev;
9459
9460         rdev_for_each(rdev, mddev) {
9461                 if (rdev->data_offset > rdev->new_data_offset)
9462                         rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9463                 else
9464                         rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9465                 rdev->data_offset = rdev->new_data_offset;
9466         }
9467 }
9468 EXPORT_SYMBOL(md_finish_reshape);
9469
9470 /* Bad block management */
9471
9472 /* Returns 1 on success, 0 on failure */
9473 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9474                        int is_new)
9475 {
9476         struct mddev *mddev = rdev->mddev;
9477         int rv;
9478         if (is_new)
9479                 s += rdev->new_data_offset;
9480         else
9481                 s += rdev->data_offset;
9482         rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9483         if (rv == 0) {
9484                 /* Make sure they get written out promptly */
9485                 if (test_bit(ExternalBbl, &rdev->flags))
9486                         sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9487                 sysfs_notify_dirent_safe(rdev->sysfs_state);
9488                 set_mask_bits(&mddev->sb_flags, 0,
9489                               BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9490                 md_wakeup_thread(rdev->mddev->thread);
9491                 return 1;
9492         } else
9493                 return 0;
9494 }
9495 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9496
9497 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9498                          int is_new)
9499 {
9500         int rv;
9501         if (is_new)
9502                 s += rdev->new_data_offset;
9503         else
9504                 s += rdev->data_offset;
9505         rv = badblocks_clear(&rdev->badblocks, s, sectors);
9506         if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9507                 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9508         return rv;
9509 }
9510 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9511
9512 static int md_notify_reboot(struct notifier_block *this,
9513                             unsigned long code, void *x)
9514 {
9515         struct list_head *tmp;
9516         struct mddev *mddev;
9517         int need_delay = 0;
9518
9519         for_each_mddev(mddev, tmp) {
9520                 if (mddev_trylock(mddev)) {
9521                         if (mddev->pers)
9522                                 __md_stop_writes(mddev);
9523                         if (mddev->persistent)
9524                                 mddev->safemode = 2;
9525                         mddev_unlock(mddev);
9526                 }
9527                 need_delay = 1;
9528         }
9529         /*
9530          * certain more exotic SCSI devices are known to be
9531          * volatile wrt too early system reboots. While the
9532          * right place to handle this issue is the given
9533          * driver, we do want to have a safe RAID driver ...
9534          */
9535         if (need_delay)
9536                 mdelay(1000*1);
9537
9538         return NOTIFY_DONE;
9539 }
9540
9541 static struct notifier_block md_notifier = {
9542         .notifier_call  = md_notify_reboot,
9543         .next           = NULL,
9544         .priority       = INT_MAX, /* before any real devices */
9545 };
9546
9547 static void md_geninit(void)
9548 {
9549         pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9550
9551         proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9552 }
9553
9554 static int __init md_init(void)
9555 {
9556         int ret = -ENOMEM;
9557
9558         md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9559         if (!md_wq)
9560                 goto err_wq;
9561
9562         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9563         if (!md_misc_wq)
9564                 goto err_misc_wq;
9565
9566         md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9567         if (!md_rdev_misc_wq)
9568                 goto err_rdev_misc_wq;
9569
9570         if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9571                 goto err_md;
9572
9573         if ((ret = register_blkdev(0, "mdp")) < 0)
9574                 goto err_mdp;
9575         mdp_major = ret;
9576
9577         blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9578                             md_probe, NULL, NULL);
9579         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9580                             md_probe, NULL, NULL);
9581
9582         register_reboot_notifier(&md_notifier);
9583         raid_table_header = register_sysctl_table(raid_root_table);
9584
9585         md_geninit();
9586         return 0;
9587
9588 err_mdp:
9589         unregister_blkdev(MD_MAJOR, "md");
9590 err_md:
9591         destroy_workqueue(md_rdev_misc_wq);
9592 err_rdev_misc_wq:
9593         destroy_workqueue(md_misc_wq);
9594 err_misc_wq:
9595         destroy_workqueue(md_wq);
9596 err_wq:
9597         return ret;
9598 }
9599
9600 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9601 {
9602         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9603         struct md_rdev *rdev2;
9604         int role, ret;
9605         char b[BDEVNAME_SIZE];
9606
9607         /*
9608          * If size is changed in another node then we need to
9609          * do resize as well.
