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 <[email protected]>
  12    - boot support for linear and striped mode by Harald Hoyer <[email protected]>
  13    - kerneld support by Boris Tobotras <[email protected]>
  14    - kmod support by: Cyrus Durgin
  15    - RAID0 bugfixes: Mark Anthony Lisher <[email protected]>
  16    - Devfs support by Richard Gooch <[email protected]>
  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 <[email protected]>.
  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/blk-integrity.h>
  45 #include <linux/badblocks.h>
  46 #include <linux/sysctl.h>
  47 #include <linux/seq_file.h>
  48 #include <linux/fs.h>
  49 #include <linux/poll.h>
  50 #include <linux/ctype.h>
  51 #include <linux/string.h>
  52 #include <linux/hdreg.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/random.h>
  55 #include <linux/major.h>
  56 #include <linux/module.h>
  57 #include <linux/reboot.h>
  58 #include <linux/file.h>
  59 #include <linux/compat.h>
  60 #include <linux/delay.h>
  61 #include <linux/raid/md_p.h>
  62 #include <linux/raid/md_u.h>
  63 #include <linux/raid/detect.h>
  64 #include <linux/slab.h>
  65 #include <linux/percpu-refcount.h>
  66 #include <linux/part_stat.h>
  67
  68 #include <trace/events/block.h>
  69 #include "md.h"
  70 #include "md-bitmap.h"
  71 #include "md-cluster.h"
  72
  73 /* pers_list is a list of registered personalities protected
  74  * by pers_lock.
  75  * pers_lock does extra service to protect accesses to
  76  * mddev->thread when the mutex cannot be held.
  77  */
  78 static LIST_HEAD(pers_list);
  79 static DEFINE_SPINLOCK(pers_lock);
  80
  81 static struct kobj_type md_ktype;
  82
  83 struct md_cluster_operations *md_cluster_ops;
  84 EXPORT_SYMBOL(md_cluster_ops);
  85 static struct module *md_cluster_mod;
  86
  87 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  88 static struct workqueue_struct *md_wq;
  89 static struct workqueue_struct *md_misc_wq;
  90 static struct workqueue_struct *md_rdev_misc_wq;
  91
  92 static int remove_and_add_spares(struct mddev *mddev,
  93                                  struct md_rdev *this);
  94 static void mddev_detach(struct mddev *mddev);
  95
  96 enum md_ro_state {
  97         MD_RDWR,
  98         MD_RDONLY,
  99         MD_AUTO_READ,
 100         MD_MAX_STATE
 101 };
 102
 103 static bool md_is_rdwr(struct mddev *mddev)
 104 {
 105         return (mddev->ro == MD_RDWR);
 106 }
 107
 108 /*
 109  * Default number of read corrections we'll attempt on an rdev
 110  * before ejecting it from the array. We divide the read error
 111  * count by 2 for every hour elapsed between read errors.
 112  */
 113 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
 114 /* Default safemode delay: 200 msec */
 115 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
 116 /*
 117  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
 118  * is 1000 KB/sec, so the extra system load does not show up that much.
 119  * Increase it if you want to have more _guaranteed_ speed. Note that
 120  * the RAID driver will use the maximum available bandwidth if the IO
 121  * subsystem is idle. There is also an 'absolute maximum' reconstruction
 122  * speed limit - in case reconstruction slows down your system despite
 123  * idle IO detection.
 124  *
 125  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
 126  * or /sys/block/mdX/md/sync_speed_{min,max}
 127  */
 128
 129 static int sysctl_speed_limit_min = 1000;
 130 static int sysctl_speed_limit_max = 200000;
 131 static inline int speed_min(struct mddev *mddev)
 132 {
 133         return mddev->sync_speed_min ?
 134                 mddev->sync_speed_min : sysctl_speed_limit_min;
 135 }
 136
 137 static inline int speed_max(struct mddev *mddev)
 138 {
 139         return mddev->sync_speed_max ?
 140                 mddev->sync_speed_max : sysctl_speed_limit_max;
 141 }
 142
 143 static void rdev_uninit_serial(struct md_rdev *rdev)
 144 {
 145         if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
 146                 return;
 147
 148         kvfree(rdev->serial);
 149         rdev->serial = NULL;
 150 }
 151
 152 static void rdevs_uninit_serial(struct mddev *mddev)
 153 {
 154         struct md_rdev *rdev;
 155
 156         rdev_for_each(rdev, mddev)
 157                 rdev_uninit_serial(rdev);
 158 }
 159
 160 static int rdev_init_serial(struct md_rdev *rdev)
 161 {
 162         /* serial_nums equals with BARRIER_BUCKETS_NR */
 163         int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
 164         struct serial_in_rdev *serial = NULL;
 165
 166         if (test_bit(CollisionCheck, &rdev->flags))
 167                 return 0;
 168
 169         serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
 170                           GFP_KERNEL);
 171         if (!serial)
 172                 return -ENOMEM;
 173
 174         for (i = 0; i < serial_nums; i++) {
 175                 struct serial_in_rdev *serial_tmp = &serial[i];
 176
 177                 spin_lock_init(&serial_tmp->serial_lock);
 178                 serial_tmp->serial_rb = RB_ROOT_CACHED;
 179                 init_waitqueue_head(&serial_tmp->serial_io_wait);
 180         }
 181
 182         rdev->serial = serial;
 183         set_bit(CollisionCheck, &rdev->flags);
 184
 185         return 0;
 186 }
 187
 188 static int rdevs_init_serial(struct mddev *mddev)
 189 {
 190         struct md_rdev *rdev;
 191         int ret = 0;
 192
 193         rdev_for_each(rdev, mddev) {
 194                 ret = rdev_init_serial(rdev);
 195                 if (ret)
 196                         break;
 197         }
 198
 199         /* Free all resources if pool is not existed */
 200         if (ret && !mddev->serial_info_pool)
 201                 rdevs_uninit_serial(mddev);
 202
 203         return ret;
 204 }
 205
 206 /*
 207  * rdev needs to enable serial stuffs if it meets the conditions:
 208  * 1. it is multi-queue device flaged with writemostly.
 209  * 2. the write-behind mode is enabled.
 210  */
 211 static int rdev_need_serial(struct md_rdev *rdev)
 212 {
 213         return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
 214                 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
 215                 test_bit(WriteMostly, &rdev->flags));
 216 }
 217
 218 /*
 219  * Init resource for rdev(s), then create serial_info_pool if:
 220  * 1. rdev is the first device which return true from rdev_enable_serial.
 221  * 2. rdev is NULL, means we want to enable serialization for all rdevs.
 222  */
 223 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
 224                               bool is_suspend)
 225 {
 226         int ret = 0;
 227
 228         if (rdev && !rdev_need_serial(rdev) &&
 229             !test_bit(CollisionCheck, &rdev->flags))
 230                 return;
 231
 232         if (!is_suspend)
 233                 mddev_suspend(mddev);
 234
 235         if (!rdev)
 236                 ret = rdevs_init_serial(mddev);
 237         else
 238                 ret = rdev_init_serial(rdev);
 239         if (ret)
 240                 goto abort;
 241
 242         if (mddev->serial_info_pool == NULL) {
 243                 /*
 244                  * already in memalloc noio context by
 245                  * mddev_suspend()
 246                  */
 247                 mddev->serial_info_pool =
 248                         mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
 249                                                 sizeof(struct serial_info));
 250                 if (!mddev->serial_info_pool) {
 251                         rdevs_uninit_serial(mddev);
 252                         pr_err("can't alloc memory pool for serialization\n");
 253                 }
 254         }
 255
 256 abort:
 257         if (!is_suspend)
 258                 mddev_resume(mddev);
 259 }
 260
 261 /*
 262  * Free resource from rdev(s), and destroy serial_info_pool under conditions:
 263  * 1. rdev is the last device flaged with CollisionCheck.
 264  * 2. when bitmap is destroyed while policy is not enabled.
 265  * 3. for disable policy, the pool is destroyed only when no rdev needs it.
 266  */
 267 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
 268                                bool is_suspend)
 269 {
 270         if (rdev && !test_bit(CollisionCheck, &rdev->flags))
 271                 return;
 272
 273         if (mddev->serial_info_pool) {
 274                 struct md_rdev *temp;
 275                 int num = 0; /* used to track if other rdevs need the pool */
 276
 277                 if (!is_suspend)
 278                         mddev_suspend(mddev);
 279                 rdev_for_each(temp, mddev) {
 280                         if (!rdev) {
 281                                 if (!mddev->serialize_policy ||
 282                                     !rdev_need_serial(temp))
 283                                         rdev_uninit_serial(temp);
 284                                 else
 285                                         num++;
 286                         } else if (temp != rdev &&
 287                                    test_bit(CollisionCheck, &temp->flags))
 288                                 num++;
 289                 }
 290
 291                 if (rdev)
 292                         rdev_uninit_serial(rdev);
 293
 294                 if (num)
 295                         pr_info("The mempool could be used by other devices\n");
 296                 else {
 297                         mempool_destroy(mddev->serial_info_pool);
 298                         mddev->serial_info_pool = NULL;
 299                 }
 300                 if (!is_suspend)
 301                         mddev_resume(mddev);
 302         }
 303 }
 304
 305 static struct ctl_table_header *raid_table_header;
 306
 307 static struct ctl_table raid_table[] = {
 308         {
 309                 .procname       = "speed_limit_min",
 310                 .data           = &sysctl_speed_limit_min,
 311                 .maxlen         = sizeof(int),
 312                 .mode           = S_IRUGO|S_IWUSR,
 313                 .proc_handler   = proc_dointvec,
 314         },
 315         {
 316                 .procname       = "speed_limit_max",
 317                 .data           = &sysctl_speed_limit_max,
 318                 .maxlen         = sizeof(int),
 319                 .mode           = S_IRUGO|S_IWUSR,
 320                 .proc_handler   = proc_dointvec,
 321         },
 322         { }
 323 };
 324
 325 static struct ctl_table raid_dir_table[] = {
 326         {
 327                 .procname       = "raid",
 328                 .maxlen         = 0,
 329                 .mode           = S_IRUGO|S_IXUGO,
 330                 .child          = raid_table,
 331         },
 332         { }
 333 };
 334
 335 static struct ctl_table raid_root_table[] = {
 336         {
 337                 .procname       = "dev",
 338                 .maxlen         = 0,
 339                 .mode           = 0555,
 340                 .child          = raid_dir_table,
 341         },
 342         {  }
 343 };
 344
 345 static int start_readonly;
 346
 347 /*
 348  * The original mechanism for creating an md device is to create
 349  * a device node in /dev and to open it.  This causes races with device-close.
 350  * The preferred method is to write to the "new_array" module parameter.
 351  * This can avoid races.
 352  * Setting create_on_open to false disables the original mechanism
 353  * so all the races disappear.
 354  */
 355 static bool create_on_open = true;
 356
 357 /*
 358  * We have a system wide 'event count' that is incremented
 359  * on any 'interesting' event, and readers of /proc/mdstat
 360  * can use 'poll' or 'select' to find out when the event
 361  * count increases.
 362  *
 363  * Events are:
 364  *  start array, stop array, error, add device, remove device,
 365  *  start build, activate spare
 366  */
 367 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 368 static atomic_t md_event_count;
 369 void md_new_event(void)
 370 {
 371         atomic_inc(&md_event_count);
 372         wake_up(&md_event_waiters);
 373 }
 374 EXPORT_SYMBOL_GPL(md_new_event);
 375
 376 /*
 377  * Enables to iterate over all existing md arrays
 378  * all_mddevs_lock protects this list.
 379  */
 380 static LIST_HEAD(all_mddevs);
 381 static DEFINE_SPINLOCK(all_mddevs_lock);
 382
 383 static bool is_md_suspended(struct mddev *mddev)
 384 {
 385         return percpu_ref_is_dying(&mddev->active_io);
 386 }
 387 /* Rather than calling directly into the personality make_request function,
 388  * IO requests come here first so that we can check if the device is
 389  * being suspended pending a reconfiguration.
 390  * We hold a refcount over the call to ->make_request.  By the time that
 391  * call has finished, the bio has been linked into some internal structure
 392  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 393  */
 394 static bool is_suspended(struct mddev *mddev, struct bio *bio)
 395 {
 396         if (is_md_suspended(mddev))
 397                 return true;
 398         if (bio_data_dir(bio) != WRITE)
 399                 return false;
 400         if (mddev->suspend_lo >= mddev->suspend_hi)
 401                 return false;
 402         if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
 403                 return false;
 404         if (bio_end_sector(bio) < mddev->suspend_lo)
 405                 return false;
 406         return true;
 407 }
 408
 409 void md_handle_request(struct mddev *mddev, struct bio *bio)
 410 {
 411 check_suspended:
 412         if (is_suspended(mddev, bio)) {
 413                 DEFINE_WAIT(__wait);
 414                 /* Bail out if REQ_NOWAIT is set for the bio */
 415                 if (bio->bi_opf & REQ_NOWAIT) {
 416                         bio_wouldblock_error(bio);
 417                         return;
 418                 }
 419                 for (;;) {
 420                         prepare_to_wait(&mddev->sb_wait, &__wait,
 421                                         TASK_UNINTERRUPTIBLE);
 422                         if (!is_suspended(mddev, bio))
 423                                 break;
 424                         schedule();
 425                 }
 426                 finish_wait(&mddev->sb_wait, &__wait);
 427         }
 428         if (!percpu_ref_tryget_live(&mddev->active_io))
 429                 goto check_suspended;
 430
 431         if (!mddev->pers->make_request(mddev, bio)) {
 432                 percpu_ref_put(&mddev->active_io);
 433                 goto check_suspended;
 434         }
 435
 436         percpu_ref_put(&mddev->active_io);
 437 }
 438 EXPORT_SYMBOL(md_handle_request);
 439
 440 static void md_submit_bio(struct bio *bio)
 441 {
 442         const int rw = bio_data_dir(bio);
 443         struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
 444
 445         if (mddev == NULL || mddev->pers == NULL) {
 446                 bio_io_error(bio);
 447                 return;
 448         }
 449
 450         if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
 451                 bio_io_error(bio);
 452                 return;
 453         }
 454
 455         bio = bio_split_to_limits(bio);
 456         if (!bio)
 457                 return;
 458
 459         if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
 460                 if (bio_sectors(bio) != 0)
 461                         bio->bi_status = BLK_STS_IOERR;
 462                 bio_endio(bio);
 463                 return;
 464         }
 465
 466         /* bio could be mergeable after passing to underlayer */
 467         bio->bi_opf &= ~REQ_NOMERGE;
 468
 469         md_handle_request(mddev, bio);
 470 }
 471
 472 /* mddev_suspend makes sure no new requests are submitted
 473  * to the device, and that any requests that have been submitted
 474  * are completely handled.
 475  * Once mddev_detach() is called and completes, the module will be
 476  * completely unused.
 477  */
 478 void mddev_suspend(struct mddev *mddev)
 479 {
 480         WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
 481         lockdep_assert_held(&mddev->reconfig_mutex);
 482         if (mddev->suspended++)
 483                 return;
 484         wake_up(&mddev->sb_wait);
 485         set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
 486         percpu_ref_kill(&mddev->active_io);
 487         wait_event(mddev->sb_wait, percpu_ref_is_zero(&mddev->active_io));
 488         mddev->pers->quiesce(mddev, 1);
 489         clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
 490         wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
 491
 492         del_timer_sync(&mddev->safemode_timer);
 493         /* restrict memory reclaim I/O during raid array is suspend */
 494         mddev->noio_flag = memalloc_noio_save();
 495 }
 496 EXPORT_SYMBOL_GPL(mddev_suspend);
 497
 498 void mddev_resume(struct mddev *mddev)
 499 {
 500         /* entred the memalloc scope from mddev_suspend() */
 501         memalloc_noio_restore(mddev->noio_flag);
 502         lockdep_assert_held(&mddev->reconfig_mutex);
 503         if (--mddev->suspended)
 504                 return;
 505         percpu_ref_resurrect(&mddev->active_io);
 506         wake_up(&mddev->sb_wait);
 507         mddev->pers->quiesce(mddev, 0);
 508
 509         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 510         md_wakeup_thread(mddev->thread);
 511         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 512 }
 513 EXPORT_SYMBOL_GPL(mddev_resume);
 514
 515 /*
 516  * Generic flush handling for md
 517  */
 518
 519 static void md_end_flush(struct bio *bio)
 520 {
 521         struct md_rdev *rdev = bio->bi_private;
 522         struct mddev *mddev = rdev->mddev;
 523
 524         bio_put(bio);
 525
 526         rdev_dec_pending(rdev, mddev);
 527
 528         if (atomic_dec_and_test(&mddev->flush_pending)) {
 529                 /* The pre-request flush has finished */
 530                 queue_work(md_wq, &mddev->flush_work);
 531         }
 532 }
 533
 534 static void md_submit_flush_data(struct work_struct *ws);
 535
 536 static void submit_flushes(struct work_struct *ws)
 537 {
 538         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 539         struct md_rdev *rdev;
 540
 541         mddev->start_flush = ktime_get_boottime();
 542         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 543         atomic_set(&mddev->flush_pending, 1);
 544         rcu_read_lock();
 545         rdev_for_each_rcu(rdev, mddev)
 546                 if (rdev->raid_disk >= 0 &&
 547                     !test_bit(Faulty, &rdev->flags)) {
 548                         /* Take two references, one is dropped
 549                          * when request finishes, one after
 550                          * we reclaim rcu_read_lock
 551                          */
 552                         struct bio *bi;
 553                         atomic_inc(&rdev->nr_pending);
 554                         atomic_inc(&rdev->nr_pending);
 555                         rcu_read_unlock();
 556                         bi = bio_alloc_bioset(rdev->bdev, 0,
 557                                               REQ_OP_WRITE | REQ_PREFLUSH,
 558                                               GFP_NOIO, &mddev->bio_set);
 559                         bi->bi_end_io = md_end_flush;
 560                         bi->bi_private = rdev;
 561                         atomic_inc(&mddev->flush_pending);
 562                         submit_bio(bi);
 563                         rcu_read_lock();
 564                         rdev_dec_pending(rdev, mddev);
 565                 }
 566         rcu_read_unlock();
 567         if (atomic_dec_and_test(&mddev->flush_pending))
 568                 queue_work(md_wq, &mddev->flush_work);
 569 }
 570
 571 static void md_submit_flush_data(struct work_struct *ws)
 572 {
 573         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 574         struct bio *bio = mddev->flush_bio;
 575
 576         /*
 577          * must reset flush_bio before calling into md_handle_request to avoid a
 578          * deadlock, because other bios passed md_handle_request suspend check
 579          * could wait for this and below md_handle_request could wait for those
 580          * bios because of suspend check
 581          */
 582         spin_lock_irq(&mddev->lock);
 583         mddev->prev_flush_start = mddev->start_flush;
 584         mddev->flush_bio = NULL;
 585         spin_unlock_irq(&mddev->lock);
 586         wake_up(&mddev->sb_wait);
 587
 588         if (bio->bi_iter.bi_size == 0) {
 589                 /* an empty barrier - all done */
 590                 bio_endio(bio);
 591         } else {
 592                 bio->bi_opf &= ~REQ_PREFLUSH;
 593                 md_handle_request(mddev, bio);
 594         }
 595 }
 596
 597 /*
 598  * Manages consolidation of flushes and submitting any flushes needed for
 599  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
 600  * being finished in another context.  Returns false if the flushing is
 601  * complete but still needs the I/O portion of the bio to be processed.
 602  */
 603 bool md_flush_request(struct mddev *mddev, struct bio *bio)
 604 {
 605         ktime_t req_start = ktime_get_boottime();
 606         spin_lock_irq(&mddev->lock);
 607         /* flush requests wait until ongoing flush completes,
 608          * hence coalescing all the pending requests.
 609          */
 610         wait_event_lock_irq(mddev->sb_wait,
 611                             !mddev->flush_bio ||
 612                             ktime_before(req_start, mddev->prev_flush_start),
 613                             mddev->lock);
 614         /* new request after previous flush is completed */
 615         if (ktime_after(req_start, mddev->prev_flush_start)) {
 616                 WARN_ON(mddev->flush_bio);
 617                 mddev->flush_bio = bio;
 618                 bio = NULL;
 619         }
 620         spin_unlock_irq(&mddev->lock);
 621
 622         if (!bio) {
 623                 INIT_WORK(&mddev->flush_work, submit_flushes);
 624                 queue_work(md_wq, &mddev->flush_work);
 625         } else {
 626                 /* flush was performed for some other bio while we waited. */
 627                 if (bio->bi_iter.bi_size == 0)
 628                         /* an empty barrier - all done */
 629                         bio_endio(bio);
 630                 else {
 631                         bio->bi_opf &= ~REQ_PREFLUSH;
 632                         return false;
 633                 }
 634         }
 635         return true;
 636 }
 637 EXPORT_SYMBOL(md_flush_request);
 638
 639 static inline struct mddev *mddev_get(struct mddev *mddev)
 640 {
 641         lockdep_assert_held(&all_mddevs_lock);
 642
 643         if (test_bit(MD_DELETED, &mddev->flags))
 644                 return NULL;
 645         atomic_inc(&mddev->active);
 646         return mddev;
 647 }
 648
 649 static void mddev_delayed_delete(struct work_struct *ws);
 650
 651 void mddev_put(struct mddev *mddev)
 652 {
 653         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 654                 return;
 655         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 656             mddev->ctime == 0 && !mddev->hold_active) {
 657                 /* Array is not configured at all, and not held active,
 658                  * so destroy it */
 659                 set_bit(MD_DELETED, &mddev->flags);
 660
 661                 /*
 662                  * Call queue_work inside the spinlock so that
 663                  * flush_workqueue() after mddev_find will succeed in waiting
 664                  * for the work to be done.
 665                  */
 666                 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 667                 queue_work(md_misc_wq, &mddev->del_work);
 668         }
 669         spin_unlock(&all_mddevs_lock);
 670 }
 671
 672 static void md_safemode_timeout(struct timer_list *t);
 673
 674 void mddev_init(struct mddev *mddev)
 675 {
 676         mutex_init(&mddev->open_mutex);
 677         mutex_init(&mddev->reconfig_mutex);
 678         mutex_init(&mddev->bitmap_info.mutex);
 679         INIT_LIST_HEAD(&mddev->disks);
 680         INIT_LIST_HEAD(&mddev->all_mddevs);
 681         timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
 682         atomic_set(&mddev->active, 1);
 683         atomic_set(&mddev->openers, 0);
 684         spin_lock_init(&mddev->lock);
 685         atomic_set(&mddev->flush_pending, 0);
 686         init_waitqueue_head(&mddev->sb_wait);
 687         init_waitqueue_head(&mddev->recovery_wait);
 688         mddev->reshape_position = MaxSector;
 689         mddev->reshape_backwards = 0;
 690         mddev->last_sync_action = "none";
 691         mddev->resync_min = 0;
 692         mddev->resync_max = MaxSector;
 693         mddev->level = LEVEL_NONE;
 694 }
 695 EXPORT_SYMBOL_GPL(mddev_init);
 696
 697 static struct mddev *mddev_find_locked(dev_t unit)
 698 {
 699         struct mddev *mddev;
 700
 701         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 702                 if (mddev->unit == unit)
 703                         return mddev;
 704
 705         return NULL;
 706 }
 707
 708 /* find an unused unit number */
 709 static dev_t mddev_alloc_unit(void)
 710 {
 711         static int next_minor = 512;
 712         int start = next_minor;
 713         bool is_free = 0;
 714         dev_t dev = 0;
 715
 716         while (!is_free) {
 717                 dev = MKDEV(MD_MAJOR, next_minor);
 718                 next_minor++;
 719                 if (next_minor > MINORMASK)
 720                         next_minor = 0;
 721                 if (next_minor == start)
 722                         return 0;               /* Oh dear, all in use. */
 723                 is_free = !mddev_find_locked(dev);
 724         }
 725
 726         return dev;
 727 }
 728
 729 static struct mddev *mddev_alloc(dev_t unit)
 730 {
 731         struct mddev *new;
 732         int error;
 733
 734         if (unit && MAJOR(unit) != MD_MAJOR)
 735                 unit &= ~((1 << MdpMinorShift) - 1);
 736
 737         new = kzalloc(sizeof(*new), GFP_KERNEL);
 738         if (!new)
 739                 return ERR_PTR(-ENOMEM);
 740         mddev_init(new);
 741
 742         spin_lock(&all_mddevs_lock);
 743         if (unit) {
 744                 error = -EEXIST;
 745                 if (mddev_find_locked(unit))
 746                         goto out_free_new;
 747                 new->unit = unit;
 748                 if (MAJOR(unit) == MD_MAJOR)
 749                         new->md_minor = MINOR(unit);
 750                 else
 751                         new->md_minor = MINOR(unit) >> MdpMinorShift;
 752                 new->hold_active = UNTIL_IOCTL;
 753         } else {
 754                 error = -ENODEV;
 755                 new->unit = mddev_alloc_unit();
 756                 if (!new->unit)
 757                         goto out_free_new;
 758                 new->md_minor = MINOR(new->unit);
 759                 new->hold_active = UNTIL_STOP;
 760         }
 761
 762         list_add(&new->all_mddevs, &all_mddevs);
 763         spin_unlock(&all_mddevs_lock);
 764         return new;
 765 out_free_new:
 766         spin_unlock(&all_mddevs_lock);
 767         kfree(new);
 768         return ERR_PTR(error);
 769 }
 770
 771 static void mddev_free(struct mddev *mddev)
 772 {
 773         spin_lock(&all_mddevs_lock);
 774         list_del(&mddev->all_mddevs);
 775         spin_unlock(&all_mddevs_lock);
 776
 777         kfree(mddev);
 778 }
 779
 780 static const struct attribute_group md_redundancy_group;
 781
 782 void mddev_unlock(struct mddev *mddev)
 783 {
 784         if (mddev->to_remove) {
 785                 /* These cannot be removed under reconfig_mutex as
 786                  * an access to the files will try to take reconfig_mutex
 787                  * while holding the file unremovable, which leads to
 788                  * a deadlock.
 789                  * So hold set sysfs_active while the remove in happeing,
 790                  * and anything else which might set ->to_remove or my
 791                  * otherwise change the sysfs namespace will fail with
 792                  * -EBUSY if sysfs_active is still set.
 793                  * We set sysfs_active under reconfig_mutex and elsewhere
 794                  * test it under the same mutex to ensure its correct value
 795                  * is seen.
 796                  */
 797                 const struct attribute_group *to_remove = mddev->to_remove;
 798                 mddev->to_remove = NULL;
 799                 mddev->sysfs_active = 1;
 800                 mutex_unlock(&mddev->reconfig_mutex);
 801
 802                 if (mddev->kobj.sd) {
 803                         if (to_remove != &md_redundancy_group)
 804                                 sysfs_remove_group(&mddev->kobj, to_remove);
 805                         if (mddev->pers == NULL ||
 806                             mddev->pers->sync_request == NULL) {
 807                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 808                                 if (mddev->sysfs_action)
 809                                         sysfs_put(mddev->sysfs_action);
 810                                 if (mddev->sysfs_completed)
 811                                         sysfs_put(mddev->sysfs_completed);
 812                                 if (mddev->sysfs_degraded)
 813                                         sysfs_put(mddev->sysfs_degraded);
 814                                 mddev->sysfs_action = NULL;
 815                                 mddev->sysfs_completed = NULL;
 816                                 mddev->sysfs_degraded = NULL;
 817                         }
 818                 }
 819                 mddev->sysfs_active = 0;
 820         } else
 821                 mutex_unlock(&mddev->reconfig_mutex);
 822
 823         /* As we've dropped the mutex we need a spinlock to
 824          * make sure the thread doesn't disappear
 825          */
 826         spin_lock(&pers_lock);
 827         md_wakeup_thread(mddev->thread);
 828         wake_up(&mddev->sb_wait);
 829         spin_unlock(&pers_lock);
 830 }
 831 EXPORT_SYMBOL_GPL(mddev_unlock);
 832
 833 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
 834 {
 835         struct md_rdev *rdev;
 836
 837         rdev_for_each_rcu(rdev, mddev)
 838                 if (rdev->desc_nr == nr)
 839                         return rdev;
 840
 841         return NULL;
 842 }
 843 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
 844
 845 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 846 {
 847         struct md_rdev *rdev;
 848
 849         rdev_for_each(rdev, mddev)
 850                 if (rdev->bdev->bd_dev == dev)
 851                         return rdev;
 852
 853         return NULL;
 854 }
 855
 856 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
 857 {
 858         struct md_rdev *rdev;
 859
 860         rdev_for_each_rcu(rdev, mddev)
 861                 if (rdev->bdev->bd_dev == dev)
 862                         return rdev;
 863
 864         return NULL;
 865 }
 866 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
 867
 868 static struct md_personality *find_pers(int level, char *clevel)
 869 {
 870         struct md_personality *pers;
 871         list_for_each_entry(pers, &pers_list, list) {
 872                 if (level != LEVEL_NONE && pers->level == level)
 873                         return pers;
 874                 if (strcmp(pers->name, clevel)==0)
 875                         return pers;
 876         }
 877         return NULL;
 878 }
 879
 880 /* return the offset of the super block in 512byte sectors */
 881 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 882 {
 883         return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
 884 }
 885
 886 static int alloc_disk_sb(struct md_rdev *rdev)
 887 {
 888         rdev->sb_page = alloc_page(GFP_KERNEL);
 889         if (!rdev->sb_page)
 890                 return -ENOMEM;
 891         return 0;
 892 }
 893
 894 void md_rdev_clear(struct md_rdev *rdev)
 895 {
 896         if (rdev->sb_page) {
 897                 put_page(rdev->sb_page);
 898                 rdev->sb_loaded = 0;
 899                 rdev->sb_page = NULL;
 900                 rdev->sb_start = 0;
 901                 rdev->sectors = 0;
 902         }
 903         if (rdev->bb_page) {
 904                 put_page(rdev->bb_page);
 905                 rdev->bb_page = NULL;
 906         }
 907         badblocks_exit(&rdev->badblocks);
 908 }
 909 EXPORT_SYMBOL_GPL(md_rdev_clear);
 910
 911 static void super_written(struct bio *bio)
 912 {
 913         struct md_rdev *rdev = bio->bi_private;
 914         struct mddev *mddev = rdev->mddev;
 915
 916         if (bio->bi_status) {
 917                 pr_err("md: %s gets error=%d\n", __func__,
 918                        blk_status_to_errno(bio->bi_status));
 919                 md_error(mddev, rdev);
 920                 if (!test_bit(Faulty, &rdev->flags)
 921                     && (bio->bi_opf & MD_FAILFAST)) {
 922                         set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
 923                         set_bit(LastDev, &rdev->flags);
 924                 }
 925         } else
 926                 clear_bit(LastDev, &rdev->flags);
 927
 928         bio_put(bio);
 929
 930         rdev_dec_pending(rdev, mddev);
 931
 932         if (atomic_dec_and_test(&mddev->pending_writes))
 933                 wake_up(&mddev->sb_wait);
 934 }
 935
 936 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 937                    sector_t sector, int size, struct page *page)
 938 {
 939         /* write first size bytes of page to sector of rdev
 940          * Increment mddev->pending_writes before returning
 941          * and decrement it on completion, waking up sb_wait
 942          * if zero is reached.
 943          * If an error occurred, call md_error
 944          */
 945         struct bio *bio;
 946
 947         if (!page)
 948                 return;
 949
 950         if (test_bit(Faulty, &rdev->flags))
 951                 return;
 952
 953         bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
 954                                1,
 955                                REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA,
 956                                GFP_NOIO, &mddev->sync_set);
 957
 958         atomic_inc(&rdev->nr_pending);
 959
 960         bio->bi_iter.bi_sector = sector;
 961         bio_add_page(bio, page, size, 0);
 962         bio->bi_private = rdev;
 963         bio->bi_end_io = super_written;
 964
 965         if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
 966             test_bit(FailFast, &rdev->flags) &&
 967             !test_bit(LastDev, &rdev->flags))
 968                 bio->bi_opf |= MD_FAILFAST;
 969
 970         atomic_inc(&mddev->pending_writes);
 971         submit_bio(bio);
 972 }
 973
 974 int md_super_wait(struct mddev *mddev)
 975 {
 976         /* wait for all superblock writes that were scheduled to complete */
 977         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
 978         if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
 979                 return -EAGAIN;
 980         return 0;
 981 }
 982
 983 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
 984                  struct page *page, blk_opf_t opf, bool metadata_op)
 985 {
 986         struct bio bio;
 987         struct bio_vec bvec;
 988
 989         if (metadata_op && rdev->meta_bdev)
 990                 bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
 991         else
 992                 bio_init(&bio, rdev->bdev, &bvec, 1, opf);
 993
 994         if (metadata_op)
 995                 bio.bi_iter.bi_sector = sector + rdev->sb_start;
 996         else if (rdev->mddev->reshape_position != MaxSector &&
 997                  (rdev->mddev->reshape_backwards ==
 998                   (sector >= rdev->mddev->reshape_position)))
 999                 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1000         else
1001                 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1002         bio_add_page(&bio, page, size, 0);
1003
1004         submit_bio_wait(&bio);
1005
1006         return !bio.bi_status;
1007 }
1008 EXPORT_SYMBOL_GPL(sync_page_io);
1009
1010 static int read_disk_sb(struct md_rdev *rdev, int size)
1011 {
1012         if (rdev->sb_loaded)
1013                 return 0;
1014
1015         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1016                 goto fail;
1017         rdev->sb_loaded = 1;
1018         return 0;
1019
1020 fail:
1021         pr_err("md: disabled device %pg, could not read superblock.\n",
1022                rdev->bdev);
1023         return -EINVAL;
1024 }
1025
1026 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1027 {
1028         return  sb1->set_uuid0 == sb2->set_uuid0 &&
1029                 sb1->set_uuid1 == sb2->set_uuid1 &&
1030                 sb1->set_uuid2 == sb2->set_uuid2 &&
1031                 sb1->set_uuid3 == sb2->set_uuid3;
1032 }
1033
1034 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1035 {
1036         int ret;
1037         mdp_super_t *tmp1, *tmp2;
1038
1039         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1040         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1041
1042         if (!tmp1 || !tmp2) {
1043                 ret = 0;
1044                 goto abort;
1045         }
1046
1047         *tmp1 = *sb1;
1048         *tmp2 = *sb2;
1049
1050         /*
1051          * nr_disks is not constant
1052          */
1053         tmp1->nr_disks = 0;
1054         tmp2->nr_disks = 0;
1055
1056         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1057 abort:
1058         kfree(tmp1);
1059         kfree(tmp2);
1060         return ret;
1061 }
1062
1063 static u32 md_csum_fold(u32 csum)
1064 {
1065         csum = (csum & 0xffff) + (csum >> 16);
1066         return (csum & 0xffff) + (csum >> 16);
1067 }
1068
1069 static unsigned int calc_sb_csum(mdp_super_t *sb)
1070 {
1071         u64 newcsum = 0;
1072         u32 *sb32 = (u32*)sb;
1073         int i;
1074         unsigned int disk_csum, csum;
1075
1076         disk_csum = sb->sb_csum;
1077         sb->sb_csum = 0;
1078
1079         for (i = 0; i < MD_SB_BYTES/4 ; i++)
1080                 newcsum += sb32[i];
1081         csum = (newcsum & 0xffffffff) + (newcsum>>32);
1082
1083 #ifdef CONFIG_ALPHA
1084         /* This used to use csum_partial, which was wrong for several
1085          * reasons including that different results are returned on
1086          * different architectures.  It isn't critical that we get exactly
1087          * the same return value as before (we always csum_fold before
1088          * testing, and that removes any differences).  However as we
1089          * know that csum_partial always returned a 16bit value on
1090          * alphas, do a fold to maximise conformity to previous behaviour.
1091          */
1092         sb->sb_csum = md_csum_fold(disk_csum);
1093 #else
1094         sb->sb_csum = disk_csum;
1095 #endif
1096         return csum;
1097 }
1098
1099 /*
1100  * Handle superblock details.
1101  * We want to be able to handle multiple superblock formats
1102  * so we have a common interface to them all, and an array of
1103  * different handlers.
1104  * We rely on user-space to write the initial superblock, and support
1105  * reading and updating of superblocks.
1106  * Interface methods are:
1107  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1108  *      loads and validates a superblock on dev.
1109  *      if refdev != NULL, compare superblocks on both devices
1110  *    Return:
1111  *      0 - dev has a superblock that is compatible with refdev
1112  *      1 - dev has a superblock that is compatible and newer than refdev
1113  *          so dev should be used as the refdev in future
1114  *     -EINVAL superblock incompatible or invalid
1115  *     -othererror e.g. -EIO
1116  *
1117  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1118  *      Verify that dev is acceptable into mddev.
1119  *       The first time, mddev->raid_disks will be 0, and data from
1120  *       dev should be merged in.  Subsequent calls check that dev
1121  *       is new enough.  Return 0 or -EINVAL
1122  *
1123  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1124  *     Update the superblock for rdev with data in mddev
1125  *     This does not write to disc.
1126  *
1127  */
1128
1129 struct super_type  {
1130         char                *name;
1131         struct module       *owner;
1132         int                 (*load_super)(struct md_rdev *rdev,
1133                                           struct md_rdev *refdev,
1134                                           int minor_version);
1135         int                 (*validate_super)(struct mddev *mddev,
1136                                               struct md_rdev *rdev);
1137         void                (*sync_super)(struct mddev *mddev,
1138                                           struct md_rdev *rdev);
1139         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1140                                                 sector_t num_sectors);
1141         int                 (*allow_new_offset)(struct md_rdev *rdev,
1142                                                 unsigned long long new_offset);
1143 };
1144
1145 /*
1146  * Check that the given mddev has no bitmap.
1147  *
1148  * This function is called from the run method of all personalities that do not
1149  * support bitmaps. It prints an error message and returns non-zero if mddev
1150  * has a bitmap. Otherwise, it returns 0.
1151  *
1152  */
1153 int md_check_no_bitmap(struct mddev *mddev)
1154 {
1155         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1156                 return 0;
1157         pr_warn("%s: bitmaps are not supported for %s\n",
1158                 mdname(mddev), mddev->pers->name);
1159         return 1;
1160 }
1161 EXPORT_SYMBOL(md_check_no_bitmap);
1162
1163 /*
1164  * load_super for 0.90.0
1165  */
1166 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1167 {
1168         mdp_super_t *sb;
1169         int ret;
1170         bool spare_disk = true;
1171
1172         /*
1173          * Calculate the position of the superblock (512byte sectors),
1174          * it's at the end of the disk.
1175          *
1176          * It also happens to be a multiple of 4Kb.
1177          */
1178         rdev->sb_start = calc_dev_sboffset(rdev);
1179
1180         ret = read_disk_sb(rdev, MD_SB_BYTES);
1181         if (ret)
1182                 return ret;
1183
1184         ret = -EINVAL;
1185
1186         sb = page_address(rdev->sb_page);
1187
1188         if (sb->md_magic != MD_SB_MAGIC) {
1189                 pr_warn("md: invalid raid superblock magic on %pg\n",
1190                         rdev->bdev);
1191                 goto abort;
1192         }
1193
1194         if (sb->major_version != 0 ||
1195             sb->minor_version < 90 ||
1196             sb->minor_version > 91) {
1197                 pr_warn("Bad version number %d.%d on %pg\n",
1198                         sb->major_version, sb->minor_version, rdev->bdev);
1199                 goto abort;
1200         }
1201
1202         if (sb->raid_disks <= 0)
1203                 goto abort;
1204
1205         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1206                 pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1207                 goto abort;
1208         }
1209
1210         rdev->preferred_minor = sb->md_minor;
1211         rdev->data_offset = 0;
1212         rdev->new_data_offset = 0;
1213         rdev->sb_size = MD_SB_BYTES;
1214         rdev->badblocks.shift = -1;
1215
1216         if (sb->level == LEVEL_MULTIPATH)
1217                 rdev->desc_nr = -1;
1218         else
1219                 rdev->desc_nr = sb->this_disk.number;
1220
1221         /* not spare disk, or LEVEL_MULTIPATH */
1222         if (sb->level == LEVEL_MULTIPATH ||
1223                 (rdev->desc_nr >= 0 &&
1224                  rdev->desc_nr < MD_SB_DISKS &&
1225                  sb->disks[rdev->desc_nr].state &
1226                  ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1227                 spare_disk = false;
1228
1229         if (!refdev) {
1230                 if (!spare_disk)
1231                         ret = 1;
1232                 else
1233                         ret = 0;
1234         } else {
1235                 __u64 ev1, ev2;
1236                 mdp_super_t *refsb = page_address(refdev->sb_page);
1237                 if (!md_uuid_equal(refsb, sb)) {
1238                         pr_warn("md: %pg has different UUID to %pg\n",
1239                                 rdev->bdev, refdev->bdev);
1240                         goto abort;
1241                 }
1242                 if (!md_sb_equal(refsb, sb)) {
1243                         pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1244                                 rdev->bdev, refdev->bdev);
1245                         goto abort;
1246                 }
1247                 ev1 = md_event(sb);
1248                 ev2 = md_event(refsb);
1249
1250                 if (!spare_disk && ev1 > ev2)
1251                         ret = 1;
1252                 else
1253                         ret = 0;
1254         }
1255         rdev->sectors = rdev->sb_start;
1256         /* Limit to 4TB as metadata cannot record more than that.
