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_locked(dev_t unit)
 738 {
 739         struct mddev *mddev;
 740
 741         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 742                 if (mddev->unit == unit)
 743                         return mddev;
 744
 745         return NULL;
 746 }
 747
 748 /* find an unused unit number */
 749 static dev_t mddev_alloc_unit(void)
 750 {
 751         static int next_minor = 512;
 752         int start = next_minor;
 753         bool is_free = 0;
 754         dev_t dev = 0;
 755
 756         while (!is_free) {
 757                 dev = MKDEV(MD_MAJOR, next_minor);
 758                 next_minor++;
 759                 if (next_minor > MINORMASK)
 760                         next_minor = 0;
 761                 if (next_minor == start)
 762                         return 0;               /* Oh dear, all in use. */
 763                 is_free = !mddev_find_locked(dev);
 764         }
 765
 766         return dev;
 767 }
 768
 769 static struct mddev *mddev_find(dev_t unit)
 770 {
 771         struct mddev *mddev;
 772
 773         if (MAJOR(unit) != MD_MAJOR)
 774                 unit &= ~((1 << MdpMinorShift) - 1);
 775
 776         spin_lock(&all_mddevs_lock);
 777         mddev = mddev_find_locked(unit);
 778         if (mddev)
 779                 mddev_get(mddev);
 780         spin_unlock(&all_mddevs_lock);
 781
 782         return mddev;
 783 }
 784
 785 static struct mddev *mddev_alloc(dev_t unit)
 786 {
 787         struct mddev *new;
 788         int error;
 789
 790         if (unit && MAJOR(unit) != MD_MAJOR)
 791                 unit &= ~((1 << MdpMinorShift) - 1);
 792
 793         new = kzalloc(sizeof(*new), GFP_KERNEL);
 794         if (!new)
 795                 return ERR_PTR(-ENOMEM);
 796         mddev_init(new);
 797
 798         spin_lock(&all_mddevs_lock);
 799         if (unit) {
 800                 error = -EEXIST;
 801                 if (mddev_find_locked(unit))
 802                         goto out_free_new;
 803                 new->unit = unit;
 804                 if (MAJOR(unit) == MD_MAJOR)
 805                         new->md_minor = MINOR(unit);
 806                 else
 807                         new->md_minor = MINOR(unit) >> MdpMinorShift;
 808                 new->hold_active = UNTIL_IOCTL;
 809         } else {
 810                 error = -ENODEV;
 811                 new->unit = mddev_alloc_unit();
 812                 if (!new->unit)
 813                         goto out_free_new;
 814                 new->md_minor = MINOR(new->unit);
 815                 new->hold_active = UNTIL_STOP;
 816         }
 817
 818         list_add(&new->all_mddevs, &all_mddevs);
 819         spin_unlock(&all_mddevs_lock);
 820         return new;
 821 out_free_new:
 822         spin_unlock(&all_mddevs_lock);
 823         kfree(new);
 824         return ERR_PTR(error);
 825 }
 826
 827 static struct attribute_group md_redundancy_group;
 828
 829 void mddev_unlock(struct mddev *mddev)
 830 {
 831         if (mddev->to_remove) {
 832                 /* These cannot be removed under reconfig_mutex as
 833                  * an access to the files will try to take reconfig_mutex
 834                  * while holding the file unremovable, which leads to
 835                  * a deadlock.
 836                  * So hold set sysfs_active while the remove in happeing,
 837                  * and anything else which might set ->to_remove or my
 838                  * otherwise change the sysfs namespace will fail with
 839                  * -EBUSY if sysfs_active is still set.
 840                  * We set sysfs_active under reconfig_mutex and elsewhere
 841                  * test it under the same mutex to ensure its correct value
 842                  * is seen.
 843                  */
 844                 struct attribute_group *to_remove = mddev->to_remove;
 845                 mddev->to_remove = NULL;
 846                 mddev->sysfs_active = 1;
 847                 mutex_unlock(&mddev->reconfig_mutex);
 848
 849                 if (mddev->kobj.sd) {
 850                         if (to_remove != &md_redundancy_group)
 851                                 sysfs_remove_group(&mddev->kobj, to_remove);
 852                         if (mddev->pers == NULL ||
 853                             mddev->pers->sync_request == NULL) {
 854                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 855                                 if (mddev->sysfs_action)
 856                                         sysfs_put(mddev->sysfs_action);
 857                                 if (mddev->sysfs_completed)
 858                                         sysfs_put(mddev->sysfs_completed);
 859                                 if (mddev->sysfs_degraded)
 860                                         sysfs_put(mddev->sysfs_degraded);
 861                                 mddev->sysfs_action = NULL;
 862                                 mddev->sysfs_completed = NULL;
 863                                 mddev->sysfs_degraded = NULL;
 864                         }
 865                 }
 866                 mddev->sysfs_active = 0;
 867         } else
 868                 mutex_unlock(&mddev->reconfig_mutex);
 869
 870         /* As we've dropped the mutex we need a spinlock to
 871          * make sure the thread doesn't disappear
 872          */
 873         spin_lock(&pers_lock);
 874         md_wakeup_thread(mddev->thread);
 875         wake_up(&mddev->sb_wait);
 876         spin_unlock(&pers_lock);
 877 }
 878 EXPORT_SYMBOL_GPL(mddev_unlock);
 879
 880 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
 881 {
 882         struct md_rdev *rdev;
 883
 884         rdev_for_each_rcu(rdev, mddev)
 885                 if (rdev->desc_nr == nr)
 886                         return rdev;
 887
 888         return NULL;
 889 }
 890 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
 891
 892 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 893 {
 894         struct md_rdev *rdev;
 895
 896         rdev_for_each(rdev, mddev)
 897                 if (rdev->bdev->bd_dev == dev)
 898                         return rdev;
 899
 900         return NULL;
 901 }
 902
 903 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
 904 {
 905         struct md_rdev *rdev;
 906
 907         rdev_for_each_rcu(rdev, mddev)
 908                 if (rdev->bdev->bd_dev == dev)
 909                         return rdev;
 910
 911         return NULL;
 912 }
 913 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
 914
 915 static struct md_personality *find_pers(int level, char *clevel)
 916 {
 917         struct md_personality *pers;
 918         list_for_each_entry(pers, &pers_list, list) {
 919                 if (level != LEVEL_NONE && pers->level == level)
 920                         return pers;
 921                 if (strcmp(pers->name, clevel)==0)
 922                         return pers;
 923         }
 924         return NULL;
 925 }
 926
 927 /* return the offset of the super block in 512byte sectors */
 928 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 929 {
 930         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 931         return MD_NEW_SIZE_SECTORS(num_sectors);
 932 }
 933
 934 static int alloc_disk_sb(struct md_rdev *rdev)
 935 {
 936         rdev->sb_page = alloc_page(GFP_KERNEL);
 937         if (!rdev->sb_page)
 938                 return -ENOMEM;
 939         return 0;
 940 }
 941
 942 void md_rdev_clear(struct md_rdev *rdev)
 943 {
 944         if (rdev->sb_page) {
 945                 put_page(rdev->sb_page);
 946                 rdev->sb_loaded = 0;
 947                 rdev->sb_page = NULL;
 948                 rdev->sb_start = 0;
 949                 rdev->sectors = 0;
 950         }
 951         if (rdev->bb_page) {
 952                 put_page(rdev->bb_page);
 953                 rdev->bb_page = NULL;
 954         }
 955         badblocks_exit(&rdev->badblocks);
 956 }
 957 EXPORT_SYMBOL_GPL(md_rdev_clear);
 958
 959 static void super_written(struct bio *bio)
 960 {
 961         struct md_rdev *rdev = bio->bi_private;
 962         struct mddev *mddev = rdev->mddev;
 963
 964         if (bio->bi_status) {
 965                 pr_err("md: %s gets error=%d\n", __func__,
 966                        blk_status_to_errno(bio->bi_status));
 967                 md_error(mddev, rdev);
 968                 if (!test_bit(Faulty, &rdev->flags)
 969                     && (bio->bi_opf & MD_FAILFAST)) {
 970                         set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
 971                         set_bit(LastDev, &rdev->flags);
 972                 }
 973         } else
 974                 clear_bit(LastDev, &rdev->flags);
 975
 976         if (atomic_dec_and_test(&mddev->pending_writes))
 977                 wake_up(&mddev->sb_wait);
 978         rdev_dec_pending(rdev, mddev);
 979         bio_put(bio);
 980 }
 981
 982 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 983                    sector_t sector, int size, struct page *page)
 984 {
 985         /* write first size bytes of page to sector of rdev
 986          * Increment mddev->pending_writes before returning
 987          * and decrement it on completion, waking up sb_wait
 988          * if zero is reached.
 989          * If an error occurred, call md_error
 990          */
 991         struct bio *bio;
 992         int ff = 0;
 993
 994         if (!page)
 995                 return;
 996
 997         if (test_bit(Faulty, &rdev->flags))
 998                 return;
 999
1000         bio = bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
1001
1002         atomic_inc(&rdev->nr_pending);
1003
1004         bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
1005         bio->bi_iter.bi_sector = sector;
1006         bio_add_page(bio, page, size, 0);
1007         bio->bi_private = rdev;
1008         bio->bi_end_io = super_written;
1009
1010         if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1011             test_bit(FailFast, &rdev->flags) &&
1012             !test_bit(LastDev, &rdev->flags))
1013                 ff = MD_FAILFAST;
1014         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
1015
1016         atomic_inc(&mddev->pending_writes);
1017         submit_bio(bio);
1018 }
1019
1020 int md_super_wait(struct mddev *mddev)
1021 {
1022         /* wait for all superblock writes that were scheduled to complete */
1023         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1024         if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1025                 return -EAGAIN;
1026         return 0;
1027 }
1028
1029 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1030                  struct page *page, int op, int op_flags, bool metadata_op)
1031 {
1032         struct bio bio;
1033         struct bio_vec bvec;
1034
1035         bio_init(&bio, &bvec, 1);
1036
1037         if (metadata_op && rdev->meta_bdev)
1038                 bio_set_dev(&bio, rdev->meta_bdev);
1039         else
1040                 bio_set_dev(&bio, rdev->bdev);
1041         bio.bi_opf = op | op_flags;
1042         if (metadata_op)
1043                 bio.bi_iter.bi_sector = sector + rdev->sb_start;
1044         else if (rdev->mddev->reshape_position != MaxSector &&
1045                  (rdev->mddev->reshape_backwards ==
1046                   (sector >= rdev->mddev->reshape_position)))
1047                 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1048         else
1049                 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1050         bio_add_page(&bio, page, size, 0);
1051
1052         submit_bio_wait(&bio);
1053
1054         return !bio.bi_status;
1055 }
1056 EXPORT_SYMBOL_GPL(sync_page_io);
1057
1058 static int read_disk_sb(struct md_rdev *rdev, int size)
1059 {
1060         char b[BDEVNAME_SIZE];
1061
1062         if (rdev->sb_loaded)
1063                 return 0;
1064
1065         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1066                 goto fail;
1067         rdev->sb_loaded = 1;
1068         return 0;
1069
1070 fail:
1071         pr_err("md: disabled device %s, could not read superblock.\n",
1072                bdevname(rdev->bdev,b));
1073         return -EINVAL;
1074 }
1075
1076 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1077 {
1078         return  sb1->set_uuid0 == sb2->set_uuid0 &&
1079                 sb1->set_uuid1 == sb2->set_uuid1 &&
1080                 sb1->set_uuid2 == sb2->set_uuid2 &&
1081                 sb1->set_uuid3 == sb2->set_uuid3;
1082 }
1083
1084 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1085 {
1086         int ret;
1087         mdp_super_t *tmp1, *tmp2;
1088
1089         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1090         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1091
1092         if (!tmp1 || !tmp2) {
1093                 ret = 0;
1094                 goto abort;
1095         }
1096
1097         *tmp1 = *sb1;
1098         *tmp2 = *sb2;
1099
1100         /*
1101          * nr_disks is not constant
1102          */
1103         tmp1->nr_disks = 0;
1104         tmp2->nr_disks = 0;
1105
1106         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1107 abort:
1108         kfree(tmp1);
1109         kfree(tmp2);
1110         return ret;
1111 }
1112
1113 static u32 md_csum_fold(u32 csum)
1114 {
1115         csum = (csum & 0xffff) + (csum >> 16);
1116         return (csum & 0xffff) + (csum >> 16);
1117 }
1118
1119 static unsigned int calc_sb_csum(mdp_super_t *sb)
1120 {
1121         u64 newcsum = 0;
1122         u32 *sb32 = (u32*)sb;
1123         int i;
1124         unsigned int disk_csum, csum;
1125
1126         disk_csum = sb->sb_csum;
1127         sb->sb_csum = 0;
1128
1129         for (i = 0; i < MD_SB_BYTES/4 ; i++)
1130                 newcsum += sb32[i];
1131         csum = (newcsum & 0xffffffff) + (newcsum>>32);
1132
1133 #ifdef CONFIG_ALPHA
1134         /* This used to use csum_partial, which was wrong for several
1135          * reasons including that different results are returned on
1136          * different architectures.  It isn't critical that we get exactly
1137          * the same return value as before (we always csum_fold before
1138          * testing, and that removes any differences).  However as we
1139          * know that csum_partial always returned a 16bit value on
1140          * alphas, do a fold to maximise conformity to previous behaviour.
1141          */
1142         sb->sb_csum = md_csum_fold(disk_csum);
1143 #else
1144         sb->sb_csum = disk_csum;
1145 #endif
1146         return csum;
1147 }
1148
1149 /*
1150  * Handle superblock details.
1151  * We want to be able to handle multiple superblock formats
1152  * so we have a common interface to them all, and an array of
1153  * different handlers.
1154  * We rely on user-space to write the initial superblock, and support
1155  * reading and updating of superblocks.
1156  * Interface methods are:
1157  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1158  *      loads and validates a superblock on dev.
1159  *      if refdev != NULL, compare superblocks on both devices
1160  *    Return:
1161  *      0 - dev has a superblock that is compatible with refdev
1162  *      1 - dev has a superblock that is compatible and newer than refdev
1163  *          so dev should be used as the refdev in future
1164  *     -EINVAL superblock incompatible or invalid
1165  *     -othererror e.g. -EIO
1166  *
1167  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1168  *      Verify that dev is acceptable into mddev.
1169  *       The first time, mddev->raid_disks will be 0, and data from
1170  *       dev should be merged in.  Subsequent calls check that dev
1171  *       is new enough.  Return 0 or -EINVAL
1172  *
1173  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1174  *     Update the superblock for rdev with data in mddev
1175  *     This does not write to disc.
1176  *
1177  */
1178
1179 struct super_type  {
1180         char                *name;
1181         struct module       *owner;
1182         int                 (*load_super)(struct md_rdev *rdev,
1183                                           struct md_rdev *refdev,
1184                                           int minor_version);
1185         int                 (*validate_super)(struct mddev *mddev,
1186                                               struct md_rdev *rdev);
1187         void                (*sync_super)(struct mddev *mddev,
1188                                           struct md_rdev *rdev);
1189         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1190                                                 sector_t num_sectors);
1191         int                 (*allow_new_offset)(struct md_rdev *rdev,
1192                                                 unsigned long long new_offset);
1193 };
1194
1195 /*
1196  * Check that the given mddev has no bitmap.
1197  *
1198  * This function is called from the run method of all personalities that do not
1199  * support bitmaps. It prints an error message and returns non-zero if mddev
1200  * has a bitmap. Otherwise, it returns 0.
1201  *
1202  */
1203 int md_check_no_bitmap(struct mddev *mddev)
1204 {
1205         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1206                 return 0;
1207         pr_warn("%s: bitmaps are not supported for %s\n",
1208                 mdname(mddev), mddev->pers->name);
1209         return 1;
1210 }
1211 EXPORT_SYMBOL(md_check_no_bitmap);
1212
1213 /*
1214  * load_super for 0.90.0
1215  */
1216 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1217 {
1218         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1219         mdp_super_t *sb;
1220         int ret;
1221         bool spare_disk = true;
1222
1223         /*
1224          * Calculate the position of the superblock (512byte sectors),
1225          * it's at the end of the disk.
1226          *
1227          * It also happens to be a multiple of 4Kb.
1228          */
1229         rdev->sb_start = calc_dev_sboffset(rdev);
1230
1231         ret = read_disk_sb(rdev, MD_SB_BYTES);
1232         if (ret)
1233                 return ret;
1234
1235         ret = -EINVAL;
1236
1237         bdevname(rdev->bdev, b);
1238         sb = page_address(rdev->sb_page);
1239
1240         if (sb->md_magic != MD_SB_MAGIC) {
1241                 pr_warn("md: invalid raid superblock magic on %s\n", b);
1242                 goto abort;
1243         }
1244
1245         if (sb->major_version != 0 ||
1246             sb->minor_version < 90 ||
1247             sb->minor_version > 91) {
1248                 pr_warn("Bad version number %d.%d on %s\n",
1249                         sb->major_version, sb->minor_version, b);
1250                 goto abort;
1251         }
1252
1253         if (sb->raid_disks <= 0)
1254                 goto abort;
1255
1256         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1257                 pr_warn("md: invalid superblock checksum on %s\n", b);
1258                 goto abort;
1259         }
1260
1261         rdev->preferred_minor = sb->md_minor;
1262         rdev->data_offset = 0;
1263         rdev->new_data_offset = 0;
1264         rdev->sb_size = MD_SB_BYTES;
1265         rdev->badblocks.shift = -1;
1266
1267         if (sb->level == LEVEL_MULTIPATH)
1268                 rdev->desc_nr = -1;
1269         else
1270                 rdev->desc_nr = sb->this_disk.number;
1271
1272         /* not spare disk, or LEVEL_MULTIPATH */
1273         if (sb->level == LEVEL_MULTIPATH ||
1274                 (rdev->desc_nr >= 0 &&
1275                  rdev->desc_nr < MD_SB_DISKS &&
1276                  sb->disks[rdev->desc_nr].state &
1277                  ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1278                 spare_disk = false;
1279
1280         if (!refdev) {
1281                 if (!spare_disk)
1282                         ret = 1;
1283                 else
1284                         ret = 0;
1285         } else {
1286                 __u64 ev1, ev2;
1287                 mdp_super_t *refsb = page_address(refdev->sb_page);
1288                 if (!md_uuid_equal(refsb, sb)) {
1289                         pr_warn("md: %s has different UUID to %s\n",
1290                                 b, bdevname(refdev->bdev,b2));
1291                         goto abort;
1292                 }
1293                 if (!md_sb_equal(refsb, sb)) {
1294                         pr_warn("md: %s has same UUID but different superblock to %s\n",
1295                                 b, bdevname(refdev->bdev, b2));
1296                         goto abort;
1297                 }
1298                 ev1 = md_event(sb);
1299                 ev2 = md_event(refsb);
1300
1301                 if (!spare_disk && ev1 > ev2)
1302                         ret = 1;
1303                 else
1304                         ret = 0;
1305         }
1306         rdev->sectors = rdev->sb_start;
1307         /* Limit to 4TB as metadata cannot record more than that.
1308          * (not needed for Linear and RAID0 as metadata doesn't
1309          * record this size)
1310          */
1311         if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1312                 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1313
1314         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1315                 /* "this cannot possibly happen" ... */
1316                 ret = -EINVAL;
1317
1318  abort:
1319         return ret;
1320 }
1321
1322 /*
1323  * validate_super for 0.90.0
1324  */
1325 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1326 {
1327         mdp_disk_t *desc;
1328         mdp_super_t *sb = page_address(rdev->sb_page);
1329         __u64 ev1 = md_event(sb);
1330
1331         rdev->raid_disk = -1;
1332         clear_bit(Faulty, &rdev->flags);
1333         clear_bit(In_sync, &rdev->flags);
1334         clear_bit(Bitmap_sync, &rdev->flags);
1335         clear_bit(WriteMostly, &rdev->flags);
1336
1337         if (mddev->raid_disks == 0) {
1338                 mddev->major_version = 0;
1339                 mddev->minor_version = sb->minor_version;
1340                 mddev->patch_version = sb->patch_version;
1341                 mddev->external = 0;
1342                 mddev->chunk_sectors = sb->chunk_size >> 9;
1343                 mddev->ctime = sb->ctime;
1344                 mddev->utime = sb->utime;
1345                 mddev->level = sb->level;
1346                 mddev->clevel[0] = 0;
1347                 mddev->layout = sb->layout;
1348                 mddev->raid_disks = sb->raid_disks;
1349                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1350                 mddev->events = ev1;
1351                 mddev->bitmap_info.offset = 0;
1352                 mddev->bitmap_info.space = 0;
1353                 /* bitmap can use 60 K after the 4K superblocks */
1354                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1355                 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1356                 mddev->reshape_backwards = 0;
1357
1358                 if (mddev->minor_version >= 91) {
1359                         mddev->reshape_position = sb->reshape_position;
1360                         mddev->delta_disks = sb->delta_disks;
1361                         mddev->new_level = sb->new_level;
1362                         mddev->new_layout = sb->new_layout;
1363                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1364                         if (mddev->delta_disks < 0)
1365                                 mddev->reshape_backwards = 1;
1366                 } else {
1367                         mddev->reshape_position = MaxSector;
1368                         mddev->delta_disks = 0;
1369                         mddev->new_level = mddev->level;
1370                         mddev->new_layout = mddev->layout;
1371                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1372                 }
1373                 if (mddev->level == 0)
1374                         mddev->layout = -1;
1375
1376                 if (sb->state & (1<<MD_SB_CLEAN))
1377                         mddev->recovery_cp = MaxSector;
1378                 else {
1379                         if (sb->events_hi == sb->cp_events_hi &&
1380                                 sb->events_lo == sb->cp_events_lo) {
1381                                 mddev->recovery_cp = sb->recovery_cp;
1382                         } else
1383                                 mddev->recovery_cp = 0;
1384                 }
1385
1386                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1387                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1388                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1389                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1390
1391                 mddev->max_disks = MD_SB_DISKS;
1392
1393                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1394                     mddev->bitmap_info.file == NULL) {
1395                         mddev->bitmap_info.offset =
1396                                 mddev->bitmap_info.default_offset;
1397                         mddev->bitmap_info.space =
1398                                 mddev->bitmap_info.default_space;
1399                 }
1400
1401         } else if (mddev->pers == NULL) {
1402                 /* Insist on good event counter while assembling, except
1403                  * for spares (which don't need an event count) */
1404                 ++ev1;
1405                 if (sb->disks[rdev->desc_nr].state & (
1406                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1407                         if (ev1 < mddev->events)
1408                                 return -EINVAL;
1409         } else if (mddev->bitmap) {
1410                 /* if adding to array with a bitmap, then we can accept an
1411                  * older device ... but not too old.
1412                  */
1413                 if (ev1 < mddev->bitmap->events_cleared)
1414                         return 0;
1415                 if (ev1 < mddev->events)
1416                         set_bit(Bitmap_sync, &rdev->flags);
1417         } else {
1418                 if (ev1 < mddev->events)
1419                         /* just a hot-add of a new device, leave raid_disk at -1 */
1420                         return 0;
1421         }
1422
1423         if (mddev->level != LEVEL_MULTIPATH) {
1424                 desc = sb->disks + rdev->desc_nr;
1425
1426                 if (desc->state & (1<<MD_DISK_FAULTY))
1427                         set_bit(Faulty, &rdev->flags);
1428                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1429                             desc->raid_disk < mddev->raid_disks */) {
1430                         set_bit(In_sync, &rdev->flags);
1431                         rdev->raid_disk = desc->raid_disk;
1432                         rdev->saved_raid_disk = desc->raid_disk;
1433                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1434                         /* active but not in sync implies recovery up to
1435                          * reshape position.  We don't know exactly where
1436                          * that is, so set to zero for now */
1437                         if (mddev->minor_version >= 91) {
1438                                 rdev->recovery_offset = 0;
1439                                 rdev->raid_disk = desc->raid_disk;
1440                         }
1441                 }
1442                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1443                         set_bit(WriteMostly, &rdev->flags);
1444                 if (desc->state & (1<<MD_DISK_FAILFAST))
1445                         set_bit(FailFast, &rdev->flags);
1446         } else /* MULTIPATH are always insync */
1447                 set_bit(In_sync, &rdev->flags);
1448         return 0;
1449 }
1450
1451 /*
1452  * sync_super for 0.90.0
1453  */
1454 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1455 {
1456         mdp_super_t *sb;
1457         struct md_rdev *rdev2;
1458         int next_spare = mddev->raid_disks;
1459
1460         /* make rdev->sb match mddev data..
1461          *
1462          * 1/ zero out disks
1463          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1464          * 3/ any empty disks < next_spare become removed
1465          *
1466          * disks[0] gets initialised to REMOVED because
1467          * we cannot be sure from other fields if it has
1468          * been initialised or not.
1469          */
1470         int i;
1471         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1472
1473         rdev->sb_size = MD_SB_BYTES;
1474
1475         sb = page_address(rdev->sb_page);
1476
1477         memset(sb, 0, sizeof(*sb));
1478
1479         sb->md_magic = MD_SB_MAGIC;
1480         sb->major_version = mddev->major_version;
1481         sb->patch_version = mddev->patch_version;
1482         sb->gvalid_words  = 0; /* ignored */
1483         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1484         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1485         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1486         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1487
1488         sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1489         sb->level = mddev->level;
1490         sb->size = mddev->dev_sectors / 2;
1491         sb->raid_disks = mddev->raid_disks;
1492         sb->md_minor = mddev->md_minor;
1493         sb->not_persistent = 0;
1494         sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1495         sb->state = 0;
1496         sb->events_hi = (mddev->events>>32);
1497         sb->events_lo = (u32)mddev->events;
1498
1499         if (mddev->reshape_position == MaxSector)
1500                 sb->minor_version = 90;
1501         else {
1502                 sb->minor_version = 91;
1503                 sb->reshape_position = mddev->reshape_position;
1504                 sb->new_level = mddev->new_level;
1505                 sb->delta_disks = mddev->delta_disks;
1506                 sb->new_layout = mddev->new_layout;
1507                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1508         }
1509         mddev->minor_version = sb->minor_version;
1510         if (mddev->in_sync)
1511         {
1512                 sb->recovery_cp = mddev->recovery_cp;
1513                 sb->cp_events_hi = (mddev->events>>32);
1514                 sb->cp_events_lo = (u32)mddev->events;
1515                 if (mddev->recovery_cp == MaxSector)
1516                         sb->state = (1<< MD_SB_CLEAN);
1517         } else
1518                 sb->recovery_cp = 0;
1519
1520         sb->layout = mddev->layout;
1521         sb->chunk_size = mddev->chunk_sectors << 9;
1522
1523         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1524                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1525
1526         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1527         rdev_for_each(rdev2, mddev) {
1528                 mdp_disk_t *d;
1529                 int desc_nr;
1530                 int is_active = test_bit(In_sync, &rdev2->flags);
1531
1532                 if (rdev2->raid_disk >= 0 &&
1533                     sb->minor_version >= 91)
1534                         /* we have nowhere to store the recovery_offset,
1535                          * but if it is not below the reshape_position,
1536                          * we can piggy-back on that.
1537                          */
1538                         is_active = 1;
1539                 if (rdev2->raid_disk < 0 ||
1540                     test_bit(Faulty, &rdev2->flags))
1541                         is_active = 0;
1542                 if (is_active)
1543                         desc_nr = rdev2->raid_disk;
1544                 else
1545                         desc_nr = next_spare++;
1546                 rdev2->desc_nr = desc_nr;
1547                 d = &sb->disks[rdev2->desc_nr];
1548                 nr_disks++;
1549                 d->number = rdev2->desc_nr;
1550                 d->major = MAJOR(rdev2->bdev->bd_dev);
1551                 d->minor = MINOR(rdev2->bdev->bd_dev);
1552                 if (is_active)
1553                         d->raid_disk = rdev2->raid_disk;
1554                 else
1555                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1556                 if (test_bit(Faulty, &rdev2->flags))
1557                         d->state = (1<<MD_DISK_FAULTY);
1558                 else if (is_active) {
1559                         d->state = (1<<MD_DISK_ACTIVE);
1560                         if (test_bit(In_sync, &rdev2->flags))
1561                                 d->state |= (1<<MD_DISK_SYNC);
1562                         active++;
1563                         working++;
1564                 } else {
1565                         d->state = 0;
1566                         spare++;
1567                         working++;
1568                 }
1569                 if (test_bit(WriteMostly, &rdev2->flags))
1570                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1571                 if (test_bit(FailFast, &rdev2->flags))
1572                         d->state |= (1<<MD_DISK_FAILFAST);
1573         }
1574         /* now set the "removed" and "faulty" bits on any missing devices */
1575         for (i=0 ; i < mddev->raid_disks ; i++) {
1576                 mdp_disk_t *d = &sb->disks[i];
1577                 if (d->state == 0 && d->number == 0) {
1578                         d->number = i;
1579                         d->raid_disk = i;
1580                         d->state = (1<<MD_DISK_REMOVED);
1581                         d->state |= (1<<MD_DISK_FAULTY);
1582                         failed++;
1583                 }
1584         }
1585         sb->nr_disks = nr_disks;
1586         sb->active_disks = active;
1587         sb->working_disks = working;
1588         sb->failed_disks = failed;
1589         sb->spare_disks = spare;
1590
1591         sb->this_disk = sb->disks[rdev->desc_nr];
1592         sb->sb_csum = calc_sb_csum(sb);
1593 }
1594
1595 /*
1596  * rdev_size_change for 0.90.0
1597  */
1598 static unsigned long long
1599 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1600 {
1601         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1602                 return 0; /* component must fit device */
1603         if (rdev->mddev->bitmap_info.offset)
1604                 return 0; /* can't move bitmap */
1605         rdev->sb_start = calc_dev_sboffset(rdev);
1606         if (!num_sectors || num_sectors > rdev->sb_start)
1607                 num_sectors = rdev->sb_start;
1608         /* Limit to 4TB as metadata cannot record more than that.
1609          * 4TB == 2^32 KB, or 2*2^32 sectors.
1610          */
1611         if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1612                 num_sectors = (sector_t)(2ULL << 32) - 2;
1613         do {
1614                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1615                        rdev->sb_page);
1616         } while (md_super_wait(rdev->mddev) < 0);
1617         return num_sectors;
1618 }
1619
1620 static int
1621 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1622 {
1623         /* non-zero offset changes not possible with v0.90 */
1624         return new_offset == 0;
1625 }
1626
1627 /*
1628  * version 1 superblock
1629  */
1630
1631 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1632 {
1633         __le32 disk_csum;
1634         u32 csum;
1635         unsigned long long newcsum;
1636         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1637         __le32 *isuper = (__le32*)sb;
1638
1639         disk_csum = sb->sb_csum;
1640         sb->sb_csum = 0;
1641         newcsum = 0;
1642         for (; size >= 4; size -= 4)
1643                 newcsum += le32_to_cpu(*isuper++);
1644
1645         if (size == 2)
1646                 newcsum += le16_to_cpu(*(__le16*) isuper);
1647
1648         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1649         sb->sb_csum = disk_csum;
1650         return cpu_to_le32(csum);
1651 }
1652
1653 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1654 {
1655         struct mdp_superblock_1 *sb;
1656         int ret;
1657         sector_t sb_start;
1658         sector_t sectors;
1659         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1660         int bmask;
1661         bool spare_disk = true;
1662
1663         /*
1664          * Calculate the position of the superblock in 512byte sectors.
1665          * It is always aligned to a 4K boundary and
1666          * depeding on minor_version, it can be:
1667          * 0: At least 8K, but less than 12K, from end of device
1668          * 1: At start of device
1669          * 2: 4K from start of device.
1670          */
1671         switch(minor_version) {
1672         case 0:
1673                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1674                 sb_start -= 8*2;
1675                 sb_start &= ~(sector_t)(4*2-1);
1676                 break;
1677         case 1:
1678                 sb_start = 0;
1679                 break;
1680         case 2:
1681                 sb_start = 8;
1682                 break;
1683         default:
1684                 return -EINVAL;
1685         }
1686         rdev->sb_start = sb_start;
1687
1688         /* superblock is rarely larger than 1K, but it can be larger,
1689          * and it is safe to read 4k, so we do that
1690          */
1691         ret = read_disk_sb(rdev, 4096);
1692         if (ret) return ret;
1693
1694         sb = page_address(rdev->sb_page);
1695
1696         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1697             sb->major_version != cpu_to_le32(1) ||
1698             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1699             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1700             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1701                 return -EINVAL;
1702
1703         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1704                 pr_warn("md: invalid superblock checksum on %s\n",
1705                         bdevname(rdev->bdev,b));
1706                 return -EINVAL;
1707         }
1708         if (le64_to_cpu(sb->data_size) < 10) {
1709                 pr_warn("md: data_size too small on %s\n",
1710                         bdevname(rdev->bdev,b));
1711                 return -EINVAL;
1712         }
1713         if (sb->pad0 ||
1714             sb->pad3[0] ||
1715             memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1716                 /* Some padding is non-zero, might be a new feature */
1717                 return -EINVAL;
1718
1719         rdev->preferred_minor = 0xffff;
1720         rdev->data_offset = le64_to_cpu(sb->data_offset);
1721         rdev->new_data_offset = rdev->data_offset;
1722         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1723             (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1724                 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1725         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1726
1727         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1728         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1729         if (rdev->sb_size & bmask)
1730                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1731
1732         if (minor_version
1733             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1734                 return -EINVAL;
1735         if (minor_version
1736             && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1737                 return -EINVAL;
1738
1739         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1740                 rdev->desc_nr = -1;
1741         else
1742                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1743
1744         if (!rdev->bb_page) {
1745                 rdev->bb_page = alloc_page(GFP_KERNEL);
1746                 if (!rdev->bb_page)
1747                         return -ENOMEM;
1748         }
1749         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1750             rdev->badblocks.count == 0) {
1751                 /* need to load the bad block list.
1752                  * Currently we limit it to one page.
1753                  */
1754                 s32 offset;
1755                 sector_t bb_sector;
1756                 __le64 *bbp;
1757                 int i;
1758                 int sectors = le16_to_cpu(sb->bblog_size);
1759                 if (sectors > (PAGE_SIZE / 512))
1760                         return -EINVAL;
1761                 offset = le32_to_cpu(sb->bblog_offset);
1762                 if (offset == 0)
1763                         return -EINVAL;
1764                 bb_sector = (long long)offset;
1765                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1766                                   rdev->bb_page, REQ_OP_READ, 0, true))
1767                         return -EIO;
1768                 bbp = (__le64 *)page_address(rdev->bb_page);
1769                 rdev->badblocks.shift = sb->bblog_shift;
1770                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1771                         u64 bb = le64_to_cpu(*bbp);
1772                         int count = bb & (0x3ff);
1773                         u64 sector = bb >> 10;
1774                         sector <<= sb->bblog_shift;
1775                         count <<= sb->bblog_shift;
1776                         if (bb + 1 == 0)
1777                                 break;
1778                         if (badblocks_set(&rdev->badblocks, sector, count, 1))
1779                                 return -EINVAL;
1780                 }
1781         } else if (sb->bblog_offset != 0)
1782                 rdev->badblocks.shift = 0;
1783
1784         if ((le32_to_cpu(sb->feature_map) &
1785             (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1786                 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1787                 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1788                 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1789         }
1790
1791         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1792             sb->level != 0)
1793                 return -EINVAL;
1794
1795         /* not spare disk, or LEVEL_MULTIPATH */
1796         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1797                 (rdev->desc_nr >= 0 &&
1798                 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1799                 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1800                  le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1801                 spare_disk = false;
1802
1803         if (!refdev) {
1804                 if (!spare_disk)
1805                         ret = 1;
1806                 else
1807                         ret = 0;
1808         } else {
1809                 __u64 ev1, ev2;
1810                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1811
1812                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1813                     sb->level != refsb->level ||
1814                     sb->layout != refsb->layout ||
1815                     sb->chunksize != refsb->chunksize) {
1816                         pr_warn("md: %s has strangely different superblock to %s\n",
1817                                 bdevname(rdev->bdev,b),
1818                                 bdevname(refdev->bdev,b2));
1819                         return -EINVAL;
1820                 }
1821                 ev1 = le64_to_cpu(sb->events);
1822                 ev2 = le64_to_cpu(refsb->events);
1823
1824                 if (!spare_disk && ev1 > ev2)
1825                         ret = 1;
1826                 else
1827                         ret = 0;
1828         }
1829         if (minor_version) {
1830                 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1831                 sectors -= rdev->data_offset;
1832         } else
1833                 sectors = rdev->sb_start;
1834         if (sectors < le64_to_cpu(sb->data_size))
1835                 return -EINVAL;
1836         rdev->sectors = le64_to_cpu(sb->data_size);
1837         return ret;
1838 }
1839
1840 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1841 {
1842         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1843         __u64 ev1 = le64_to_cpu(sb->events);
1844
1845         rdev->raid_disk = -1;
1846         clear_bit(Faulty, &rdev->flags);
1847         clear_bit(In_sync, &rdev->flags);
1848         clear_bit(Bitmap_sync, &rdev->flags);
1849         clear_bit(WriteMostly, &rdev->flags);
1850
1851         if (mddev->raid_disks == 0) {
1852                 mddev->major_version = 1;
1853                 mddev->patch_version = 0;
1854                 mddev->external = 0;
1855                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1856                 mddev->ctime = le64_to_cpu(sb->ctime);
1857                 mddev->utime = le64_to_cpu(sb->utime);
1858                 mddev->level = le32_to_cpu(sb->level);
1859                 mddev->clevel[0] = 0;
1860                 mddev->layout = le32_to_cpu(sb->layout);
1861                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1862                 mddev->dev_sectors = le64_to_cpu(sb->size);
1863                 mddev->events = ev1;
1864                 mddev->bitmap_info.offset = 0;
1865                 mddev->bitmap_info.space = 0;
1866                 /* Default location for bitmap is 1K after superblock
1867                  * using 3K - total of 4K
1868                  */
1869                 mddev->bitmap_info.default_offset = 1024 >> 9;
1870                 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1871                 mddev->reshape_backwards = 0;
1872
1873                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1874                 memcpy(mddev->uuid, sb->set_uuid, 16);
1875
1876                 mddev->max_disks =  (4096-256)/2;
1877
1878                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1879                     mddev->bitmap_info.file == NULL) {
1880                         mddev->bitmap_info.offset =
1881                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1882                         /* Metadata doesn't record how much space is available.
