drivers/md/md.c

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