1257          * (not needed for Linear and RAID0 as metadata doesn't
1258          * record this size)
1259          */
1260         if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1261                 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1262
1263         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1264                 /* "this cannot possibly happen" ... */
1265                 ret = -EINVAL;
1266
1267  abort:
1268         return ret;
1269 }
1270
1271 /*
1272  * validate_super for 0.90.0
1273  */
1274 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1275 {
1276         mdp_disk_t *desc;
1277         mdp_super_t *sb = page_address(rdev->sb_page);
1278         __u64 ev1 = md_event(sb);
1279
1280         rdev->raid_disk = -1;
1281         clear_bit(Faulty, &rdev->flags);
1282         clear_bit(In_sync, &rdev->flags);
1283         clear_bit(Bitmap_sync, &rdev->flags);
1284         clear_bit(WriteMostly, &rdev->flags);
1285
1286         if (mddev->raid_disks == 0) {
1287                 mddev->major_version = 0;
1288                 mddev->minor_version = sb->minor_version;
1289                 mddev->patch_version = sb->patch_version;
1290                 mddev->external = 0;
1291                 mddev->chunk_sectors = sb->chunk_size >> 9;
1292                 mddev->ctime = sb->ctime;
1293                 mddev->utime = sb->utime;
1294                 mddev->level = sb->level;
1295                 mddev->clevel[0] = 0;
1296                 mddev->layout = sb->layout;
1297                 mddev->raid_disks = sb->raid_disks;
1298                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1299                 mddev->events = ev1;
1300                 mddev->bitmap_info.offset = 0;
1301                 mddev->bitmap_info.space = 0;
1302                 /* bitmap can use 60 K after the 4K superblocks */
1303                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1304                 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1305                 mddev->reshape_backwards = 0;
1306
1307                 if (mddev->minor_version >= 91) {
1308                         mddev->reshape_position = sb->reshape_position;
1309                         mddev->delta_disks = sb->delta_disks;
1310                         mddev->new_level = sb->new_level;
1311                         mddev->new_layout = sb->new_layout;
1312                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1313                         if (mddev->delta_disks < 0)
1314                                 mddev->reshape_backwards = 1;
1315                 } else {
1316                         mddev->reshape_position = MaxSector;
1317                         mddev->delta_disks = 0;
1318                         mddev->new_level = mddev->level;
1319                         mddev->new_layout = mddev->layout;
1320                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1321                 }
1322                 if (mddev->level == 0)
1323                         mddev->layout = -1;
1324
1325                 if (sb->state & (1<<MD_SB_CLEAN))
1326                         mddev->recovery_cp = MaxSector;
1327                 else {
1328                         if (sb->events_hi == sb->cp_events_hi &&
1329                                 sb->events_lo == sb->cp_events_lo) {
1330                                 mddev->recovery_cp = sb->recovery_cp;
1331                         } else
1332                                 mddev->recovery_cp = 0;
1333                 }
1334
1335                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1336                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1337                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1338                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1339
1340                 mddev->max_disks = MD_SB_DISKS;
1341
1342                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1343                     mddev->bitmap_info.file == NULL) {
1344                         mddev->bitmap_info.offset =
1345                                 mddev->bitmap_info.default_offset;
1346                         mddev->bitmap_info.space =
1347                                 mddev->bitmap_info.default_space;
1348                 }
1349
1350         } else if (mddev->pers == NULL) {
1351                 /* Insist on good event counter while assembling, except
1352                  * for spares (which don't need an event count) */
1353                 ++ev1;
1354                 if (sb->disks[rdev->desc_nr].state & (
1355                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1356                         if (ev1 < mddev->events)
1357                                 return -EINVAL;
1358         } else if (mddev->bitmap) {
1359                 /* if adding to array with a bitmap, then we can accept an
1360                  * older device ... but not too old.
1361                  */
1362                 if (ev1 < mddev->bitmap->events_cleared)
1363                         return 0;
1364                 if (ev1 < mddev->events)
1365                         set_bit(Bitmap_sync, &rdev->flags);
1366         } else {
1367                 if (ev1 < mddev->events)
1368                         /* just a hot-add of a new device, leave raid_disk at -1 */
1369                         return 0;
1370         }
1371
1372         if (mddev->level != LEVEL_MULTIPATH) {
1373                 desc = sb->disks + rdev->desc_nr;
1374
1375                 if (desc->state & (1<<MD_DISK_FAULTY))
1376                         set_bit(Faulty, &rdev->flags);
1377                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1378                             desc->raid_disk < mddev->raid_disks */) {
1379                         set_bit(In_sync, &rdev->flags);
1380                         rdev->raid_disk = desc->raid_disk;
1381                         rdev->saved_raid_disk = desc->raid_disk;
1382                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1383                         /* active but not in sync implies recovery up to
1384                          * reshape position.  We don't know exactly where
1385                          * that is, so set to zero for now */
1386                         if (mddev->minor_version >= 91) {
1387                                 rdev->recovery_offset = 0;
1388                                 rdev->raid_disk = desc->raid_disk;
1389                         }
1390                 }
1391                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1392                         set_bit(WriteMostly, &rdev->flags);
1393                 if (desc->state & (1<<MD_DISK_FAILFAST))
1394                         set_bit(FailFast, &rdev->flags);
1395         } else /* MULTIPATH are always insync */
1396                 set_bit(In_sync, &rdev->flags);
1397         return 0;
1398 }
1399
1400 /*
1401  * sync_super for 0.90.0
1402  */
1403 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1404 {
1405         mdp_super_t *sb;
1406         struct md_rdev *rdev2;
1407         int next_spare = mddev->raid_disks;
1408
1409         /* make rdev->sb match mddev data..
1410          *
1411          * 1/ zero out disks
1412          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1413          * 3/ any empty disks < next_spare become removed
1414          *
1415          * disks[0] gets initialised to REMOVED because
1416          * we cannot be sure from other fields if it has
1417          * been initialised or not.
1418          */
1419         int i;
1420         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1421
1422         rdev->sb_size = MD_SB_BYTES;
1423
1424         sb = page_address(rdev->sb_page);
1425
1426         memset(sb, 0, sizeof(*sb));
1427
1428         sb->md_magic = MD_SB_MAGIC;
1429         sb->major_version = mddev->major_version;
1430         sb->patch_version = mddev->patch_version;
1431         sb->gvalid_words  = 0; /* ignored */
1432         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1433         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1434         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1435         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1436
1437         sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1438         sb->level = mddev->level;
1439         sb->size = mddev->dev_sectors / 2;
1440         sb->raid_disks = mddev->raid_disks;
1441         sb->md_minor = mddev->md_minor;
1442         sb->not_persistent = 0;
1443         sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1444         sb->state = 0;
1445         sb->events_hi = (mddev->events>>32);
1446         sb->events_lo = (u32)mddev->events;
1447
1448         if (mddev->reshape_position == MaxSector)
1449                 sb->minor_version = 90;
1450         else {
1451                 sb->minor_version = 91;
1452                 sb->reshape_position = mddev->reshape_position;
1453                 sb->new_level = mddev->new_level;
1454                 sb->delta_disks = mddev->delta_disks;
1455                 sb->new_layout = mddev->new_layout;
1456                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1457         }
1458         mddev->minor_version = sb->minor_version;
1459         if (mddev->in_sync)
1460         {
1461                 sb->recovery_cp = mddev->recovery_cp;
1462                 sb->cp_events_hi = (mddev->events>>32);
1463                 sb->cp_events_lo = (u32)mddev->events;
1464                 if (mddev->recovery_cp == MaxSector)
1465                         sb->state = (1<< MD_SB_CLEAN);
1466         } else
1467                 sb->recovery_cp = 0;
1468
1469         sb->layout = mddev->layout;
1470         sb->chunk_size = mddev->chunk_sectors << 9;
1471
1472         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1473                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1474
1475         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1476         rdev_for_each(rdev2, mddev) {
1477                 mdp_disk_t *d;
1478                 int desc_nr;
1479                 int is_active = test_bit(In_sync, &rdev2->flags);
1480
1481                 if (rdev2->raid_disk >= 0 &&
1482                     sb->minor_version >= 91)
1483                         /* we have nowhere to store the recovery_offset,
1484                          * but if it is not below the reshape_position,
1485                          * we can piggy-back on that.
1486                          */
1487                         is_active = 1;
1488                 if (rdev2->raid_disk < 0 ||
1489                     test_bit(Faulty, &rdev2->flags))
1490                         is_active = 0;
1491                 if (is_active)
1492                         desc_nr = rdev2->raid_disk;
1493                 else
1494                         desc_nr = next_spare++;
1495                 rdev2->desc_nr = desc_nr;
1496                 d = &sb->disks[rdev2->desc_nr];
1497                 nr_disks++;
1498                 d->number = rdev2->desc_nr;
1499                 d->major = MAJOR(rdev2->bdev->bd_dev);
1500                 d->minor = MINOR(rdev2->bdev->bd_dev);
1501                 if (is_active)
1502                         d->raid_disk = rdev2->raid_disk;
1503                 else
1504                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1505                 if (test_bit(Faulty, &rdev2->flags))
1506                         d->state = (1<<MD_DISK_FAULTY);
1507                 else if (is_active) {
1508                         d->state = (1<<MD_DISK_ACTIVE);
1509                         if (test_bit(In_sync, &rdev2->flags))
1510                                 d->state |= (1<<MD_DISK_SYNC);
1511                         active++;
1512                         working++;
1513                 } else {
1514                         d->state = 0;
1515                         spare++;
1516                         working++;
1517                 }
1518                 if (test_bit(WriteMostly, &rdev2->flags))
1519                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1520                 if (test_bit(FailFast, &rdev2->flags))
1521                         d->state |= (1<<MD_DISK_FAILFAST);
1522         }
1523         /* now set the "removed" and "faulty" bits on any missing devices */
1524         for (i=0 ; i < mddev->raid_disks ; i++) {
1525                 mdp_disk_t *d = &sb->disks[i];
1526                 if (d->state == 0 && d->number == 0) {
1527                         d->number = i;
1528                         d->raid_disk = i;
1529                         d->state = (1<<MD_DISK_REMOVED);
1530                         d->state |= (1<<MD_DISK_FAULTY);
1531                         failed++;
1532                 }
1533         }
1534         sb->nr_disks = nr_disks;
1535         sb->active_disks = active;
1536         sb->working_disks = working;
1537         sb->failed_disks = failed;
1538         sb->spare_disks = spare;
1539
1540         sb->this_disk = sb->disks[rdev->desc_nr];
1541         sb->sb_csum = calc_sb_csum(sb);
1542 }
1543
1544 /*
1545  * rdev_size_change for 0.90.0
1546  */
1547 static unsigned long long
1548 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1549 {
1550         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1551                 return 0; /* component must fit device */
1552         if (rdev->mddev->bitmap_info.offset)
1553                 return 0; /* can't move bitmap */
1554         rdev->sb_start = calc_dev_sboffset(rdev);
1555         if (!num_sectors || num_sectors > rdev->sb_start)
1556                 num_sectors = rdev->sb_start;
1557         /* Limit to 4TB as metadata cannot record more than that.
1558          * 4TB == 2^32 KB, or 2*2^32 sectors.
1559          */
1560         if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1561                 num_sectors = (sector_t)(2ULL << 32) - 2;
1562         do {
1563                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1564                        rdev->sb_page);
1565         } while (md_super_wait(rdev->mddev) < 0);
1566         return num_sectors;
1567 }
1568
1569 static int
1570 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1571 {
1572         /* non-zero offset changes not possible with v0.90 */
1573         return new_offset == 0;
1574 }
1575
1576 /*
1577  * version 1 superblock
1578  */
1579
1580 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1581 {
1582         __le32 disk_csum;
1583         u32 csum;
1584         unsigned long long newcsum;
1585         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1586         __le32 *isuper = (__le32*)sb;
1587
1588         disk_csum = sb->sb_csum;
1589         sb->sb_csum = 0;
1590         newcsum = 0;
1591         for (; size >= 4; size -= 4)
1592                 newcsum += le32_to_cpu(*isuper++);
1593
1594         if (size == 2)
1595                 newcsum += le16_to_cpu(*(__le16*) isuper);
1596
1597         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1598         sb->sb_csum = disk_csum;
1599         return cpu_to_le32(csum);
1600 }
1601
1602 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1603 {
1604         struct mdp_superblock_1 *sb;
1605         int ret;
1606         sector_t sb_start;
1607         sector_t sectors;
1608         int bmask;
1609         bool spare_disk = true;
1610
1611         /*
1612          * Calculate the position of the superblock in 512byte sectors.
1613          * It is always aligned to a 4K boundary and
1614          * depeding on minor_version, it can be:
1615          * 0: At least 8K, but less than 12K, from end of device
1616          * 1: At start of device
1617          * 2: 4K from start of device.
1618          */
1619         switch(minor_version) {
1620         case 0:
1621                 sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1622                 sb_start &= ~(sector_t)(4*2-1);
1623                 break;
1624         case 1:
1625                 sb_start = 0;
1626                 break;
1627         case 2:
1628                 sb_start = 8;
1629                 break;
1630         default:
1631                 return -EINVAL;
1632         }
1633         rdev->sb_start = sb_start;
1634
1635         /* superblock is rarely larger than 1K, but it can be larger,
1636          * and it is safe to read 4k, so we do that
1637          */
1638         ret = read_disk_sb(rdev, 4096);
1639         if (ret) return ret;
1640
1641         sb = page_address(rdev->sb_page);
1642
1643         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1644             sb->major_version != cpu_to_le32(1) ||
1645             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1646             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1647             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1648                 return -EINVAL;
1649
1650         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1651                 pr_warn("md: invalid superblock checksum on %pg\n",
1652                         rdev->bdev);
1653                 return -EINVAL;
1654         }
1655         if (le64_to_cpu(sb->data_size) < 10) {
1656                 pr_warn("md: data_size too small on %pg\n",
1657                         rdev->bdev);
1658                 return -EINVAL;
1659         }
1660         if (sb->pad0 ||
1661             sb->pad3[0] ||
1662             memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1663                 /* Some padding is non-zero, might be a new feature */
1664                 return -EINVAL;
1665
1666         rdev->preferred_minor = 0xffff;
1667         rdev->data_offset = le64_to_cpu(sb->data_offset);
1668         rdev->new_data_offset = rdev->data_offset;
1669         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1670             (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1671                 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1672         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1673
1674         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1675         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1676         if (rdev->sb_size & bmask)
1677                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1678
1679         if (minor_version
1680             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1681                 return -EINVAL;
1682         if (minor_version
1683             && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1684                 return -EINVAL;
1685
1686         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1687                 rdev->desc_nr = -1;
1688         else
1689                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1690
1691         if (!rdev->bb_page) {
1692                 rdev->bb_page = alloc_page(GFP_KERNEL);
1693                 if (!rdev->bb_page)
1694                         return -ENOMEM;
1695         }
1696         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1697             rdev->badblocks.count == 0) {
1698                 /* need to load the bad block list.
1699                  * Currently we limit it to one page.
1700                  */
1701                 s32 offset;
1702                 sector_t bb_sector;
1703                 __le64 *bbp;
1704                 int i;
1705                 int sectors = le16_to_cpu(sb->bblog_size);
1706                 if (sectors > (PAGE_SIZE / 512))
1707                         return -EINVAL;
1708                 offset = le32_to_cpu(sb->bblog_offset);
1709                 if (offset == 0)
1710                         return -EINVAL;
1711                 bb_sector = (long long)offset;
1712                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1713                                   rdev->bb_page, REQ_OP_READ, true))
1714                         return -EIO;
1715                 bbp = (__le64 *)page_address(rdev->bb_page);
1716                 rdev->badblocks.shift = sb->bblog_shift;
1717                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1718                         u64 bb = le64_to_cpu(*bbp);
1719                         int count = bb & (0x3ff);
1720                         u64 sector = bb >> 10;
1721                         sector <<= sb->bblog_shift;
1722                         count <<= sb->bblog_shift;
1723                         if (bb + 1 == 0)
1724                                 break;
1725                         if (badblocks_set(&rdev->badblocks, sector, count, 1))
1726                                 return -EINVAL;
1727                 }
1728         } else if (sb->bblog_offset != 0)
1729                 rdev->badblocks.shift = 0;
1730
1731         if ((le32_to_cpu(sb->feature_map) &
1732             (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1733                 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1734                 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1735                 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1736         }
1737
1738         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1739             sb->level != 0)
1740                 return -EINVAL;
1741
1742         /* not spare disk, or LEVEL_MULTIPATH */
1743         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1744                 (rdev->desc_nr >= 0 &&
1745                 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1746                 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1747                  le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1748                 spare_disk = false;
1749
1750         if (!refdev) {
1751                 if (!spare_disk)
1752                         ret = 1;
1753                 else
1754                         ret = 0;
1755         } else {
1756                 __u64 ev1, ev2;
1757                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1758
1759                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1760                     sb->level != refsb->level ||
1761                     sb->layout != refsb->layout ||
1762                     sb->chunksize != refsb->chunksize) {
1763                         pr_warn("md: %pg has strangely different superblock to %pg\n",
1764                                 rdev->bdev,
1765                                 refdev->bdev);
1766                         return -EINVAL;
1767                 }
1768                 ev1 = le64_to_cpu(sb->events);
1769                 ev2 = le64_to_cpu(refsb->events);
1770
1771                 if (!spare_disk && ev1 > ev2)
1772                         ret = 1;
1773                 else
1774                         ret = 0;
1775         }
1776         if (minor_version)
1777                 sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1778         else
1779                 sectors = rdev->sb_start;
1780         if (sectors < le64_to_cpu(sb->data_size))
1781                 return -EINVAL;
1782         rdev->sectors = le64_to_cpu(sb->data_size);
1783         return ret;
1784 }
1785
1786 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1787 {
1788         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1789         __u64 ev1 = le64_to_cpu(sb->events);
1790
1791         rdev->raid_disk = -1;
1792         clear_bit(Faulty, &rdev->flags);
1793         clear_bit(In_sync, &rdev->flags);
1794         clear_bit(Bitmap_sync, &rdev->flags);
1795         clear_bit(WriteMostly, &rdev->flags);
1796
1797         if (mddev->raid_disks == 0) {
1798                 mddev->major_version = 1;
1799                 mddev->patch_version = 0;
1800                 mddev->external = 0;
1801                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1802                 mddev->ctime = le64_to_cpu(sb->ctime);
1803                 mddev->utime = le64_to_cpu(sb->utime);
1804                 mddev->level = le32_to_cpu(sb->level);
1805                 mddev->clevel[0] = 0;
1806                 mddev->layout = le32_to_cpu(sb->layout);
1807                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1808                 mddev->dev_sectors = le64_to_cpu(sb->size);
1809                 mddev->events = ev1;
1810                 mddev->bitmap_info.offset = 0;
1811                 mddev->bitmap_info.space = 0;
1812                 /* Default location for bitmap is 1K after superblock
1813                  * using 3K - total of 4K
1814                  */
1815                 mddev->bitmap_info.default_offset = 1024 >> 9;
1816                 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1817                 mddev->reshape_backwards = 0;
1818
1819                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1820                 memcpy(mddev->uuid, sb->set_uuid, 16);
1821
1822                 mddev->max_disks =  (4096-256)/2;
1823
1824                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1825                     mddev->bitmap_info.file == NULL) {
1826                         mddev->bitmap_info.offset =
1827                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1828                         /* Metadata doesn't record how much space is available.
1829                          * For 1.0, we assume we can use up to the superblock
1830                          * if before, else to 4K beyond superblock.
1831                          * For others, assume no change is possible.
1832                          */
1833                         if (mddev->minor_version > 0)
1834                                 mddev->bitmap_info.space = 0;
1835                         else if (mddev->bitmap_info.offset > 0)
1836                                 mddev->bitmap_info.space =
1837                                         8 - mddev->bitmap_info.offset;
1838                         else
1839                                 mddev->bitmap_info.space =
1840                                         -mddev->bitmap_info.offset;
1841                 }
1842
1843                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1844                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1845                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1846                         mddev->new_level = le32_to_cpu(sb->new_level);
1847                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1848                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1849                         if (mddev->delta_disks < 0 ||
1850                             (mddev->delta_disks == 0 &&
1851                              (le32_to_cpu(sb->feature_map)
1852                               & MD_FEATURE_RESHAPE_BACKWARDS)))
1853                                 mddev->reshape_backwards = 1;
1854                 } else {
1855                         mddev->reshape_position = MaxSector;
1856                         mddev->delta_disks = 0;
1857                         mddev->new_level = mddev->level;
1858                         mddev->new_layout = mddev->layout;
1859                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1860                 }
1861
1862                 if (mddev->level == 0 &&
1863                     !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1864                         mddev->layout = -1;
1865
1866                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1867                         set_bit(MD_HAS_JOURNAL, &mddev->flags);
1868
1869                 if (le32_to_cpu(sb->feature_map) &
1870                     (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1871                         if (le32_to_cpu(sb->feature_map) &
1872                             (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1873                                 return -EINVAL;
1874                         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1875                             (le32_to_cpu(sb->feature_map) &
1876                                             MD_FEATURE_MULTIPLE_PPLS))
1877                                 return -EINVAL;
1878                         set_bit(MD_HAS_PPL, &mddev->flags);
1879                 }
1880         } else if (mddev->pers == NULL) {
1881                 /* Insist of good event counter while assembling, except for
1882                  * spares (which don't need an event count) */
1883                 ++ev1;
1884                 if (rdev->desc_nr >= 0 &&
1885                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1886                     (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1887                      le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1888                         if (ev1 < mddev->events)
1889                                 return -EINVAL;
1890         } else if (mddev->bitmap) {
1891                 /* If adding to array with a bitmap, then we can accept an
1892                  * older device, but not too old.
1893                  */
1894                 if (ev1 < mddev->bitmap->events_cleared)
1895                         return 0;
1896                 if (ev1 < mddev->events)
1897                         set_bit(Bitmap_sync, &rdev->flags);
1898         } else {
1899                 if (ev1 < mddev->events)
1900                         /* just a hot-add of a new device, leave raid_disk at -1 */
1901                         return 0;
1902         }
1903         if (mddev->level != LEVEL_MULTIPATH) {
1904                 int role;
1905                 if (rdev->desc_nr < 0 ||
1906                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1907                         role = MD_DISK_ROLE_SPARE;
1908                         rdev->desc_nr = -1;
1909                 } else
1910                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1911                 switch(role) {
1912                 case MD_DISK_ROLE_SPARE: /* spare */
1913                         break;
1914                 case MD_DISK_ROLE_FAULTY: /* faulty */
1915                         set_bit(Faulty, &rdev->flags);
1916                         break;
1917                 case MD_DISK_ROLE_JOURNAL: /* journal device */
1918                         if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1919                                 /* journal device without journal feature */
1920                                 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1921                                 return -EINVAL;
1922                         }
1923                         set_bit(Journal, &rdev->flags);
1924                         rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1925                         rdev->raid_disk = 0;
1926                         break;
1927                 default:
1928                         rdev->saved_raid_disk = role;
1929                         if ((le32_to_cpu(sb->feature_map) &
1930                              MD_FEATURE_RECOVERY_OFFSET)) {
1931                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1932                                 if (!(le32_to_cpu(sb->feature_map) &
1933                                       MD_FEATURE_RECOVERY_BITMAP))
1934                                         rdev->saved_raid_disk = -1;
1935                         } else {
1936                                 /*
1937                                  * If the array is FROZEN, then the device can't
1938                                  * be in_sync with rest of array.
1939                                  */
1940                                 if (!test_bit(MD_RECOVERY_FROZEN,
1941                                               &mddev->recovery))
1942                                         set_bit(In_sync, &rdev->flags);
1943                         }
1944                         rdev->raid_disk = role;
1945                         break;
1946                 }
1947                 if (sb->devflags & WriteMostly1)
1948                         set_bit(WriteMostly, &rdev->flags);
1949                 if (sb->devflags & FailFast1)
1950                         set_bit(FailFast, &rdev->flags);
1951                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1952                         set_bit(Replacement, &rdev->flags);
1953         } else /* MULTIPATH are always insync */
1954                 set_bit(In_sync, &rdev->flags);
1955
1956         return 0;
1957 }
1958
1959 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1960 {
1961         struct mdp_superblock_1 *sb;
1962         struct md_rdev *rdev2;
1963         int max_dev, i;
1964         /* make rdev->sb match mddev and rdev data. */
1965
1966         sb = page_address(rdev->sb_page);
1967
1968         sb->feature_map = 0;
1969         sb->pad0 = 0;
1970         sb->recovery_offset = cpu_to_le64(0);
1971         memset(sb->pad3, 0, sizeof(sb->pad3));
1972
1973         sb->utime = cpu_to_le64((__u64)mddev->utime);
1974         sb->events = cpu_to_le64(mddev->events);
1975         if (mddev->in_sync)
1976                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1977         else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1978                 sb->resync_offset = cpu_to_le64(MaxSector);
1979         else
1980                 sb->resync_offset = cpu_to_le64(0);
1981
1982         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1983
1984         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1985         sb->size = cpu_to_le64(mddev->dev_sectors);
1986         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1987         sb->level = cpu_to_le32(mddev->level);
1988         sb->layout = cpu_to_le32(mddev->layout);
1989         if (test_bit(FailFast, &rdev->flags))
1990                 sb->devflags |= FailFast1;
1991         else
1992                 sb->devflags &= ~FailFast1;
1993
1994         if (test_bit(WriteMostly, &rdev->flags))
1995                 sb->devflags |= WriteMostly1;
1996         else
1997                 sb->devflags &= ~WriteMostly1;
1998         sb->data_offset = cpu_to_le64(rdev->data_offset);
1999         sb->data_size = cpu_to_le64(rdev->sectors);
2000
2001         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2002                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2003                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2004         }
2005
2006         if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2007             !test_bit(In_sync, &rdev->flags)) {
2008                 sb->feature_map |=
2009                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2010                 sb->recovery_offset =
2011                         cpu_to_le64(rdev->recovery_offset);
2012                 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2013                         sb->feature_map |=
2014                                 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2015         }
2016         /* Note: recovery_offset and journal_tail share space  */
2017         if (test_bit(Journal, &rdev->flags))
2018                 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2019         if (test_bit(Replacement, &rdev->flags))
2020                 sb->feature_map |=
2021                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
2022
2023         if (mddev->reshape_position != MaxSector) {
2024                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2025                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2026                 sb->new_layout = cpu_to_le32(mddev->new_layout);
2027                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2028                 sb->new_level = cpu_to_le32(mddev->new_level);
2029                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2030                 if (mddev->delta_disks == 0 &&
2031                     mddev->reshape_backwards)
2032                         sb->feature_map
2033                                 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2034                 if (rdev->new_data_offset != rdev->data_offset) {
2035                         sb->feature_map
2036                                 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2037                         sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2038                                                              - rdev->data_offset));
2039                 }
2040         }
2041
2042         if (mddev_is_clustered(mddev))
2043                 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2044
2045         if (rdev->badblocks.count == 0)
2046                 /* Nothing to do for bad blocks*/ ;
2047         else if (sb->bblog_offset == 0)
2048                 /* Cannot record bad blocks on this device */
2049                 md_error(mddev, rdev);
2050         else {
2051                 struct badblocks *bb = &rdev->badblocks;
2052                 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2053                 u64 *p = bb->page;
2054                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2055                 if (bb->changed) {
2056                         unsigned seq;
2057
2058 retry:
2059                         seq = read_seqbegin(&bb->lock);
2060
2061                         memset(bbp, 0xff, PAGE_SIZE);
2062
2063                         for (i = 0 ; i < bb->count ; i++) {
2064                                 u64 internal_bb = p[i];
2065                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2066                                                 | BB_LEN(internal_bb));
2067                                 bbp[i] = cpu_to_le64(store_bb);
2068                         }
2069                         bb->changed = 0;
2070                         if (read_seqretry(&bb->lock, seq))
2071                                 goto retry;
2072
2073                         bb->sector = (rdev->sb_start +
2074                                       (int)le32_to_cpu(sb->bblog_offset));
2075                         bb->size = le16_to_cpu(sb->bblog_size);
2076                 }
2077         }
2078
2079         max_dev = 0;
2080         rdev_for_each(rdev2, mddev)
2081                 if (rdev2->desc_nr+1 > max_dev)
2082                         max_dev = rdev2->desc_nr+1;
2083
2084         if (max_dev > le32_to_cpu(sb->max_dev)) {
2085                 int bmask;
2086                 sb->max_dev = cpu_to_le32(max_dev);
2087                 rdev->sb_size = max_dev * 2 + 256;
2088                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2089                 if (rdev->sb_size & bmask)
2090                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
2091         } else
2092                 max_dev = le32_to_cpu(sb->max_dev);
2093
2094         for (i=0; i<max_dev;i++)
2095                 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2096
2097         if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2098                 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2099
2100         if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2101                 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2102                         sb->feature_map |=
2103                             cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2104                 else
2105                         sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2106                 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2107                 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2108         }
2109
2110         rdev_for_each(rdev2, mddev) {
2111                 i = rdev2->desc_nr;
2112                 if (test_bit(Faulty, &rdev2->flags))
2113                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2114                 else if (test_bit(In_sync, &rdev2->flags))
2115                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2116                 else if (test_bit(Journal, &rdev2->flags))
2117                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2118                 else if (rdev2->raid_disk >= 0)
2119                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2120                 else
2121                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2122         }
2123
2124         sb->sb_csum = calc_sb_1_csum(sb);
2125 }
2126
2127 static sector_t super_1_choose_bm_space(sector_t dev_size)
2128 {
2129         sector_t bm_space;
2130
2131         /* if the device is bigger than 8Gig, save 64k for bitmap
2132          * usage, if bigger than 200Gig, save 128k
2133          */
2134         if (dev_size < 64*2)
2135                 bm_space = 0;
2136         else if (dev_size - 64*2 >= 200*1024*1024*2)
2137                 bm_space = 128*2;
2138         else if (dev_size - 4*2 > 8*1024*1024*2)
2139                 bm_space = 64*2;
2140         else
2141                 bm_space = 4*2;
2142         return bm_space;
2143 }
2144
2145 static unsigned long long
2146 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2147 {
2148         struct mdp_superblock_1 *sb;
2149         sector_t max_sectors;
2150         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2151                 return 0; /* component must fit device */
2152         if (rdev->data_offset != rdev->new_data_offset)
2153                 return 0; /* too confusing */
2154         if (rdev->sb_start < rdev->data_offset) {
2155                 /* minor versions 1 and 2; superblock before data */
2156                 max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2157                 if (!num_sectors || num_sectors > max_sectors)
2158                         num_sectors = max_sectors;
2159         } else if (rdev->mddev->bitmap_info.offset) {
2160                 /* minor version 0 with bitmap we can't move */
2161                 return 0;
2162         } else {
2163                 /* minor version 0; superblock after data */
2164                 sector_t sb_start, bm_space;
2165                 sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2166
2167                 /* 8K is for superblock */
2168                 sb_start = dev_size - 8*2;
2169                 sb_start &= ~(sector_t)(4*2 - 1);
2170
2171                 bm_space = super_1_choose_bm_space(dev_size);
2172
2173                 /* Space that can be used to store date needs to decrease
2174                  * superblock bitmap space and bad block space(4K)
2175                  */
2176                 max_sectors = sb_start - bm_space - 4*2;
2177
2178                 if (!num_sectors || num_sectors > max_sectors)
2179                         num_sectors = max_sectors;
2180                 rdev->sb_start = sb_start;
2181         }
2182         sb = page_address(rdev->sb_page);
2183         sb->data_size = cpu_to_le64(num_sectors);
2184         sb->super_offset = cpu_to_le64(rdev->sb_start);
2185         sb->sb_csum = calc_sb_1_csum(sb);
2186         do {
2187                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2188                                rdev->sb_page);
2189         } while (md_super_wait(rdev->mddev) < 0);
2190         return num_sectors;
2191
2192 }
2193
2194 static int
2195 super_1_allow_new_offset(struct md_rdev *rdev,
2196                          unsigned long long new_offset)
2197 {
2198         /* All necessary checks on new >= old have been done */
2199         struct bitmap *bitmap;
2200         if (new_offset >= rdev->data_offset)
2201                 return 1;
2202
2203         /* with 1.0 metadata, there is no metadata to tread on
2204          * so we can always move back */
2205         if (rdev->mddev->minor_version == 0)
2206                 return 1;
2207
2208         /* otherwise we must be sure not to step on
2209          * any metadata, so stay:
2210          * 36K beyond start of superblock
2211          * beyond end of badblocks
2212          * beyond write-intent bitmap
2213          */
2214         if (rdev->sb_start + (32+4)*2 > new_offset)
2215                 return 0;
2216         bitmap = rdev->mddev->bitmap;
2217         if (bitmap && !rdev->mddev->bitmap_info.file &&
2218             rdev->sb_start + rdev->mddev->bitmap_info.offset +
2219             bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2220                 return 0;
2221         if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2222                 return 0;
2223
2224         return 1;
2225 }
2226
2227 static struct super_type super_types[] = {
2228         [0] = {
2229                 .name   = "0.90.0",
2230                 .owner  = THIS_MODULE,
2231                 .load_super         = super_90_load,
2232                 .validate_super     = super_90_validate,
2233                 .sync_super         = super_90_sync,
2234                 .rdev_size_change   = super_90_rdev_size_change,
2235                 .allow_new_offset   = super_90_allow_new_offset,
2236         },
2237         [1] = {
2238                 .name   = "md-1",
2239                 .owner  = THIS_MODULE,
2240                 .load_super         = super_1_load,
2241                 .validate_super     = super_1_validate,
2242                 .sync_super         = super_1_sync,
2243                 .rdev_size_change   = super_1_rdev_size_change,
2244                 .allow_new_offset   = super_1_allow_new_offset,
2245         },
2246 };
2247
2248 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2249 {
2250         if (mddev->sync_super) {
2251                 mddev->sync_super(mddev, rdev);
2252                 return;
2253         }
2254
2255         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2256
2257         super_types[mddev->major_version].sync_super(mddev, rdev);
2258 }
2259
2260 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2261 {
2262         struct md_rdev *rdev, *rdev2;
2263
2264         rcu_read_lock();
2265         rdev_for_each_rcu(rdev, mddev1) {
2266                 if (test_bit(Faulty, &rdev->flags) ||
2267                     test_bit(Journal, &rdev->flags) ||
2268                     rdev->raid_disk == -1)
2269                         continue;
2270                 rdev_for_each_rcu(rdev2, mddev2) {
2271                         if (test_bit(Faulty, &rdev2->flags) ||
2272                             test_bit(Journal, &rdev2->flags) ||
2273                             rdev2->raid_disk == -1)
2274                                 continue;
2275                         if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2276                                 rcu_read_unlock();
2277                                 return 1;
2278                         }
2279                 }
2280         }
2281         rcu_read_unlock();
2282         return 0;
2283 }
2284
2285 static LIST_HEAD(pending_raid_disks);
2286
2287 /*
2288  * Try to register data integrity profile for an mddev
2289  *
2290  * This is called when an array is started and after a disk has been kicked
2291  * from the array. It only succeeds if all working and active component devices
2292  * are integrity capable with matching profiles.
2293  */
2294 int md_integrity_register(struct mddev *mddev)
2295 {
2296         struct md_rdev *rdev, *reference = NULL;
2297
2298         if (list_empty(&mddev->disks))
2299                 return 0; /* nothing to do */
2300         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2301                 return 0; /* shouldn't register, or already is */
2302         rdev_for_each(rdev, mddev) {
2303                 /* skip spares and non-functional disks */
2304                 if (test_bit(Faulty, &rdev->flags))
2305                         continue;
2306                 if (rdev->raid_disk < 0)
2307                         continue;
2308                 if (!reference) {
2309                         /* Use the first rdev as the reference */
2310                         reference = rdev;
2311                         continue;
2312                 }
2313                 /* does this rdev's profile match the reference profile? */
2314                 if (blk_integrity_compare(reference->bdev->bd_disk,
2315                                 rdev->bdev->bd_disk) < 0)
2316                         return -EINVAL;
2317         }
2318         if (!reference || !bdev_get_integrity(reference->bdev))
2319                 return 0;
2320         /*
2321          * All component devices are integrity capable and have matching
2322          * profiles, register the common profile for the md device.
2323          */
2324         blk_integrity_register(mddev->gendisk,
2325                                bdev_get_integrity(reference->bdev));
2326
2327         pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2328         if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2329             (mddev->level != 1 && mddev->level != 10 &&
2330              bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2331                 /*
2332                  * No need to handle the failure of bioset_integrity_create,
2333                  * because the function is called by md_run() -> pers->run(),
2334                  * md_run calls bioset_exit -> bioset_integrity_free in case
2335                  * of failure case.
2336                  */
2337                 pr_err("md: failed to create integrity pool for %s\n",
2338                        mdname(mddev));
2339                 return -EINVAL;
2340         }
2341         return 0;
2342 }
2343 EXPORT_SYMBOL(md_integrity_register);
2344
2345 /*
2346  * Attempt to add an rdev, but only if it is consistent with the current
2347  * integrity profile
2348  */
2349 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2350 {
2351         struct blk_integrity *bi_mddev;
2352
2353         if (!mddev->gendisk)
2354                 return 0;
2355
2356         bi_mddev = blk_get_integrity(mddev->gendisk);
2357
2358         if (!bi_mddev) /* nothing to do */
2359                 return 0;
2360
2361         if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2362                 pr_err("%s: incompatible integrity profile for %pg\n",
2363                        mdname(mddev), rdev->bdev);
2364                 return -ENXIO;
2365         }
2366
2367         return 0;
2368 }
2369 EXPORT_SYMBOL(md_integrity_add_rdev);
2370
2371 static bool rdev_read_only(struct md_rdev *rdev)
2372 {
2373         return bdev_read_only(rdev->bdev) ||
2374                 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2375 }
2376
2377 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2378 {
2379         char b[BDEVNAME_SIZE];
2380         int err;
2381
2382         /* prevent duplicates */
2383         if (find_rdev(mddev, rdev->bdev->bd_dev))
2384                 return -EEXIST;
2385
2386         if (rdev_read_only(rdev) && mddev->pers)
2387                 return -EROFS;
2388
2389         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2390         if (!test_bit(Journal, &rdev->flags) &&
2391             rdev->sectors &&
2392             (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2393                 if (mddev->pers) {
2394                         /* Cannot change size, so fail
2395                          * If mddev->level <= 0, then we don't care
2396                          * about aligning sizes (e.g. linear)
2397                          */
2398                         if (mddev->level > 0)
2399                                 return -ENOSPC;
2400                 } else
2401                         mddev->dev_sectors = rdev->sectors;
2402         }
2403
2404         /* Verify rdev->desc_nr is unique.
2405          * If it is -1, assign a free number, else
2406          * check number is not in use
2407          */
2408         rcu_read_lock();
2409         if (rdev->desc_nr < 0) {
2410                 int choice = 0;
2411                 if (mddev->pers)
2412                         choice = mddev->raid_disks;
2413                 while (md_find_rdev_nr_rcu(mddev, choice))
2414                         choice++;
2415                 rdev->desc_nr = choice;
2416         } else {
2417                 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2418                         rcu_read_unlock();
2419                         return -EBUSY;
2420                 }
2421         }
2422         rcu_read_unlock();
2423         if (!test_bit(Journal, &rdev->flags) &&
2424             mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2425                 pr_warn("md: %s: array is limited to %d devices\n",
2426                         mdname(mddev), mddev->max_disks);
2427                 return -EBUSY;
2428         }
2429         snprintf(b, sizeof(b), "%pg", rdev->bdev);
2430         strreplace(b, '/', '!');
2431
2432         rdev->mddev = mddev;
2433         pr_debug("md: bind<%s>\n", b);
2434
2435         if (mddev->raid_disks)
2436                 mddev_create_serial_pool(mddev, rdev, false);
2437
2438         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2439                 goto fail;
2440
2441         /* failure here is OK */
2442         err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2443         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2444         rdev->sysfs_unack_badblocks =
2445                 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2446         rdev->sysfs_badblocks =
2447                 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2448
2449         list_add_rcu(&rdev->same_set, &mddev->disks);
2450         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2451
2452         /* May as well allow recovery to be retried once */
2453         mddev->recovery_disabled++;
2454
2455         return 0;
2456
2457  fail:
2458         pr_warn("md: failed to register dev-%s for %s\n",
2459                 b, mdname(mddev));
2460         return err;
2461 }
2462
2463 static void rdev_delayed_delete(struct work_struct *ws)
2464 {
2465         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2466         kobject_del(&rdev->kobj);
2467         kobject_put(&rdev->kobj);
2468 }
2469
2470 void md_autodetect_dev(dev_t dev);
2471
2472 static void export_rdev(struct md_rdev *rdev)
2473 {
2474         pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2475         md_rdev_clear(rdev);
2476 #ifndef MODULE
2477         if (test_bit(AutoDetected, &rdev->flags))
2478                 md_autodetect_dev(rdev->bdev->bd_dev);
2479 #endif
2480         blkdev_put(rdev->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2481         rdev->bdev = NULL;
2482         kobject_put(&rdev->kobj);
2483 }
2484
2485 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2486 {
2487         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2488         list_del_rcu(&rdev->same_set);
2489         pr_debug("md: unbind<%pg>\n", rdev->bdev);
2490         mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2491         rdev->mddev = NULL;
2492         sysfs_remove_link(&rdev->kobj, "block");
2493         sysfs_put(rdev->sysfs_state);
2494         sysfs_put(rdev->sysfs_unack_badblocks);
2495         sysfs_put(rdev->sysfs_badblocks);
2496         rdev->sysfs_state = NULL;
2497         rdev->sysfs_unack_badblocks = NULL;
2498         rdev->sysfs_badblocks = NULL;
2499         rdev->badblocks.count = 0;
2500         /* We need to delay this, otherwise we can deadlock when
2501          * writing to 'remove' to "dev/state".  We also need
2502          * to delay it due to rcu usage.