1883                          * For 1.0, we assume we can use up to the superblock
1884                          * if before, else to 4K beyond superblock.
1885                          * For others, assume no change is possible.
1886                          */
1887                         if (mddev->minor_version > 0)
1888                                 mddev->bitmap_info.space = 0;
1889                         else if (mddev->bitmap_info.offset > 0)
1890                                 mddev->bitmap_info.space =
1891                                         8 - mddev->bitmap_info.offset;
1892                         else
1893                                 mddev->bitmap_info.space =
1894                                         -mddev->bitmap_info.offset;
1895                 }
1896
1897                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1898                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1899                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1900                         mddev->new_level = le32_to_cpu(sb->new_level);
1901                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1902                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1903                         if (mddev->delta_disks < 0 ||
1904                             (mddev->delta_disks == 0 &&
1905                              (le32_to_cpu(sb->feature_map)
1906                               & MD_FEATURE_RESHAPE_BACKWARDS)))
1907                                 mddev->reshape_backwards = 1;
1908                 } else {
1909                         mddev->reshape_position = MaxSector;
1910                         mddev->delta_disks = 0;
1911                         mddev->new_level = mddev->level;
1912                         mddev->new_layout = mddev->layout;
1913                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1914                 }
1915
1916                 if (mddev->level == 0 &&
1917                     !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1918                         mddev->layout = -1;
1919
1920                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1921                         set_bit(MD_HAS_JOURNAL, &mddev->flags);
1922
1923                 if (le32_to_cpu(sb->feature_map) &
1924                     (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1925                         if (le32_to_cpu(sb->feature_map) &
1926                             (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1927                                 return -EINVAL;
1928                         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1929                             (le32_to_cpu(sb->feature_map) &
1930                                             MD_FEATURE_MULTIPLE_PPLS))
1931                                 return -EINVAL;
1932                         set_bit(MD_HAS_PPL, &mddev->flags);
1933                 }
1934         } else if (mddev->pers == NULL) {
1935                 /* Insist of good event counter while assembling, except for
1936                  * spares (which don't need an event count) */
1937                 ++ev1;
1938                 if (rdev->desc_nr >= 0 &&
1939                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1940                     (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1941                      le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1942                         if (ev1 < mddev->events)
1943                                 return -EINVAL;
1944         } else if (mddev->bitmap) {
1945                 /* If adding to array with a bitmap, then we can accept an
1946                  * older device, but not too old.
1947                  */
1948                 if (ev1 < mddev->bitmap->events_cleared)
1949                         return 0;
1950                 if (ev1 < mddev->events)
1951                         set_bit(Bitmap_sync, &rdev->flags);
1952         } else {
1953                 if (ev1 < mddev->events)
1954                         /* just a hot-add of a new device, leave raid_disk at -1 */
1955                         return 0;
1956         }
1957         if (mddev->level != LEVEL_MULTIPATH) {
1958                 int role;
1959                 if (rdev->desc_nr < 0 ||
1960                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1961                         role = MD_DISK_ROLE_SPARE;
1962                         rdev->desc_nr = -1;
1963                 } else
1964                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1965                 switch(role) {
1966                 case MD_DISK_ROLE_SPARE: /* spare */
1967                         break;
1968                 case MD_DISK_ROLE_FAULTY: /* faulty */
1969                         set_bit(Faulty, &rdev->flags);
1970                         break;
1971                 case MD_DISK_ROLE_JOURNAL: /* journal device */
1972                         if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1973                                 /* journal device without journal feature */
1974                                 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1975                                 return -EINVAL;
1976                         }
1977                         set_bit(Journal, &rdev->flags);
1978                         rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1979                         rdev->raid_disk = 0;
1980                         break;
1981                 default:
1982                         rdev->saved_raid_disk = role;
1983                         if ((le32_to_cpu(sb->feature_map) &
1984                              MD_FEATURE_RECOVERY_OFFSET)) {
1985                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1986                                 if (!(le32_to_cpu(sb->feature_map) &
1987                                       MD_FEATURE_RECOVERY_BITMAP))
1988                                         rdev->saved_raid_disk = -1;
1989                         } else {
1990                                 /*
1991                                  * If the array is FROZEN, then the device can't
1992                                  * be in_sync with rest of array.
1993                                  */
1994                                 if (!test_bit(MD_RECOVERY_FROZEN,
1995                                               &mddev->recovery))
1996                                         set_bit(In_sync, &rdev->flags);
1997                         }
1998                         rdev->raid_disk = role;
1999                         break;
2000                 }
2001                 if (sb->devflags & WriteMostly1)
2002                         set_bit(WriteMostly, &rdev->flags);
2003                 if (sb->devflags & FailFast1)
2004                         set_bit(FailFast, &rdev->flags);
2005                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2006                         set_bit(Replacement, &rdev->flags);
2007         } else /* MULTIPATH are always insync */
2008                 set_bit(In_sync, &rdev->flags);
2009
2010         return 0;
2011 }
2012
2013 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2014 {
2015         struct mdp_superblock_1 *sb;
2016         struct md_rdev *rdev2;
2017         int max_dev, i;
2018         /* make rdev->sb match mddev and rdev data. */
2019
2020         sb = page_address(rdev->sb_page);
2021
2022         sb->feature_map = 0;
2023         sb->pad0 = 0;
2024         sb->recovery_offset = cpu_to_le64(0);
2025         memset(sb->pad3, 0, sizeof(sb->pad3));
2026
2027         sb->utime = cpu_to_le64((__u64)mddev->utime);
2028         sb->events = cpu_to_le64(mddev->events);
2029         if (mddev->in_sync)
2030                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2031         else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2032                 sb->resync_offset = cpu_to_le64(MaxSector);
2033         else
2034                 sb->resync_offset = cpu_to_le64(0);
2035
2036         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2037
2038         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2039         sb->size = cpu_to_le64(mddev->dev_sectors);
2040         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2041         sb->level = cpu_to_le32(mddev->level);
2042         sb->layout = cpu_to_le32(mddev->layout);
2043         if (test_bit(FailFast, &rdev->flags))
2044                 sb->devflags |= FailFast1;
2045         else
2046                 sb->devflags &= ~FailFast1;
2047
2048         if (test_bit(WriteMostly, &rdev->flags))
2049                 sb->devflags |= WriteMostly1;
2050         else
2051                 sb->devflags &= ~WriteMostly1;
2052         sb->data_offset = cpu_to_le64(rdev->data_offset);
2053         sb->data_size = cpu_to_le64(rdev->sectors);
2054
2055         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2056                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2057                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2058         }
2059
2060         if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2061             !test_bit(In_sync, &rdev->flags)) {
2062                 sb->feature_map |=
2063                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2064                 sb->recovery_offset =
2065                         cpu_to_le64(rdev->recovery_offset);
2066                 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2067                         sb->feature_map |=
2068                                 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2069         }
2070         /* Note: recovery_offset and journal_tail share space  */
2071         if (test_bit(Journal, &rdev->flags))
2072                 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2073         if (test_bit(Replacement, &rdev->flags))
2074                 sb->feature_map |=
2075                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
2076
2077         if (mddev->reshape_position != MaxSector) {
2078                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2079                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2080                 sb->new_layout = cpu_to_le32(mddev->new_layout);
2081                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2082                 sb->new_level = cpu_to_le32(mddev->new_level);
2083                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2084                 if (mddev->delta_disks == 0 &&
2085                     mddev->reshape_backwards)
2086                         sb->feature_map
2087                                 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2088                 if (rdev->new_data_offset != rdev->data_offset) {
2089                         sb->feature_map
2090                                 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2091                         sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2092                                                              - rdev->data_offset));
2093                 }
2094         }
2095
2096         if (mddev_is_clustered(mddev))
2097                 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2098
2099         if (rdev->badblocks.count == 0)
2100                 /* Nothing to do for bad blocks*/ ;
2101         else if (sb->bblog_offset == 0)
2102                 /* Cannot record bad blocks on this device */
2103                 md_error(mddev, rdev);
2104         else {
2105                 struct badblocks *bb = &rdev->badblocks;
2106                 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2107                 u64 *p = bb->page;
2108                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2109                 if (bb->changed) {
2110                         unsigned seq;
2111
2112 retry:
2113                         seq = read_seqbegin(&bb->lock);
2114
2115                         memset(bbp, 0xff, PAGE_SIZE);
2116
2117                         for (i = 0 ; i < bb->count ; i++) {
2118                                 u64 internal_bb = p[i];
2119                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2120                                                 | BB_LEN(internal_bb));
2121                                 bbp[i] = cpu_to_le64(store_bb);
2122                         }
2123                         bb->changed = 0;
2124                         if (read_seqretry(&bb->lock, seq))
2125                                 goto retry;
2126
2127                         bb->sector = (rdev->sb_start +
2128                                       (int)le32_to_cpu(sb->bblog_offset));
2129                         bb->size = le16_to_cpu(sb->bblog_size);
2130                 }
2131         }
2132
2133         max_dev = 0;
2134         rdev_for_each(rdev2, mddev)
2135                 if (rdev2->desc_nr+1 > max_dev)
2136                         max_dev = rdev2->desc_nr+1;
2137
2138         if (max_dev > le32_to_cpu(sb->max_dev)) {
2139                 int bmask;
2140                 sb->max_dev = cpu_to_le32(max_dev);
2141                 rdev->sb_size = max_dev * 2 + 256;
2142                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2143                 if (rdev->sb_size & bmask)
2144                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
2145         } else
2146                 max_dev = le32_to_cpu(sb->max_dev);
2147
2148         for (i=0; i<max_dev;i++)
2149                 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2150
2151         if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2152                 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2153
2154         if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2155                 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2156                         sb->feature_map |=
2157                             cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2158                 else
2159                         sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2160                 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2161                 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2162         }
2163
2164         rdev_for_each(rdev2, mddev) {
2165                 i = rdev2->desc_nr;
2166                 if (test_bit(Faulty, &rdev2->flags))
2167                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2168                 else if (test_bit(In_sync, &rdev2->flags))
2169                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2170                 else if (test_bit(Journal, &rdev2->flags))
2171                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2172                 else if (rdev2->raid_disk >= 0)
2173                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2174                 else
2175                         sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2176         }
2177
2178         sb->sb_csum = calc_sb_1_csum(sb);
2179 }
2180
2181 static sector_t super_1_choose_bm_space(sector_t dev_size)
2182 {
2183         sector_t bm_space;
2184
2185         /* if the device is bigger than 8Gig, save 64k for bitmap
2186          * usage, if bigger than 200Gig, save 128k
2187          */
2188         if (dev_size < 64*2)
2189                 bm_space = 0;
2190         else if (dev_size - 64*2 >= 200*1024*1024*2)
2191                 bm_space = 128*2;
2192         else if (dev_size - 4*2 > 8*1024*1024*2)
2193                 bm_space = 64*2;
2194         else
2195                 bm_space = 4*2;
2196         return bm_space;
2197 }
2198
2199 static unsigned long long
2200 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2201 {
2202         struct mdp_superblock_1 *sb;
2203         sector_t max_sectors;
2204         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2205                 return 0; /* component must fit device */
2206         if (rdev->data_offset != rdev->new_data_offset)
2207                 return 0; /* too confusing */
2208         if (rdev->sb_start < rdev->data_offset) {
2209                 /* minor versions 1 and 2; superblock before data */
2210                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2211                 max_sectors -= rdev->data_offset;
2212                 if (!num_sectors || num_sectors > max_sectors)
2213                         num_sectors = max_sectors;
2214         } else if (rdev->mddev->bitmap_info.offset) {
2215                 /* minor version 0 with bitmap we can't move */
2216                 return 0;
2217         } else {
2218                 /* minor version 0; superblock after data */
2219                 sector_t sb_start, bm_space;
2220                 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2221
2222                 /* 8K is for superblock */
2223                 sb_start = dev_size - 8*2;
2224                 sb_start &= ~(sector_t)(4*2 - 1);
2225
2226                 bm_space = super_1_choose_bm_space(dev_size);
2227
2228                 /* Space that can be used to store date needs to decrease
2229                  * superblock bitmap space and bad block space(4K)
2230                  */
2231                 max_sectors = sb_start - bm_space - 4*2;
2232
2233                 if (!num_sectors || num_sectors > max_sectors)
2234                         num_sectors = max_sectors;
2235         }
2236         sb = page_address(rdev->sb_page);
2237         sb->data_size = cpu_to_le64(num_sectors);
2238         sb->super_offset = cpu_to_le64(rdev->sb_start);
2239         sb->sb_csum = calc_sb_1_csum(sb);
2240         do {
2241                 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2242                                rdev->sb_page);
2243         } while (md_super_wait(rdev->mddev) < 0);
2244         return num_sectors;
2245
2246 }
2247
2248 static int
2249 super_1_allow_new_offset(struct md_rdev *rdev,
2250                          unsigned long long new_offset)
2251 {
2252         /* All necessary checks on new >= old have been done */
2253         struct bitmap *bitmap;
2254         if (new_offset >= rdev->data_offset)
2255                 return 1;
2256
2257         /* with 1.0 metadata, there is no metadata to tread on
2258          * so we can always move back */
2259         if (rdev->mddev->minor_version == 0)
2260                 return 1;
2261
2262         /* otherwise we must be sure not to step on
2263          * any metadata, so stay:
2264          * 36K beyond start of superblock
2265          * beyond end of badblocks
2266          * beyond write-intent bitmap
2267          */
2268         if (rdev->sb_start + (32+4)*2 > new_offset)
2269                 return 0;
2270         bitmap = rdev->mddev->bitmap;
2271         if (bitmap && !rdev->mddev->bitmap_info.file &&
2272             rdev->sb_start + rdev->mddev->bitmap_info.offset +
2273             bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2274                 return 0;
2275         if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2276                 return 0;
2277
2278         return 1;
2279 }
2280
2281 static struct super_type super_types[] = {
2282         [0] = {
2283                 .name   = "0.90.0",
2284                 .owner  = THIS_MODULE,
2285                 .load_super         = super_90_load,
2286                 .validate_super     = super_90_validate,
2287                 .sync_super         = super_90_sync,
2288                 .rdev_size_change   = super_90_rdev_size_change,
2289                 .allow_new_offset   = super_90_allow_new_offset,
2290         },
2291         [1] = {
2292                 .name   = "md-1",
2293                 .owner  = THIS_MODULE,
2294                 .load_super         = super_1_load,
2295                 .validate_super     = super_1_validate,
2296                 .sync_super         = super_1_sync,
2297                 .rdev_size_change   = super_1_rdev_size_change,
2298                 .allow_new_offset   = super_1_allow_new_offset,
2299         },
2300 };
2301
2302 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2303 {
2304         if (mddev->sync_super) {
2305                 mddev->sync_super(mddev, rdev);
2306                 return;
2307         }
2308
2309         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2310
2311         super_types[mddev->major_version].sync_super(mddev, rdev);
2312 }
2313
2314 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2315 {
2316         struct md_rdev *rdev, *rdev2;
2317
2318         rcu_read_lock();
2319         rdev_for_each_rcu(rdev, mddev1) {
2320                 if (test_bit(Faulty, &rdev->flags) ||
2321                     test_bit(Journal, &rdev->flags) ||
2322                     rdev->raid_disk == -1)
2323                         continue;
2324                 rdev_for_each_rcu(rdev2, mddev2) {
2325                         if (test_bit(Faulty, &rdev2->flags) ||
2326                             test_bit(Journal, &rdev2->flags) ||
2327                             rdev2->raid_disk == -1)
2328                                 continue;
2329                         if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2330                                 rcu_read_unlock();
2331                                 return 1;
2332                         }
2333                 }
2334         }
2335         rcu_read_unlock();
2336         return 0;
2337 }
2338
2339 static LIST_HEAD(pending_raid_disks);
2340
2341 /*
2342  * Try to register data integrity profile for an mddev
2343  *
2344  * This is called when an array is started and after a disk has been kicked
2345  * from the array. It only succeeds if all working and active component devices
2346  * are integrity capable with matching profiles.
2347  */
2348 int md_integrity_register(struct mddev *mddev)
2349 {
2350         struct md_rdev *rdev, *reference = NULL;
2351
2352         if (list_empty(&mddev->disks))
2353                 return 0; /* nothing to do */
2354         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2355                 return 0; /* shouldn't register, or already is */
2356         rdev_for_each(rdev, mddev) {
2357                 /* skip spares and non-functional disks */
2358                 if (test_bit(Faulty, &rdev->flags))
2359                         continue;
2360                 if (rdev->raid_disk < 0)
2361                         continue;
2362                 if (!reference) {
2363                         /* Use the first rdev as the reference */
2364                         reference = rdev;
2365                         continue;
2366                 }
2367                 /* does this rdev's profile match the reference profile? */
2368                 if (blk_integrity_compare(reference->bdev->bd_disk,
2369                                 rdev->bdev->bd_disk) < 0)
2370                         return -EINVAL;
2371         }
2372         if (!reference || !bdev_get_integrity(reference->bdev))
2373                 return 0;
2374         /*
2375          * All component devices are integrity capable and have matching
2376          * profiles, register the common profile for the md device.
2377          */
2378         blk_integrity_register(mddev->gendisk,
2379                                bdev_get_integrity(reference->bdev));
2380
2381         pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2382         if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2383                 pr_err("md: failed to create integrity pool for %s\n",
2384                        mdname(mddev));
2385                 return -EINVAL;
2386         }
2387         return 0;
2388 }
2389 EXPORT_SYMBOL(md_integrity_register);
2390
2391 /*
2392  * Attempt to add an rdev, but only if it is consistent with the current
2393  * integrity profile
2394  */
2395 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2396 {
2397         struct blk_integrity *bi_mddev;
2398         char name[BDEVNAME_SIZE];
2399
2400         if (!mddev->gendisk)
2401                 return 0;
2402
2403         bi_mddev = blk_get_integrity(mddev->gendisk);
2404
2405         if (!bi_mddev) /* nothing to do */
2406                 return 0;
2407
2408         if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2409                 pr_err("%s: incompatible integrity profile for %s\n",
2410                        mdname(mddev), bdevname(rdev->bdev, name));
2411                 return -ENXIO;
2412         }
2413
2414         return 0;
2415 }
2416 EXPORT_SYMBOL(md_integrity_add_rdev);
2417
2418 static bool rdev_read_only(struct md_rdev *rdev)
2419 {
2420         return bdev_read_only(rdev->bdev) ||
2421                 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2422 }
2423
2424 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2425 {
2426         char b[BDEVNAME_SIZE];
2427         int err;
2428
2429         /* prevent duplicates */
2430         if (find_rdev(mddev, rdev->bdev->bd_dev))
2431                 return -EEXIST;
2432
2433         if (rdev_read_only(rdev) && mddev->pers)
2434                 return -EROFS;
2435
2436         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2437         if (!test_bit(Journal, &rdev->flags) &&
2438             rdev->sectors &&
2439             (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2440                 if (mddev->pers) {
2441                         /* Cannot change size, so fail
2442                          * If mddev->level <= 0, then we don't care
2443                          * about aligning sizes (e.g. linear)
2444                          */
2445                         if (mddev->level > 0)
2446                                 return -ENOSPC;
2447                 } else
2448                         mddev->dev_sectors = rdev->sectors;
2449         }
2450
2451         /* Verify rdev->desc_nr is unique.
2452          * If it is -1, assign a free number, else
2453          * check number is not in use
2454          */
2455         rcu_read_lock();
2456         if (rdev->desc_nr < 0) {
2457                 int choice = 0;
2458                 if (mddev->pers)
2459                         choice = mddev->raid_disks;
2460                 while (md_find_rdev_nr_rcu(mddev, choice))
2461                         choice++;
2462                 rdev->desc_nr = choice;
2463         } else {
2464                 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2465                         rcu_read_unlock();
2466                         return -EBUSY;
2467                 }
2468         }
2469         rcu_read_unlock();
2470         if (!test_bit(Journal, &rdev->flags) &&
2471             mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2472                 pr_warn("md: %s: array is limited to %d devices\n",
2473                         mdname(mddev), mddev->max_disks);
2474                 return -EBUSY;
2475         }
2476         bdevname(rdev->bdev,b);
2477         strreplace(b, '/', '!');
2478
2479         rdev->mddev = mddev;
2480         pr_debug("md: bind<%s>\n", b);
2481
2482         if (mddev->raid_disks)
2483                 mddev_create_serial_pool(mddev, rdev, false);
2484
2485         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2486                 goto fail;
2487
2488         /* failure here is OK */
2489         err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2490         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2491         rdev->sysfs_unack_badblocks =
2492                 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2493         rdev->sysfs_badblocks =
2494                 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2495
2496         list_add_rcu(&rdev->same_set, &mddev->disks);
2497         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2498
2499         /* May as well allow recovery to be retried once */
2500         mddev->recovery_disabled++;
2501
2502         return 0;
2503
2504  fail:
2505         pr_warn("md: failed to register dev-%s for %s\n",
2506                 b, mdname(mddev));
2507         return err;
2508 }
2509
2510 static void rdev_delayed_delete(struct work_struct *ws)
2511 {
2512         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2513         kobject_del(&rdev->kobj);
2514         kobject_put(&rdev->kobj);
2515 }
2516
2517 static void unbind_rdev_from_array(struct md_rdev *rdev)
2518 {
2519         char b[BDEVNAME_SIZE];
2520
2521         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2522         list_del_rcu(&rdev->same_set);
2523         pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2524         mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2525         rdev->mddev = NULL;
2526         sysfs_remove_link(&rdev->kobj, "block");
2527         sysfs_put(rdev->sysfs_state);
2528         sysfs_put(rdev->sysfs_unack_badblocks);
2529         sysfs_put(rdev->sysfs_badblocks);
2530         rdev->sysfs_state = NULL;
2531         rdev->sysfs_unack_badblocks = NULL;
2532         rdev->sysfs_badblocks = NULL;
2533         rdev->badblocks.count = 0;
2534         /* We need to delay this, otherwise we can deadlock when
2535          * writing to 'remove' to "dev/state".  We also need
2536          * to delay it due to rcu usage.
2537          */
2538         synchronize_rcu();
2539         INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2540         kobject_get(&rdev->kobj);
2541         queue_work(md_rdev_misc_wq, &rdev->del_work);
2542 }
2543
2544 /*
2545  * prevent the device from being mounted, repartitioned or
2546  * otherwise reused by a RAID array (or any other kernel
2547  * subsystem), by bd_claiming the device.
2548  */
2549 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2550 {
2551         int err = 0;
2552         struct block_device *bdev;
2553
2554         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2555                                  shared ? (struct md_rdev *)lock_rdev : rdev);
2556         if (IS_ERR(bdev)) {
2557                 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2558                         MAJOR(dev), MINOR(dev));
2559                 return PTR_ERR(bdev);
2560         }
2561         rdev->bdev = bdev;
2562         return err;
2563 }
2564
2565 static void unlock_rdev(struct md_rdev *rdev)
2566 {
2567         struct block_device *bdev = rdev->bdev;
2568         rdev->bdev = NULL;
2569         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2570 }
2571
2572 void md_autodetect_dev(dev_t dev);
2573
2574 static void export_rdev(struct md_rdev *rdev)
2575 {
2576         char b[BDEVNAME_SIZE];
2577
2578         pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2579         md_rdev_clear(rdev);
2580 #ifndef MODULE
2581         if (test_bit(AutoDetected, &rdev->flags))
2582                 md_autodetect_dev(rdev->bdev->bd_dev);
2583 #endif
2584         unlock_rdev(rdev);
2585         kobject_put(&rdev->kobj);
2586 }
2587
2588 void md_kick_rdev_from_array(struct md_rdev *rdev)
2589 {
2590         unbind_rdev_from_array(rdev);
2591         export_rdev(rdev);
2592 }
2593 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2594
2595 static void export_array(struct mddev *mddev)
2596 {
2597         struct md_rdev *rdev;
2598
2599         while (!list_empty(&mddev->disks)) {
2600                 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2601                                         same_set);
2602                 md_kick_rdev_from_array(rdev);
2603         }
2604         mddev->raid_disks = 0;
2605         mddev->major_version = 0;
2606 }
2607
2608 static bool set_in_sync(struct mddev *mddev)
2609 {
2610         lockdep_assert_held(&mddev->lock);
2611         if (!mddev->in_sync) {
2612                 mddev->sync_checkers++;
2613                 spin_unlock(&mddev->lock);
2614                 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2615                 spin_lock(&mddev->lock);
2616                 if (!mddev->in_sync &&
2617                     percpu_ref_is_zero(&mddev->writes_pending)) {
2618                         mddev->in_sync = 1;
2619                         /*
2620                          * Ensure ->in_sync is visible before we clear
2621                          * ->sync_checkers.
2622                          */
2623                         smp_mb();
2624                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2625                         sysfs_notify_dirent_safe(mddev->sysfs_state);
2626                 }
2627                 if (--mddev->sync_checkers == 0)
2628                         percpu_ref_switch_to_percpu(&mddev->writes_pending);
2629         }
2630         if (mddev->safemode == 1)
2631                 mddev->safemode = 0;
2632         return mddev->in_sync;
2633 }
2634
2635 static void sync_sbs(struct mddev *mddev, int nospares)
2636 {
2637         /* Update each superblock (in-memory image), but
2638          * if we are allowed to, skip spares which already
2639          * have the right event counter, or have one earlier
2640          * (which would mean they aren't being marked as dirty
2641          * with the rest of the array)
2642          */
2643         struct md_rdev *rdev;
2644         rdev_for_each(rdev, mddev) {
2645                 if (rdev->sb_events == mddev->events ||
2646                     (nospares &&
2647                      rdev->raid_disk < 0 &&
2648                      rdev->sb_events+1 == mddev->events)) {
2649                         /* Don't update this superblock */
2650                         rdev->sb_loaded = 2;
2651                 } else {
2652                         sync_super(mddev, rdev);
2653                         rdev->sb_loaded = 1;
2654                 }
2655         }
2656 }
2657
2658 static bool does_sb_need_changing(struct mddev *mddev)
2659 {
2660         struct md_rdev *rdev;
2661         struct mdp_superblock_1 *sb;
2662         int role;
2663
2664         /* Find a good rdev */
2665         rdev_for_each(rdev, mddev)
2666                 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2667                         break;
2668
2669         /* No good device found. */
2670         if (!rdev)
2671                 return false;
2672
2673         sb = page_address(rdev->sb_page);
2674         /* Check if a device has become faulty or a spare become active */
2675         rdev_for_each(rdev, mddev) {
2676                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2677                 /* Device activated? */
2678                 if (role == 0xffff && rdev->raid_disk >=0 &&
2679                     !test_bit(Faulty, &rdev->flags))
2680                         return true;
2681                 /* Device turned faulty? */
2682                 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2683                         return true;
2684         }
2685
2686         /* Check if any mddev parameters have changed */
2687         if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2688             (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2689             (mddev->layout != le32_to_cpu(sb->layout)) ||
2690             (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2691             (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2692                 return true;
2693
2694         return false;
2695 }
2696
2697 void md_update_sb(struct mddev *mddev, int force_change)
2698 {
2699         struct md_rdev *rdev;
2700         int sync_req;
2701         int nospares = 0;
2702         int any_badblocks_changed = 0;
2703         int ret = -1;
2704
2705         if (mddev->ro) {
2706                 if (force_change)
2707                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2708                 return;
2709         }
2710
2711 repeat:
2712         if (mddev_is_clustered(mddev)) {
2713                 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2714                         force_change = 1;
2715                 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2716                         nospares = 1;
2717                 ret = md_cluster_ops->metadata_update_start(mddev);
2718                 /* Has someone else has updated the sb */
2719                 if (!does_sb_need_changing(mddev)) {
2720                         if (ret == 0)
2721                                 md_cluster_ops->metadata_update_cancel(mddev);
2722                         bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2723                                                          BIT(MD_SB_CHANGE_DEVS) |
2724                                                          BIT(MD_SB_CHANGE_CLEAN));
2725                         return;
2726                 }
2727         }
2728
2729         /*
2730          * First make sure individual recovery_offsets are correct
2731          * curr_resync_completed can only be used during recovery.
2732          * During reshape/resync it might use array-addresses rather
2733          * that device addresses.
2734          */
2735         rdev_for_each(rdev, mddev) {
2736                 if (rdev->raid_disk >= 0 &&
2737                     mddev->delta_disks >= 0 &&
2738                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2739                     test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2740                     !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2741                     !test_bit(Journal, &rdev->flags) &&
2742                     !test_bit(In_sync, &rdev->flags) &&
2743                     mddev->curr_resync_completed > rdev->recovery_offset)
2744                                 rdev->recovery_offset = mddev->curr_resync_completed;
2745
2746         }
2747         if (!mddev->persistent) {
2748                 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2749                 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2750                 if (!mddev->external) {
2751                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2752                         rdev_for_each(rdev, mddev) {
2753                                 if (rdev->badblocks.changed) {
2754                                         rdev->badblocks.changed = 0;
2755                                         ack_all_badblocks(&rdev->badblocks);
2756                                         md_error(mddev, rdev);
2757                                 }
2758                                 clear_bit(Blocked, &rdev->flags);
2759                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2760                                 wake_up(&rdev->blocked_wait);
2761                         }
2762                 }
2763                 wake_up(&mddev->sb_wait);
2764                 return;
2765         }
2766
2767         spin_lock(&mddev->lock);
2768
2769         mddev->utime = ktime_get_real_seconds();
2770
2771         if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2772                 force_change = 1;
2773         if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2774                 /* just a clean<-> dirty transition, possibly leave spares alone,
2775                  * though if events isn't the right even/odd, we will have to do
2776                  * spares after all
2777                  */
2778                 nospares = 1;
2779         if (force_change)
2780                 nospares = 0;
2781         if (mddev->degraded)
2782                 /* If the array is degraded, then skipping spares is both
2783                  * dangerous and fairly pointless.
2784                  * Dangerous because a device that was removed from the array
2785                  * might have a event_count that still looks up-to-date,
2786                  * so it can be re-added without a resync.
2787                  * Pointless because if there are any spares to skip,
2788                  * then a recovery will happen and soon that array won't
2789                  * be degraded any more and the spare can go back to sleep then.
2790                  */
2791                 nospares = 0;
2792
2793         sync_req = mddev->in_sync;
2794
2795         /* If this is just a dirty<->clean transition, and the array is clean
2796          * and 'events' is odd, we can roll back to the previous clean state */
2797         if (nospares
2798             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2799             && mddev->can_decrease_events
2800             && mddev->events != 1) {
2801                 mddev->events--;
2802                 mddev->can_decrease_events = 0;
2803         } else {
2804                 /* otherwise we have to go forward and ... */
2805                 mddev->events ++;
2806                 mddev->can_decrease_events = nospares;
2807         }
2808
2809         /*
2810          * This 64-bit counter should never wrap.
2811          * Either we are in around ~1 trillion A.C., assuming
2812          * 1 reboot per second, or we have a bug...
2813          */
2814         WARN_ON(mddev->events == 0);
2815
2816         rdev_for_each(rdev, mddev) {
2817                 if (rdev->badblocks.changed)
2818                         any_badblocks_changed++;
2819                 if (test_bit(Faulty, &rdev->flags))
2820                         set_bit(FaultRecorded, &rdev->flags);
2821         }
2822
2823         sync_sbs(mddev, nospares);
2824         spin_unlock(&mddev->lock);
2825
2826         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2827                  mdname(mddev), mddev->in_sync);
2828
2829         if (mddev->queue)
2830                 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2831 rewrite:
2832         md_bitmap_update_sb(mddev->bitmap);
2833         rdev_for_each(rdev, mddev) {
2834                 char b[BDEVNAME_SIZE];
2835
2836                 if (rdev->sb_loaded != 1)
2837                         continue; /* no noise on spare devices */
2838
2839                 if (!test_bit(Faulty, &rdev->flags)) {
2840                         md_super_write(mddev,rdev,
2841                                        rdev->sb_start, rdev->sb_size,
2842                                        rdev->sb_page);
2843                         pr_debug("md: (write) %s's sb offset: %llu\n",
2844                                  bdevname(rdev->bdev, b),
2845                                  (unsigned long long)rdev->sb_start);
2846                         rdev->sb_events = mddev->events;
2847                         if (rdev->badblocks.size) {
2848                                 md_super_write(mddev, rdev,
2849                                                rdev->badblocks.sector,
2850                                                rdev->badblocks.size << 9,
2851                                                rdev->bb_page);
2852                                 rdev->badblocks.size = 0;
2853                         }
2854
2855                 } else
2856                         pr_debug("md: %s (skipping faulty)\n",
2857                                  bdevname(rdev->bdev, b));
2858
2859                 if (mddev->level == LEVEL_MULTIPATH)
2860                         /* only need to write one superblock... */
2861                         break;
2862         }
2863         if (md_super_wait(mddev) < 0)
2864                 goto rewrite;
2865         /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2866
2867         if (mddev_is_clustered(mddev) && ret == 0)
2868                 md_cluster_ops->metadata_update_finish(mddev);
2869
2870         if (mddev->in_sync != sync_req ||
2871             !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2872                                BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2873                 /* have to write it out again */
2874                 goto repeat;
2875         wake_up(&mddev->sb_wait);
2876         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2877                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2878
2879         rdev_for_each(rdev, mddev) {
2880                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2881                         clear_bit(Blocked, &rdev->flags);
2882
2883                 if (any_badblocks_changed)
2884                         ack_all_badblocks(&rdev->badblocks);
2885                 clear_bit(BlockedBadBlocks, &rdev->flags);
2886                 wake_up(&rdev->blocked_wait);
2887         }
2888 }
2889 EXPORT_SYMBOL(md_update_sb);
2890
2891 static int add_bound_rdev(struct md_rdev *rdev)
2892 {
2893         struct mddev *mddev = rdev->mddev;
2894         int err = 0;
2895         bool add_journal = test_bit(Journal, &rdev->flags);
2896
2897         if (!mddev->pers->hot_remove_disk || add_journal) {
2898                 /* If there is hot_add_disk but no hot_remove_disk
2899                  * then added disks for geometry changes,
2900                  * and should be added immediately.
2901                  */
2902                 super_types[mddev->major_version].
2903                         validate_super(mddev, rdev);
2904                 if (add_journal)
2905                         mddev_suspend(mddev);
2906                 err = mddev->pers->hot_add_disk(mddev, rdev);
2907                 if (add_journal)
2908                         mddev_resume(mddev);
2909                 if (err) {
2910                         md_kick_rdev_from_array(rdev);
2911                         return err;
2912                 }
2913         }
2914         sysfs_notify_dirent_safe(rdev->sysfs_state);
2915
2916         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2917         if (mddev->degraded)
2918                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2919         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2920         md_new_event(mddev);
2921         md_wakeup_thread(mddev->thread);
2922         return 0;
2923 }
2924
2925 /* words written to sysfs files may, or may not, be \n terminated.
2926  * We want to accept with case. For this we use cmd_match.
