drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/kthread.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/sysctl.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/fs.h>
  40 #include <linux/poll.h>
  41 #include <linux/ctype.h>
  42 #include <linux/string.h>
  43 #include <linux/hdreg.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/random.h>
  46 #include <linux/module.h>
  47 #include <linux/reboot.h>
  48 #include <linux/file.h>
  49 #include <linux/compat.h>
  50 #include <linux/delay.h>
  51 #include <linux/raid/md_p.h>
  52 #include <linux/raid/md_u.h>
  53 #include <linux/slab.h>
  54 #include "md.h"
  55 #include "bitmap.h"
  56
  57 #ifndef MODULE
  58 static void autostart_arrays(int part);
  59 #endif
  60
  61 /* pers_list is a list of registered personalities protected
  62  * by pers_lock.
  63  * pers_lock does extra service to protect accesses to
  64  * mddev->thread when the mutex cannot be held.
  65  */
  66 static LIST_HEAD(pers_list);
  67 static DEFINE_SPINLOCK(pers_lock);
  68
  69 static void md_print_devices(void);
  70
  71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  72 static struct workqueue_struct *md_wq;
  73 static struct workqueue_struct *md_misc_wq;
  74
  75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  76
  77 /*
  78  * Default number of read corrections we'll attempt on an rdev
  79  * before ejecting it from the array. We divide the read error
  80  * count by 2 for every hour elapsed between read errors.
  81  */
  82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  83 /*
  84  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  85  * is 1000 KB/sec, so the extra system load does not show up that much.
  86  * Increase it if you want to have more _guaranteed_ speed. Note that
  87  * the RAID driver will use the maximum available bandwidth if the IO
  88  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  89  * speed limit - in case reconstruction slows down your system despite
  90  * idle IO detection.
  91  *
  92  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  93  * or /sys/block/mdX/md/sync_speed_{min,max}
  94  */
  95
  96 static int sysctl_speed_limit_min = 1000;
  97 static int sysctl_speed_limit_max = 200000;
  98 static inline int speed_min(struct mddev *mddev)
  99 {
 100         return mddev->sync_speed_min ?
 101                 mddev->sync_speed_min : sysctl_speed_limit_min;
 102 }
 103
 104 static inline int speed_max(struct mddev *mddev)
 105 {
 106         return mddev->sync_speed_max ?
 107                 mddev->sync_speed_max : sysctl_speed_limit_max;
 108 }
 109
 110 static struct ctl_table_header *raid_table_header;
 111
 112 static ctl_table raid_table[] = {
 113         {
 114                 .procname       = "speed_limit_min",
 115                 .data           = &sysctl_speed_limit_min,
 116                 .maxlen         = sizeof(int),
 117                 .mode           = S_IRUGO|S_IWUSR,
 118                 .proc_handler   = proc_dointvec,
 119         },
 120         {
 121                 .procname       = "speed_limit_max",
 122                 .data           = &sysctl_speed_limit_max,
 123                 .maxlen         = sizeof(int),
 124                 .mode           = S_IRUGO|S_IWUSR,
 125                 .proc_handler   = proc_dointvec,
 126         },
 127         { }
 128 };
 129
 130 static ctl_table raid_dir_table[] = {
 131         {
 132                 .procname       = "raid",
 133                 .maxlen         = 0,
 134                 .mode           = S_IRUGO|S_IXUGO,
 135                 .child          = raid_table,
 136         },
 137         { }
 138 };
 139
 140 static ctl_table raid_root_table[] = {
 141         {
 142                 .procname       = "dev",
 143                 .maxlen         = 0,
 144                 .mode           = 0555,
 145                 .child          = raid_dir_table,
 146         },
 147         {  }
 148 };
 149
 150 static const struct block_device_operations md_fops;
 151
 152 static int start_readonly;
 153
 154 /* bio_clone_mddev
 155  * like bio_clone, but with a local bio set
 156  */
 157
 158 static void mddev_bio_destructor(struct bio *bio)
 159 {
 160         struct mddev *mddev, **mddevp;
 161
 162         mddevp = (void*)bio;
 163         mddev = mddevp[-1];
 164
 165         bio_free(bio, mddev->bio_set);
 166 }
 167
 168 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 169                             struct mddev *mddev)
 170 {
 171         struct bio *b;
 172         struct mddev **mddevp;
 173
 174         if (!mddev || !mddev->bio_set)
 175                 return bio_alloc(gfp_mask, nr_iovecs);
 176
 177         b = bio_alloc_bioset(gfp_mask, nr_iovecs,
 178                              mddev->bio_set);
 179         if (!b)
 180                 return NULL;
 181         mddevp = (void*)b;
 182         mddevp[-1] = mddev;
 183         b->bi_destructor = mddev_bio_destructor;
 184         return b;
 185 }
 186 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 187
 188 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
 189                             struct mddev *mddev)
 190 {
 191         struct bio *b;
 192         struct mddev **mddevp;
 193
 194         if (!mddev || !mddev->bio_set)
 195                 return bio_clone(bio, gfp_mask);
 196
 197         b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
 198                              mddev->bio_set);
 199         if (!b)
 200                 return NULL;
 201         mddevp = (void*)b;
 202         mddevp[-1] = mddev;
 203         b->bi_destructor = mddev_bio_destructor;
 204         __bio_clone(b, bio);
 205         if (bio_integrity(bio)) {
 206                 int ret;
 207
 208                 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
 209
 210                 if (ret < 0) {
 211                         bio_put(b);
 212                         return NULL;
 213                 }
 214         }
 215
 216         return b;
 217 }
 218 EXPORT_SYMBOL_GPL(bio_clone_mddev);
 219
 220 void md_trim_bio(struct bio *bio, int offset, int size)
 221 {
 222         /* 'bio' is a cloned bio which we need to trim to match
 223          * the given offset and size.
 224          * This requires adjusting bi_sector, bi_size, and bi_io_vec
 225          */
 226         int i;
 227         struct bio_vec *bvec;
 228         int sofar = 0;
 229
 230         size <<= 9;
 231         if (offset == 0 && size == bio->bi_size)
 232                 return;
 233
 234         bio->bi_sector += offset;
 235         bio->bi_size = size;
 236         offset <<= 9;
 237         clear_bit(BIO_SEG_VALID, &bio->bi_flags);
 238
 239         while (bio->bi_idx < bio->bi_vcnt &&
 240                bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
 241                 /* remove this whole bio_vec */
 242                 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
 243                 bio->bi_idx++;
 244         }
 245         if (bio->bi_idx < bio->bi_vcnt) {
 246                 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
 247                 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
 248         }
 249         /* avoid any complications with bi_idx being non-zero*/
 250         if (bio->bi_idx) {
 251                 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
 252                         (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
 253                 bio->bi_vcnt -= bio->bi_idx;
 254                 bio->bi_idx = 0;
 255         }
 256         /* Make sure vcnt and last bv are not too big */
 257         bio_for_each_segment(bvec, bio, i) {
 258                 if (sofar + bvec->bv_len > size)
 259                         bvec->bv_len = size - sofar;
 260                 if (bvec->bv_len == 0) {
 261                         bio->bi_vcnt = i;
 262                         break;
 263                 }
 264                 sofar += bvec->bv_len;
 265         }
 266 }
 267 EXPORT_SYMBOL_GPL(md_trim_bio);
 268
 269 /*
 270  * We have a system wide 'event count' that is incremented
 271  * on any 'interesting' event, and readers of /proc/mdstat
 272  * can use 'poll' or 'select' to find out when the event
 273  * count increases.