2503          */
2504         synchronize_rcu();
2505         INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2506         kobject_get(&rdev->kobj);
2507         queue_work(md_rdev_misc_wq, &rdev->del_work);
2508         export_rdev(rdev);
2509 }
2510
2511 static void export_array(struct mddev *mddev)
2512 {
2513         struct md_rdev *rdev;
2514
2515         while (!list_empty(&mddev->disks)) {
2516                 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2517                                         same_set);
2518                 md_kick_rdev_from_array(rdev);
2519         }
2520         mddev->raid_disks = 0;
2521         mddev->major_version = 0;
2522 }
2523
2524 static bool set_in_sync(struct mddev *mddev)
2525 {
2526         lockdep_assert_held(&mddev->lock);
2527         if (!mddev->in_sync) {
2528                 mddev->sync_checkers++;
2529                 spin_unlock(&mddev->lock);
2530                 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2531                 spin_lock(&mddev->lock);
2532                 if (!mddev->in_sync &&
2533                     percpu_ref_is_zero(&mddev->writes_pending)) {
2534                         mddev->in_sync = 1;
2535                         /*
2536                          * Ensure ->in_sync is visible before we clear
2537                          * ->sync_checkers.
2538                          */
2539                         smp_mb();
2540                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2541                         sysfs_notify_dirent_safe(mddev->sysfs_state);
2542                 }
2543                 if (--mddev->sync_checkers == 0)
2544                         percpu_ref_switch_to_percpu(&mddev->writes_pending);
2545         }
2546         if (mddev->safemode == 1)
2547                 mddev->safemode = 0;
2548         return mddev->in_sync;
2549 }
2550
2551 static void sync_sbs(struct mddev *mddev, int nospares)
2552 {
2553         /* Update each superblock (in-memory image), but
2554          * if we are allowed to, skip spares which already
2555          * have the right event counter, or have one earlier
2556          * (which would mean they aren't being marked as dirty
2557          * with the rest of the array)
2558          */
2559         struct md_rdev *rdev;
2560         rdev_for_each(rdev, mddev) {
2561                 if (rdev->sb_events == mddev->events ||
2562                     (nospares &&
2563                      rdev->raid_disk < 0 &&
2564                      rdev->sb_events+1 == mddev->events)) {
2565                         /* Don't update this superblock */
2566                         rdev->sb_loaded = 2;
2567                 } else {
2568                         sync_super(mddev, rdev);
2569                         rdev->sb_loaded = 1;
2570                 }
2571         }
2572 }
2573
2574 static bool does_sb_need_changing(struct mddev *mddev)
2575 {
2576         struct md_rdev *rdev = NULL, *iter;
2577         struct mdp_superblock_1 *sb;
2578         int role;
2579
2580         /* Find a good rdev */
2581         rdev_for_each(iter, mddev)
2582                 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2583                         rdev = iter;
2584                         break;
2585                 }
2586
2587         /* No good device found. */
2588         if (!rdev)
2589                 return false;
2590
2591         sb = page_address(rdev->sb_page);
2592         /* Check if a device has become faulty or a spare become active */
2593         rdev_for_each(rdev, mddev) {
2594                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2595                 /* Device activated? */
2596                 if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2597                     !test_bit(Faulty, &rdev->flags))
2598                         return true;
2599                 /* Device turned faulty? */
2600                 if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2601                         return true;
2602         }
2603
2604         /* Check if any mddev parameters have changed */
2605         if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2606             (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2607             (mddev->layout != le32_to_cpu(sb->layout)) ||
2608             (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2609             (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2610                 return true;
2611
2612         return false;
2613 }
2614
2615 void md_update_sb(struct mddev *mddev, int force_change)
2616 {
2617         struct md_rdev *rdev;
2618         int sync_req;
2619         int nospares = 0;
2620         int any_badblocks_changed = 0;
2621         int ret = -1;
2622
2623         if (!md_is_rdwr(mddev)) {
2624                 if (force_change)
2625                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2626                 return;
2627         }
2628
2629 repeat:
2630         if (mddev_is_clustered(mddev)) {
2631                 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2632                         force_change = 1;
2633                 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2634                         nospares = 1;
2635                 ret = md_cluster_ops->metadata_update_start(mddev);
2636                 /* Has someone else has updated the sb */
2637                 if (!does_sb_need_changing(mddev)) {
2638                         if (ret == 0)
2639                                 md_cluster_ops->metadata_update_cancel(mddev);
2640                         bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2641                                                          BIT(MD_SB_CHANGE_DEVS) |
2642                                                          BIT(MD_SB_CHANGE_CLEAN));
2643                         return;
2644                 }
2645         }
2646
2647         /*
2648          * First make sure individual recovery_offsets are correct
2649          * curr_resync_completed can only be used during recovery.
2650          * During reshape/resync it might use array-addresses rather
2651          * that device addresses.
2652          */
2653         rdev_for_each(rdev, mddev) {
2654                 if (rdev->raid_disk >= 0 &&
2655                     mddev->delta_disks >= 0 &&
2656                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2657                     test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2658                     !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2659                     !test_bit(Journal, &rdev->flags) &&
2660                     !test_bit(In_sync, &rdev->flags) &&
2661                     mddev->curr_resync_completed > rdev->recovery_offset)
2662                                 rdev->recovery_offset = mddev->curr_resync_completed;
2663
2664         }
2665         if (!mddev->persistent) {
2666                 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2667                 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2668                 if (!mddev->external) {
2669                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2670                         rdev_for_each(rdev, mddev) {
2671                                 if (rdev->badblocks.changed) {
2672                                         rdev->badblocks.changed = 0;
2673                                         ack_all_badblocks(&rdev->badblocks);
2674                                         md_error(mddev, rdev);
2675                                 }
2676                                 clear_bit(Blocked, &rdev->flags);
2677                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2678                                 wake_up(&rdev->blocked_wait);
2679                         }
2680                 }
2681                 wake_up(&mddev->sb_wait);
2682                 return;
2683         }
2684
2685         spin_lock(&mddev->lock);
2686
2687         mddev->utime = ktime_get_real_seconds();
2688
2689         if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2690                 force_change = 1;
2691         if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2692                 /* just a clean<-> dirty transition, possibly leave spares alone,
2693                  * though if events isn't the right even/odd, we will have to do
2694                  * spares after all
2695                  */
2696                 nospares = 1;
2697         if (force_change)
2698                 nospares = 0;
2699         if (mddev->degraded)
2700                 /* If the array is degraded, then skipping spares is both
2701                  * dangerous and fairly pointless.
2702                  * Dangerous because a device that was removed from the array
2703                  * might have a event_count that still looks up-to-date,
2704                  * so it can be re-added without a resync.
2705                  * Pointless because if there are any spares to skip,
2706                  * then a recovery will happen and soon that array won't
2707                  * be degraded any more and the spare can go back to sleep then.
2708                  */
2709                 nospares = 0;
2710
2711         sync_req = mddev->in_sync;
2712
2713         /* If this is just a dirty<->clean transition, and the array is clean
2714          * and 'events' is odd, we can roll back to the previous clean state */
2715         if (nospares
2716             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2717             && mddev->can_decrease_events
2718             && mddev->events != 1) {
2719                 mddev->events--;
2720                 mddev->can_decrease_events = 0;
2721         } else {
2722                 /* otherwise we have to go forward and ... */
2723                 mddev->events ++;
2724                 mddev->can_decrease_events = nospares;
2725         }
2726
2727         /*
2728          * This 64-bit counter should never wrap.
2729          * Either we are in around ~1 trillion A.C., assuming
2730          * 1 reboot per second, or we have a bug...
2731          */
2732         WARN_ON(mddev->events == 0);
2733
2734         rdev_for_each(rdev, mddev) {
2735                 if (rdev->badblocks.changed)
2736                         any_badblocks_changed++;
2737                 if (test_bit(Faulty, &rdev->flags))
2738                         set_bit(FaultRecorded, &rdev->flags);
2739         }
2740
2741         sync_sbs(mddev, nospares);
2742         spin_unlock(&mddev->lock);
2743
2744         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2745                  mdname(mddev), mddev->in_sync);
2746
2747         if (mddev->queue)
2748                 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2749 rewrite:
2750         md_bitmap_update_sb(mddev->bitmap);
2751         rdev_for_each(rdev, mddev) {
2752                 if (rdev->sb_loaded != 1)
2753                         continue; /* no noise on spare devices */
2754
2755                 if (!test_bit(Faulty, &rdev->flags)) {
2756                         md_super_write(mddev,rdev,
2757                                        rdev->sb_start, rdev->sb_size,
2758                                        rdev->sb_page);
2759                         pr_debug("md: (write) %pg's sb offset: %llu\n",
2760                                  rdev->bdev,
2761                                  (unsigned long long)rdev->sb_start);
2762                         rdev->sb_events = mddev->events;
2763                         if (rdev->badblocks.size) {
2764                                 md_super_write(mddev, rdev,
2765                                                rdev->badblocks.sector,
2766                                                rdev->badblocks.size << 9,
2767                                                rdev->bb_page);
2768                                 rdev->badblocks.size = 0;
2769                         }
2770
2771                 } else
2772                         pr_debug("md: %pg (skipping faulty)\n",
2773                                  rdev->bdev);
2774
2775                 if (mddev->level == LEVEL_MULTIPATH)
2776                         /* only need to write one superblock... */
2777                         break;
2778         }
2779         if (md_super_wait(mddev) < 0)
2780                 goto rewrite;
2781         /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2782
2783         if (mddev_is_clustered(mddev) && ret == 0)
2784                 md_cluster_ops->metadata_update_finish(mddev);
2785
2786         if (mddev->in_sync != sync_req ||
2787             !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2788                                BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2789                 /* have to write it out again */
2790                 goto repeat;
2791         wake_up(&mddev->sb_wait);
2792         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2793                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2794
2795         rdev_for_each(rdev, mddev) {
2796                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2797                         clear_bit(Blocked, &rdev->flags);
2798
2799                 if (any_badblocks_changed)
2800                         ack_all_badblocks(&rdev->badblocks);
2801                 clear_bit(BlockedBadBlocks, &rdev->flags);
2802                 wake_up(&rdev->blocked_wait);
2803         }
2804 }
2805 EXPORT_SYMBOL(md_update_sb);
2806
2807 static int add_bound_rdev(struct md_rdev *rdev)
2808 {
2809         struct mddev *mddev = rdev->mddev;
2810         int err = 0;
2811         bool add_journal = test_bit(Journal, &rdev->flags);
2812
2813         if (!mddev->pers->hot_remove_disk || add_journal) {
2814                 /* If there is hot_add_disk but no hot_remove_disk
2815                  * then added disks for geometry changes,
2816                  * and should be added immediately.
2817                  */
2818                 super_types[mddev->major_version].
2819                         validate_super(mddev, rdev);
2820                 if (add_journal)
2821                         mddev_suspend(mddev);
2822                 err = mddev->pers->hot_add_disk(mddev, rdev);
2823                 if (add_journal)
2824                         mddev_resume(mddev);
2825                 if (err) {
2826                         md_kick_rdev_from_array(rdev);
2827                         return err;
2828                 }
2829         }
2830         sysfs_notify_dirent_safe(rdev->sysfs_state);
2831
2832         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2833         if (mddev->degraded)
2834                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2835         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2836         md_new_event();
2837         md_wakeup_thread(mddev->thread);
2838         return 0;
2839 }
2840
2841 /* words written to sysfs files may, or may not, be \n terminated.
2842  * We want to accept with case. For this we use cmd_match.
2843  */
2844 static int cmd_match(const char *cmd, const char *str)
2845 {
2846         /* See if cmd, written into a sysfs file, matches
2847          * str.  They must either be the same, or cmd can
2848          * have a trailing newline
2849          */
2850         while (*cmd && *str && *cmd == *str) {
2851                 cmd++;
2852                 str++;
2853         }
2854         if (*cmd == '\n')
2855                 cmd++;
2856         if (*str || *cmd)
2857                 return 0;
2858         return 1;
2859 }
2860
2861 struct rdev_sysfs_entry {
2862         struct attribute attr;
2863         ssize_t (*show)(struct md_rdev *, char *);
2864         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2865 };
2866
2867 static ssize_t
2868 state_show(struct md_rdev *rdev, char *page)
2869 {
2870         char *sep = ",";
2871         size_t len = 0;
2872         unsigned long flags = READ_ONCE(rdev->flags);
2873
2874         if (test_bit(Faulty, &flags) ||
2875             (!test_bit(ExternalBbl, &flags) &&
2876             rdev->badblocks.unacked_exist))
2877                 len += sprintf(page+len, "faulty%s", sep);
2878         if (test_bit(In_sync, &flags))
2879                 len += sprintf(page+len, "in_sync%s", sep);
2880         if (test_bit(Journal, &flags))
2881                 len += sprintf(page+len, "journal%s", sep);
2882         if (test_bit(WriteMostly, &flags))
2883                 len += sprintf(page+len, "write_mostly%s", sep);
2884         if (test_bit(Blocked, &flags) ||
2885             (rdev->badblocks.unacked_exist
2886              && !test_bit(Faulty, &flags)))
2887                 len += sprintf(page+len, "blocked%s", sep);
2888         if (!test_bit(Faulty, &flags) &&
2889             !test_bit(Journal, &flags) &&
2890             !test_bit(In_sync, &flags))
2891                 len += sprintf(page+len, "spare%s", sep);
2892         if (test_bit(WriteErrorSeen, &flags))
2893                 len += sprintf(page+len, "write_error%s", sep);
2894         if (test_bit(WantReplacement, &flags))
2895                 len += sprintf(page+len, "want_replacement%s", sep);
2896         if (test_bit(Replacement, &flags))
2897                 len += sprintf(page+len, "replacement%s", sep);
2898         if (test_bit(ExternalBbl, &flags))
2899                 len += sprintf(page+len, "external_bbl%s", sep);
2900         if (test_bit(FailFast, &flags))
2901                 len += sprintf(page+len, "failfast%s", sep);
2902
2903         if (len)
2904                 len -= strlen(sep);
2905
2906         return len+sprintf(page+len, "\n");
2907 }
2908
2909 static ssize_t
2910 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2911 {
2912         /* can write
2913          *  faulty  - simulates an error
2914          *  remove  - disconnects the device
2915          *  writemostly - sets write_mostly
2916          *  -writemostly - clears write_mostly
2917          *  blocked - sets the Blocked flags
2918          *  -blocked - clears the Blocked and possibly simulates an error
2919          *  insync - sets Insync providing device isn't active
2920          *  -insync - clear Insync for a device with a slot assigned,
2921          *            so that it gets rebuilt based on bitmap
2922          *  write_error - sets WriteErrorSeen
2923          *  -write_error - clears WriteErrorSeen
2924          *  {,-}failfast - set/clear FailFast
2925          */
2926
2927         struct mddev *mddev = rdev->mddev;
2928         int err = -EINVAL;
2929         bool need_update_sb = false;
2930
2931         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2932                 md_error(rdev->mddev, rdev);
2933
2934                 if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2935                         err = -EBUSY;
2936                 else
2937                         err = 0;
2938         } else if (cmd_match(buf, "remove")) {
2939                 if (rdev->mddev->pers) {
2940                         clear_bit(Blocked, &rdev->flags);
2941                         remove_and_add_spares(rdev->mddev, rdev);
2942                 }
2943                 if (rdev->raid_disk >= 0)
2944                         err = -EBUSY;
2945                 else {
2946                         err = 0;
2947                         if (mddev_is_clustered(mddev))
2948                                 err = md_cluster_ops->remove_disk(mddev, rdev);
2949
2950                         if (err == 0) {
2951                                 md_kick_rdev_from_array(rdev);
2952                                 if (mddev->pers) {
2953                                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2954                                         md_wakeup_thread(mddev->thread);
2955                                 }
2956                                 md_new_event();
2957                         }
2958                 }
2959         } else if (cmd_match(buf, "writemostly")) {
2960                 set_bit(WriteMostly, &rdev->flags);
2961                 mddev_create_serial_pool(rdev->mddev, rdev, false);
2962                 need_update_sb = true;
2963                 err = 0;
2964         } else if (cmd_match(buf, "-writemostly")) {
2965                 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2966                 clear_bit(WriteMostly, &rdev->flags);
2967                 need_update_sb = true;
2968                 err = 0;
2969         } else if (cmd_match(buf, "blocked")) {
2970                 set_bit(Blocked, &rdev->flags);
2971                 err = 0;
2972         } else if (cmd_match(buf, "-blocked")) {
2973                 if (!test_bit(Faulty, &rdev->flags) &&
2974                     !test_bit(ExternalBbl, &rdev->flags) &&
2975                     rdev->badblocks.unacked_exist) {
2976                         /* metadata handler doesn't understand badblocks,
2977                          * so we need to fail the device
2978                          */
2979                         md_error(rdev->mddev, rdev);
2980                 }
2981                 clear_bit(Blocked, &rdev->flags);
2982                 clear_bit(BlockedBadBlocks, &rdev->flags);
2983                 wake_up(&rdev->blocked_wait);
2984                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2985                 md_wakeup_thread(rdev->mddev->thread);
2986
2987                 err = 0;
2988         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2989                 set_bit(In_sync, &rdev->flags);
2990                 err = 0;
2991         } else if (cmd_match(buf, "failfast")) {
2992                 set_bit(FailFast, &rdev->flags);
2993                 need_update_sb = true;
2994                 err = 0;
2995         } else if (cmd_match(buf, "-failfast")) {
2996                 clear_bit(FailFast, &rdev->flags);
2997                 need_update_sb = true;
2998                 err = 0;
2999         } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3000                    !test_bit(Journal, &rdev->flags)) {
3001                 if (rdev->mddev->pers == NULL) {
3002                         clear_bit(In_sync, &rdev->flags);
3003                         rdev->saved_raid_disk = rdev->raid_disk;
3004                         rdev->raid_disk = -1;
3005                         err = 0;
3006                 }
3007         } else if (cmd_match(buf, "write_error")) {
3008                 set_bit(WriteErrorSeen, &rdev->flags);
3009                 err = 0;
3010         } else if (cmd_match(buf, "-write_error")) {
3011                 clear_bit(WriteErrorSeen, &rdev->flags);
3012                 err = 0;
3013         } else if (cmd_match(buf, "want_replacement")) {
3014                 /* Any non-spare device that is not a replacement can
3015                  * become want_replacement at any time, but we then need to
3016                  * check if recovery is needed.
3017                  */
3018                 if (rdev->raid_disk >= 0 &&
3019                     !test_bit(Journal, &rdev->flags) &&
3020                     !test_bit(Replacement, &rdev->flags))
3021                         set_bit(WantReplacement, &rdev->flags);
3022                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3023                 md_wakeup_thread(rdev->mddev->thread);
3024                 err = 0;
3025         } else if (cmd_match(buf, "-want_replacement")) {
3026                 /* Clearing 'want_replacement' is always allowed.
3027                  * Once replacements starts it is too late though.
3028                  */
3029                 err = 0;
3030                 clear_bit(WantReplacement, &rdev->flags);
3031         } else if (cmd_match(buf, "replacement")) {
3032                 /* Can only set a device as a replacement when array has not
3033                  * yet been started.  Once running, replacement is automatic
3034                  * from spares, or by assigning 'slot'.
3035                  */
3036                 if (rdev->mddev->pers)
3037                         err = -EBUSY;
3038                 else {
3039                         set_bit(Replacement, &rdev->flags);
3040                         err = 0;
3041                 }
3042         } else if (cmd_match(buf, "-replacement")) {
3043                 /* Similarly, can only clear Replacement before start */
3044                 if (rdev->mddev->pers)
3045                         err = -EBUSY;
3046                 else {
3047                         clear_bit(Replacement, &rdev->flags);
3048                         err = 0;
3049                 }
3050         } else if (cmd_match(buf, "re-add")) {
3051                 if (!rdev->mddev->pers)
3052                         err = -EINVAL;
3053                 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3054                                 rdev->saved_raid_disk >= 0) {
3055                         /* clear_bit is performed _after_ all the devices
3056                          * have their local Faulty bit cleared. If any writes
3057                          * happen in the meantime in the local node, they
3058                          * will land in the local bitmap, which will be synced
3059                          * by this node eventually
3060                          */
3061                         if (!mddev_is_clustered(rdev->mddev) ||
3062                             (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3063                                 clear_bit(Faulty, &rdev->flags);
3064                                 err = add_bound_rdev(rdev);
3065                         }
3066                 } else
3067                         err = -EBUSY;
3068         } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3069                 set_bit(ExternalBbl, &rdev->flags);
3070                 rdev->badblocks.shift = 0;
3071                 err = 0;
3072         } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3073                 clear_bit(ExternalBbl, &rdev->flags);
3074                 err = 0;
3075         }
3076         if (need_update_sb)
3077                 md_update_sb(mddev, 1);
3078         if (!err)
3079                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3080         return err ? err : len;
3081 }
3082 static struct rdev_sysfs_entry rdev_state =
3083 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3084
3085 static ssize_t
3086 errors_show(struct md_rdev *rdev, char *page)
3087 {
3088         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3089 }
3090
3091 static ssize_t
3092 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3093 {
3094         unsigned int n;
3095         int rv;
3096
3097         rv = kstrtouint(buf, 10, &n);
3098         if (rv < 0)
3099                 return rv;
3100         atomic_set(&rdev->corrected_errors, n);
3101         return len;
3102 }
3103 static struct rdev_sysfs_entry rdev_errors =
3104 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3105
3106 static ssize_t
3107 slot_show(struct md_rdev *rdev, char *page)
3108 {
3109         if (test_bit(Journal, &rdev->flags))
3110                 return sprintf(page, "journal\n");
3111         else if (rdev->raid_disk < 0)
3112                 return sprintf(page, "none\n");
3113         else
3114                 return sprintf(page, "%d\n", rdev->raid_disk);
3115 }
3116
3117 static ssize_t
3118 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3119 {
3120         int slot;
3121         int err;
3122
3123         if (test_bit(Journal, &rdev->flags))
3124                 return -EBUSY;
3125         if (strncmp(buf, "none", 4)==0)
3126                 slot = -1;
3127         else {
3128                 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3129                 if (err < 0)
3130                         return err;
3131                 if (slot < 0)
3132                         /* overflow */
3133                         return -ENOSPC;
3134         }
3135         if (rdev->mddev->pers && slot == -1) {
3136                 /* Setting 'slot' on an active array requires also
3137                  * updating the 'rd%d' link, and communicating
3138                  * with the personality with ->hot_*_disk.
3139                  * For now we only support removing
3140                  * failed/spare devices.  This normally happens automatically,
3141                  * but not when the metadata is externally managed.
3142                  */
3143                 if (rdev->raid_disk == -1)
3144                         return -EEXIST;
3145                 /* personality does all needed checks */
3146                 if (rdev->mddev->pers->hot_remove_disk == NULL)
3147                         return -EINVAL;
3148                 clear_bit(Blocked, &rdev->flags);
3149                 remove_and_add_spares(rdev->mddev, rdev);
3150                 if (rdev->raid_disk >= 0)
3151                         return -EBUSY;
3152                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3153                 md_wakeup_thread(rdev->mddev->thread);
3154         } else if (rdev->mddev->pers) {
3155                 /* Activating a spare .. or possibly reactivating
3156                  * if we ever get bitmaps working here.
3157                  */
3158                 int err;
3159
3160                 if (rdev->raid_disk != -1)
3161                         return -EBUSY;
3162
3163                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3164                         return -EBUSY;
3165
3166                 if (rdev->mddev->pers->hot_add_disk == NULL)
3167                         return -EINVAL;
3168
3169                 if (slot >= rdev->mddev->raid_disks &&
3170                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3171                         return -ENOSPC;
3172
3173                 rdev->raid_disk = slot;
3174                 if (test_bit(In_sync, &rdev->flags))
3175                         rdev->saved_raid_disk = slot;
3176                 else
3177                         rdev->saved_raid_disk = -1;
3178                 clear_bit(In_sync, &rdev->flags);
3179                 clear_bit(Bitmap_sync, &rdev->flags);
3180                 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3181                 if (err) {
3182                         rdev->raid_disk = -1;
3183                         return err;
3184                 } else
3185                         sysfs_notify_dirent_safe(rdev->sysfs_state);
3186                 /* failure here is OK */;
3187                 sysfs_link_rdev(rdev->mddev, rdev);
3188                 /* don't wakeup anyone, leave that to userspace. */
3189         } else {
3190                 if (slot >= rdev->mddev->raid_disks &&
3191                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3192                         return -ENOSPC;
3193                 rdev->raid_disk = slot;
3194                 /* assume it is working */
3195                 clear_bit(Faulty, &rdev->flags);
3196                 clear_bit(WriteMostly, &rdev->flags);
3197                 set_bit(In_sync, &rdev->flags);
3198                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3199         }
3200         return len;
3201 }
3202
3203 static struct rdev_sysfs_entry rdev_slot =
3204 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3205
3206 static ssize_t
3207 offset_show(struct md_rdev *rdev, char *page)
3208 {
3209         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3210 }
3211
3212 static ssize_t
3213 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3214 {
3215         unsigned long long offset;
3216         if (kstrtoull(buf, 10, &offset) < 0)
3217                 return -EINVAL;
3218         if (rdev->mddev->pers && rdev->raid_disk >= 0)
3219                 return -EBUSY;
3220         if (rdev->sectors && rdev->mddev->external)
3221                 /* Must set offset before size, so overlap checks
3222                  * can be sane */
3223                 return -EBUSY;
3224         rdev->data_offset = offset;
3225         rdev->new_data_offset = offset;
3226         return len;
3227 }
3228
3229 static struct rdev_sysfs_entry rdev_offset =
3230 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3231
3232 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3233 {
3234         return sprintf(page, "%llu\n",
3235                        (unsigned long long)rdev->new_data_offset);
3236 }
3237
3238 static ssize_t new_offset_store(struct md_rdev *rdev,
3239                                 const char *buf, size_t len)
3240 {
3241         unsigned long long new_offset;
3242         struct mddev *mddev = rdev->mddev;
3243
3244         if (kstrtoull(buf, 10, &new_offset) < 0)
3245                 return -EINVAL;
3246
3247         if (mddev->sync_thread ||
3248             test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3249                 return -EBUSY;
3250         if (new_offset == rdev->data_offset)
3251                 /* reset is always permitted */
3252                 ;
3253         else if (new_offset > rdev->data_offset) {
3254                 /* must not push array size beyond rdev_sectors */
3255                 if (new_offset - rdev->data_offset
3256                     + mddev->dev_sectors > rdev->sectors)
3257                                 return -E2BIG;
3258         }
3259         /* Metadata worries about other space details. */
3260
3261         /* decreasing the offset is inconsistent with a backwards
3262          * reshape.
3263          */
3264         if (new_offset < rdev->data_offset &&
3265             mddev->reshape_backwards)
3266                 return -EINVAL;
3267         /* Increasing offset is inconsistent with forwards
3268          * reshape.  reshape_direction should be set to
3269          * 'backwards' first.
3270          */
3271         if (new_offset > rdev->data_offset &&
3272             !mddev->reshape_backwards)
3273                 return -EINVAL;
3274
3275         if (mddev->pers && mddev->persistent &&
3276             !super_types[mddev->major_version]
3277             .allow_new_offset(rdev, new_offset))
3278                 return -E2BIG;
3279         rdev->new_data_offset = new_offset;
3280         if (new_offset > rdev->data_offset)
3281                 mddev->reshape_backwards = 1;
3282         else if (new_offset < rdev->data_offset)
3283                 mddev->reshape_backwards = 0;
3284
3285         return len;
3286 }
3287 static struct rdev_sysfs_entry rdev_new_offset =
3288 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3289
3290 static ssize_t
3291 rdev_size_show(struct md_rdev *rdev, char *page)
3292 {
3293         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3294 }
3295
3296 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3297 {
3298         /* check if two start/length pairs overlap */
3299         if (a->data_offset + a->sectors <= b->data_offset)
3300                 return false;
3301         if (b->data_offset + b->sectors <= a->data_offset)
3302                 return false;
3303         return true;
3304 }
3305
3306 static bool md_rdev_overlaps(struct md_rdev *rdev)
3307 {
3308         struct mddev *mddev;
3309         struct md_rdev *rdev2;
3310
3311         spin_lock(&all_mddevs_lock);
3312         list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3313                 if (test_bit(MD_DELETED, &mddev->flags))
3314                         continue;
3315                 rdev_for_each(rdev2, mddev) {
3316                         if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3317                             md_rdevs_overlap(rdev, rdev2)) {
3318                                 spin_unlock(&all_mddevs_lock);
3319                                 return true;
3320                         }
3321                 }
3322         }
3323         spin_unlock(&all_mddevs_lock);
3324         return false;
3325 }
3326
3327 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3328 {
3329         unsigned long long blocks;
3330         sector_t new;
3331
3332         if (kstrtoull(buf, 10, &blocks) < 0)
3333                 return -EINVAL;
3334
3335         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3336                 return -EINVAL; /* sector conversion overflow */
3337
3338         new = blocks * 2;
3339         if (new != blocks * 2)
3340                 return -EINVAL; /* unsigned long long to sector_t overflow */
3341
3342         *sectors = new;
3343         return 0;
3344 }
3345
3346 static ssize_t
3347 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3348 {
3349         struct mddev *my_mddev = rdev->mddev;
3350         sector_t oldsectors = rdev->sectors;
3351         sector_t sectors;
3352
3353         if (test_bit(Journal, &rdev->flags))
3354                 return -EBUSY;
3355         if (strict_blocks_to_sectors(buf, &sectors) < 0)
3356                 return -EINVAL;
3357         if (rdev->data_offset != rdev->new_data_offset)
3358                 return -EINVAL; /* too confusing */
3359         if (my_mddev->pers && rdev->raid_disk >= 0) {
3360                 if (my_mddev->persistent) {
3361                         sectors = super_types[my_mddev->major_version].
3362                                 rdev_size_change(rdev, sectors);
3363                         if (!sectors)
3364                                 return -EBUSY;
3365                 } else if (!sectors)
3366                         sectors = bdev_nr_sectors(rdev->bdev) -
3367                                 rdev->data_offset;
3368                 if (!my_mddev->pers->resize)
3369                         /* Cannot change size for RAID0 or Linear etc */
3370                         return -EINVAL;
3371         }
3372         if (sectors < my_mddev->dev_sectors)
3373                 return -EINVAL; /* component must fit device */
3374
3375         rdev->sectors = sectors;
3376
3377         /*
3378          * Check that all other rdevs with the same bdev do not overlap.  This
3379          * check does not provide a hard guarantee, it just helps avoid
3380          * dangerous mistakes.
3381          */
3382         if (sectors > oldsectors && my_mddev->external &&
3383             md_rdev_overlaps(rdev)) {
3384                 /*
3385                  * Someone else could have slipped in a size change here, but
3386                  * doing so is just silly.  We put oldsectors back because we
3387                  * know it is safe, and trust userspace not to race with itself.
3388                  */
3389                 rdev->sectors = oldsectors;
3390                 return -EBUSY;
3391         }
3392         return len;
3393 }
3394
3395 static struct rdev_sysfs_entry rdev_size =
3396 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3397
3398 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3399 {
3400         unsigned long long recovery_start = rdev->recovery_offset;
3401
3402         if (test_bit(In_sync, &rdev->flags) ||
3403             recovery_start == MaxSector)
3404                 return sprintf(page, "none\n");
3405
3406         return sprintf(page, "%llu\n", recovery_start);
3407 }
3408
3409 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3410 {
3411         unsigned long long recovery_start;
3412
3413         if (cmd_match(buf, "none"))
3414                 recovery_start = MaxSector;
3415         else if (kstrtoull(buf, 10, &recovery_start))
3416                 return -EINVAL;
3417
3418         if (rdev->mddev->pers &&
3419             rdev->raid_disk >= 0)
3420                 return -EBUSY;
3421
3422         rdev->recovery_offset = recovery_start;
3423         if (recovery_start == MaxSector)
3424                 set_bit(In_sync, &rdev->flags);
3425         else
3426                 clear_bit(In_sync, &rdev->flags);
3427         return len;
3428 }
3429
3430 static struct rdev_sysfs_entry rdev_recovery_start =
3431 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3432
3433 /* sysfs access to bad-blocks list.
3434  * We present two files.
3435  * 'bad-blocks' lists sector numbers and lengths of ranges that
3436  *    are recorded as bad.  The list is truncated to fit within
3437  *    the one-page limit of sysfs.
3438  *    Writing "sector length" to this file adds an acknowledged
3439  *    bad block list.
3440  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3441  *    been acknowledged.  Writing to this file adds bad blocks
3442  *    without acknowledging them.  This is largely for testing.
3443  */
3444 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3445 {
3446         return badblocks_show(&rdev->badblocks, page, 0);
3447 }
3448 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3449 {
3450         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3451         /* Maybe that ack was all we needed */
3452         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3453                 wake_up(&rdev->blocked_wait);
3454         return rv;
3455 }
3456 static struct rdev_sysfs_entry rdev_bad_blocks =
3457 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3458
3459 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3460 {
3461         return badblocks_show(&rdev->badblocks, page, 1);
3462 }
3463 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3464 {
3465         return badblocks_store(&rdev->badblocks, page, len, 1);
3466 }
3467 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3468 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3469
3470 static ssize_t
3471 ppl_sector_show(struct md_rdev *rdev, char *page)
3472 {
3473         return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3474 }
3475
3476 static ssize_t
3477 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3478 {
3479         unsigned long long sector;
3480
3481         if (kstrtoull(buf, 10, &sector) < 0)
3482                 return -EINVAL;
3483         if (sector != (sector_t)sector)
3484                 return -EINVAL;
3485
3486         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3487             rdev->raid_disk >= 0)
3488                 return -EBUSY;
3489
3490         if (rdev->mddev->persistent) {
3491                 if (rdev->mddev->major_version == 0)
3492                         return -EINVAL;
3493                 if ((sector > rdev->sb_start &&
3494                      sector - rdev->sb_start > S16_MAX) ||
3495                     (sector < rdev->sb_start &&
3496                      rdev->sb_start - sector > -S16_MIN))
3497                         return -EINVAL;
3498                 rdev->ppl.offset = sector - rdev->sb_start;
3499         } else if (!rdev->mddev->external) {
3500                 return -EBUSY;
3501         }
3502         rdev->ppl.sector = sector;
3503         return len;
3504 }
3505
3506 static struct rdev_sysfs_entry rdev_ppl_sector =
3507 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3508
3509 static ssize_t
3510 ppl_size_show(struct md_rdev *rdev, char *page)
3511 {
3512         return sprintf(page, "%u\n", rdev->ppl.size);
3513 }
3514
3515 static ssize_t
3516 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3517 {
3518         unsigned int size;
3519
3520         if (kstrtouint(buf, 10, &size) < 0)
3521                 return -EINVAL;
3522
3523         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3524             rdev->raid_disk >= 0)
3525                 return -EBUSY;
3526
3527         if (rdev->mddev->persistent) {
3528                 if (rdev->mddev->major_version == 0)
3529                         return -EINVAL;
3530                 if (size > U16_MAX)
3531                         return -EINVAL;
3532         } else if (!rdev->mddev->external) {
3533                 return -EBUSY;
3534         }
3535         rdev->ppl.size = size;
3536         return len;
3537 }
3538
3539 static struct rdev_sysfs_entry rdev_ppl_size =
3540 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3541
3542 static struct attribute *rdev_default_attrs[] = {
3543         &rdev_state.attr,
3544         &rdev_errors.attr,
3545         &rdev_slot.attr,
3546         &rdev_offset.attr,
3547         &rdev_new_offset.attr,
3548         &rdev_size.attr,
3549         &rdev_recovery_start.attr,
3550         &rdev_bad_blocks.attr,
3551         &rdev_unack_bad_blocks.attr,
3552         &rdev_ppl_sector.attr,
3553         &rdev_ppl_size.attr,
3554         NULL,
3555 };
3556 ATTRIBUTE_GROUPS(rdev_default);
3557 static ssize_t
3558 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3559 {
3560         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3561         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3562
3563         if (!entry->show)
3564                 return -EIO;
3565         if (!rdev->mddev)
3566                 return -ENODEV;
3567         return entry->show(rdev, page);
3568 }
3569
3570 static ssize_t
3571 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3572               const char *page, size_t length)
3573 {
3574         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3575         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3576         ssize_t rv;
3577         struct mddev *mddev = rdev->mddev;
3578
3579         if (!entry->store)
3580                 return -EIO;
3581         if (!capable(CAP_SYS_ADMIN))
3582                 return -EACCES;
3583         rv = mddev ? mddev_lock(mddev) : -ENODEV;
3584         if (!rv) {
3585                 if (rdev->mddev == NULL)
3586                         rv = -ENODEV;
3587                 else
3588                         rv = entry->store(rdev, page, length);
3589                 mddev_unlock(mddev);
3590         }
3591         return rv;
3592 }
3593
3594 static void rdev_free(struct kobject *ko)
3595 {
3596         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3597         kfree(rdev);
3598 }
3599 static const struct sysfs_ops rdev_sysfs_ops = {
3600         .show           = rdev_attr_show,
3601         .store          = rdev_attr_store,
3602 };
3603 static struct kobj_type rdev_ktype = {
3604         .release        = rdev_free,
3605         .sysfs_ops      = &rdev_sysfs_ops,
3606         .default_groups = rdev_default_groups,
3607 };
3608
3609 int md_rdev_init(struct md_rdev *rdev)
3610 {
3611         rdev->desc_nr = -1;
3612         rdev->saved_raid_disk = -1;
3613         rdev->raid_disk = -1;
3614         rdev->flags = 0;
3615         rdev->data_offset = 0;
3616         rdev->new_data_offset = 0;
3617         rdev->sb_events = 0;
3618         rdev->last_read_error = 0;
3619         rdev->sb_loaded = 0;
3620         rdev->bb_page = NULL;
3621         atomic_set(&rdev->nr_pending, 0);
3622         atomic_set(&rdev->read_errors, 0);
3623         atomic_set(&rdev->corrected_errors, 0);
3624
3625         INIT_LIST_HEAD(&rdev->same_set);
3626         init_waitqueue_head(&rdev->blocked_wait);
3627
3628         /* Add space to store bad block list.
3629          * This reserves the space even on arrays where it cannot
3630          * be used - I wonder if that matters
3631          */
3632         return badblocks_init(&rdev->badblocks, 0);
3633 }
3634 EXPORT_SYMBOL_GPL(md_rdev_init);
3635 /*
3636  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3637  *
3638  * mark the device faulty if:
3639  *
3640  *   - the device is nonexistent (zero size)
3641  *   - the device has no valid superblock
3642  *
3643  * a faulty rdev _never_ has rdev->sb set.
3644  */
3645 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3646 {
3647         static struct md_rdev claim_rdev; /* just for claiming the bdev */
3648         struct md_rdev *rdev;
3649         sector_t size;
3650         int err;
3651
3652         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3653         if (!rdev)
3654                 return ERR_PTR(-ENOMEM);
3655
3656         err = md_rdev_init(rdev);
3657         if (err)
3658                 goto out_free_rdev;
3659         err = alloc_disk_sb(rdev);
3660         if (err)
3661                 goto out_clear_rdev;
3662
3663         rdev->bdev = blkdev_get_by_dev(newdev,
3664                         FMODE_READ | FMODE_WRITE | FMODE_EXCL,
3665                         super_format == -2 ? &claim_rdev : rdev);
3666         if (IS_ERR(rdev->bdev)) {
3667                 pr_warn("md: could not open device unknown-block(%u,%u).\n",
3668                         MAJOR(newdev), MINOR(newdev));
3669                 err = PTR_ERR(rdev->bdev);
3670                 goto out_clear_rdev;
3671         }
3672
3673         kobject_init(&rdev->kobj, &rdev_ktype);
3674
3675         size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3676         if (!size) {
3677                 pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3678                         rdev->bdev);
3679                 err = -EINVAL;
3680                 goto out_blkdev_put;
3681         }
3682
3683         if (super_format >= 0) {
3684                 err = super_types[super_format].