9610          */
9611         if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9612                 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9613                 if (ret)
9614                         pr_info("md-cluster: resize failed\n");
9615                 else
9616                         md_bitmap_update_sb(mddev->bitmap);
9617         }
9618
9619         /* Check for change of roles in the active devices */
9620         rdev_for_each(rdev2, mddev) {
9621                 if (test_bit(Faulty, &rdev2->flags))
9622                         continue;
9623
9624                 /* Check if the roles changed */
9625                 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9626
9627                 if (test_bit(Candidate, &rdev2->flags)) {
9628                         if (role == 0xfffe) {
9629                                 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9630                                 md_kick_rdev_from_array(rdev2);
9631                                 continue;
9632                         }
9633                         else
9634                                 clear_bit(Candidate, &rdev2->flags);
9635                 }
9636
9637                 if (role != rdev2->raid_disk) {
9638                         /*
9639                          * got activated except reshape is happening.
9640                          */
9641                         if (rdev2->raid_disk == -1 && role != 0xffff &&
9642                             !(le32_to_cpu(sb->feature_map) &
9643                               MD_FEATURE_RESHAPE_ACTIVE)) {
9644                                 rdev2->saved_raid_disk = role;
9645                                 ret = remove_and_add_spares(mddev, rdev2);
9646                                 pr_info("Activated spare: %s\n",
9647                                         bdevname(rdev2->bdev,b));
9648                                 /* wakeup mddev->thread here, so array could
9649                                  * perform resync with the new activated disk */
9650                                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9651                                 md_wakeup_thread(mddev->thread);
9652                         }
9653                         /* device faulty
9654                          * We just want to do the minimum to mark the disk
9655                          * as faulty. The recovery is performed by the
9656                          * one who initiated the error.
9657                          */
9658                         if ((role == 0xfffe) || (role == 0xfffd)) {
9659                                 md_error(mddev, rdev2);
9660                                 clear_bit(Blocked, &rdev2->flags);
9661                         }
9662                 }
9663         }
9664
9665         if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9666                 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9667
9668         /*
9669          * Since mddev->delta_disks has already updated in update_raid_disks,
9670          * so it is time to check reshape.
9671          */
9672         if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9673             (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9674                 /*
9675                  * reshape is happening in the remote node, we need to
9676                  * update reshape_position and call start_reshape.
9677                  */
9678                 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9679                 if (mddev->pers->update_reshape_pos)
9680                         mddev->pers->update_reshape_pos(mddev);
9681                 if (mddev->pers->start_reshape)
9682                         mddev->pers->start_reshape(mddev);
9683         } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9684                    mddev->reshape_position != MaxSector &&
9685                    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9686                 /* reshape is just done in another node. */
9687                 mddev->reshape_position = MaxSector;
9688                 if (mddev->pers->update_reshape_pos)
9689                         mddev->pers->update_reshape_pos(mddev);
9690         }
9691
9692         /* Finally set the event to be up to date */
9693         mddev->events = le64_to_cpu(sb->events);
9694 }
9695
9696 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9697 {
9698         int err;
9699         struct page *swapout = rdev->sb_page;
9700         struct mdp_superblock_1 *sb;
9701
9702         /* Store the sb page of the rdev in the swapout temporary
9703          * variable in case we err in the future
9704          */
9705         rdev->sb_page = NULL;
9706         err = alloc_disk_sb(rdev);
9707         if (err == 0) {
9708                 ClearPageUptodate(rdev->sb_page);
9709                 rdev->sb_loaded = 0;
9710                 err = super_types[mddev->major_version].
9711                         load_super(rdev, NULL, mddev->minor_version);
9712         }
9713         if (err < 0) {
9714                 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9715                                 __func__, __LINE__, rdev->desc_nr, err);
9716                 if (rdev->sb_page)
9717                         put_page(rdev->sb_page);
9718                 rdev->sb_page = swapout;
9719                 rdev->sb_loaded = 1;
9720                 return err;
9721         }
9722
9723         sb = page_address(rdev->sb_page);
9724         /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9725          * is not set
9726          */
9727
9728         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9729                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9730
9731         /* The other node finished recovery, call spare_active to set
9732          * device In_sync and mddev->degraded
9733          */
9734         if (rdev->recovery_offset == MaxSector &&
9735             !test_bit(In_sync, &rdev->flags) &&
9736             mddev->pers->spare_active(mddev))
9737                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9738
9739         put_page(swapout);
9740         return 0;
9741 }
9742
9743 void md_reload_sb(struct mddev *mddev, int nr)
9744 {
9745         struct md_rdev *rdev;
9746         int err;
9747
9748         /* Find the rdev */
9749         rdev_for_each_rcu(rdev, mddev) {
9750                 if (rdev->desc_nr == nr)
9751                         break;
9752         }
9753
9754         if (!rdev || rdev->desc_nr != nr) {
9755                 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9756                 return;
9757         }
9758
9759         err = read_rdev(mddev, rdev);
9760         if (err < 0)
9761                 return;
9762
9763         check_sb_changes(mddev, rdev);
9764
9765         /* Read all rdev's to update recovery_offset */
9766         rdev_for_each_rcu(rdev, mddev) {
9767                 if (!test_bit(Faulty, &rdev->flags))
9768                         read_rdev(mddev, rdev);
9769         }
9770 }
9771 EXPORT_SYMBOL(md_reload_sb);
9772
9773 #ifndef MODULE
9774
9775 /*
9776  * Searches all registered partitions for autorun RAID arrays
9777  * at boot time.