 274  *
 275  * Events are:
 276  *  start array, stop array, error, add device, remove device,
 277  *  start build, activate spare
 278  */
 279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 280 static atomic_t md_event_count;
 281 void md_new_event(struct mddev *mddev)
 282 {
 283         atomic_inc(&md_event_count);
 284         wake_up(&md_event_waiters);
 285 }
 286 EXPORT_SYMBOL_GPL(md_new_event);
 287
 288 /* Alternate version that can be called from interrupts
 289  * when calling sysfs_notify isn't needed.
 290  */
 291 static void md_new_event_inintr(struct mddev *mddev)
 292 {
 293         atomic_inc(&md_event_count);
 294         wake_up(&md_event_waiters);
 295 }
 296
 297 /*
 298  * Enables to iterate over all existing md arrays
 299  * all_mddevs_lock protects this list.
 300  */
 301 static LIST_HEAD(all_mddevs);
 302 static DEFINE_SPINLOCK(all_mddevs_lock);
 303
 304
 305 /*
 306  * iterates through all used mddevs in the system.
 307  * We take care to grab the all_mddevs_lock whenever navigating
 308  * the list, and to always hold a refcount when unlocked.
 309  * Any code which breaks out of this loop while own
 310  * a reference to the current mddev and must mddev_put it.
 311  */
 312 #define for_each_mddev(_mddev,_tmp)                                     \
 313                                                                         \
 314         for (({ spin_lock(&all_mddevs_lock);                            \
 315                 _tmp = all_mddevs.next;                                 \
 316                 _mddev = NULL;});                                       \
 317              ({ if (_tmp != &all_mddevs)                                \
 318                         mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
 319                 spin_unlock(&all_mddevs_lock);                          \
 320                 if (_mddev) mddev_put(_mddev);                          \
 321                 _mddev = list_entry(_tmp, struct mddev, all_mddevs);    \
 322                 _tmp != &all_mddevs;});                                 \
 323              ({ spin_lock(&all_mddevs_lock);                            \
 324                 _tmp = _tmp->next;})                                    \
 325                 )
 326
 327
 328 /* Rather than calling directly into the personality make_request function,
 329  * IO requests come here first so that we can check if the device is
 330  * being suspended pending a reconfiguration.
 331  * We hold a refcount over the call to ->make_request.  By the time that
 332  * call has finished, the bio has been linked into some internal structure
 333  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 334  */
 335 static void md_make_request(struct request_queue *q, struct bio *bio)
 336 {
 337         const int rw = bio_data_dir(bio);
 338         struct mddev *mddev = q->queuedata;
 339         int cpu;
 340         unsigned int sectors;
 341
 342         if (mddev == NULL || mddev->pers == NULL
 343             || !mddev->ready) {
 344                 bio_io_error(bio);
 345                 return;
 346         }
 347         smp_rmb(); /* Ensure implications of  'active' are visible */
 348         rcu_read_lock();
 349         if (mddev->suspended) {
 350                 DEFINE_WAIT(__wait);
 351                 for (;;) {
 352                         prepare_to_wait(&mddev->sb_wait, &__wait,
 353                                         TASK_UNINTERRUPTIBLE);
 354                         if (!mddev->suspended)
 355                                 break;
 356                         rcu_read_unlock();
 357                         schedule();
 358                         rcu_read_lock();
 359                 }
 360                 finish_wait(&mddev->sb_wait, &__wait);
 361         }
 362         atomic_inc(&mddev->active_io);
 363         rcu_read_unlock();
 364
 365         /*
 366          * save the sectors now since our bio can
 367          * go away inside make_request
 368          */
 369         sectors = bio_sectors(bio);
 370         mddev->pers->make_request(mddev, bio);
 371
 372         cpu = part_stat_lock();
 373         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 374         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
 375         part_stat_unlock();
 376
 377         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 378                 wake_up(&mddev->sb_wait);
 379 }
 380
 381 /* mddev_suspend makes sure no new requests are submitted
 382  * to the device, and that any requests that have been submitted
 383  * are completely handled.
 384  * Once ->stop is called and completes, the module will be completely
 385  * unused.
 386  */
 387 void mddev_suspend(struct mddev *mddev)
 388 {
 389         BUG_ON(mddev->suspended);
 390         mddev->suspended = 1;
 391         synchronize_rcu();
 392         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 393         mddev->pers->quiesce(mddev, 1);
 394
 395         del_timer_sync(&mddev->safemode_timer);
 396 }
 397 EXPORT_SYMBOL_GPL(mddev_suspend);
 398
 399 void mddev_resume(struct mddev *mddev)
 400 {
 401         mddev->suspended = 0;
 402         wake_up(&mddev->sb_wait);
 403         mddev->pers->quiesce(mddev, 0);
 404
 405         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 406         md_wakeup_thread(mddev->thread);
 407         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 408 }
 409 EXPORT_SYMBOL_GPL(mddev_resume);
 410
 411 int mddev_congested(struct mddev *mddev, int bits)
 412 {
 413         return mddev->suspended;
 414 }
 415 EXPORT_SYMBOL(mddev_congested);
 416
 417 /*
 418  * Generic flush handling for md
 419  */
 420
 421 static void md_end_flush(struct bio *bio, int err)
 422 {
 423         struct md_rdev *rdev = bio->bi_private;
 424         struct mddev *mddev = rdev->mddev;
 425
 426         rdev_dec_pending(rdev, mddev);
 427
 428         if (atomic_dec_and_test(&mddev->flush_pending)) {
 429                 /* The pre-request flush has finished */
 430                 queue_work(md_wq, &mddev->flush_work);
 431         }
 432         bio_put(bio);
 433 }
 434
 435 static void md_submit_flush_data(struct work_struct *ws);
 436
 437 static void submit_flushes(struct work_struct *ws)
 438 {
 439         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 440         struct md_rdev *rdev;
 441
 442         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 443         atomic_set(&mddev->flush_pending, 1);
 444         rcu_read_lock();
 445         rdev_for_each_rcu(rdev, mddev)
 446                 if (rdev->raid_disk >= 0 &&
 447                     !test_bit(Faulty, &rdev->flags)) {
 448                         /* Take two references, one is dropped
 449                          * when request finishes, one after
 450                          * we reclaim rcu_read_lock
 451                          */
 452                         struct bio *bi;
 453                         atomic_inc(&rdev->nr_pending);
 454                         atomic_inc(&rdev->nr_pending);
 455                         rcu_read_unlock();
 456                         bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
 457                         bi->bi_end_io = md_end_flush;
 458                         bi->bi_private = rdev;
 459                         bi->bi_bdev = rdev->bdev;
 460                         atomic_inc(&mddev->flush_pending);
 461                         submit_bio(WRITE_FLUSH, bi);
 462                         rcu_read_lock();
 463                         rdev_dec_pending(rdev, mddev);
 464                 }
 465         rcu_read_unlock();