3685                         load_super(rdev, NULL, super_minor);
3686                 if (err == -EINVAL) {
3687                         pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3688                                 rdev->bdev,
3689                                 super_format, super_minor);
3690                         goto out_blkdev_put;
3691                 }
3692                 if (err < 0) {
3693                         pr_warn("md: could not read %pg's sb, not importing!\n",
3694                                 rdev->bdev);
3695                         goto out_blkdev_put;
3696                 }
3697         }
3698
3699         return rdev;
3700
3701 out_blkdev_put:
3702         blkdev_put(rdev->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3703 out_clear_rdev:
3704         md_rdev_clear(rdev);
3705 out_free_rdev:
3706         kfree(rdev);
3707         return ERR_PTR(err);
3708 }
3709
3710 /*
3711  * Check a full RAID array for plausibility
3712  */
3713
3714 static int analyze_sbs(struct mddev *mddev)
3715 {
3716         int i;
3717         struct md_rdev *rdev, *freshest, *tmp;
3718
3719         freshest = NULL;
3720         rdev_for_each_safe(rdev, tmp, mddev)
3721                 switch (super_types[mddev->major_version].
3722                         load_super(rdev, freshest, mddev->minor_version)) {
3723                 case 1:
3724                         freshest = rdev;
3725                         break;
3726                 case 0:
3727                         break;
3728                 default:
3729                         pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3730                                 rdev->bdev);
3731                         md_kick_rdev_from_array(rdev);
3732                 }
3733
3734         /* Cannot find a valid fresh disk */
3735         if (!freshest) {
3736                 pr_warn("md: cannot find a valid disk\n");
3737                 return -EINVAL;
3738         }
3739
3740         super_types[mddev->major_version].
3741                 validate_super(mddev, freshest);
3742
3743         i = 0;
3744         rdev_for_each_safe(rdev, tmp, mddev) {
3745                 if (mddev->max_disks &&
3746                     (rdev->desc_nr >= mddev->max_disks ||
3747                      i > mddev->max_disks)) {
3748                         pr_warn("md: %s: %pg: only %d devices permitted\n",
3749                                 mdname(mddev), rdev->bdev,
3750                                 mddev->max_disks);
3751                         md_kick_rdev_from_array(rdev);
3752                         continue;
3753                 }
3754                 if (rdev != freshest) {
3755                         if (super_types[mddev->major_version].
3756                             validate_super(mddev, rdev)) {
3757                                 pr_warn("md: kicking non-fresh %pg from array!\n",
3758                                         rdev->bdev);
3759                                 md_kick_rdev_from_array(rdev);
3760                                 continue;
3761                         }
3762                 }
3763                 if (mddev->level == LEVEL_MULTIPATH) {
3764                         rdev->desc_nr = i++;
3765                         rdev->raid_disk = rdev->desc_nr;
3766                         set_bit(In_sync, &rdev->flags);
3767                 } else if (rdev->raid_disk >=
3768                             (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3769                            !test_bit(Journal, &rdev->flags)) {
3770                         rdev->raid_disk = -1;
3771                         clear_bit(In_sync, &rdev->flags);
3772                 }
3773         }
3774
3775         return 0;
3776 }
3777
3778 /* Read a fixed-point number.
3779  * Numbers in sysfs attributes should be in "standard" units where
3780  * possible, so time should be in seconds.
3781  * However we internally use a a much smaller unit such as
3782  * milliseconds or jiffies.
3783  * This function takes a decimal number with a possible fractional
3784  * component, and produces an integer which is the result of
3785  * multiplying that number by 10^'scale'.
3786  * all without any floating-point arithmetic.
3787  */
3788 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3789 {
3790         unsigned long result = 0;
3791         long decimals = -1;
3792         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3793                 if (*cp == '.')
3794                         decimals = 0;
3795                 else if (decimals < scale) {
3796                         unsigned int value;
3797                         value = *cp - '0';
3798                         result = result * 10 + value;
3799                         if (decimals >= 0)
3800                                 decimals++;
3801                 }
3802                 cp++;
3803         }
3804         if (*cp == '\n')
3805                 cp++;
3806         if (*cp)
3807                 return -EINVAL;
3808         if (decimals < 0)
3809                 decimals = 0;
3810         *res = result * int_pow(10, scale - decimals);
3811         return 0;
3812 }
3813
3814 static ssize_t
3815 safe_delay_show(struct mddev *mddev, char *page)
3816 {
3817         int msec = (mddev->safemode_delay*1000)/HZ;
3818         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3819 }
3820 static ssize_t
3821 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3822 {
3823         unsigned long msec;
3824
3825         if (mddev_is_clustered(mddev)) {
3826                 pr_warn("md: Safemode is disabled for clustered mode\n");
3827                 return -EINVAL;
3828         }
3829
3830         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3831                 return -EINVAL;
3832         if (msec == 0)
3833                 mddev->safemode_delay = 0;
3834         else {
3835                 unsigned long old_delay = mddev->safemode_delay;
3836                 unsigned long new_delay = (msec*HZ)/1000;
3837
3838                 if (new_delay == 0)
3839                         new_delay = 1;
3840                 mddev->safemode_delay = new_delay;
3841                 if (new_delay < old_delay || old_delay == 0)
3842                         mod_timer(&mddev->safemode_timer, jiffies+1);
3843         }
3844         return len;
3845 }
3846 static struct md_sysfs_entry md_safe_delay =
3847 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3848
3849 static ssize_t
3850 level_show(struct mddev *mddev, char *page)
3851 {
3852         struct md_personality *p;
3853         int ret;
3854         spin_lock(&mddev->lock);
3855         p = mddev->pers;
3856         if (p)
3857                 ret = sprintf(page, "%s\n", p->name);
3858         else if (mddev->clevel[0])
3859                 ret = sprintf(page, "%s\n", mddev->clevel);
3860         else if (mddev->level != LEVEL_NONE)
3861                 ret = sprintf(page, "%d\n", mddev->level);
3862         else
3863                 ret = 0;
3864         spin_unlock(&mddev->lock);
3865         return ret;
3866 }
3867
3868 static ssize_t
3869 level_store(struct mddev *mddev, const char *buf, size_t len)
3870 {
3871         char clevel[16];
3872         ssize_t rv;
3873         size_t slen = len;
3874         struct md_personality *pers, *oldpers;
3875         long level;
3876         void *priv, *oldpriv;
3877         struct md_rdev *rdev;
3878
3879         if (slen == 0 || slen >= sizeof(clevel))
3880                 return -EINVAL;
3881
3882         rv = mddev_lock(mddev);
3883         if (rv)
3884                 return rv;
3885
3886         if (mddev->pers == NULL) {
3887                 strncpy(mddev->clevel, buf, slen);
3888                 if (mddev->clevel[slen-1] == '\n')
3889                         slen--;
3890                 mddev->clevel[slen] = 0;
3891                 mddev->level = LEVEL_NONE;
3892                 rv = len;
3893                 goto out_unlock;
3894         }
3895         rv = -EROFS;
3896         if (!md_is_rdwr(mddev))
3897                 goto out_unlock;
3898
3899         /* request to change the personality.  Need to ensure:
3900          *  - array is not engaged in resync/recovery/reshape
3901          *  - old personality can be suspended
3902          *  - new personality will access other array.
3903          */
3904
3905         rv = -EBUSY;
3906         if (mddev->sync_thread ||
3907             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3908             mddev->reshape_position != MaxSector ||
3909             mddev->sysfs_active)
3910                 goto out_unlock;
3911
3912         rv = -EINVAL;
3913         if (!mddev->pers->quiesce) {
3914                 pr_warn("md: %s: %s does not support online personality change\n",
3915                         mdname(mddev), mddev->pers->name);
3916                 goto out_unlock;
3917         }
3918
3919         /* Now find the new personality */
3920         strncpy(clevel, buf, slen);
3921         if (clevel[slen-1] == '\n')
3922                 slen--;
3923         clevel[slen] = 0;
3924         if (kstrtol(clevel, 10, &level))
3925                 level = LEVEL_NONE;
3926
3927         if (request_module("md-%s", clevel) != 0)
3928                 request_module("md-level-%s", clevel);
3929         spin_lock(&pers_lock);
3930         pers = find_pers(level, clevel);
3931         if (!pers || !try_module_get(pers->owner)) {
3932                 spin_unlock(&pers_lock);
3933                 pr_warn("md: personality %s not loaded\n", clevel);
3934                 rv = -EINVAL;
3935                 goto out_unlock;
3936         }
3937         spin_unlock(&pers_lock);
3938
3939         if (pers == mddev->pers) {
3940                 /* Nothing to do! */
3941                 module_put(pers->owner);
3942                 rv = len;
3943                 goto out_unlock;
3944         }
3945         if (!pers->takeover) {
3946                 module_put(pers->owner);
3947                 pr_warn("md: %s: %s does not support personality takeover\n",
3948                         mdname(mddev), clevel);
3949                 rv = -EINVAL;
3950                 goto out_unlock;
3951         }
3952
3953         rdev_for_each(rdev, mddev)
3954                 rdev->new_raid_disk = rdev->raid_disk;
3955
3956         /* ->takeover must set new_* and/or delta_disks
3957          * if it succeeds, and may set them when it fails.
3958          */
3959         priv = pers->takeover(mddev);
3960         if (IS_ERR(priv)) {
3961                 mddev->new_level = mddev->level;
3962                 mddev->new_layout = mddev->layout;
3963                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3964                 mddev->raid_disks -= mddev->delta_disks;
3965                 mddev->delta_disks = 0;
3966                 mddev->reshape_backwards = 0;
3967                 module_put(pers->owner);
3968                 pr_warn("md: %s: %s would not accept array\n",
3969                         mdname(mddev), clevel);
3970                 rv = PTR_ERR(priv);
3971                 goto out_unlock;
3972         }
3973
3974         /* Looks like we have a winner */
3975         mddev_suspend(mddev);
3976         mddev_detach(mddev);
3977
3978         spin_lock(&mddev->lock);
3979         oldpers = mddev->pers;
3980         oldpriv = mddev->private;
3981         mddev->pers = pers;
3982         mddev->private = priv;
3983         strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3984         mddev->level = mddev->new_level;
3985         mddev->layout = mddev->new_layout;
3986         mddev->chunk_sectors = mddev->new_chunk_sectors;
3987         mddev->delta_disks = 0;
3988         mddev->reshape_backwards = 0;
3989         mddev->degraded = 0;
3990         spin_unlock(&mddev->lock);
3991
3992         if (oldpers->sync_request == NULL &&
3993             mddev->external) {
3994                 /* We are converting from a no-redundancy array
3995                  * to a redundancy array and metadata is managed
3996                  * externally so we need to be sure that writes
3997                  * won't block due to a need to transition
3998                  *      clean->dirty
3999                  * until external management is started.
4000                  */
4001                 mddev->in_sync = 0;
4002                 mddev->safemode_delay = 0;
4003                 mddev->safemode = 0;
4004         }
4005
4006         oldpers->free(mddev, oldpriv);
4007
4008         if (oldpers->sync_request == NULL &&
4009             pers->sync_request != NULL) {
4010                 /* need to add the md_redundancy_group */
4011                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4012                         pr_warn("md: cannot register extra attributes for %s\n",
4013                                 mdname(mddev));
4014                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4015                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4016                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4017         }
4018         if (oldpers->sync_request != NULL &&
4019             pers->sync_request == NULL) {
4020                 /* need to remove the md_redundancy_group */
4021                 if (mddev->to_remove == NULL)
4022                         mddev->to_remove = &md_redundancy_group;
4023         }
4024
4025         module_put(oldpers->owner);
4026
4027         rdev_for_each(rdev, mddev) {
4028                 if (rdev->raid_disk < 0)
4029                         continue;
4030                 if (rdev->new_raid_disk >= mddev->raid_disks)
4031                         rdev->new_raid_disk = -1;
4032                 if (rdev->new_raid_disk == rdev->raid_disk)
4033                         continue;
4034                 sysfs_unlink_rdev(mddev, rdev);
4035         }
4036         rdev_for_each(rdev, mddev) {
4037                 if (rdev->raid_disk < 0)
4038                         continue;
4039                 if (rdev->new_raid_disk == rdev->raid_disk)
4040                         continue;
4041                 rdev->raid_disk = rdev->new_raid_disk;
4042                 if (rdev->raid_disk < 0)
4043                         clear_bit(In_sync, &rdev->flags);
4044                 else {
4045                         if (sysfs_link_rdev(mddev, rdev))
4046                                 pr_warn("md: cannot register rd%d for %s after level change\n",
4047                                         rdev->raid_disk, mdname(mddev));
4048                 }
4049         }
4050
4051         if (pers->sync_request == NULL) {
4052                 /* this is now an array without redundancy, so
4053                  * it must always be in_sync
4054                  */
4055                 mddev->in_sync = 1;
4056                 del_timer_sync(&mddev->safemode_timer);
4057         }
4058         blk_set_stacking_limits(&mddev->queue->limits);
4059         pers->run(mddev);
4060         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4061         mddev_resume(mddev);
4062         if (!mddev->thread)
4063                 md_update_sb(mddev, 1);
4064         sysfs_notify_dirent_safe(mddev->sysfs_level);
4065         md_new_event();
4066         rv = len;
4067 out_unlock:
4068         mddev_unlock(mddev);
4069         return rv;
4070 }
4071
4072 static struct md_sysfs_entry md_level =
4073 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4074
4075 static ssize_t
4076 layout_show(struct mddev *mddev, char *page)
4077 {
4078         /* just a number, not meaningful for all levels */
4079         if (mddev->reshape_position != MaxSector &&
4080             mddev->layout != mddev->new_layout)
4081                 return sprintf(page, "%d (%d)\n",
4082                                mddev->new_layout, mddev->layout);
4083         return sprintf(page, "%d\n", mddev->layout);
4084 }
4085
4086 static ssize_t
4087 layout_store(struct mddev *mddev, const char *buf, size_t len)
4088 {
4089         unsigned int n;
4090         int err;
4091
4092         err = kstrtouint(buf, 10, &n);
4093         if (err < 0)
4094                 return err;
4095         err = mddev_lock(mddev);
4096         if (err)
4097                 return err;
4098
4099         if (mddev->pers) {
4100                 if (mddev->pers->check_reshape == NULL)
4101                         err = -EBUSY;
4102                 else if (!md_is_rdwr(mddev))
4103                         err = -EROFS;
4104                 else {
4105                         mddev->new_layout = n;
4106                         err = mddev->pers->check_reshape(mddev);
4107                         if (err)
4108                                 mddev->new_layout = mddev->layout;
4109                 }
4110         } else {
4111                 mddev->new_layout = n;
4112                 if (mddev->reshape_position == MaxSector)
4113                         mddev->layout = n;
4114         }
4115         mddev_unlock(mddev);
4116         return err ?: len;
4117 }
4118 static struct md_sysfs_entry md_layout =
4119 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4120
4121 static ssize_t
4122 raid_disks_show(struct mddev *mddev, char *page)
4123 {
4124         if (mddev->raid_disks == 0)
4125                 return 0;
4126         if (mddev->reshape_position != MaxSector &&
4127             mddev->delta_disks != 0)
4128                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4129                                mddev->raid_disks - mddev->delta_disks);
4130         return sprintf(page, "%d\n", mddev->raid_disks);
4131 }
4132
4133 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4134
4135 static ssize_t
4136 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4137 {
4138         unsigned int n;
4139         int err;
4140
4141         err = kstrtouint(buf, 10, &n);
4142         if (err < 0)
4143                 return err;
4144
4145         err = mddev_lock(mddev);
4146         if (err)
4147                 return err;
4148         if (mddev->pers)
4149                 err = update_raid_disks(mddev, n);
4150         else if (mddev->reshape_position != MaxSector) {
4151                 struct md_rdev *rdev;
4152                 int olddisks = mddev->raid_disks - mddev->delta_disks;
4153
4154                 err = -EINVAL;
4155                 rdev_for_each(rdev, mddev) {
4156                         if (olddisks < n &&
4157                             rdev->data_offset < rdev->new_data_offset)
4158                                 goto out_unlock;
4159                         if (olddisks > n &&
4160                             rdev->data_offset > rdev->new_data_offset)
4161                                 goto out_unlock;
4162                 }
4163                 err = 0;
4164                 mddev->delta_disks = n - olddisks;
4165                 mddev->raid_disks = n;
4166                 mddev->reshape_backwards = (mddev->delta_disks < 0);
4167         } else
4168                 mddev->raid_disks = n;
4169 out_unlock:
4170         mddev_unlock(mddev);
4171         return err ? err : len;
4172 }
4173 static struct md_sysfs_entry md_raid_disks =
4174 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4175
4176 static ssize_t
4177 uuid_show(struct mddev *mddev, char *page)
4178 {
4179         return sprintf(page, "%pU\n", mddev->uuid);
4180 }
4181 static struct md_sysfs_entry md_uuid =
4182 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4183
4184 static ssize_t
4185 chunk_size_show(struct mddev *mddev, char *page)
4186 {
4187         if (mddev->reshape_position != MaxSector &&
4188             mddev->chunk_sectors != mddev->new_chunk_sectors)
4189                 return sprintf(page, "%d (%d)\n",
4190                                mddev->new_chunk_sectors << 9,
4191                                mddev->chunk_sectors << 9);
4192         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4193 }
4194
4195 static ssize_t
4196 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4197 {
4198         unsigned long n;
4199         int err;
4200
4201         err = kstrtoul(buf, 10, &n);
4202         if (err < 0)
4203                 return err;
4204
4205         err = mddev_lock(mddev);
4206         if (err)
4207                 return err;
4208         if (mddev->pers) {
4209                 if (mddev->pers->check_reshape == NULL)
4210                         err = -EBUSY;
4211                 else if (!md_is_rdwr(mddev))
4212                         err = -EROFS;
4213                 else {
4214                         mddev->new_chunk_sectors = n >> 9;
4215                         err = mddev->pers->check_reshape(mddev);
4216                         if (err)
4217                                 mddev->new_chunk_sectors = mddev->chunk_sectors;
4218                 }
4219         } else {
4220                 mddev->new_chunk_sectors = n >> 9;
4221                 if (mddev->reshape_position == MaxSector)
4222                         mddev->chunk_sectors = n >> 9;
4223         }
4224         mddev_unlock(mddev);
4225         return err ?: len;
4226 }
4227 static struct md_sysfs_entry md_chunk_size =
4228 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4229
4230 static ssize_t
4231 resync_start_show(struct mddev *mddev, char *page)
4232 {
4233         if (mddev->recovery_cp == MaxSector)
4234                 return sprintf(page, "none\n");
4235         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4236 }
4237
4238 static ssize_t
4239 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4240 {
4241         unsigned long long n;
4242         int err;
4243
4244         if (cmd_match(buf, "none"))
4245                 n = MaxSector;
4246         else {
4247                 err = kstrtoull(buf, 10, &n);
4248                 if (err < 0)
4249                         return err;
4250                 if (n != (sector_t)n)
4251                         return -EINVAL;
4252         }
4253
4254         err = mddev_lock(mddev);
4255         if (err)
4256                 return err;
4257         if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4258                 err = -EBUSY;
4259
4260         if (!err) {
4261                 mddev->recovery_cp = n;
4262                 if (mddev->pers)
4263                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4264         }
4265         mddev_unlock(mddev);
4266         return err ?: len;
4267 }
4268 static struct md_sysfs_entry md_resync_start =
4269 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4270                 resync_start_show, resync_start_store);
4271
4272 /*
4273  * The array state can be:
4274  *
4275  * clear
4276  *     No devices, no size, no level
4277  *     Equivalent to STOP_ARRAY ioctl
4278  * inactive
4279  *     May have some settings, but array is not active
4280  *        all IO results in error
4281  *     When written, doesn't tear down array, but just stops it
4282  * suspended (not supported yet)
4283  *     All IO requests will block. The array can be reconfigured.
4284  *     Writing this, if accepted, will block until array is quiescent
4285  * readonly
4286  *     no resync can happen.  no superblocks get written.
4287  *     write requests fail
4288  * read-auto
4289  *     like readonly, but behaves like 'clean' on a write request.
4290  *
4291  * clean - no pending writes, but otherwise active.
4292  *     When written to inactive array, starts without resync
4293  *     If a write request arrives then
4294  *       if metadata is known, mark 'dirty' and switch to 'active'.
4295  *       if not known, block and switch to write-pending
4296  *     If written to an active array that has pending writes, then fails.
4297  * active
4298  *     fully active: IO and resync can be happening.
4299  *     When written to inactive array, starts with resync
4300  *
4301  * write-pending
4302  *     clean, but writes are blocked waiting for 'active' to be written.
4303  *
4304  * active-idle
4305  *     like active, but no writes have been seen for a while (100msec).
4306  *
4307  * broken
4308 *     Array is failed. It's useful because mounted-arrays aren't stopped
4309 *     when array is failed, so this state will at least alert the user that
4310 *     something is wrong.
4311  */
4312 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4313                    write_pending, active_idle, broken, bad_word};
4314 static char *array_states[] = {
4315         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4316         "write-pending", "active-idle", "broken", NULL };
4317
4318 static int match_word(const char *word, char **list)
4319 {
4320         int n;
4321         for (n=0; list[n]; n++)
4322                 if (cmd_match(word, list[n]))
4323                         break;
4324         return n;
4325 }
4326
4327 static ssize_t
4328 array_state_show(struct mddev *mddev, char *page)
4329 {
4330         enum array_state st = inactive;
4331
4332         if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4333                 switch(mddev->ro) {
4334                 case MD_RDONLY:
4335                         st = readonly;
4336                         break;
4337                 case MD_AUTO_READ:
4338                         st = read_auto;
4339                         break;
4340                 case MD_RDWR:
4341                         spin_lock(&mddev->lock);
4342                         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4343                                 st = write_pending;
4344                         else if (mddev->in_sync)
4345                                 st = clean;
4346                         else if (mddev->safemode)
4347                                 st = active_idle;
4348                         else
4349                                 st = active;
4350                         spin_unlock(&mddev->lock);
4351                 }
4352
4353                 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4354                         st = broken;
4355         } else {
4356                 if (list_empty(&mddev->disks) &&
4357                     mddev->raid_disks == 0 &&
4358                     mddev->dev_sectors == 0)
4359                         st = clear;
4360                 else
4361                         st = inactive;
4362         }
4363         return sprintf(page, "%s\n", array_states[st]);
4364 }
4365
4366 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4367 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4368 static int restart_array(struct mddev *mddev);
4369
4370 static ssize_t
4371 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4372 {
4373         int err = 0;
4374         enum array_state st = match_word(buf, array_states);
4375
4376         if (mddev->pers && (st == active || st == clean) &&
4377             mddev->ro != MD_RDONLY) {
4378                 /* don't take reconfig_mutex when toggling between
4379                  * clean and active
4380                  */
4381                 spin_lock(&mddev->lock);
4382                 if (st == active) {
4383                         restart_array(mddev);
4384                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4385                         md_wakeup_thread(mddev->thread);
4386                         wake_up(&mddev->sb_wait);
4387                 } else /* st == clean */ {
4388                         restart_array(mddev);
4389                         if (!set_in_sync(mddev))
4390                                 err = -EBUSY;
4391                 }
4392                 if (!err)
4393                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4394                 spin_unlock(&mddev->lock);
4395                 return err ?: len;
4396         }
4397         err = mddev_lock(mddev);
4398         if (err)
4399                 return err;
4400         err = -EINVAL;
4401         switch(st) {
4402         case bad_word:
4403                 break;
4404         case clear:
4405                 /* stopping an active array */
4406                 err = do_md_stop(mddev, 0, NULL);
4407                 break;
4408         case inactive:
4409                 /* stopping an active array */
4410                 if (mddev->pers)
4411                         err = do_md_stop(mddev, 2, NULL);
4412                 else
4413                         err = 0; /* already inactive */
4414                 break;
4415         case suspended:
4416                 break; /* not supported yet */
4417         case readonly:
4418                 if (mddev->pers)
4419                         err = md_set_readonly(mddev, NULL);
4420                 else {
4421                         mddev->ro = MD_RDONLY;
4422                         set_disk_ro(mddev->gendisk, 1);
4423                         err = do_md_run(mddev);
4424                 }
4425                 break;
4426         case read_auto:
4427                 if (mddev->pers) {
4428                         if (md_is_rdwr(mddev))
4429                                 err = md_set_readonly(mddev, NULL);
4430                         else if (mddev->ro == MD_RDONLY)
4431                                 err = restart_array(mddev);
4432                         if (err == 0) {
4433                                 mddev->ro = MD_AUTO_READ;
4434                                 set_disk_ro(mddev->gendisk, 0);
4435                         }
4436                 } else {
4437                         mddev->ro = MD_AUTO_READ;
4438                         err = do_md_run(mddev);
4439                 }
4440                 break;
4441         case clean:
4442                 if (mddev->pers) {
4443                         err = restart_array(mddev);
4444                         if (err)
4445                                 break;
4446                         spin_lock(&mddev->lock);
4447                         if (!set_in_sync(mddev))
4448                                 err = -EBUSY;
4449                         spin_unlock(&mddev->lock);
4450                 } else
4451                         err = -EINVAL;
4452                 break;
4453         case active:
4454                 if (mddev->pers) {
4455                         err = restart_array(mddev);
4456                         if (err)
4457                                 break;
4458                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4459                         wake_up(&mddev->sb_wait);
4460                         err = 0;
4461                 } else {
4462                         mddev->ro = MD_RDWR;
4463                         set_disk_ro(mddev->gendisk, 0);
4464                         err = do_md_run(mddev);
4465                 }
4466                 break;
4467         case write_pending:
4468         case active_idle:
4469         case broken:
4470                 /* these cannot be set */
4471                 break;
4472         }
4473
4474         if (!err) {
4475                 if (mddev->hold_active == UNTIL_IOCTL)
4476                         mddev->hold_active = 0;
4477                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4478         }
4479         mddev_unlock(mddev);
4480         return err ?: len;
4481 }
4482 static struct md_sysfs_entry md_array_state =
4483 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4484
4485 static ssize_t
4486 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4487         return sprintf(page, "%d\n",
4488                        atomic_read(&mddev->max_corr_read_errors));
4489 }
4490
4491 static ssize_t
4492 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4493 {
4494         unsigned int n;
4495         int rv;
4496
4497         rv = kstrtouint(buf, 10, &n);
4498         if (rv < 0)
4499                 return rv;
4500         atomic_set(&mddev->max_corr_read_errors, n);
4501         return len;
4502 }
4503
4504 static struct md_sysfs_entry max_corr_read_errors =
4505 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4506         max_corrected_read_errors_store);
4507
4508 static ssize_t
4509 null_show(struct mddev *mddev, char *page)
4510 {
4511         return -EINVAL;
4512 }
4513
4514 /* need to ensure rdev_delayed_delete() has completed */
4515 static void flush_rdev_wq(struct mddev *mddev)
4516 {
4517         struct md_rdev *rdev;
4518
4519         rcu_read_lock();
4520         rdev_for_each_rcu(rdev, mddev)
4521                 if (work_pending(&rdev->del_work)) {
4522                         flush_workqueue(md_rdev_misc_wq);
4523                         break;
4524                 }
4525         rcu_read_unlock();
4526 }
4527
4528 static ssize_t
4529 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4530 {
4531         /* buf must be %d:%d\n? giving major and minor numbers */
4532         /* The new device is added to the array.
4533          * If the array has a persistent superblock, we read the
4534          * superblock to initialise info and check validity.
4535          * Otherwise, only checking done is that in bind_rdev_to_array,
4536          * which mainly checks size.
4537          */
4538         char *e;
4539         int major = simple_strtoul(buf, &e, 10);
4540         int minor;
4541         dev_t dev;
4542         struct md_rdev *rdev;
4543         int err;
4544
4545         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4546                 return -EINVAL;
4547         minor = simple_strtoul(e+1, &e, 10);
4548         if (*e && *e != '\n')
4549                 return -EINVAL;
4550         dev = MKDEV(major, minor);
4551         if (major != MAJOR(dev) ||
4552             minor != MINOR(dev))
4553                 return -EOVERFLOW;
4554
4555         flush_rdev_wq(mddev);
4556         err = mddev_lock(mddev);
4557         if (err)
4558                 return err;
4559         if (mddev->persistent) {
4560                 rdev = md_import_device(dev, mddev->major_version,
4561                                         mddev->minor_version);
4562                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4563                         struct md_rdev *rdev0
4564                                 = list_entry(mddev->disks.next,
4565                                              struct md_rdev, same_set);
4566                         err = super_types[mddev->major_version]
4567                                 .load_super(rdev, rdev0, mddev->minor_version);
4568                         if (err < 0)
4569                                 goto out;
4570                 }
4571         } else if (mddev->external)
4572                 rdev = md_import_device(dev, -2, -1);
4573         else
4574                 rdev = md_import_device(dev, -1, -1);
4575
4576         if (IS_ERR(rdev)) {
4577                 mddev_unlock(mddev);
4578                 return PTR_ERR(rdev);
4579         }
4580         err = bind_rdev_to_array(rdev, mddev);
4581  out:
4582         if (err)
4583                 export_rdev(rdev);
4584         mddev_unlock(mddev);
4585         if (!err)
4586                 md_new_event();
4587         return err ? err : len;
4588 }
4589
4590 static struct md_sysfs_entry md_new_device =
4591 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4592
4593 static ssize_t
4594 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4595 {
4596         char *end;
4597         unsigned long chunk, end_chunk;
4598         int err;
4599
4600         err = mddev_lock(mddev);
4601         if (err)
4602                 return err;
4603         if (!mddev->bitmap)
4604                 goto out;
4605         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4606         while (*buf) {
4607                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4608                 if (buf == end) break;
4609                 if (*end == '-') { /* range */
4610                         buf = end + 1;
4611                         end_chunk = simple_strtoul(buf, &end, 0);
4612                         if (buf == end) break;
4613                 }
4614                 if (*end && !isspace(*end)) break;
4615                 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4616                 buf = skip_spaces(end);
4617         }
4618         md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4619 out:
4620         mddev_unlock(mddev);
4621         return len;
4622 }
4623
4624 static struct md_sysfs_entry md_bitmap =
4625 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4626
4627 static ssize_t
4628 size_show(struct mddev *mddev, char *page)
4629 {
4630         return sprintf(page, "%llu\n",
4631                 (unsigned long long)mddev->dev_sectors / 2);
4632 }
4633
4634 static int update_size(struct mddev *mddev, sector_t num_sectors);
4635
4636 static ssize_t
4637 size_store(struct mddev *mddev, const char *buf, size_t len)
4638 {
4639         /* If array is inactive, we can reduce the component size, but
4640          * not increase it (except from 0).
4641          * If array is active, we can try an on-line resize
4642          */
4643         sector_t sectors;
4644         int err = strict_blocks_to_sectors(buf, &sectors);
4645
4646         if (err < 0)
4647                 return err;
4648         err = mddev_lock(mddev);
4649         if (err)
4650                 return err;
4651         if (mddev->pers) {
4652                 err = update_size(mddev, sectors);
4653                 if (err == 0)
4654                         md_update_sb(mddev, 1);
4655         } else {
4656                 if (mddev->dev_sectors == 0 ||
4657                     mddev->dev_sectors > sectors)
4658                         mddev->dev_sectors = sectors;
4659                 else
4660                         err = -ENOSPC;
4661         }
4662         mddev_unlock(mddev);
4663         return err ? err : len;
4664 }
4665
4666 static struct md_sysfs_entry md_size =
4667 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4668
4669 /* Metadata version.
4670  * This is one of
4671  *   'none' for arrays with no metadata (good luck...)
4672  *   'external' for arrays with externally managed metadata,
4673  * or N.M for internally known formats
4674  */
4675 static ssize_t
4676 metadata_show(struct mddev *mddev, char *page)
4677 {
4678         if (mddev->persistent)
4679                 return sprintf(page, "%d.%d\n",
4680                                mddev->major_version, mddev->minor_version);
4681         else if (mddev->external)
4682                 return sprintf(page, "external:%s\n", mddev->metadata_type);
4683         else
4684                 return sprintf(page, "none\n");
4685 }
4686
4687 static ssize_t
4688 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4689 {
4690         int major, minor;
4691         char *e;
4692         int err;
4693         /* Changing the details of 'external' metadata is
4694          * always permitted.  Otherwise there must be
4695          * no devices attached to the array.
4696          */
4697
4698         err = mddev_lock(mddev);
4699         if (err)
4700                 return err;
4701         err = -EBUSY;
4702         if (mddev->external && strncmp(buf, "external:", 9) == 0)
4703                 ;
4704         else if (!list_empty(&mddev->disks))
4705                 goto out_unlock;
4706
4707         err = 0;
4708         if (cmd_match(buf, "none")) {
4709                 mddev->persistent = 0;
4710                 mddev->external = 0;
4711                 mddev->major_version = 0;
4712                 mddev->minor_version = 90;
4713                 goto out_unlock;
4714         }
4715         if (strncmp(buf, "external:", 9) == 0) {
4716                 size_t namelen = len-9;
4717                 if (namelen >= sizeof(mddev->metadata_type))
4718                         namelen = sizeof(mddev->metadata_type)-1;
4719                 strncpy(mddev->metadata_type, buf+9, namelen);
4720                 mddev->metadata_type[namelen] = 0;
4721                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4722                         mddev->metadata_type[--namelen] = 0;
4723                 mddev->persistent = 0;
4724                 mddev->external = 1;
4725                 mddev->major_version = 0;
4726                 mddev->minor_version = 90;
4727                 goto out_unlock;
4728         }
4729         major = simple_strtoul(buf, &e, 10);
4730         err = -EINVAL;
4731         if (e==buf || *e != '.')
4732                 goto out_unlock;
4733         buf = e+1;
4734         minor = simple_strtoul(buf, &e, 10);
4735         if (e==buf || (*e && *e != '\n') )
4736                 goto out_unlock;
4737         err = -ENOENT;
4738         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4739                 goto out_unlock;
4740         mddev->major_version = major;
4741         mddev->minor_version = minor;
4742         mddev->persistent = 1;
4743         mddev->external = 0;
4744         err = 0;
4745 out_unlock:
4746         mddev_unlock(mddev);
4747         return err ?: len;
4748 }
4749
4750 static struct md_sysfs_entry md_metadata =
4751 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4752
4753 static ssize_t
4754 action_show(struct mddev *mddev, char *page)
4755 {
4756         char *type = "idle";
4757         unsigned long recovery = mddev->recovery;
4758         if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4759                 type = "frozen";
4760         else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4761             (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4762                 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4763                         type = "reshape";
4764                 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4765                         if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4766                                 type = "resync";
4767                         else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4768                                 type = "check";
4769                         else
4770                                 type = "repair";
4771                 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4772                         type = "recover";
4773                 else if (mddev->reshape_position != MaxSector)
4774                         type = "reshape";
4775         }
4776         return sprintf(page, "%s\n", type);
4777 }
4778
4779 static ssize_t
4780 action_store(struct mddev *mddev, const char *page, size_t len)
4781 {
4782         if (!mddev->pers || !mddev->pers->sync_request)
4783                 return -EINVAL;
4784
4785
4786         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4787                 if (cmd_match(page, "frozen"))
4788                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4789                 else
4790                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4791                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4792                     mddev_lock(mddev) == 0) {
4793                         if (work_pending(&mddev->del_work))
4794                                 flush_workqueue(md_misc_wq);
4795                         if (mddev->sync_thread) {
4796                                 sector_t save_rp = mddev->reshape_position;
4797
4798                                 mddev_unlock(mddev);
4799                                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4800                                 md_unregister_thread(&mddev->sync_thread);
4801                                 mddev_lock_nointr(mddev);
4802                                 /*
4803                                  * set RECOVERY_INTR again and restore reshape
4804                                  * position in case others changed them after
4805                                  * got lock, eg, reshape_position_store and
4806                                  * md_check_recovery.