2927  */
2928 static int cmd_match(const char *cmd, const char *str)
2929 {
2930         /* See if cmd, written into a sysfs file, matches
2931          * str.  They must either be the same, or cmd can
2932          * have a trailing newline
2933          */
2934         while (*cmd && *str && *cmd == *str) {
2935                 cmd++;
2936                 str++;
2937         }
2938         if (*cmd == '\n')
2939                 cmd++;
2940         if (*str || *cmd)
2941                 return 0;
2942         return 1;
2943 }
2944
2945 struct rdev_sysfs_entry {
2946         struct attribute attr;
2947         ssize_t (*show)(struct md_rdev *, char *);
2948         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2949 };
2950
2951 static ssize_t
2952 state_show(struct md_rdev *rdev, char *page)
2953 {
2954         char *sep = ",";
2955         size_t len = 0;
2956         unsigned long flags = READ_ONCE(rdev->flags);
2957
2958         if (test_bit(Faulty, &flags) ||
2959             (!test_bit(ExternalBbl, &flags) &&
2960             rdev->badblocks.unacked_exist))
2961                 len += sprintf(page+len, "faulty%s", sep);
2962         if (test_bit(In_sync, &flags))
2963                 len += sprintf(page+len, "in_sync%s", sep);
2964         if (test_bit(Journal, &flags))
2965                 len += sprintf(page+len, "journal%s", sep);
2966         if (test_bit(WriteMostly, &flags))
2967                 len += sprintf(page+len, "write_mostly%s", sep);
2968         if (test_bit(Blocked, &flags) ||
2969             (rdev->badblocks.unacked_exist
2970              && !test_bit(Faulty, &flags)))
2971                 len += sprintf(page+len, "blocked%s", sep);
2972         if (!test_bit(Faulty, &flags) &&
2973             !test_bit(Journal, &flags) &&
2974             !test_bit(In_sync, &flags))
2975                 len += sprintf(page+len, "spare%s", sep);
2976         if (test_bit(WriteErrorSeen, &flags))
2977                 len += sprintf(page+len, "write_error%s", sep);
2978         if (test_bit(WantReplacement, &flags))
2979                 len += sprintf(page+len, "want_replacement%s", sep);
2980         if (test_bit(Replacement, &flags))
2981                 len += sprintf(page+len, "replacement%s", sep);
2982         if (test_bit(ExternalBbl, &flags))
2983                 len += sprintf(page+len, "external_bbl%s", sep);
2984         if (test_bit(FailFast, &flags))
2985                 len += sprintf(page+len, "failfast%s", sep);
2986
2987         if (len)
2988                 len -= strlen(sep);
2989
2990         return len+sprintf(page+len, "\n");
2991 }
2992
2993 static ssize_t
2994 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2995 {
2996         /* can write
2997          *  faulty  - simulates an error
2998          *  remove  - disconnects the device
2999          *  writemostly - sets write_mostly
3000          *  -writemostly - clears write_mostly
3001          *  blocked - sets the Blocked flags
3002          *  -blocked - clears the Blocked and possibly simulates an error
3003          *  insync - sets Insync providing device isn't active
3004          *  -insync - clear Insync for a device with a slot assigned,
3005          *            so that it gets rebuilt based on bitmap
3006          *  write_error - sets WriteErrorSeen
3007          *  -write_error - clears WriteErrorSeen
3008          *  {,-}failfast - set/clear FailFast
3009          */
3010         int err = -EINVAL;
3011         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
3012                 md_error(rdev->mddev, rdev);
3013                 if (test_bit(Faulty, &rdev->flags))
3014                         err = 0;
3015                 else
3016                         err = -EBUSY;
3017         } else if (cmd_match(buf, "remove")) {
3018                 if (rdev->mddev->pers) {
3019                         clear_bit(Blocked, &rdev->flags);
3020                         remove_and_add_spares(rdev->mddev, rdev);
3021                 }
3022                 if (rdev->raid_disk >= 0)
3023                         err = -EBUSY;
3024                 else {
3025                         struct mddev *mddev = rdev->mddev;
3026                         err = 0;
3027                         if (mddev_is_clustered(mddev))
3028                                 err = md_cluster_ops->remove_disk(mddev, rdev);
3029
3030                         if (err == 0) {
3031                                 md_kick_rdev_from_array(rdev);
3032                                 if (mddev->pers) {
3033                                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3034                                         md_wakeup_thread(mddev->thread);
3035                                 }
3036                                 md_new_event(mddev);
3037                         }
3038                 }
3039         } else if (cmd_match(buf, "writemostly")) {
3040                 set_bit(WriteMostly, &rdev->flags);
3041                 mddev_create_serial_pool(rdev->mddev, rdev, false);
3042                 err = 0;
3043         } else if (cmd_match(buf, "-writemostly")) {
3044                 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3045                 clear_bit(WriteMostly, &rdev->flags);
3046                 err = 0;
3047         } else if (cmd_match(buf, "blocked")) {
3048                 set_bit(Blocked, &rdev->flags);
3049                 err = 0;
3050         } else if (cmd_match(buf, "-blocked")) {
3051                 if (!test_bit(Faulty, &rdev->flags) &&
3052                     !test_bit(ExternalBbl, &rdev->flags) &&
3053                     rdev->badblocks.unacked_exist) {
3054                         /* metadata handler doesn't understand badblocks,
3055                          * so we need to fail the device
3056                          */
3057                         md_error(rdev->mddev, rdev);
3058                 }
3059                 clear_bit(Blocked, &rdev->flags);
3060                 clear_bit(BlockedBadBlocks, &rdev->flags);
3061                 wake_up(&rdev->blocked_wait);
3062                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3063                 md_wakeup_thread(rdev->mddev->thread);
3064
3065                 err = 0;
3066         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3067                 set_bit(In_sync, &rdev->flags);
3068                 err = 0;
3069         } else if (cmd_match(buf, "failfast")) {
3070                 set_bit(FailFast, &rdev->flags);
3071                 err = 0;
3072         } else if (cmd_match(buf, "-failfast")) {
3073                 clear_bit(FailFast, &rdev->flags);
3074                 err = 0;
3075         } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3076                    !test_bit(Journal, &rdev->flags)) {
3077                 if (rdev->mddev->pers == NULL) {
3078                         clear_bit(In_sync, &rdev->flags);
3079                         rdev->saved_raid_disk = rdev->raid_disk;
3080                         rdev->raid_disk = -1;
3081                         err = 0;
3082                 }
3083         } else if (cmd_match(buf, "write_error")) {
3084                 set_bit(WriteErrorSeen, &rdev->flags);
3085                 err = 0;
3086         } else if (cmd_match(buf, "-write_error")) {
3087                 clear_bit(WriteErrorSeen, &rdev->flags);
3088                 err = 0;
3089         } else if (cmd_match(buf, "want_replacement")) {
3090                 /* Any non-spare device that is not a replacement can
3091                  * become want_replacement at any time, but we then need to
3092                  * check if recovery is needed.
3093                  */
3094                 if (rdev->raid_disk >= 0 &&
3095                     !test_bit(Journal, &rdev->flags) &&
3096                     !test_bit(Replacement, &rdev->flags))
3097                         set_bit(WantReplacement, &rdev->flags);
3098                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3099                 md_wakeup_thread(rdev->mddev->thread);
3100                 err = 0;
3101         } else if (cmd_match(buf, "-want_replacement")) {
3102                 /* Clearing 'want_replacement' is always allowed.
3103                  * Once replacements starts it is too late though.
3104                  */
3105                 err = 0;
3106                 clear_bit(WantReplacement, &rdev->flags);
3107         } else if (cmd_match(buf, "replacement")) {
3108                 /* Can only set a device as a replacement when array has not
3109                  * yet been started.  Once running, replacement is automatic
3110                  * from spares, or by assigning 'slot'.
3111                  */
3112                 if (rdev->mddev->pers)
3113                         err = -EBUSY;
3114                 else {
3115                         set_bit(Replacement, &rdev->flags);
3116                         err = 0;
3117                 }
3118         } else if (cmd_match(buf, "-replacement")) {
3119                 /* Similarly, can only clear Replacement before start */
3120                 if (rdev->mddev->pers)
3121                         err = -EBUSY;
3122                 else {
3123                         clear_bit(Replacement, &rdev->flags);
3124                         err = 0;
3125                 }
3126         } else if (cmd_match(buf, "re-add")) {
3127                 if (!rdev->mddev->pers)
3128                         err = -EINVAL;
3129                 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3130                                 rdev->saved_raid_disk >= 0) {
3131                         /* clear_bit is performed _after_ all the devices
3132                          * have their local Faulty bit cleared. If any writes
3133                          * happen in the meantime in the local node, they
3134                          * will land in the local bitmap, which will be synced
3135                          * by this node eventually
3136                          */
3137                         if (!mddev_is_clustered(rdev->mddev) ||
3138                             (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3139                                 clear_bit(Faulty, &rdev->flags);
3140                                 err = add_bound_rdev(rdev);
3141                         }
3142                 } else
3143                         err = -EBUSY;
3144         } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3145                 set_bit(ExternalBbl, &rdev->flags);
3146                 rdev->badblocks.shift = 0;
3147                 err = 0;
3148         } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3149                 clear_bit(ExternalBbl, &rdev->flags);
3150                 err = 0;
3151         }
3152         if (!err)
3153                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3154         return err ? err : len;
3155 }
3156 static struct rdev_sysfs_entry rdev_state =
3157 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3158
3159 static ssize_t
3160 errors_show(struct md_rdev *rdev, char *page)
3161 {
3162         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3163 }
3164
3165 static ssize_t
3166 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3167 {
3168         unsigned int n;
3169         int rv;
3170
3171         rv = kstrtouint(buf, 10, &n);
3172         if (rv < 0)
3173                 return rv;
3174         atomic_set(&rdev->corrected_errors, n);
3175         return len;
3176 }
3177 static struct rdev_sysfs_entry rdev_errors =
3178 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3179
3180 static ssize_t
3181 slot_show(struct md_rdev *rdev, char *page)
3182 {
3183         if (test_bit(Journal, &rdev->flags))
3184                 return sprintf(page, "journal\n");
3185         else if (rdev->raid_disk < 0)
3186                 return sprintf(page, "none\n");
3187         else
3188                 return sprintf(page, "%d\n", rdev->raid_disk);
3189 }
3190
3191 static ssize_t
3192 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3193 {
3194         int slot;
3195         int err;
3196
3197         if (test_bit(Journal, &rdev->flags))
3198                 return -EBUSY;
3199         if (strncmp(buf, "none", 4)==0)
3200                 slot = -1;
3201         else {
3202                 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3203                 if (err < 0)
3204                         return err;
3205         }
3206         if (rdev->mddev->pers && slot == -1) {
3207                 /* Setting 'slot' on an active array requires also
3208                  * updating the 'rd%d' link, and communicating
3209                  * with the personality with ->hot_*_disk.
3210                  * For now we only support removing
3211                  * failed/spare devices.  This normally happens automatically,
3212                  * but not when the metadata is externally managed.
3213                  */
3214                 if (rdev->raid_disk == -1)
3215                         return -EEXIST;
3216                 /* personality does all needed checks */
3217                 if (rdev->mddev->pers->hot_remove_disk == NULL)
3218                         return -EINVAL;
3219                 clear_bit(Blocked, &rdev->flags);
3220                 remove_and_add_spares(rdev->mddev, rdev);
3221                 if (rdev->raid_disk >= 0)
3222                         return -EBUSY;
3223                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3224                 md_wakeup_thread(rdev->mddev->thread);
3225         } else if (rdev->mddev->pers) {
3226                 /* Activating a spare .. or possibly reactivating
3227                  * if we ever get bitmaps working here.
3228                  */
3229                 int err;
3230
3231                 if (rdev->raid_disk != -1)
3232                         return -EBUSY;
3233
3234                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3235                         return -EBUSY;
3236
3237                 if (rdev->mddev->pers->hot_add_disk == NULL)
3238                         return -EINVAL;
3239
3240                 if (slot >= rdev->mddev->raid_disks &&
3241                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3242                         return -ENOSPC;
3243
3244                 rdev->raid_disk = slot;
3245                 if (test_bit(In_sync, &rdev->flags))
3246                         rdev->saved_raid_disk = slot;
3247                 else
3248                         rdev->saved_raid_disk = -1;
3249                 clear_bit(In_sync, &rdev->flags);
3250                 clear_bit(Bitmap_sync, &rdev->flags);
3251                 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3252                 if (err) {
3253                         rdev->raid_disk = -1;
3254                         return err;
3255                 } else
3256                         sysfs_notify_dirent_safe(rdev->sysfs_state);
3257                 /* failure here is OK */;
3258                 sysfs_link_rdev(rdev->mddev, rdev);
3259                 /* don't wakeup anyone, leave that to userspace. */
3260         } else {
3261                 if (slot >= rdev->mddev->raid_disks &&
3262                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3263                         return -ENOSPC;
3264                 rdev->raid_disk = slot;
3265                 /* assume it is working */
3266                 clear_bit(Faulty, &rdev->flags);
3267                 clear_bit(WriteMostly, &rdev->flags);
3268                 set_bit(In_sync, &rdev->flags);
3269                 sysfs_notify_dirent_safe(rdev->sysfs_state);
3270         }
3271         return len;
3272 }
3273
3274 static struct rdev_sysfs_entry rdev_slot =
3275 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3276
3277 static ssize_t
3278 offset_show(struct md_rdev *rdev, char *page)
3279 {
3280         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3281 }
3282
3283 static ssize_t
3284 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3285 {
3286         unsigned long long offset;
3287         if (kstrtoull(buf, 10, &offset) < 0)
3288                 return -EINVAL;
3289         if (rdev->mddev->pers && rdev->raid_disk >= 0)
3290                 return -EBUSY;
3291         if (rdev->sectors && rdev->mddev->external)
3292                 /* Must set offset before size, so overlap checks
3293                  * can be sane */
3294                 return -EBUSY;
3295         rdev->data_offset = offset;
3296         rdev->new_data_offset = offset;
3297         return len;
3298 }
3299
3300 static struct rdev_sysfs_entry rdev_offset =
3301 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3302
3303 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3304 {
3305         return sprintf(page, "%llu\n",
3306                        (unsigned long long)rdev->new_data_offset);
3307 }
3308
3309 static ssize_t new_offset_store(struct md_rdev *rdev,
3310                                 const char *buf, size_t len)
3311 {
3312         unsigned long long new_offset;
3313         struct mddev *mddev = rdev->mddev;
3314
3315         if (kstrtoull(buf, 10, &new_offset) < 0)
3316                 return -EINVAL;
3317
3318         if (mddev->sync_thread ||
3319             test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3320                 return -EBUSY;
3321         if (new_offset == rdev->data_offset)
3322                 /* reset is always permitted */
3323                 ;
3324         else if (new_offset > rdev->data_offset) {
3325                 /* must not push array size beyond rdev_sectors */
3326                 if (new_offset - rdev->data_offset
3327                     + mddev->dev_sectors > rdev->sectors)
3328                                 return -E2BIG;
3329         }
3330         /* Metadata worries about other space details. */
3331
3332         /* decreasing the offset is inconsistent with a backwards
3333          * reshape.
3334          */
3335         if (new_offset < rdev->data_offset &&
3336             mddev->reshape_backwards)
3337                 return -EINVAL;
3338         /* Increasing offset is inconsistent with forwards
3339          * reshape.  reshape_direction should be set to
3340          * 'backwards' first.
3341          */
3342         if (new_offset > rdev->data_offset &&
3343             !mddev->reshape_backwards)
3344                 return -EINVAL;
3345
3346         if (mddev->pers && mddev->persistent &&
3347             !super_types[mddev->major_version]
3348             .allow_new_offset(rdev, new_offset))
3349                 return -E2BIG;
3350         rdev->new_data_offset = new_offset;
3351         if (new_offset > rdev->data_offset)
3352                 mddev->reshape_backwards = 1;
3353         else if (new_offset < rdev->data_offset)
3354                 mddev->reshape_backwards = 0;
3355
3356         return len;
3357 }
3358 static struct rdev_sysfs_entry rdev_new_offset =
3359 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3360
3361 static ssize_t
3362 rdev_size_show(struct md_rdev *rdev, char *page)
3363 {
3364         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3365 }
3366
3367 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3368 {
3369         /* check if two start/length pairs overlap */
3370         if (s1+l1 <= s2)
3371                 return 0;
3372         if (s2+l2 <= s1)
3373                 return 0;
3374         return 1;
3375 }
3376
3377 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3378 {
3379         unsigned long long blocks;
3380         sector_t new;
3381
3382         if (kstrtoull(buf, 10, &blocks) < 0)
3383                 return -EINVAL;
3384
3385         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3386                 return -EINVAL; /* sector conversion overflow */
3387
3388         new = blocks * 2;
3389         if (new != blocks * 2)
3390                 return -EINVAL; /* unsigned long long to sector_t overflow */
3391
3392         *sectors = new;
3393         return 0;
3394 }
3395
3396 static ssize_t
3397 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3398 {
3399         struct mddev *my_mddev = rdev->mddev;
3400         sector_t oldsectors = rdev->sectors;
3401         sector_t sectors;
3402
3403         if (test_bit(Journal, &rdev->flags))
3404                 return -EBUSY;
3405         if (strict_blocks_to_sectors(buf, &sectors) < 0)
3406                 return -EINVAL;
3407         if (rdev->data_offset != rdev->new_data_offset)
3408                 return -EINVAL; /* too confusing */
3409         if (my_mddev->pers && rdev->raid_disk >= 0) {
3410                 if (my_mddev->persistent) {
3411                         sectors = super_types[my_mddev->major_version].
3412                                 rdev_size_change(rdev, sectors);
3413                         if (!sectors)
3414                                 return -EBUSY;
3415                 } else if (!sectors)
3416                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3417                                 rdev->data_offset;
3418                 if (!my_mddev->pers->resize)
3419                         /* Cannot change size for RAID0 or Linear etc */
3420                         return -EINVAL;
3421         }
3422         if (sectors < my_mddev->dev_sectors)
3423                 return -EINVAL; /* component must fit device */
3424
3425         rdev->sectors = sectors;
3426         if (sectors > oldsectors && my_mddev->external) {
3427                 /* Need to check that all other rdevs with the same
3428                  * ->bdev do not overlap.  'rcu' is sufficient to walk
3429                  * the rdev lists safely.
3430                  * This check does not provide a hard guarantee, it
3431                  * just helps avoid dangerous mistakes.
3432                  */
3433                 struct mddev *mddev;
3434                 int overlap = 0;
3435                 struct list_head *tmp;
3436
3437                 rcu_read_lock();
3438                 for_each_mddev(mddev, tmp) {
3439                         struct md_rdev *rdev2;
3440
3441                         rdev_for_each(rdev2, mddev)
3442                                 if (rdev->bdev == rdev2->bdev &&
3443                                     rdev != rdev2 &&
3444                                     overlaps(rdev->data_offset, rdev->sectors,
3445                                              rdev2->data_offset,
3446                                              rdev2->sectors)) {
3447                                         overlap = 1;
3448                                         break;
3449                                 }
3450                         if (overlap) {
3451                                 mddev_put(mddev);
3452                                 break;
3453                         }
3454                 }
3455                 rcu_read_unlock();
3456                 if (overlap) {
3457                         /* Someone else could have slipped in a size
3458                          * change here, but doing so is just silly.
3459                          * We put oldsectors back because we *know* it is
3460                          * safe, and trust userspace not to race with
3461                          * itself
3462                          */
3463                         rdev->sectors = oldsectors;
3464                         return -EBUSY;
3465                 }
3466         }
3467         return len;
3468 }
3469
3470 static struct rdev_sysfs_entry rdev_size =
3471 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3472
3473 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3474 {
3475         unsigned long long recovery_start = rdev->recovery_offset;
3476
3477         if (test_bit(In_sync, &rdev->flags) ||
3478             recovery_start == MaxSector)
3479                 return sprintf(page, "none\n");
3480
3481         return sprintf(page, "%llu\n", recovery_start);
3482 }
3483
3484 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3485 {
3486         unsigned long long recovery_start;
3487
3488         if (cmd_match(buf, "none"))
3489                 recovery_start = MaxSector;
3490         else if (kstrtoull(buf, 10, &recovery_start))
3491                 return -EINVAL;
3492
3493         if (rdev->mddev->pers &&
3494             rdev->raid_disk >= 0)
3495                 return -EBUSY;
3496
3497         rdev->recovery_offset = recovery_start;
3498         if (recovery_start == MaxSector)
3499                 set_bit(In_sync, &rdev->flags);
3500         else
3501                 clear_bit(In_sync, &rdev->flags);
3502         return len;
3503 }
3504
3505 static struct rdev_sysfs_entry rdev_recovery_start =
3506 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3507
3508 /* sysfs access to bad-blocks list.
3509  * We present two files.
3510  * 'bad-blocks' lists sector numbers and lengths of ranges that
3511  *    are recorded as bad.  The list is truncated to fit within
3512  *    the one-page limit of sysfs.
3513  *    Writing "sector length" to this file adds an acknowledged
3514  *    bad block list.
3515  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3516  *    been acknowledged.  Writing to this file adds bad blocks
3517  *    without acknowledging them.  This is largely for testing.
3518  */
3519 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3520 {
3521         return badblocks_show(&rdev->badblocks, page, 0);
3522 }
3523 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3524 {
3525         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3526         /* Maybe that ack was all we needed */
3527         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3528                 wake_up(&rdev->blocked_wait);
3529         return rv;
3530 }
3531 static struct rdev_sysfs_entry rdev_bad_blocks =
3532 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3533
3534 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3535 {
3536         return badblocks_show(&rdev->badblocks, page, 1);
3537 }
3538 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3539 {
3540         return badblocks_store(&rdev->badblocks, page, len, 1);
3541 }
3542 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3543 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3544
3545 static ssize_t
3546 ppl_sector_show(struct md_rdev *rdev, char *page)
3547 {
3548         return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3549 }
3550
3551 static ssize_t
3552 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3553 {
3554         unsigned long long sector;
3555
3556         if (kstrtoull(buf, 10, &sector) < 0)
3557                 return -EINVAL;
3558         if (sector != (sector_t)sector)
3559                 return -EINVAL;
3560
3561         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3562             rdev->raid_disk >= 0)
3563                 return -EBUSY;
3564
3565         if (rdev->mddev->persistent) {
3566                 if (rdev->mddev->major_version == 0)
3567                         return -EINVAL;
3568                 if ((sector > rdev->sb_start &&
3569                      sector - rdev->sb_start > S16_MAX) ||
3570                     (sector < rdev->sb_start &&
3571                      rdev->sb_start - sector > -S16_MIN))
3572                         return -EINVAL;
3573                 rdev->ppl.offset = sector - rdev->sb_start;
3574         } else if (!rdev->mddev->external) {
3575                 return -EBUSY;
3576         }
3577         rdev->ppl.sector = sector;
3578         return len;
3579 }
3580
3581 static struct rdev_sysfs_entry rdev_ppl_sector =
3582 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3583
3584 static ssize_t
3585 ppl_size_show(struct md_rdev *rdev, char *page)
3586 {
3587         return sprintf(page, "%u\n", rdev->ppl.size);
3588 }
3589
3590 static ssize_t
3591 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3592 {
3593         unsigned int size;
3594
3595         if (kstrtouint(buf, 10, &size) < 0)
3596                 return -EINVAL;
3597
3598         if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3599             rdev->raid_disk >= 0)
3600                 return -EBUSY;
3601
3602         if (rdev->mddev->persistent) {
3603                 if (rdev->mddev->major_version == 0)
3604                         return -EINVAL;
3605                 if (size > U16_MAX)
3606                         return -EINVAL;
3607         } else if (!rdev->mddev->external) {
3608                 return -EBUSY;
3609         }
3610         rdev->ppl.size = size;
3611         return len;
3612 }
3613
3614 static struct rdev_sysfs_entry rdev_ppl_size =
3615 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3616
3617 static struct attribute *rdev_default_attrs[] = {
3618         &rdev_state.attr,
3619         &rdev_errors.attr,
3620         &rdev_slot.attr,
3621         &rdev_offset.attr,
3622         &rdev_new_offset.attr,
3623         &rdev_size.attr,
3624         &rdev_recovery_start.attr,
3625         &rdev_bad_blocks.attr,
3626         &rdev_unack_bad_blocks.attr,
3627         &rdev_ppl_sector.attr,
3628         &rdev_ppl_size.attr,
3629         NULL,
3630 };
3631 static ssize_t
3632 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3633 {
3634         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3635         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3636
3637         if (!entry->show)
3638                 return -EIO;
3639         if (!rdev->mddev)
3640                 return -ENODEV;
3641         return entry->show(rdev, page);
3642 }
3643
3644 static ssize_t
3645 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3646               const char *page, size_t length)
3647 {
3648         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3649         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3650         ssize_t rv;
3651         struct mddev *mddev = rdev->mddev;
3652
3653         if (!entry->store)
3654                 return -EIO;
3655         if (!capable(CAP_SYS_ADMIN))
3656                 return -EACCES;
3657         rv = mddev ? mddev_lock(mddev) : -ENODEV;
3658         if (!rv) {
3659                 if (rdev->mddev == NULL)
3660                         rv = -ENODEV;
3661                 else
3662                         rv = entry->store(rdev, page, length);
3663                 mddev_unlock(mddev);
3664         }
3665         return rv;
3666 }
3667
3668 static void rdev_free(struct kobject *ko)
3669 {
3670         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3671         kfree(rdev);
3672 }
3673 static const struct sysfs_ops rdev_sysfs_ops = {
3674         .show           = rdev_attr_show,
3675         .store          = rdev_attr_store,
3676 };
3677 static struct kobj_type rdev_ktype = {
3678         .release        = rdev_free,
3679         .sysfs_ops      = &rdev_sysfs_ops,
3680         .default_attrs  = rdev_default_attrs,
3681 };
3682
3683 int md_rdev_init(struct md_rdev *rdev)
3684 {
3685         rdev->desc_nr = -1;
3686         rdev->saved_raid_disk = -1;
3687         rdev->raid_disk = -1;
3688         rdev->flags = 0;
3689         rdev->data_offset = 0;
3690         rdev->new_data_offset = 0;
3691         rdev->sb_events = 0;
3692         rdev->last_read_error = 0;
3693         rdev->sb_loaded = 0;
3694         rdev->bb_page = NULL;
3695         atomic_set(&rdev->nr_pending, 0);
3696         atomic_set(&rdev->read_errors, 0);
3697         atomic_set(&rdev->corrected_errors, 0);
3698
3699         INIT_LIST_HEAD(&rdev->same_set);
3700         init_waitqueue_head(&rdev->blocked_wait);
3701
3702         /* Add space to store bad block list.
3703          * This reserves the space even on arrays where it cannot
3704          * be used - I wonder if that matters
3705          */
3706         return badblocks_init(&rdev->badblocks, 0);
3707 }
3708 EXPORT_SYMBOL_GPL(md_rdev_init);
3709 /*
3710  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3711  *
3712  * mark the device faulty if:
3713  *
3714  *   - the device is nonexistent (zero size)
3715  *   - the device has no valid superblock
3716  *
3717  * a faulty rdev _never_ has rdev->sb set.
3718  */
3719 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3720 {
3721         char b[BDEVNAME_SIZE];
3722         int err;
3723         struct md_rdev *rdev;
3724         sector_t size;
3725
3726         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3727         if (!rdev)
3728                 return ERR_PTR(-ENOMEM);
3729
3730         err = md_rdev_init(rdev);
3731         if (err)
3732                 goto abort_free;
3733         err = alloc_disk_sb(rdev);
3734         if (err)
3735                 goto abort_free;
3736
3737         err = lock_rdev(rdev, newdev, super_format == -2);
3738         if (err)
3739                 goto abort_free;
3740
3741         kobject_init(&rdev->kobj, &rdev_ktype);
3742
3743         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3744         if (!size) {
3745                 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3746                         bdevname(rdev->bdev,b));
3747                 err = -EINVAL;
3748                 goto abort_free;
3749         }
3750
3751         if (super_format >= 0) {
3752                 err = super_types[super_format].
3753                         load_super(rdev, NULL, super_minor);
3754                 if (err == -EINVAL) {
3755                         pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3756                                 bdevname(rdev->bdev,b),
3757                                 super_format, super_minor);
3758                         goto abort_free;
3759                 }
3760                 if (err < 0) {
3761                         pr_warn("md: could not read %s's sb, not importing!\n",
3762                                 bdevname(rdev->bdev,b));
3763                         goto abort_free;
3764                 }
3765         }
3766
3767         return rdev;
3768
3769 abort_free:
3770         if (rdev->bdev)
3771                 unlock_rdev(rdev);
3772         md_rdev_clear(rdev);
3773         kfree(rdev);
3774         return ERR_PTR(err);
3775 }
3776
3777 /*
3778  * Check a full RAID array for plausibility
3779  */
3780
3781 static int analyze_sbs(struct mddev *mddev)
3782 {
3783         int i;
3784         struct md_rdev *rdev, *freshest, *tmp;
3785         char b[BDEVNAME_SIZE];
3786
3787         freshest = NULL;
3788         rdev_for_each_safe(rdev, tmp, mddev)
3789                 switch (super_types[mddev->major_version].
3790                         load_super(rdev, freshest, mddev->minor_version)) {
3791                 case 1:
3792                         freshest = rdev;
3793                         break;
3794                 case 0:
3795                         break;
3796                 default:
3797                         pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3798                                 bdevname(rdev->bdev,b));
3799                         md_kick_rdev_from_array(rdev);
3800                 }
3801
3802         /* Cannot find a valid fresh disk */
3803         if (!freshest) {
3804                 pr_warn("md: cannot find a valid disk\n");
3805                 return -EINVAL;
3806         }
3807
3808         super_types[mddev->major_version].
3809                 validate_super(mddev, freshest);
3810
3811         i = 0;
3812         rdev_for_each_safe(rdev, tmp, mddev) {
3813                 if (mddev->max_disks &&
3814                     (rdev->desc_nr >= mddev->max_disks ||
3815                      i > mddev->max_disks)) {
3816                         pr_warn("md: %s: %s: only %d devices permitted\n",
3817                                 mdname(mddev), bdevname(rdev->bdev, b),
3818                                 mddev->max_disks);
3819                         md_kick_rdev_from_array(rdev);
3820                         continue;
3821                 }
3822                 if (rdev != freshest) {
3823                         if (super_types[mddev->major_version].
3824                             validate_super(mddev, rdev)) {
3825                                 pr_warn("md: kicking non-fresh %s from array!\n",
3826                                         bdevname(rdev->bdev,b));
3827                                 md_kick_rdev_from_array(rdev);
3828                                 continue;
3829                         }
3830                 }
3831                 if (mddev->level == LEVEL_MULTIPATH) {
3832                         rdev->desc_nr = i++;
3833                         rdev->raid_disk = rdev->desc_nr;
3834                         set_bit(In_sync, &rdev->flags);
3835                 } else if (rdev->raid_disk >=
3836                             (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3837                            !test_bit(Journal, &rdev->flags)) {
3838                         rdev->raid_disk = -1;
3839                         clear_bit(In_sync, &rdev->flags);
3840                 }
3841         }
3842
3843         return 0;
3844 }
3845
3846 /* Read a fixed-point number.
3847  * Numbers in sysfs attributes should be in "standard" units where
3848  * possible, so time should be in seconds.
3849  * However we internally use a a much smaller unit such as
3850  * milliseconds or jiffies.
3851  * This function takes a decimal number with a possible fractional
3852  * component, and produces an integer which is the result of
3853  * multiplying that number by 10^'scale'.
3854  * all without any floating-point arithmetic.
3855  */
3856 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3857 {
3858         unsigned long result = 0;
3859         long decimals = -1;
3860         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3861                 if (*cp == '.')
3862                         decimals = 0;
3863                 else if (decimals < scale) {
3864                         unsigned int value;
3865                         value = *cp - '0';
3866                         result = result * 10 + value;
3867                         if (decimals >= 0)
3868                                 decimals++;
3869                 }
3870                 cp++;
3871         }
3872         if (*cp == '\n')
3873                 cp++;
3874         if (*cp)
3875                 return -EINVAL;
3876         if (decimals < 0)
3877                 decimals = 0;
3878         *res = result * int_pow(10, scale - decimals);
3879         return 0;
3880 }
3881
3882 static ssize_t
3883 safe_delay_show(struct mddev *mddev, char *page)
3884 {
3885         int msec = (mddev->safemode_delay*1000)/HZ;
3886         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3887 }
3888 static ssize_t
3889 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3890 {
3891         unsigned long msec;
3892
3893         if (mddev_is_clustered(mddev)) {
3894                 pr_warn("md: Safemode is disabled for clustered mode\n");
3895                 return -EINVAL;
3896         }
3897
3898         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3899                 return -EINVAL;
3900         if (msec == 0)
3901                 mddev->safemode_delay = 0;
3902         else {
3903                 unsigned long old_delay = mddev->safemode_delay;
3904                 unsigned long new_delay = (msec*HZ)/1000;
3905
3906                 if (new_delay == 0)
3907                         new_delay = 1;
3908                 mddev->safemode_delay = new_delay;
3909                 if (new_delay < old_delay || old_delay == 0)
3910                         mod_timer(&mddev->safemode_timer, jiffies+1);
3911         }
3912         return len;
3913 }
3914 static struct md_sysfs_entry md_safe_delay =
3915 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3916
3917 static ssize_t
3918 level_show(struct mddev *mddev, char *page)
3919 {
3920         struct md_personality *p;
3921         int ret;
3922         spin_lock(&mddev->lock);
3923         p = mddev->pers;
3924         if (p)
3925                 ret = sprintf(page, "%s\n", p->name);
3926         else if (mddev->clevel[0])
3927                 ret = sprintf(page, "%s\n", mddev->clevel);
3928         else if (mddev->level != LEVEL_NONE)
3929                 ret = sprintf(page, "%d\n", mddev->level);
3930         else
3931                 ret = 0;
3932         spin_unlock(&mddev->lock);
3933         return ret;
3934 }
3935
3936 static ssize_t
3937 level_store(struct mddev *mddev, const char *buf, size_t len)
3938 {
3939         char clevel[16];
3940         ssize_t rv;
3941         size_t slen = len;
3942         struct md_personality *pers, *oldpers;
3943         long level;
3944         void *priv, *oldpriv;
3945         struct md_rdev *rdev;
3946
3947         if (slen == 0 || slen >= sizeof(clevel))
3948                 return -EINVAL;
3949
3950         rv = mddev_lock(mddev);
3951         if (rv)
3952                 return rv;
3953
3954         if (mddev->pers == NULL) {
3955                 strncpy(mddev->clevel, buf, slen);
3956                 if (mddev->clevel[slen-1] == '\n')
3957                         slen--;
3958                 mddev->clevel[slen] = 0;
3959                 mddev->level = LEVEL_NONE;
3960                 rv = len;
3961                 goto out_unlock;
3962         }
3963         rv = -EROFS;
3964         if (mddev->ro)
3965                 goto out_unlock;
3966
3967         /* request to change the personality.  Need to ensure:
3968          *  - array is not engaged in resync/recovery/reshape
3969          *  - old personality can be suspended
3970          *  - new personality will access other array.
3971          */
3972
3973         rv = -EBUSY;
3974         if (mddev->sync_thread ||
3975             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3976             mddev->reshape_position != MaxSector ||
3977             mddev->sysfs_active)
3978                 goto out_unlock;
3979
3980         rv = -EINVAL;
3981         if (!mddev->pers->quiesce) {
3982                 pr_warn("md: %s: %s does not support online personality change\n",
3983                         mdname(mddev), mddev->pers->name);
3984                 goto out_unlock;
3985         }
3986
3987         /* Now find the new personality */
3988         strncpy(clevel, buf, slen);
3989         if (clevel[slen-1] == '\n')
3990                 slen--;
3991         clevel[slen] = 0;
3992         if (kstrtol(clevel, 10, &level))
3993                 level = LEVEL_NONE;
3994
3995         if (request_module("md-%s", clevel) != 0)
3996                 request_module("md-level-%s", clevel);
3997         spin_lock(&pers_lock);
3998         pers = find_pers(level, clevel);
3999         if (!pers || !try_module_get(pers->owner)) {
4000                 spin_unlock(&pers_lock);
4001                 pr_warn("md: personality %s not loaded\n", clevel);
4002                 rv = -EINVAL;
4003                 goto out_unlock;
4004         }
4005         spin_unlock(&pers_lock);
4006
4007         if (pers == mddev->pers) {
4008                 /* Nothing to do! */
4009                 module_put(pers->owner);
4010                 rv = len;
4011                 goto out_unlock;
4012         }
4013         if (!pers->takeover) {
4014                 module_put(pers->owner);
4015                 pr_warn("md: %s: %s does not support personality takeover\n",
4016                         mdname(mddev), clevel);
4017                 rv = -EINVAL;
4018                 goto out_unlock;
4019         }
4020
4021         rdev_for_each(rdev, mddev)
4022                 rdev->new_raid_disk = rdev->raid_disk;
4023
4024         /* ->takeover must set new_* and/or delta_disks
4025          * if it succeeds, and may set them when it fails.
4026          */
4027         priv = pers->takeover(mddev);
4028         if (IS_ERR(priv)) {
4029                 mddev->new_level = mddev->level;
4030                 mddev->new_layout = mddev->layout;
4031                 mddev->new_chunk_sectors = mddev->chunk_sectors;
4032                 mddev->raid_disks -= mddev->delta_disks;
4033                 mddev->delta_disks = 0;
4034                 mddev->reshape_backwards = 0;
4035                 module_put(pers->owner);
4036                 pr_warn("md: %s: %s would not accept array\n",
4037                         mdname(mddev), clevel);
4038                 rv = PTR_ERR(priv);
4039                 goto out_unlock;
4040         }
4041
4042         /* Looks like we have a winner */
4043         mddev_suspend(mddev);
4044         mddev_detach(mddev);
4045
4046         spin_lock(&mddev->lock);
4047         oldpers = mddev->pers;
4048         oldpriv = mddev->private;
4049         mddev->pers = pers;
4050         mddev->private = priv;
4051         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4052         mddev->level = mddev->new_level;
4053         mddev->layout = mddev->new_layout;
4054         mddev->chunk_sectors = mddev->new_chunk_sectors;
4055         mddev->delta_disks = 0;
4056         mddev->reshape_backwards = 0;
4057         mddev->degraded = 0;
4058         spin_unlock(&mddev->lock);
4059
4060         if (oldpers->sync_request == NULL &&
4061             mddev->external) {
4062                 /* We are converting from a no-redundancy array
4063                  * to a redundancy array and metadata is managed
4064                  * externally so we need to be sure that writes
4065                  * won't block due to a need to transition
4066                  *      clean->dirty
4067                  * until external management is started.