 466         if (atomic_dec_and_test(&mddev->flush_pending))
 467                 queue_work(md_wq, &mddev->flush_work);
 468 }
 469
 470 static void md_submit_flush_data(struct work_struct *ws)
 471 {
 472         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 473         struct bio *bio = mddev->flush_bio;
 474
 475         if (bio->bi_size == 0)
 476                 /* an empty barrier - all done */
 477                 bio_endio(bio, 0);
 478         else {
 479                 bio->bi_rw &= ~REQ_FLUSH;
 480                 mddev->pers->make_request(mddev, bio);
 481         }
 482
 483         mddev->flush_bio = NULL;
 484         wake_up(&mddev->sb_wait);
 485 }
 486
 487 void md_flush_request(struct mddev *mddev, struct bio *bio)
 488 {
 489         spin_lock_irq(&mddev->write_lock);
 490         wait_event_lock_irq(mddev->sb_wait,
 491                             !mddev->flush_bio,
 492                             mddev->write_lock, /*nothing*/);
 493         mddev->flush_bio = bio;
 494         spin_unlock_irq(&mddev->write_lock);
 495
 496         INIT_WORK(&mddev->flush_work, submit_flushes);
 497         queue_work(md_wq, &mddev->flush_work);
 498 }
 499 EXPORT_SYMBOL(md_flush_request);
 500
 501 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
 502 {
 503         struct mddev *mddev = cb->data;
 504         md_wakeup_thread(mddev->thread);
 505         kfree(cb);
 506 }
 507 EXPORT_SYMBOL(md_unplug);
 508
 509 static inline struct mddev *mddev_get(struct mddev *mddev)
 510 {
 511         atomic_inc(&mddev->active);
 512         return mddev;
 513 }
 514
 515 static void mddev_delayed_delete(struct work_struct *ws);
 516
 517 static void mddev_put(struct mddev *mddev)
 518 {
 519         struct bio_set *bs = NULL;
 520
 521         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 522                 return;
 523         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 524             mddev->ctime == 0 && !mddev->hold_active) {
 525                 /* Array is not configured at all, and not held active,
 526                  * so destroy it */
 527                 list_del_init(&mddev->all_mddevs);
 528                 bs = mddev->bio_set;
 529                 mddev->bio_set = NULL;
 530                 if (mddev->gendisk) {
 531                         /* We did a probe so need to clean up.  Call
 532                          * queue_work inside the spinlock so that
 533                          * flush_workqueue() after mddev_find will
 534                          * succeed in waiting for the work to be done.
 535                          */
 536                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 537                         queue_work(md_misc_wq, &mddev->del_work);
 538                 } else
 539                         kfree(mddev);
 540         }
 541         spin_unlock(&all_mddevs_lock);
 542         if (bs)
 543                 bioset_free(bs);
 544 }
 545
 546 void mddev_init(struct mddev *mddev)
 547 {
 548         mutex_init(&mddev->open_mutex);
 549         mutex_init(&mddev->reconfig_mutex);
 550         mutex_init(&mddev->bitmap_info.mutex);
 551         INIT_LIST_HEAD(&mddev->disks);
 552         INIT_LIST_HEAD(&mddev->all_mddevs);
 553         init_timer(&mddev->safemode_timer);
 554         atomic_set(&mddev->active, 1);
 555         atomic_set(&mddev->openers, 0);
 556         atomic_set(&mddev->active_io, 0);
 557         spin_lock_init(&mddev->write_lock);
 558         atomic_set(&mddev->flush_pending, 0);
 559         init_waitqueue_head(&mddev->sb_wait);
 560         init_waitqueue_head(&mddev->recovery_wait);
 561         mddev->reshape_position = MaxSector;
 562         mddev->reshape_backwards = 0;
 563         mddev->resync_min = 0;
 564         mddev->resync_max = MaxSector;
 565         mddev->level = LEVEL_NONE;
 566 }
 567 EXPORT_SYMBOL_GPL(mddev_init);
 568
 569 static struct mddev * mddev_find(dev_t unit)
 570 {
 571         struct mddev *mddev, *new = NULL;
 572
 573         if (unit && MAJOR(unit) != MD_MAJOR)
 574                 unit &= ~((1<<MdpMinorShift)-1);
 575
 576  retry:
 577         spin_lock(&all_mddevs_lock);
 578
 579         if (unit) {
 580                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 581                         if (mddev->unit == unit) {
 582                                 mddev_get(mddev);
 583                                 spin_unlock(&all_mddevs_lock);
 584                                 kfree(new);
 585                                 return mddev;
 586                         }
 587
 588                 if (new) {
 589                         list_add(&new->all_mddevs, &all_mddevs);
 590                         spin_unlock(&all_mddevs_lock);
 591                         new->hold_active = UNTIL_IOCTL;
 592                         return new;
 593                 }
 594         } else if (new) {
 595                 /* find an unused unit number */
 596                 static int next_minor = 512;
 597                 int start = next_minor;
 598                 int is_free = 0;
 599                 int dev = 0;
 600                 while (!is_free) {
 601                         dev = MKDEV(MD_MAJOR, next_minor);
 602                         next_minor++;
 603                         if (next_minor > MINORMASK)
 604                                 next_minor = 0;
 605                         if (next_minor == start) {
 606                                 /* Oh dear, all in use. */
 607                                 spin_unlock(&all_mddevs_lock);
 608                                 kfree(new);
 609                                 return NULL;
 610                         }
 611
 612                         is_free = 1;
 613                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 614                                 if (mddev->unit == dev) {
 615                                         is_free = 0;
 616                                         break;
 617                                 }
 618                 }
 619                 new->unit = dev;
 620                 new->md_minor = MINOR(dev);
 621                 new->hold_active = UNTIL_STOP;
 622                 list_add(&new->all_mddevs, &all_mddevs);
 623                 spin_unlock(&all_mddevs_lock);
 624                 return new;
 625         }
 626         spin_unlock(&all_mddevs_lock);
 627
 628         new = kzalloc(sizeof(*new), GFP_KERNEL);
 629         if (!new)
 630                 return NULL;
 631
 632         new->unit = unit;
 633         if (MAJOR(unit) == MD_MAJOR)
 634                 new->md_minor = MINOR(unit);
 635         else
 636                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 637
 638         mddev_init(new);
 639
 640         goto retry;
 641 }
 642
 643 static inline int mddev_lock(struct mddev * mddev)
 644 {
 645         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 646 }
 647
 648 static inline int mddev_is_locked(struct mddev *mddev)
 649 {
 650         return mutex_is_locked(&mddev->reconfig_mutex);
 651 }
 652
 653 static inline int mddev_trylock(struct mddev * mddev)
 654 {
 655         return mutex_trylock(&mddev->reconfig_mutex);
 656 }
 657
 658 static struct attribute_group md_redundancy_group;
 659
 660 static void mddev_unlock(struct mddev * mddev)
 661 {
 662         if (mddev->to_remove) {
 663                 /* These cannot be removed under reconfig_mutex as
 664                  * an access to the files will try to take reconfig_mutex
 665                  * while holding the file unremovable, which leads to
 666                  * a deadlock.