4807                                  */
4808                                 mddev->reshape_position = save_rp;
4809                                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4810                                 md_reap_sync_thread(mddev);
4811                         }
4812                         mddev_unlock(mddev);
4813                 }
4814         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4815                 return -EBUSY;
4816         else if (cmd_match(page, "resync"))
4817                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4818         else if (cmd_match(page, "recover")) {
4819                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4820                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4821         } else if (cmd_match(page, "reshape")) {
4822                 int err;
4823                 if (mddev->pers->start_reshape == NULL)
4824                         return -EINVAL;
4825                 err = mddev_lock(mddev);
4826                 if (!err) {
4827                         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4828                                 err =  -EBUSY;
4829                         else {
4830                                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4831                                 err = mddev->pers->start_reshape(mddev);
4832                         }
4833                         mddev_unlock(mddev);
4834                 }
4835                 if (err)
4836                         return err;
4837                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4838         } else {
4839                 if (cmd_match(page, "check"))
4840                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4841                 else if (!cmd_match(page, "repair"))
4842                         return -EINVAL;
4843                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4844                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4845                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4846         }
4847         if (mddev->ro == MD_AUTO_READ) {
4848                 /* A write to sync_action is enough to justify
4849                  * canceling read-auto mode
4850                  */
4851                 mddev->ro = MD_RDWR;
4852                 md_wakeup_thread(mddev->sync_thread);
4853         }
4854         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4855         md_wakeup_thread(mddev->thread);
4856         sysfs_notify_dirent_safe(mddev->sysfs_action);
4857         return len;
4858 }
4859
4860 static struct md_sysfs_entry md_scan_mode =
4861 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4862
4863 static ssize_t
4864 last_sync_action_show(struct mddev *mddev, char *page)
4865 {
4866         return sprintf(page, "%s\n", mddev->last_sync_action);
4867 }
4868
4869 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4870
4871 static ssize_t
4872 mismatch_cnt_show(struct mddev *mddev, char *page)
4873 {
4874         return sprintf(page, "%llu\n",
4875                        (unsigned long long)
4876                        atomic64_read(&mddev->resync_mismatches));
4877 }
4878
4879 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4880
4881 static ssize_t
4882 sync_min_show(struct mddev *mddev, char *page)
4883 {
4884         return sprintf(page, "%d (%s)\n", speed_min(mddev),
4885                        mddev->sync_speed_min ? "local": "system");
4886 }
4887
4888 static ssize_t
4889 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4890 {
4891         unsigned int min;
4892         int rv;
4893
4894         if (strncmp(buf, "system", 6)==0) {
4895                 min = 0;
4896         } else {
4897                 rv = kstrtouint(buf, 10, &min);
4898                 if (rv < 0)
4899                         return rv;
4900                 if (min == 0)
4901                         return -EINVAL;
4902         }
4903         mddev->sync_speed_min = min;
4904         return len;
4905 }
4906
4907 static struct md_sysfs_entry md_sync_min =
4908 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4909
4910 static ssize_t
4911 sync_max_show(struct mddev *mddev, char *page)
4912 {
4913         return sprintf(page, "%d (%s)\n", speed_max(mddev),
4914                        mddev->sync_speed_max ? "local": "system");
4915 }
4916
4917 static ssize_t
4918 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4919 {
4920         unsigned int max;
4921         int rv;
4922
4923         if (strncmp(buf, "system", 6)==0) {
4924                 max = 0;
4925         } else {
4926                 rv = kstrtouint(buf, 10, &max);
4927                 if (rv < 0)
4928                         return rv;
4929                 if (max == 0)
4930                         return -EINVAL;
4931         }
4932         mddev->sync_speed_max = max;
4933         return len;
4934 }
4935
4936 static struct md_sysfs_entry md_sync_max =
4937 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4938
4939 static ssize_t
4940 degraded_show(struct mddev *mddev, char *page)
4941 {
4942         return sprintf(page, "%d\n", mddev->degraded);
4943 }
4944 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4945
4946 static ssize_t
4947 sync_force_parallel_show(struct mddev *mddev, char *page)
4948 {
4949         return sprintf(page, "%d\n", mddev->parallel_resync);
4950 }
4951
4952 static ssize_t
4953 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4954 {
4955         long n;
4956
4957         if (kstrtol(buf, 10, &n))
4958                 return -EINVAL;
4959
4960         if (n != 0 && n != 1)
4961                 return -EINVAL;
4962
4963         mddev->parallel_resync = n;
4964
4965         if (mddev->sync_thread)
4966                 wake_up(&resync_wait);
4967
4968         return len;
4969 }
4970
4971 /* force parallel resync, even with shared block devices */
4972 static struct md_sysfs_entry md_sync_force_parallel =
4973 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4974        sync_force_parallel_show, sync_force_parallel_store);
4975
4976 static ssize_t
4977 sync_speed_show(struct mddev *mddev, char *page)
4978 {
4979         unsigned long resync, dt, db;
4980         if (mddev->curr_resync == MD_RESYNC_NONE)
4981                 return sprintf(page, "none\n");
4982         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4983         dt = (jiffies - mddev->resync_mark) / HZ;
4984         if (!dt) dt++;
4985         db = resync - mddev->resync_mark_cnt;
4986         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4987 }
4988
4989 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4990
4991 static ssize_t
4992 sync_completed_show(struct mddev *mddev, char *page)
4993 {
4994         unsigned long long max_sectors, resync;
4995
4996         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4997                 return sprintf(page, "none\n");
4998
4999         if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5000             mddev->curr_resync == MD_RESYNC_DELAYED)
5001                 return sprintf(page, "delayed\n");
5002
5003         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5004             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5005                 max_sectors = mddev->resync_max_sectors;
5006         else
5007                 max_sectors = mddev->dev_sectors;
5008
5009         resync = mddev->curr_resync_completed;
5010         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5011 }
5012
5013 static struct md_sysfs_entry md_sync_completed =
5014         __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5015
5016 static ssize_t
5017 min_sync_show(struct mddev *mddev, char *page)
5018 {
5019         return sprintf(page, "%llu\n",
5020                        (unsigned long long)mddev->resync_min);
5021 }
5022 static ssize_t
5023 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5024 {
5025         unsigned long long min;
5026         int err;
5027
5028         if (kstrtoull(buf, 10, &min))
5029                 return -EINVAL;
5030
5031         spin_lock(&mddev->lock);
5032         err = -EINVAL;
5033         if (min > mddev->resync_max)
5034                 goto out_unlock;
5035
5036         err = -EBUSY;
5037         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5038                 goto out_unlock;
5039
5040         /* Round down to multiple of 4K for safety */
5041         mddev->resync_min = round_down(min, 8);
5042         err = 0;
5043
5044 out_unlock:
5045         spin_unlock(&mddev->lock);
5046         return err ?: len;
5047 }
5048
5049 static struct md_sysfs_entry md_min_sync =
5050 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5051
5052 static ssize_t
5053 max_sync_show(struct mddev *mddev, char *page)
5054 {
5055         if (mddev->resync_max == MaxSector)
5056                 return sprintf(page, "max\n");
5057         else
5058                 return sprintf(page, "%llu\n",
5059                                (unsigned long long)mddev->resync_max);
5060 }
5061 static ssize_t
5062 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5063 {
5064         int err;
5065         spin_lock(&mddev->lock);
5066         if (strncmp(buf, "max", 3) == 0)
5067                 mddev->resync_max = MaxSector;
5068         else {
5069                 unsigned long long max;
5070                 int chunk;
5071
5072                 err = -EINVAL;
5073                 if (kstrtoull(buf, 10, &max))
5074                         goto out_unlock;
5075                 if (max < mddev->resync_min)
5076                         goto out_unlock;
5077
5078                 err = -EBUSY;
5079                 if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5080                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5081                         goto out_unlock;
5082
5083                 /* Must be a multiple of chunk_size */
5084                 chunk = mddev->chunk_sectors;
5085                 if (chunk) {
5086                         sector_t temp = max;
5087
5088                         err = -EINVAL;
5089                         if (sector_div(temp, chunk))
5090                                 goto out_unlock;
5091                 }
5092                 mddev->resync_max = max;
5093         }
5094         wake_up(&mddev->recovery_wait);
5095         err = 0;
5096 out_unlock:
5097         spin_unlock(&mddev->lock);
5098         return err ?: len;
5099 }
5100
5101 static struct md_sysfs_entry md_max_sync =
5102 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5103
5104 static ssize_t
5105 suspend_lo_show(struct mddev *mddev, char *page)
5106 {
5107         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5108 }
5109
5110 static ssize_t
5111 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5112 {
5113         unsigned long long new;
5114         int err;
5115
5116         err = kstrtoull(buf, 10, &new);
5117         if (err < 0)
5118                 return err;
5119         if (new != (sector_t)new)
5120                 return -EINVAL;
5121
5122         err = mddev_lock(mddev);
5123         if (err)
5124                 return err;
5125         err = -EINVAL;
5126         if (mddev->pers == NULL ||
5127             mddev->pers->quiesce == NULL)
5128                 goto unlock;
5129         mddev_suspend(mddev);
5130         mddev->suspend_lo = new;
5131         mddev_resume(mddev);
5132
5133         err = 0;
5134 unlock:
5135         mddev_unlock(mddev);
5136         return err ?: len;
5137 }
5138 static struct md_sysfs_entry md_suspend_lo =
5139 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5140
5141 static ssize_t
5142 suspend_hi_show(struct mddev *mddev, char *page)
5143 {
5144         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5145 }
5146
5147 static ssize_t
5148 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5149 {
5150         unsigned long long new;
5151         int err;
5152
5153         err = kstrtoull(buf, 10, &new);
5154         if (err < 0)
5155                 return err;
5156         if (new != (sector_t)new)
5157                 return -EINVAL;
5158
5159         err = mddev_lock(mddev);
5160         if (err)
5161                 return err;
5162         err = -EINVAL;
5163         if (mddev->pers == NULL)
5164                 goto unlock;
5165
5166         mddev_suspend(mddev);
5167         mddev->suspend_hi = new;
5168         mddev_resume(mddev);
5169
5170         err = 0;
5171 unlock:
5172         mddev_unlock(mddev);
5173         return err ?: len;
5174 }
5175 static struct md_sysfs_entry md_suspend_hi =
5176 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5177
5178 static ssize_t
5179 reshape_position_show(struct mddev *mddev, char *page)
5180 {
5181         if (mddev->reshape_position != MaxSector)
5182                 return sprintf(page, "%llu\n",
5183                                (unsigned long long)mddev->reshape_position);
5184         strcpy(page, "none\n");
5185         return 5;
5186 }
5187
5188 static ssize_t
5189 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5190 {
5191         struct md_rdev *rdev;
5192         unsigned long long new;
5193         int err;
5194
5195         err = kstrtoull(buf, 10, &new);
5196         if (err < 0)
5197                 return err;
5198         if (new != (sector_t)new)
5199                 return -EINVAL;
5200         err = mddev_lock(mddev);
5201         if (err)
5202                 return err;
5203         err = -EBUSY;
5204         if (mddev->pers)
5205                 goto unlock;
5206         mddev->reshape_position = new;
5207         mddev->delta_disks = 0;
5208         mddev->reshape_backwards = 0;
5209         mddev->new_level = mddev->level;
5210         mddev->new_layout = mddev->layout;
5211         mddev->new_chunk_sectors = mddev->chunk_sectors;
5212         rdev_for_each(rdev, mddev)
5213                 rdev->new_data_offset = rdev->data_offset;
5214         err = 0;
5215 unlock:
5216         mddev_unlock(mddev);
5217         return err ?: len;
5218 }
5219
5220 static struct md_sysfs_entry md_reshape_position =
5221 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5222        reshape_position_store);
5223
5224 static ssize_t
5225 reshape_direction_show(struct mddev *mddev, char *page)
5226 {
5227         return sprintf(page, "%s\n",
5228                        mddev->reshape_backwards ? "backwards" : "forwards");
5229 }
5230
5231 static ssize_t
5232 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5233 {
5234         int backwards = 0;
5235         int err;
5236
5237         if (cmd_match(buf, "forwards"))
5238                 backwards = 0;
5239         else if (cmd_match(buf, "backwards"))
5240                 backwards = 1;
5241         else
5242                 return -EINVAL;
5243         if (mddev->reshape_backwards == backwards)
5244                 return len;
5245
5246         err = mddev_lock(mddev);
5247         if (err)
5248                 return err;
5249         /* check if we are allowed to change */
5250         if (mddev->delta_disks)
5251                 err = -EBUSY;
5252         else if (mddev->persistent &&
5253             mddev->major_version == 0)
5254                 err =  -EINVAL;
5255         else
5256                 mddev->reshape_backwards = backwards;
5257         mddev_unlock(mddev);
5258         return err ?: len;
5259 }
5260
5261 static struct md_sysfs_entry md_reshape_direction =
5262 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5263        reshape_direction_store);
5264
5265 static ssize_t
5266 array_size_show(struct mddev *mddev, char *page)
5267 {
5268         if (mddev->external_size)
5269                 return sprintf(page, "%llu\n",
5270                                (unsigned long long)mddev->array_sectors/2);
5271         else
5272                 return sprintf(page, "default\n");
5273 }
5274
5275 static ssize_t
5276 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5277 {
5278         sector_t sectors;
5279         int err;
5280
5281         err = mddev_lock(mddev);
5282         if (err)
5283                 return err;
5284
5285         /* cluster raid doesn't support change array_sectors */
5286         if (mddev_is_clustered(mddev)) {
5287                 mddev_unlock(mddev);
5288                 return -EINVAL;
5289         }
5290
5291         if (strncmp(buf, "default", 7) == 0) {
5292                 if (mddev->pers)
5293                         sectors = mddev->pers->size(mddev, 0, 0);
5294                 else
5295                         sectors = mddev->array_sectors;
5296
5297                 mddev->external_size = 0;
5298         } else {
5299                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
5300                         err = -EINVAL;
5301                 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5302                         err = -E2BIG;
5303                 else
5304                         mddev->external_size = 1;
5305         }
5306
5307         if (!err) {
5308                 mddev->array_sectors = sectors;
5309                 if (mddev->pers)
5310                         set_capacity_and_notify(mddev->gendisk,
5311                                                 mddev->array_sectors);
5312         }
5313         mddev_unlock(mddev);
5314         return err ?: len;
5315 }
5316
5317 static struct md_sysfs_entry md_array_size =
5318 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5319        array_size_store);
5320
5321 static ssize_t
5322 consistency_policy_show(struct mddev *mddev, char *page)
5323 {
5324         int ret;
5325
5326         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5327                 ret = sprintf(page, "journal\n");
5328         } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5329                 ret = sprintf(page, "ppl\n");
5330         } else if (mddev->bitmap) {
5331                 ret = sprintf(page, "bitmap\n");
5332         } else if (mddev->pers) {
5333                 if (mddev->pers->sync_request)
5334                         ret = sprintf(page, "resync\n");
5335                 else
5336                         ret = sprintf(page, "none\n");
5337         } else {
5338                 ret = sprintf(page, "unknown\n");
5339         }
5340
5341         return ret;
5342 }
5343
5344 static ssize_t
5345 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5346 {
5347         int err = 0;
5348
5349         if (mddev->pers) {
5350                 if (mddev->pers->change_consistency_policy)
5351                         err = mddev->pers->change_consistency_policy(mddev, buf);
5352                 else
5353                         err = -EBUSY;
5354         } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5355                 set_bit(MD_HAS_PPL, &mddev->flags);
5356         } else {
5357                 err = -EINVAL;
5358         }
5359
5360         return err ? err : len;
5361 }
5362
5363 static struct md_sysfs_entry md_consistency_policy =
5364 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5365        consistency_policy_store);
5366
5367 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5368 {
5369         return sprintf(page, "%d\n", mddev->fail_last_dev);
5370 }
5371
5372 /*
5373  * Setting fail_last_dev to true to allow last device to be forcibly removed
5374  * from RAID1/RAID10.
5375  */
5376 static ssize_t
5377 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5378 {
5379         int ret;
5380         bool value;
5381
5382         ret = kstrtobool(buf, &value);
5383         if (ret)
5384                 return ret;
5385
5386         if (value != mddev->fail_last_dev)
5387                 mddev->fail_last_dev = value;
5388
5389         return len;
5390 }
5391 static struct md_sysfs_entry md_fail_last_dev =
5392 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5393        fail_last_dev_store);
5394
5395 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5396 {
5397         if (mddev->pers == NULL || (mddev->pers->level != 1))
5398                 return sprintf(page, "n/a\n");
5399         else
5400                 return sprintf(page, "%d\n", mddev->serialize_policy);
5401 }
5402
5403 /*
5404  * Setting serialize_policy to true to enforce write IO is not reordered
5405  * for raid1.
5406  */
5407 static ssize_t
5408 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5409 {
5410         int err;
5411         bool value;
5412
5413         err = kstrtobool(buf, &value);
5414         if (err)
5415                 return err;
5416
5417         if (value == mddev->serialize_policy)
5418                 return len;
5419
5420         err = mddev_lock(mddev);
5421         if (err)
5422                 return err;
5423         if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5424                 pr_err("md: serialize_policy is only effective for raid1\n");
5425                 err = -EINVAL;
5426                 goto unlock;
5427         }
5428
5429         mddev_suspend(mddev);
5430         if (value)
5431                 mddev_create_serial_pool(mddev, NULL, true);
5432         else
5433                 mddev_destroy_serial_pool(mddev, NULL, true);
5434         mddev->serialize_policy = value;
5435         mddev_resume(mddev);
5436 unlock:
5437         mddev_unlock(mddev);
5438         return err ?: len;
5439 }
5440
5441 static struct md_sysfs_entry md_serialize_policy =
5442 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5443        serialize_policy_store);
5444
5445
5446 static struct attribute *md_default_attrs[] = {
5447         &md_level.attr,
5448         &md_layout.attr,
5449         &md_raid_disks.attr,
5450         &md_uuid.attr,
5451         &md_chunk_size.attr,
5452         &md_size.attr,
5453         &md_resync_start.attr,
5454         &md_metadata.attr,
5455         &md_new_device.attr,
5456         &md_safe_delay.attr,
5457         &md_array_state.attr,
5458         &md_reshape_position.attr,
5459         &md_reshape_direction.attr,
5460         &md_array_size.attr,
5461         &max_corr_read_errors.attr,
5462         &md_consistency_policy.attr,
5463         &md_fail_last_dev.attr,
5464         &md_serialize_policy.attr,
5465         NULL,
5466 };
5467
5468 static const struct attribute_group md_default_group = {
5469         .attrs = md_default_attrs,
5470 };
5471
5472 static struct attribute *md_redundancy_attrs[] = {
5473         &md_scan_mode.attr,
5474         &md_last_scan_mode.attr,
5475         &md_mismatches.attr,
5476         &md_sync_min.attr,
5477         &md_sync_max.attr,
5478         &md_sync_speed.attr,
5479         &md_sync_force_parallel.attr,
5480         &md_sync_completed.attr,
5481         &md_min_sync.attr,
5482         &md_max_sync.attr,
5483         &md_suspend_lo.attr,
5484         &md_suspend_hi.attr,
5485         &md_bitmap.attr,
5486         &md_degraded.attr,
5487         NULL,
5488 };
5489 static const struct attribute_group md_redundancy_group = {
5490         .name = NULL,
5491         .attrs = md_redundancy_attrs,
5492 };
5493
5494 static const struct attribute_group *md_attr_groups[] = {
5495         &md_default_group,
5496         &md_bitmap_group,
5497         NULL,
5498 };
5499
5500 static ssize_t
5501 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5502 {
5503         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5504         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5505         ssize_t rv;
5506
5507         if (!entry->show)
5508                 return -EIO;
5509         spin_lock(&all_mddevs_lock);
5510         if (!mddev_get(mddev)) {
5511                 spin_unlock(&all_mddevs_lock);
5512                 return -EBUSY;
5513         }
5514         spin_unlock(&all_mddevs_lock);
5515
5516         rv = entry->show(mddev, page);
5517         mddev_put(mddev);
5518         return rv;
5519 }
5520
5521 static ssize_t
5522 md_attr_store(struct kobject *kobj, struct attribute *attr,
5523               const char *page, size_t length)
5524 {
5525         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5526         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5527         ssize_t rv;
5528
5529         if (!entry->store)
5530                 return -EIO;
5531         if (!capable(CAP_SYS_ADMIN))
5532                 return -EACCES;
5533         spin_lock(&all_mddevs_lock);
5534         if (!mddev_get(mddev)) {
5535                 spin_unlock(&all_mddevs_lock);
5536                 return -EBUSY;
5537         }
5538         spin_unlock(&all_mddevs_lock);
5539         rv = entry->store(mddev, page, length);
5540         mddev_put(mddev);
5541         return rv;
5542 }
5543
5544 static void md_kobj_release(struct kobject *ko)
5545 {
5546         struct mddev *mddev = container_of(ko, struct mddev, kobj);
5547
5548         if (mddev->sysfs_state)
5549                 sysfs_put(mddev->sysfs_state);
5550         if (mddev->sysfs_level)
5551                 sysfs_put(mddev->sysfs_level);
5552
5553         del_gendisk(mddev->gendisk);
5554         put_disk(mddev->gendisk);
5555 }
5556
5557 static const struct sysfs_ops md_sysfs_ops = {
5558         .show   = md_attr_show,
5559         .store  = md_attr_store,
5560 };
5561 static struct kobj_type md_ktype = {
5562         .release        = md_kobj_release,
5563         .sysfs_ops      = &md_sysfs_ops,
5564         .default_groups = md_attr_groups,
5565 };
5566
5567 int mdp_major = 0;
5568
5569 static void mddev_delayed_delete(struct work_struct *ws)
5570 {
5571         struct mddev *mddev = container_of(ws, struct mddev, del_work);
5572
5573         kobject_put(&mddev->kobj);
5574 }
5575
5576 static void no_op(struct percpu_ref *r) {}
5577
5578 int mddev_init_writes_pending(struct mddev *mddev)
5579 {
5580         if (mddev->writes_pending.percpu_count_ptr)
5581                 return 0;
5582         if (percpu_ref_init(&mddev->writes_pending, no_op,
5583                             PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5584                 return -ENOMEM;
5585         /* We want to start with the refcount at zero */
5586         percpu_ref_put(&mddev->writes_pending);
5587         return 0;
5588 }
5589 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5590
5591 struct mddev *md_alloc(dev_t dev, char *name)
5592 {
5593         /*
5594          * If dev is zero, name is the name of a device to allocate with
5595          * an arbitrary minor number.  It will be "md_???"
5596          * If dev is non-zero it must be a device number with a MAJOR of
5597          * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5598          * the device is being created by opening a node in /dev.
5599          * If "name" is not NULL, the device is being created by
5600          * writing to /sys/module/md_mod/parameters/new_array.
5601          */
5602         static DEFINE_MUTEX(disks_mutex);
5603         struct mddev *mddev;
5604         struct gendisk *disk;
5605         int partitioned;
5606         int shift;
5607         int unit;
5608         int error ;
5609
5610         /*
5611          * Wait for any previous instance of this device to be completely
5612          * removed (mddev_delayed_delete).
5613          */
5614         flush_workqueue(md_misc_wq);
5615         flush_workqueue(md_rdev_misc_wq);
5616
5617         mutex_lock(&disks_mutex);
5618         mddev = mddev_alloc(dev);
5619         if (IS_ERR(mddev)) {
5620                 error = PTR_ERR(mddev);
5621                 goto out_unlock;
5622         }
5623
5624         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5625         shift = partitioned ? MdpMinorShift : 0;
5626         unit = MINOR(mddev->unit) >> shift;
5627
5628         if (name && !dev) {
5629                 /* Need to ensure that 'name' is not a duplicate.
5630                  */
5631                 struct mddev *mddev2;
5632                 spin_lock(&all_mddevs_lock);
5633
5634                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5635                         if (mddev2->gendisk &&
5636                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
5637                                 spin_unlock(&all_mddevs_lock);
5638                                 error = -EEXIST;
5639                                 goto out_free_mddev;
5640                         }
5641                 spin_unlock(&all_mddevs_lock);
5642         }
5643         if (name && dev)
5644                 /*
5645                  * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5646                  */
5647                 mddev->hold_active = UNTIL_STOP;
5648
5649         error = -ENOMEM;
5650         disk = blk_alloc_disk(NUMA_NO_NODE);
5651         if (!disk)
5652                 goto out_free_mddev;
5653
5654         disk->major = MAJOR(mddev->unit);
5655         disk->first_minor = unit << shift;
5656         disk->minors = 1 << shift;
5657         if (name)
5658                 strcpy(disk->disk_name, name);
5659         else if (partitioned)
5660                 sprintf(disk->disk_name, "md_d%d", unit);
5661         else
5662                 sprintf(disk->disk_name, "md%d", unit);
5663         disk->fops = &md_fops;
5664         disk->private_data = mddev;
5665
5666         mddev->queue = disk->queue;
5667         blk_set_stacking_limits(&mddev->queue->limits);
5668         blk_queue_write_cache(mddev->queue, true, true);
5669         disk->events |= DISK_EVENT_MEDIA_CHANGE;
5670         mddev->gendisk = disk;
5671         error = add_disk(disk);
5672         if (error)
5673                 goto out_put_disk;
5674
5675         kobject_init(&mddev->kobj, &md_ktype);
5676         error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5677         if (error) {
5678                 /*
5679                  * The disk is already live at this point.  Clear the hold flag
5680                  * and let mddev_put take care of the deletion, as it isn't any
5681                  * different from a normal close on last release now.
5682                  */
5683                 mddev->hold_active = 0;
5684                 mutex_unlock(&disks_mutex);
5685                 mddev_put(mddev);
5686                 return ERR_PTR(error);
5687         }
5688
5689         kobject_uevent(&mddev->kobj, KOBJ_ADD);
5690         mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5691         mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5692         mutex_unlock(&disks_mutex);
5693         return mddev;
5694
5695 out_put_disk:
5696         put_disk(disk);
5697 out_free_mddev:
5698         mddev_free(mddev);
5699 out_unlock:
5700         mutex_unlock(&disks_mutex);
5701         return ERR_PTR(error);
5702 }
5703
5704 static int md_alloc_and_put(dev_t dev, char *name)
5705 {
5706         struct mddev *mddev = md_alloc(dev, name);
5707
5708         if (IS_ERR(mddev))
5709                 return PTR_ERR(mddev);
5710         mddev_put(mddev);
5711         return 0;
5712 }
5713
5714 static void md_probe(dev_t dev)
5715 {
5716         if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5717                 return;
5718         if (create_on_open)
5719                 md_alloc_and_put(dev, NULL);
5720 }
5721
5722 static int add_named_array(const char *val, const struct kernel_param *kp)
5723 {
5724         /*
5725          * val must be "md_*" or "mdNNN".
5726          * For "md_*" we allocate an array with a large free minor number, and
5727          * set the name to val.  val must not already be an active name.
5728          * For "mdNNN" we allocate an array with the minor number NNN
5729          * which must not already be in use.
5730          */
5731         int len = strlen(val);
5732         char buf[DISK_NAME_LEN];
5733         unsigned long devnum;
5734
5735         while (len && val[len-1] == '\n')
5736                 len--;
5737         if (len >= DISK_NAME_LEN)
5738                 return -E2BIG;
5739         strscpy(buf, val, len+1);
5740         if (strncmp(buf, "md_", 3) == 0)
5741                 return md_alloc_and_put(0, buf);
5742         if (strncmp(buf, "md", 2) == 0 &&
5743             isdigit(buf[2]) &&
5744             kstrtoul(buf+2, 10, &devnum) == 0 &&
5745             devnum <= MINORMASK)
5746                 return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5747
5748         return -EINVAL;
5749 }
5750
5751 static void md_safemode_timeout(struct timer_list *t)
5752 {
5753         struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5754
5755         mddev->safemode = 1;
5756         if (mddev->external)
5757                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5758
5759         md_wakeup_thread(mddev->thread);
5760 }
5761
5762 static int start_dirty_degraded;
5763 static void active_io_release(struct percpu_ref *ref)
5764 {
5765         struct mddev *mddev = container_of(ref, struct mddev, active_io);
5766
5767         wake_up(&mddev->sb_wait);
5768 }
5769
5770 int md_run(struct mddev *mddev)
5771 {
5772         int err;
5773         struct md_rdev *rdev;
5774         struct md_personality *pers;
5775         bool nowait = true;
5776
5777         if (list_empty(&mddev->disks))
5778                 /* cannot run an array with no devices.. */
5779                 return -EINVAL;
5780
5781         if (mddev->pers)
5782                 return -EBUSY;
5783         /* Cannot run until previous stop completes properly */
5784         if (mddev->sysfs_active)
5785                 return -EBUSY;
5786
5787         /*
5788          * Analyze all RAID superblock(s)
5789          */
5790         if (!mddev->raid_disks) {
5791                 if (!mddev->persistent)
5792                         return -EINVAL;
5793                 err = analyze_sbs(mddev);
5794                 if (err)
5795                         return -EINVAL;
5796         }
5797
5798         if (mddev->level != LEVEL_NONE)
5799                 request_module("md-level-%d", mddev->level);
5800         else if (mddev->clevel[0])
5801                 request_module("md-%s", mddev->clevel);
5802
5803         /*
5804          * Drop all container device buffers, from now on
5805          * the only valid external interface is through the md
5806          * device.
5807          */
5808         mddev->has_superblocks = false;
5809         rdev_for_each(rdev, mddev) {
5810                 if (test_bit(Faulty, &rdev->flags))
5811                         continue;
5812                 sync_blockdev(rdev->bdev);
5813                 invalidate_bdev(rdev->bdev);
5814                 if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5815                         mddev->ro = MD_RDONLY;
5816                         if (mddev->gendisk)
5817                                 set_disk_ro(mddev->gendisk, 1);
5818                 }
5819
5820                 if (rdev->sb_page)
5821                         mddev->has_superblocks = true;
5822
5823                 /* perform some consistency tests on the device.
5824                  * We don't want the data to overlap the metadata,
5825                  * Internal Bitmap issues have been handled elsewhere.
5826                  */
5827                 if (rdev->meta_bdev) {
5828                         /* Nothing to check */;
5829                 } else if (rdev->data_offset < rdev->sb_start) {
5830                         if (mddev->dev_sectors &&
5831                             rdev->data_offset + mddev->dev_sectors
5832                             > rdev->sb_start) {
5833                                 pr_warn("md: %s: data overlaps metadata\n",
5834                                         mdname(mddev));
5835                                 return -EINVAL;
5836                         }
5837                 } else {
5838                         if (rdev->sb_start + rdev->sb_size/512
5839                             > rdev->data_offset) {
5840                                 pr_warn("md: %s: metadata overlaps data\n",
5841                                         mdname(mddev));
5842                                 return -EINVAL;
5843                         }
5844                 }
5845                 sysfs_notify_dirent_safe(rdev->sysfs_state);
5846                 nowait = nowait && bdev_nowait(rdev->bdev);
5847         }
5848
5849         err = percpu_ref_init(&mddev->active_io, active_io_release,
5850                                 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5851         if (err)
5852                 return err;
5853
5854         if (!bioset_initialized(&mddev->bio_set)) {
5855                 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5856                 if (err)
5857                         goto exit_active_io;
5858         }
5859         if (!bioset_initialized(&mddev->sync_set)) {
5860                 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5861                 if (err)
5862                         goto exit_bio_set;
5863         }
5864
5865         spin_lock(&pers_lock);
5866         pers = find_pers(mddev->level, mddev->clevel);
5867         if (!pers || !try_module_get(pers->owner)) {
5868                 spin_unlock(&pers_lock);
5869                 if (mddev->level != LEVEL_NONE)
5870                         pr_warn("md: personality for level %d is not loaded!\n",
5871                                 mddev->level);
5872                 else
5873                         pr_warn("md: personality for level %s is not loaded!\n",
5874                                 mddev->clevel);
5875                 err = -EINVAL;
5876                 goto abort;
5877         }
5878         spin_unlock(&pers_lock);
5879         if (mddev->level != pers->level) {
5880                 mddev->level = pers->level;
5881                 mddev->new_level = pers->level;
5882         }
5883         strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5884
5885         if (mddev->reshape_position != MaxSector &&
5886             pers->start_reshape == NULL) {
5887                 /* This personality cannot handle reshaping... */
5888                 module_put(pers->owner);
5889                 err = -EINVAL;
5890                 goto abort;
5891         }
5892
5893         if (pers->sync_request) {
5894                 /* Warn if this is a potentially silly
5895                  * configuration.
5896                  */
5897                 struct md_rdev *rdev2;
5898                 int warned = 0;
5899
5900                 rdev_for_each(rdev, mddev)
5901                         rdev_for_each(rdev2, mddev) {
5902                                 if (rdev < rdev2 &&
5903                                     rdev->bdev->bd_disk ==
5904                                     rdev2->bdev->bd_disk) {
5905                                         pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5906                                                 mdname(mddev),
5907                                                 rdev->bdev,
5908                                                 rdev2->bdev);
5909                                         warned = 1;
5910                                 }
5911                         }
5912
5913                 if (warned)
5914                         pr_warn("True protection against single-disk failure might be compromised.\n");
5915         }
5916
5917         mddev->recovery = 0;
5918         /* may be over-ridden by personality */
5919         mddev->resync_max_sectors = mddev->dev_sectors;
5920
5921         mddev->ok_start_degraded = start_dirty_degraded;
5922
5923         if (start_readonly && md_is_rdwr(mddev))
5924                 mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
5925
5926         err = pers->run(mddev);
5927         if (err)
5928                 pr_warn("md: pers->run() failed ...\n");
5929         else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5930                 WARN_ONCE(!mddev->external_size,
5931                           "%s: default size too small, but 'external_size' not in effect?\n",
5932                           __func__);
5933                 pr_warn("md: invalid array_size %llu > default size %llu\n",
5934                         (unsigned long long)mddev->array_sectors / 2,
5935                         (unsigned long long)pers->size(mddev, 0, 0) / 2);
5936                 err = -EINVAL;
5937         }
5938         if (err == 0 && pers->sync_request &&
5939             (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5940                 struct bitmap *bitmap;
5941
5942                 bitmap = md_bitmap_create(mddev, -1);
5943                 if (IS_ERR(bitmap)) {
5944                         err = PTR_ERR(bitmap);
5945                         pr_warn("%s: failed to create bitmap (%d)\n",
5946                                 mdname(mddev), err);
5947                 } else
5948                         mddev->bitmap = bitmap;
5949
5950         }
5951         if (err)
5952                 goto bitmap_abort;
5953
5954         if (mddev->bitmap_info.max_write_behind > 0) {
5955                 bool create_pool = false;
5956
5957                 rdev_for_each(rdev, mddev) {
5958                         if (test_bit(WriteMostly, &rdev->flags) &&
5959                             rdev_init_serial(rdev))
5960                                 create_pool = true;
5961                 }
5962                 if (create_pool && mddev->serial_info_pool == NULL) {
5963                         mddev->serial_info_pool =
5964                                 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5965                                                     sizeof(struct serial_info));
5966                         if (!mddev->serial_info_pool) {
5967                                 err = -ENOMEM;
5968                                 goto bitmap_abort;
5969                         }
5970                 }
5971         }
5972
5973         if (mddev->queue) {
5974                 bool nonrot = true;
5975
5976                 rdev_for_each(rdev, mddev) {
5977                         if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
5978                                 nonrot = false;
5979                                 break;
5980                         }
5981                 }
5982                 if (mddev->degraded)
5983                         nonrot = false;
5984                 if (nonrot)
5985                         blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5986                 else
5987                         blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5988                 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
5989
5990                 /* Set the NOWAIT flags if all underlying devices support it */
5991                 if (nowait)
5992                         blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
5993         }
5994         if (pers->sync_request) {
5995                 if (mddev->kobj.sd &&
5996                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5997                         pr_warn("md: cannot register extra attributes for %s\n",
5998                                 mdname(mddev));
5999                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6000                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6001                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6002         } else if (mddev->ro == MD_AUTO_READ)
6003                 mddev->ro = MD_RDWR;
6004
6005         atomic_set(&mddev->max_corr_read_errors,
6006                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6007         mddev->safemode = 0;
6008         if (mddev_is_clustered(mddev))
6009                 mddev->safemode_delay = 0;
6010         else
6011                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6012         mddev->in_sync = 1;
6013         smp_wmb();
6014         spin_lock(&mddev->lock);
6015         mddev->pers = pers;
6016         spin_unlock(&mddev->lock);
6017         rdev_for_each(rdev, mddev)
6018                 if (rdev->raid_disk >= 0)
6019                         sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6020
6021         if (mddev->degraded && md_is_rdwr(mddev))
6022                 /* This ensures that recovering status is reported immediately
6023                  * via sysfs - until a lack of spares is confirmed.
6024                  */
6025                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6026         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6027
6028         if (mddev->sb_flags)
6029                 md_update_sb(mddev, 0);
6030
6031         md_new_event();
6032         return 0;
6033
6034 bitmap_abort:
6035         mddev_detach(mddev);
6036         if (mddev->private)
6037                 pers->free(mddev, mddev->private);
6038         mddev->private = NULL;
6039         module_put(pers->owner);
6040         md_bitmap_destroy(mddev);
6041 abort:
6042         bioset_exit(&mddev->sync_set);
6043 exit_bio_set:
6044         bioset_exit(&mddev->bio_set);
6045 exit_active_io:
6046         percpu_ref_exit(&mddev->active_io);
6047         return err;
6048 }
6049 EXPORT_SYMBOL_GPL(md_run);
6050
6051 int do_md_run(struct mddev *mddev)
6052 {
6053         int err;
6054
6055         set_bit(MD_NOT_READY, &mddev->flags);
6056         err = md_run(mddev);
6057         if (err)
6058                 goto out;
6059         err = md_bitmap_load(mddev);
6060         if (err) {
6061                 md_bitmap_destroy(mddev);
6062                 goto out;
6063         }
6064
6065         if (mddev_is_clustered(mddev))
6066                 md_allow_write(mddev);
6067
6068         /* run start up tasks that require md_thread */
6069         md_start(mddev);
6070
6071         md_wakeup_thread(mddev->thread);
6072         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6073
6074         set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6075         clear_bit(MD_NOT_READY, &mddev->flags);
6076         mddev->changed = 1;
6077         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6078         sysfs_notify_dirent_safe(mddev->sysfs_state);
6079         sysfs_notify_dirent_safe(mddev->sysfs_action);
6080         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6081 out:
6082         clear_bit(MD_NOT_READY, &mddev->flags);
6083         return err;
6084 }
6085
6086 int md_start(struct mddev *mddev)
6087 {
6088         int ret = 0;
6089
6090         if (mddev->pers->start) {
6091                 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6092                 md_wakeup_thread(mddev->thread);
6093                 ret = mddev->pers->start(mddev);
6094                 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6095                 md_wakeup_thread(mddev->sync_thread);
6096         }
6097         return ret;
6098 }
6099 EXPORT_SYMBOL_GPL(md_start);
6100
6101 static int restart_array(struct mddev *mddev)
6102 {
6103         struct gendisk *disk = mddev->gendisk;
6104         struct md_rdev *rdev;
6105         bool has_journal = false;
6106         bool has_readonly = false;
6107
6108         /* Complain if it has no devices */
6109         if (list_empty(&mddev->disks))
6110                 return -ENXIO;
6111         if (!mddev->pers)
6112                 return -EINVAL;
6113         if (md_is_rdwr(mddev))
6114                 return -EBUSY;
6115
6116         rcu_read_lock();
6117         rdev_for_each_rcu(rdev, mddev) {
6118                 if (test_bit(Journal, &rdev->flags) &&
6119                     !test_bit(Faulty, &rdev->flags))
6120                         has_journal = true;
6121                 if (rdev_read_only(rdev))
6122                         has_readonly = true;
6123         }
6124         rcu_read_unlock();
6125         if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6126                 /* Don't restart rw with journal missing/faulty */
6127                         return -EINVAL;
6128         if (has_readonly)
6129                 return -EROFS;
6130
6131         mddev->safemode = 0;
6132         mddev->ro = MD_RDWR;
6133         set_disk_ro(disk, 0);
6134         pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6135         /* Kick recovery or resync if necessary */
6136         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6137         md_wakeup_thread(mddev->thread);
6138         md_wakeup_thread(mddev->sync_thread);
6139         sysfs_notify_dirent_safe(mddev->sysfs_state);
6140         return 0;
6141 }
6142
6143 static void md_clean(struct mddev *mddev)
6144 {
6145         mddev->array_sectors = 0;
6146         mddev->external_size = 0;
6147         mddev->dev_sectors = 0;
6148         mddev->raid_disks = 0;
6149         mddev->recovery_cp = 0;
6150         mddev->resync_min = 0;
6151         mddev->resync_max = MaxSector;
6152         mddev->reshape_position = MaxSector;
6153         mddev->external = 0;
6154         mddev->persistent = 0;
6155         mddev->level = LEVEL_NONE;
6156         mddev->clevel[0] = 0;
6157         mddev->flags = 0;
6158         mddev->sb_flags = 0;
6159         mddev->ro = MD_RDWR;
6160         mddev->metadata_type[0] = 0;
6161         mddev->chunk_sectors = 0;
6162         mddev->ctime = mddev->utime = 0;
6163         mddev->layout = 0;
6164         mddev->max_disks = 0;
6165         mddev->events = 0;
6166         mddev->can_decrease_events = 0;
6167         mddev->delta_disks = 0;
6168         mddev->reshape_backwards = 0;
6169         mddev->new_level = LEVEL_NONE;
6170         mddev->new_layout = 0;
6171         mddev->new_chunk_sectors = 0;
6172         mddev->curr_resync = MD_RESYNC_NONE;
6173         atomic64_set(&mddev->resync_mismatches, 0);
6174         mddev->suspend_lo = mddev->suspend_hi = 0;
6175         mddev->sync_speed_min = mddev->sync_speed_max = 0;
6176         mddev->recovery = 0;
6177         mddev->in_sync = 0;
6178         mddev->changed = 0;
6179         mddev->degraded = 0;
6180         mddev->safemode = 0;
6181         mddev->private = NULL;
6182         mddev->cluster_info = NULL;
6183         mddev->bitmap_info.offset = 0;
6184         mddev->bitmap_info.default_offset = 0;
6185         mddev->bitmap_info.default_space = 0;
6186         mddev->bitmap_info.chunksize = 0;
6187         mddev->bitmap_info.daemon_sleep = 0;
6188         mddev->bitmap_info.max_write_behind = 0;
6189         mddev->bitmap_info.nodes = 0;
6190 }
6191
6192 static void __md_stop_writes(struct mddev *mddev)
6193 {
6194         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6195         if (work_pending(&mddev->del_work))
6196                 flush_workqueue(md_misc_wq);
6197         if (mddev->sync_thread) {
6198                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6199                 md_unregister_thread(&mddev->sync_thread);
6200                 md_reap_sync_thread(mddev);
6201         }
6202
6203         del_timer_sync(&mddev->safemode_timer);
6204
6205         if (mddev->pers && mddev->pers->quiesce) {
6206                 mddev->pers->quiesce(mddev, 1);
6207                 mddev->pers->quiesce(mddev, 0);
6208         }
6209         md_bitmap_flush(mddev);
6210
6211         if (md_is_rdwr(mddev) &&
6212             ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6213              mddev->sb_flags)) {
6214                 /* mark array as shutdown cleanly */
6215                 if (!mddev_is_clustered(mddev))
6216                         mddev->in_sync = 1;
6217                 md_update_sb(mddev, 1);
6218         }
6219         /* disable policy to guarantee rdevs free resources for serialization */
6220         mddev->serialize_policy = 0;
6221         mddev_destroy_serial_pool(mddev, NULL, true);
6222 }
6223
6224 void md_stop_writes(struct mddev *mddev)
6225 {
6226         mddev_lock_nointr(mddev);
6227         __md_stop_writes(mddev);
6228         mddev_unlock(mddev);
6229 }
6230 EXPORT_SYMBOL_GPL(md_stop_writes);
6231
6232 static void mddev_detach(struct mddev *mddev)
6233 {
6234         md_bitmap_wait_behind_writes(mddev);
6235         if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6236                 mddev->pers->quiesce(mddev, 1);
6237                 mddev->pers->quiesce(mddev, 0);
6238         }
6239         md_unregister_thread(&mddev->thread);
6240         if (mddev->queue)
6241                 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6242 }
6243
6244 static void __md_stop(struct mddev *mddev)
6245 {
6246         struct md_personality *pers = mddev->pers;
6247         md_bitmap_destroy(mddev);
6248         mddev_detach(mddev);
6249         /* Ensure ->event_work is done */
6250         if (mddev->event_work.func)
6251                 flush_workqueue(md_misc_wq);
6252         spin_lock(&mddev->lock);
6253         mddev->pers = NULL;
6254         spin_unlock(&mddev->lock);
6255         if (mddev->private)
6256                 pers->free(mddev, mddev->private);
6257         mddev->private = NULL;
6258         if (pers->sync_request && mddev->to_remove == NULL)
6259                 mddev->to_remove = &md_redundancy_group;
6260         module_put(pers->owner);
6261         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6262
6263         percpu_ref_exit(&mddev->active_io);
6264         bioset_exit(&mddev->bio_set);
6265         bioset_exit(&mddev->sync_set);
6266 }
6267
6268 void md_stop(struct mddev *mddev)
6269 {
6270         /* stop the array and free an attached data structures.