4068                  */
4069                 mddev->in_sync = 0;
4070                 mddev->safemode_delay = 0;
4071                 mddev->safemode = 0;
4072         }
4073
4074         oldpers->free(mddev, oldpriv);
4075
4076         if (oldpers->sync_request == NULL &&
4077             pers->sync_request != NULL) {
4078                 /* need to add the md_redundancy_group */
4079                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4080                         pr_warn("md: cannot register extra attributes for %s\n",
4081                                 mdname(mddev));
4082                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4083                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4084                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4085         }
4086         if (oldpers->sync_request != NULL &&
4087             pers->sync_request == NULL) {
4088                 /* need to remove the md_redundancy_group */
4089                 if (mddev->to_remove == NULL)
4090                         mddev->to_remove = &md_redundancy_group;
4091         }
4092
4093         module_put(oldpers->owner);
4094
4095         rdev_for_each(rdev, mddev) {
4096                 if (rdev->raid_disk < 0)
4097                         continue;
4098                 if (rdev->new_raid_disk >= mddev->raid_disks)
4099                         rdev->new_raid_disk = -1;
4100                 if (rdev->new_raid_disk == rdev->raid_disk)
4101                         continue;
4102                 sysfs_unlink_rdev(mddev, rdev);
4103         }
4104         rdev_for_each(rdev, mddev) {
4105                 if (rdev->raid_disk < 0)
4106                         continue;
4107                 if (rdev->new_raid_disk == rdev->raid_disk)
4108                         continue;
4109                 rdev->raid_disk = rdev->new_raid_disk;
4110                 if (rdev->raid_disk < 0)
4111                         clear_bit(In_sync, &rdev->flags);
4112                 else {
4113                         if (sysfs_link_rdev(mddev, rdev))
4114                                 pr_warn("md: cannot register rd%d for %s after level change\n",
4115                                         rdev->raid_disk, mdname(mddev));
4116                 }
4117         }
4118
4119         if (pers->sync_request == NULL) {
4120                 /* this is now an array without redundancy, so
4121                  * it must always be in_sync
4122                  */
4123                 mddev->in_sync = 1;
4124                 del_timer_sync(&mddev->safemode_timer);
4125         }
4126         blk_set_stacking_limits(&mddev->queue->limits);
4127         pers->run(mddev);
4128         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4129         mddev_resume(mddev);
4130         if (!mddev->thread)
4131                 md_update_sb(mddev, 1);
4132         sysfs_notify_dirent_safe(mddev->sysfs_level);
4133         md_new_event(mddev);
4134         rv = len;
4135 out_unlock:
4136         mddev_unlock(mddev);
4137         return rv;
4138 }
4139
4140 static struct md_sysfs_entry md_level =
4141 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4142
4143 static ssize_t
4144 layout_show(struct mddev *mddev, char *page)
4145 {
4146         /* just a number, not meaningful for all levels */
4147         if (mddev->reshape_position != MaxSector &&
4148             mddev->layout != mddev->new_layout)
4149                 return sprintf(page, "%d (%d)\n",
4150                                mddev->new_layout, mddev->layout);
4151         return sprintf(page, "%d\n", mddev->layout);
4152 }
4153
4154 static ssize_t
4155 layout_store(struct mddev *mddev, const char *buf, size_t len)
4156 {
4157         unsigned int n;
4158         int err;
4159
4160         err = kstrtouint(buf, 10, &n);
4161         if (err < 0)
4162                 return err;
4163         err = mddev_lock(mddev);
4164         if (err)
4165                 return err;
4166
4167         if (mddev->pers) {
4168                 if (mddev->pers->check_reshape == NULL)
4169                         err = -EBUSY;
4170                 else if (mddev->ro)
4171                         err = -EROFS;
4172                 else {
4173                         mddev->new_layout = n;
4174                         err = mddev->pers->check_reshape(mddev);
4175                         if (err)
4176                                 mddev->new_layout = mddev->layout;
4177                 }
4178         } else {
4179                 mddev->new_layout = n;
4180                 if (mddev->reshape_position == MaxSector)
4181                         mddev->layout = n;
4182         }
4183         mddev_unlock(mddev);
4184         return err ?: len;
4185 }
4186 static struct md_sysfs_entry md_layout =
4187 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4188
4189 static ssize_t
4190 raid_disks_show(struct mddev *mddev, char *page)
4191 {
4192         if (mddev->raid_disks == 0)
4193                 return 0;
4194         if (mddev->reshape_position != MaxSector &&
4195             mddev->delta_disks != 0)
4196                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4197                                mddev->raid_disks - mddev->delta_disks);
4198         return sprintf(page, "%d\n", mddev->raid_disks);
4199 }
4200
4201 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4202
4203 static ssize_t
4204 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4205 {
4206         unsigned int n;
4207         int err;
4208
4209         err = kstrtouint(buf, 10, &n);
4210         if (err < 0)
4211                 return err;
4212
4213         err = mddev_lock(mddev);
4214         if (err)
4215                 return err;
4216         if (mddev->pers)
4217                 err = update_raid_disks(mddev, n);
4218         else if (mddev->reshape_position != MaxSector) {
4219                 struct md_rdev *rdev;
4220                 int olddisks = mddev->raid_disks - mddev->delta_disks;
4221
4222                 err = -EINVAL;
4223                 rdev_for_each(rdev, mddev) {
4224                         if (olddisks < n &&
4225                             rdev->data_offset < rdev->new_data_offset)
4226                                 goto out_unlock;
4227                         if (olddisks > n &&
4228                             rdev->data_offset > rdev->new_data_offset)
4229                                 goto out_unlock;
4230                 }
4231                 err = 0;
4232                 mddev->delta_disks = n - olddisks;
4233                 mddev->raid_disks = n;
4234                 mddev->reshape_backwards = (mddev->delta_disks < 0);
4235         } else
4236                 mddev->raid_disks = n;
4237 out_unlock:
4238         mddev_unlock(mddev);
4239         return err ? err : len;
4240 }
4241 static struct md_sysfs_entry md_raid_disks =
4242 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4243
4244 static ssize_t
4245 uuid_show(struct mddev *mddev, char *page)
4246 {
4247         return sprintf(page, "%pU\n", mddev->uuid);
4248 }
4249 static struct md_sysfs_entry md_uuid =
4250 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4251
4252 static ssize_t
4253 chunk_size_show(struct mddev *mddev, char *page)
4254 {
4255         if (mddev->reshape_position != MaxSector &&
4256             mddev->chunk_sectors != mddev->new_chunk_sectors)
4257                 return sprintf(page, "%d (%d)\n",
4258                                mddev->new_chunk_sectors << 9,
4259                                mddev->chunk_sectors << 9);
4260         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4261 }
4262
4263 static ssize_t
4264 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4265 {
4266         unsigned long n;
4267         int err;
4268
4269         err = kstrtoul(buf, 10, &n);
4270         if (err < 0)
4271                 return err;
4272
4273         err = mddev_lock(mddev);
4274         if (err)
4275                 return err;
4276         if (mddev->pers) {
4277                 if (mddev->pers->check_reshape == NULL)
4278                         err = -EBUSY;
4279                 else if (mddev->ro)
4280                         err = -EROFS;
4281                 else {
4282                         mddev->new_chunk_sectors = n >> 9;
4283                         err = mddev->pers->check_reshape(mddev);
4284                         if (err)
4285                                 mddev->new_chunk_sectors = mddev->chunk_sectors;
4286                 }
4287         } else {
4288                 mddev->new_chunk_sectors = n >> 9;
4289                 if (mddev->reshape_position == MaxSector)
4290                         mddev->chunk_sectors = n >> 9;
4291         }
4292         mddev_unlock(mddev);
4293         return err ?: len;
4294 }
4295 static struct md_sysfs_entry md_chunk_size =
4296 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4297
4298 static ssize_t
4299 resync_start_show(struct mddev *mddev, char *page)
4300 {
4301         if (mddev->recovery_cp == MaxSector)
4302                 return sprintf(page, "none\n");
4303         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4304 }
4305
4306 static ssize_t
4307 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4308 {
4309         unsigned long long n;
4310         int err;
4311
4312         if (cmd_match(buf, "none"))
4313                 n = MaxSector;
4314         else {
4315                 err = kstrtoull(buf, 10, &n);
4316                 if (err < 0)
4317                         return err;
4318                 if (n != (sector_t)n)
4319                         return -EINVAL;
4320         }
4321
4322         err = mddev_lock(mddev);
4323         if (err)
4324                 return err;
4325         if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4326                 err = -EBUSY;
4327
4328         if (!err) {
4329                 mddev->recovery_cp = n;
4330                 if (mddev->pers)
4331                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4332         }
4333         mddev_unlock(mddev);
4334         return err ?: len;
4335 }
4336 static struct md_sysfs_entry md_resync_start =
4337 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4338                 resync_start_show, resync_start_store);
4339
4340 /*
4341  * The array state can be:
4342  *
4343  * clear
4344  *     No devices, no size, no level
4345  *     Equivalent to STOP_ARRAY ioctl
4346  * inactive
4347  *     May have some settings, but array is not active
4348  *        all IO results in error
4349  *     When written, doesn't tear down array, but just stops it
4350  * suspended (not supported yet)
4351  *     All IO requests will block. The array can be reconfigured.
4352  *     Writing this, if accepted, will block until array is quiescent
4353  * readonly
4354  *     no resync can happen.  no superblocks get written.
4355  *     write requests fail
4356  * read-auto
4357  *     like readonly, but behaves like 'clean' on a write request.
4358  *
4359  * clean - no pending writes, but otherwise active.
4360  *     When written to inactive array, starts without resync
4361  *     If a write request arrives then
4362  *       if metadata is known, mark 'dirty' and switch to 'active'.
4363  *       if not known, block and switch to write-pending
4364  *     If written to an active array that has pending writes, then fails.
4365  * active
4366  *     fully active: IO and resync can be happening.
4367  *     When written to inactive array, starts with resync
4368  *
4369  * write-pending
4370  *     clean, but writes are blocked waiting for 'active' to be written.
4371  *
4372  * active-idle
4373  *     like active, but no writes have been seen for a while (100msec).
4374  *
4375  * broken
4376  *     RAID0/LINEAR-only: same as clean, but array is missing a member.
4377  *     It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4378  *     when a member is gone, so this state will at least alert the
4379  *     user that something is wrong.
4380  */
4381 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4382                    write_pending, active_idle, broken, bad_word};
4383 static char *array_states[] = {
4384         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4385         "write-pending", "active-idle", "broken", NULL };
4386
4387 static int match_word(const char *word, char **list)
4388 {
4389         int n;
4390         for (n=0; list[n]; n++)
4391                 if (cmd_match(word, list[n]))
4392                         break;
4393         return n;
4394 }
4395
4396 static ssize_t
4397 array_state_show(struct mddev *mddev, char *page)
4398 {
4399         enum array_state st = inactive;
4400
4401         if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4402                 switch(mddev->ro) {
4403                 case 1:
4404                         st = readonly;
4405                         break;
4406                 case 2:
4407                         st = read_auto;
4408                         break;
4409                 case 0:
4410                         spin_lock(&mddev->lock);
4411                         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4412                                 st = write_pending;
4413                         else if (mddev->in_sync)
4414                                 st = clean;
4415                         else if (mddev->safemode)
4416                                 st = active_idle;
4417                         else
4418                                 st = active;
4419                         spin_unlock(&mddev->lock);
4420                 }
4421
4422                 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4423                         st = broken;
4424         } else {
4425                 if (list_empty(&mddev->disks) &&
4426                     mddev->raid_disks == 0 &&
4427                     mddev->dev_sectors == 0)
4428                         st = clear;
4429                 else
4430                         st = inactive;
4431         }
4432         return sprintf(page, "%s\n", array_states[st]);
4433 }
4434
4435 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4436 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4437 static int restart_array(struct mddev *mddev);
4438
4439 static ssize_t
4440 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4441 {
4442         int err = 0;
4443         enum array_state st = match_word(buf, array_states);
4444
4445         if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4446                 /* don't take reconfig_mutex when toggling between
4447                  * clean and active
4448                  */
4449                 spin_lock(&mddev->lock);
4450                 if (st == active) {
4451                         restart_array(mddev);
4452                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4453                         md_wakeup_thread(mddev->thread);
4454                         wake_up(&mddev->sb_wait);
4455                 } else /* st == clean */ {
4456                         restart_array(mddev);
4457                         if (!set_in_sync(mddev))
4458                                 err = -EBUSY;
4459                 }
4460                 if (!err)
4461                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4462                 spin_unlock(&mddev->lock);
4463                 return err ?: len;
4464         }
4465         err = mddev_lock(mddev);
4466         if (err)
4467                 return err;
4468         err = -EINVAL;
4469         switch(st) {
4470         case bad_word:
4471                 break;
4472         case clear:
4473                 /* stopping an active array */
4474                 err = do_md_stop(mddev, 0, NULL);
4475                 break;
4476         case inactive:
4477                 /* stopping an active array */
4478                 if (mddev->pers)
4479                         err = do_md_stop(mddev, 2, NULL);
4480                 else
4481                         err = 0; /* already inactive */
4482                 break;
4483         case suspended:
4484                 break; /* not supported yet */
4485         case readonly:
4486                 if (mddev->pers)
4487                         err = md_set_readonly(mddev, NULL);
4488                 else {
4489                         mddev->ro = 1;
4490                         set_disk_ro(mddev->gendisk, 1);
4491                         err = do_md_run(mddev);
4492                 }
4493                 break;
4494         case read_auto:
4495                 if (mddev->pers) {
4496                         if (mddev->ro == 0)
4497                                 err = md_set_readonly(mddev, NULL);
4498                         else if (mddev->ro == 1)
4499                                 err = restart_array(mddev);
4500                         if (err == 0) {
4501                                 mddev->ro = 2;
4502                                 set_disk_ro(mddev->gendisk, 0);
4503                         }
4504                 } else {
4505                         mddev->ro = 2;
4506                         err = do_md_run(mddev);
4507                 }
4508                 break;
4509         case clean:
4510                 if (mddev->pers) {
4511                         err = restart_array(mddev);
4512                         if (err)
4513                                 break;
4514                         spin_lock(&mddev->lock);
4515                         if (!set_in_sync(mddev))
4516                                 err = -EBUSY;
4517                         spin_unlock(&mddev->lock);
4518                 } else
4519                         err = -EINVAL;
4520                 break;
4521         case active:
4522                 if (mddev->pers) {
4523                         err = restart_array(mddev);
4524                         if (err)
4525                                 break;
4526                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4527                         wake_up(&mddev->sb_wait);
4528                         err = 0;
4529                 } else {
4530                         mddev->ro = 0;
4531                         set_disk_ro(mddev->gendisk, 0);
4532                         err = do_md_run(mddev);
4533                 }
4534                 break;
4535         case write_pending:
4536         case active_idle:
4537         case broken:
4538                 /* these cannot be set */
4539                 break;
4540         }
4541
4542         if (!err) {
4543                 if (mddev->hold_active == UNTIL_IOCTL)
4544                         mddev->hold_active = 0;
4545                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4546         }
4547         mddev_unlock(mddev);
4548         return err ?: len;
4549 }
4550 static struct md_sysfs_entry md_array_state =
4551 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4552
4553 static ssize_t
4554 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4555         return sprintf(page, "%d\n",
4556                        atomic_read(&mddev->max_corr_read_errors));
4557 }
4558
4559 static ssize_t
4560 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4561 {
4562         unsigned int n;
4563         int rv;
4564
4565         rv = kstrtouint(buf, 10, &n);
4566         if (rv < 0)
4567                 return rv;
4568         atomic_set(&mddev->max_corr_read_errors, n);
4569         return len;
4570 }
4571
4572 static struct md_sysfs_entry max_corr_read_errors =
4573 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4574         max_corrected_read_errors_store);
4575
4576 static ssize_t
4577 null_show(struct mddev *mddev, char *page)
4578 {
4579         return -EINVAL;
4580 }
4581
4582 /* need to ensure rdev_delayed_delete() has completed */
4583 static void flush_rdev_wq(struct mddev *mddev)
4584 {
4585         struct md_rdev *rdev;
4586
4587         rcu_read_lock();
4588         rdev_for_each_rcu(rdev, mddev)
4589                 if (work_pending(&rdev->del_work)) {
4590                         flush_workqueue(md_rdev_misc_wq);
4591                         break;
4592                 }
4593         rcu_read_unlock();
4594 }
4595
4596 static ssize_t
4597 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4598 {
4599         /* buf must be %d:%d\n? giving major and minor numbers */
4600         /* The new device is added to the array.
4601          * If the array has a persistent superblock, we read the
4602          * superblock to initialise info and check validity.
4603          * Otherwise, only checking done is that in bind_rdev_to_array,
4604          * which mainly checks size.
4605          */
4606         char *e;
4607         int major = simple_strtoul(buf, &e, 10);
4608         int minor;
4609         dev_t dev;
4610         struct md_rdev *rdev;
4611         int err;
4612
4613         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4614                 return -EINVAL;
4615         minor = simple_strtoul(e+1, &e, 10);
4616         if (*e && *e != '\n')
4617                 return -EINVAL;
4618         dev = MKDEV(major, minor);
4619         if (major != MAJOR(dev) ||
4620             minor != MINOR(dev))
4621                 return -EOVERFLOW;
4622
4623         flush_rdev_wq(mddev);
4624         err = mddev_lock(mddev);
4625         if (err)
4626                 return err;
4627         if (mddev->persistent) {
4628                 rdev = md_import_device(dev, mddev->major_version,
4629                                         mddev->minor_version);
4630                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4631                         struct md_rdev *rdev0
4632                                 = list_entry(mddev->disks.next,
4633                                              struct md_rdev, same_set);
4634                         err = super_types[mddev->major_version]
4635                                 .load_super(rdev, rdev0, mddev->minor_version);
4636                         if (err < 0)
4637                                 goto out;
4638                 }
4639         } else if (mddev->external)
4640                 rdev = md_import_device(dev, -2, -1);
4641         else
4642                 rdev = md_import_device(dev, -1, -1);
4643
4644         if (IS_ERR(rdev)) {
4645                 mddev_unlock(mddev);
4646                 return PTR_ERR(rdev);
4647         }
4648         err = bind_rdev_to_array(rdev, mddev);
4649  out:
4650         if (err)
4651                 export_rdev(rdev);
4652         mddev_unlock(mddev);
4653         if (!err)
4654                 md_new_event(mddev);
4655         return err ? err : len;
4656 }
4657
4658 static struct md_sysfs_entry md_new_device =
4659 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4660
4661 static ssize_t
4662 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4663 {
4664         char *end;
4665         unsigned long chunk, end_chunk;
4666         int err;
4667
4668         err = mddev_lock(mddev);
4669         if (err)
4670                 return err;
4671         if (!mddev->bitmap)
4672                 goto out;
4673         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4674         while (*buf) {
4675                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4676                 if (buf == end) break;
4677                 if (*end == '-') { /* range */
4678                         buf = end + 1;
4679                         end_chunk = simple_strtoul(buf, &end, 0);
4680                         if (buf == end) break;
4681                 }
4682                 if (*end && !isspace(*end)) break;
4683                 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4684                 buf = skip_spaces(end);
4685         }
4686         md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4687 out:
4688         mddev_unlock(mddev);
4689         return len;
4690 }
4691
4692 static struct md_sysfs_entry md_bitmap =
4693 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4694
4695 static ssize_t
4696 size_show(struct mddev *mddev, char *page)
4697 {
4698         return sprintf(page, "%llu\n",
4699                 (unsigned long long)mddev->dev_sectors / 2);
4700 }
4701
4702 static int update_size(struct mddev *mddev, sector_t num_sectors);
4703
4704 static ssize_t
4705 size_store(struct mddev *mddev, const char *buf, size_t len)
4706 {
4707         /* If array is inactive, we can reduce the component size, but
4708          * not increase it (except from 0).
4709          * If array is active, we can try an on-line resize
4710          */
4711         sector_t sectors;
4712         int err = strict_blocks_to_sectors(buf, &sectors);
4713
4714         if (err < 0)
4715                 return err;
4716         err = mddev_lock(mddev);
4717         if (err)
4718                 return err;
4719         if (mddev->pers) {
4720                 err = update_size(mddev, sectors);
4721                 if (err == 0)
4722                         md_update_sb(mddev, 1);
4723         } else {
4724                 if (mddev->dev_sectors == 0 ||
4725                     mddev->dev_sectors > sectors)
4726                         mddev->dev_sectors = sectors;
4727                 else
4728                         err = -ENOSPC;
4729         }
4730         mddev_unlock(mddev);
4731         return err ? err : len;
4732 }
4733
4734 static struct md_sysfs_entry md_size =
4735 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4736
4737 /* Metadata version.
4738  * This is one of
4739  *   'none' for arrays with no metadata (good luck...)
4740  *   'external' for arrays with externally managed metadata,
4741  * or N.M for internally known formats
4742  */
4743 static ssize_t
4744 metadata_show(struct mddev *mddev, char *page)
4745 {
4746         if (mddev->persistent)
4747                 return sprintf(page, "%d.%d\n",
4748                                mddev->major_version, mddev->minor_version);
4749         else if (mddev->external)
4750                 return sprintf(page, "external:%s\n", mddev->metadata_type);
4751         else
4752                 return sprintf(page, "none\n");
4753 }
4754
4755 static ssize_t
4756 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4757 {
4758         int major, minor;
4759         char *e;
4760         int err;
4761         /* Changing the details of 'external' metadata is
4762          * always permitted.  Otherwise there must be
4763          * no devices attached to the array.
4764          */
4765
4766         err = mddev_lock(mddev);
4767         if (err)
4768                 return err;
4769         err = -EBUSY;
4770         if (mddev->external && strncmp(buf, "external:", 9) == 0)
4771                 ;
4772         else if (!list_empty(&mddev->disks))
4773                 goto out_unlock;
4774
4775         err = 0;
4776         if (cmd_match(buf, "none")) {
4777                 mddev->persistent = 0;
4778                 mddev->external = 0;
4779                 mddev->major_version = 0;
4780                 mddev->minor_version = 90;
4781                 goto out_unlock;
4782         }
4783         if (strncmp(buf, "external:", 9) == 0) {
4784                 size_t namelen = len-9;
4785                 if (namelen >= sizeof(mddev->metadata_type))
4786                         namelen = sizeof(mddev->metadata_type)-1;
4787                 strncpy(mddev->metadata_type, buf+9, namelen);
4788                 mddev->metadata_type[namelen] = 0;
4789                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4790                         mddev->metadata_type[--namelen] = 0;
4791                 mddev->persistent = 0;
4792                 mddev->external = 1;
4793                 mddev->major_version = 0;
4794                 mddev->minor_version = 90;
4795                 goto out_unlock;
4796         }
4797         major = simple_strtoul(buf, &e, 10);
4798         err = -EINVAL;
4799         if (e==buf || *e != '.')
4800                 goto out_unlock;
4801         buf = e+1;
4802         minor = simple_strtoul(buf, &e, 10);
4803         if (e==buf || (*e && *e != '\n') )
4804                 goto out_unlock;
4805         err = -ENOENT;
4806         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4807                 goto out_unlock;
4808         mddev->major_version = major;
4809         mddev->minor_version = minor;
4810         mddev->persistent = 1;
4811         mddev->external = 0;
4812         err = 0;
4813 out_unlock:
4814         mddev_unlock(mddev);
4815         return err ?: len;
4816 }
4817
4818 static struct md_sysfs_entry md_metadata =
4819 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4820
4821 static ssize_t
4822 action_show(struct mddev *mddev, char *page)
4823 {
4824         char *type = "idle";
4825         unsigned long recovery = mddev->recovery;
4826         if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4827                 type = "frozen";
4828         else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4829             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4830                 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4831                         type = "reshape";
4832                 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4833                         if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4834                                 type = "resync";
4835                         else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4836                                 type = "check";
4837                         else
4838                                 type = "repair";
4839                 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4840                         type = "recover";
4841                 else if (mddev->reshape_position != MaxSector)
4842                         type = "reshape";
4843         }
4844         return sprintf(page, "%s\n", type);
4845 }
4846
4847 static ssize_t
4848 action_store(struct mddev *mddev, const char *page, size_t len)
4849 {
4850         if (!mddev->pers || !mddev->pers->sync_request)
4851                 return -EINVAL;
4852
4853
4854         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4855                 if (cmd_match(page, "frozen"))
4856                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4857                 else
4858                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4859                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4860                     mddev_lock(mddev) == 0) {
4861                         if (work_pending(&mddev->del_work))
4862                                 flush_workqueue(md_misc_wq);
4863                         if (mddev->sync_thread) {
4864                                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4865                                 md_reap_sync_thread(mddev);
4866                         }
4867                         mddev_unlock(mddev);
4868                 }
4869         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4870                 return -EBUSY;
4871         else if (cmd_match(page, "resync"))
4872                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4873         else if (cmd_match(page, "recover")) {
4874                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4875                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4876         } else if (cmd_match(page, "reshape")) {
4877                 int err;
4878                 if (mddev->pers->start_reshape == NULL)
4879                         return -EINVAL;
4880                 err = mddev_lock(mddev);
4881                 if (!err) {
4882                         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4883                                 err =  -EBUSY;
4884                         else {
4885                                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4886                                 err = mddev->pers->start_reshape(mddev);
4887                         }
4888                         mddev_unlock(mddev);
4889                 }
4890                 if (err)
4891                         return err;
4892                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4893         } else {
4894                 if (cmd_match(page, "check"))
4895                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4896                 else if (!cmd_match(page, "repair"))
4897                         return -EINVAL;
4898                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4899                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4900                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4901         }
4902         if (mddev->ro == 2) {
4903                 /* A write to sync_action is enough to justify
4904                  * canceling read-auto mode
4905                  */
4906                 mddev->ro = 0;
4907                 md_wakeup_thread(mddev->sync_thread);
4908         }
4909         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4910         md_wakeup_thread(mddev->thread);
4911         sysfs_notify_dirent_safe(mddev->sysfs_action);
4912         return len;
4913 }
4914
4915 static struct md_sysfs_entry md_scan_mode =
4916 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4917
4918 static ssize_t
4919 last_sync_action_show(struct mddev *mddev, char *page)
4920 {
4921         return sprintf(page, "%s\n", mddev->last_sync_action);
4922 }
4923
4924 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4925
4926 static ssize_t
4927 mismatch_cnt_show(struct mddev *mddev, char *page)
4928 {
4929         return sprintf(page, "%llu\n",
4930                        (unsigned long long)
4931                        atomic64_read(&mddev->resync_mismatches));
4932 }
4933
4934 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4935
4936 static ssize_t
4937 sync_min_show(struct mddev *mddev, char *page)
4938 {
4939         return sprintf(page, "%d (%s)\n", speed_min(mddev),
4940                        mddev->sync_speed_min ? "local": "system");
4941 }
4942
4943 static ssize_t
4944 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4945 {
4946         unsigned int min;
4947         int rv;
4948
4949         if (strncmp(buf, "system", 6)==0) {
4950                 min = 0;
4951         } else {
4952                 rv = kstrtouint(buf, 10, &min);
4953                 if (rv < 0)
4954                         return rv;
4955                 if (min == 0)
4956                         return -EINVAL;
4957         }
4958         mddev->sync_speed_min = min;
4959         return len;
4960 }
4961
4962 static struct md_sysfs_entry md_sync_min =
4963 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4964
4965 static ssize_t
4966 sync_max_show(struct mddev *mddev, char *page)
4967 {
4968         return sprintf(page, "%d (%s)\n", speed_max(mddev),
4969                        mddev->sync_speed_max ? "local": "system");
4970 }
4971
4972 static ssize_t
4973 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4974 {
4975         unsigned int max;
4976         int rv;
4977
4978         if (strncmp(buf, "system", 6)==0) {
4979                 max = 0;
4980         } else {
4981                 rv = kstrtouint(buf, 10, &max);
4982                 if (rv < 0)
4983                         return rv;
4984                 if (max == 0)
4985                         return -EINVAL;
4986         }
4987         mddev->sync_speed_max = max;
4988         return len;
4989 }
4990
4991 static struct md_sysfs_entry md_sync_max =
4992 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4993
4994 static ssize_t
4995 degraded_show(struct mddev *mddev, char *page)
4996 {
4997         return sprintf(page, "%d\n", mddev->degraded);
4998 }
4999 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5000
5001 static ssize_t
5002 sync_force_parallel_show(struct mddev *mddev, char *page)
5003 {
5004         return sprintf(page, "%d\n", mddev->parallel_resync);
5005 }
5006
5007 static ssize_t
5008 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5009 {
5010         long n;
5011
5012         if (kstrtol(buf, 10, &n))
5013                 return -EINVAL;
5014
5015         if (n != 0 && n != 1)
5016                 return -EINVAL;
5017
5018         mddev->parallel_resync = n;
5019
5020         if (mddev->sync_thread)
5021                 wake_up(&resync_wait);
5022
5023         return len;
5024 }
5025
5026 /* force parallel resync, even with shared block devices */
5027 static struct md_sysfs_entry md_sync_force_parallel =
5028 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5029        sync_force_parallel_show, sync_force_parallel_store);
5030
5031 static ssize_t
5032 sync_speed_show(struct mddev *mddev, char *page)
5033 {
5034         unsigned long resync, dt, db;
5035         if (mddev->curr_resync == 0)
5036                 return sprintf(page, "none\n");
5037         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5038         dt = (jiffies - mddev->resync_mark) / HZ;
5039         if (!dt) dt++;
5040         db = resync - mddev->resync_mark_cnt;
5041         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5042 }
5043
5044 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5045
5046 static ssize_t
5047 sync_completed_show(struct mddev *mddev, char *page)
5048 {
5049         unsigned long long max_sectors, resync;
5050
5051         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5052                 return sprintf(page, "none\n");
5053
5054         if (mddev->curr_resync == 1 ||
5055             mddev->curr_resync == 2)
5056                 return sprintf(page, "delayed\n");
5057
5058         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5059             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5060                 max_sectors = mddev->resync_max_sectors;
5061         else
5062                 max_sectors = mddev->dev_sectors;
5063
5064         resync = mddev->curr_resync_completed;
5065         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5066 }
5067
5068 static struct md_sysfs_entry md_sync_completed =
5069         __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5070
5071 static ssize_t
5072 min_sync_show(struct mddev *mddev, char *page)
5073 {
5074         return sprintf(page, "%llu\n",
5075                        (unsigned long long)mddev->resync_min);
5076 }
5077 static ssize_t
5078 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5079 {
5080         unsigned long long min;
5081         int err;
5082
5083         if (kstrtoull(buf, 10, &min))
5084                 return -EINVAL;
5085
5086         spin_lock(&mddev->lock);
5087         err = -EINVAL;
5088         if (min > mddev->resync_max)
5089                 goto out_unlock;
5090
5091         err = -EBUSY;
5092         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5093                 goto out_unlock;
5094
5095         /* Round down to multiple of 4K for safety */
5096         mddev->resync_min = round_down(min, 8);
5097         err = 0;
5098
5099 out_unlock:
5100         spin_unlock(&mddev->lock);
5101         return err ?: len;
5102 }
5103
5104 static struct md_sysfs_entry md_min_sync =
5105 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5106
5107 static ssize_t
5108 max_sync_show(struct mddev *mddev, char *page)
5109 {
5110         if (mddev->resync_max == MaxSector)
5111                 return sprintf(page, "max\n");
5112         else
5113                 return sprintf(page, "%llu\n",
5114                                (unsigned long long)mddev->resync_max);
5115 }
5116 static ssize_t
5117 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5118 {
5119         int err;
5120         spin_lock(&mddev->lock);
5121         if (strncmp(buf, "max", 3) == 0)
5122                 mddev->resync_max = MaxSector;
5123         else {
5124                 unsigned long long max;
5125                 int chunk;
5126
5127                 err = -EINVAL;
5128                 if (kstrtoull(buf, 10, &max))
5129                         goto out_unlock;
5130                 if (max < mddev->resync_min)
5131                         goto out_unlock;
5132
5133                 err = -EBUSY;
5134                 if (max < mddev->resync_max &&
5135                     mddev->ro == 0 &&
5136                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5137                         goto out_unlock;
5138
5139                 /* Must be a multiple of chunk_size */
5140                 chunk = mddev->chunk_sectors;
5141                 if (chunk) {
5142                         sector_t temp = max;
5143
5144                         err = -EINVAL;
5145                         if (sector_div(temp, chunk))
5146                                 goto out_unlock;
5147                 }
5148                 mddev->resync_max = max;
5149         }
5150         wake_up(&mddev->recovery_wait);
5151         err = 0;
5152 out_unlock:
5153         spin_unlock(&mddev->lock);
5154         return err ?: len;
5155 }
5156
5157 static struct md_sysfs_entry md_max_sync =
5158 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5159
5160 static ssize_t
5161 suspend_lo_show(struct mddev *mddev, char *page)
5162 {
5163         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5164 }
5165
5166 static ssize_t
5167 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5168 {
5169         unsigned long long new;
5170         int err;
5171
5172         err = kstrtoull(buf, 10, &new);
5173         if (err < 0)
5174                 return err;
5175         if (new != (sector_t)new)
5176                 return -EINVAL;
5177
5178         err = mddev_lock(mddev);
5179         if (err)
5180                 return err;
5181         err = -EINVAL;
5182         if (mddev->pers == NULL ||
5183             mddev->pers->quiesce == NULL)
5184                 goto unlock;
5185         mddev_suspend(mddev);
5186         mddev->suspend_lo = new;
5187         mddev_resume(mddev);
5188
5189         err = 0;
5190 unlock:
5191         mddev_unlock(mddev);
5192         return err ?: len;
5193 }
5194 static struct md_sysfs_entry md_suspend_lo =
5195 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5196
5197 static ssize_t
5198 suspend_hi_show(struct mddev *mddev, char *page)
5199 {
5200         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5201 }
5202
5203 static ssize_t
5204 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5205 {
5206         unsigned long long new;
5207         int err;
5208
5209         err = kstrtoull(buf, 10, &new);
5210         if (err < 0)
5211                 return err;
5212         if (new != (sector_t)new)
5213                 return -EINVAL;
5214
5215         err = mddev_lock(mddev);
5216         if (err)
5217                 return err;
5218         err = -EINVAL;
5219         if (mddev->pers == NULL)
5220                 goto unlock;
5221
5222         mddev_suspend(mddev);
5223         mddev->suspend_hi = new;
5224         mddev_resume(mddev);
5225
5226         err = 0;
5227 unlock:
5228         mddev_unlock(mddev);
5229         return err ?: len;
5230 }
5231 static struct md_sysfs_entry md_suspend_hi =
5232 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5233
5234 static ssize_t
5235 reshape_position_show(struct mddev *mddev, char *page)
5236 {
5237         if (mddev->reshape_position != MaxSector)
5238                 return sprintf(page, "%llu\n",
5239                                (unsigned long long)mddev->reshape_position);
5240         strcpy(page, "none\n");
5241         return 5;
5242 }
5243
5244 static ssize_t
5245 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5246 {
5247         struct md_rdev *rdev;
5248         unsigned long long new;
5249         int err;
5250
5251         err = kstrtoull(buf, 10, &new);
5252         if (err < 0)
5253                 return err;
5254         if (new != (sector_t)new)
5255                 return -EINVAL;
5256         err = mddev_lock(mddev);
5257         if (err)
5258                 return err;
5259         err = -EBUSY;
5260         if (mddev->pers)
5261                 goto unlock;
5262         mddev->reshape_position = new;
5263         mddev->delta_disks = 0;
5264         mddev->reshape_backwards = 0;
5265         mddev->new_level = mddev->level;
5266         mddev->new_layout = mddev->layout;
5267         mddev->new_chunk_sectors = mddev->chunk_sectors;
5268         rdev_for_each(rdev, mddev)
5269                 rdev->new_data_offset = rdev->data_offset;
5270         err = 0;
5271 unlock:
5272         mddev_unlock(mddev);
5273         return err ?: len;
5274 }
5275
5276 static struct md_sysfs_entry md_reshape_position =
5277 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5278        reshape_position_store);
5279
5280 static ssize_t
5281 reshape_direction_show(struct mddev *mddev, char *page)
5282 {
5283         return sprintf(page, "%s\n",
5284                        mddev->reshape_backwards ? "backwards" : "forwards");
5285 }
5286
5287 static ssize_t
5288 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5289 {
5290         int backwards = 0;
5291         int err;
5292
5293         if (cmd_match(buf, "forwards"))
5294                 backwards = 0;
5295         else if (cmd_match(buf, "backwards"))
5296                 backwards = 1;
5297         else
5298                 return -EINVAL;
5299         if (mddev->reshape_backwards == backwards)
5300                 return len;
5301
5302         err = mddev_lock(mddev);
5303         if (err)
5304                 return err;
5305         /* check if we are allowed to change */
5306         if (mddev->delta_disks)
5307                 err = -EBUSY;
5308         else if (mddev->persistent &&
5309             mddev->major_version == 0)
5310                 err =  -EINVAL;
5311         else
5312                 mddev->reshape_backwards = backwards;
5313         mddev_unlock(mddev);
5314         return err ?: len;
5315 }
5316
5317 static struct md_sysfs_entry md_reshape_direction =
5318 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5319        reshape_direction_store);
5320
5321 static ssize_t
5322 array_size_show(struct mddev *mddev, char *page)
5323 {
5324         if (mddev->external_size)
5325                 return sprintf(page, "%llu\n",
5326                                (unsigned long long)mddev->array_sectors/2);
5327         else
5328                 return sprintf(page, "default\n");
5329 }
5330
5331 static ssize_t
5332 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5333 {
5334         sector_t sectors;
5335         int err;
5336
5337         err = mddev_lock(mddev);
5338         if (err)
5339                 return err;
5340
5341         /* cluster raid doesn't support change array_sectors */
5342         if (mddev_is_clustered(mddev)) {
5343                 mddev_unlock(mddev);
5344                 return -EINVAL;
5345         }
5346
5347         if (strncmp(buf, "default", 7) == 0) {
5348                 if (mddev->pers)
5349                         sectors = mddev->pers->size(mddev, 0, 0);
5350                 else
5351                         sectors = mddev->array_sectors;
5352
5353                 mddev->external_size = 0;
5354         } else {
5355                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
5356                         err = -EINVAL;
5357                 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5358                         err = -E2BIG;
5359                 else
5360                         mddev->external_size = 1;
5361         }
5362
5363         if (!err) {
5364                 mddev->array_sectors = sectors;
5365                 if (mddev->pers)
5366                         set_capacity_and_notify(mddev->gendisk,
5367                                                 mddev->array_sectors);
5368         }
5369         mddev_unlock(mddev);
5370         return err ?: len;
5371 }
5372
5373 static struct md_sysfs_entry md_array_size =
5374 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5375        array_size_store);
5376
5377 static ssize_t
5378 consistency_policy_show(struct mddev *mddev, char *page)
5379 {
5380         int ret;
5381
5382         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5383                 ret = sprintf(page, "journal\n");
5384         } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5385                 ret = sprintf(page, "ppl\n");
5386         } else if (mddev->bitmap) {
5387                 ret = sprintf(page, "bitmap\n");
5388         } else if (mddev->pers) {
5389                 if (mddev->pers->sync_request)
5390                         ret = sprintf(page, "resync\n");
5391                 else
5392                         ret = sprintf(page, "none\n");
5393         } else {
5394                 ret = sprintf(page, "unknown\n");
5395         }
5396
5397         return ret;
5398 }
5399
5400 static ssize_t
5401 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5402 {
5403         int err = 0;
5404
5405         if (mddev->pers) {
5406                 if (mddev->pers->change_consistency_policy)
5407                         err = mddev->pers->change_consistency_policy(mddev, buf);
5408                 else
5409                         err = -EBUSY;
5410         } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5411                 set_bit(MD_HAS_PPL, &mddev->flags);
5412         } else {
5413                 err = -EINVAL;
5414         }
5415
5416         return err ? err : len;
5417 }
5418
5419 static struct md_sysfs_entry md_consistency_policy =
5420 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5421        consistency_policy_store);
5422
5423 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5424 {
5425         return sprintf(page, "%d\n", mddev->fail_last_dev);
5426 }
5427
5428 /*
5429  * Setting fail_last_dev to true to allow last device to be forcibly removed
5430  * from RAID1/RAID10.