 667                  * So hold set sysfs_active while the remove in happeing,
 668                  * and anything else which might set ->to_remove or my
 669                  * otherwise change the sysfs namespace will fail with
 670                  * -EBUSY if sysfs_active is still set.
 671                  * We set sysfs_active under reconfig_mutex and elsewhere
 672                  * test it under the same mutex to ensure its correct value
 673                  * is seen.
 674                  */
 675                 struct attribute_group *to_remove = mddev->to_remove;
 676                 mddev->to_remove = NULL;
 677                 mddev->sysfs_active = 1;
 678                 mutex_unlock(&mddev->reconfig_mutex);
 679
 680                 if (mddev->kobj.sd) {
 681                         if (to_remove != &md_redundancy_group)
 682                                 sysfs_remove_group(&mddev->kobj, to_remove);
 683                         if (mddev->pers == NULL ||
 684                             mddev->pers->sync_request == NULL) {
 685                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 686                                 if (mddev->sysfs_action)
 687                                         sysfs_put(mddev->sysfs_action);
 688                                 mddev->sysfs_action = NULL;
 689                         }
 690                 }
 691                 mddev->sysfs_active = 0;
 692         } else
 693                 mutex_unlock(&mddev->reconfig_mutex);
 694
 695         /* As we've dropped the mutex we need a spinlock to
 696          * make sure the thread doesn't disappear
 697          */
 698         spin_lock(&pers_lock);
 699         md_wakeup_thread(mddev->thread);
 700         spin_unlock(&pers_lock);
 701 }
 702
 703 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
 704 {
 705         struct md_rdev *rdev;
 706
 707         rdev_for_each(rdev, mddev)
 708                 if (rdev->desc_nr == nr)
 709                         return rdev;
 710
 711         return NULL;
 712 }
 713
 714 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
 715 {
 716         struct md_rdev *rdev;
 717
 718         rdev_for_each(rdev, mddev)
 719                 if (rdev->bdev->bd_dev == dev)
 720                         return rdev;
 721
 722         return NULL;
 723 }
 724
 725 static struct md_personality *find_pers(int level, char *clevel)
 726 {
 727         struct md_personality *pers;
 728         list_for_each_entry(pers, &pers_list, list) {
 729                 if (level != LEVEL_NONE && pers->level == level)
 730                         return pers;
 731                 if (strcmp(pers->name, clevel)==0)
 732                         return pers;
 733         }
 734         return NULL;
 735 }
 736
 737 /* return the offset of the super block in 512byte sectors */
 738 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 739 {
 740         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 741         return MD_NEW_SIZE_SECTORS(num_sectors);
 742 }
 743
 744 static int alloc_disk_sb(struct md_rdev * rdev)
 745 {
 746         if (rdev->sb_page)
 747                 MD_BUG();
 748
 749         rdev->sb_page = alloc_page(GFP_KERNEL);
 750         if (!rdev->sb_page) {
 751                 printk(KERN_ALERT "md: out of memory.\n");
 752                 return -ENOMEM;
 753         }
 754
 755         return 0;
 756 }
 757
 758 void md_rdev_clear(struct md_rdev *rdev)
 759 {
 760         if (rdev->sb_page) {
 761                 put_page(rdev->sb_page);
 762                 rdev->sb_loaded = 0;
 763                 rdev->sb_page = NULL;
 764                 rdev->sb_start = 0;
 765                 rdev->sectors = 0;
 766         }
 767         if (rdev->bb_page) {
 768                 put_page(rdev->bb_page);
 769                 rdev->bb_page = NULL;
 770         }
 771         kfree(rdev->badblocks.page);
 772         rdev->badblocks.page = NULL;
 773 }
 774 EXPORT_SYMBOL_GPL(md_rdev_clear);
 775
 776 static void super_written(struct bio *bio, int error)
 777 {
 778         struct md_rdev *rdev = bio->bi_private;
 779         struct mddev *mddev = rdev->mddev;
 780
 781         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 782                 printk("md: super_written gets error=%d, uptodate=%d\n",
 783                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 784                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 785                 md_error(mddev, rdev);
 786         }
 787
 788         if (atomic_dec_and_test(&mddev->pending_writes))
 789                 wake_up(&mddev->sb_wait);
 790         bio_put(bio);
 791 }
 792
 793 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 794                    sector_t sector, int size, struct page *page)
 795 {
 796         /* write first size bytes of page to sector of rdev
 797          * Increment mddev->pending_writes before returning
 798          * and decrement it on completion, waking up sb_wait
 799          * if zero is reached.