6271          * This is called from dm-raid
6272          */
6273         __md_stop_writes(mddev);
6274         __md_stop(mddev);
6275         percpu_ref_exit(&mddev->writes_pending);
6276 }
6277
6278 EXPORT_SYMBOL_GPL(md_stop);
6279
6280 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6281 {
6282         int err = 0;
6283         int did_freeze = 0;
6284
6285         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6286                 did_freeze = 1;
6287                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6288                 md_wakeup_thread(mddev->thread);
6289         }
6290         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6291                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6292         if (mddev->sync_thread)
6293                 /* Thread might be blocked waiting for metadata update
6294                  * which will now never happen */
6295                 wake_up_process(mddev->sync_thread->tsk);
6296
6297         if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6298                 return -EBUSY;
6299         mddev_unlock(mddev);
6300         wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6301                                           &mddev->recovery));
6302         wait_event(mddev->sb_wait,
6303                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6304         mddev_lock_nointr(mddev);
6305
6306         mutex_lock(&mddev->open_mutex);
6307         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6308             mddev->sync_thread ||
6309             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6310                 pr_warn("md: %s still in use.\n",mdname(mddev));
6311                 if (did_freeze) {
6312                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6313                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6314                         md_wakeup_thread(mddev->thread);
6315                 }
6316                 err = -EBUSY;
6317                 goto out;
6318         }
6319         if (mddev->pers) {
6320                 __md_stop_writes(mddev);
6321
6322                 err  = -ENXIO;
6323                 if (mddev->ro == MD_RDONLY)
6324                         goto out;
6325                 mddev->ro = MD_RDONLY;
6326                 set_disk_ro(mddev->gendisk, 1);
6327                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6328                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6329                 md_wakeup_thread(mddev->thread);
6330                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6331                 err = 0;
6332         }
6333 out:
6334         mutex_unlock(&mddev->open_mutex);
6335         return err;
6336 }
6337
6338 /* mode:
6339  *   0 - completely stop and dis-assemble array
6340  *   2 - stop but do not disassemble array
6341  */
6342 static int do_md_stop(struct mddev *mddev, int mode,
6343                       struct block_device *bdev)
6344 {
6345         struct gendisk *disk = mddev->gendisk;
6346         struct md_rdev *rdev;
6347         int did_freeze = 0;
6348
6349         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6350                 did_freeze = 1;
6351                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6352                 md_wakeup_thread(mddev->thread);
6353         }
6354         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6355                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6356         if (mddev->sync_thread)
6357                 /* Thread might be blocked waiting for metadata update
6358                  * which will now never happen */
6359                 wake_up_process(mddev->sync_thread->tsk);
6360
6361         mddev_unlock(mddev);
6362         wait_event(resync_wait, (mddev->sync_thread == NULL &&
6363                                  !test_bit(MD_RECOVERY_RUNNING,
6364                                            &mddev->recovery)));
6365         mddev_lock_nointr(mddev);
6366
6367         mutex_lock(&mddev->open_mutex);
6368         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6369             mddev->sysfs_active ||
6370             mddev->sync_thread ||
6371             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6372                 pr_warn("md: %s still in use.\n",mdname(mddev));
6373                 mutex_unlock(&mddev->open_mutex);
6374                 if (did_freeze) {
6375                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6376                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6377                         md_wakeup_thread(mddev->thread);
6378                 }
6379                 return -EBUSY;
6380         }
6381         if (mddev->pers) {
6382                 if (!md_is_rdwr(mddev))
6383                         set_disk_ro(disk, 0);
6384
6385                 __md_stop_writes(mddev);
6386                 __md_stop(mddev);
6387
6388                 /* tell userspace to handle 'inactive' */
6389                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6390
6391                 rdev_for_each(rdev, mddev)
6392                         if (rdev->raid_disk >= 0)
6393                                 sysfs_unlink_rdev(mddev, rdev);
6394
6395                 set_capacity_and_notify(disk, 0);
6396                 mutex_unlock(&mddev->open_mutex);
6397                 mddev->changed = 1;
6398
6399                 if (!md_is_rdwr(mddev))
6400                         mddev->ro = MD_RDWR;
6401         } else
6402                 mutex_unlock(&mddev->open_mutex);
6403         /*
6404          * Free resources if final stop
6405          */
6406         if (mode == 0) {
6407                 pr_info("md: %s stopped.\n", mdname(mddev));
6408
6409                 if (mddev->bitmap_info.file) {
6410                         struct file *f = mddev->bitmap_info.file;
6411                         spin_lock(&mddev->lock);
6412                         mddev->bitmap_info.file = NULL;
6413                         spin_unlock(&mddev->lock);
6414                         fput(f);
6415                 }
6416                 mddev->bitmap_info.offset = 0;
6417
6418                 export_array(mddev);
6419
6420                 md_clean(mddev);
6421                 if (mddev->hold_active == UNTIL_STOP)
6422                         mddev->hold_active = 0;
6423         }
6424         md_new_event();
6425         sysfs_notify_dirent_safe(mddev->sysfs_state);
6426         return 0;
6427 }
6428
6429 #ifndef MODULE
6430 static void autorun_array(struct mddev *mddev)
6431 {
6432         struct md_rdev *rdev;
6433         int err;
6434
6435         if (list_empty(&mddev->disks))
6436                 return;
6437
6438         pr_info("md: running: ");
6439
6440         rdev_for_each(rdev, mddev) {
6441                 pr_cont("<%pg>", rdev->bdev);
6442         }
6443         pr_cont("\n");
6444
6445         err = do_md_run(mddev);
6446         if (err) {
6447                 pr_warn("md: do_md_run() returned %d\n", err);
6448                 do_md_stop(mddev, 0, NULL);
6449         }
6450 }
6451
6452 /*
6453  * lets try to run arrays based on all disks that have arrived
6454  * until now. (those are in pending_raid_disks)
6455  *
6456  * the method: pick the first pending disk, collect all disks with
6457  * the same UUID, remove all from the pending list and put them into
6458  * the 'same_array' list. Then order this list based on superblock
6459  * update time (freshest comes first), kick out 'old' disks and
6460  * compare superblocks. If everything's fine then run it.
6461  *
6462  * If "unit" is allocated, then bump its reference count
6463  */
6464 static void autorun_devices(int part)
6465 {
6466         struct md_rdev *rdev0, *rdev, *tmp;
6467         struct mddev *mddev;
6468
6469         pr_info("md: autorun ...\n");
6470         while (!list_empty(&pending_raid_disks)) {
6471                 int unit;
6472                 dev_t dev;
6473                 LIST_HEAD(candidates);
6474                 rdev0 = list_entry(pending_raid_disks.next,
6475                                          struct md_rdev, same_set);
6476
6477                 pr_debug("md: considering %pg ...\n", rdev0->bdev);
6478                 INIT_LIST_HEAD(&candidates);
6479                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6480                         if (super_90_load(rdev, rdev0, 0) >= 0) {
6481                                 pr_debug("md:  adding %pg ...\n",
6482                                          rdev->bdev);
6483                                 list_move(&rdev->same_set, &candidates);
6484                         }
6485                 /*
6486                  * now we have a set of devices, with all of them having
6487                  * mostly sane superblocks. It's time to allocate the
6488                  * mddev.
6489                  */
6490                 if (part) {
6491                         dev = MKDEV(mdp_major,
6492                                     rdev0->preferred_minor << MdpMinorShift);
6493                         unit = MINOR(dev) >> MdpMinorShift;
6494                 } else {
6495                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6496                         unit = MINOR(dev);
6497                 }
6498                 if (rdev0->preferred_minor != unit) {
6499                         pr_warn("md: unit number in %pg is bad: %d\n",
6500                                 rdev0->bdev, rdev0->preferred_minor);
6501                         break;
6502                 }
6503
6504                 mddev = md_alloc(dev, NULL);
6505                 if (IS_ERR(mddev))
6506                         break;
6507
6508                 if (mddev_lock(mddev))
6509                         pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6510                 else if (mddev->raid_disks || mddev->major_version
6511                          || !list_empty(&mddev->disks)) {
6512                         pr_warn("md: %s already running, cannot run %pg\n",
6513                                 mdname(mddev), rdev0->bdev);
6514                         mddev_unlock(mddev);
6515                 } else {
6516                         pr_debug("md: created %s\n", mdname(mddev));
6517                         mddev->persistent = 1;
6518                         rdev_for_each_list(rdev, tmp, &candidates) {
6519                                 list_del_init(&rdev->same_set);
6520                                 if (bind_rdev_to_array(rdev, mddev))
6521                                         export_rdev(rdev);
6522                         }
6523                         autorun_array(mddev);
6524                         mddev_unlock(mddev);
6525                 }
6526                 /* on success, candidates will be empty, on error
6527                  * it won't...
6528                  */
6529                 rdev_for_each_list(rdev, tmp, &candidates) {
6530                         list_del_init(&rdev->same_set);
6531                         export_rdev(rdev);
6532                 }
6533                 mddev_put(mddev);
6534         }
6535         pr_info("md: ... autorun DONE.\n");
6536 }
6537 #endif /* !MODULE */
6538
6539 static int get_version(void __user *arg)
6540 {
6541         mdu_version_t ver;
6542
6543         ver.major = MD_MAJOR_VERSION;
6544         ver.minor = MD_MINOR_VERSION;
6545         ver.patchlevel = MD_PATCHLEVEL_VERSION;
6546
6547         if (copy_to_user(arg, &ver, sizeof(ver)))
6548                 return -EFAULT;
6549
6550         return 0;
6551 }
6552
6553 static int get_array_info(struct mddev *mddev, void __user *arg)
6554 {
6555         mdu_array_info_t info;
6556         int nr,working,insync,failed,spare;
6557         struct md_rdev *rdev;
6558
6559         nr = working = insync = failed = spare = 0;
6560         rcu_read_lock();
6561         rdev_for_each_rcu(rdev, mddev) {
6562                 nr++;
6563                 if (test_bit(Faulty, &rdev->flags))
6564                         failed++;
6565                 else {
6566                         working++;
6567                         if (test_bit(In_sync, &rdev->flags))
6568                                 insync++;
6569                         else if (test_bit(Journal, &rdev->flags))
6570                                 /* TODO: add journal count to md_u.h */
6571                                 ;
6572                         else
6573                                 spare++;
6574                 }
6575         }
6576         rcu_read_unlock();
6577
6578         info.major_version = mddev->major_version;
6579         info.minor_version = mddev->minor_version;
6580         info.patch_version = MD_PATCHLEVEL_VERSION;
6581         info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6582         info.level         = mddev->level;
6583         info.size          = mddev->dev_sectors / 2;
6584         if (info.size != mddev->dev_sectors / 2) /* overflow */
6585                 info.size = -1;
6586         info.nr_disks      = nr;
6587         info.raid_disks    = mddev->raid_disks;
6588         info.md_minor      = mddev->md_minor;
6589         info.not_persistent= !mddev->persistent;
6590
6591         info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6592         info.state         = 0;
6593         if (mddev->in_sync)
6594                 info.state = (1<<MD_SB_CLEAN);
6595         if (mddev->bitmap && mddev->bitmap_info.offset)
6596                 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6597         if (mddev_is_clustered(mddev))
6598                 info.state |= (1<<MD_SB_CLUSTERED);
6599         info.active_disks  = insync;
6600         info.working_disks = working;
6601         info.failed_disks  = failed;
6602         info.spare_disks   = spare;
6603
6604         info.layout        = mddev->layout;
6605         info.chunk_size    = mddev->chunk_sectors << 9;
6606
6607         if (copy_to_user(arg, &info, sizeof(info)))
6608                 return -EFAULT;
6609
6610         return 0;
6611 }
6612
6613 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6614 {
6615         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6616         char *ptr;
6617         int err;
6618
6619         file = kzalloc(sizeof(*file), GFP_NOIO);
6620         if (!file)
6621                 return -ENOMEM;
6622
6623         err = 0;
6624         spin_lock(&mddev->lock);
6625         /* bitmap enabled */
6626         if (mddev->bitmap_info.file) {
6627                 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6628                                 sizeof(file->pathname));
6629                 if (IS_ERR(ptr))
6630                         err = PTR_ERR(ptr);
6631                 else
6632                         memmove(file->pathname, ptr,
6633                                 sizeof(file->pathname)-(ptr-file->pathname));
6634         }
6635         spin_unlock(&mddev->lock);
6636
6637         if (err == 0 &&
6638             copy_to_user(arg, file, sizeof(*file)))
6639                 err = -EFAULT;
6640
6641         kfree(file);
6642         return err;
6643 }
6644
6645 static int get_disk_info(struct mddev *mddev, void __user * arg)
6646 {
6647         mdu_disk_info_t info;
6648         struct md_rdev *rdev;
6649
6650         if (copy_from_user(&info, arg, sizeof(info)))
6651                 return -EFAULT;
6652
6653         rcu_read_lock();
6654         rdev = md_find_rdev_nr_rcu(mddev, info.number);
6655         if (rdev) {
6656                 info.major = MAJOR(rdev->bdev->bd_dev);
6657                 info.minor = MINOR(rdev->bdev->bd_dev);
6658                 info.raid_disk = rdev->raid_disk;
6659                 info.state = 0;
6660                 if (test_bit(Faulty, &rdev->flags))
6661                         info.state |= (1<<MD_DISK_FAULTY);
6662                 else if (test_bit(In_sync, &rdev->flags)) {
6663                         info.state |= (1<<MD_DISK_ACTIVE);
6664                         info.state |= (1<<MD_DISK_SYNC);
6665                 }
6666                 if (test_bit(Journal, &rdev->flags))
6667                         info.state |= (1<<MD_DISK_JOURNAL);
6668                 if (test_bit(WriteMostly, &rdev->flags))
6669                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
6670                 if (test_bit(FailFast, &rdev->flags))
6671                         info.state |= (1<<MD_DISK_FAILFAST);
6672         } else {
6673                 info.major = info.minor = 0;
6674                 info.raid_disk = -1;
6675                 info.state = (1<<MD_DISK_REMOVED);
6676         }
6677         rcu_read_unlock();
6678
6679         if (copy_to_user(arg, &info, sizeof(info)))
6680                 return -EFAULT;
6681
6682         return 0;
6683 }
6684
6685 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6686 {
6687         struct md_rdev *rdev;
6688         dev_t dev = MKDEV(info->major,info->minor);
6689
6690         if (mddev_is_clustered(mddev) &&
6691                 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6692                 pr_warn("%s: Cannot add to clustered mddev.\n",
6693                         mdname(mddev));
6694                 return -EINVAL;
6695         }
6696
6697         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6698                 return -EOVERFLOW;
6699
6700         if (!mddev->raid_disks) {
6701                 int err;
6702                 /* expecting a device which has a superblock */
6703                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6704                 if (IS_ERR(rdev)) {
6705                         pr_warn("md: md_import_device returned %ld\n",
6706                                 PTR_ERR(rdev));
6707                         return PTR_ERR(rdev);
6708                 }
6709                 if (!list_empty(&mddev->disks)) {
6710                         struct md_rdev *rdev0
6711                                 = list_entry(mddev->disks.next,
6712                                              struct md_rdev, same_set);
6713                         err = super_types[mddev->major_version]
6714                                 .load_super(rdev, rdev0, mddev->minor_version);
6715                         if (err < 0) {
6716                                 pr_warn("md: %pg has different UUID to %pg\n",
6717                                         rdev->bdev,
6718                                         rdev0->bdev);
6719                                 export_rdev(rdev);
6720                                 return -EINVAL;
6721                         }
6722                 }
6723                 err = bind_rdev_to_array(rdev, mddev);
6724                 if (err)
6725                         export_rdev(rdev);
6726                 return err;
6727         }
6728
6729         /*
6730          * md_add_new_disk can be used once the array is assembled
6731          * to add "hot spares".  They must already have a superblock
6732          * written
6733          */
6734         if (mddev->pers) {
6735                 int err;
6736                 if (!mddev->pers->hot_add_disk) {
6737                         pr_warn("%s: personality does not support diskops!\n",
6738                                 mdname(mddev));
6739                         return -EINVAL;
6740                 }
6741                 if (mddev->persistent)
6742                         rdev = md_import_device(dev, mddev->major_version,
6743                                                 mddev->minor_version);
6744                 else
6745                         rdev = md_import_device(dev, -1, -1);
6746                 if (IS_ERR(rdev)) {
6747                         pr_warn("md: md_import_device returned %ld\n",
6748                                 PTR_ERR(rdev));
6749                         return PTR_ERR(rdev);
6750                 }
6751                 /* set saved_raid_disk if appropriate */
6752                 if (!mddev->persistent) {
6753                         if (info->state & (1<<MD_DISK_SYNC)  &&
6754                             info->raid_disk < mddev->raid_disks) {
6755                                 rdev->raid_disk = info->raid_disk;
6756                                 set_bit(In_sync, &rdev->flags);
6757                                 clear_bit(Bitmap_sync, &rdev->flags);
6758                         } else
6759                                 rdev->raid_disk = -1;
6760                         rdev->saved_raid_disk = rdev->raid_disk;
6761                 } else
6762                         super_types[mddev->major_version].
6763                                 validate_super(mddev, rdev);
6764                 if ((info->state & (1<<MD_DISK_SYNC)) &&
6765                      rdev->raid_disk != info->raid_disk) {
6766                         /* This was a hot-add request, but events doesn't
6767                          * match, so reject it.
6768                          */
6769                         export_rdev(rdev);
6770                         return -EINVAL;
6771                 }
6772
6773                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6774                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6775                         set_bit(WriteMostly, &rdev->flags);
6776                 else
6777                         clear_bit(WriteMostly, &rdev->flags);
6778                 if (info->state & (1<<MD_DISK_FAILFAST))
6779                         set_bit(FailFast, &rdev->flags);
6780                 else
6781                         clear_bit(FailFast, &rdev->flags);
6782
6783                 if (info->state & (1<<MD_DISK_JOURNAL)) {
6784                         struct md_rdev *rdev2;
6785                         bool has_journal = false;
6786
6787                         /* make sure no existing journal disk */
6788                         rdev_for_each(rdev2, mddev) {
6789                                 if (test_bit(Journal, &rdev2->flags)) {
6790                                         has_journal = true;
6791                                         break;
6792                                 }
6793                         }
6794                         if (has_journal || mddev->bitmap) {
6795                                 export_rdev(rdev);
6796                                 return -EBUSY;
6797                         }
6798                         set_bit(Journal, &rdev->flags);
6799                 }
6800                 /*
6801                  * check whether the device shows up in other nodes
6802                  */
6803                 if (mddev_is_clustered(mddev)) {
6804                         if (info->state & (1 << MD_DISK_CANDIDATE))
6805                                 set_bit(Candidate, &rdev->flags);
6806                         else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6807                                 /* --add initiated by this node */
6808                                 err = md_cluster_ops->add_new_disk(mddev, rdev);
6809                                 if (err) {
6810                                         export_rdev(rdev);
6811                                         return err;
6812                                 }
6813                         }
6814                 }
6815
6816                 rdev->raid_disk = -1;
6817                 err = bind_rdev_to_array(rdev, mddev);
6818
6819                 if (err)
6820                         export_rdev(rdev);
6821
6822                 if (mddev_is_clustered(mddev)) {
6823                         if (info->state & (1 << MD_DISK_CANDIDATE)) {
6824                                 if (!err) {
6825                                         err = md_cluster_ops->new_disk_ack(mddev,
6826                                                 err == 0);
6827                                         if (err)
6828                                                 md_kick_rdev_from_array(rdev);
6829                                 }
6830                         } else {
6831                                 if (err)
6832                                         md_cluster_ops->add_new_disk_cancel(mddev);
6833                                 else
6834                                         err = add_bound_rdev(rdev);
6835                         }
6836
6837                 } else if (!err)
6838                         err = add_bound_rdev(rdev);
6839
6840                 return err;
6841         }
6842
6843         /* otherwise, md_add_new_disk is only allowed
6844          * for major_version==0 superblocks
6845          */
6846         if (mddev->major_version != 0) {
6847                 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6848                 return -EINVAL;
6849         }
6850
6851         if (!(info->state & (1<<MD_DISK_FAULTY))) {
6852                 int err;
6853                 rdev = md_import_device(dev, -1, 0);
6854                 if (IS_ERR(rdev)) {
6855                         pr_warn("md: error, md_import_device() returned %ld\n",
6856                                 PTR_ERR(rdev));
6857                         return PTR_ERR(rdev);
6858                 }
6859                 rdev->desc_nr = info->number;
6860                 if (info->raid_disk < mddev->raid_disks)
6861                         rdev->raid_disk = info->raid_disk;
6862                 else
6863                         rdev->raid_disk = -1;
6864
6865                 if (rdev->raid_disk < mddev->raid_disks)
6866                         if (info->state & (1<<MD_DISK_SYNC))
6867                                 set_bit(In_sync, &rdev->flags);
6868
6869                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6870                         set_bit(WriteMostly, &rdev->flags);
6871                 if (info->state & (1<<MD_DISK_FAILFAST))
6872                         set_bit(FailFast, &rdev->flags);
6873
6874                 if (!mddev->persistent) {
6875                         pr_debug("md: nonpersistent superblock ...\n");
6876                         rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6877                 } else
6878                         rdev->sb_start = calc_dev_sboffset(rdev);
6879                 rdev->sectors = rdev->sb_start;
6880
6881                 err = bind_rdev_to_array(rdev, mddev);
6882                 if (err) {
6883                         export_rdev(rdev);
6884                         return err;
6885                 }
6886         }
6887
6888         return 0;
6889 }
6890
6891 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6892 {
6893         struct md_rdev *rdev;
6894
6895         if (!mddev->pers)
6896                 return -ENODEV;
6897
6898         rdev = find_rdev(mddev, dev);
6899         if (!rdev)
6900                 return -ENXIO;
6901
6902         if (rdev->raid_disk < 0)
6903                 goto kick_rdev;
6904
6905         clear_bit(Blocked, &rdev->flags);
6906         remove_and_add_spares(mddev, rdev);
6907
6908         if (rdev->raid_disk >= 0)
6909                 goto busy;
6910
6911 kick_rdev:
6912         if (mddev_is_clustered(mddev)) {
6913                 if (md_cluster_ops->remove_disk(mddev, rdev))
6914                         goto busy;
6915         }
6916
6917         md_kick_rdev_from_array(rdev);
6918         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6919         if (mddev->thread)
6920                 md_wakeup_thread(mddev->thread);
6921         else
6922                 md_update_sb(mddev, 1);
6923         md_new_event();
6924
6925         return 0;
6926 busy:
6927         pr_debug("md: cannot remove active disk %pg from %s ...\n",
6928                  rdev->bdev, mdname(mddev));
6929         return -EBUSY;
6930 }
6931
6932 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6933 {
6934         int err;
6935         struct md_rdev *rdev;
6936
6937         if (!mddev->pers)
6938                 return -ENODEV;
6939
6940         if (mddev->major_version != 0) {
6941                 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6942                         mdname(mddev));
6943                 return -EINVAL;
6944         }
6945         if (!mddev->pers->hot_add_disk) {
6946                 pr_warn("%s: personality does not support diskops!\n",
6947                         mdname(mddev));
6948                 return -EINVAL;
6949         }
6950
6951         rdev = md_import_device(dev, -1, 0);
6952         if (IS_ERR(rdev)) {
6953                 pr_warn("md: error, md_import_device() returned %ld\n",
6954                         PTR_ERR(rdev));
6955                 return -EINVAL;
6956         }
6957
6958         if (mddev->persistent)
6959                 rdev->sb_start = calc_dev_sboffset(rdev);
6960         else
6961                 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6962
6963         rdev->sectors = rdev->sb_start;
6964
6965         if (test_bit(Faulty, &rdev->flags)) {
6966                 pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
6967                         rdev->bdev, mdname(mddev));
6968                 err = -EINVAL;
6969                 goto abort_export;
6970         }
6971
6972         clear_bit(In_sync, &rdev->flags);
6973         rdev->desc_nr = -1;
6974         rdev->saved_raid_disk = -1;
6975         err = bind_rdev_to_array(rdev, mddev);
6976         if (err)
6977                 goto abort_export;
6978
6979         /*
6980          * The rest should better be atomic, we can have disk failures
6981          * noticed in interrupt contexts ...
6982          */
6983
6984         rdev->raid_disk = -1;
6985
6986         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6987         if (!mddev->thread)
6988                 md_update_sb(mddev, 1);
6989         /*
6990          * If the new disk does not support REQ_NOWAIT,
6991          * disable on the whole MD.
6992          */
6993         if (!bdev_nowait(rdev->bdev)) {
6994                 pr_info("%s: Disabling nowait because %pg does not support nowait\n",
6995                         mdname(mddev), rdev->bdev);
6996                 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
6997         }
6998         /*
6999          * Kick recovery, maybe this spare has to be added to the
7000          * array immediately.
7001          */
7002         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7003         md_wakeup_thread(mddev->thread);
7004         md_new_event();
7005         return 0;
7006
7007 abort_export:
7008         export_rdev(rdev);
7009         return err;
7010 }
7011
7012 static int set_bitmap_file(struct mddev *mddev, int fd)
7013 {
7014         int err = 0;
7015
7016         if (mddev->pers) {
7017                 if (!mddev->pers->quiesce || !mddev->thread)
7018                         return -EBUSY;
7019                 if (mddev->recovery || mddev->sync_thread)
7020                         return -EBUSY;
7021                 /* we should be able to change the bitmap.. */
7022         }
7023
7024         if (fd >= 0) {
7025                 struct inode *inode;
7026                 struct file *f;
7027
7028                 if (mddev->bitmap || mddev->bitmap_info.file)
7029                         return -EEXIST; /* cannot add when bitmap is present */
7030                 f = fget(fd);
7031
7032                 if (f == NULL) {
7033                         pr_warn("%s: error: failed to get bitmap file\n",
7034                                 mdname(mddev));
7035                         return -EBADF;
7036                 }
7037
7038                 inode = f->f_mapping->host;
7039                 if (!S_ISREG(inode->i_mode)) {
7040                         pr_warn("%s: error: bitmap file must be a regular file\n",
7041                                 mdname(mddev));
7042                         err = -EBADF;
7043                 } else if (!(f->f_mode & FMODE_WRITE)) {
7044                         pr_warn("%s: error: bitmap file must open for write\n",
7045                                 mdname(mddev));
7046                         err = -EBADF;
7047                 } else if (atomic_read(&inode->i_writecount) != 1) {
7048                         pr_warn("%s: error: bitmap file is already in use\n",
7049                                 mdname(mddev));
7050                         err = -EBUSY;
7051                 }
7052                 if (err) {
7053                         fput(f);
7054                         return err;
7055                 }
7056                 mddev->bitmap_info.file = f;
7057                 mddev->bitmap_info.offset = 0; /* file overrides offset */
7058         } else if (mddev->bitmap == NULL)
7059                 return -ENOENT; /* cannot remove what isn't there */
7060         err = 0;
7061         if (mddev->pers) {
7062                 if (fd >= 0) {
7063                         struct bitmap *bitmap;
7064
7065                         bitmap = md_bitmap_create(mddev, -1);
7066                         mddev_suspend(mddev);
7067                         if (!IS_ERR(bitmap)) {
7068                                 mddev->bitmap = bitmap;
7069                                 err = md_bitmap_load(mddev);
7070                         } else
7071                                 err = PTR_ERR(bitmap);
7072                         if (err) {
7073                                 md_bitmap_destroy(mddev);
7074                                 fd = -1;
7075                         }
7076                         mddev_resume(mddev);
7077                 } else if (fd < 0) {
7078                         mddev_suspend(mddev);
7079                         md_bitmap_destroy(mddev);
7080                         mddev_resume(mddev);
7081                 }
7082         }
7083         if (fd < 0) {
7084                 struct file *f = mddev->bitmap_info.file;
7085                 if (f) {
7086                         spin_lock(&mddev->lock);
7087                         mddev->bitmap_info.file = NULL;
7088                         spin_unlock(&mddev->lock);
7089                         fput(f);
7090                 }
7091         }
7092
7093         return err;
7094 }
7095
7096 /*
7097  * md_set_array_info is used two different ways
7098  * The original usage is when creating a new array.
7099  * In this usage, raid_disks is > 0 and it together with
7100  *  level, size, not_persistent,layout,chunksize determine the
7101  *  shape of the array.
7102  *  This will always create an array with a type-0.90.0 superblock.
7103  * The newer usage is when assembling an array.
7104  *  In this case raid_disks will be 0, and the major_version field is
7105  *  use to determine which style super-blocks are to be found on the devices.
7106  *  The minor and patch _version numbers are also kept incase the
7107  *  super_block handler wishes to interpret them.
7108  */
7109 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7110 {
7111         if (info->raid_disks == 0) {
7112                 /* just setting version number for superblock loading */
7113                 if (info->major_version < 0 ||
7114                     info->major_version >= ARRAY_SIZE(super_types) ||
7115                     super_types[info->major_version].name == NULL) {
7116                         /* maybe try to auto-load a module? */
7117                         pr_warn("md: superblock version %d not known\n",
7118                                 info->major_version);
7119                         return -EINVAL;
7120                 }
7121                 mddev->major_version = info->major_version;
7122                 mddev->minor_version = info->minor_version;
7123                 mddev->patch_version = info->patch_version;
7124                 mddev->persistent = !info->not_persistent;
7125                 /* ensure mddev_put doesn't delete this now that there
7126                  * is some minimal configuration.
7127                  */
7128                 mddev->ctime         = ktime_get_real_seconds();
7129                 return 0;
7130         }
7131         mddev->major_version = MD_MAJOR_VERSION;
7132         mddev->minor_version = MD_MINOR_VERSION;
7133         mddev->patch_version = MD_PATCHLEVEL_VERSION;
7134         mddev->ctime         = ktime_get_real_seconds();
7135
7136         mddev->level         = info->level;
7137         mddev->clevel[0]     = 0;
7138         mddev->dev_sectors   = 2 * (sector_t)info->size;
7139         mddev->raid_disks    = info->raid_disks;
7140         /* don't set md_minor, it is determined by which /dev/md* was
7141          * openned
7142          */
7143         if (info->state & (1<<MD_SB_CLEAN))
7144                 mddev->recovery_cp = MaxSector;
7145         else
7146                 mddev->recovery_cp = 0;
7147         mddev->persistent    = ! info->not_persistent;
7148         mddev->external      = 0;
7149
7150         mddev->layout        = info->layout;
7151         if (mddev->level == 0)
7152                 /* Cannot trust RAID0 layout info here */
7153                 mddev->layout = -1;
7154         mddev->chunk_sectors = info->chunk_size >> 9;
7155
7156         if (mddev->persistent) {
7157                 mddev->max_disks = MD_SB_DISKS;
7158                 mddev->flags = 0;
7159                 mddev->sb_flags = 0;
7160         }
7161         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7162
7163         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7164         mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7165         mddev->bitmap_info.offset = 0;
7166
7167         mddev->reshape_position = MaxSector;
7168
7169         /*
7170          * Generate a 128 bit UUID
7171          */
7172         get_random_bytes(mddev->uuid, 16);
7173
7174         mddev->new_level = mddev->level;
7175         mddev->new_chunk_sectors = mddev->chunk_sectors;
7176         mddev->new_layout = mddev->layout;
7177         mddev->delta_disks = 0;
7178         mddev->reshape_backwards = 0;
7179
7180         return 0;
7181 }
7182
7183 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7184 {
7185         lockdep_assert_held(&mddev->reconfig_mutex);
7186
7187         if (mddev->external_size)
7188                 return;
7189
7190         mddev->array_sectors = array_sectors;
7191 }
7192 EXPORT_SYMBOL(md_set_array_sectors);
7193
7194 static int update_size(struct mddev *mddev, sector_t num_sectors)
7195 {
7196         struct md_rdev *rdev;
7197         int rv;
7198         int fit = (num_sectors == 0);
7199         sector_t old_dev_sectors = mddev->dev_sectors;
7200
7201         if (mddev->pers->resize == NULL)
7202                 return -EINVAL;
7203         /* The "num_sectors" is the number of sectors of each device that
7204          * is used.  This can only make sense for arrays with redundancy.
7205          * linear and raid0 always use whatever space is available. We can only
7206          * consider changing this number if no resync or reconstruction is
7207          * happening, and if the new size is acceptable. It must fit before the
7208          * sb_start or, if that is <data_offset, it must fit before the size
7209          * of each device.  If num_sectors is zero, we find the largest size
7210          * that fits.
7211          */
7212         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7213             mddev->sync_thread)
7214                 return -EBUSY;
7215         if (!md_is_rdwr(mddev))
7216                 return -EROFS;
7217
7218         rdev_for_each(rdev, mddev) {
7219                 sector_t avail = rdev->sectors;
7220
7221                 if (fit && (num_sectors == 0 || num_sectors > avail))
7222                         num_sectors = avail;
7223                 if (avail < num_sectors)
7224                         return -ENOSPC;
7225         }
7226         rv = mddev->pers->resize(mddev, num_sectors);
7227         if (!rv) {
7228                 if (mddev_is_clustered(mddev))
7229                         md_cluster_ops->update_size(mddev, old_dev_sectors);
7230                 else if (mddev->queue) {
7231                         set_capacity_and_notify(mddev->gendisk,
7232                                                 mddev->array_sectors);
7233                 }
7234         }
7235         return rv;
7236 }
7237
7238 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7239 {
7240         int rv;
7241         struct md_rdev *rdev;
7242         /* change the number of raid disks */
7243         if (mddev->pers->check_reshape == NULL)
7244                 return -EINVAL;
7245         if (!md_is_rdwr(mddev))
7246                 return -EROFS;
7247         if (raid_disks <= 0 ||
7248             (mddev->max_disks && raid_disks >= mddev->max_disks))
7249                 return -EINVAL;
7250         if (mddev->sync_thread ||
7251             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7252             test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7253             mddev->reshape_position != MaxSector)
7254                 return -EBUSY;
7255
7256         rdev_for_each(rdev, mddev) {
7257                 if (mddev->raid_disks < raid_disks &&
7258                     rdev->data_offset < rdev->new_data_offset)
7259                         return -EINVAL;
7260                 if (mddev->raid_disks > raid_disks &&
7261                     rdev->data_offset > rdev->new_data_offset)
7262                         return -EINVAL;
7263         }
7264
7265         mddev->delta_disks = raid_disks - mddev->raid_disks;
7266         if (mddev->delta_disks < 0)
7267                 mddev->reshape_backwards = 1;
7268         else if (mddev->delta_disks > 0)
7269                 mddev->reshape_backwards = 0;
7270
7271         rv = mddev->pers->check_reshape(mddev);
7272         if (rv < 0) {
7273                 mddev->delta_disks = 0;
7274                 mddev->reshape_backwards = 0;
7275         }
7276         return rv;
7277 }
7278
7279 /*
7280  * update_array_info is used to change the configuration of an
7281  * on-line array.
7282  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7283  * fields in the info are checked against the array.
7284  * Any differences that cannot be handled will cause an error.
7285  * Normally, only one change can be managed at a time.
7286  */
7287 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7288 {
7289         int rv = 0;
7290         int cnt = 0;
7291         int state = 0;
7292
7293         /* calculate expected state,ignoring low bits */
7294         if (mddev->bitmap && mddev->bitmap_info.offset)
7295                 state |= (1 << MD_SB_BITMAP_PRESENT);
7296
7297         if (mddev->major_version != info->major_version ||
7298             mddev->minor_version != info->minor_version ||
7299 /*          mddev->patch_version != info->patch_version || */
7300             mddev->ctime         != info->ctime         ||
7301             mddev->level         != info->level         ||
7302 /*          mddev->layout        != info->layout        || */
7303             mddev->persistent    != !info->not_persistent ||
7304             mddev->chunk_sectors != info->chunk_size >> 9 ||
7305             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7306             ((state^info->state) & 0xfffffe00)
7307                 )
7308                 return -EINVAL;
7309         /* Check there is only one change */
7310         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7311                 cnt++;
7312         if (mddev->raid_disks != info->raid_disks)
7313                 cnt++;
7314         if (mddev->layout != info->layout)
7315                 cnt++;
7316         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7317                 cnt++;
7318         if (cnt == 0)
7319                 return 0;
7320         if (cnt > 1)
7321                 return -EINVAL;
7322
7323         if (mddev->layout != info->layout) {
7324                 /* Change layout
7325                  * we don't need to do anything at the md level, the
7326                  * personality will take care of it all.