5431  */
5432 static ssize_t
5433 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5434 {
5435         int ret;
5436         bool value;
5437
5438         ret = kstrtobool(buf, &value);
5439         if (ret)
5440                 return ret;
5441
5442         if (value != mddev->fail_last_dev)
5443                 mddev->fail_last_dev = value;
5444
5445         return len;
5446 }
5447 static struct md_sysfs_entry md_fail_last_dev =
5448 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5449        fail_last_dev_store);
5450
5451 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5452 {
5453         if (mddev->pers == NULL || (mddev->pers->level != 1))
5454                 return sprintf(page, "n/a\n");
5455         else
5456                 return sprintf(page, "%d\n", mddev->serialize_policy);
5457 }
5458
5459 /*
5460  * Setting serialize_policy to true to enforce write IO is not reordered
5461  * for raid1.
5462  */
5463 static ssize_t
5464 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5465 {
5466         int err;
5467         bool value;
5468
5469         err = kstrtobool(buf, &value);
5470         if (err)
5471                 return err;
5472
5473         if (value == mddev->serialize_policy)
5474                 return len;
5475
5476         err = mddev_lock(mddev);
5477         if (err)
5478                 return err;
5479         if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5480                 pr_err("md: serialize_policy is only effective for raid1\n");
5481                 err = -EINVAL;
5482                 goto unlock;
5483         }
5484
5485         mddev_suspend(mddev);
5486         if (value)
5487                 mddev_create_serial_pool(mddev, NULL, true);
5488         else
5489                 mddev_destroy_serial_pool(mddev, NULL, true);
5490         mddev->serialize_policy = value;
5491         mddev_resume(mddev);
5492 unlock:
5493         mddev_unlock(mddev);
5494         return err ?: len;
5495 }
5496
5497 static struct md_sysfs_entry md_serialize_policy =
5498 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5499        serialize_policy_store);
5500
5501
5502 static struct attribute *md_default_attrs[] = {
5503         &md_level.attr,
5504         &md_layout.attr,
5505         &md_raid_disks.attr,
5506         &md_uuid.attr,
5507         &md_chunk_size.attr,
5508         &md_size.attr,
5509         &md_resync_start.attr,
5510         &md_metadata.attr,
5511         &md_new_device.attr,
5512         &md_safe_delay.attr,
5513         &md_array_state.attr,
5514         &md_reshape_position.attr,
5515         &md_reshape_direction.attr,
5516         &md_array_size.attr,
5517         &max_corr_read_errors.attr,
5518         &md_consistency_policy.attr,
5519         &md_fail_last_dev.attr,
5520         &md_serialize_policy.attr,
5521         NULL,
5522 };
5523
5524 static struct attribute *md_redundancy_attrs[] = {
5525         &md_scan_mode.attr,
5526         &md_last_scan_mode.attr,
5527         &md_mismatches.attr,
5528         &md_sync_min.attr,
5529         &md_sync_max.attr,
5530         &md_sync_speed.attr,
5531         &md_sync_force_parallel.attr,
5532         &md_sync_completed.attr,
5533         &md_min_sync.attr,
5534         &md_max_sync.attr,
5535         &md_suspend_lo.attr,
5536         &md_suspend_hi.attr,
5537         &md_bitmap.attr,
5538         &md_degraded.attr,
5539         NULL,
5540 };
5541 static struct attribute_group md_redundancy_group = {
5542         .name = NULL,
5543         .attrs = md_redundancy_attrs,
5544 };
5545
5546 static ssize_t
5547 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5548 {
5549         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5550         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5551         ssize_t rv;
5552
5553         if (!entry->show)
5554                 return -EIO;
5555         spin_lock(&all_mddevs_lock);
5556         if (list_empty(&mddev->all_mddevs)) {
5557                 spin_unlock(&all_mddevs_lock);
5558                 return -EBUSY;
5559         }
5560         mddev_get(mddev);
5561         spin_unlock(&all_mddevs_lock);
5562
5563         rv = entry->show(mddev, page);
5564         mddev_put(mddev);
5565         return rv;
5566 }
5567
5568 static ssize_t
5569 md_attr_store(struct kobject *kobj, struct attribute *attr,
5570               const char *page, size_t length)
5571 {
5572         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5573         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5574         ssize_t rv;
5575
5576         if (!entry->store)
5577                 return -EIO;
5578         if (!capable(CAP_SYS_ADMIN))
5579                 return -EACCES;
5580         spin_lock(&all_mddevs_lock);
5581         if (list_empty(&mddev->all_mddevs)) {
5582                 spin_unlock(&all_mddevs_lock);
5583                 return -EBUSY;
5584         }
5585         mddev_get(mddev);
5586         spin_unlock(&all_mddevs_lock);
5587         rv = entry->store(mddev, page, length);
5588         mddev_put(mddev);
5589         return rv;
5590 }
5591
5592 static void md_free(struct kobject *ko)
5593 {
5594         struct mddev *mddev = container_of(ko, struct mddev, kobj);
5595
5596         if (mddev->sysfs_state)
5597                 sysfs_put(mddev->sysfs_state);
5598         if (mddev->sysfs_level)
5599                 sysfs_put(mddev->sysfs_level);
5600
5601         if (mddev->gendisk)
5602                 del_gendisk(mddev->gendisk);
5603         if (mddev->queue)
5604                 blk_cleanup_queue(mddev->queue);
5605         if (mddev->gendisk)
5606                 put_disk(mddev->gendisk);
5607         percpu_ref_exit(&mddev->writes_pending);
5608
5609         bioset_exit(&mddev->bio_set);
5610         bioset_exit(&mddev->sync_set);
5611         mempool_exit(&mddev->md_io_pool);
5612         kfree(mddev);
5613 }
5614
5615 static const struct sysfs_ops md_sysfs_ops = {
5616         .show   = md_attr_show,
5617         .store  = md_attr_store,
5618 };
5619 static struct kobj_type md_ktype = {
5620         .release        = md_free,
5621         .sysfs_ops      = &md_sysfs_ops,
5622         .default_attrs  = md_default_attrs,
5623 };
5624
5625 int mdp_major = 0;
5626
5627 static void mddev_delayed_delete(struct work_struct *ws)
5628 {
5629         struct mddev *mddev = container_of(ws, struct mddev, del_work);
5630
5631         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5632         kobject_del(&mddev->kobj);
5633         kobject_put(&mddev->kobj);
5634 }
5635
5636 static void no_op(struct percpu_ref *r) {}
5637
5638 int mddev_init_writes_pending(struct mddev *mddev)
5639 {
5640         if (mddev->writes_pending.percpu_count_ptr)
5641                 return 0;
5642         if (percpu_ref_init(&mddev->writes_pending, no_op,
5643                             PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5644                 return -ENOMEM;
5645         /* We want to start with the refcount at zero */
5646         percpu_ref_put(&mddev->writes_pending);
5647         return 0;
5648 }
5649 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5650
5651 static int md_alloc(dev_t dev, char *name)
5652 {
5653         /*
5654          * If dev is zero, name is the name of a device to allocate with
5655          * an arbitrary minor number.  It will be "md_???"
5656          * If dev is non-zero it must be a device number with a MAJOR of
5657          * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5658          * the device is being created by opening a node in /dev.
5659          * If "name" is not NULL, the device is being created by
5660          * writing to /sys/module/md_mod/parameters/new_array.
5661          */
5662         static DEFINE_MUTEX(disks_mutex);
5663         struct mddev *mddev;
5664         struct gendisk *disk;
5665         int partitioned;
5666         int shift;
5667         int unit;
5668         int error ;
5669
5670         /*
5671          * Wait for any previous instance of this device to be completely
5672          * removed (mddev_delayed_delete).
5673          */
5674         flush_workqueue(md_misc_wq);
5675
5676         mutex_lock(&disks_mutex);
5677         mddev = mddev_alloc(dev);
5678         if (IS_ERR(mddev)) {
5679                 mutex_unlock(&disks_mutex);
5680                 return PTR_ERR(mddev);
5681         }
5682
5683         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5684         shift = partitioned ? MdpMinorShift : 0;
5685         unit = MINOR(mddev->unit) >> shift;
5686
5687         if (name && !dev) {
5688                 /* Need to ensure that 'name' is not a duplicate.
5689                  */
5690                 struct mddev *mddev2;
5691                 spin_lock(&all_mddevs_lock);
5692
5693                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5694                         if (mddev2->gendisk &&
5695                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
5696                                 spin_unlock(&all_mddevs_lock);
5697                                 error = -EEXIST;
5698                                 goto abort;
5699                         }
5700                 spin_unlock(&all_mddevs_lock);
5701         }
5702         if (name && dev)
5703                 /*
5704                  * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5705                  */
5706                 mddev->hold_active = UNTIL_STOP;
5707
5708         error = mempool_init_kmalloc_pool(&mddev->md_io_pool, BIO_POOL_SIZE,
5709                                           sizeof(struct md_io));
5710         if (error)
5711                 goto abort;
5712
5713         error = -ENOMEM;
5714         mddev->queue = blk_alloc_queue(NUMA_NO_NODE);
5715         if (!mddev->queue)
5716                 goto abort;
5717
5718         blk_set_stacking_limits(&mddev->queue->limits);
5719
5720         disk = alloc_disk(1 << shift);
5721         if (!disk) {
5722                 blk_cleanup_queue(mddev->queue);
5723                 mddev->queue = NULL;
5724                 goto abort;
5725         }
5726         disk->major = MAJOR(mddev->unit);
5727         disk->first_minor = unit << shift;
5728         if (name)
5729                 strcpy(disk->disk_name, name);
5730         else if (partitioned)
5731                 sprintf(disk->disk_name, "md_d%d", unit);
5732         else
5733                 sprintf(disk->disk_name, "md%d", unit);
5734         disk->fops = &md_fops;
5735         disk->private_data = mddev;
5736         disk->queue = mddev->queue;
5737         blk_queue_write_cache(mddev->queue, true, true);
5738         /* Allow extended partitions.  This makes the
5739          * 'mdp' device redundant, but we can't really
5740          * remove it now.
5741          */
5742         disk->flags |= GENHD_FL_EXT_DEVT;
5743         disk->events |= DISK_EVENT_MEDIA_CHANGE;
5744         mddev->gendisk = disk;
5745         /* As soon as we call add_disk(), another thread could get
5746          * through to md_open, so make sure it doesn't get too far
5747          */
5748         mutex_lock(&mddev->open_mutex);
5749         add_disk(disk);
5750
5751         error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5752         if (error) {
5753                 /* This isn't possible, but as kobject_init_and_add is marked
5754                  * __must_check, we must do something with the result
5755                  */
5756                 pr_debug("md: cannot register %s/md - name in use\n",
5757                          disk->disk_name);
5758                 error = 0;
5759         }
5760         if (mddev->kobj.sd &&
5761             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5762                 pr_debug("pointless warning\n");
5763         mutex_unlock(&mddev->open_mutex);
5764  abort:
5765         mutex_unlock(&disks_mutex);
5766         if (!error && mddev->kobj.sd) {
5767                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5768                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5769                 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5770         }
5771         mddev_put(mddev);
5772         return error;
5773 }
5774
5775 static void md_probe(dev_t dev)
5776 {
5777         if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5778                 return;
5779         if (create_on_open)
5780                 md_alloc(dev, NULL);
5781 }
5782
5783 static int add_named_array(const char *val, const struct kernel_param *kp)
5784 {
5785         /*
5786          * val must be "md_*" or "mdNNN".
5787          * For "md_*" we allocate an array with a large free minor number, and
5788          * set the name to val.  val must not already be an active name.
5789          * For "mdNNN" we allocate an array with the minor number NNN
5790          * which must not already be in use.
5791          */
5792         int len = strlen(val);
5793         char buf[DISK_NAME_LEN];
5794         unsigned long devnum;
5795
5796         while (len && val[len-1] == '\n')
5797                 len--;
5798         if (len >= DISK_NAME_LEN)
5799                 return -E2BIG;
5800         strlcpy(buf, val, len+1);
5801         if (strncmp(buf, "md_", 3) == 0)
5802                 return md_alloc(0, buf);
5803         if (strncmp(buf, "md", 2) == 0 &&
5804             isdigit(buf[2]) &&
5805             kstrtoul(buf+2, 10, &devnum) == 0 &&
5806             devnum <= MINORMASK)
5807                 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5808
5809         return -EINVAL;
5810 }
5811
5812 static void md_safemode_timeout(struct timer_list *t)
5813 {
5814         struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5815
5816         mddev->safemode = 1;
5817         if (mddev->external)
5818                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5819
5820         md_wakeup_thread(mddev->thread);
5821 }
5822
5823 static int start_dirty_degraded;
5824
5825 int md_run(struct mddev *mddev)
5826 {
5827         int err;
5828         struct md_rdev *rdev;
5829         struct md_personality *pers;
5830
5831         if (list_empty(&mddev->disks))
5832                 /* cannot run an array with no devices.. */
5833                 return -EINVAL;
5834
5835         if (mddev->pers)
5836                 return -EBUSY;
5837         /* Cannot run until previous stop completes properly */
5838         if (mddev->sysfs_active)
5839                 return -EBUSY;
5840
5841         /*
5842          * Analyze all RAID superblock(s)
5843          */
5844         if (!mddev->raid_disks) {
5845                 if (!mddev->persistent)
5846                         return -EINVAL;
5847                 err = analyze_sbs(mddev);
5848                 if (err)
5849                         return -EINVAL;
5850         }
5851
5852         if (mddev->level != LEVEL_NONE)
5853                 request_module("md-level-%d", mddev->level);
5854         else if (mddev->clevel[0])
5855                 request_module("md-%s", mddev->clevel);
5856
5857         /*
5858          * Drop all container device buffers, from now on
5859          * the only valid external interface is through the md
5860          * device.
5861          */
5862         mddev->has_superblocks = false;
5863         rdev_for_each(rdev, mddev) {
5864                 if (test_bit(Faulty, &rdev->flags))
5865                         continue;
5866                 sync_blockdev(rdev->bdev);
5867                 invalidate_bdev(rdev->bdev);
5868                 if (mddev->ro != 1 && rdev_read_only(rdev)) {
5869                         mddev->ro = 1;
5870                         if (mddev->gendisk)
5871                                 set_disk_ro(mddev->gendisk, 1);
5872                 }
5873
5874                 if (rdev->sb_page)
5875                         mddev->has_superblocks = true;
5876
5877                 /* perform some consistency tests on the device.
5878                  * We don't want the data to overlap the metadata,
5879                  * Internal Bitmap issues have been handled elsewhere.
5880                  */
5881                 if (rdev->meta_bdev) {
5882                         /* Nothing to check */;
5883                 } else if (rdev->data_offset < rdev->sb_start) {
5884                         if (mddev->dev_sectors &&
5885                             rdev->data_offset + mddev->dev_sectors
5886                             > rdev->sb_start) {
5887                                 pr_warn("md: %s: data overlaps metadata\n",
5888                                         mdname(mddev));
5889                                 return -EINVAL;
5890                         }
5891                 } else {
5892                         if (rdev->sb_start + rdev->sb_size/512
5893                             > rdev->data_offset) {
5894                                 pr_warn("md: %s: metadata overlaps data\n",
5895                                         mdname(mddev));
5896                                 return -EINVAL;
5897                         }
5898                 }
5899                 sysfs_notify_dirent_safe(rdev->sysfs_state);
5900         }
5901
5902         if (!bioset_initialized(&mddev->bio_set)) {
5903                 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5904                 if (err)
5905                         return err;
5906         }
5907         if (!bioset_initialized(&mddev->sync_set)) {
5908                 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5909                 if (err)
5910                         return err;
5911         }
5912
5913         spin_lock(&pers_lock);
5914         pers = find_pers(mddev->level, mddev->clevel);
5915         if (!pers || !try_module_get(pers->owner)) {
5916                 spin_unlock(&pers_lock);
5917                 if (mddev->level != LEVEL_NONE)
5918                         pr_warn("md: personality for level %d is not loaded!\n",
5919                                 mddev->level);
5920                 else
5921                         pr_warn("md: personality for level %s is not loaded!\n",
5922                                 mddev->clevel);
5923                 err = -EINVAL;
5924                 goto abort;
5925         }
5926         spin_unlock(&pers_lock);
5927         if (mddev->level != pers->level) {
5928                 mddev->level = pers->level;
5929                 mddev->new_level = pers->level;
5930         }
5931         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5932
5933         if (mddev->reshape_position != MaxSector &&
5934             pers->start_reshape == NULL) {
5935                 /* This personality cannot handle reshaping... */
5936                 module_put(pers->owner);
5937                 err = -EINVAL;
5938                 goto abort;
5939         }
5940
5941         if (pers->sync_request) {
5942                 /* Warn if this is a potentially silly
5943                  * configuration.
5944                  */
5945                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5946                 struct md_rdev *rdev2;
5947                 int warned = 0;
5948
5949                 rdev_for_each(rdev, mddev)
5950                         rdev_for_each(rdev2, mddev) {
5951                                 if (rdev < rdev2 &&
5952                                     rdev->bdev->bd_disk ==
5953                                     rdev2->bdev->bd_disk) {
5954                                         pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5955                                                 mdname(mddev),
5956                                                 bdevname(rdev->bdev,b),
5957                                                 bdevname(rdev2->bdev,b2));
5958                                         warned = 1;
5959                                 }
5960                         }
5961
5962                 if (warned)
5963                         pr_warn("True protection against single-disk failure might be compromised.\n");
5964         }
5965
5966         mddev->recovery = 0;
5967         /* may be over-ridden by personality */
5968         mddev->resync_max_sectors = mddev->dev_sectors;
5969
5970         mddev->ok_start_degraded = start_dirty_degraded;
5971
5972         if (start_readonly && mddev->ro == 0)
5973                 mddev->ro = 2; /* read-only, but switch on first write */
5974
5975         err = pers->run(mddev);
5976         if (err)
5977                 pr_warn("md: pers->run() failed ...\n");
5978         else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5979                 WARN_ONCE(!mddev->external_size,
5980                           "%s: default size too small, but 'external_size' not in effect?\n",
5981                           __func__);
5982                 pr_warn("md: invalid array_size %llu > default size %llu\n",
5983                         (unsigned long long)mddev->array_sectors / 2,
5984                         (unsigned long long)pers->size(mddev, 0, 0) / 2);
5985                 err = -EINVAL;
5986         }
5987         if (err == 0 && pers->sync_request &&
5988             (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5989                 struct bitmap *bitmap;
5990
5991                 bitmap = md_bitmap_create(mddev, -1);
5992                 if (IS_ERR(bitmap)) {
5993                         err = PTR_ERR(bitmap);
5994                         pr_warn("%s: failed to create bitmap (%d)\n",
5995                                 mdname(mddev), err);
5996                 } else
5997                         mddev->bitmap = bitmap;
5998
5999         }
6000         if (err)
6001                 goto bitmap_abort;
6002
6003         if (mddev->bitmap_info.max_write_behind > 0) {
6004                 bool create_pool = false;
6005
6006                 rdev_for_each(rdev, mddev) {
6007                         if (test_bit(WriteMostly, &rdev->flags) &&
6008                             rdev_init_serial(rdev))
6009                                 create_pool = true;
6010                 }
6011                 if (create_pool && mddev->serial_info_pool == NULL) {
6012                         mddev->serial_info_pool =
6013                                 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6014                                                     sizeof(struct serial_info));
6015                         if (!mddev->serial_info_pool) {
6016                                 err = -ENOMEM;
6017                                 goto bitmap_abort;
6018                         }
6019                 }
6020         }
6021
6022         if (mddev->queue) {
6023                 bool nonrot = true;
6024
6025                 rdev_for_each(rdev, mddev) {
6026                         if (rdev->raid_disk >= 0 &&
6027                             !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6028                                 nonrot = false;
6029                                 break;
6030                         }
6031                 }
6032                 if (mddev->degraded)
6033                         nonrot = false;
6034                 if (nonrot)
6035                         blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6036                 else
6037                         blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6038         }
6039         if (pers->sync_request) {
6040                 if (mddev->kobj.sd &&
6041                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6042                         pr_warn("md: cannot register extra attributes for %s\n",
6043                                 mdname(mddev));
6044                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6045                 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6046                 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6047         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6048                 mddev->ro = 0;
6049
6050         atomic_set(&mddev->max_corr_read_errors,
6051                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6052         mddev->safemode = 0;
6053         if (mddev_is_clustered(mddev))
6054                 mddev->safemode_delay = 0;
6055         else
6056                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6057         mddev->in_sync = 1;
6058         smp_wmb();
6059         spin_lock(&mddev->lock);
6060         mddev->pers = pers;
6061         spin_unlock(&mddev->lock);
6062         rdev_for_each(rdev, mddev)
6063                 if (rdev->raid_disk >= 0)
6064                         sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6065
6066         if (mddev->degraded && !mddev->ro)
6067                 /* This ensures that recovering status is reported immediately
6068                  * via sysfs - until a lack of spares is confirmed.
6069                  */
6070                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6071         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6072
6073         if (mddev->sb_flags)
6074                 md_update_sb(mddev, 0);
6075
6076         md_new_event(mddev);
6077         return 0;
6078
6079 bitmap_abort:
6080         mddev_detach(mddev);
6081         if (mddev->private)
6082                 pers->free(mddev, mddev->private);
6083         mddev->private = NULL;
6084         module_put(pers->owner);
6085         md_bitmap_destroy(mddev);
6086 abort:
6087         bioset_exit(&mddev->bio_set);
6088         bioset_exit(&mddev->sync_set);
6089         return err;
6090 }
6091 EXPORT_SYMBOL_GPL(md_run);
6092
6093 int do_md_run(struct mddev *mddev)
6094 {
6095         int err;
6096
6097         set_bit(MD_NOT_READY, &mddev->flags);
6098         err = md_run(mddev);
6099         if (err)
6100                 goto out;
6101         err = md_bitmap_load(mddev);
6102         if (err) {
6103                 md_bitmap_destroy(mddev);
6104                 goto out;
6105         }
6106
6107         if (mddev_is_clustered(mddev))
6108                 md_allow_write(mddev);
6109
6110         /* run start up tasks that require md_thread */
6111         md_start(mddev);
6112
6113         md_wakeup_thread(mddev->thread);
6114         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6115
6116         set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6117         clear_bit(MD_NOT_READY, &mddev->flags);
6118         mddev->changed = 1;
6119         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6120         sysfs_notify_dirent_safe(mddev->sysfs_state);
6121         sysfs_notify_dirent_safe(mddev->sysfs_action);
6122         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6123 out:
6124         clear_bit(MD_NOT_READY, &mddev->flags);
6125         return err;
6126 }
6127
6128 int md_start(struct mddev *mddev)
6129 {
6130         int ret = 0;
6131
6132         if (mddev->pers->start) {
6133                 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6134                 md_wakeup_thread(mddev->thread);
6135                 ret = mddev->pers->start(mddev);
6136                 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6137                 md_wakeup_thread(mddev->sync_thread);
6138         }
6139         return ret;
6140 }
6141 EXPORT_SYMBOL_GPL(md_start);
6142
6143 static int restart_array(struct mddev *mddev)
6144 {
6145         struct gendisk *disk = mddev->gendisk;
6146         struct md_rdev *rdev;
6147         bool has_journal = false;
6148         bool has_readonly = false;
6149
6150         /* Complain if it has no devices */
6151         if (list_empty(&mddev->disks))
6152                 return -ENXIO;
6153         if (!mddev->pers)
6154                 return -EINVAL;
6155         if (!mddev->ro)
6156                 return -EBUSY;
6157
6158         rcu_read_lock();
6159         rdev_for_each_rcu(rdev, mddev) {
6160                 if (test_bit(Journal, &rdev->flags) &&
6161                     !test_bit(Faulty, &rdev->flags))
6162                         has_journal = true;
6163                 if (rdev_read_only(rdev))
6164                         has_readonly = true;
6165         }
6166         rcu_read_unlock();
6167         if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6168                 /* Don't restart rw with journal missing/faulty */
6169                         return -EINVAL;
6170         if (has_readonly)
6171                 return -EROFS;
6172
6173         mddev->safemode = 0;
6174         mddev->ro = 0;
6175         set_disk_ro(disk, 0);
6176         pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6177         /* Kick recovery or resync if necessary */
6178         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6179         md_wakeup_thread(mddev->thread);
6180         md_wakeup_thread(mddev->sync_thread);
6181         sysfs_notify_dirent_safe(mddev->sysfs_state);
6182         return 0;
6183 }
6184
6185 static void md_clean(struct mddev *mddev)
6186 {
6187         mddev->array_sectors = 0;
6188         mddev->external_size = 0;
6189         mddev->dev_sectors = 0;
6190         mddev->raid_disks = 0;
6191         mddev->recovery_cp = 0;
6192         mddev->resync_min = 0;
6193         mddev->resync_max = MaxSector;
6194         mddev->reshape_position = MaxSector;
6195         mddev->external = 0;
6196         mddev->persistent = 0;
6197         mddev->level = LEVEL_NONE;
6198         mddev->clevel[0] = 0;
6199         mddev->flags = 0;
6200         mddev->sb_flags = 0;
6201         mddev->ro = 0;
6202         mddev->metadata_type[0] = 0;
6203         mddev->chunk_sectors = 0;
6204         mddev->ctime = mddev->utime = 0;
6205         mddev->layout = 0;
6206         mddev->max_disks = 0;
6207         mddev->events = 0;
6208         mddev->can_decrease_events = 0;
6209         mddev->delta_disks = 0;
6210         mddev->reshape_backwards = 0;
6211         mddev->new_level = LEVEL_NONE;
6212         mddev->new_layout = 0;
6213         mddev->new_chunk_sectors = 0;
6214         mddev->curr_resync = 0;
6215         atomic64_set(&mddev->resync_mismatches, 0);
6216         mddev->suspend_lo = mddev->suspend_hi = 0;
6217         mddev->sync_speed_min = mddev->sync_speed_max = 0;
6218         mddev->recovery = 0;
6219         mddev->in_sync = 0;
6220         mddev->changed = 0;
6221         mddev->degraded = 0;
6222         mddev->safemode = 0;
6223         mddev->private = NULL;
6224         mddev->cluster_info = NULL;
6225         mddev->bitmap_info.offset = 0;
6226         mddev->bitmap_info.default_offset = 0;
6227         mddev->bitmap_info.default_space = 0;
6228         mddev->bitmap_info.chunksize = 0;
6229         mddev->bitmap_info.daemon_sleep = 0;
6230         mddev->bitmap_info.max_write_behind = 0;
6231         mddev->bitmap_info.nodes = 0;
6232 }
6233
6234 static void __md_stop_writes(struct mddev *mddev)
6235 {
6236         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6237         if (work_pending(&mddev->del_work))
6238                 flush_workqueue(md_misc_wq);
6239         if (mddev->sync_thread) {
6240                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6241                 md_reap_sync_thread(mddev);
6242         }
6243
6244         del_timer_sync(&mddev->safemode_timer);
6245
6246         if (mddev->pers && mddev->pers->quiesce) {
6247                 mddev->pers->quiesce(mddev, 1);
6248                 mddev->pers->quiesce(mddev, 0);
6249         }
6250         md_bitmap_flush(mddev);
6251
6252         if (mddev->ro == 0 &&
6253             ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6254              mddev->sb_flags)) {
6255                 /* mark array as shutdown cleanly */
6256                 if (!mddev_is_clustered(mddev))
6257                         mddev->in_sync = 1;
6258                 md_update_sb(mddev, 1);
6259         }
6260         /* disable policy to guarantee rdevs free resources for serialization */
6261         mddev->serialize_policy = 0;
6262         mddev_destroy_serial_pool(mddev, NULL, true);
6263 }
6264
6265 void md_stop_writes(struct mddev *mddev)
6266 {
6267         mddev_lock_nointr(mddev);
6268         __md_stop_writes(mddev);
6269         mddev_unlock(mddev);
6270 }
6271 EXPORT_SYMBOL_GPL(md_stop_writes);
6272
6273 static void mddev_detach(struct mddev *mddev)
6274 {
6275         md_bitmap_wait_behind_writes(mddev);
6276         if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6277                 mddev->pers->quiesce(mddev, 1);
6278                 mddev->pers->quiesce(mddev, 0);
6279         }
6280         md_unregister_thread(&mddev->thread);
6281         if (mddev->queue)
6282                 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6283 }
6284
6285 static void __md_stop(struct mddev *mddev)
6286 {
6287         struct md_personality *pers = mddev->pers;
6288         md_bitmap_destroy(mddev);
6289         mddev_detach(mddev);
6290         /* Ensure ->event_work is done */
6291         if (mddev->event_work.func)
6292                 flush_workqueue(md_misc_wq);
6293         spin_lock(&mddev->lock);
6294         mddev->pers = NULL;
6295         spin_unlock(&mddev->lock);
6296         pers->free(mddev, mddev->private);
6297         mddev->private = NULL;
6298         if (pers->sync_request && mddev->to_remove == NULL)
6299                 mddev->to_remove = &md_redundancy_group;
6300         module_put(pers->owner);
6301         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6302 }
6303
6304 void md_stop(struct mddev *mddev)
6305 {
6306         /* stop the array and free an attached data structures.
6307          * This is called from dm-raid
6308          */
6309         __md_stop(mddev);
6310         bioset_exit(&mddev->bio_set);
6311         bioset_exit(&mddev->sync_set);
6312 }
6313
6314 EXPORT_SYMBOL_GPL(md_stop);
6315
6316 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6317 {
6318         int err = 0;
6319         int did_freeze = 0;
6320
6321         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6322                 did_freeze = 1;
6323                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6324                 md_wakeup_thread(mddev->thread);
6325         }
6326         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6327                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6328         if (mddev->sync_thread)
6329                 /* Thread might be blocked waiting for metadata update
6330                  * which will now never happen */
6331                 wake_up_process(mddev->sync_thread->tsk);
6332
6333         if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6334                 return -EBUSY;
6335         mddev_unlock(mddev);
6336         wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6337                                           &mddev->recovery));
6338         wait_event(mddev->sb_wait,
6339                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6340         mddev_lock_nointr(mddev);
6341
6342         mutex_lock(&mddev->open_mutex);
6343         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6344             mddev->sync_thread ||
6345             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6346                 pr_warn("md: %s still in use.\n",mdname(mddev));
6347                 if (did_freeze) {
6348                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6349                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6350                         md_wakeup_thread(mddev->thread);
6351                 }
6352                 err = -EBUSY;
6353                 goto out;
6354         }
6355         if (mddev->pers) {
6356                 __md_stop_writes(mddev);
6357
6358                 err  = -ENXIO;
6359                 if (mddev->ro==1)
6360                         goto out;
6361                 mddev->ro = 1;
6362                 set_disk_ro(mddev->gendisk, 1);
6363                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6364                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6365                 md_wakeup_thread(mddev->thread);
6366                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6367                 err = 0;
6368         }
6369 out:
6370         mutex_unlock(&mddev->open_mutex);
6371         return err;
6372 }
6373
6374 /* mode:
6375  *   0 - completely stop and dis-assemble array
6376  *   2 - stop but do not disassemble array
6377  */
6378 static int do_md_stop(struct mddev *mddev, int mode,
6379                       struct block_device *bdev)
6380 {
6381         struct gendisk *disk = mddev->gendisk;
6382         struct md_rdev *rdev;
6383         int did_freeze = 0;
6384
6385         if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6386                 did_freeze = 1;
6387                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6388                 md_wakeup_thread(mddev->thread);
6389         }
6390         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6391                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6392         if (mddev->sync_thread)
6393                 /* Thread might be blocked waiting for metadata update
6394                  * which will now never happen */
6395                 wake_up_process(mddev->sync_thread->tsk);
6396
6397         mddev_unlock(mddev);
6398         wait_event(resync_wait, (mddev->sync_thread == NULL &&
6399                                  !test_bit(MD_RECOVERY_RUNNING,
6400                                            &mddev->recovery)));
6401         mddev_lock_nointr(mddev);
6402
6403         mutex_lock(&mddev->open_mutex);
6404         if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6405             mddev->sysfs_active ||
6406             mddev->sync_thread ||
6407             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6408                 pr_warn("md: %s still in use.\n",mdname(mddev));
6409                 mutex_unlock(&mddev->open_mutex);
6410                 if (did_freeze) {
6411                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6412                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6413                         md_wakeup_thread(mddev->thread);
6414                 }
6415                 return -EBUSY;
6416         }
6417         if (mddev->pers) {
6418                 if (mddev->ro)
6419                         set_disk_ro(disk, 0);
6420
6421                 __md_stop_writes(mddev);
6422                 __md_stop(mddev);
6423
6424                 /* tell userspace to handle 'inactive' */
6425                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6426
6427                 rdev_for_each(rdev, mddev)
6428                         if (rdev->raid_disk >= 0)
6429                                 sysfs_unlink_rdev(mddev, rdev);
6430
6431                 set_capacity_and_notify(disk, 0);
6432                 mutex_unlock(&mddev->open_mutex);
6433                 mddev->changed = 1;
6434
6435                 if (mddev->ro)
6436                         mddev->ro = 0;
6437         } else
6438                 mutex_unlock(&mddev->open_mutex);
6439         /*
6440          * Free resources if final stop
6441          */
6442         if (mode == 0) {
6443                 pr_info("md: %s stopped.\n", mdname(mddev));
6444
6445                 if (mddev->bitmap_info.file) {
6446                         struct file *f = mddev->bitmap_info.file;
6447                         spin_lock(&mddev->lock);
6448                         mddev->bitmap_info.file = NULL;
6449                         spin_unlock(&mddev->lock);
6450                         fput(f);
6451                 }
6452                 mddev->bitmap_info.offset = 0;
6453
6454                 export_array(mddev);
6455
6456                 md_clean(mddev);
6457                 if (mddev->hold_active == UNTIL_STOP)
6458                         mddev->hold_active = 0;
6459         }
6460         md_new_event(mddev);
6461         sysfs_notify_dirent_safe(mddev->sysfs_state);
6462         return 0;
6463 }
6464
6465 #ifndef MODULE
6466 static void autorun_array(struct mddev *mddev)
6467 {
6468         struct md_rdev *rdev;
6469         int err;
6470
6471         if (list_empty(&mddev->disks))
6472                 return;
6473
6474         pr_info("md: running: ");
6475
6476         rdev_for_each(rdev, mddev) {
6477                 char b[BDEVNAME_SIZE];
6478                 pr_cont("<%s>", bdevname(rdev->bdev,b));
6479         }
6480         pr_cont("\n");
6481
6482         err = do_md_run(mddev);
6483         if (err) {
6484                 pr_warn("md: do_md_run() returned %d\n", err);
6485                 do_md_stop(mddev, 0, NULL);
6486         }
6487 }
6488
6489 /*
6490  * lets try to run arrays based on all disks that have arrived
6491  * until now. (those are in pending_raid_disks)
6492  *
6493  * the method: pick the first pending disk, collect all disks with
6494  * the same UUID, remove all from the pending list and put them into
6495  * the 'same_array' list. Then order this list based on superblock
6496  * update time (freshest comes first), kick out 'old' disks and
6497  * compare superblocks. If everything's fine then run it.
6498  *
6499  * If "unit" is allocated, then bump its reference count
6500  */
6501 static void autorun_devices(int part)
6502 {
6503         struct md_rdev *rdev0, *rdev, *tmp;
6504         struct mddev *mddev;
6505         char b[BDEVNAME_SIZE];
6506
6507         pr_info("md: autorun ...\n");
6508         while (!list_empty(&pending_raid_disks)) {
6509                 int unit;
6510                 dev_t dev;
6511                 LIST_HEAD(candidates);
6512                 rdev0 = list_entry(pending_raid_disks.next,
6513                                          struct md_rdev, same_set);
6514
6515                 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6516                 INIT_LIST_HEAD(&candidates);
6517                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6518                         if (super_90_load(rdev, rdev0, 0) >= 0) {
6519                                 pr_debug("md:  adding %s ...\n",
6520                                          bdevname(rdev->bdev,b));
6521                                 list_move(&rdev->same_set, &candidates);
6522                         }
6523                 /*
6524                  * now we have a set of devices, with all of them having
6525                  * mostly sane superblocks. It's time to allocate the
6526                  * mddev.