 800          * If an error occurred, call md_error
 801          */
 802         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 803
 804         bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
 805         bio->bi_sector = sector;
 806         bio_add_page(bio, page, size, 0);
 807         bio->bi_private = rdev;
 808         bio->bi_end_io = super_written;
 809
 810         atomic_inc(&mddev->pending_writes);
 811         submit_bio(WRITE_FLUSH_FUA, bio);
 812 }
 813
 814 void md_super_wait(struct mddev *mddev)
 815 {
 816         /* wait for all superblock writes that were scheduled to complete */
 817         DEFINE_WAIT(wq);
 818         for(;;) {
 819                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 820                 if (atomic_read(&mddev->pending_writes)==0)
 821                         break;
 822                 schedule();
 823         }
 824         finish_wait(&mddev->sb_wait, &wq);
 825 }
 826
 827 static void bi_complete(struct bio *bio, int error)
 828 {
 829         complete((struct completion*)bio->bi_private);
 830 }
 831
 832 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
 833                  struct page *page, int rw, bool metadata_op)
 834 {
 835         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 836         struct completion event;
 837         int ret;
 838
 839         rw |= REQ_SYNC;
 840
 841         bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
 842                 rdev->meta_bdev : rdev->bdev;
 843         if (metadata_op)
 844                 bio->bi_sector = sector + rdev->sb_start;
 845         else if (rdev->mddev->reshape_position != MaxSector &&
 846                  (rdev->mddev->reshape_backwards ==
 847                   (sector >= rdev->mddev->reshape_position)))
 848                 bio->bi_sector = sector + rdev->new_data_offset;
 849         else
 850                 bio->bi_sector = sector + rdev->data_offset;
 851         bio_add_page(bio, page, size, 0);
 852         init_completion(&event);
 853         bio->bi_private = &event;
 854         bio->bi_end_io = bi_complete;
 855         submit_bio(rw, bio);
 856         wait_for_completion(&event);
 857
 858         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 859         bio_put(bio);
 860         return ret;
 861 }
 862 EXPORT_SYMBOL_GPL(sync_page_io);
 863
 864 static int read_disk_sb(struct md_rdev * rdev, int size)
 865 {
 866         char b[BDEVNAME_SIZE];
 867         if (!rdev->sb_page) {
 868                 MD_BUG();
 869                 return -EINVAL;
 870         }
 871         if (rdev->sb_loaded)
 872                 return 0;
 873
 874
 875         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
 876                 goto fail;
 877         rdev->sb_loaded = 1;
 878         return 0;
 879
 880 fail:
 881         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 882                 bdevname(rdev->bdev,b));
 883         return -EINVAL;
 884 }
 885
 886 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 887 {
 888         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 889                 sb1->set_uuid1 == sb2->set_uuid1 &&
 890                 sb1->set_uuid2 == sb2->set_uuid2 &&
 891                 sb1->set_uuid3 == sb2->set_uuid3;
 892 }
 893
 894 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 895 {
 896         int ret;
 897         mdp_super_t *tmp1, *tmp2;
 898
 899         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 900         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 901
 902         if (!tmp1 || !tmp2) {
 903                 ret = 0;
 904                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 905                 goto abort;
 906         }
 907
 908         *tmp1 = *sb1;
 909         *tmp2 = *sb2;
 910
 911         /*
 912          * nr_disks is not constant
 913          */
 914         tmp1->nr_disks = 0;
 915         tmp2->nr_disks = 0;
 916
 917         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 918 abort:
 919         kfree(tmp1);
 920         kfree(tmp2);
 921         return ret;
 922 }
 923
 924
 925 static u32 md_csum_fold(u32 csum)
 926 {
 927         csum = (csum & 0xffff) + (csum >> 16);
 928         return (csum & 0xffff) + (csum >> 16);
 929 }
 930
 931 static unsigned int calc_sb_csum(mdp_super_t * sb)
 932 {
 933         u64 newcsum = 0;
 934         u32 *sb32 = (u32*)sb;
 935         int i;
 936         unsigned int disk_csum, csum;
 937
 938         disk_csum = sb->sb_csum;
 939         sb->sb_csum = 0;
 940
 941         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 942                 newcsum += sb32[i];
 943         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 944
 945
 946 #ifdef CONFIG_ALPHA
 947         /* This used to use csum_partial, which was wrong for several
 948          * reasons including that different results are returned on
 949          * different architectures.  It isn't critical that we get exactly
 950          * the same return value as before (we always csum_fold before
 951          * testing, and that removes any differences).  However as we
 952          * know that csum_partial always returned a 16bit value on
 953          * alphas, do a fold to maximise conformity to previous behaviour.
 954          */
 955         sb->sb_csum = md_csum_fold(disk_csum);
 956 #else
 957         sb->sb_csum = disk_csum;
 958 #endif
 959         return csum;
 960 }
 961
 962
 963 /*
 964  * Handle superblock details.
 965  * We want to be able to handle multiple superblock formats
 966  * so we have a common interface to them all, and an array of
 967  * different handlers.
 968  * We rely on user-space to write the initial superblock, and support
 969  * reading and updating of superblocks.
 970  * Interface methods are:
 971  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
 972  *      loads and validates a superblock on dev.
 973  *      if refdev != NULL, compare superblocks on both devices
 974  *    Return:
 975  *      0 - dev has a superblock that is compatible with refdev
 976  *      1 - dev has a superblock that is compatible and newer than refdev
 977  *          so dev should be used as the refdev in future
 978  *     -EINVAL superblock incompatible or invalid
 979  *     -othererror e.g. -EIO
 980  *
 981  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
 982  *      Verify that dev is acceptable into mddev.
 983  *       The first time, mddev->raid_disks will be 0, and data from
 984  *       dev should be merged in.  Subsequent calls check that dev
 985  *       is new enough.  Return 0 or -EINVAL
 986  *
 987  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
 988  *     Update the superblock for rdev with data in mddev
 989  *     This does not write to disc.
 990  *
 991  */
 992
 993 struct super_type  {
 994         char                *name;
 995         struct module       *owner;
 996         int                 (*load_super)(struct md_rdev *rdev,
 997                                           struct md_rdev *refdev,
 998                                           int minor_version);
 999         int                 (*validate_super)(struct mddev *mddev,
1000                                               struct md_rdev *rdev);
1001         void                (*sync_super)(struct mddev *mddev,
1002                                           struct md_rdev *rdev);
1003         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1004                                                 sector_t num_sectors);
1005         int                 (*allow_new_offset)(struct md_rdev *rdev,
1006                                                 unsigned long long new_offset);
1007 };
1008
1009 /*
1010  * Check that the given mddev has no bitmap.
1011  *
1012  * This function is called from the run method of all personalities that do not
1013  * support bitmaps. It prints an error message and returns non-zero if mddev
1014  * has a bitmap. Otherwise, it returns 0.
1015  *
1016  */
1017 int md_check_no_bitmap(struct mddev *mddev)
1018 {
1019         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1020                 return 0;
1021         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1022                 mdname(mddev), mddev->pers->name);
1023         return 1;
1024 }
1025 EXPORT_SYMBOL(md_check_no_bitmap);
1026
1027 /*
1028  * load_super for 0.90.0
1029  */
1030 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1031 {
1032         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1033         mdp_super_t *sb;
1034         int ret;
1035
1036         /*
1037          * Calculate the position of the superblock (512byte sectors),
1038          * it's at the end of the disk.
1039          *
1040          * It also happens to be a multiple of 4Kb.