7327                  */
7328                 if (mddev->pers->check_reshape == NULL)
7329                         return -EINVAL;
7330                 else {
7331                         mddev->new_layout = info->layout;
7332                         rv = mddev->pers->check_reshape(mddev);
7333                         if (rv)
7334                                 mddev->new_layout = mddev->layout;
7335                         return rv;
7336                 }
7337         }
7338         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7339                 rv = update_size(mddev, (sector_t)info->size * 2);
7340
7341         if (mddev->raid_disks    != info->raid_disks)
7342                 rv = update_raid_disks(mddev, info->raid_disks);
7343
7344         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7345                 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7346                         rv = -EINVAL;
7347                         goto err;
7348                 }
7349                 if (mddev->recovery || mddev->sync_thread) {
7350                         rv = -EBUSY;
7351                         goto err;
7352                 }
7353                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7354                         struct bitmap *bitmap;
7355                         /* add the bitmap */
7356                         if (mddev->bitmap) {
7357                                 rv = -EEXIST;
7358                                 goto err;
7359                         }
7360                         if (mddev->bitmap_info.default_offset == 0) {
7361                                 rv = -EINVAL;
7362                                 goto err;
7363                         }
7364                         mddev->bitmap_info.offset =
7365                                 mddev->bitmap_info.default_offset;
7366                         mddev->bitmap_info.space =
7367                                 mddev->bitmap_info.default_space;
7368                         bitmap = md_bitmap_create(mddev, -1);
7369                         mddev_suspend(mddev);
7370                         if (!IS_ERR(bitmap)) {
7371                                 mddev->bitmap = bitmap;
7372                                 rv = md_bitmap_load(mddev);
7373                         } else
7374                                 rv = PTR_ERR(bitmap);
7375                         if (rv)
7376                                 md_bitmap_destroy(mddev);
7377                         mddev_resume(mddev);
7378                 } else {
7379                         /* remove the bitmap */
7380                         if (!mddev->bitmap) {
7381                                 rv = -ENOENT;
7382                                 goto err;
7383                         }
7384                         if (mddev->bitmap->storage.file) {
7385                                 rv = -EINVAL;
7386                                 goto err;
7387                         }
7388                         if (mddev->bitmap_info.nodes) {
7389                                 /* hold PW on all the bitmap lock */
7390                                 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7391                                         pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7392                                         rv = -EPERM;
7393                                         md_cluster_ops->unlock_all_bitmaps(mddev);
7394                                         goto err;
7395                                 }
7396
7397                                 mddev->bitmap_info.nodes = 0;
7398                                 md_cluster_ops->leave(mddev);
7399                                 module_put(md_cluster_mod);
7400                                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7401                         }
7402                         mddev_suspend(mddev);
7403                         md_bitmap_destroy(mddev);
7404                         mddev_resume(mddev);
7405                         mddev->bitmap_info.offset = 0;
7406                 }
7407         }
7408         md_update_sb(mddev, 1);
7409         return rv;
7410 err:
7411         return rv;
7412 }
7413
7414 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7415 {
7416         struct md_rdev *rdev;
7417         int err = 0;
7418
7419         if (mddev->pers == NULL)
7420                 return -ENODEV;
7421
7422         rcu_read_lock();
7423         rdev = md_find_rdev_rcu(mddev, dev);
7424         if (!rdev)
7425                 err =  -ENODEV;
7426         else {
7427                 md_error(mddev, rdev);
7428                 if (test_bit(MD_BROKEN, &mddev->flags))
7429                         err = -EBUSY;
7430         }
7431         rcu_read_unlock();
7432         return err;
7433 }
7434
7435 /*
7436  * We have a problem here : there is no easy way to give a CHS
7437  * virtual geometry. We currently pretend that we have a 2 heads
7438  * 4 sectors (with a BIG number of cylinders...). This drives
7439  * dosfs just mad... ;-)
7440  */
7441 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7442 {
7443         struct mddev *mddev = bdev->bd_disk->private_data;
7444
7445         geo->heads = 2;
7446         geo->sectors = 4;
7447         geo->cylinders = mddev->array_sectors / 8;
7448         return 0;
7449 }
7450
7451 static inline bool md_ioctl_valid(unsigned int cmd)
7452 {
7453         switch (cmd) {
7454         case ADD_NEW_DISK:
7455         case GET_ARRAY_INFO:
7456         case GET_BITMAP_FILE:
7457         case GET_DISK_INFO:
7458         case HOT_ADD_DISK:
7459         case HOT_REMOVE_DISK:
7460         case RAID_VERSION:
7461         case RESTART_ARRAY_RW:
7462         case RUN_ARRAY:
7463         case SET_ARRAY_INFO:
7464         case SET_BITMAP_FILE:
7465         case SET_DISK_FAULTY:
7466         case STOP_ARRAY:
7467         case STOP_ARRAY_RO:
7468         case CLUSTERED_DISK_NACK:
7469                 return true;
7470         default:
7471                 return false;
7472         }
7473 }
7474
7475 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7476 {
7477         mdu_array_info_t info;
7478         int err;
7479
7480         if (!argp)
7481                 memset(&info, 0, sizeof(info));
7482         else if (copy_from_user(&info, argp, sizeof(info)))
7483                 return -EFAULT;
7484
7485         if (mddev->pers) {
7486                 err = update_array_info(mddev, &info);
7487                 if (err)
7488                         pr_warn("md: couldn't update array info. %d\n", err);
7489                 return err;
7490         }
7491
7492         if (!list_empty(&mddev->disks)) {
7493                 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7494                 return -EBUSY;
7495         }
7496
7497         if (mddev->raid_disks) {
7498                 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7499                 return -EBUSY;
7500         }
7501
7502         err = md_set_array_info(mddev, &info);
7503         if (err)
7504                 pr_warn("md: couldn't set array info. %d\n", err);
7505
7506         return err;
7507 }
7508
7509 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7510                         unsigned int cmd, unsigned long arg)
7511 {
7512         int err = 0;
7513         void __user *argp = (void __user *)arg;
7514         struct mddev *mddev = NULL;
7515         bool did_set_md_closing = false;
7516
7517         if (!md_ioctl_valid(cmd))
7518                 return -ENOTTY;
7519
7520         switch (cmd) {
7521         case RAID_VERSION:
7522         case GET_ARRAY_INFO:
7523         case GET_DISK_INFO:
7524                 break;
7525         default:
7526                 if (!capable(CAP_SYS_ADMIN))
7527                         return -EACCES;
7528         }
7529
7530         /*
7531          * Commands dealing with the RAID driver but not any
7532          * particular array:
7533          */
7534         switch (cmd) {
7535         case RAID_VERSION:
7536                 err = get_version(argp);
7537                 goto out;
7538         default:;
7539         }
7540
7541         /*
7542          * Commands creating/starting a new array:
7543          */
7544
7545         mddev = bdev->bd_disk->private_data;
7546
7547         if (!mddev) {
7548                 BUG();
7549                 goto out;
7550         }
7551
7552         /* Some actions do not requires the mutex */
7553         switch (cmd) {
7554         case GET_ARRAY_INFO:
7555                 if (!mddev->raid_disks && !mddev->external)
7556                         err = -ENODEV;
7557                 else
7558                         err = get_array_info(mddev, argp);
7559                 goto out;
7560
7561         case GET_DISK_INFO:
7562                 if (!mddev->raid_disks && !mddev->external)
7563                         err = -ENODEV;
7564                 else
7565                         err = get_disk_info(mddev, argp);
7566                 goto out;
7567
7568         case SET_DISK_FAULTY:
7569                 err = set_disk_faulty(mddev, new_decode_dev(arg));
7570                 goto out;
7571
7572         case GET_BITMAP_FILE:
7573                 err = get_bitmap_file(mddev, argp);
7574                 goto out;
7575
7576         }
7577
7578         if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7579                 flush_rdev_wq(mddev);
7580
7581         if (cmd == HOT_REMOVE_DISK)
7582                 /* need to ensure recovery thread has run */
7583                 wait_event_interruptible_timeout(mddev->sb_wait,
7584                                                  !test_bit(MD_RECOVERY_NEEDED,
7585                                                            &mddev->recovery),
7586                                                  msecs_to_jiffies(5000));
7587         if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7588                 /* Need to flush page cache, and ensure no-one else opens
7589                  * and writes
7590                  */
7591                 mutex_lock(&mddev->open_mutex);
7592                 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7593                         mutex_unlock(&mddev->open_mutex);
7594                         err = -EBUSY;
7595                         goto out;
7596                 }
7597                 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7598                         mutex_unlock(&mddev->open_mutex);
7599                         err = -EBUSY;
7600                         goto out;
7601                 }
7602                 did_set_md_closing = true;
7603                 mutex_unlock(&mddev->open_mutex);
7604                 sync_blockdev(bdev);
7605         }
7606         err = mddev_lock(mddev);
7607         if (err) {
7608                 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7609                          err, cmd);
7610                 goto out;
7611         }
7612
7613         if (cmd == SET_ARRAY_INFO) {
7614                 err = __md_set_array_info(mddev, argp);
7615                 goto unlock;
7616         }
7617
7618         /*
7619          * Commands querying/configuring an existing array:
7620          */
7621         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7622          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7623         if ((!mddev->raid_disks && !mddev->external)
7624             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7625             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7626             && cmd != GET_BITMAP_FILE) {
7627                 err = -ENODEV;
7628                 goto unlock;
7629         }
7630
7631         /*
7632          * Commands even a read-only array can execute:
7633          */
7634         switch (cmd) {
7635         case RESTART_ARRAY_RW:
7636                 err = restart_array(mddev);
7637                 goto unlock;
7638
7639         case STOP_ARRAY:
7640                 err = do_md_stop(mddev, 0, bdev);
7641                 goto unlock;
7642
7643         case STOP_ARRAY_RO:
7644                 err = md_set_readonly(mddev, bdev);
7645                 goto unlock;
7646
7647         case HOT_REMOVE_DISK:
7648                 err = hot_remove_disk(mddev, new_decode_dev(arg));
7649                 goto unlock;
7650
7651         case ADD_NEW_DISK:
7652                 /* We can support ADD_NEW_DISK on read-only arrays
7653                  * only if we are re-adding a preexisting device.
7654                  * So require mddev->pers and MD_DISK_SYNC.
7655                  */
7656                 if (mddev->pers) {
7657                         mdu_disk_info_t info;
7658                         if (copy_from_user(&info, argp, sizeof(info)))
7659                                 err = -EFAULT;
7660                         else if (!(info.state & (1<<MD_DISK_SYNC)))
7661                                 /* Need to clear read-only for this */
7662                                 break;
7663                         else
7664                                 err = md_add_new_disk(mddev, &info);
7665                         goto unlock;
7666                 }
7667                 break;
7668         }
7669
7670         /*
7671          * The remaining ioctls are changing the state of the
7672          * superblock, so we do not allow them on read-only arrays.
7673          */
7674         if (!md_is_rdwr(mddev) && mddev->pers) {
7675                 if (mddev->ro != MD_AUTO_READ) {
7676                         err = -EROFS;
7677                         goto unlock;
7678                 }
7679                 mddev->ro = MD_RDWR;
7680                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7681                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7682                 /* mddev_unlock will wake thread */
7683                 /* If a device failed while we were read-only, we
7684                  * need to make sure the metadata is updated now.
7685                  */
7686                 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7687                         mddev_unlock(mddev);
7688                         wait_event(mddev->sb_wait,
7689                                    !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7690                                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7691                         mddev_lock_nointr(mddev);
7692                 }
7693         }
7694
7695         switch (cmd) {
7696         case ADD_NEW_DISK:
7697         {
7698                 mdu_disk_info_t info;
7699                 if (copy_from_user(&info, argp, sizeof(info)))
7700                         err = -EFAULT;
7701                 else
7702                         err = md_add_new_disk(mddev, &info);
7703                 goto unlock;
7704         }
7705
7706         case CLUSTERED_DISK_NACK:
7707                 if (mddev_is_clustered(mddev))
7708                         md_cluster_ops->new_disk_ack(mddev, false);
7709                 else
7710                         err = -EINVAL;
7711                 goto unlock;
7712
7713         case HOT_ADD_DISK:
7714                 err = hot_add_disk(mddev, new_decode_dev(arg));
7715                 goto unlock;
7716
7717         case RUN_ARRAY:
7718                 err = do_md_run(mddev);
7719                 goto unlock;
7720
7721         case SET_BITMAP_FILE:
7722                 err = set_bitmap_file(mddev, (int)arg);
7723                 goto unlock;
7724
7725         default:
7726                 err = -EINVAL;
7727                 goto unlock;
7728         }
7729
7730 unlock:
7731         if (mddev->hold_active == UNTIL_IOCTL &&
7732             err != -EINVAL)
7733                 mddev->hold_active = 0;
7734         mddev_unlock(mddev);
7735 out:
7736         if(did_set_md_closing)
7737                 clear_bit(MD_CLOSING, &mddev->flags);
7738         return err;
7739 }
7740 #ifdef CONFIG_COMPAT
7741 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7742                     unsigned int cmd, unsigned long arg)
7743 {
7744         switch (cmd) {
7745         case HOT_REMOVE_DISK:
7746         case HOT_ADD_DISK:
7747         case SET_DISK_FAULTY:
7748         case SET_BITMAP_FILE:
7749                 /* These take in integer arg, do not convert */
7750                 break;
7751         default:
7752                 arg = (unsigned long)compat_ptr(arg);
7753                 break;
7754         }
7755
7756         return md_ioctl(bdev, mode, cmd, arg);
7757 }
7758 #endif /* CONFIG_COMPAT */
7759
7760 static int md_set_read_only(struct block_device *bdev, bool ro)
7761 {
7762         struct mddev *mddev = bdev->bd_disk->private_data;
7763         int err;
7764
7765         err = mddev_lock(mddev);
7766         if (err)
7767                 return err;
7768
7769         if (!mddev->raid_disks && !mddev->external) {
7770                 err = -ENODEV;
7771                 goto out_unlock;
7772         }
7773
7774         /*
7775          * Transitioning to read-auto need only happen for arrays that call
7776          * md_write_start and which are not ready for writes yet.
7777          */
7778         if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7779                 err = restart_array(mddev);
7780                 if (err)
7781                         goto out_unlock;
7782                 mddev->ro = MD_AUTO_READ;
7783         }
7784
7785 out_unlock:
7786         mddev_unlock(mddev);
7787         return err;
7788 }
7789
7790 static int md_open(struct block_device *bdev, fmode_t mode)
7791 {
7792         struct mddev *mddev;
7793         int err;
7794
7795         spin_lock(&all_mddevs_lock);
7796         mddev = mddev_get(bdev->bd_disk->private_data);
7797         spin_unlock(&all_mddevs_lock);
7798         if (!mddev)
7799                 return -ENODEV;
7800
7801         err = mutex_lock_interruptible(&mddev->open_mutex);
7802         if (err)
7803                 goto out;
7804
7805         err = -ENODEV;
7806         if (test_bit(MD_CLOSING, &mddev->flags))
7807                 goto out_unlock;
7808
7809         atomic_inc(&mddev->openers);
7810         mutex_unlock(&mddev->open_mutex);
7811
7812         bdev_check_media_change(bdev);
7813         return 0;
7814
7815 out_unlock:
7816         mutex_unlock(&mddev->open_mutex);
7817 out:
7818         mddev_put(mddev);
7819         return err;
7820 }
7821
7822 static void md_release(struct gendisk *disk, fmode_t mode)
7823 {
7824         struct mddev *mddev = disk->private_data;
7825
7826         BUG_ON(!mddev);
7827         atomic_dec(&mddev->openers);
7828         mddev_put(mddev);
7829 }
7830
7831 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7832 {
7833         struct mddev *mddev = disk->private_data;
7834         unsigned int ret = 0;
7835
7836         if (mddev->changed)
7837                 ret = DISK_EVENT_MEDIA_CHANGE;
7838         mddev->changed = 0;
7839         return ret;
7840 }
7841
7842 static void md_free_disk(struct gendisk *disk)
7843 {
7844         struct mddev *mddev = disk->private_data;
7845
7846         percpu_ref_exit(&mddev->writes_pending);
7847         mddev_free(mddev);
7848 }
7849
7850 const struct block_device_operations md_fops =
7851 {
7852         .owner          = THIS_MODULE,
7853         .submit_bio     = md_submit_bio,
7854         .open           = md_open,
7855         .release        = md_release,
7856         .ioctl          = md_ioctl,
7857 #ifdef CONFIG_COMPAT
7858         .compat_ioctl   = md_compat_ioctl,
7859 #endif
7860         .getgeo         = md_getgeo,
7861         .check_events   = md_check_events,
7862         .set_read_only  = md_set_read_only,
7863         .free_disk      = md_free_disk,
7864 };
7865
7866 static int md_thread(void *arg)
7867 {
7868         struct md_thread *thread = arg;
7869
7870         /*
7871          * md_thread is a 'system-thread', it's priority should be very
7872          * high. We avoid resource deadlocks individually in each
7873          * raid personality. (RAID5 does preallocation) We also use RR and
7874          * the very same RT priority as kswapd, thus we will never get
7875          * into a priority inversion deadlock.
7876          *
7877          * we definitely have to have equal or higher priority than
7878          * bdflush, otherwise bdflush will deadlock if there are too
7879          * many dirty RAID5 blocks.
7880          */
7881
7882         allow_signal(SIGKILL);
7883         while (!kthread_should_stop()) {
7884
7885                 /* We need to wait INTERRUPTIBLE so that
7886                  * we don't add to the load-average.
7887                  * That means we need to be sure no signals are
7888                  * pending
7889                  */
7890                 if (signal_pending(current))
7891                         flush_signals(current);
7892
7893                 wait_event_interruptible_timeout
7894                         (thread->wqueue,
7895                          test_bit(THREAD_WAKEUP, &thread->flags)
7896                          || kthread_should_stop() || kthread_should_park(),
7897                          thread->timeout);
7898
7899                 clear_bit(THREAD_WAKEUP, &thread->flags);
7900                 if (kthread_should_park())
7901                         kthread_parkme();
7902                 if (!kthread_should_stop())
7903                         thread->run(thread);
7904         }
7905
7906         return 0;
7907 }
7908
7909 void md_wakeup_thread(struct md_thread *thread)
7910 {
7911         if (thread) {
7912                 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7913                 set_bit(THREAD_WAKEUP, &thread->flags);
7914                 wake_up(&thread->wqueue);
7915         }
7916 }
7917 EXPORT_SYMBOL(md_wakeup_thread);
7918
7919 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7920                 struct mddev *mddev, const char *name)
7921 {
7922         struct md_thread *thread;
7923
7924         thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7925         if (!thread)
7926                 return NULL;
7927
7928         init_waitqueue_head(&thread->wqueue);
7929
7930         thread->run = run;
7931         thread->mddev = mddev;
7932         thread->timeout = MAX_SCHEDULE_TIMEOUT;
7933         thread->tsk = kthread_run(md_thread, thread,
7934                                   "%s_%s",
7935                                   mdname(thread->mddev),
7936                                   name);
7937         if (IS_ERR(thread->tsk)) {
7938                 kfree(thread);
7939                 return NULL;
7940         }
7941         return thread;
7942 }
7943 EXPORT_SYMBOL(md_register_thread);
7944
7945 void md_unregister_thread(struct md_thread **threadp)
7946 {
7947         struct md_thread *thread;
7948
7949         /*
7950          * Locking ensures that mddev_unlock does not wake_up a
7951          * non-existent thread
7952          */
7953         spin_lock(&pers_lock);
7954         thread = *threadp;
7955         if (!thread) {
7956                 spin_unlock(&pers_lock);
7957                 return;
7958         }
7959         *threadp = NULL;
7960         spin_unlock(&pers_lock);
7961
7962         pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7963         kthread_stop(thread->tsk);
7964         kfree(thread);
7965 }
7966 EXPORT_SYMBOL(md_unregister_thread);
7967
7968 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7969 {
7970         if (!rdev || test_bit(Faulty, &rdev->flags))
7971                 return;
7972
7973         if (!mddev->pers || !mddev->pers->error_handler)
7974                 return;
7975         mddev->pers->error_handler(mddev, rdev);
7976
7977         if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
7978                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7979         sysfs_notify_dirent_safe(rdev->sysfs_state);
7980         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7981         if (!test_bit(MD_BROKEN, &mddev->flags)) {
7982                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7983                 md_wakeup_thread(mddev->thread);
7984         }
7985         if (mddev->event_work.func)
7986                 queue_work(md_misc_wq, &mddev->event_work);
7987         md_new_event();
7988 }
7989 EXPORT_SYMBOL(md_error);
7990
7991 /* seq_file implementation /proc/mdstat */
7992
7993 static void status_unused(struct seq_file *seq)
7994 {
7995         int i = 0;
7996         struct md_rdev *rdev;
7997
7998         seq_printf(seq, "unused devices: ");
7999
8000         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8001                 i++;
8002                 seq_printf(seq, "%pg ", rdev->bdev);
8003         }
8004         if (!i)
8005                 seq_printf(seq, "<none>");
8006
8007         seq_printf(seq, "\n");
8008 }
8009
8010 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8011 {
8012         sector_t max_sectors, resync, res;
8013         unsigned long dt, db = 0;
8014         sector_t rt, curr_mark_cnt, resync_mark_cnt;
8015         int scale, recovery_active;
8016         unsigned int per_milli;
8017
8018         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8019             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8020                 max_sectors = mddev->resync_max_sectors;
8021         else
8022                 max_sectors = mddev->dev_sectors;
8023
8024         resync = mddev->curr_resync;
8025         if (resync < MD_RESYNC_ACTIVE) {
8026                 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8027                         /* Still cleaning up */
8028                         resync = max_sectors;
8029         } else if (resync > max_sectors) {
8030                 resync = max_sectors;
8031         } else {
8032                 resync -= atomic_read(&mddev->recovery_active);
8033                 if (resync < MD_RESYNC_ACTIVE) {
8034                         /*
8035                          * Resync has started, but the subtraction has
8036                          * yielded one of the special values. Force it
8037                          * to active to ensure the status reports an
8038                          * active resync.
8039                          */
8040                         resync = MD_RESYNC_ACTIVE;
8041                 }
8042         }
8043
8044         if (resync == MD_RESYNC_NONE) {
8045                 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8046                         struct md_rdev *rdev;
8047
8048                         rdev_for_each(rdev, mddev)
8049                                 if (rdev->raid_disk >= 0 &&
8050                                     !test_bit(Faulty, &rdev->flags) &&
8051                                     rdev->recovery_offset != MaxSector &&
8052                                     rdev->recovery_offset) {
8053                                         seq_printf(seq, "\trecover=REMOTE");
8054                                         return 1;
8055                                 }
8056                         if (mddev->reshape_position != MaxSector)
8057                                 seq_printf(seq, "\treshape=REMOTE");
8058                         else
8059                                 seq_printf(seq, "\tresync=REMOTE");
8060                         return 1;
8061                 }
8062                 if (mddev->recovery_cp < MaxSector) {
8063                         seq_printf(seq, "\tresync=PENDING");
8064                         return 1;
8065                 }
8066                 return 0;
8067         }
8068         if (resync < MD_RESYNC_ACTIVE) {
8069                 seq_printf(seq, "\tresync=DELAYED");
8070                 return 1;
8071         }
8072
8073         WARN_ON(max_sectors == 0);
8074         /* Pick 'scale' such that (resync>>scale)*1000 will fit
8075          * in a sector_t, and (max_sectors>>scale) will fit in a
8076          * u32, as those are the requirements for sector_div.
8077          * Thus 'scale' must be at least 10
8078          */
8079         scale = 10;
8080         if (sizeof(sector_t) > sizeof(unsigned long)) {
8081                 while ( max_sectors/2 > (1ULL<<(scale+32)))
8082                         scale++;
8083         }
8084         res = (resync>>scale)*1000;
8085         sector_div(res, (u32)((max_sectors>>scale)+1));
8086
8087         per_milli = res;
8088         {
8089                 int i, x = per_milli/50, y = 20-x;
8090                 seq_printf(seq, "[");
8091                 for (i = 0; i < x; i++)
8092                         seq_printf(seq, "=");
8093                 seq_printf(seq, ">");
8094                 for (i = 0; i < y; i++)
8095                         seq_printf(seq, ".");
8096                 seq_printf(seq, "] ");
8097         }
8098         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8099                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8100                     "reshape" :
8101                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8102                      "check" :
8103                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8104                       "resync" : "recovery"))),
8105                    per_milli/10, per_milli % 10,
8106                    (unsigned long long) resync/2,
8107                    (unsigned long long) max_sectors/2);
8108
8109         /*
8110          * dt: time from mark until now
8111          * db: blocks written from mark until now
8112          * rt: remaining time
8113          *
8114          * rt is a sector_t, which is always 64bit now. We are keeping
8115          * the original algorithm, but it is not really necessary.
8116          *
8117          * Original algorithm:
8118          *   So we divide before multiply in case it is 32bit and close
8119          *   to the limit.
8120          *   We scale the divisor (db) by 32 to avoid losing precision
8121          *   near the end of resync when the number of remaining sectors
8122          *   is close to 'db'.
8123          *   We then divide rt by 32 after multiplying by db to compensate.
8124          *   The '+1' avoids division by zero if db is very small.
8125          */
8126         dt = ((jiffies - mddev->resync_mark) / HZ);
8127         if (!dt) dt++;
8128
8129         curr_mark_cnt = mddev->curr_mark_cnt;
8130         recovery_active = atomic_read(&mddev->recovery_active);
8131         resync_mark_cnt = mddev->resync_mark_cnt;
8132
8133         if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8134                 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8135
8136         rt = max_sectors - resync;    /* number of remaining sectors */
8137         rt = div64_u64(rt, db/32+1);
8138         rt *= dt;
8139         rt >>= 5;
8140
8141         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8142                    ((unsigned long)rt % 60)/6);
8143
8144         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8145         return 1;
8146 }
8147
8148 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8149 {
8150         struct list_head *tmp;
8151         loff_t l = *pos;
8152         struct mddev *mddev;
8153
8154         if (l == 0x10000) {
8155                 ++*pos;
8156                 return (void *)2;
8157         }
8158         if (l > 0x10000)
8159                 return NULL;
8160         if (!l--)
8161                 /* header */
8162                 return (void*)1;
8163
8164         spin_lock(&all_mddevs_lock);
8165         list_for_each(tmp,&all_mddevs)
8166                 if (!l--) {
8167                         mddev = list_entry(tmp, struct mddev, all_mddevs);
8168                         if (!mddev_get(mddev))
8169                                 continue;
8170                         spin_unlock(&all_mddevs_lock);
8171                         return mddev;
8172                 }
8173         spin_unlock(&all_mddevs_lock);
8174         if (!l--)
8175                 return (void*)2;/* tail */
8176         return NULL;
8177 }
8178
8179 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8180 {
8181         struct list_head *tmp;
8182         struct mddev *next_mddev, *mddev = v;
8183         struct mddev *to_put = NULL;
8184
8185         ++*pos;
8186         if (v == (void*)2)
8187                 return NULL;
8188
8189         spin_lock(&all_mddevs_lock);
8190         if (v == (void*)1) {
8191                 tmp = all_mddevs.next;
8192         } else {
8193                 to_put = mddev;
8194                 tmp = mddev->all_mddevs.next;
8195         }
8196
8197         for (;;) {
8198                 if (tmp == &all_mddevs) {
8199                         next_mddev = (void*)2;
8200                         *pos = 0x10000;
8201                         break;
8202                 }
8203                 next_mddev = list_entry(tmp, struct mddev, all_mddevs);
8204                 if (mddev_get(next_mddev))
8205                         break;
8206                 mddev = next_mddev;
8207                 tmp = mddev->all_mddevs.next;
8208         }
8209         spin_unlock(&all_mddevs_lock);
8210
8211         if (to_put)
8212                 mddev_put(mddev);
8213         return next_mddev;
8214
8215 }
8216
8217 static void md_seq_stop(struct seq_file *seq, void *v)
8218 {
8219         struct mddev *mddev = v;
8220
8221         if (mddev && v != (void*)1 && v != (void*)2)
8222                 mddev_put(mddev);
8223 }
8224
8225 static int md_seq_show(struct seq_file *seq, void *v)
8226 {
8227         struct mddev *mddev = v;
8228         sector_t sectors;
8229         struct md_rdev *rdev;
8230
8231         if (v == (void*)1) {
8232                 struct md_personality *pers;
8233                 seq_printf(seq, "Personalities : ");
8234                 spin_lock(&pers_lock);
8235                 list_for_each_entry(pers, &pers_list, list)
8236                         seq_printf(seq, "[%s] ", pers->name);
8237
8238                 spin_unlock(&pers_lock);
8239                 seq_printf(seq, "\n");
8240                 seq->poll_event = atomic_read(&md_event_count);
8241                 return 0;
8242         }
8243         if (v == (void*)2) {
8244                 status_unused(seq);
8245                 return 0;
8246         }
8247
8248         spin_lock(&mddev->lock);
8249         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8250                 seq_printf(seq, "%s : %sactive", mdname(mddev),
8251                                                 mddev->pers ? "" : "in");
8252                 if (mddev->pers) {
8253                         if (mddev->ro == MD_RDONLY)
8254                                 seq_printf(seq, " (read-only)");
8255                         if (mddev->ro == MD_AUTO_READ)
8256                                 seq_printf(seq, " (auto-read-only)");
8257                         seq_printf(seq, " %s", mddev->pers->name);
8258                 }
8259
8260                 sectors = 0;
8261                 rcu_read_lock();
8262                 rdev_for_each_rcu(rdev, mddev) {
8263                         seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8264
8265                         if (test_bit(WriteMostly, &rdev->flags))
8266                                 seq_printf(seq, "(W)");
8267                         if (test_bit(Journal, &rdev->flags))
8268                                 seq_printf(seq, "(J)");
8269                         if (test_bit(Faulty, &rdev->flags)) {
8270                                 seq_printf(seq, "(F)");
8271                                 continue;
8272                         }
8273                         if (rdev->raid_disk < 0)
8274                                 seq_printf(seq, "(S)"); /* spare */
8275                         if (test_bit(Replacement, &rdev->flags))
8276                                 seq_printf(seq, "(R)");
8277                         sectors += rdev->sectors;
8278                 }
8279                 rcu_read_unlock();
8280
8281                 if (!list_empty(&mddev->disks)) {
8282                         if (mddev->pers)
8283                                 seq_printf(seq, "\n      %llu blocks",
8284                                            (unsigned long long)
8285                                            mddev->array_sectors / 2);
8286                         else
8287                                 seq_printf(seq, "\n      %llu blocks",
8288                                            (unsigned long long)sectors / 2);
8289                 }
8290                 if (mddev->persistent) {
8291                         if (mddev->major_version != 0 ||
8292                             mddev->minor_version != 90) {
8293                                 seq_printf(seq," super %d.%d",
8294                                            mddev->major_version,
8295                                            mddev->minor_version);
8296                         }
8297                 } else if (mddev->external)
8298                         seq_printf(seq, " super external:%s",
8299                                    mddev->metadata_type);
8300                 else
8301                         seq_printf(seq, " super non-persistent");
8302
8303                 if (mddev->pers) {
8304                         mddev->pers->status(seq, mddev);
8305                         seq_printf(seq, "\n      ");
8306                         if (mddev->pers->sync_request) {
8307                                 if (status_resync(seq, mddev))
8308                                         seq_printf(seq, "\n      ");
8309                         }
8310                 } else
8311                         seq_printf(seq, "\n       ");
8312
8313                 md_bitmap_status(seq, mddev->bitmap);
8314
8315                 seq_printf(seq, "\n");
8316         }
8317         spin_unlock(&mddev->lock);
8318
8319         return 0;
8320 }
8321
8322 static const struct seq_operations md_seq_ops = {
8323         .start  = md_seq_start,
8324         .next   = md_seq_next,
8325         .stop   = md_seq_stop,
8326         .show   = md_seq_show,
8327 };
8328
8329 static int md_seq_open(struct inode *inode, struct file *file)
8330 {
8331         struct seq_file *seq;
8332         int error;
8333
8334         error = seq_open(file, &md_seq_ops);
8335         if (error)
8336                 return error;
8337
8338         seq = file->private_data;
8339         seq->poll_event = atomic_read(&md_event_count);
8340         return error;
8341 }
8342
8343 static int md_unloading;
8344 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8345 {
8346         struct seq_file *seq = filp->private_data;
8347         __poll_t mask;
8348
8349         if (md_unloading)
8350                 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8351         poll_wait(filp, &md_event_waiters, wait);
8352
8353         /* always allow read */
8354         mask = EPOLLIN | EPOLLRDNORM;
8355
8356         if (seq->poll_event != atomic_read(&md_event_count))
8357                 mask |= EPOLLERR | EPOLLPRI;
8358         return mask;
8359 }
8360
8361 static const struct proc_ops mdstat_proc_ops = {
8362         .proc_open      = md_seq_open,
8363         .proc_read      = seq_read,
8364         .proc_lseek     = seq_lseek,
8365         .proc_release   = seq_release,
8366         .proc_poll      = mdstat_poll,
8367 };
8368
8369 int register_md_personality(struct md_personality *p)
8370 {
8371         pr_debug("md: %s personality registered for level %d\n",
8372                  p->name, p->level);
8373         spin_lock(&pers_lock);
8374         list_add_tail(&p->list, &pers_list);
8375         spin_unlock(&pers_lock);
8376         return 0;
8377 }
8378 EXPORT_SYMBOL(register_md_personality);
8379
8380 int unregister_md_personality(struct md_personality *p)
8381 {
8382         pr_debug("md: %s personality unregistered\n", p->name);
8383         spin_lock(&pers_lock);
8384         list_del_init(&p->list);
8385         spin_unlock(&pers_lock);
8386         return 0;
8387 }
8388 EXPORT_SYMBOL(unregister_md_personality);
8389
8390 int register_md_cluster_operations(struct md_cluster_operations *ops,
8391                                    struct module *module)
8392 {
8393         int ret = 0;
8394         spin_lock(&pers_lock);
8395         if (md_cluster_ops != NULL)
8396                 ret = -EALREADY;
8397         else {
8398                 md_cluster_ops = ops;
8399                 md_cluster_mod = module;
8400         }
8401         spin_unlock(&pers_lock);
8402         return ret;
8403 }
8404 EXPORT_SYMBOL(register_md_cluster_operations);
8405
8406 int unregister_md_cluster_operations(void)
8407 {
8408         spin_lock(&pers_lock);
8409         md_cluster_ops = NULL;
8410         spin_unlock(&pers_lock);
8411         return 0;
8412 }
8413 EXPORT_SYMBOL(unregister_md_cluster_operations);
8414
8415 int md_setup_cluster(struct mddev *mddev, int nodes)
8416 {
8417         int ret;
8418         if (!md_cluster_ops)
8419                 request_module("md-cluster");
8420         spin_lock(&pers_lock);
8421         /* ensure module won't be unloaded */
8422         if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8423                 pr_warn("can't find md-cluster module or get its reference.\n");
8424                 spin_unlock(&pers_lock);
8425                 return -ENOENT;
8426         }
8427         spin_unlock(&pers_lock);
8428
8429         ret = md_cluster_ops->join(mddev, nodes);
8430         if (!ret)
8431                 mddev->safemode_delay = 0;
8432         return ret;
8433 }
8434
8435 void md_cluster_stop(struct mddev *mddev)
8436 {
8437         if (!md_cluster_ops)
8438                 return;
8439         md_cluster_ops->leave(mddev);
8440         module_put(md_cluster_mod);
8441 }
8442
8443 static int is_mddev_idle(struct mddev *mddev, int init)
8444 {
8445         struct md_rdev *rdev;
8446         int idle;
8447         int curr_events;
8448
8449         idle = 1;
8450         rcu_read_lock();
8451         rdev_for_each_rcu(rdev, mddev) {
8452                 struct gendisk *disk = rdev->bdev->bd_disk;
8453                 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8454                               atomic_read(&disk->sync_io);
8455                 /* sync IO will cause sync_io to increase before the disk_stats
8456                  * as sync_io is counted when a request starts, and
8457                  * disk_stats is counted when it completes.
8458                  * So resync activity will cause curr_events to be smaller than
8459                  * when there was no such activity.
8460                  * non-sync IO will cause disk_stat to increase without
8461                  * increasing sync_io so curr_events will (eventually)
8462                  * be larger than it was before.  Once it becomes
8463                  * substantially larger, the test below will cause
8464                  * the array to appear non-idle, and resync will slow
8465                  * down.
8466                  * If there is a lot of outstanding resync activity when
8467                  * we set last_event to curr_events, then all that activity
8468                  * completing might cause the array to appear non-idle
8469                  * and resync will be slowed down even though there might
8470                  * not have been non-resync activity.  This will only
8471                  * happen once though.  'last_events' will soon reflect
8472                  * the state where there is little or no outstanding
8473                  * resync requests, and further resync activity will
8474                  * always make curr_events less than last_events.
8475                  *
8476                  */
8477                 if (init || curr_events - rdev->last_events > 64) {
8478                         rdev->last_events = curr_events;
8479                         idle = 0;
8480                 }
8481         }
8482         rcu_read_unlock();
8483         return idle;
8484 }
8485
8486 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8487 {
8488         /* another "blocks" (512byte) blocks have been synced */
8489         atomic_sub(blocks, &mddev->recovery_active);
8490         wake_up(&mddev->recovery_wait);
8491         if (!ok) {
8492                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8493                 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8494                 md_wakeup_thread(mddev->thread);
8495                 // stop recovery, signal do_sync ....
8496         }
8497 }
8498 EXPORT_SYMBOL(md_done_sync);
8499
8500 /* md_write_start(mddev, bi)
8501  * If we need to update some array metadata (e.g. 'active' flag
8502  * in superblock) before writing, schedule a superblock update
8503  * and wait for it to complete.
8504  * A return value of 'false' means that the write wasn't recorded
8505  * and cannot proceed as the array is being suspend.
8506  */
8507 bool md_write_start(struct mddev *mddev, struct bio *bi)
8508 {
8509         int did_change = 0;
8510
8511         if (bio_data_dir(bi) != WRITE)
8512                 return true;
8513
8514         BUG_ON(mddev->ro == MD_RDONLY);
8515         if (mddev->ro == MD_AUTO_READ) {
8516                 /* need to switch to read/write */
8517                 mddev->ro = MD_RDWR;
8518                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8519                 md_wakeup_thread(mddev->thread);
8520                 md_wakeup_thread(mddev->sync_thread);
8521                 did_change = 1;
8522         }
8523         rcu_read_lock();
8524         percpu_ref_get(&mddev->writes_pending);
8525         smp_mb(); /* Match smp_mb in set_in_sync() */
8526         if (mddev->safemode == 1)
8527                 mddev->safemode = 0;
8528         /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8529         if (mddev->in_sync || mddev->sync_checkers) {
8530                 spin_lock(&mddev->lock);
8531                 if (mddev->in_sync) {
8532                         mddev->in_sync = 0;
8533                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8534                         set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8535                         md_wakeup_thread(mddev->thread);
8536                         did_change = 1;
8537                 }
8538                 spin_unlock(&mddev->lock);
8539         }
8540         rcu_read_unlock();
8541         if (did_change)
8542                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8543         if (!mddev->has_superblocks)
8544                 return true;
8545         wait_event(mddev->sb_wait,
8546                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8547                    is_md_suspended(mddev));
8548         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8549                 percpu_ref_put(&mddev->writes_pending);
8550                 return false;
8551         }
8552         return true;
8553 }
8554 EXPORT_SYMBOL(md_write_start);
8555
8556 /* md_write_inc can only be called when md_write_start() has
8557  * already been called at least once of the current request.
8558  * It increments the counter and is useful when a single request
8559  * is split into several parts.  Each part causes an increment and
8560  * so needs a matching md_write_end().
8561  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8562  * a spinlocked region.
8563  */
8564 void md_write_inc(struct mddev *mddev, struct bio *bi)
8565 {
8566         if (bio_data_dir(bi) != WRITE)
8567                 return;
8568         WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8569         percpu_ref_get(&mddev->writes_pending);
8570 }
8571 EXPORT_SYMBOL(md_write_inc);
8572
8573 void md_write_end(struct mddev *mddev)
8574 {
8575         percpu_ref_put(&mddev->writes_pending);
8576
8577         if (mddev->safemode == 2)
8578                 md_wakeup_thread(mddev->thread);
8579         else if (mddev->safemode_delay)
8580                 /* The roundup() ensures this only performs locking once
8581                  * every ->safemode_delay jiffies
8582                  */
8583                 mod_timer(&mddev->safemode_timer,
8584                           roundup(jiffies, mddev->safemode_delay) +
8585                           mddev->safemode_delay);
8586 }
8587
8588 EXPORT_SYMBOL(md_write_end);
8589
8590 /* This is used by raid0 and raid10 */
8591 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8592                         struct bio *bio, sector_t start, sector_t size)
8593 {
8594         struct bio *discard_bio = NULL;
8595
8596         if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8597                         &discard_bio) || !discard_bio)
8598                 return;
8599
8600         bio_chain(discard_bio, bio);
8601         bio_clone_blkg_association(discard_bio, bio);
8602         if (mddev->gendisk)
8603                 trace_block_bio_remap(discard_bio,
8604                                 disk_devt(mddev->gendisk),
8605                                 bio->bi_iter.bi_sector);
8606         submit_bio_noacct(discard_bio);
8607 }
8608 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8609
8610 int acct_bioset_init(struct mddev *mddev)
8611 {
8612         int err = 0;
8613
8614         if (!bioset_initialized(&mddev->io_acct_set))
8615                 err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
8616                         offsetof(struct md_io_acct, bio_clone), 0);
8617         return err;
8618 }
8619 EXPORT_SYMBOL_GPL(acct_bioset_init);
8620
8621 void acct_bioset_exit(struct mddev *mddev)
8622 {
8623         bioset_exit(&mddev->io_acct_set);
8624 }
8625 EXPORT_SYMBOL_GPL(acct_bioset_exit);
8626
8627 static void md_end_io_acct(struct bio *bio)
8628 {
8629         struct md_io_acct *md_io_acct = bio->bi_private;
8630         struct bio *orig_bio = md_io_acct->orig_bio;
8631         struct mddev *mddev = md_io_acct->mddev;
8632
8633         orig_bio->bi_status = bio->bi_status;
8634
8635         bio_end_io_acct(orig_bio, md_io_acct->start_time);
8636         bio_put(bio);
8637         bio_endio(orig_bio);
8638
8639         percpu_ref_put(&mddev->active_io);
8640 }
8641
8642 /*
8643  * Used by personalities that don't already clone the bio and thus can't
8644  * easily add the timestamp to their extended bio structure.
8645  */
8646 void md_account_bio(struct mddev *mddev, struct bio **bio)
8647 {
8648         struct block_device *bdev = (*bio)->bi_bdev;
8649         struct md_io_acct *md_io_acct;
8650         struct bio *clone;
8651
8652         if (!blk_queue_io_stat(bdev->bd_disk->queue))
8653                 return;
8654
8655         percpu_ref_get(&mddev->active_io);
8656
8657         clone = bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_acct_set);
8658         md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8659         md_io_acct->orig_bio = *bio;
8660         md_io_acct->start_time = bio_start_io_acct(*bio);
8661         md_io_acct->mddev = mddev;
8662
8663         clone->bi_end_io = md_end_io_acct;
8664         clone->bi_private = md_io_acct;
8665         *bio = clone;
8666 }
8667 EXPORT_SYMBOL_GPL(md_account_bio);
8668
8669 /* md_allow_write(mddev)
8670  * Calling this ensures that the array is marked 'active' so that writes
8671  * may proceed without blocking.  It is important to call this before
8672  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8673  * Must be called with mddev_lock held.
8674  */
8675 void md_allow_write(struct mddev *mddev)
8676 {
8677         if (!mddev->pers)
8678                 return;
8679         if (!md_is_rdwr(mddev))
8680                 return;
8681         if (!mddev->pers->sync_request)
8682                 return;
8683
8684         spin_lock(&mddev->lock);
8685         if (mddev->in_sync) {
8686                 mddev->in_sync = 0;
8687                 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8688                 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8689                 if (mddev->safemode_delay &&
8690                     mddev->safemode == 0)
8691                         mddev->safemode = 1;
8692                 spin_unlock(&mddev->lock);
8693                 md_update_sb(mddev, 0);
8694                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8695                 /* wait for the dirty state to be recorded in the metadata */
8696                 wait_event(mddev->sb_wait,
8697                            !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8698         } else
8699                 spin_unlock(&mddev->lock);
8700 }
8701 EXPORT_SYMBOL_GPL(md_allow_write);
8702
8703 #define SYNC_MARKS      10
8704 #define SYNC_MARK_STEP  (3*HZ)
8705 #define UPDATE_FREQUENCY (5*60*HZ)
8706 void md_do_sync(struct md_thread *thread)
8707 {
8708         struct mddev *mddev = thread->mddev;
8709         struct mddev *mddev2;
8710         unsigned int currspeed = 0, window;
8711         sector_t max_sectors,j, io_sectors, recovery_done;
8712         unsigned long mark[SYNC_MARKS];
8713         unsigned long update_time;
8714         sector_t mark_cnt[SYNC_MARKS];
8715         int last_mark,m;
8716         sector_t last_check;
8717         int skipped = 0;
8718         struct md_rdev *rdev;
8719         char *desc, *action = NULL;
8720         struct blk_plug plug;
8721         int ret;
8722
8723         /* just incase thread restarts... */
8724         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8725             test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8726                 return;
8727         if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8728                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8729                 return;
8730         }
8731
8732         if (mddev_is_clustered(mddev)) {
8733                 ret = md_cluster_ops->resync_start(mddev);
8734                 if (ret)
8735                         goto skip;
8736
8737                 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8738                 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8739                         test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8740                         test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8741                      && ((unsigned long long)mddev->curr_resync_completed
8742                          < (unsigned long long)mddev->resync_max_sectors))
8743                         goto skip;
8744         }
8745
8746         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8747                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8748                         desc = "data-check";
8749                         action = "check";
8750                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8751                         desc = "requested-resync";
8752                         action = "repair";
8753                 } else
8754                         desc = "resync";
8755         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8756                 desc = "reshape";
8757         else
8758                 desc = "recovery";
8759
8760         mddev->last_sync_action = action ?: desc;
8761
8762         /*
8763          * Before starting a resync we must have set curr_resync to
8764          * 2, and then checked that every "conflicting" array has curr_resync
8765          * less than ours.  When we find one that is the same or higher
8766          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8767          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8768          * This will mean we have to start checking from the beginning again.
