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