1041          */
1042         rdev->sb_start = calc_dev_sboffset(rdev);
1043
1044         ret = read_disk_sb(rdev, MD_SB_BYTES);
1045         if (ret) return ret;
1046
1047         ret = -EINVAL;
1048
1049         bdevname(rdev->bdev, b);
1050         sb = page_address(rdev->sb_page);
1051
1052         if (sb->md_magic != MD_SB_MAGIC) {
1053                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1054                        b);
1055                 goto abort;
1056         }
1057
1058         if (sb->major_version != 0 ||
1059             sb->minor_version < 90 ||
1060             sb->minor_version > 91) {
1061                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1062                         sb->major_version, sb->minor_version,
1063                         b);
1064                 goto abort;
1065         }
1066
1067         if (sb->raid_disks <= 0)
1068                 goto abort;
1069
1070         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1071                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1072                         b);
1073                 goto abort;
1074         }
1075
1076         rdev->preferred_minor = sb->md_minor;
1077         rdev->data_offset = 0;
1078         rdev->new_data_offset = 0;
1079         rdev->sb_size = MD_SB_BYTES;
1080         rdev->badblocks.shift = -1;
1081
1082         if (sb->level == LEVEL_MULTIPATH)
1083                 rdev->desc_nr = -1;
1084         else
1085                 rdev->desc_nr = sb->this_disk.number;
1086
1087         if (!refdev) {
1088                 ret = 1;
1089         } else {
1090                 __u64 ev1, ev2;
1091                 mdp_super_t *refsb = page_address(refdev->sb_page);
1092                 if (!uuid_equal(refsb, sb)) {
1093                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
1094                                 b, bdevname(refdev->bdev,b2));
1095                         goto abort;
1096                 }
1097                 if (!sb_equal(refsb, sb)) {
1098                         printk(KERN_WARNING "md: %s has same UUID"
1099                                " but different superblock to %s\n",
1100                                b, bdevname(refdev->bdev, b2));
1101                         goto abort;
1102                 }
1103                 ev1 = md_event(sb);
1104                 ev2 = md_event(refsb);
1105                 if (ev1 > ev2)
1106                         ret = 1;
1107                 else
1108                         ret = 0;
1109         }
1110         rdev->sectors = rdev->sb_start;
1111         /* Limit to 4TB as metadata cannot record more than that */
1112         if (rdev->sectors >= (2ULL << 32))
1113                 rdev->sectors = (2ULL << 32) - 2;
1114
1115         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1116                 /* "this cannot possibly happen" ... */
1117                 ret = -EINVAL;
1118
1119  abort:
1120         return ret;
1121 }
1122
1123 /*
1124  * validate_super for 0.90.0
1125  */
1126 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1127 {
1128         mdp_disk_t *desc;
1129         mdp_super_t *sb = page_address(rdev->sb_page);
1130         __u64 ev1 = md_event(sb);
1131
1132         rdev->raid_disk = -1;
1133         clear_bit(Faulty, &rdev->flags);
1134         clear_bit(In_sync, &rdev->flags);
1135         clear_bit(WriteMostly, &rdev->flags);
1136
1137         if (mddev->raid_disks == 0) {
1138                 mddev->major_version = 0;
1139                 mddev->minor_version = sb->minor_version;
1140                 mddev->patch_version = sb->patch_version;
1141                 mddev->external = 0;
1142                 mddev->chunk_sectors = sb->chunk_size >> 9;
1143                 mddev->ctime = sb->ctime;
1144                 mddev->utime = sb->utime;
1145                 mddev->level = sb->level;
1146                 mddev->clevel[0] = 0;
1147                 mddev->layout = sb->layout;
1148                 mddev->raid_disks = sb->raid_disks;
1149                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1150                 mddev->events = ev1;
1151                 mddev->bitmap_info.offset = 0;
1152                 mddev->bitmap_info.space = 0;
1153                 /* bitmap can use 60 K after the 4K superblocks */
1154                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1155                 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1156                 mddev->reshape_backwards = 0;
1157
1158                 if (mddev->minor_version >= 91) {
1159                         mddev->reshape_position = sb->reshape_position;
1160                         mddev->delta_disks = sb->delta_disks;
1161                         mddev->new_level = sb->new_level;
1162                         mddev->new_layout = sb->new_layout;
1163                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1164                         if (mddev->delta_disks < 0)
1165                                 mddev->reshape_backwards = 1;
1166                 } else {
1167                         mddev->reshape_position = MaxSector;
1168                         mddev->delta_disks = 0;
1169                         mddev->new_level = mddev->level;
1170                         mddev->new_layout = mddev->layout;
1171                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1172                 }
1173
1174                 if (sb->state & (1<<MD_SB_CLEAN))
1175                         mddev->recovery_cp = MaxSector;
1176                 else {
1177                         if (sb->events_hi == sb->cp_events_hi &&
1178                                 sb->events_lo == sb->cp_events_lo) {
1179                                 mddev->recovery_cp = sb->recovery_cp;
1180                         } else
1181                                 mddev->recovery_cp = 0;
1182                 }
1183
1184                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1185                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1186                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1187                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1188
1189                 mddev->max_disks = MD_SB_DISKS;
1190
1191                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1192                     mddev->bitmap_info.file == NULL) {
1193                         mddev->bitmap_info.offset =
1194                                 mddev->bitmap_info.default_offset;
1195                         mddev->bitmap_info.space =
1196                                 mddev->bitmap_info.space;
1197                 }
1198
1199         } else if (mddev->pers == NULL) {
1200                 /* Insist on good event counter while assembling, except
1201                  * for spares (which don't need an event count) */
1202                 ++ev1;
1203                 if (sb->disks[rdev->desc_nr].state & (
1204                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1205                         if (ev1 < mddev->events)
1206                                 return -EINVAL;
1207         } else if (mddev->bitmap) {
1208                 /* if adding to array with a bitmap, then we can accept an
1209                  * older device ... but not too old.
1210                  */
1211                 if (ev1 < mddev->bitmap->events_cleared)
1212                         return 0;
1213         } else {
1214                 if (ev1 < mddev->events)
1215                         /* just a hot-add of a new device, leave raid_disk at -1 */
1216                         return 0;
1217         }
1218
1219         if (mddev->level != LEVEL_MULTIPATH) {
1220                 desc = sb->disks + rdev->desc_nr;
1221
1222                 if (desc->state & (1<<MD_DISK_FAULTY))
1223                         set_bit(Faulty, &rdev->flags);
1224                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1225                             desc->raid_disk < mddev->raid_disks */) {
1226                         set_bit(In_sync, &rdev->flags);
1227                         rdev->raid_disk = desc->raid_disk;
1228                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1229                         /* active but not in sync implies recovery up to
1230                          * reshape position.  We don't know exactly where
1231                          * that is, so set to zero for now */
1232                         if (mddev->minor_version >= 91) {
1233                                 rdev->recovery_offset = 0;
1234                                 rdev->raid_disk = desc->raid_disk;
1235                         }
1236                 }
1237                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1238                         set_bit(WriteMostly, &rdev->flags);
1239         } else /* MULTIPATH are always insync */
1240                 set_bit(In_sync, &rdev->flags);
1241         return 0;
1242 }
1243
1244 /*
1245  * sync_super for 0.90.0
1246  */
1247 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1248 {
1249         mdp_super_t *sb;
1250         struct md_rdev *rdev2;
1251         int next_spare = mddev->raid_disks;
1252
1253
1254         /* make rdev->sb match mddev data..
1255          *
1256          * 1/ zero out disks
1257          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1258          * 3/ any empty disks < next_spare become removed
1259          *
1260          * disks[0] gets initialised to REMOVED because
1261          * we cannot be sure from other fields if it has
1262          * been initialised or not.
