drivers/md/md.c

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