drivers/md/md.c

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