6527                  */
6528                 if (part) {
6529                         dev = MKDEV(mdp_major,
6530                                     rdev0->preferred_minor << MdpMinorShift);
6531                         unit = MINOR(dev) >> MdpMinorShift;
6532                 } else {
6533                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6534                         unit = MINOR(dev);
6535                 }
6536                 if (rdev0->preferred_minor != unit) {
6537                         pr_warn("md: unit number in %s is bad: %d\n",
6538                                 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6539                         break;
6540                 }
6541
6542                 md_probe(dev);
6543                 mddev = mddev_find(dev);
6544                 if (!mddev)
6545                         break;
6546
6547                 if (mddev_lock(mddev))
6548                         pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6549                 else if (mddev->raid_disks || mddev->major_version
6550                          || !list_empty(&mddev->disks)) {
6551                         pr_warn("md: %s already running, cannot run %s\n",
6552                                 mdname(mddev), bdevname(rdev0->bdev,b));
6553                         mddev_unlock(mddev);
6554                 } else {
6555                         pr_debug("md: created %s\n", mdname(mddev));
6556                         mddev->persistent = 1;
6557                         rdev_for_each_list(rdev, tmp, &candidates) {
6558                                 list_del_init(&rdev->same_set);
6559                                 if (bind_rdev_to_array(rdev, mddev))
6560                                         export_rdev(rdev);
6561                         }
6562                         autorun_array(mddev);
6563                         mddev_unlock(mddev);
6564                 }
6565                 /* on success, candidates will be empty, on error
6566                  * it won't...
6567                  */
6568                 rdev_for_each_list(rdev, tmp, &candidates) {
6569                         list_del_init(&rdev->same_set);
6570                         export_rdev(rdev);
6571                 }
6572                 mddev_put(mddev);
6573         }
6574         pr_info("md: ... autorun DONE.\n");
6575 }
6576 #endif /* !MODULE */
6577
6578 static int get_version(void __user *arg)
6579 {
6580         mdu_version_t ver;
6581
6582         ver.major = MD_MAJOR_VERSION;
6583         ver.minor = MD_MINOR_VERSION;
6584         ver.patchlevel = MD_PATCHLEVEL_VERSION;
6585
6586         if (copy_to_user(arg, &ver, sizeof(ver)))
6587                 return -EFAULT;
6588
6589         return 0;
6590 }
6591
6592 static int get_array_info(struct mddev *mddev, void __user *arg)
6593 {
6594         mdu_array_info_t info;
6595         int nr,working,insync,failed,spare;
6596         struct md_rdev *rdev;
6597
6598         nr = working = insync = failed = spare = 0;
6599         rcu_read_lock();
6600         rdev_for_each_rcu(rdev, mddev) {
6601                 nr++;
6602                 if (test_bit(Faulty, &rdev->flags))
6603                         failed++;
6604                 else {
6605                         working++;
6606                         if (test_bit(In_sync, &rdev->flags))
6607                                 insync++;
6608                         else if (test_bit(Journal, &rdev->flags))
6609                                 /* TODO: add journal count to md_u.h */
6610                                 ;
6611                         else
6612                                 spare++;
6613                 }
6614         }
6615         rcu_read_unlock();
6616
6617         info.major_version = mddev->major_version;
6618         info.minor_version = mddev->minor_version;
6619         info.patch_version = MD_PATCHLEVEL_VERSION;
6620         info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6621         info.level         = mddev->level;
6622         info.size          = mddev->dev_sectors / 2;
6623         if (info.size != mddev->dev_sectors / 2) /* overflow */
6624                 info.size = -1;
6625         info.nr_disks      = nr;
6626         info.raid_disks    = mddev->raid_disks;
6627         info.md_minor      = mddev->md_minor;
6628         info.not_persistent= !mddev->persistent;
6629
6630         info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6631         info.state         = 0;
6632         if (mddev->in_sync)
6633                 info.state = (1<<MD_SB_CLEAN);
6634         if (mddev->bitmap && mddev->bitmap_info.offset)
6635                 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6636         if (mddev_is_clustered(mddev))
6637                 info.state |= (1<<MD_SB_CLUSTERED);
6638         info.active_disks  = insync;
6639         info.working_disks = working;
6640         info.failed_disks  = failed;
6641         info.spare_disks   = spare;
6642
6643         info.layout        = mddev->layout;
6644         info.chunk_size    = mddev->chunk_sectors << 9;
6645
6646         if (copy_to_user(arg, &info, sizeof(info)))
6647                 return -EFAULT;
6648
6649         return 0;
6650 }
6651
6652 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6653 {
6654         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6655         char *ptr;
6656         int err;
6657
6658         file = kzalloc(sizeof(*file), GFP_NOIO);
6659         if (!file)
6660                 return -ENOMEM;
6661
6662         err = 0;
6663         spin_lock(&mddev->lock);
6664         /* bitmap enabled */
6665         if (mddev->bitmap_info.file) {
6666                 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6667                                 sizeof(file->pathname));
6668                 if (IS_ERR(ptr))
6669                         err = PTR_ERR(ptr);
6670                 else
6671                         memmove(file->pathname, ptr,
6672                                 sizeof(file->pathname)-(ptr-file->pathname));
6673         }
6674         spin_unlock(&mddev->lock);
6675
6676         if (err == 0 &&
6677             copy_to_user(arg, file, sizeof(*file)))
6678                 err = -EFAULT;
6679
6680         kfree(file);
6681         return err;
6682 }
6683
6684 static int get_disk_info(struct mddev *mddev, void __user * arg)
6685 {
6686         mdu_disk_info_t info;
6687         struct md_rdev *rdev;
6688
6689         if (copy_from_user(&info, arg, sizeof(info)))
6690                 return -EFAULT;
6691
6692         rcu_read_lock();
6693         rdev = md_find_rdev_nr_rcu(mddev, info.number);
6694         if (rdev) {
6695                 info.major = MAJOR(rdev->bdev->bd_dev);
6696                 info.minor = MINOR(rdev->bdev->bd_dev);
6697                 info.raid_disk = rdev->raid_disk;
6698                 info.state = 0;
6699                 if (test_bit(Faulty, &rdev->flags))
6700                         info.state |= (1<<MD_DISK_FAULTY);
6701                 else if (test_bit(In_sync, &rdev->flags)) {
6702                         info.state |= (1<<MD_DISK_ACTIVE);
6703                         info.state |= (1<<MD_DISK_SYNC);
6704                 }
6705                 if (test_bit(Journal, &rdev->flags))
6706                         info.state |= (1<<MD_DISK_JOURNAL);
6707                 if (test_bit(WriteMostly, &rdev->flags))
6708                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
6709                 if (test_bit(FailFast, &rdev->flags))
6710                         info.state |= (1<<MD_DISK_FAILFAST);
6711         } else {
6712                 info.major = info.minor = 0;
6713                 info.raid_disk = -1;
6714                 info.state = (1<<MD_DISK_REMOVED);
6715         }
6716         rcu_read_unlock();
6717
6718         if (copy_to_user(arg, &info, sizeof(info)))
6719                 return -EFAULT;
6720
6721         return 0;
6722 }
6723
6724 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6725 {
6726         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6727         struct md_rdev *rdev;
6728         dev_t dev = MKDEV(info->major,info->minor);
6729
6730         if (mddev_is_clustered(mddev) &&
6731                 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6732                 pr_warn("%s: Cannot add to clustered mddev.\n",
6733                         mdname(mddev));
6734                 return -EINVAL;
6735         }
6736
6737         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6738                 return -EOVERFLOW;
6739
6740         if (!mddev->raid_disks) {
6741                 int err;
6742                 /* expecting a device which has a superblock */
6743                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6744                 if (IS_ERR(rdev)) {
6745                         pr_warn("md: md_import_device returned %ld\n",
6746                                 PTR_ERR(rdev));
6747                         return PTR_ERR(rdev);
6748                 }
6749                 if (!list_empty(&mddev->disks)) {
6750                         struct md_rdev *rdev0
6751                                 = list_entry(mddev->disks.next,
6752                                              struct md_rdev, same_set);
6753                         err = super_types[mddev->major_version]
6754                                 .load_super(rdev, rdev0, mddev->minor_version);
6755                         if (err < 0) {
6756                                 pr_warn("md: %s has different UUID to %s\n",
6757                                         bdevname(rdev->bdev,b),
6758                                         bdevname(rdev0->bdev,b2));
6759                                 export_rdev(rdev);
6760                                 return -EINVAL;
6761                         }
6762                 }
6763                 err = bind_rdev_to_array(rdev, mddev);
6764                 if (err)
6765                         export_rdev(rdev);
6766                 return err;
6767         }
6768
6769         /*
6770          * md_add_new_disk can be used once the array is assembled
6771          * to add "hot spares".  They must already have a superblock
6772          * written
6773          */
6774         if (mddev->pers) {
6775                 int err;
6776                 if (!mddev->pers->hot_add_disk) {
6777                         pr_warn("%s: personality does not support diskops!\n",
6778                                 mdname(mddev));
6779                         return -EINVAL;
6780                 }
6781                 if (mddev->persistent)
6782                         rdev = md_import_device(dev, mddev->major_version,
6783                                                 mddev->minor_version);
6784                 else
6785                         rdev = md_import_device(dev, -1, -1);
6786                 if (IS_ERR(rdev)) {
6787                         pr_warn("md: md_import_device returned %ld\n",
6788                                 PTR_ERR(rdev));
6789                         return PTR_ERR(rdev);
6790                 }
6791                 /* set saved_raid_disk if appropriate */
6792                 if (!mddev->persistent) {
6793                         if (info->state & (1<<MD_DISK_SYNC)  &&
6794                             info->raid_disk < mddev->raid_disks) {
6795                                 rdev->raid_disk = info->raid_disk;
6796                                 set_bit(In_sync, &rdev->flags);
6797                                 clear_bit(Bitmap_sync, &rdev->flags);
6798                         } else
6799                                 rdev->raid_disk = -1;
6800                         rdev->saved_raid_disk = rdev->raid_disk;
6801                 } else
6802                         super_types[mddev->major_version].
6803                                 validate_super(mddev, rdev);
6804                 if ((info->state & (1<<MD_DISK_SYNC)) &&
6805                      rdev->raid_disk != info->raid_disk) {
6806                         /* This was a hot-add request, but events doesn't
6807                          * match, so reject it.
6808                          */
6809                         export_rdev(rdev);
6810                         return -EINVAL;
6811                 }
6812
6813                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6814                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6815                         set_bit(WriteMostly, &rdev->flags);
6816                 else
6817                         clear_bit(WriteMostly, &rdev->flags);
6818                 if (info->state & (1<<MD_DISK_FAILFAST))
6819                         set_bit(FailFast, &rdev->flags);
6820                 else
6821                         clear_bit(FailFast, &rdev->flags);
6822
6823                 if (info->state & (1<<MD_DISK_JOURNAL)) {
6824                         struct md_rdev *rdev2;
6825                         bool has_journal = false;
6826
6827                         /* make sure no existing journal disk */
6828                         rdev_for_each(rdev2, mddev) {
6829                                 if (test_bit(Journal, &rdev2->flags)) {
6830                                         has_journal = true;
6831                                         break;
6832                                 }
6833                         }
6834                         if (has_journal || mddev->bitmap) {
6835                                 export_rdev(rdev);
6836                                 return -EBUSY;
6837                         }
6838                         set_bit(Journal, &rdev->flags);
6839                 }
6840                 /*
6841                  * check whether the device shows up in other nodes
6842                  */
6843                 if (mddev_is_clustered(mddev)) {
6844                         if (info->state & (1 << MD_DISK_CANDIDATE))
6845                                 set_bit(Candidate, &rdev->flags);
6846                         else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6847                                 /* --add initiated by this node */
6848                                 err = md_cluster_ops->add_new_disk(mddev, rdev);
6849                                 if (err) {
6850                                         export_rdev(rdev);
6851                                         return err;
6852                                 }
6853                         }
6854                 }
6855
6856                 rdev->raid_disk = -1;
6857                 err = bind_rdev_to_array(rdev, mddev);
6858
6859                 if (err)
6860                         export_rdev(rdev);
6861
6862                 if (mddev_is_clustered(mddev)) {
6863                         if (info->state & (1 << MD_DISK_CANDIDATE)) {
6864                                 if (!err) {
6865                                         err = md_cluster_ops->new_disk_ack(mddev,
6866                                                 err == 0);
6867                                         if (err)
6868                                                 md_kick_rdev_from_array(rdev);
6869                                 }
6870                         } else {
6871                                 if (err)
6872                                         md_cluster_ops->add_new_disk_cancel(mddev);
6873                                 else
6874                                         err = add_bound_rdev(rdev);
6875                         }
6876
6877                 } else if (!err)
6878                         err = add_bound_rdev(rdev);
6879
6880                 return err;
6881         }
6882
6883         /* otherwise, md_add_new_disk is only allowed
6884          * for major_version==0 superblocks
6885          */
6886         if (mddev->major_version != 0) {
6887                 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6888                 return -EINVAL;
6889         }
6890
6891         if (!(info->state & (1<<MD_DISK_FAULTY))) {
6892                 int err;
6893                 rdev = md_import_device(dev, -1, 0);
6894                 if (IS_ERR(rdev)) {
6895                         pr_warn("md: error, md_import_device() returned %ld\n",
6896                                 PTR_ERR(rdev));
6897                         return PTR_ERR(rdev);
6898                 }
6899                 rdev->desc_nr = info->number;
6900                 if (info->raid_disk < mddev->raid_disks)
6901                         rdev->raid_disk = info->raid_disk;
6902                 else
6903                         rdev->raid_disk = -1;
6904
6905                 if (rdev->raid_disk < mddev->raid_disks)
6906                         if (info->state & (1<<MD_DISK_SYNC))
6907                                 set_bit(In_sync, &rdev->flags);
6908
6909                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6910                         set_bit(WriteMostly, &rdev->flags);
6911                 if (info->state & (1<<MD_DISK_FAILFAST))
6912                         set_bit(FailFast, &rdev->flags);
6913
6914                 if (!mddev->persistent) {
6915                         pr_debug("md: nonpersistent superblock ...\n");
6916                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6917                 } else
6918                         rdev->sb_start = calc_dev_sboffset(rdev);
6919                 rdev->sectors = rdev->sb_start;
6920
6921                 err = bind_rdev_to_array(rdev, mddev);
6922                 if (err) {
6923                         export_rdev(rdev);
6924                         return err;
6925                 }
6926         }
6927
6928         return 0;
6929 }
6930
6931 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6932 {
6933         char b[BDEVNAME_SIZE];
6934         struct md_rdev *rdev;
6935
6936         if (!mddev->pers)
6937                 return -ENODEV;
6938
6939         rdev = find_rdev(mddev, dev);
6940         if (!rdev)
6941                 return -ENXIO;
6942
6943         if (rdev->raid_disk < 0)
6944                 goto kick_rdev;
6945
6946         clear_bit(Blocked, &rdev->flags);
6947         remove_and_add_spares(mddev, rdev);
6948
6949         if (rdev->raid_disk >= 0)
6950                 goto busy;
6951
6952 kick_rdev:
6953         if (mddev_is_clustered(mddev)) {
6954                 if (md_cluster_ops->remove_disk(mddev, rdev))
6955                         goto busy;
6956         }
6957
6958         md_kick_rdev_from_array(rdev);
6959         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6960         if (mddev->thread)
6961                 md_wakeup_thread(mddev->thread);
6962         else
6963                 md_update_sb(mddev, 1);
6964         md_new_event(mddev);
6965
6966         return 0;
6967 busy:
6968         pr_debug("md: cannot remove active disk %s from %s ...\n",
6969                  bdevname(rdev->bdev,b), mdname(mddev));
6970         return -EBUSY;
6971 }
6972
6973 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6974 {
6975         char b[BDEVNAME_SIZE];
6976         int err;
6977         struct md_rdev *rdev;
6978
6979         if (!mddev->pers)
6980                 return -ENODEV;
6981
6982         if (mddev->major_version != 0) {
6983                 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6984                         mdname(mddev));
6985                 return -EINVAL;
6986         }
6987         if (!mddev->pers->hot_add_disk) {
6988                 pr_warn("%s: personality does not support diskops!\n",
6989                         mdname(mddev));
6990                 return -EINVAL;
6991         }
6992
6993         rdev = md_import_device(dev, -1, 0);
6994         if (IS_ERR(rdev)) {
6995                 pr_warn("md: error, md_import_device() returned %ld\n",
6996                         PTR_ERR(rdev));
6997                 return -EINVAL;
6998         }
6999
7000         if (mddev->persistent)
7001                 rdev->sb_start = calc_dev_sboffset(rdev);
7002         else
7003                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
7004
7005         rdev->sectors = rdev->sb_start;
7006
7007         if (test_bit(Faulty, &rdev->flags)) {
7008                 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
7009                         bdevname(rdev->bdev,b), mdname(mddev));
7010                 err = -EINVAL;
7011                 goto abort_export;
7012         }
7013
7014         clear_bit(In_sync, &rdev->flags);
7015         rdev->desc_nr = -1;
7016         rdev->saved_raid_disk = -1;
7017         err = bind_rdev_to_array(rdev, mddev);
7018         if (err)
7019                 goto abort_export;
7020
7021         /*
7022          * The rest should better be atomic, we can have disk failures
7023          * noticed in interrupt contexts ...
7024          */
7025
7026         rdev->raid_disk = -1;
7027
7028         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7029         if (!mddev->thread)
7030                 md_update_sb(mddev, 1);
7031         /*
7032          * Kick recovery, maybe this spare has to be added to the
7033          * array immediately.
7034          */
7035         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7036         md_wakeup_thread(mddev->thread);
7037         md_new_event(mddev);
7038         return 0;
7039
7040 abort_export:
7041         export_rdev(rdev);
7042         return err;
7043 }
7044
7045 static int set_bitmap_file(struct mddev *mddev, int fd)
7046 {
7047         int err = 0;
7048
7049         if (mddev->pers) {
7050                 if (!mddev->pers->quiesce || !mddev->thread)
7051                         return -EBUSY;
7052                 if (mddev->recovery || mddev->sync_thread)
7053                         return -EBUSY;
7054                 /* we should be able to change the bitmap.. */
7055         }
7056
7057         if (fd >= 0) {
7058                 struct inode *inode;
7059                 struct file *f;
7060
7061                 if (mddev->bitmap || mddev->bitmap_info.file)
7062                         return -EEXIST; /* cannot add when bitmap is present */
7063                 f = fget(fd);
7064
7065                 if (f == NULL) {
7066                         pr_warn("%s: error: failed to get bitmap file\n",
7067                                 mdname(mddev));
7068                         return -EBADF;
7069                 }
7070
7071                 inode = f->f_mapping->host;
7072                 if (!S_ISREG(inode->i_mode)) {
7073                         pr_warn("%s: error: bitmap file must be a regular file\n",
7074                                 mdname(mddev));
7075                         err = -EBADF;
7076                 } else if (!(f->f_mode & FMODE_WRITE)) {
7077                         pr_warn("%s: error: bitmap file must open for write\n",
7078                                 mdname(mddev));
7079                         err = -EBADF;
7080                 } else if (atomic_read(&inode->i_writecount) != 1) {
7081                         pr_warn("%s: error: bitmap file is already in use\n",
7082                                 mdname(mddev));
7083                         err = -EBUSY;
7084                 }
7085                 if (err) {
7086                         fput(f);
7087                         return err;
7088                 }
7089                 mddev->bitmap_info.file = f;
7090                 mddev->bitmap_info.offset = 0; /* file overrides offset */
7091         } else if (mddev->bitmap == NULL)
7092                 return -ENOENT; /* cannot remove what isn't there */
7093         err = 0;
7094         if (mddev->pers) {
7095                 if (fd >= 0) {
7096                         struct bitmap *bitmap;
7097
7098                         bitmap = md_bitmap_create(mddev, -1);
7099                         mddev_suspend(mddev);
7100                         if (!IS_ERR(bitmap)) {
7101                                 mddev->bitmap = bitmap;
7102                                 err = md_bitmap_load(mddev);
7103                         } else
7104                                 err = PTR_ERR(bitmap);
7105                         if (err) {
7106                                 md_bitmap_destroy(mddev);
7107                                 fd = -1;
7108                         }
7109                         mddev_resume(mddev);
7110                 } else if (fd < 0) {
7111                         mddev_suspend(mddev);
7112                         md_bitmap_destroy(mddev);
7113                         mddev_resume(mddev);
7114                 }
7115         }
7116         if (fd < 0) {
7117                 struct file *f = mddev->bitmap_info.file;
7118                 if (f) {
7119                         spin_lock(&mddev->lock);
7120                         mddev->bitmap_info.file = NULL;
7121                         spin_unlock(&mddev->lock);
7122                         fput(f);
7123                 }
7124         }
7125
7126         return err;
7127 }
7128
7129 /*
7130  * md_set_array_info is used two different ways
7131  * The original usage is when creating a new array.
7132  * In this usage, raid_disks is > 0 and it together with
7133  *  level, size, not_persistent,layout,chunksize determine the
7134  *  shape of the array.
7135  *  This will always create an array with a type-0.90.0 superblock.
7136  * The newer usage is when assembling an array.
7137  *  In this case raid_disks will be 0, and the major_version field is
7138  *  use to determine which style super-blocks are to be found on the devices.
7139  *  The minor and patch _version numbers are also kept incase the
7140  *  super_block handler wishes to interpret them.
7141  */
7142 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7143 {
7144         if (info->raid_disks == 0) {
7145                 /* just setting version number for superblock loading */
7146                 if (info->major_version < 0 ||
7147                     info->major_version >= ARRAY_SIZE(super_types) ||
7148                     super_types[info->major_version].name == NULL) {
7149                         /* maybe try to auto-load a module? */
7150                         pr_warn("md: superblock version %d not known\n",
7151                                 info->major_version);
7152                         return -EINVAL;
7153                 }
7154                 mddev->major_version = info->major_version;
7155                 mddev->minor_version = info->minor_version;
7156                 mddev->patch_version = info->patch_version;
7157                 mddev->persistent = !info->not_persistent;
7158                 /* ensure mddev_put doesn't delete this now that there
7159                  * is some minimal configuration.
7160                  */
7161                 mddev->ctime         = ktime_get_real_seconds();
7162                 return 0;
7163         }
7164         mddev->major_version = MD_MAJOR_VERSION;
7165         mddev->minor_version = MD_MINOR_VERSION;
7166         mddev->patch_version = MD_PATCHLEVEL_VERSION;
7167         mddev->ctime         = ktime_get_real_seconds();
7168
7169         mddev->level         = info->level;
7170         mddev->clevel[0]     = 0;
7171         mddev->dev_sectors   = 2 * (sector_t)info->size;
7172         mddev->raid_disks    = info->raid_disks;
7173         /* don't set md_minor, it is determined by which /dev/md* was
7174          * openned
7175          */
7176         if (info->state & (1<<MD_SB_CLEAN))
7177                 mddev->recovery_cp = MaxSector;
7178         else
7179                 mddev->recovery_cp = 0;
7180         mddev->persistent    = ! info->not_persistent;
7181         mddev->external      = 0;
7182
7183         mddev->layout        = info->layout;
7184         if (mddev->level == 0)
7185                 /* Cannot trust RAID0 layout info here */
7186                 mddev->layout = -1;
7187         mddev->chunk_sectors = info->chunk_size >> 9;
7188
7189         if (mddev->persistent) {
7190                 mddev->max_disks = MD_SB_DISKS;
7191                 mddev->flags = 0;
7192                 mddev->sb_flags = 0;
7193         }
7194         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7195
7196         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7197         mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7198         mddev->bitmap_info.offset = 0;
7199
7200         mddev->reshape_position = MaxSector;
7201
7202         /*
7203          * Generate a 128 bit UUID
7204          */
7205         get_random_bytes(mddev->uuid, 16);
7206
7207         mddev->new_level = mddev->level;
7208         mddev->new_chunk_sectors = mddev->chunk_sectors;
7209         mddev->new_layout = mddev->layout;
7210         mddev->delta_disks = 0;
7211         mddev->reshape_backwards = 0;
7212
7213         return 0;
7214 }
7215
7216 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7217 {
7218         lockdep_assert_held(&mddev->reconfig_mutex);
7219
7220         if (mddev->external_size)
7221                 return;
7222
7223         mddev->array_sectors = array_sectors;
7224 }
7225 EXPORT_SYMBOL(md_set_array_sectors);
7226
7227 static int update_size(struct mddev *mddev, sector_t num_sectors)
7228 {
7229         struct md_rdev *rdev;
7230         int rv;
7231         int fit = (num_sectors == 0);
7232         sector_t old_dev_sectors = mddev->dev_sectors;
7233
7234         if (mddev->pers->resize == NULL)
7235                 return -EINVAL;
7236         /* The "num_sectors" is the number of sectors of each device that
7237          * is used.  This can only make sense for arrays with redundancy.
7238          * linear and raid0 always use whatever space is available. We can only
7239          * consider changing this number if no resync or reconstruction is
7240          * happening, and if the new size is acceptable. It must fit before the
7241          * sb_start or, if that is <data_offset, it must fit before the size
7242          * of each device.  If num_sectors is zero, we find the largest size
7243          * that fits.
7244          */
7245         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7246             mddev->sync_thread)
7247                 return -EBUSY;
7248         if (mddev->ro)
7249                 return -EROFS;
7250
7251         rdev_for_each(rdev, mddev) {
7252                 sector_t avail = rdev->sectors;
7253
7254                 if (fit && (num_sectors == 0 || num_sectors > avail))
7255                         num_sectors = avail;
7256                 if (avail < num_sectors)
7257                         return -ENOSPC;
7258         }
7259         rv = mddev->pers->resize(mddev, num_sectors);
7260         if (!rv) {
7261                 if (mddev_is_clustered(mddev))
7262                         md_cluster_ops->update_size(mddev, old_dev_sectors);
7263                 else if (mddev->queue) {
7264                         set_capacity_and_notify(mddev->gendisk,
7265                                                 mddev->array_sectors);
7266                 }
7267         }
7268         return rv;
7269 }
7270
7271 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7272 {
7273         int rv;
7274         struct md_rdev *rdev;
7275         /* change the number of raid disks */
7276         if (mddev->pers->check_reshape == NULL)
7277                 return -EINVAL;
7278         if (mddev->ro)
7279                 return -EROFS;
7280         if (raid_disks <= 0 ||
7281             (mddev->max_disks && raid_disks >= mddev->max_disks))
7282                 return -EINVAL;
7283         if (mddev->sync_thread ||
7284             test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7285             test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7286             mddev->reshape_position != MaxSector)
7287                 return -EBUSY;
7288
7289         rdev_for_each(rdev, mddev) {
7290                 if (mddev->raid_disks < raid_disks &&
7291                     rdev->data_offset < rdev->new_data_offset)
7292                         return -EINVAL;
7293                 if (mddev->raid_disks > raid_disks &&
7294                     rdev->data_offset > rdev->new_data_offset)
7295                         return -EINVAL;
7296         }
7297
7298         mddev->delta_disks = raid_disks - mddev->raid_disks;
7299         if (mddev->delta_disks < 0)
7300                 mddev->reshape_backwards = 1;
7301         else if (mddev->delta_disks > 0)
7302                 mddev->reshape_backwards = 0;
7303
7304         rv = mddev->pers->check_reshape(mddev);
7305         if (rv < 0) {
7306                 mddev->delta_disks = 0;
7307                 mddev->reshape_backwards = 0;
7308         }
7309         return rv;
7310 }
7311
7312 /*
7313  * update_array_info is used to change the configuration of an
7314  * on-line array.
7315  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7316  * fields in the info are checked against the array.
7317  * Any differences that cannot be handled will cause an error.
7318  * Normally, only one change can be managed at a time.
7319  */
7320 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7321 {
7322         int rv = 0;
7323         int cnt = 0;
7324         int state = 0;
7325
7326         /* calculate expected state,ignoring low bits */
7327         if (mddev->bitmap && mddev->bitmap_info.offset)
7328                 state |= (1 << MD_SB_BITMAP_PRESENT);
7329
7330         if (mddev->major_version != info->major_version ||
7331             mddev->minor_version != info->minor_version ||
7332 /*          mddev->patch_version != info->patch_version || */
7333             mddev->ctime         != info->ctime         ||
7334             mddev->level         != info->level         ||
7335 /*          mddev->layout        != info->layout        || */
7336             mddev->persistent    != !info->not_persistent ||
7337             mddev->chunk_sectors != info->chunk_size >> 9 ||
7338             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7339             ((state^info->state) & 0xfffffe00)
7340                 )
7341                 return -EINVAL;
7342         /* Check there is only one change */
7343         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7344                 cnt++;
7345         if (mddev->raid_disks != info->raid_disks)
7346                 cnt++;
7347         if (mddev->layout != info->layout)
7348                 cnt++;
7349         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7350                 cnt++;
7351         if (cnt == 0)
7352                 return 0;
7353         if (cnt > 1)
7354                 return -EINVAL;
7355
7356         if (mddev->layout != info->layout) {
7357                 /* Change layout
7358                  * we don't need to do anything at the md level, the
7359                  * personality will take care of it all.
7360                  */
7361                 if (mddev->pers->check_reshape == NULL)
7362                         return -EINVAL;
7363                 else {
7364                         mddev->new_layout = info->layout;
7365                         rv = mddev->pers->check_reshape(mddev);
7366                         if (rv)
7367                                 mddev->new_layout = mddev->layout;
7368                         return rv;
7369                 }
7370         }
7371         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7372                 rv = update_size(mddev, (sector_t)info->size * 2);
7373
7374         if (mddev->raid_disks    != info->raid_disks)
7375                 rv = update_raid_disks(mddev, info->raid_disks);
7376
7377         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7378                 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7379                         rv = -EINVAL;
7380                         goto err;
7381                 }
7382                 if (mddev->recovery || mddev->sync_thread) {
7383                         rv = -EBUSY;
7384                         goto err;
7385                 }
7386                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7387                         struct bitmap *bitmap;
7388                         /* add the bitmap */
7389                         if (mddev->bitmap) {
7390                                 rv = -EEXIST;
7391                                 goto err;
7392                         }
7393                         if (mddev->bitmap_info.default_offset == 0) {
7394                                 rv = -EINVAL;
7395                                 goto err;
7396                         }
7397                         mddev->bitmap_info.offset =
7398                                 mddev->bitmap_info.default_offset;
7399                         mddev->bitmap_info.space =
7400                                 mddev->bitmap_info.default_space;
7401                         bitmap = md_bitmap_create(mddev, -1);
7402                         mddev_suspend(mddev);
7403                         if (!IS_ERR(bitmap)) {
7404                                 mddev->bitmap = bitmap;
7405                                 rv = md_bitmap_load(mddev);
7406                         } else
7407                                 rv = PTR_ERR(bitmap);
7408                         if (rv)
7409                                 md_bitmap_destroy(mddev);
7410                         mddev_resume(mddev);
7411                 } else {
7412                         /* remove the bitmap */
7413                         if (!mddev->bitmap) {
7414                                 rv = -ENOENT;
7415                                 goto err;
7416                         }
7417                         if (mddev->bitmap->storage.file) {
7418                                 rv = -EINVAL;
7419                                 goto err;
7420                         }
7421                         if (mddev->bitmap_info.nodes) {
7422                                 /* hold PW on all the bitmap lock */
7423                                 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7424                                         pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7425                                         rv = -EPERM;
7426                                         md_cluster_ops->unlock_all_bitmaps(mddev);
7427                                         goto err;
7428                                 }
7429
7430                                 mddev->bitmap_info.nodes = 0;
7431                                 md_cluster_ops->leave(mddev);
7432                                 module_put(md_cluster_mod);
7433                                 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7434                         }
7435                         mddev_suspend(mddev);
7436                         md_bitmap_destroy(mddev);
7437                         mddev_resume(mddev);
7438                         mddev->bitmap_info.offset = 0;
7439                 }
7440         }
7441         md_update_sb(mddev, 1);
7442         return rv;
7443 err:
7444         return rv;
7445 }
7446
7447 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7448 {
7449         struct md_rdev *rdev;
7450         int err = 0;
7451
7452         if (mddev->pers == NULL)
7453                 return -ENODEV;
7454
7455         rcu_read_lock();
7456         rdev = md_find_rdev_rcu(mddev, dev);
7457         if (!rdev)
7458                 err =  -ENODEV;
7459         else {
7460                 md_error(mddev, rdev);
7461                 if (!test_bit(Faulty, &rdev->flags))
7462                         err = -EBUSY;
7463         }
7464         rcu_read_unlock();
7465         return err;
7466 }
7467
7468 /*
7469  * We have a problem here : there is no easy way to give a CHS
7470  * virtual geometry. We currently pretend that we have a 2 heads
7471  * 4 sectors (with a BIG number of cylinders...). This drives
7472  * dosfs just mad... ;-)
7473  */
7474 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7475 {
7476         struct mddev *mddev = bdev->bd_disk->private_data;
7477
7478         geo->heads = 2;
7479         geo->sectors = 4;
7480         geo->cylinders = mddev->array_sectors / 8;
7481         return 0;
7482 }
7483
7484 static inline bool md_ioctl_valid(unsigned int cmd)
7485 {
7486         switch (cmd) {
7487         case ADD_NEW_DISK:
7488         case GET_ARRAY_INFO:
7489         case GET_BITMAP_FILE:
7490         case GET_DISK_INFO:
7491         case HOT_ADD_DISK:
7492         case HOT_REMOVE_DISK:
7493         case RAID_VERSION:
7494         case RESTART_ARRAY_RW:
7495         case RUN_ARRAY:
7496         case SET_ARRAY_INFO:
7497         case SET_BITMAP_FILE:
7498         case SET_DISK_FAULTY:
7499         case STOP_ARRAY:
7500         case STOP_ARRAY_RO:
7501         case CLUSTERED_DISK_NACK:
7502                 return true;
7503         default:
7504                 return false;
7505         }
7506 }
7507
7508 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7509                         unsigned int cmd, unsigned long arg)
7510 {
7511         int err = 0;
7512         void __user *argp = (void __user *)arg;
7513         struct mddev *mddev = NULL;
7514         bool did_set_md_closing = false;
7515
7516         if (!md_ioctl_valid(cmd))
7517                 return -ENOTTY;
7518
7519         switch (cmd) {
7520         case RAID_VERSION:
7521         case GET_ARRAY_INFO:
7522         case GET_DISK_INFO:
7523                 break;
7524         default:
7525                 if (!capable(CAP_SYS_ADMIN))
7526                         return -EACCES;
7527         }
7528
7529         /*
7530          * Commands dealing with the RAID driver but not any
7531          * particular array:
7532          */
7533         switch (cmd) {
7534         case RAID_VERSION:
7535                 err = get_version(argp);
7536                 goto out;
7537         default:;
7538         }
7539
7540         /*
7541          * Commands creating/starting a new array:
7542          */
7543
7544         mddev = bdev->bd_disk->private_data;
7545
7546         if (!mddev) {
7547                 BUG();
7548                 goto out;
7549         }
7550
7551         /* Some actions do not requires the mutex */
7552         switch (cmd) {
7553         case GET_ARRAY_INFO:
7554                 if (!mddev->raid_disks && !mddev->external)
7555                         err = -ENODEV;
7556                 else
7557                         err = get_array_info(mddev, argp);
7558                 goto out;
7559
7560         case GET_DISK_INFO:
7561                 if (!mddev->raid_disks && !mddev->external)
7562                         err = -ENODEV;
7563                 else
7564                         err = get_disk_info(mddev, argp);
7565                 goto out;
7566
7567         case SET_DISK_FAULTY:
7568                 err = set_disk_faulty(mddev, new_decode_dev(arg));
7569                 goto out;
7570
7571         case GET_BITMAP_FILE:
7572                 err = get_bitmap_file(mddev, argp);
7573                 goto out;
7574
7575         }
7576
7577         if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7578                 flush_rdev_wq(mddev);
7579
7580         if (cmd == HOT_REMOVE_DISK)
7581                 /* need to ensure recovery thread has run */
7582                 wait_event_interruptible_timeout(mddev->sb_wait,
7583                                                  !test_bit(MD_RECOVERY_NEEDED,
7584                                                            &mddev->recovery),
7585                                                  msecs_to_jiffies(5000));
7586         if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7587                 /* Need to flush page cache, and ensure no-one else opens
7588                  * and writes
7589                  */
7590                 mutex_lock(&mddev->open_mutex);
7591                 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7592                         mutex_unlock(&mddev->open_mutex);
7593                         err = -EBUSY;
7594                         goto out;
7595                 }
7596                 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7597                         mutex_unlock(&mddev->open_mutex);
7598                         err = -EBUSY;
7599                         goto out;
7600                 }
7601                 did_set_md_closing = true;
7602                 mutex_unlock(&mddev->open_mutex);
7603                 sync_blockdev(bdev);
7604         }
7605         err = mddev_lock(mddev);
7606         if (err) {
7607                 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7608                          err, cmd);
7609                 goto out;
7610         }
7611
7612         if (cmd == SET_ARRAY_INFO) {
7613                 mdu_array_info_t info;
7614                 if (!arg)
7615                         memset(&info, 0, sizeof(info));
7616                 else if (copy_from_user(&info, argp, sizeof(info))) {
7617                         err = -EFAULT;
7618                         goto unlock;
7619                 }
7620                 if (mddev->pers) {
7621                         err = update_array_info(mddev, &info);
7622                         if (err) {
7623                                 pr_warn("md: couldn't update array info. %d\n", err);
7624                                 goto unlock;
7625                         }
7626                         goto unlock;
7627                 }
7628                 if (!list_empty(&mddev->disks)) {
7629                         pr_warn("md: array %s already has disks!\n", mdname(mddev));
7630                         err = -EBUSY;
7631                         goto unlock;
7632                 }
7633                 if (mddev->raid_disks) {
7634                         pr_warn("md: array %s already initialised!\n", mdname(mddev));
7635                         err = -EBUSY;
7636                         goto unlock;
7637                 }
7638                 err = md_set_array_info(mddev, &info);
7639                 if (err) {
7640                         pr_warn("md: couldn't set array info. %d\n", err);
7641                         goto unlock;
7642                 }
7643                 goto unlock;
7644         }
7645
7646         /*
7647          * Commands querying/configuring an existing array:
7648          */
7649         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7650          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7651         if ((!mddev->raid_disks && !mddev->external)
7652             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7653             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7654             && cmd != GET_BITMAP_FILE) {
7655                 err = -ENODEV;
7656                 goto unlock;
7657         }
7658
7659         /*
7660          * Commands even a read-only array can execute:
7661          */
7662         switch (cmd) {
7663         case RESTART_ARRAY_RW:
7664                 err = restart_array(mddev);
7665                 goto unlock;
7666
7667         case STOP_ARRAY:
7668                 err = do_md_stop(mddev, 0, bdev);
7669                 goto unlock;
7670
7671         case STOP_ARRAY_RO:
7672                 err = md_set_readonly(mddev, bdev);
7673                 goto unlock;
7674
7675         case HOT_REMOVE_DISK:
7676                 err = hot_remove_disk(mddev, new_decode_dev(arg));
7677                 goto unlock;
7678
7679         case ADD_NEW_DISK:
7680                 /* We can support ADD_NEW_DISK on read-only arrays
7681                  * only if we are re-adding a preexisting device.