1263          */
1264         int i;
1265         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1266
1267         rdev->sb_size = MD_SB_BYTES;
1268
1269         sb = page_address(rdev->sb_page);
1270
1271         memset(sb, 0, sizeof(*sb));
1272
1273         sb->md_magic = MD_SB_MAGIC;
1274         sb->major_version = mddev->major_version;
1275         sb->patch_version = mddev->patch_version;
1276         sb->gvalid_words  = 0; /* ignored */
1277         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1278         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1279         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1280         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1281
1282         sb->ctime = mddev->ctime;
1283         sb->level = mddev->level;
1284         sb->size = mddev->dev_sectors / 2;
1285         sb->raid_disks = mddev->raid_disks;
1286         sb->md_minor = mddev->md_minor;
1287         sb->not_persistent = 0;
1288         sb->utime = mddev->utime;
1289         sb->state = 0;
1290         sb->events_hi = (mddev->events>>32);
1291         sb->events_lo = (u32)mddev->events;
1292
1293         if (mddev->reshape_position == MaxSector)
1294                 sb->minor_version = 90;
1295         else {
1296                 sb->minor_version = 91;
1297                 sb->reshape_position = mddev->reshape_position;
1298                 sb->new_level = mddev->new_level;
1299                 sb->delta_disks = mddev->delta_disks;
1300                 sb->new_layout = mddev->new_layout;
1301                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1302         }
1303         mddev->minor_version = sb->minor_version;
1304         if (mddev->in_sync)
1305         {
1306                 sb->recovery_cp = mddev->recovery_cp;
1307                 sb->cp_events_hi = (mddev->events>>32);
1308                 sb->cp_events_lo = (u32)mddev->events;
1309                 if (mddev->recovery_cp == MaxSector)
1310                         sb->state = (1<< MD_SB_CLEAN);
1311         } else
1312                 sb->recovery_cp = 0;
1313
1314         sb->layout = mddev->layout;
1315         sb->chunk_size = mddev->chunk_sectors << 9;
1316
1317         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1318                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1319
1320         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1321         rdev_for_each(rdev2, mddev) {
1322                 mdp_disk_t *d;
1323                 int desc_nr;
1324                 int is_active = test_bit(In_sync, &rdev2->flags);
1325
1326                 if (rdev2->raid_disk >= 0 &&
1327                     sb->minor_version >= 91)
1328                         /* we have nowhere to store the recovery_offset,
1329                          * but if it is not below the reshape_position,
1330                          * we can piggy-back on that.
1331                          */
1332                         is_active = 1;
1333                 if (rdev2->raid_disk < 0 ||
1334                     test_bit(Faulty, &rdev2->flags))
1335                         is_active = 0;
1336                 if (is_active)
1337                         desc_nr = rdev2->raid_disk;
1338                 else
1339                         desc_nr = next_spare++;
1340                 rdev2->desc_nr = desc_nr;
1341                 d = &sb->disks[rdev2->desc_nr];
1342                 nr_disks++;
1343                 d->number = rdev2->desc_nr;
1344                 d->major = MAJOR(rdev2->bdev->bd_dev);
1345                 d->minor = MINOR(rdev2->bdev->bd_dev);
1346                 if (is_active)
1347                         d->raid_disk = rdev2->raid_disk;
1348                 else
1349                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1350                 if (test_bit(Faulty, &rdev2->flags))
1351                         d->state = (1<<MD_DISK_FAULTY);
1352                 else if (is_active) {
1353                         d->state = (1<<MD_DISK_ACTIVE);
1354                         if (test_bit(In_sync, &rdev2->flags))
1355                                 d->state |= (1<<MD_DISK_SYNC);
1356                         active++;
1357                         working++;
1358                 } else {
1359                         d->state = 0;
1360                         spare++;
1361                         working++;
1362                 }
1363                 if (test_bit(WriteMostly, &rdev2->flags))
1364                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1365         }
1366         /* now set the "removed" and "faulty" bits on any missing devices */
1367         for (i=0 ; i < mddev->raid_disks ; i++) {
1368                 mdp_disk_t *d = &sb->disks[i];
1369                 if (d->state == 0 && d->number == 0) {
1370                         d->number = i;
1371                         d->raid_disk = i;
1372                         d->state = (1<<MD_DISK_REMOVED);
1373                         d->state |= (1<<MD_DISK_FAULTY);
1374                         failed++;
1375                 }
1376         }
1377         sb->nr_disks = nr_disks;
1378         sb->active_disks = active;
1379         sb->working_disks = working;
1380         sb->failed_disks = failed;
1381         sb->spare_disks = spare;
1382
1383         sb->this_disk = sb->disks[rdev->desc_nr];
1384         sb->sb_csum = calc_sb_csum(sb);
1385 }
1386
1387 /*
1388  * rdev_size_change for 0.90.0
1389  */
1390 static unsigned long long
1391 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1392 {
1393         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1394                 return 0; /* component must fit device */
1395         if (rdev->mddev->bitmap_info.offset)
1396                 return 0; /* can't move bitmap */
1397         rdev->sb_start = calc_dev_sboffset(rdev);
1398         if (!num_sectors || num_sectors > rdev->sb_start)
1399                 num_sectors = rdev->sb_start;
1400         /* Limit to 4TB as metadata cannot record more than that.
1401          * 4TB == 2^32 KB, or 2*2^32 sectors.
1402          */
1403         if (num_sectors >= (2ULL << 32))
1404                 num_sectors = (2ULL << 32) - 2;
1405         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1406                        rdev->sb_page);
1407         md_super_wait(rdev->mddev);
1408         return num_sectors;
1409 }
1410
1411 static int
1412 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1413 {
1414         /* non-zero offset changes not possible with v0.90 */
1415         return new_offset == 0;
1416 }
1417
1418 /*
1419  * version 1 superblock
1420  */
1421
1422 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1423 {
1424         __le32 disk_csum;
1425         u32 csum;
1426         unsigned long long newcsum;
1427         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1428         __le32 *isuper = (__le32*)sb;
1429         int i;
1430
1431         disk_csum = sb->sb_csum;
1432         sb->sb_csum = 0;
1433         newcsum = 0;
1434         for (i=0; size>=4; size -= 4 )
1435                 newcsum += le32_to_cpu(*isuper++);
1436
1437         if (size == 2)
1438                 newcsum += le16_to_cpu(*(__le16*) isuper);
1439
1440         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1441         sb->sb_csum = disk_csum;
1442         return cpu_to_le32(csum);
1443 }
1444
1445 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1446                             int acknowledged);
1447 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1448 {
1449         struct mdp_superblock_1 *sb;
1450         int ret;
1451         sector_t sb_start;
1452         sector_t sectors;
1453         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1454         int bmask;
1455
1456         /*
1457          * Calculate the position of the superblock in 512byte sectors.
1458          * It is always aligned to a 4K boundary and
1459          * depeding on minor_version, it can be:
1460          * 0: At least 8K, but less than 12K, from end of device
1461          * 1: At start of device
1462          * 2: 4K from start of device.