8769          *
8770          */
8771
8772         do {
8773                 int mddev2_minor = -1;
8774                 mddev->curr_resync = MD_RESYNC_DELAYED;
8775
8776         try_again:
8777                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8778                         goto skip;
8779                 spin_lock(&all_mddevs_lock);
8780                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8781                         if (test_bit(MD_DELETED, &mddev2->flags))
8782                                 continue;
8783                         if (mddev2 == mddev)
8784                                 continue;
8785                         if (!mddev->parallel_resync
8786                         &&  mddev2->curr_resync
8787                         &&  match_mddev_units(mddev, mddev2)) {
8788                                 DEFINE_WAIT(wq);
8789                                 if (mddev < mddev2 &&
8790                                     mddev->curr_resync == MD_RESYNC_DELAYED) {
8791                                         /* arbitrarily yield */
8792                                         mddev->curr_resync = MD_RESYNC_YIELDED;
8793                                         wake_up(&resync_wait);
8794                                 }
8795                                 if (mddev > mddev2 &&
8796                                     mddev->curr_resync == MD_RESYNC_YIELDED)
8797                                         /* no need to wait here, we can wait the next
8798                                          * time 'round when curr_resync == 2
8799                                          */
8800                                         continue;
8801                                 /* We need to wait 'interruptible' so as not to
8802                                  * contribute to the load average, and not to
8803                                  * be caught by 'softlockup'
8804                                  */
8805                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8806                                 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8807                                     mddev2->curr_resync >= mddev->curr_resync) {
8808                                         if (mddev2_minor != mddev2->md_minor) {
8809                                                 mddev2_minor = mddev2->md_minor;
8810                                                 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8811                                                         desc, mdname(mddev),
8812                                                         mdname(mddev2));
8813                                         }
8814                                         spin_unlock(&all_mddevs_lock);
8815
8816                                         if (signal_pending(current))
8817                                                 flush_signals(current);
8818                                         schedule();
8819                                         finish_wait(&resync_wait, &wq);
8820                                         goto try_again;
8821                                 }
8822                                 finish_wait(&resync_wait, &wq);
8823                         }
8824                 }
8825                 spin_unlock(&all_mddevs_lock);
8826         } while (mddev->curr_resync < MD_RESYNC_DELAYED);
8827
8828         j = 0;
8829         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8830                 /* resync follows the size requested by the personality,
8831                  * which defaults to physical size, but can be virtual size
8832                  */
8833                 max_sectors = mddev->resync_max_sectors;
8834                 atomic64_set(&mddev->resync_mismatches, 0);
8835                 /* we don't use the checkpoint if there's a bitmap */
8836                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8837                         j = mddev->resync_min;
8838                 else if (!mddev->bitmap)
8839                         j = mddev->recovery_cp;
8840
8841         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8842                 max_sectors = mddev->resync_max_sectors;
8843                 /*
8844                  * If the original node aborts reshaping then we continue the
8845                  * reshaping, so set j again to avoid restart reshape from the
8846                  * first beginning
8847                  */
8848                 if (mddev_is_clustered(mddev) &&
8849                     mddev->reshape_position != MaxSector)
8850                         j = mddev->reshape_position;
8851         } else {
8852                 /* recovery follows the physical size of devices */
8853                 max_sectors = mddev->dev_sectors;
8854                 j = MaxSector;
8855                 rcu_read_lock();
8856                 rdev_for_each_rcu(rdev, mddev)
8857                         if (rdev->raid_disk >= 0 &&
8858                             !test_bit(Journal, &rdev->flags) &&
8859                             !test_bit(Faulty, &rdev->flags) &&
8860                             !test_bit(In_sync, &rdev->flags) &&
8861                             rdev->recovery_offset < j)
8862                                 j = rdev->recovery_offset;
8863                 rcu_read_unlock();
8864
8865                 /* If there is a bitmap, we need to make sure all
8866                  * writes that started before we added a spare
8867                  * complete before we start doing a recovery.
8868                  * Otherwise the write might complete and (via
8869                  * bitmap_endwrite) set a bit in the bitmap after the
8870                  * recovery has checked that bit and skipped that
8871                  * region.
8872                  */
8873                 if (mddev->bitmap) {
8874                         mddev->pers->quiesce(mddev, 1);
8875                         mddev->pers->quiesce(mddev, 0);
8876                 }
8877         }
8878
8879         pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8880         pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8881         pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8882                  speed_max(mddev), desc);
8883
8884         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8885
8886         io_sectors = 0;
8887         for (m = 0; m < SYNC_MARKS; m++) {
8888                 mark[m] = jiffies;
8889                 mark_cnt[m] = io_sectors;
8890         }
8891         last_mark = 0;
8892         mddev->resync_mark = mark[last_mark];
8893         mddev->resync_mark_cnt = mark_cnt[last_mark];
8894
8895         /*
8896          * Tune reconstruction:
8897          */
8898         window = 32 * (PAGE_SIZE / 512);
8899         pr_debug("md: using %dk window, over a total of %lluk.\n",
8900                  window/2, (unsigned long long)max_sectors/2);
8901
8902         atomic_set(&mddev->recovery_active, 0);
8903         last_check = 0;
8904
8905         if (j >= MD_RESYNC_ACTIVE) {
8906                 pr_debug("md: resuming %s of %s from checkpoint.\n",
8907                          desc, mdname(mddev));
8908                 mddev->curr_resync = j;
8909         } else
8910                 mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
8911         mddev->curr_resync_completed = j;
8912         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8913         md_new_event();
8914         update_time = jiffies;
8915
8916         blk_start_plug(&plug);
8917         while (j < max_sectors) {
8918                 sector_t sectors;
8919
8920                 skipped = 0;
8921
8922                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8923                     ((mddev->curr_resync > mddev->curr_resync_completed &&
8924                       (mddev->curr_resync - mddev->curr_resync_completed)
8925                       > (max_sectors >> 4)) ||
8926                      time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8927                      (j - mddev->curr_resync_completed)*2
8928                      >= mddev->resync_max - mddev->curr_resync_completed ||
8929                      mddev->curr_resync_completed > mddev->resync_max
8930                             )) {
8931                         /* time to update curr_resync_completed */
8932                         wait_event(mddev->recovery_wait,
8933                                    atomic_read(&mddev->recovery_active) == 0);
8934                         mddev->curr_resync_completed = j;
8935                         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8936                             j > mddev->recovery_cp)
8937                                 mddev->recovery_cp = j;
8938                         update_time = jiffies;
8939                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8940                         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8941                 }
8942
8943                 while (j >= mddev->resync_max &&
8944                        !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8945                         /* As this condition is controlled by user-space,
8946                          * we can block indefinitely, so use '_interruptible'
8947                          * to avoid triggering warnings.
8948                          */
8949                         flush_signals(current); /* just in case */
8950                         wait_event_interruptible(mddev->recovery_wait,
8951                                                  mddev->resync_max > j
8952                                                  || test_bit(MD_RECOVERY_INTR,
8953                                                              &mddev->recovery));
8954                 }
8955
8956                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8957                         break;
8958
8959                 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8960                 if (sectors == 0) {
8961                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8962                         break;
8963                 }
8964
8965                 if (!skipped) { /* actual IO requested */
8966                         io_sectors += sectors;
8967                         atomic_add(sectors, &mddev->recovery_active);
8968                 }
8969
8970                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8971                         break;
8972
8973                 j += sectors;
8974                 if (j > max_sectors)
8975                         /* when skipping, extra large numbers can be returned. */
8976                         j = max_sectors;
8977                 if (j >= MD_RESYNC_ACTIVE)
8978                         mddev->curr_resync = j;
8979                 mddev->curr_mark_cnt = io_sectors;
8980                 if (last_check == 0)
8981                         /* this is the earliest that rebuild will be
8982                          * visible in /proc/mdstat
8983                          */
8984                         md_new_event();
8985
8986                 if (last_check + window > io_sectors || j == max_sectors)
8987                         continue;
8988
8989                 last_check = io_sectors;
8990         repeat:
8991                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8992                         /* step marks */
8993                         int next = (last_mark+1) % SYNC_MARKS;
8994
8995                         mddev->resync_mark = mark[next];
8996                         mddev->resync_mark_cnt = mark_cnt[next];
8997                         mark[next] = jiffies;
8998                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8999                         last_mark = next;
9000                 }
9001
9002                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9003                         break;
9004
9005                 /*
9006                  * this loop exits only if either when we are slower than
9007                  * the 'hard' speed limit, or the system was IO-idle for
9008                  * a jiffy.
9009                  * the system might be non-idle CPU-wise, but we only care
9010                  * about not overloading the IO subsystem. (things like an
9011                  * e2fsck being done on the RAID array should execute fast)
9012                  */
9013                 cond_resched();
9014
9015                 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9016                 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9017                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
9018
9019                 if (currspeed > speed_min(mddev)) {
9020                         if (currspeed > speed_max(mddev)) {
9021                                 msleep(500);
9022                                 goto repeat;
9023                         }
9024                         if (!is_mddev_idle(mddev, 0)) {
9025                                 /*
9026                                  * Give other IO more of a chance.
9027                                  * The faster the devices, the less we wait.
9028                                  */
9029                                 wait_event(mddev->recovery_wait,
9030                                            !atomic_read(&mddev->recovery_active));
9031                         }
9032                 }
9033         }
9034         pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9035                 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9036                 ? "interrupted" : "done");
9037         /*
9038          * this also signals 'finished resyncing' to md_stop
9039          */
9040         blk_finish_plug(&plug);
9041         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9042
9043         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9044             !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9045             mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9046                 mddev->curr_resync_completed = mddev->curr_resync;
9047                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9048         }
9049         mddev->pers->sync_request(mddev, max_sectors, &skipped);
9050
9051         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9052             mddev->curr_resync > MD_RESYNC_ACTIVE) {
9053                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9054                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9055                                 if (mddev->curr_resync >= mddev->recovery_cp) {
9056                                         pr_debug("md: checkpointing %s of %s.\n",
9057                                                  desc, mdname(mddev));
9058                                         if (test_bit(MD_RECOVERY_ERROR,
9059                                                 &mddev->recovery))
9060                                                 mddev->recovery_cp =
9061                                                         mddev->curr_resync_completed;
9062                                         else
9063                                                 mddev->recovery_cp =
9064                                                         mddev->curr_resync;
9065                                 }
9066                         } else
9067                                 mddev->recovery_cp = MaxSector;
9068                 } else {
9069                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9070                                 mddev->curr_resync = MaxSector;
9071                         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9072                             test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9073                                 rcu_read_lock();
9074                                 rdev_for_each_rcu(rdev, mddev)
9075                                         if (rdev->raid_disk >= 0 &&
9076                                             mddev->delta_disks >= 0 &&
9077                                             !test_bit(Journal, &rdev->flags) &&
9078                                             !test_bit(Faulty, &rdev->flags) &&
9079                                             !test_bit(In_sync, &rdev->flags) &&
9080                                             rdev->recovery_offset < mddev->curr_resync)
9081                                                 rdev->recovery_offset = mddev->curr_resync;
9082                                 rcu_read_unlock();
9083                         }
9084                 }
9085         }
9086  skip:
9087         /* set CHANGE_PENDING here since maybe another update is needed,
9088          * so other nodes are informed. It should be harmless for normal
9089          * raid */
9090         set_mask_bits(&mddev->sb_flags, 0,
9091                       BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9092
9093         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9094                         !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9095                         mddev->delta_disks > 0 &&
9096                         mddev->pers->finish_reshape &&
9097                         mddev->pers->size &&
9098                         mddev->queue) {
9099                 mddev_lock_nointr(mddev);
9100                 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9101                 mddev_unlock(mddev);
9102                 if (!mddev_is_clustered(mddev))
9103                         set_capacity_and_notify(mddev->gendisk,
9104                                                 mddev->array_sectors);
9105         }
9106
9107         spin_lock(&mddev->lock);
9108         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9109                 /* We completed so min/max setting can be forgotten if used. */
9110                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9111                         mddev->resync_min = 0;
9112                 mddev->resync_max = MaxSector;
9113         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9114                 mddev->resync_min = mddev->curr_resync_completed;
9115         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9116         mddev->curr_resync = MD_RESYNC_NONE;
9117         spin_unlock(&mddev->lock);
9118
9119         wake_up(&resync_wait);
9120         md_wakeup_thread(mddev->thread);
9121         return;
9122 }
9123 EXPORT_SYMBOL_GPL(md_do_sync);
9124
9125 static int remove_and_add_spares(struct mddev *mddev,
9126                                  struct md_rdev *this)
9127 {
9128         struct md_rdev *rdev;
9129         int spares = 0;
9130         int removed = 0;
9131         bool remove_some = false;
9132
9133         if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9134                 /* Mustn't remove devices when resync thread is running */
9135                 return 0;
9136
9137         rdev_for_each(rdev, mddev) {
9138                 if ((this == NULL || rdev == this) &&
9139                     rdev->raid_disk >= 0 &&
9140                     !test_bit(Blocked, &rdev->flags) &&
9141                     test_bit(Faulty, &rdev->flags) &&
9142                     atomic_read(&rdev->nr_pending)==0) {
9143                         /* Faulty non-Blocked devices with nr_pending == 0
9144                          * never get nr_pending incremented,
9145                          * never get Faulty cleared, and never get Blocked set.
9146                          * So we can synchronize_rcu now rather than once per device
9147                          */
9148                         remove_some = true;
9149                         set_bit(RemoveSynchronized, &rdev->flags);
9150                 }
9151         }
9152
9153         if (remove_some)
9154                 synchronize_rcu();
9155         rdev_for_each(rdev, mddev) {
9156                 if ((this == NULL || rdev == this) &&
9157                     rdev->raid_disk >= 0 &&
9158                     !test_bit(Blocked, &rdev->flags) &&
9159                     ((test_bit(RemoveSynchronized, &rdev->flags) ||
9160                      (!test_bit(In_sync, &rdev->flags) &&
9161                       !test_bit(Journal, &rdev->flags))) &&
9162                     atomic_read(&rdev->nr_pending)==0)) {
9163                         if (mddev->pers->hot_remove_disk(
9164                                     mddev, rdev) == 0) {
9165                                 sysfs_unlink_rdev(mddev, rdev);
9166                                 rdev->saved_raid_disk = rdev->raid_disk;
9167                                 rdev->raid_disk = -1;
9168                                 removed++;
9169                         }
9170                 }
9171                 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9172                         clear_bit(RemoveSynchronized, &rdev->flags);
9173         }
9174
9175         if (removed && mddev->kobj.sd)
9176                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9177
9178         if (this && removed)
9179                 goto no_add;
9180
9181         rdev_for_each(rdev, mddev) {
9182                 if (this && this != rdev)
9183                         continue;
9184                 if (test_bit(Candidate, &rdev->flags))
9185                         continue;
9186                 if (rdev->raid_disk >= 0 &&
9187                     !test_bit(In_sync, &rdev->flags) &&
9188                     !test_bit(Journal, &rdev->flags) &&
9189                     !test_bit(Faulty, &rdev->flags))
9190                         spares++;
9191                 if (rdev->raid_disk >= 0)
9192                         continue;
9193                 if (test_bit(Faulty, &rdev->flags))
9194                         continue;
9195                 if (!test_bit(Journal, &rdev->flags)) {
9196                         if (!md_is_rdwr(mddev) &&
9197                             !(rdev->saved_raid_disk >= 0 &&
9198                               !test_bit(Bitmap_sync, &rdev->flags)))
9199                                 continue;
9200
9201                         rdev->recovery_offset = 0;
9202                 }
9203                 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9204                         /* failure here is OK */
9205                         sysfs_link_rdev(mddev, rdev);
9206                         if (!test_bit(Journal, &rdev->flags))
9207                                 spares++;
9208                         md_new_event();
9209                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9210                 }
9211         }
9212 no_add:
9213         if (removed)
9214                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9215         return spares;
9216 }
9217
9218 static void md_start_sync(struct work_struct *ws)
9219 {
9220         struct mddev *mddev = container_of(ws, struct mddev, del_work);
9221
9222         mddev->sync_thread = md_register_thread(md_do_sync,
9223                                                 mddev,
9224                                                 "resync");
9225         if (!mddev->sync_thread) {
9226                 pr_warn("%s: could not start resync thread...\n",
9227                         mdname(mddev));
9228                 /* leave the spares where they are, it shouldn't hurt */
9229                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9230                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9231                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9232                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9233                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9234                 wake_up(&resync_wait);
9235                 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9236                                        &mddev->recovery))
9237                         if (mddev->sysfs_action)
9238                                 sysfs_notify_dirent_safe(mddev->sysfs_action);
9239         } else
9240                 md_wakeup_thread(mddev->sync_thread);
9241         sysfs_notify_dirent_safe(mddev->sysfs_action);
9242         md_new_event();
9243 }
9244
9245 /*
9246  * This routine is regularly called by all per-raid-array threads to
9247  * deal with generic issues like resync and super-block update.
9248  * Raid personalities that don't have a thread (linear/raid0) do not
9249  * need this as they never do any recovery or update the superblock.
9250  *
9251  * It does not do any resync itself, but rather "forks" off other threads
9252  * to do that as needed.
9253  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9254  * "->recovery" and create a thread at ->sync_thread.
9255  * When the thread finishes it sets MD_RECOVERY_DONE
9256  * and wakeups up this thread which will reap the thread and finish up.
9257  * This thread also removes any faulty devices (with nr_pending == 0).
9258  *
9259  * The overall approach is:
9260  *  1/ if the superblock needs updating, update it.
9261  *  2/ If a recovery thread is running, don't do anything else.
9262  *  3/ If recovery has finished, clean up, possibly marking spares active.
9263  *  4/ If there are any faulty devices, remove them.
9264  *  5/ If array is degraded, try to add spares devices
9265  *  6/ If array has spares or is not in-sync, start a resync thread.
9266  */
9267 void md_check_recovery(struct mddev *mddev)
9268 {
9269         if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9270                 /* Write superblock - thread that called mddev_suspend()
9271                  * holds reconfig_mutex for us.
9272                  */
9273                 set_bit(MD_UPDATING_SB, &mddev->flags);
9274                 smp_mb__after_atomic();
9275                 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9276                         md_update_sb(mddev, 0);
9277                 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9278                 wake_up(&mddev->sb_wait);
9279         }
9280
9281         if (is_md_suspended(mddev))
9282                 return;
9283
9284         if (mddev->bitmap)
9285                 md_bitmap_daemon_work(mddev);
9286
9287         if (signal_pending(current)) {
9288                 if (mddev->pers->sync_request && !mddev->external) {
9289                         pr_debug("md: %s in immediate safe mode\n",
9290                                  mdname(mddev));
9291                         mddev->safemode = 2;
9292                 }
9293                 flush_signals(current);
9294         }
9295
9296         if (!md_is_rdwr(mddev) &&
9297             !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9298                 return;
9299         if ( ! (
9300                 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9301                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9302                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9303                 (mddev->external == 0 && mddev->safemode == 1) ||
9304                 (mddev->safemode == 2
9305                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9306                 ))
9307                 return;
9308
9309         if (mddev_trylock(mddev)) {
9310                 int spares = 0;
9311                 bool try_set_sync = mddev->safemode != 0;
9312
9313                 if (!mddev->external && mddev->safemode == 1)
9314                         mddev->safemode = 0;
9315
9316                 if (!md_is_rdwr(mddev)) {
9317                         struct md_rdev *rdev;
9318                         if (!mddev->external && mddev->in_sync)
9319                                 /* 'Blocked' flag not needed as failed devices
9320                                  * will be recorded if array switched to read/write.
9321                                  * Leaving it set will prevent the device
9322                                  * from being removed.
9323                                  */
9324                                 rdev_for_each(rdev, mddev)
9325                                         clear_bit(Blocked, &rdev->flags);
9326                         /* On a read-only array we can:
9327                          * - remove failed devices
9328                          * - add already-in_sync devices if the array itself
9329                          *   is in-sync.
9330                          * As we only add devices that are already in-sync,
9331                          * we can activate the spares immediately.
9332                          */
9333                         remove_and_add_spares(mddev, NULL);
9334                         /* There is no thread, but we need to call
9335                          * ->spare_active and clear saved_raid_disk
9336                          */
9337                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9338                         md_unregister_thread(&mddev->sync_thread);
9339                         md_reap_sync_thread(mddev);
9340                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9341                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9342                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9343                         goto unlock;
9344                 }
9345
9346                 if (mddev_is_clustered(mddev)) {
9347                         struct md_rdev *rdev, *tmp;
9348                         /* kick the device if another node issued a
9349                          * remove disk.
9350                          */
9351                         rdev_for_each_safe(rdev, tmp, mddev) {
9352                                 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9353                                                 rdev->raid_disk < 0)
9354                                         md_kick_rdev_from_array(rdev);
9355                         }
9356                 }
9357
9358                 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9359                         spin_lock(&mddev->lock);
9360                         set_in_sync(mddev);
9361                         spin_unlock(&mddev->lock);
9362                 }
9363
9364                 if (mddev->sb_flags)
9365                         md_update_sb(mddev, 0);
9366
9367                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9368                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9369                         /* resync/recovery still happening */
9370                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9371                         goto unlock;
9372                 }
9373                 if (mddev->sync_thread) {
9374                         md_unregister_thread(&mddev->sync_thread);
9375                         md_reap_sync_thread(mddev);
9376                         goto unlock;
9377                 }
9378                 /* Set RUNNING before clearing NEEDED to avoid
9379                  * any transients in the value of "sync_action".
9380                  */
9381                 mddev->curr_resync_completed = 0;
9382                 spin_lock(&mddev->lock);
9383                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9384                 spin_unlock(&mddev->lock);
9385                 /* Clear some bits that don't mean anything, but
9386                  * might be left set
9387                  */
9388                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9389                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9390
9391                 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9392                     test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9393                         goto not_running;
9394                 /* no recovery is running.
9395                  * remove any failed drives, then
9396                  * add spares if possible.
9397                  * Spares are also removed and re-added, to allow
9398                  * the personality to fail the re-add.
9399                  */
9400
9401                 if (mddev->reshape_position != MaxSector) {
9402                         if (mddev->pers->check_reshape == NULL ||
9403                             mddev->pers->check_reshape(mddev) != 0)
9404                                 /* Cannot proceed */
9405                                 goto not_running;
9406                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9407                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9408                 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9409                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9410                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9411                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9412                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9413                 } else if (mddev->recovery_cp < MaxSector) {
9414                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9415                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9416                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9417                         /* nothing to be done ... */
9418                         goto not_running;
9419
9420                 if (mddev->pers->sync_request) {
9421                         if (spares) {
9422                                 /* We are adding a device or devices to an array
9423                                  * which has the bitmap stored on all devices.
9424                                  * So make sure all bitmap pages get written
9425                                  */
9426                                 md_bitmap_write_all(mddev->bitmap);
9427                         }
9428                         INIT_WORK(&mddev->del_work, md_start_sync);
9429                         queue_work(md_misc_wq, &mddev->del_work);
9430                         goto unlock;
9431                 }
9432         not_running:
9433                 if (!mddev->sync_thread) {
9434                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9435                         wake_up(&resync_wait);
9436                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9437                                                &mddev->recovery))
9438                                 if (mddev->sysfs_action)
9439                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
9440                 }
9441         unlock:
9442                 wake_up(&mddev->sb_wait);
9443                 mddev_unlock(mddev);
9444         }
9445 }
9446 EXPORT_SYMBOL(md_check_recovery);
9447
9448 void md_reap_sync_thread(struct mddev *mddev)
9449 {
9450         struct md_rdev *rdev;
9451         sector_t old_dev_sectors = mddev->dev_sectors;
9452         bool is_reshaped = false;
9453
9454         /* sync_thread should be unregistered, collect result */
9455         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9456             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9457             mddev->degraded != mddev->raid_disks) {
9458                 /* success...*/
9459                 /* activate any spares */
9460                 if (mddev->pers->spare_active(mddev)) {
9461                         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9462                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9463                 }
9464         }
9465         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9466             mddev->pers->finish_reshape) {
9467                 mddev->pers->finish_reshape(mddev);
9468                 if (mddev_is_clustered(mddev))
9469                         is_reshaped = true;
9470         }
9471
9472         /* If array is no-longer degraded, then any saved_raid_disk
9473          * information must be scrapped.
9474          */
9475         if (!mddev->degraded)
9476                 rdev_for_each(rdev, mddev)
9477                         rdev->saved_raid_disk = -1;
9478
9479         md_update_sb(mddev, 1);
9480         /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9481          * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9482          * clustered raid */
9483         if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9484                 md_cluster_ops->resync_finish(mddev);
9485         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9486         clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9487         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9488         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9489         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9490         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9491         /*
9492          * We call md_cluster_ops->update_size here because sync_size could
9493          * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9494          * so it is time to update size across cluster.
9495          */
9496         if (mddev_is_clustered(mddev) && is_reshaped
9497                                       && !test_bit(MD_CLOSING, &mddev->flags))
9498                 md_cluster_ops->update_size(mddev, old_dev_sectors);
9499         wake_up(&resync_wait);
9500         /* flag recovery needed just to double check */
9501         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9502         sysfs_notify_dirent_safe(mddev->sysfs_completed);
9503         sysfs_notify_dirent_safe(mddev->sysfs_action);
9504         md_new_event();
9505         if (mddev->event_work.func)
9506                 queue_work(md_misc_wq, &mddev->event_work);
9507 }
9508 EXPORT_SYMBOL(md_reap_sync_thread);
9509
9510 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9511 {
9512         sysfs_notify_dirent_safe(rdev->sysfs_state);
9513         wait_event_timeout(rdev->blocked_wait,
9514                            !test_bit(Blocked, &rdev->flags) &&
9515                            !test_bit(BlockedBadBlocks, &rdev->flags),
9516                            msecs_to_jiffies(5000));
9517         rdev_dec_pending(rdev, mddev);
9518 }
9519 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9520
9521 void md_finish_reshape(struct mddev *mddev)
9522 {
9523         /* called be personality module when reshape completes. */
9524         struct md_rdev *rdev;
9525
9526         rdev_for_each(rdev, mddev) {
9527                 if (rdev->data_offset > rdev->new_data_offset)
9528                         rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9529                 else
9530                         rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9531                 rdev->data_offset = rdev->new_data_offset;
9532         }
9533 }
9534 EXPORT_SYMBOL(md_finish_reshape);
9535
9536 /* Bad block management */
9537
9538 /* Returns 1 on success, 0 on failure */
9539 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9540                        int is_new)
9541 {
9542         struct mddev *mddev = rdev->mddev;
9543         int rv;
9544         if (is_new)
9545                 s += rdev->new_data_offset;
9546         else
9547                 s += rdev->data_offset;
9548         rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9549         if (rv == 0) {
9550                 /* Make sure they get written out promptly */
9551                 if (test_bit(ExternalBbl, &rdev->flags))
9552                         sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9553                 sysfs_notify_dirent_safe(rdev->sysfs_state);
9554                 set_mask_bits(&mddev->sb_flags, 0,
9555                               BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9556                 md_wakeup_thread(rdev->mddev->thread);
9557                 return 1;
9558         } else
9559                 return 0;
9560 }
9561 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9562
9563 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9564                          int is_new)
9565 {
9566         int rv;
9567         if (is_new)
9568                 s += rdev->new_data_offset;
9569         else
9570                 s += rdev->data_offset;
9571         rv = badblocks_clear(&rdev->badblocks, s, sectors);
9572         if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9573                 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9574         return rv;
9575 }
9576 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9577
9578 static int md_notify_reboot(struct notifier_block *this,
9579                             unsigned long code, void *x)
9580 {
9581         struct mddev *mddev, *n;
9582         int need_delay = 0;
9583
9584         spin_lock(&all_mddevs_lock);
9585         list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9586                 if (!mddev_get(mddev))
9587                         continue;
9588                 spin_unlock(&all_mddevs_lock);
9589                 if (mddev_trylock(mddev)) {
9590                         if (mddev->pers)
9591                                 __md_stop_writes(mddev);
9592                         if (mddev->persistent)
9593                                 mddev->safemode = 2;
9594                         mddev_unlock(mddev);
9595                 }
9596                 need_delay = 1;
9597                 mddev_put(mddev);
9598                 spin_lock(&all_mddevs_lock);
9599         }
9600         spin_unlock(&all_mddevs_lock);
9601
9602         /*
9603          * certain more exotic SCSI devices are known to be
9604          * volatile wrt too early system reboots. While the
9605          * right place to handle this issue is the given
9606          * driver, we do want to have a safe RAID driver ...
9607          */
9608         if (need_delay)
9609                 msleep(1000);
9610
9611         return NOTIFY_DONE;
9612 }
9613
9614 static struct notifier_block md_notifier = {
9615         .notifier_call  = md_notify_reboot,
9616         .next           = NULL,
9617         .priority       = INT_MAX, /* before any real devices */
9618 };
9619
9620 static void md_geninit(void)
9621 {
9622         pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9623
9624         proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9625 }
9626
9627 static int __init md_init(void)
9628 {
9629         int ret = -ENOMEM;
9630
9631         md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9632         if (!md_wq)
9633                 goto err_wq;
9634
9635         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9636         if (!md_misc_wq)
9637                 goto err_misc_wq;
9638
9639         md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9640         if (!md_rdev_misc_wq)
9641                 goto err_rdev_misc_wq;
9642
9643         ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9644         if (ret < 0)
9645                 goto err_md;
9646
9647         ret = __register_blkdev(0, "mdp", md_probe);
9648         if (ret < 0)
9649                 goto err_mdp;
9650         mdp_major = ret;
9651
9652         register_reboot_notifier(&md_notifier);
9653         raid_table_header = register_sysctl_table(raid_root_table);
9654
9655         md_geninit();
9656         return 0;
9657
9658 err_mdp:
9659         unregister_blkdev(MD_MAJOR, "md");
9660 err_md:
9661         destroy_workqueue(md_rdev_misc_wq);
9662 err_rdev_misc_wq:
9663         destroy_workqueue(md_misc_wq);
9664 err_misc_wq:
9665         destroy_workqueue(md_wq);
9666 err_wq:
9667         return ret;
9668 }
9669
9670 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9671 {
9672         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9673         struct md_rdev *rdev2, *tmp;
9674         int role, ret;
9675
9676         /*
9677          * If size is changed in another node then we need to
9678          * do resize as well.
9679          */
9680         if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9681                 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9682                 if (ret)
9683                         pr_info("md-cluster: resize failed\n");
9684                 else
9685                         md_bitmap_update_sb(mddev->bitmap);
9686         }
9687
9688         /* Check for change of roles in the active devices */
9689         rdev_for_each_safe(rdev2, tmp, mddev) {
9690                 if (test_bit(Faulty, &rdev2->flags))
9691                         continue;
9692
9693                 /* Check if the roles changed */
9694                 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9695
9696                 if (test_bit(Candidate, &rdev2->flags)) {
9697                         if (role == MD_DISK_ROLE_FAULTY) {
9698                                 pr_info("md: Removing Candidate device %pg because add failed\n",
9699                                         rdev2->bdev);
9700                                 md_kick_rdev_from_array(rdev2);
9701                                 continue;
9702                         }
9703                         else
9704                                 clear_bit(Candidate, &rdev2->flags);
9705                 }
9706
9707                 if (role != rdev2->raid_disk) {
9708                         /*
9709                          * got activated except reshape is happening.
9710                          */
9711                         if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9712                             !(le32_to_cpu(sb->feature_map) &
9713                               MD_FEATURE_RESHAPE_ACTIVE)) {
9714                                 rdev2->saved_raid_disk = role;
9715                                 ret = remove_and_add_spares(mddev, rdev2);
9716                                 pr_info("Activated spare: %pg\n",
9717                                         rdev2->bdev);
9718                                 /* wakeup mddev->thread here, so array could
9719                                  * perform resync with the new activated disk */
9720                                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9721                                 md_wakeup_thread(mddev->thread);
9722                         }
9723                         /* device faulty
9724                          * We just want to do the minimum to mark the disk
9725                          * as faulty. The recovery is performed by the
9726                          * one who initiated the error.
9727                          */
9728                         if (role == MD_DISK_ROLE_FAULTY ||
9729                             role == MD_DISK_ROLE_JOURNAL) {
9730                                 md_error(mddev, rdev2);
9731                                 clear_bit(Blocked, &rdev2->flags);
9732                         }
9733                 }
9734         }
9735
9736         if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9737                 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9738                 if (ret)
9739                         pr_warn("md: updating array disks failed. %d\n", ret);
9740         }
9741
9742         /*
9743          * Since mddev->delta_disks has already updated in update_raid_disks,
9744          * so it is time to check reshape.
9745          */
9746         if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9747             (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9748                 /*
9749                  * reshape is happening in the remote node, we need to
9750                  * update reshape_position and call start_reshape.
9751                  */
9752                 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9753                 if (mddev->pers->update_reshape_pos)
9754                         mddev->pers->update_reshape_pos(mddev);
9755                 if (mddev->pers->start_reshape)
9756                         mddev->pers->start_reshape(mddev);
9757         } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9758                    mddev->reshape_position != MaxSector &&
9759                    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9760                 /* reshape is just done in another node. */
9761                 mddev->reshape_position = MaxSector;
9762                 if (mddev->pers->update_reshape_pos)
9763                         mddev->pers->update_reshape_pos(mddev);
9764         }
9765
9766         /* Finally set the event to be up to date */
9767         mddev->events = le64_to_cpu(sb->events);
9768 }
9769
9770 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9771 {
9772         int err;
9773         struct page *swapout = rdev->sb_page;
9774         struct mdp_superblock_1 *sb;
9775
9776         /* Store the sb page of the rdev in the swapout temporary
9777          * variable in case we err in the future
9778          */
9779         rdev->sb_page = NULL;
9780         err = alloc_disk_sb(rdev);
9781         if (err == 0) {
9782                 ClearPageUptodate(rdev->sb_page);
9783                 rdev->sb_loaded = 0;
9784                 err = super_types[mddev->major_version].
9785                         load_super(rdev, NULL, mddev->minor_version);
9786         }
9787         if (err < 0) {
9788                 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9789                                 __func__, __LINE__, rdev->desc_nr, err);
9790                 if (rdev->sb_page)
9791                         put_page(rdev->sb_page);
9792                 rdev->sb_page = swapout;
9793                 rdev->sb_loaded = 1;
9794                 return err;
9795         }
9796
9797         sb = page_address(rdev->sb_page);
9798         /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9799          * is not set
9800          */
9801
9802         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9803                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9804
9805         /* The other node finished recovery, call spare_active to set
9806          * device In_sync and mddev->degraded
9807          */
9808         if (rdev->recovery_offset == MaxSector &&
9809             !test_bit(In_sync, &rdev->flags) &&
9810             mddev->pers->spare_active(mddev))
9811                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9812
9813         put_page(swapout);
9814         return 0;
9815 }
9816
9817 void md_reload_sb(struct mddev *mddev, int nr)
9818 {
9819         struct md_rdev *rdev = NULL, *iter;
9820         int err;
9821
9822         /* Find the rdev */
9823         rdev_for_each_rcu(iter, mddev) {
9824                 if (iter->desc_nr == nr) {
9825                         rdev = iter;
9826                         break;
9827                 }
9828         }
9829
9830         if (!rdev) {
9831                 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9832                 return;
9833         }
9834
9835         err = read_rdev(mddev, rdev);
9836         if (err < 0)
9837                 return;
9838
9839         check_sb_changes(mddev, rdev);
9840
9841         /* Read all rdev's to update recovery_offset */
9842         rdev_for_each_rcu(rdev, mddev) {
9843                 if (!test_bit(Faulty, &rdev->flags))
9844                         read_rdev(mddev, rdev);
9845         }
9846 }
9847 EXPORT_SYMBOL(md_reload_sb);
9848
9849 #ifndef MODULE
9850
9851 /*
9852  * Searches all registered partitions for autorun RAID arrays
9853  * at boot time.
9854  */
9855
9856 static DEFINE_MUTEX(detected_devices_mutex);
9857 static LIST_HEAD(all_detected_devices);
9858 struct detected_devices_node {
9859         struct list_head list;
9860         dev_t dev;
9861 };
9862
9863 void md_autodetect_dev(dev_t dev)
9864 {
9865         struct detected_devices_node *node_detected_dev;
9866
9867         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9868         if (node_detected_dev) {
9869                 node_detected_dev->dev = dev;
9870                 mutex_lock(&detected_devices_mutex);
9871                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9872                 mutex_unlock(&detected_devices_mutex);
9873         }
9874 }
9875
9876 void md_autostart_arrays(int part)
9877 {
9878         struct md_rdev *rdev;
9879         struct detected_devices_node *node_detected_dev;
9880         dev_t dev;
9881         int i_scanned, i_passed;
9882
9883         i_scanned = 0;
9884         i_passed = 0;
9885
9886         pr_info("md: Autodetecting RAID arrays.\n");
9887
9888         mutex_lock(&detected_devices_mutex);
9889         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9890                 i_scanned++;
9891                 node_detected_dev = list_entry(all_detected_devices.next,
9892                                         struct detected_devices_node, list);
9893                 list_del(&node_detected_dev->list);
9894                 dev = node_detected_dev->dev;
9895                 kfree(node_detected_dev);
9896                 mutex_unlock(&detected_devices_mutex);
9897                 rdev = md_import_device(dev,0, 90);
9898                 mutex_lock(&detected_devices_mutex);
9899                 if (IS_ERR(rdev))
9900                         continue;
9901
9902                 if (test_bit(Faulty, &rdev->flags))
9903                         continue;
9904
9905                 set_bit(AutoDetected, &rdev->flags);
9906                 list_add(&rdev->same_set, &pending_raid_disks);
9907                 i_passed++;
9908         }
9909         mutex_unlock(&detected_devices_mutex);
9910
9911         pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9912
9913         autorun_devices(part);
9914 }
9915
9916 #endif /* !MODULE */
9917
9918 static __exit void md_exit(void)
9919 {
9920         struct mddev *mddev, *n;
9921         int delay = 1;
9922
9923         unregister_blkdev(MD_MAJOR,"md");
9924         unregister_blkdev(mdp_major, "mdp");
9925         unregister_reboot_notifier(&md_notifier);
9926         unregister_sysctl_table(raid_table_header);
9927
9928         /* We cannot unload the modules while some process is
9929          * waiting for us in select() or poll() - wake them up
9930          */
9931         md_unloading = 1;
9932         while (waitqueue_active(&md_event_waiters)) {
9933                 /* not safe to leave yet */
9934                 wake_up(&md_event_waiters);
9935                 msleep(delay);
9936                 delay += delay;
9937         }
9938         remove_proc_entry("mdstat", NULL);
9939
9940         spin_lock(&all_mddevs_lock);
9941         list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9942                 if (!mddev_get(mddev))
9943                         continue;
9944                 spin_unlock(&all_mddevs_lock);
9945                 export_array(mddev);
9946                 mddev->ctime = 0;
9947                 mddev->hold_active = 0;
9948                 /*
9949                  * As the mddev is now fully clear, mddev_put will schedule
9950                  * the mddev for destruction by a workqueue, and the
9951                  * destroy_workqueue() below will wait for that to complete.
9952                  */
9953                 mddev_put(mddev);
9954                 spin_lock(&all_mddevs_lock);
9955         }
9956         spin_unlock(&all_mddevs_lock);
9957
9958         destroy_workqueue(md_rdev_misc_wq);
9959         destroy_workqueue(md_misc_wq);
9960         destroy_workqueue(md_wq);
9961 }
9962
9963 subsys_initcall(md_init);
9964 module_exit(md_exit)
9965
9966 static int get_ro(char *buffer, const struct kernel_param *kp)
9967 {
9968         return sprintf(buffer, "%d\n", start_readonly);
9969 }
9970 static int set_ro(const char *val, const struct kernel_param *kp)
9971 {
9972         return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9973 }
9974
9975 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9976 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9977 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9978 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9979
9980 MODULE_LICENSE("GPL");
9981 MODULE_DESCRIPTION("MD RAID framework");
9982 MODULE_ALIAS("md");
9983 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);