7682                  * So require mddev->pers and MD_DISK_SYNC.
7683                  */
7684                 if (mddev->pers) {
7685                         mdu_disk_info_t info;
7686                         if (copy_from_user(&info, argp, sizeof(info)))
7687                                 err = -EFAULT;
7688                         else if (!(info.state & (1<<MD_DISK_SYNC)))
7689                                 /* Need to clear read-only for this */
7690                                 break;
7691                         else
7692                                 err = md_add_new_disk(mddev, &info);
7693                         goto unlock;
7694                 }
7695                 break;
7696         }
7697
7698         /*
7699          * The remaining ioctls are changing the state of the
7700          * superblock, so we do not allow them on read-only arrays.
7701          */
7702         if (mddev->ro && mddev->pers) {
7703                 if (mddev->ro == 2) {
7704                         mddev->ro = 0;
7705                         sysfs_notify_dirent_safe(mddev->sysfs_state);
7706                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7707                         /* mddev_unlock will wake thread */
7708                         /* If a device failed while we were read-only, we
7709                          * need to make sure the metadata is updated now.
7710                          */
7711                         if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7712                                 mddev_unlock(mddev);
7713                                 wait_event(mddev->sb_wait,
7714                                            !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7715                                            !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7716                                 mddev_lock_nointr(mddev);
7717                         }
7718                 } else {
7719                         err = -EROFS;
7720                         goto unlock;
7721                 }
7722         }
7723
7724         switch (cmd) {
7725         case ADD_NEW_DISK:
7726         {
7727                 mdu_disk_info_t info;
7728                 if (copy_from_user(&info, argp, sizeof(info)))
7729                         err = -EFAULT;
7730                 else
7731                         err = md_add_new_disk(mddev, &info);
7732                 goto unlock;
7733         }
7734
7735         case CLUSTERED_DISK_NACK:
7736                 if (mddev_is_clustered(mddev))
7737                         md_cluster_ops->new_disk_ack(mddev, false);
7738                 else
7739                         err = -EINVAL;
7740                 goto unlock;
7741
7742         case HOT_ADD_DISK:
7743                 err = hot_add_disk(mddev, new_decode_dev(arg));
7744                 goto unlock;
7745
7746         case RUN_ARRAY:
7747                 err = do_md_run(mddev);
7748                 goto unlock;
7749
7750         case SET_BITMAP_FILE:
7751                 err = set_bitmap_file(mddev, (int)arg);
7752                 goto unlock;
7753
7754         default:
7755                 err = -EINVAL;
7756                 goto unlock;
7757         }
7758
7759 unlock:
7760         if (mddev->hold_active == UNTIL_IOCTL &&
7761             err != -EINVAL)
7762                 mddev->hold_active = 0;
7763         mddev_unlock(mddev);
7764 out:
7765         if(did_set_md_closing)
7766                 clear_bit(MD_CLOSING, &mddev->flags);
7767         return err;
7768 }
7769 #ifdef CONFIG_COMPAT
7770 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7771                     unsigned int cmd, unsigned long arg)
7772 {
7773         switch (cmd) {
7774         case HOT_REMOVE_DISK:
7775         case HOT_ADD_DISK:
7776         case SET_DISK_FAULTY:
7777         case SET_BITMAP_FILE:
7778                 /* These take in integer arg, do not convert */
7779                 break;
7780         default:
7781                 arg = (unsigned long)compat_ptr(arg);
7782                 break;
7783         }
7784
7785         return md_ioctl(bdev, mode, cmd, arg);
7786 }
7787 #endif /* CONFIG_COMPAT */
7788
7789 static int md_set_read_only(struct block_device *bdev, bool ro)
7790 {
7791         struct mddev *mddev = bdev->bd_disk->private_data;
7792         int err;
7793
7794         err = mddev_lock(mddev);
7795         if (err)
7796                 return err;
7797
7798         if (!mddev->raid_disks && !mddev->external) {
7799                 err = -ENODEV;
7800                 goto out_unlock;
7801         }
7802
7803         /*
7804          * Transitioning to read-auto need only happen for arrays that call
7805          * md_write_start and which are not ready for writes yet.
7806          */
7807         if (!ro && mddev->ro == 1 && mddev->pers) {
7808                 err = restart_array(mddev);
7809                 if (err)
7810                         goto out_unlock;
7811                 mddev->ro = 2;
7812         }
7813
7814 out_unlock:
7815         mddev_unlock(mddev);
7816         return err;
7817 }
7818
7819 static int md_open(struct block_device *bdev, fmode_t mode)
7820 {
7821         /*
7822          * Succeed if we can lock the mddev, which confirms that
7823          * it isn't being stopped right now.
7824          */
7825         struct mddev *mddev = mddev_find(bdev->bd_dev);
7826         int err;
7827
7828         if (!mddev)
7829                 return -ENODEV;
7830
7831         if (mddev->gendisk != bdev->bd_disk) {
7832                 /* we are racing with mddev_put which is discarding this
7833                  * bd_disk.
7834                  */
7835                 mddev_put(mddev);
7836                 /* Wait until bdev->bd_disk is definitely gone */
7837                 if (work_pending(&mddev->del_work))
7838                         flush_workqueue(md_misc_wq);
7839                 return -EBUSY;
7840         }
7841         BUG_ON(mddev != bdev->bd_disk->private_data);
7842
7843         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7844                 goto out;
7845
7846         if (test_bit(MD_CLOSING, &mddev->flags)) {
7847                 mutex_unlock(&mddev->open_mutex);
7848                 err = -ENODEV;
7849                 goto out;
7850         }
7851
7852         err = 0;
7853         atomic_inc(&mddev->openers);
7854         mutex_unlock(&mddev->open_mutex);
7855
7856         bdev_check_media_change(bdev);
7857  out:
7858         if (err)
7859                 mddev_put(mddev);
7860         return err;
7861 }
7862
7863 static void md_release(struct gendisk *disk, fmode_t mode)
7864 {
7865         struct mddev *mddev = disk->private_data;
7866
7867         BUG_ON(!mddev);
7868         atomic_dec(&mddev->openers);
7869         mddev_put(mddev);
7870 }
7871
7872 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7873 {
7874         struct mddev *mddev = disk->private_data;
7875         unsigned int ret = 0;
7876
7877         if (mddev->changed)
7878                 ret = DISK_EVENT_MEDIA_CHANGE;
7879         mddev->changed = 0;
7880         return ret;
7881 }
7882
7883 const struct block_device_operations md_fops =
7884 {
7885         .owner          = THIS_MODULE,
7886         .submit_bio     = md_submit_bio,
7887         .open           = md_open,
7888         .release        = md_release,
7889         .ioctl          = md_ioctl,
7890 #ifdef CONFIG_COMPAT
7891         .compat_ioctl   = md_compat_ioctl,
7892 #endif
7893         .getgeo         = md_getgeo,
7894         .check_events   = md_check_events,
7895         .set_read_only  = md_set_read_only,
7896 };
7897
7898 static int md_thread(void *arg)
7899 {
7900         struct md_thread *thread = arg;
7901
7902         /*
7903          * md_thread is a 'system-thread', it's priority should be very
7904          * high. We avoid resource deadlocks individually in each
7905          * raid personality. (RAID5 does preallocation) We also use RR and
7906          * the very same RT priority as kswapd, thus we will never get
7907          * into a priority inversion deadlock.
7908          *
7909          * we definitely have to have equal or higher priority than
7910          * bdflush, otherwise bdflush will deadlock if there are too
7911          * many dirty RAID5 blocks.
7912          */
7913
7914         allow_signal(SIGKILL);
7915         while (!kthread_should_stop()) {
7916
7917                 /* We need to wait INTERRUPTIBLE so that
7918                  * we don't add to the load-average.
7919                  * That means we need to be sure no signals are
7920                  * pending
7921                  */
7922                 if (signal_pending(current))
7923                         flush_signals(current);
7924
7925                 wait_event_interruptible_timeout
7926                         (thread->wqueue,
7927                          test_bit(THREAD_WAKEUP, &thread->flags)
7928                          || kthread_should_stop() || kthread_should_park(),
7929                          thread->timeout);
7930
7931                 clear_bit(THREAD_WAKEUP, &thread->flags);
7932                 if (kthread_should_park())
7933                         kthread_parkme();
7934                 if (!kthread_should_stop())
7935                         thread->run(thread);
7936         }
7937
7938         return 0;
7939 }
7940
7941 void md_wakeup_thread(struct md_thread *thread)
7942 {
7943         if (thread) {
7944                 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7945                 set_bit(THREAD_WAKEUP, &thread->flags);
7946                 wake_up(&thread->wqueue);
7947         }
7948 }
7949 EXPORT_SYMBOL(md_wakeup_thread);
7950
7951 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7952                 struct mddev *mddev, const char *name)
7953 {
7954         struct md_thread *thread;
7955
7956         thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7957         if (!thread)
7958                 return NULL;
7959
7960         init_waitqueue_head(&thread->wqueue);
7961
7962         thread->run = run;
7963         thread->mddev = mddev;
7964         thread->timeout = MAX_SCHEDULE_TIMEOUT;
7965         thread->tsk = kthread_run(md_thread, thread,
7966                                   "%s_%s",
7967                                   mdname(thread->mddev),
7968                                   name);
7969         if (IS_ERR(thread->tsk)) {
7970                 kfree(thread);
7971                 return NULL;
7972         }
7973         return thread;
7974 }
7975 EXPORT_SYMBOL(md_register_thread);
7976
7977 void md_unregister_thread(struct md_thread **threadp)
7978 {
7979         struct md_thread *thread = *threadp;
7980         if (!thread)
7981                 return;
7982         pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7983         /* Locking ensures that mddev_unlock does not wake_up a
7984          * non-existent thread
7985          */
7986         spin_lock(&pers_lock);
7987         *threadp = NULL;
7988         spin_unlock(&pers_lock);
7989
7990         kthread_stop(thread->tsk);
7991         kfree(thread);
7992 }
7993 EXPORT_SYMBOL(md_unregister_thread);
7994
7995 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7996 {
7997         if (!rdev || test_bit(Faulty, &rdev->flags))
7998                 return;
7999
8000         if (!mddev->pers || !mddev->pers->error_handler)
8001                 return;
8002         mddev->pers->error_handler(mddev,rdev);
8003         if (mddev->degraded)
8004                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8005         sysfs_notify_dirent_safe(rdev->sysfs_state);
8006         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8007         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8008         md_wakeup_thread(mddev->thread);
8009         if (mddev->event_work.func)
8010                 queue_work(md_misc_wq, &mddev->event_work);
8011         md_new_event(mddev);
8012 }
8013 EXPORT_SYMBOL(md_error);
8014
8015 /* seq_file implementation /proc/mdstat */
8016
8017 static void status_unused(struct seq_file *seq)
8018 {
8019         int i = 0;
8020         struct md_rdev *rdev;
8021
8022         seq_printf(seq, "unused devices: ");
8023
8024         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8025                 char b[BDEVNAME_SIZE];
8026                 i++;
8027                 seq_printf(seq, "%s ",
8028                               bdevname(rdev->bdev,b));
8029         }
8030         if (!i)
8031                 seq_printf(seq, "<none>");
8032
8033         seq_printf(seq, "\n");
8034 }
8035
8036 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8037 {
8038         sector_t max_sectors, resync, res;
8039         unsigned long dt, db = 0;
8040         sector_t rt, curr_mark_cnt, resync_mark_cnt;
8041         int scale, recovery_active;
8042         unsigned int per_milli;
8043
8044         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8045             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8046                 max_sectors = mddev->resync_max_sectors;
8047         else
8048                 max_sectors = mddev->dev_sectors;
8049
8050         resync = mddev->curr_resync;
8051         if (resync <= 3) {
8052                 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8053                         /* Still cleaning up */
8054                         resync = max_sectors;
8055         } else if (resync > max_sectors)
8056                 resync = max_sectors;
8057         else
8058                 resync -= atomic_read(&mddev->recovery_active);
8059
8060         if (resync == 0) {
8061                 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8062                         struct md_rdev *rdev;
8063
8064                         rdev_for_each(rdev, mddev)
8065                                 if (rdev->raid_disk >= 0 &&
8066                                     !test_bit(Faulty, &rdev->flags) &&
8067                                     rdev->recovery_offset != MaxSector &&
8068                                     rdev->recovery_offset) {
8069                                         seq_printf(seq, "\trecover=REMOTE");
8070                                         return 1;
8071                                 }
8072                         if (mddev->reshape_position != MaxSector)
8073                                 seq_printf(seq, "\treshape=REMOTE");
8074                         else
8075                                 seq_printf(seq, "\tresync=REMOTE");
8076                         return 1;
8077                 }
8078                 if (mddev->recovery_cp < MaxSector) {
8079                         seq_printf(seq, "\tresync=PENDING");
8080                         return 1;
8081                 }
8082                 return 0;
8083         }
8084         if (resync < 3) {
8085                 seq_printf(seq, "\tresync=DELAYED");
8086                 return 1;
8087         }
8088
8089         WARN_ON(max_sectors == 0);
8090         /* Pick 'scale' such that (resync>>scale)*1000 will fit
8091          * in a sector_t, and (max_sectors>>scale) will fit in a
8092          * u32, as those are the requirements for sector_div.
8093          * Thus 'scale' must be at least 10
8094          */
8095         scale = 10;
8096         if (sizeof(sector_t) > sizeof(unsigned long)) {
8097                 while ( max_sectors/2 > (1ULL<<(scale+32)))
8098                         scale++;
8099         }
8100         res = (resync>>scale)*1000;
8101         sector_div(res, (u32)((max_sectors>>scale)+1));
8102
8103         per_milli = res;
8104         {
8105                 int i, x = per_milli/50, y = 20-x;
8106                 seq_printf(seq, "[");
8107                 for (i = 0; i < x; i++)
8108                         seq_printf(seq, "=");
8109                 seq_printf(seq, ">");
8110                 for (i = 0; i < y; i++)
8111                         seq_printf(seq, ".");
8112                 seq_printf(seq, "] ");
8113         }
8114         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8115                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8116                     "reshape" :
8117                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8118                      "check" :
8119                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8120                       "resync" : "recovery"))),
8121                    per_milli/10, per_milli % 10,
8122                    (unsigned long long) resync/2,
8123                    (unsigned long long) max_sectors/2);
8124
8125         /*
8126          * dt: time from mark until now
8127          * db: blocks written from mark until now
8128          * rt: remaining time
8129          *
8130          * rt is a sector_t, which is always 64bit now. We are keeping
8131          * the original algorithm, but it is not really necessary.
8132          *
8133          * Original algorithm:
8134          *   So we divide before multiply in case it is 32bit and close
8135          *   to the limit.
8136          *   We scale the divisor (db) by 32 to avoid losing precision
8137          *   near the end of resync when the number of remaining sectors
8138          *   is close to 'db'.
8139          *   We then divide rt by 32 after multiplying by db to compensate.
8140          *   The '+1' avoids division by zero if db is very small.
8141          */
8142         dt = ((jiffies - mddev->resync_mark) / HZ);
8143         if (!dt) dt++;
8144
8145         curr_mark_cnt = mddev->curr_mark_cnt;
8146         recovery_active = atomic_read(&mddev->recovery_active);
8147         resync_mark_cnt = mddev->resync_mark_cnt;
8148
8149         if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8150                 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8151
8152         rt = max_sectors - resync;    /* number of remaining sectors */
8153         rt = div64_u64(rt, db/32+1);
8154         rt *= dt;
8155         rt >>= 5;
8156
8157         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8158                    ((unsigned long)rt % 60)/6);
8159
8160         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8161         return 1;
8162 }
8163
8164 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8165 {
8166         struct list_head *tmp;
8167         loff_t l = *pos;
8168         struct mddev *mddev;
8169
8170         if (l == 0x10000) {
8171                 ++*pos;
8172                 return (void *)2;
8173         }
8174         if (l > 0x10000)
8175                 return NULL;
8176         if (!l--)
8177                 /* header */
8178                 return (void*)1;
8179
8180         spin_lock(&all_mddevs_lock);
8181         list_for_each(tmp,&all_mddevs)
8182                 if (!l--) {
8183                         mddev = list_entry(tmp, struct mddev, all_mddevs);
8184                         mddev_get(mddev);
8185                         spin_unlock(&all_mddevs_lock);
8186                         return mddev;
8187                 }
8188         spin_unlock(&all_mddevs_lock);
8189         if (!l--)
8190                 return (void*)2;/* tail */
8191         return NULL;
8192 }
8193
8194 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8195 {
8196         struct list_head *tmp;
8197         struct mddev *next_mddev, *mddev = v;
8198
8199         ++*pos;
8200         if (v == (void*)2)
8201                 return NULL;
8202
8203         spin_lock(&all_mddevs_lock);
8204         if (v == (void*)1)
8205                 tmp = all_mddevs.next;
8206         else
8207                 tmp = mddev->all_mddevs.next;
8208         if (tmp != &all_mddevs)
8209                 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8210         else {
8211                 next_mddev = (void*)2;
8212                 *pos = 0x10000;
8213         }
8214         spin_unlock(&all_mddevs_lock);
8215
8216         if (v != (void*)1)
8217                 mddev_put(mddev);
8218         return next_mddev;
8219
8220 }
8221
8222 static void md_seq_stop(struct seq_file *seq, void *v)
8223 {
8224         struct mddev *mddev = v;
8225
8226         if (mddev && v != (void*)1 && v != (void*)2)
8227                 mddev_put(mddev);
8228 }
8229
8230 static int md_seq_show(struct seq_file *seq, void *v)
8231 {
8232         struct mddev *mddev = v;
8233         sector_t sectors;
8234         struct md_rdev *rdev;
8235
8236         if (v == (void*)1) {
8237                 struct md_personality *pers;
8238                 seq_printf(seq, "Personalities : ");
8239                 spin_lock(&pers_lock);
8240                 list_for_each_entry(pers, &pers_list, list)
8241                         seq_printf(seq, "[%s] ", pers->name);
8242
8243                 spin_unlock(&pers_lock);
8244                 seq_printf(seq, "\n");
8245                 seq->poll_event = atomic_read(&md_event_count);
8246                 return 0;
8247         }
8248         if (v == (void*)2) {
8249                 status_unused(seq);
8250                 return 0;
8251         }
8252
8253         spin_lock(&mddev->lock);
8254         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8255                 seq_printf(seq, "%s : %sactive", mdname(mddev),
8256                                                 mddev->pers ? "" : "in");
8257                 if (mddev->pers) {
8258                         if (mddev->ro==1)
8259                                 seq_printf(seq, " (read-only)");
8260                         if (mddev->ro==2)
8261                                 seq_printf(seq, " (auto-read-only)");
8262                         seq_printf(seq, " %s", mddev->pers->name);
8263                 }
8264
8265                 sectors = 0;
8266                 rcu_read_lock();
8267                 rdev_for_each_rcu(rdev, mddev) {
8268                         char b[BDEVNAME_SIZE];
8269                         seq_printf(seq, " %s[%d]",
8270                                 bdevname(rdev->bdev,b), rdev->desc_nr);
8271                         if (test_bit(WriteMostly, &rdev->flags))
8272                                 seq_printf(seq, "(W)");
8273                         if (test_bit(Journal, &rdev->flags))
8274                                 seq_printf(seq, "(J)");
8275                         if (test_bit(Faulty, &rdev->flags)) {
8276                                 seq_printf(seq, "(F)");
8277                                 continue;
8278                         }
8279                         if (rdev->raid_disk < 0)
8280                                 seq_printf(seq, "(S)"); /* spare */
8281                         if (test_bit(Replacement, &rdev->flags))
8282                                 seq_printf(seq, "(R)");
8283                         sectors += rdev->sectors;
8284                 }
8285                 rcu_read_unlock();
8286
8287                 if (!list_empty(&mddev->disks)) {
8288                         if (mddev->pers)
8289                                 seq_printf(seq, "\n      %llu blocks",
8290                                            (unsigned long long)
8291                                            mddev->array_sectors / 2);
8292                         else
8293                                 seq_printf(seq, "\n      %llu blocks",
8294                                            (unsigned long long)sectors / 2);
8295                 }
8296                 if (mddev->persistent) {
8297                         if (mddev->major_version != 0 ||
8298                             mddev->minor_version != 90) {
8299                                 seq_printf(seq," super %d.%d",
8300                                            mddev->major_version,
8301                                            mddev->minor_version);
8302                         }
8303                 } else if (mddev->external)
8304                         seq_printf(seq, " super external:%s",
8305                                    mddev->metadata_type);
8306                 else
8307                         seq_printf(seq, " super non-persistent");
8308
8309                 if (mddev->pers) {
8310                         mddev->pers->status(seq, mddev);
8311                         seq_printf(seq, "\n      ");
8312                         if (mddev->pers->sync_request) {
8313                                 if (status_resync(seq, mddev))
8314                                         seq_printf(seq, "\n      ");
8315                         }
8316                 } else
8317                         seq_printf(seq, "\n       ");
8318
8319                 md_bitmap_status(seq, mddev->bitmap);
8320
8321                 seq_printf(seq, "\n");
8322         }
8323         spin_unlock(&mddev->lock);
8324
8325         return 0;
8326 }
8327
8328 static const struct seq_operations md_seq_ops = {
8329         .start  = md_seq_start,
8330         .next   = md_seq_next,
8331         .stop   = md_seq_stop,
8332         .show   = md_seq_show,
8333 };
8334
8335 static int md_seq_open(struct inode *inode, struct file *file)
8336 {
8337         struct seq_file *seq;
8338         int error;
8339
8340         error = seq_open(file, &md_seq_ops);
8341         if (error)
8342                 return error;
8343
8344         seq = file->private_data;
8345         seq->poll_event = atomic_read(&md_event_count);
8346         return error;
8347 }
8348
8349 static int md_unloading;
8350 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8351 {
8352         struct seq_file *seq = filp->private_data;
8353         __poll_t mask;
8354
8355         if (md_unloading)
8356                 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8357         poll_wait(filp, &md_event_waiters, wait);
8358
8359         /* always allow read */
8360         mask = EPOLLIN | EPOLLRDNORM;
8361
8362         if (seq->poll_event != atomic_read(&md_event_count))
8363                 mask |= EPOLLERR | EPOLLPRI;
8364         return mask;
8365 }
8366
8367 static const struct proc_ops mdstat_proc_ops = {
8368         .proc_open      = md_seq_open,
8369         .proc_read      = seq_read,
8370         .proc_lseek     = seq_lseek,
8371         .proc_release   = seq_release,
8372         .proc_poll      = mdstat_poll,
8373 };
8374
8375 int register_md_personality(struct md_personality *p)
8376 {
8377         pr_debug("md: %s personality registered for level %d\n",
8378                  p->name, p->level);
8379         spin_lock(&pers_lock);
8380         list_add_tail(&p->list, &pers_list);
8381         spin_unlock(&pers_lock);
8382         return 0;
8383 }
8384 EXPORT_SYMBOL(register_md_personality);
8385
8386 int unregister_md_personality(struct md_personality *p)
8387 {
8388         pr_debug("md: %s personality unregistered\n", p->name);
8389         spin_lock(&pers_lock);
8390         list_del_init(&p->list);
8391         spin_unlock(&pers_lock);
8392         return 0;
8393 }
8394 EXPORT_SYMBOL(unregister_md_personality);
8395
8396 int register_md_cluster_operations(struct md_cluster_operations *ops,
8397                                    struct module *module)
8398 {
8399         int ret = 0;
8400         spin_lock(&pers_lock);
8401         if (md_cluster_ops != NULL)
8402                 ret = -EALREADY;
8403         else {
8404                 md_cluster_ops = ops;
8405                 md_cluster_mod = module;
8406         }
8407         spin_unlock(&pers_lock);
8408         return ret;
8409 }
8410 EXPORT_SYMBOL(register_md_cluster_operations);
8411
8412 int unregister_md_cluster_operations(void)
8413 {
8414         spin_lock(&pers_lock);
8415         md_cluster_ops = NULL;
8416         spin_unlock(&pers_lock);
8417         return 0;
8418 }
8419 EXPORT_SYMBOL(unregister_md_cluster_operations);
8420
8421 int md_setup_cluster(struct mddev *mddev, int nodes)
8422 {
8423         int ret;
8424         if (!md_cluster_ops)
8425                 request_module("md-cluster");
8426         spin_lock(&pers_lock);
8427         /* ensure module won't be unloaded */
8428         if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8429                 pr_warn("can't find md-cluster module or get it's reference.\n");
8430                 spin_unlock(&pers_lock);
8431                 return -ENOENT;
8432         }
8433         spin_unlock(&pers_lock);
8434
8435         ret = md_cluster_ops->join(mddev, nodes);
8436         if (!ret)
8437                 mddev->safemode_delay = 0;
8438         return ret;
8439 }
8440
8441 void md_cluster_stop(struct mddev *mddev)
8442 {
8443         if (!md_cluster_ops)
8444                 return;
8445         md_cluster_ops->leave(mddev);
8446         module_put(md_cluster_mod);
8447 }
8448
8449 static int is_mddev_idle(struct mddev *mddev, int init)
8450 {
8451         struct md_rdev *rdev;
8452         int idle;
8453         int curr_events;
8454
8455         idle = 1;
8456         rcu_read_lock();
8457         rdev_for_each_rcu(rdev, mddev) {
8458                 struct gendisk *disk = rdev->bdev->bd_disk;
8459                 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8460                               atomic_read(&disk->sync_io);
8461                 /* sync IO will cause sync_io to increase before the disk_stats
8462                  * as sync_io is counted when a request starts, and
8463                  * disk_stats is counted when it completes.
8464                  * So resync activity will cause curr_events to be smaller than
8465                  * when there was no such activity.
8466                  * non-sync IO will cause disk_stat to increase without
8467                  * increasing sync_io so curr_events will (eventually)
8468                  * be larger than it was before.  Once it becomes
8469                  * substantially larger, the test below will cause
8470                  * the array to appear non-idle, and resync will slow
8471                  * down.
8472                  * If there is a lot of outstanding resync activity when
8473                  * we set last_event to curr_events, then all that activity
8474                  * completing might cause the array to appear non-idle
8475                  * and resync will be slowed down even though there might
8476                  * not have been non-resync activity.  This will only
8477                  * happen once though.  'last_events' will soon reflect
8478                  * the state where there is little or no outstanding
8479                  * resync requests, and further resync activity will
8480                  * always make curr_events less than last_events.
8481                  *
8482                  */
8483                 if (init || curr_events - rdev->last_events > 64) {
8484                         rdev->last_events = curr_events;
8485                         idle = 0;
8486                 }
8487         }
8488         rcu_read_unlock();
8489         return idle;
8490 }
8491
8492 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8493 {
8494         /* another "blocks" (512byte) blocks have been synced */
8495         atomic_sub(blocks, &mddev->recovery_active);
8496         wake_up(&mddev->recovery_wait);
8497         if (!ok) {
8498                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8499                 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8500                 md_wakeup_thread(mddev->thread);
8501                 // stop recovery, signal do_sync ....
8502         }
8503 }
8504 EXPORT_SYMBOL(md_done_sync);
8505
8506 /* md_write_start(mddev, bi)
8507  * If we need to update some array metadata (e.g. 'active' flag
8508  * in superblock) before writing, schedule a superblock update
8509  * and wait for it to complete.
8510  * A return value of 'false' means that the write wasn't recorded
8511  * and cannot proceed as the array is being suspend.
8512  */
8513 bool md_write_start(struct mddev *mddev, struct bio *bi)
8514 {
8515         int did_change = 0;
8516
8517         if (bio_data_dir(bi) != WRITE)
8518                 return true;
8519
8520         BUG_ON(mddev->ro == 1);
8521         if (mddev->ro == 2) {
8522                 /* need to switch to read/write */
8523                 mddev->ro = 0;
8524                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8525                 md_wakeup_thread(mddev->thread);
8526                 md_wakeup_thread(mddev->sync_thread);
8527                 did_change = 1;
8528         }
8529         rcu_read_lock();
8530         percpu_ref_get(&mddev->writes_pending);
8531         smp_mb(); /* Match smp_mb in set_in_sync() */
8532         if (mddev->safemode == 1)
8533                 mddev->safemode = 0;
8534         /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8535         if (mddev->in_sync || mddev->sync_checkers) {
8536                 spin_lock(&mddev->lock);
8537                 if (mddev->in_sync) {
8538                         mddev->in_sync = 0;
8539                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8540                         set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8541                         md_wakeup_thread(mddev->thread);
8542                         did_change = 1;
8543                 }
8544                 spin_unlock(&mddev->lock);
8545         }
8546         rcu_read_unlock();
8547         if (did_change)
8548                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8549         if (!mddev->has_superblocks)
8550                 return true;
8551         wait_event(mddev->sb_wait,
8552                    !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8553                    mddev->suspended);
8554         if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8555                 percpu_ref_put(&mddev->writes_pending);
8556                 return false;
8557         }
8558         return true;
8559 }
8560 EXPORT_SYMBOL(md_write_start);
8561
8562 /* md_write_inc can only be called when md_write_start() has
8563  * already been called at least once of the current request.
8564  * It increments the counter and is useful when a single request
8565  * is split into several parts.  Each part causes an increment and
8566  * so needs a matching md_write_end().
8567  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8568  * a spinlocked region.
8569  */
8570 void md_write_inc(struct mddev *mddev, struct bio *bi)
8571 {
8572         if (bio_data_dir(bi) != WRITE)
8573                 return;
8574         WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8575         percpu_ref_get(&mddev->writes_pending);
8576 }
8577 EXPORT_SYMBOL(md_write_inc);
8578
8579 void md_write_end(struct mddev *mddev)
8580 {
8581         percpu_ref_put(&mddev->writes_pending);
8582
8583         if (mddev->safemode == 2)
8584                 md_wakeup_thread(mddev->thread);
8585         else if (mddev->safemode_delay)
8586                 /* The roundup() ensures this only performs locking once
8587                  * every ->safemode_delay jiffies
8588                  */
8589                 mod_timer(&mddev->safemode_timer,
8590                           roundup(jiffies, mddev->safemode_delay) +
8591                           mddev->safemode_delay);
8592 }
8593
8594 EXPORT_SYMBOL(md_write_end);
8595
8596 /* This is used by raid0 and raid10 */
8597 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8598                         struct bio *bio, sector_t start, sector_t size)
8599 {
8600         struct bio *discard_bio = NULL;
8601
8602         if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO, 0,
8603                         &discard_bio) || !discard_bio)
8604                 return;
8605
8606         bio_chain(discard_bio, bio);
8607         bio_clone_blkg_association(discard_bio, bio);
8608         if (mddev->gendisk)
8609                 trace_block_bio_remap(discard_bio,
8610                                 disk_devt(mddev->gendisk),
8611                                 bio->bi_iter.bi_sector);
8612         submit_bio_noacct(discard_bio);
8613 }
8614 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8615
8616 /* md_allow_write(mddev)
8617  * Calling this ensures that the array is marked 'active' so that writes
8618  * may proceed without blocking.  It is important to call this before
8619  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8620  * Must be called with mddev_lock held.
8621  */
8622 void md_allow_write(struct mddev *mddev)
8623 {
8624         if (!mddev->pers)
8625                 return;
8626         if (mddev->ro)
8627                 return;
8628         if (!mddev->pers->sync_request)
8629                 return;
8630
8631         spin_lock(&mddev->lock);
8632         if (mddev->in_sync) {
8633                 mddev->in_sync = 0;
8634                 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8635                 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8636                 if (mddev->safemode_delay &&
8637                     mddev->safemode == 0)
8638                         mddev->safemode = 1;
8639                 spin_unlock(&mddev->lock);
8640                 md_update_sb(mddev, 0);
8641                 sysfs_notify_dirent_safe(mddev->sysfs_state);
8642                 /* wait for the dirty state to be recorded in the metadata */
8643                 wait_event(mddev->sb_wait,
8644                            !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8645         } else
8646                 spin_unlock(&mddev->lock);
8647 }
8648 EXPORT_SYMBOL_GPL(md_allow_write);
8649
8650 #define SYNC_MARKS      10
8651 #define SYNC_MARK_STEP  (3*HZ)
8652 #define UPDATE_FREQUENCY (5*60*HZ)
8653 void md_do_sync(struct md_thread *thread)
8654 {
8655         struct mddev *mddev = thread->mddev;
8656         struct mddev *mddev2;
8657         unsigned int currspeed = 0, window;
8658         sector_t max_sectors,j, io_sectors, recovery_done;
8659         unsigned long mark[SYNC_MARKS];
8660         unsigned long update_time;
8661         sector_t mark_cnt[SYNC_MARKS];
8662         int last_mark,m;
8663         struct list_head *tmp;
8664         sector_t last_check;
8665         int skipped = 0;
8666         struct md_rdev *rdev;
8667         char *desc, *action = NULL;
8668         struct blk_plug plug;
8669         int ret;
8670
8671         /* just incase thread restarts... */
8672         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8673             test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8674                 return;
8675         if (mddev->ro) {/* never try to sync a read-only array */
8676                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8677                 return;
8678         }
8679
8680         if (mddev_is_clustered(mddev)) {
8681                 ret = md_cluster_ops->resync_start(mddev);
8682                 if (ret)
8683                         goto skip;
8684
8685                 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8686                 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8687                         test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8688                         test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8689                      && ((unsigned long long)mddev->curr_resync_completed
8690                          < (unsigned long long)mddev->resync_max_sectors))
8691                         goto skip;
8692         }
8693
8694         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8695                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8696                         desc = "data-check";
8697                         action = "check";
8698                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8699                         desc = "requested-resync";
8700                         action = "repair";
8701                 } else
8702                         desc = "resync";
8703         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8704                 desc = "reshape";
8705         else
8706                 desc = "recovery";
8707
8708         mddev->last_sync_action = action ?: desc;
8709
8710         /* we overload curr_resync somewhat here.
8711          * 0 == not engaged in resync at all
8712          * 2 == checking that there is no conflict with another sync
8713          * 1 == like 2, but have yielded to allow conflicting resync to
8714          *              commence
8715          * other == active in resync - this many blocks
8716          *
8717          * Before starting a resync we must have set curr_resync to
8718          * 2, and then checked that every "conflicting" array has curr_resync
8719          * less than ours.  When we find one that is the same or higher
8720          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8721          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8722          * This will mean we have to start checking from the beginning again.