9778  */
9779
9780 static DEFINE_MUTEX(detected_devices_mutex);
9781 static LIST_HEAD(all_detected_devices);
9782 struct detected_devices_node {
9783         struct list_head list;
9784         dev_t dev;
9785 };
9786
9787 void md_autodetect_dev(dev_t dev)
9788 {
9789         struct detected_devices_node *node_detected_dev;
9790
9791         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9792         if (node_detected_dev) {
9793                 node_detected_dev->dev = dev;
9794                 mutex_lock(&detected_devices_mutex);
9795                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9796                 mutex_unlock(&detected_devices_mutex);
9797         }
9798 }
9799
9800 void md_autostart_arrays(int part)
9801 {
9802         struct md_rdev *rdev;
9803         struct detected_devices_node *node_detected_dev;
9804         dev_t dev;
9805         int i_scanned, i_passed;
9806
9807         i_scanned = 0;
9808         i_passed = 0;
9809
9810         pr_info("md: Autodetecting RAID arrays.\n");
9811
9812         mutex_lock(&detected_devices_mutex);
9813         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9814                 i_scanned++;
9815                 node_detected_dev = list_entry(all_detected_devices.next,
9816                                         struct detected_devices_node, list);
9817                 list_del(&node_detected_dev->list);
9818                 dev = node_detected_dev->dev;
9819                 kfree(node_detected_dev);
9820                 mutex_unlock(&detected_devices_mutex);
9821                 rdev = md_import_device(dev,0, 90);
9822                 mutex_lock(&detected_devices_mutex);
9823                 if (IS_ERR(rdev))
9824                         continue;
9825
9826                 if (test_bit(Faulty, &rdev->flags))
9827                         continue;
9828
9829                 set_bit(AutoDetected, &rdev->flags);
9830                 list_add(&rdev->same_set, &pending_raid_disks);
9831                 i_passed++;
9832         }
9833         mutex_unlock(&detected_devices_mutex);
9834
9835         pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9836
9837         autorun_devices(part);
9838 }
9839
9840 #endif /* !MODULE */
9841
9842 static __exit void md_exit(void)
9843 {
9844         struct mddev *mddev;
9845         struct list_head *tmp;
9846         int delay = 1;
9847
9848         blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9849         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9850
9851         unregister_blkdev(MD_MAJOR,"md");
9852         unregister_blkdev(mdp_major, "mdp");
9853         unregister_reboot_notifier(&md_notifier);
9854         unregister_sysctl_table(raid_table_header);
9855
9856         /* We cannot unload the modules while some process is
9857          * waiting for us in select() or poll() - wake them up
9858          */
9859         md_unloading = 1;
9860         while (waitqueue_active(&md_event_waiters)) {
9861                 /* not safe to leave yet */
9862                 wake_up(&md_event_waiters);
9863                 msleep(delay);
9864                 delay += delay;
9865         }
9866         remove_proc_entry("mdstat", NULL);
9867
9868         for_each_mddev(mddev, tmp) {
9869                 export_array(mddev);
9870                 mddev->ctime = 0;
9871                 mddev->hold_active = 0;
9872                 /*
9873                  * for_each_mddev() will call mddev_put() at the end of each
9874                  * iteration.  As the mddev is now fully clear, this will
9875                  * schedule the mddev for destruction by a workqueue, and the
9876                  * destroy_workqueue() below will wait for that to complete.
9877                  */
9878         }
9879         destroy_workqueue(md_rdev_misc_wq);
9880         destroy_workqueue(md_misc_wq);
9881         destroy_workqueue(md_wq);
9882 }
9883
9884 subsys_initcall(md_init);
9885 module_exit(md_exit)
9886
9887 static int get_ro(char *buffer, const struct kernel_param *kp)
9888 {
9889         return sprintf(buffer, "%d\n", start_readonly);
9890 }
9891 static int set_ro(const char *val, const struct kernel_param *kp)
9892 {
9893         return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9894 }
9895
9896 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9897 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9898 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9899 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9900
9901 MODULE_LICENSE("GPL");
9902 MODULE_DESCRIPTION("MD RAID framework");
9903 MODULE_ALIAS("md");
9904 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);