drivers/md/md.c

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