8723          *
8724          */
8725
8726         do {
8727                 int mddev2_minor = -1;
8728                 mddev->curr_resync = 2;
8729
8730         try_again:
8731                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8732                         goto skip;
8733                 for_each_mddev(mddev2, tmp) {
8734                         if (mddev2 == mddev)
8735                                 continue;
8736                         if (!mddev->parallel_resync
8737                         &&  mddev2->curr_resync
8738                         &&  match_mddev_units(mddev, mddev2)) {
8739                                 DEFINE_WAIT(wq);
8740                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
8741                                         /* arbitrarily yield */
8742                                         mddev->curr_resync = 1;
8743                                         wake_up(&resync_wait);
8744                                 }
8745                                 if (mddev > mddev2 && mddev->curr_resync == 1)
8746                                         /* no need to wait here, we can wait the next
8747                                          * time 'round when curr_resync == 2
8748                                          */
8749                                         continue;
8750                                 /* We need to wait 'interruptible' so as not to
8751                                  * contribute to the load average, and not to
8752                                  * be caught by 'softlockup'
8753                                  */
8754                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8755                                 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8756                                     mddev2->curr_resync >= mddev->curr_resync) {
8757                                         if (mddev2_minor != mddev2->md_minor) {
8758                                                 mddev2_minor = mddev2->md_minor;
8759                                                 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8760                                                         desc, mdname(mddev),
8761                                                         mdname(mddev2));
8762                                         }
8763                                         mddev_put(mddev2);
8764                                         if (signal_pending(current))
8765                                                 flush_signals(current);
8766                                         schedule();
8767                                         finish_wait(&resync_wait, &wq);
8768                                         goto try_again;
8769                                 }
8770                                 finish_wait(&resync_wait, &wq);
8771                         }
8772                 }
8773         } while (mddev->curr_resync < 2);
8774
8775         j = 0;
8776         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8777                 /* resync follows the size requested by the personality,
8778                  * which defaults to physical size, but can be virtual size
8779                  */
8780                 max_sectors = mddev->resync_max_sectors;
8781                 atomic64_set(&mddev->resync_mismatches, 0);
8782                 /* we don't use the checkpoint if there's a bitmap */
8783                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8784                         j = mddev->resync_min;
8785                 else if (!mddev->bitmap)
8786                         j = mddev->recovery_cp;
8787
8788         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8789                 max_sectors = mddev->resync_max_sectors;
8790                 /*
8791                  * If the original node aborts reshaping then we continue the
8792                  * reshaping, so set j again to avoid restart reshape from the
8793                  * first beginning
8794                  */
8795                 if (mddev_is_clustered(mddev) &&
8796                     mddev->reshape_position != MaxSector)
8797                         j = mddev->reshape_position;
8798         } else {
8799                 /* recovery follows the physical size of devices */
8800                 max_sectors = mddev->dev_sectors;
8801                 j = MaxSector;
8802                 rcu_read_lock();
8803                 rdev_for_each_rcu(rdev, mddev)
8804                         if (rdev->raid_disk >= 0 &&
8805                             !test_bit(Journal, &rdev->flags) &&
8806                             !test_bit(Faulty, &rdev->flags) &&
8807                             !test_bit(In_sync, &rdev->flags) &&
8808                             rdev->recovery_offset < j)
8809                                 j = rdev->recovery_offset;
8810                 rcu_read_unlock();
8811
8812                 /* If there is a bitmap, we need to make sure all
8813                  * writes that started before we added a spare
8814                  * complete before we start doing a recovery.
8815                  * Otherwise the write might complete and (via
8816                  * bitmap_endwrite) set a bit in the bitmap after the
8817                  * recovery has checked that bit and skipped that
8818                  * region.
8819                  */
8820                 if (mddev->bitmap) {
8821                         mddev->pers->quiesce(mddev, 1);
8822                         mddev->pers->quiesce(mddev, 0);
8823                 }
8824         }
8825
8826         pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8827         pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8828         pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8829                  speed_max(mddev), desc);
8830
8831         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8832
8833         io_sectors = 0;
8834         for (m = 0; m < SYNC_MARKS; m++) {
8835                 mark[m] = jiffies;
8836                 mark_cnt[m] = io_sectors;
8837         }
8838         last_mark = 0;
8839         mddev->resync_mark = mark[last_mark];
8840         mddev->resync_mark_cnt = mark_cnt[last_mark];
8841
8842         /*
8843          * Tune reconstruction:
8844          */
8845         window = 32 * (PAGE_SIZE / 512);
8846         pr_debug("md: using %dk window, over a total of %lluk.\n",
8847                  window/2, (unsigned long long)max_sectors/2);
8848
8849         atomic_set(&mddev->recovery_active, 0);
8850         last_check = 0;
8851
8852         if (j>2) {
8853                 pr_debug("md: resuming %s of %s from checkpoint.\n",
8854                          desc, mdname(mddev));
8855                 mddev->curr_resync = j;
8856         } else
8857                 mddev->curr_resync = 3; /* no longer delayed */
8858         mddev->curr_resync_completed = j;
8859         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8860         md_new_event(mddev);
8861         update_time = jiffies;
8862
8863         blk_start_plug(&plug);
8864         while (j < max_sectors) {
8865                 sector_t sectors;
8866
8867                 skipped = 0;
8868
8869                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8870                     ((mddev->curr_resync > mddev->curr_resync_completed &&
8871                       (mddev->curr_resync - mddev->curr_resync_completed)
8872                       > (max_sectors >> 4)) ||
8873                      time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8874                      (j - mddev->curr_resync_completed)*2
8875                      >= mddev->resync_max - mddev->curr_resync_completed ||
8876                      mddev->curr_resync_completed > mddev->resync_max
8877                             )) {
8878                         /* time to update curr_resync_completed */
8879                         wait_event(mddev->recovery_wait,
8880                                    atomic_read(&mddev->recovery_active) == 0);
8881                         mddev->curr_resync_completed = j;
8882                         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8883                             j > mddev->recovery_cp)
8884                                 mddev->recovery_cp = j;
8885                         update_time = jiffies;
8886                         set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8887                         sysfs_notify_dirent_safe(mddev->sysfs_completed);
8888                 }
8889
8890                 while (j >= mddev->resync_max &&
8891                        !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8892                         /* As this condition is controlled by user-space,
8893                          * we can block indefinitely, so use '_interruptible'
8894                          * to avoid triggering warnings.
8895                          */
8896                         flush_signals(current); /* just in case */
8897                         wait_event_interruptible(mddev->recovery_wait,
8898                                                  mddev->resync_max > j
8899                                                  || test_bit(MD_RECOVERY_INTR,
8900                                                              &mddev->recovery));
8901                 }
8902
8903                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8904                         break;
8905
8906                 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8907                 if (sectors == 0) {
8908                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8909                         break;
8910                 }
8911
8912                 if (!skipped) { /* actual IO requested */
8913                         io_sectors += sectors;
8914                         atomic_add(sectors, &mddev->recovery_active);
8915                 }
8916
8917                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8918                         break;
8919
8920                 j += sectors;
8921                 if (j > max_sectors)
8922                         /* when skipping, extra large numbers can be returned. */
8923                         j = max_sectors;
8924                 if (j > 2)
8925                         mddev->curr_resync = j;
8926                 mddev->curr_mark_cnt = io_sectors;
8927                 if (last_check == 0)
8928                         /* this is the earliest that rebuild will be
8929                          * visible in /proc/mdstat
8930                          */
8931                         md_new_event(mddev);
8932
8933                 if (last_check + window > io_sectors || j == max_sectors)
8934                         continue;
8935
8936                 last_check = io_sectors;
8937         repeat:
8938                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8939                         /* step marks */
8940                         int next = (last_mark+1) % SYNC_MARKS;
8941
8942                         mddev->resync_mark = mark[next];
8943                         mddev->resync_mark_cnt = mark_cnt[next];
8944                         mark[next] = jiffies;
8945                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8946                         last_mark = next;
8947                 }
8948
8949                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8950                         break;
8951
8952                 /*
8953                  * this loop exits only if either when we are slower than
8954                  * the 'hard' speed limit, or the system was IO-idle for
8955                  * a jiffy.
8956                  * the system might be non-idle CPU-wise, but we only care
8957                  * about not overloading the IO subsystem. (things like an
8958                  * e2fsck being done on the RAID array should execute fast)
8959                  */
8960                 cond_resched();
8961
8962                 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8963                 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8964                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
8965
8966                 if (currspeed > speed_min(mddev)) {
8967                         if (currspeed > speed_max(mddev)) {
8968                                 msleep(500);
8969                                 goto repeat;
8970                         }
8971                         if (!is_mddev_idle(mddev, 0)) {
8972                                 /*
8973                                  * Give other IO more of a chance.
8974                                  * The faster the devices, the less we wait.
8975                                  */
8976                                 wait_event(mddev->recovery_wait,
8977                                            !atomic_read(&mddev->recovery_active));
8978                         }
8979                 }
8980         }
8981         pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8982                 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8983                 ? "interrupted" : "done");
8984         /*
8985          * this also signals 'finished resyncing' to md_stop
8986          */
8987         blk_finish_plug(&plug);
8988         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8989
8990         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8991             !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8992             mddev->curr_resync > 3) {
8993                 mddev->curr_resync_completed = mddev->curr_resync;
8994                 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8995         }
8996         mddev->pers->sync_request(mddev, max_sectors, &skipped);
8997
8998         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8999             mddev->curr_resync > 3) {
9000                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9001                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9002                                 if (mddev->curr_resync >= mddev->recovery_cp) {
9003                                         pr_debug("md: checkpointing %s of %s.\n",
9004                                                  desc, mdname(mddev));
9005                                         if (test_bit(MD_RECOVERY_ERROR,
9006                                                 &mddev->recovery))
9007                                                 mddev->recovery_cp =
9008                                                         mddev->curr_resync_completed;
9009                                         else
9010                                                 mddev->recovery_cp =
9011                                                         mddev->curr_resync;
9012                                 }
9013                         } else
9014                                 mddev->recovery_cp = MaxSector;
9015                 } else {
9016                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9017                                 mddev->curr_resync = MaxSector;
9018                         if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9019                             test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9020                                 rcu_read_lock();
9021                                 rdev_for_each_rcu(rdev, mddev)
9022                                         if (rdev->raid_disk >= 0 &&
9023                                             mddev->delta_disks >= 0 &&
9024                                             !test_bit(Journal, &rdev->flags) &&
9025                                             !test_bit(Faulty, &rdev->flags) &&
9026                                             !test_bit(In_sync, &rdev->flags) &&
9027                                             rdev->recovery_offset < mddev->curr_resync)
9028                                                 rdev->recovery_offset = mddev->curr_resync;
9029                                 rcu_read_unlock();
9030                         }
9031                 }
9032         }
9033  skip:
9034         /* set CHANGE_PENDING here since maybe another update is needed,
9035          * so other nodes are informed. It should be harmless for normal
9036          * raid */
9037         set_mask_bits(&mddev->sb_flags, 0,
9038                       BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9039
9040         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9041                         !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9042                         mddev->delta_disks > 0 &&
9043                         mddev->pers->finish_reshape &&
9044                         mddev->pers->size &&
9045                         mddev->queue) {
9046                 mddev_lock_nointr(mddev);
9047                 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9048                 mddev_unlock(mddev);
9049                 if (!mddev_is_clustered(mddev))
9050                         set_capacity_and_notify(mddev->gendisk,
9051                                                 mddev->array_sectors);
9052         }
9053
9054         spin_lock(&mddev->lock);
9055         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9056                 /* We completed so min/max setting can be forgotten if used. */
9057                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9058                         mddev->resync_min = 0;
9059                 mddev->resync_max = MaxSector;
9060         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9061                 mddev->resync_min = mddev->curr_resync_completed;
9062         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9063         mddev->curr_resync = 0;
9064         spin_unlock(&mddev->lock);
9065
9066         wake_up(&resync_wait);
9067         md_wakeup_thread(mddev->thread);
9068         return;
9069 }
9070 EXPORT_SYMBOL_GPL(md_do_sync);
9071
9072 static int remove_and_add_spares(struct mddev *mddev,
9073                                  struct md_rdev *this)
9074 {
9075         struct md_rdev *rdev;
9076         int spares = 0;
9077         int removed = 0;
9078         bool remove_some = false;
9079
9080         if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9081                 /* Mustn't remove devices when resync thread is running */
9082                 return 0;
9083
9084         rdev_for_each(rdev, mddev) {
9085                 if ((this == NULL || rdev == this) &&
9086                     rdev->raid_disk >= 0 &&
9087                     !test_bit(Blocked, &rdev->flags) &&
9088                     test_bit(Faulty, &rdev->flags) &&
9089                     atomic_read(&rdev->nr_pending)==0) {
9090                         /* Faulty non-Blocked devices with nr_pending == 0
9091                          * never get nr_pending incremented,
9092                          * never get Faulty cleared, and never get Blocked set.
9093                          * So we can synchronize_rcu now rather than once per device
9094                          */
9095                         remove_some = true;
9096                         set_bit(RemoveSynchronized, &rdev->flags);
9097                 }
9098         }
9099
9100         if (remove_some)
9101                 synchronize_rcu();
9102         rdev_for_each(rdev, mddev) {
9103                 if ((this == NULL || rdev == this) &&
9104                     rdev->raid_disk >= 0 &&
9105                     !test_bit(Blocked, &rdev->flags) &&
9106                     ((test_bit(RemoveSynchronized, &rdev->flags) ||
9107                      (!test_bit(In_sync, &rdev->flags) &&
9108                       !test_bit(Journal, &rdev->flags))) &&
9109                     atomic_read(&rdev->nr_pending)==0)) {
9110                         if (mddev->pers->hot_remove_disk(
9111                                     mddev, rdev) == 0) {
9112                                 sysfs_unlink_rdev(mddev, rdev);
9113                                 rdev->saved_raid_disk = rdev->raid_disk;
9114                                 rdev->raid_disk = -1;
9115                                 removed++;
9116                         }
9117                 }
9118                 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9119                         clear_bit(RemoveSynchronized, &rdev->flags);
9120         }
9121
9122         if (removed && mddev->kobj.sd)
9123                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9124
9125         if (this && removed)
9126                 goto no_add;
9127
9128         rdev_for_each(rdev, mddev) {
9129                 if (this && this != rdev)
9130                         continue;
9131                 if (test_bit(Candidate, &rdev->flags))
9132                         continue;
9133                 if (rdev->raid_disk >= 0 &&
9134                     !test_bit(In_sync, &rdev->flags) &&
9135                     !test_bit(Journal, &rdev->flags) &&
9136                     !test_bit(Faulty, &rdev->flags))
9137                         spares++;
9138                 if (rdev->raid_disk >= 0)
9139                         continue;
9140                 if (test_bit(Faulty, &rdev->flags))
9141                         continue;
9142                 if (!test_bit(Journal, &rdev->flags)) {
9143                         if (mddev->ro &&
9144                             ! (rdev->saved_raid_disk >= 0 &&
9145                                !test_bit(Bitmap_sync, &rdev->flags)))
9146                                 continue;
9147
9148                         rdev->recovery_offset = 0;
9149                 }
9150                 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9151                         /* failure here is OK */
9152                         sysfs_link_rdev(mddev, rdev);
9153                         if (!test_bit(Journal, &rdev->flags))
9154                                 spares++;
9155                         md_new_event(mddev);
9156                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9157                 }
9158         }
9159 no_add:
9160         if (removed)
9161                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9162         return spares;
9163 }
9164
9165 static void md_start_sync(struct work_struct *ws)
9166 {
9167         struct mddev *mddev = container_of(ws, struct mddev, del_work);
9168
9169         mddev->sync_thread = md_register_thread(md_do_sync,
9170                                                 mddev,
9171                                                 "resync");
9172         if (!mddev->sync_thread) {
9173                 pr_warn("%s: could not start resync thread...\n",
9174                         mdname(mddev));
9175                 /* leave the spares where they are, it shouldn't hurt */
9176                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9177                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9178                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9179                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9180                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9181                 wake_up(&resync_wait);
9182                 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9183                                        &mddev->recovery))
9184                         if (mddev->sysfs_action)
9185                                 sysfs_notify_dirent_safe(mddev->sysfs_action);
9186         } else
9187                 md_wakeup_thread(mddev->sync_thread);
9188         sysfs_notify_dirent_safe(mddev->sysfs_action);
9189         md_new_event(mddev);
9190 }
9191
9192 /*
9193  * This routine is regularly called by all per-raid-array threads to
9194  * deal with generic issues like resync and super-block update.
9195  * Raid personalities that don't have a thread (linear/raid0) do not
9196  * need this as they never do any recovery or update the superblock.
9197  *
9198  * It does not do any resync itself, but rather "forks" off other threads
9199  * to do that as needed.
9200  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9201  * "->recovery" and create a thread at ->sync_thread.
9202  * When the thread finishes it sets MD_RECOVERY_DONE
9203  * and wakeups up this thread which will reap the thread and finish up.
9204  * This thread also removes any faulty devices (with nr_pending == 0).
9205  *
9206  * The overall approach is:
9207  *  1/ if the superblock needs updating, update it.
9208  *  2/ If a recovery thread is running, don't do anything else.
9209  *  3/ If recovery has finished, clean up, possibly marking spares active.
9210  *  4/ If there are any faulty devices, remove them.
9211  *  5/ If array is degraded, try to add spares devices
9212  *  6/ If array has spares or is not in-sync, start a resync thread.
9213  */
9214 void md_check_recovery(struct mddev *mddev)
9215 {
9216         if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9217                 /* Write superblock - thread that called mddev_suspend()
9218                  * holds reconfig_mutex for us.
9219                  */
9220                 set_bit(MD_UPDATING_SB, &mddev->flags);
9221                 smp_mb__after_atomic();
9222                 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9223                         md_update_sb(mddev, 0);
9224                 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9225                 wake_up(&mddev->sb_wait);
9226         }
9227
9228         if (mddev->suspended)
9229                 return;
9230
9231         if (mddev->bitmap)
9232                 md_bitmap_daemon_work(mddev);
9233
9234         if (signal_pending(current)) {
9235                 if (mddev->pers->sync_request && !mddev->external) {
9236                         pr_debug("md: %s in immediate safe mode\n",
9237                                  mdname(mddev));
9238                         mddev->safemode = 2;
9239                 }
9240                 flush_signals(current);
9241         }
9242
9243         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9244                 return;
9245         if ( ! (
9246                 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9247                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9248                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9249                 (mddev->external == 0 && mddev->safemode == 1) ||
9250                 (mddev->safemode == 2
9251                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9252                 ))
9253                 return;
9254
9255         if (mddev_trylock(mddev)) {
9256                 int spares = 0;
9257                 bool try_set_sync = mddev->safemode != 0;
9258
9259                 if (!mddev->external && mddev->safemode == 1)
9260                         mddev->safemode = 0;
9261
9262                 if (mddev->ro) {
9263                         struct md_rdev *rdev;
9264                         if (!mddev->external && mddev->in_sync)
9265                                 /* 'Blocked' flag not needed as failed devices
9266                                  * will be recorded if array switched to read/write.
9267                                  * Leaving it set will prevent the device
9268                                  * from being removed.
9269                                  */
9270                                 rdev_for_each(rdev, mddev)
9271                                         clear_bit(Blocked, &rdev->flags);
9272                         /* On a read-only array we can:
9273                          * - remove failed devices
9274                          * - add already-in_sync devices if the array itself
9275                          *   is in-sync.
9276                          * As we only add devices that are already in-sync,
9277                          * we can activate the spares immediately.
9278                          */
9279                         remove_and_add_spares(mddev, NULL);
9280                         /* There is no thread, but we need to call
9281                          * ->spare_active and clear saved_raid_disk
9282                          */
9283                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9284                         md_reap_sync_thread(mddev);
9285                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9286                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9287                         clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9288                         goto unlock;
9289                 }
9290
9291                 if (mddev_is_clustered(mddev)) {
9292                         struct md_rdev *rdev, *tmp;
9293                         /* kick the device if another node issued a
9294                          * remove disk.
9295                          */
9296                         rdev_for_each_safe(rdev, tmp, mddev) {
9297                                 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9298                                                 rdev->raid_disk < 0)
9299                                         md_kick_rdev_from_array(rdev);
9300                         }
9301                 }
9302
9303                 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9304                         spin_lock(&mddev->lock);
9305                         set_in_sync(mddev);
9306                         spin_unlock(&mddev->lock);
9307                 }
9308
9309                 if (mddev->sb_flags)
9310                         md_update_sb(mddev, 0);
9311
9312                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9313                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9314                         /* resync/recovery still happening */
9315                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9316                         goto unlock;
9317                 }
9318                 if (mddev->sync_thread) {
9319                         md_reap_sync_thread(mddev);
9320                         goto unlock;
9321                 }
9322                 /* Set RUNNING before clearing NEEDED to avoid
9323                  * any transients in the value of "sync_action".
9324                  */
9325                 mddev->curr_resync_completed = 0;
9326                 spin_lock(&mddev->lock);
9327                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9328                 spin_unlock(&mddev->lock);
9329                 /* Clear some bits that don't mean anything, but
9330                  * might be left set
9331                  */
9332                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9333                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9334
9335                 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9336                     test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9337                         goto not_running;
9338                 /* no recovery is running.
9339                  * remove any failed drives, then
9340                  * add spares if possible.
9341                  * Spares are also removed and re-added, to allow
9342                  * the personality to fail the re-add.
9343                  */
9344
9345                 if (mddev->reshape_position != MaxSector) {
9346                         if (mddev->pers->check_reshape == NULL ||
9347                             mddev->pers->check_reshape(mddev) != 0)
9348                                 /* Cannot proceed */
9349                                 goto not_running;
9350                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9351                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9352                 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9353                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9354                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9355                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9356                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9357                 } else if (mddev->recovery_cp < MaxSector) {
9358                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9359                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9360                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9361                         /* nothing to be done ... */
9362                         goto not_running;
9363
9364                 if (mddev->pers->sync_request) {
9365                         if (spares) {
9366                                 /* We are adding a device or devices to an array
9367                                  * which has the bitmap stored on all devices.
9368                                  * So make sure all bitmap pages get written
9369                                  */
9370                                 md_bitmap_write_all(mddev->bitmap);
9371                         }
9372                         INIT_WORK(&mddev->del_work, md_start_sync);
9373                         queue_work(md_misc_wq, &mddev->del_work);
9374                         goto unlock;
9375                 }
9376         not_running:
9377                 if (!mddev->sync_thread) {
9378                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9379                         wake_up(&resync_wait);
9380                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9381                                                &mddev->recovery))
9382                                 if (mddev->sysfs_action)
9383                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
9384                 }
9385         unlock:
9386                 wake_up(&mddev->sb_wait);
9387                 mddev_unlock(mddev);
9388         }
9389 }
9390 EXPORT_SYMBOL(md_check_recovery);
9391
9392 void md_reap_sync_thread(struct mddev *mddev)
9393 {
9394         struct md_rdev *rdev;
9395         sector_t old_dev_sectors = mddev->dev_sectors;
9396         bool is_reshaped = false;
9397
9398         /* resync has finished, collect result */
9399         md_unregister_thread(&mddev->sync_thread);
9400         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9401             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9402             mddev->degraded != mddev->raid_disks) {
9403                 /* success...*/
9404                 /* activate any spares */
9405                 if (mddev->pers->spare_active(mddev)) {
9406                         sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9407                         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9408                 }
9409         }
9410         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9411             mddev->pers->finish_reshape) {
9412                 mddev->pers->finish_reshape(mddev);
9413                 if (mddev_is_clustered(mddev))
9414                         is_reshaped = true;
9415         }
9416
9417         /* If array is no-longer degraded, then any saved_raid_disk
9418          * information must be scrapped.
9419          */
9420         if (!mddev->degraded)
9421                 rdev_for_each(rdev, mddev)
9422                         rdev->saved_raid_disk = -1;
9423
9424         md_update_sb(mddev, 1);
9425         /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9426          * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9427          * clustered raid */
9428         if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9429                 md_cluster_ops->resync_finish(mddev);
9430         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9431         clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9432         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9433         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9434         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9435         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9436         /*
9437          * We call md_cluster_ops->update_size here because sync_size could
9438          * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9439          * so it is time to update size across cluster.
9440          */
9441         if (mddev_is_clustered(mddev) && is_reshaped
9442                                       && !test_bit(MD_CLOSING, &mddev->flags))
9443                 md_cluster_ops->update_size(mddev, old_dev_sectors);
9444         wake_up(&resync_wait);
9445         /* flag recovery needed just to double check */
9446         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9447         sysfs_notify_dirent_safe(mddev->sysfs_action);
9448         md_new_event(mddev);
9449         if (mddev->event_work.func)
9450                 queue_work(md_misc_wq, &mddev->event_work);
9451 }
9452 EXPORT_SYMBOL(md_reap_sync_thread);
9453
9454 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9455 {
9456         sysfs_notify_dirent_safe(rdev->sysfs_state);
9457         wait_event_timeout(rdev->blocked_wait,
9458                            !test_bit(Blocked, &rdev->flags) &&
9459                            !test_bit(BlockedBadBlocks, &rdev->flags),
9460                            msecs_to_jiffies(5000));
9461         rdev_dec_pending(rdev, mddev);
9462 }
9463 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9464
9465 void md_finish_reshape(struct mddev *mddev)
9466 {
9467         /* called be personality module when reshape completes. */
9468         struct md_rdev *rdev;
9469
9470         rdev_for_each(rdev, mddev) {
9471                 if (rdev->data_offset > rdev->new_data_offset)
9472                         rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9473                 else
9474                         rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9475                 rdev->data_offset = rdev->new_data_offset;
9476         }
9477 }
9478 EXPORT_SYMBOL(md_finish_reshape);
9479
9480 /* Bad block management */
9481
9482 /* Returns 1 on success, 0 on failure */
9483 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9484                        int is_new)
9485 {
9486         struct mddev *mddev = rdev->mddev;
9487         int rv;
9488         if (is_new)
9489                 s += rdev->new_data_offset;
9490         else
9491                 s += rdev->data_offset;
9492         rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9493         if (rv == 0) {
9494                 /* Make sure they get written out promptly */
9495                 if (test_bit(ExternalBbl, &rdev->flags))
9496                         sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9497                 sysfs_notify_dirent_safe(rdev->sysfs_state);
9498                 set_mask_bits(&mddev->sb_flags, 0,
9499                               BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9500                 md_wakeup_thread(rdev->mddev->thread);
9501                 return 1;
9502         } else
9503                 return 0;
9504 }
9505 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9506
9507 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9508                          int is_new)
9509 {
9510         int rv;
9511         if (is_new)
9512                 s += rdev->new_data_offset;
9513         else
9514                 s += rdev->data_offset;
9515         rv = badblocks_clear(&rdev->badblocks, s, sectors);
9516         if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9517                 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9518         return rv;
9519 }
9520 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9521
9522 static int md_notify_reboot(struct notifier_block *this,
9523                             unsigned long code, void *x)
9524 {
9525         struct list_head *tmp;
9526         struct mddev *mddev;
9527         int need_delay = 0;
9528
9529         for_each_mddev(mddev, tmp) {
9530                 if (mddev_trylock(mddev)) {
9531                         if (mddev->pers)
9532                                 __md_stop_writes(mddev);
9533                         if (mddev->persistent)
9534                                 mddev->safemode = 2;
9535                         mddev_unlock(mddev);
9536                 }
9537                 need_delay = 1;
9538         }
9539         /*
9540          * certain more exotic SCSI devices are known to be
9541          * volatile wrt too early system reboots. While the
9542          * right place to handle this issue is the given
9543          * driver, we do want to have a safe RAID driver ...
9544          */
9545         if (need_delay)
9546                 mdelay(1000*1);
9547
9548         return NOTIFY_DONE;
9549 }
9550
9551 static struct notifier_block md_notifier = {
9552         .notifier_call  = md_notify_reboot,
9553         .next           = NULL,
9554         .priority       = INT_MAX, /* before any real devices */
9555 };
9556
9557 static void md_geninit(void)
9558 {
9559         pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9560
9561         proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9562 }
9563
9564 static int __init md_init(void)
9565 {
9566         int ret = -ENOMEM;
9567
9568         md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9569         if (!md_wq)
9570                 goto err_wq;
9571
9572         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9573         if (!md_misc_wq)
9574                 goto err_misc_wq;
9575
9576         md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9577         if (!md_rdev_misc_wq)
9578                 goto err_rdev_misc_wq;
9579
9580         ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9581         if (ret < 0)
9582                 goto err_md;
9583
9584         ret = __register_blkdev(0, "mdp", md_probe);
9585         if (ret < 0)
9586                 goto err_mdp;
9587         mdp_major = ret;
9588
9589         register_reboot_notifier(&md_notifier);
9590         raid_table_header = register_sysctl_table(raid_root_table);
9591
9592         md_geninit();
9593         return 0;
9594
9595 err_mdp:
9596         unregister_blkdev(MD_MAJOR, "md");
9597 err_md:
9598         destroy_workqueue(md_rdev_misc_wq);
9599 err_rdev_misc_wq:
9600         destroy_workqueue(md_misc_wq);
9601 err_misc_wq:
9602         destroy_workqueue(md_wq);
9603 err_wq:
9604         return ret;
9605 }
9606
9607 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9608 {
9609         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9610         struct md_rdev *rdev2, *tmp;
9611         int role, ret;
9612         char b[BDEVNAME_SIZE];
9613
9614         /*
9615          * If size is changed in another node then we need to
9616          * do resize as well.
9617          */
9618         if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9619                 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9620                 if (ret)
9621                         pr_info("md-cluster: resize failed\n");
9622                 else
9623                         md_bitmap_update_sb(mddev->bitmap);
9624         }
9625
9626         /* Check for change of roles in the active devices */
9627         rdev_for_each_safe(rdev2, tmp, mddev) {
9628                 if (test_bit(Faulty, &rdev2->flags))
9629                         continue;
9630
9631                 /* Check if the roles changed */
9632                 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9633
9634                 if (test_bit(Candidate, &rdev2->flags)) {
9635                         if (role == 0xfffe) {
9636                                 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9637                                 md_kick_rdev_from_array(rdev2);
9638                                 continue;
9639                         }
9640                         else
9641                                 clear_bit(Candidate, &rdev2->flags);
9642                 }
9643
9644                 if (role != rdev2->raid_disk) {
9645                         /*
9646                          * got activated except reshape is happening.
9647                          */
9648                         if (rdev2->raid_disk == -1 && role != 0xffff &&
9649                             !(le32_to_cpu(sb->feature_map) &
9650                               MD_FEATURE_RESHAPE_ACTIVE)) {
9651                                 rdev2->saved_raid_disk = role;
9652                                 ret = remove_and_add_spares(mddev, rdev2);
9653                                 pr_info("Activated spare: %s\n",
9654                                         bdevname(rdev2->bdev,b));
9655                                 /* wakeup mddev->thread here, so array could
9656                                  * perform resync with the new activated disk */
9657                                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9658                                 md_wakeup_thread(mddev->thread);
9659                         }
9660                         /* device faulty
9661                          * We just want to do the minimum to mark the disk
9662                          * as faulty. The recovery is performed by the
9663                          * one who initiated the error.
9664                          */
9665                         if ((role == 0xfffe) || (role == 0xfffd)) {
9666                                 md_error(mddev, rdev2);
9667                                 clear_bit(Blocked, &rdev2->flags);
9668                         }
9669                 }
9670         }
9671
9672         if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9673                 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9674                 if (ret)
9675                         pr_warn("md: updating array disks failed. %d\n", ret);
9676         }
9677
9678         /*
9679          * Since mddev->delta_disks has already updated in update_raid_disks,
9680          * so it is time to check reshape.
9681          */
9682         if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9683             (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9684                 /*
9685                  * reshape is happening in the remote node, we need to
9686                  * update reshape_position and call start_reshape.
9687                  */
9688                 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9689                 if (mddev->pers->update_reshape_pos)
9690                         mddev->pers->update_reshape_pos(mddev);
9691                 if (mddev->pers->start_reshape)
9692                         mddev->pers->start_reshape(mddev);
9693         } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9694                    mddev->reshape_position != MaxSector &&
9695                    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9696                 /* reshape is just done in another node. */
9697                 mddev->reshape_position = MaxSector;
9698                 if (mddev->pers->update_reshape_pos)
9699                         mddev->pers->update_reshape_pos(mddev);
9700         }
9701
9702         /* Finally set the event to be up to date */
9703         mddev->events = le64_to_cpu(sb->events);
9704 }
9705
9706 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9707 {
9708         int err;
9709         struct page *swapout = rdev->sb_page;
9710         struct mdp_superblock_1 *sb;
9711
9712         /* Store the sb page of the rdev in the swapout temporary
9713          * variable in case we err in the future
9714          */
9715         rdev->sb_page = NULL;
9716         err = alloc_disk_sb(rdev);
9717         if (err == 0) {
9718                 ClearPageUptodate(rdev->sb_page);
9719                 rdev->sb_loaded = 0;
9720                 err = super_types[mddev->major_version].
9721                         load_super(rdev, NULL, mddev->minor_version);
9722         }
9723         if (err < 0) {
9724                 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9725                                 __func__, __LINE__, rdev->desc_nr, err);
9726                 if (rdev->sb_page)
9727                         put_page(rdev->sb_page);
9728                 rdev->sb_page = swapout;
9729                 rdev->sb_loaded = 1;
9730                 return err;
9731         }
9732
9733         sb = page_address(rdev->sb_page);
9734         /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9735          * is not set
9736          */
9737
9738         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9739                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9740
9741         /* The other node finished recovery, call spare_active to set
9742          * device In_sync and mddev->degraded
9743          */
9744         if (rdev->recovery_offset == MaxSector &&
9745             !test_bit(In_sync, &rdev->flags) &&
9746             mddev->pers->spare_active(mddev))
9747                 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9748
9749         put_page(swapout);
9750         return 0;
9751 }
9752
9753 void md_reload_sb(struct mddev *mddev, int nr)
9754 {
9755         struct md_rdev *rdev;
9756         int err;
9757
9758         /* Find the rdev */
9759         rdev_for_each_rcu(rdev, mddev) {
9760                 if (rdev->desc_nr == nr)
9761                         break;
9762         }
9763
9764         if (!rdev || rdev->desc_nr != nr) {
9765                 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9766                 return;
9767         }
9768
9769         err = read_rdev(mddev, rdev);
9770         if (err < 0)
9771                 return;
9772
9773         check_sb_changes(mddev, rdev);
9774
9775         /* Read all rdev's to update recovery_offset */
9776         rdev_for_each_rcu(rdev, mddev) {
9777                 if (!test_bit(Faulty, &rdev->flags))
9778                         read_rdev(mddev, rdev);
9779         }
9780 }
9781 EXPORT_SYMBOL(md_reload_sb);
9782
9783 #ifndef MODULE
9784
9785 /*
9786  * Searches all registered partitions for autorun RAID arrays
9787  * at boot time.
9788  */
9789
9790 static DEFINE_MUTEX(detected_devices_mutex);
9791 static LIST_HEAD(all_detected_devices);
9792 struct detected_devices_node {
9793         struct list_head list;
9794         dev_t dev;
9795 };
9796
9797 void md_autodetect_dev(dev_t dev)
9798 {
9799         struct detected_devices_node *node_detected_dev;
9800
9801         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9802         if (node_detected_dev) {
9803                 node_detected_dev->dev = dev;
9804                 mutex_lock(&detected_devices_mutex);
9805                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9806                 mutex_unlock(&detected_devices_mutex);
9807         }
9808 }
9809
9810 void md_autostart_arrays(int part)
9811 {
9812         struct md_rdev *rdev;
9813         struct detected_devices_node *node_detected_dev;
9814         dev_t dev;
9815         int i_scanned, i_passed;
9816
9817         i_scanned = 0;
9818         i_passed = 0;
9819
9820         pr_info("md: Autodetecting RAID arrays.\n");
9821
9822         mutex_lock(&detected_devices_mutex);
9823         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9824                 i_scanned++;
9825                 node_detected_dev = list_entry(all_detected_devices.next,
9826                                         struct detected_devices_node, list);
9827                 list_del(&node_detected_dev->list);
9828                 dev = node_detected_dev->dev;
9829                 kfree(node_detected_dev);
9830                 mutex_unlock(&detected_devices_mutex);
9831                 rdev = md_import_device(dev,0, 90);
9832                 mutex_lock(&detected_devices_mutex);
9833                 if (IS_ERR(rdev))
9834                         continue;
9835
9836                 if (test_bit(Faulty, &rdev->flags))
9837                         continue;
9838
9839                 set_bit(AutoDetected, &rdev->flags);
9840                 list_add(&rdev->same_set, &pending_raid_disks);
9841                 i_passed++;
9842         }
9843         mutex_unlock(&detected_devices_mutex);
9844
9845         pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9846
9847         autorun_devices(part);
9848 }
9849
9850 #endif /* !MODULE */
9851
9852 static __exit void md_exit(void)
9853 {
9854         struct mddev *mddev;
9855         struct list_head *tmp;
9856         int delay = 1;
9857
9858         unregister_blkdev(MD_MAJOR,"md");
9859         unregister_blkdev(mdp_major, "mdp");
9860         unregister_reboot_notifier(&md_notifier);
9861         unregister_sysctl_table(raid_table_header);
9862
9863         /* We cannot unload the modules while some process is
9864          * waiting for us in select() or poll() - wake them up
9865          */
9866         md_unloading = 1;
9867         while (waitqueue_active(&md_event_waiters)) {
9868                 /* not safe to leave yet */
9869                 wake_up(&md_event_waiters);
9870                 msleep(delay);
9871                 delay += delay;
9872         }
9873         remove_proc_entry("mdstat", NULL);
9874
9875         for_each_mddev(mddev, tmp) {
9876                 export_array(mddev);
9877                 mddev->ctime = 0;
9878                 mddev->hold_active = 0;
9879                 /*
9880                  * for_each_mddev() will call mddev_put() at the end of each
9881                  * iteration.  As the mddev is now fully clear, this will
9882                  * schedule the mddev for destruction by a workqueue, and the
9883                  * destroy_workqueue() below will wait for that to complete.
9884                  */
9885         }
9886         destroy_workqueue(md_rdev_misc_wq);
9887         destroy_workqueue(md_misc_wq);
9888         destroy_workqueue(md_wq);
9889 }
9890
9891 subsys_initcall(md_init);
9892 module_exit(md_exit)
9893
9894 static int get_ro(char *buffer, const struct kernel_param *kp)
9895 {
9896         return sprintf(buffer, "%d\n", start_readonly);
9897 }
9898 static int set_ro(const char *val, const struct kernel_param *kp)
9899 {
9900         return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9901 }
9902
9903 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9904 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9905 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9906 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9907
9908 MODULE_LICENSE("GPL");
9909 MODULE_DESCRIPTION("MD RAID framework");
9910 MODULE_ALIAS("md");
9911 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);