1463          */
1464         switch(minor_version) {
1465         case 0:
1466                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1467                 sb_start -= 8*2;
1468                 sb_start &= ~(sector_t)(4*2-1);
1469                 break;
1470         case 1:
1471                 sb_start = 0;
1472                 break;
1473         case 2:
1474                 sb_start = 8;
1475                 break;
1476         default:
1477                 return -EINVAL;
1478         }
1479         rdev->sb_start = sb_start;
1480
1481         /* superblock is rarely larger than 1K, but it can be larger,
1482          * and it is safe to read 4k, so we do that
1483          */
1484         ret = read_disk_sb(rdev, 4096);
1485         if (ret) return ret;
1486
1487
1488         sb = page_address(rdev->sb_page);
1489
1490         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1491             sb->major_version != cpu_to_le32(1) ||
1492             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1493             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1494             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1495                 return -EINVAL;
1496
1497         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1498                 printk("md: invalid superblock checksum on %s\n",
1499                         bdevname(rdev->bdev,b));
1500                 return -EINVAL;
1501         }
1502         if (le64_to_cpu(sb->data_size) < 10) {
1503                 printk("md: data_size too small on %s\n",
1504                        bdevname(rdev->bdev,b));
1505                 return -EINVAL;
1506         }
1507         if (sb->pad0 ||
1508             sb->pad3[0] ||
1509             memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1510                 /* Some padding is non-zero, might be a new feature */
1511                 return -EINVAL;
1512
1513         rdev->preferred_minor = 0xffff;
1514         rdev->data_offset = le64_to_cpu(sb->data_offset);
1515         rdev->new_data_offset = rdev->data_offset;
1516         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1517             (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1518                 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1519         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1520
1521         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1522         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1523         if (rdev->sb_size & bmask)
1524                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1525
1526         if (minor_version
1527             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1528                 return -EINVAL;
1529         if (minor_version
1530             && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1531                 return -EINVAL;
1532
1533         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1534                 rdev->desc_nr = -1;
1535         else
1536                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1537
1538         if (!rdev->bb_page) {
1539                 rdev->bb_page = alloc_page(GFP_KERNEL);
1540                 if (!rdev->bb_page)
1541                         return -ENOMEM;
1542         }
1543         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1544             rdev->badblocks.count == 0) {
1545                 /* need to load the bad block list.
1546                  * Currently we limit it to one page.
1547                  */
1548                 s32 offset;
1549                 sector_t bb_sector;
1550                 u64 *bbp;
1551                 int i;
1552                 int sectors = le16_to_cpu(sb->bblog_size);
1553                 if (sectors > (PAGE_SIZE / 512))
1554                         return -EINVAL;
1555                 offset = le32_to_cpu(sb->bblog_offset);
1556                 if (offset == 0)
1557                         return -EINVAL;
1558                 bb_sector = (long long)offset;
1559                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1560                                   rdev->bb_page, READ, true))
1561                         return -EIO;
1562                 bbp = (u64 *)page_address(rdev->bb_page);
1563                 rdev->badblocks.shift = sb->bblog_shift;
1564                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1565                         u64 bb = le64_to_cpu(*bbp);
1566                         int count = bb & (0x3ff);
1567                         u64 sector = bb >> 10;
1568                         sector <<= sb->bblog_shift;
1569                         count <<= sb->bblog_shift;
1570                         if (bb + 1 == 0)
1571                                 break;
1572                         if (md_set_badblocks(&rdev->badblocks,
1573                                              sector, count, 1) == 0)
1574                                 return -EINVAL;
1575                 }
1576         } else if (sb->bblog_offset == 0)
1577                 rdev->badblocks.shift = -1;
1578
1579         if (!refdev) {
1580                 ret = 1;
1581         } else {
1582                 __u64 ev1, ev2;
1583                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1584
1585                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1586                     sb->level != refsb->level ||
1587                     sb->layout != refsb->layout ||
1588                     sb->chunksize != refsb->chunksize) {
1589                         printk(KERN_WARNING "md: %s has strangely different"
1590                                 " superblock to %s\n",
1591                                 bdevname(rdev->bdev,b),
1592                                 bdevname(refdev->bdev,b2));
1593                         return -EINVAL;
1594                 }
1595                 ev1 = le64_to_cpu(sb->events);
1596                 ev2 = le64_to_cpu(refsb->events);
1597
1598                 if (ev1 > ev2)
1599                         ret = 1;
1600                 else
1601                         ret = 0;
1602         }
1603         if (minor_version) {
1604                 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1605                 sectors -= rdev->data_offset;
1606         } else
1607                 sectors = rdev->sb_start;
1608         if (sectors < le64_to_cpu(sb->data_size))
1609                 return -EINVAL;
1610         rdev->sectors = le64_to_cpu(sb->data_size);
1611         return ret;
1612 }
1613
1614 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1615 {
1616         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1617         __u64 ev1 = le64_to_cpu(sb->events);
1618
1619         rdev->raid_disk = -1;
1620         clear_bit(Faulty, &rdev->flags);
1621         clear_bit(In_sync, &rdev->flags);
1622         clear_bit(WriteMostly, &rdev->flags);
1623
1624         if (mddev->raid_disks == 0) {
1625                 mddev->major_version = 1;
1626                 mddev->patch_version = 0;
1627                 mddev->external = 0;
1628                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1629                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1630                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1631                 mddev->level = le32_to_cpu(sb->level);
1632                 mddev->clevel[0] = 0;
1633                 mddev->layout = le32_to_cpu(sb->layout);
1634                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1635                 mddev->dev_sectors = le64_to_cpu(sb->size);
1636                 mddev->events = ev1;
1637                 mddev->bitmap_info.offset = 0;
1638                 mddev->bitmap_info.space = 0;
1639                 /* Default location for bitmap is 1K after superblock
1640                  * using 3K - total of 4K
1641                  */
1642                 mddev->bitmap_info.default_offset = 1024 >> 9;
1643                 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1644                 mddev->reshape_backwards = 0;
1645
1646                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1647                 memcpy(mddev->uuid, sb->set_uuid, 16);
1648
1649                 mddev->max_disks =  (4096-256)/2;
1650
1651                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1652                     mddev->bitmap_info.file == NULL) {
1653                         mddev->bitmap_info.offset =
1654                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1655                         /* Metadata doesn't record how much space is available.
1656                          * For 1.0, we assume we can use up to the superblock
1657                          * if before, else to 4K beyond superblock.
1658                          * For others, assume no change is possible.
1659                          */
1660                         if (mddev->minor_version > 0)
1661                                 mddev->bitmap_info.space = 0;
1662                         else if (mddev->bitmap_info.offset > 0)
1663                                 mddev->bitmap_info.space =
1664                                         8 - mddev->bitmap_info.offset;
1665                         else
1666                                 mddev->bitmap_info.space =
1667                                         -mddev->bitmap_info.offset;
1668                 }
1669
1670                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1671                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1672                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1673                         mddev->new_level = le32_to_cpu(sb->new_level);
1674                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1675                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1676                         if (mddev->delta_disks < 0 ||
1677                             (mddev->delta_disks == 0 &&
1678                              (le32_to_cpu(sb->feature_map)
1679                               & MD_FEATURE_RESHAPE_BACKWARDS)))
1680                                 mddev->reshape_backwards = 1;
1681                 } else {
1682                         mddev->reshape_position = MaxSector;
1683                         mddev->delta_disks = 0;
1684                         mddev->new_level = mddev->level;
1685                         mddev->new_layout = mddev->layout;
1686                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1687                 }
1688
1689         } else if (mddev->pers == NULL) {
1690                 /* Insist of good event counter while assembling, except for
1691                  * spares (which don't need an event count) */
1692                 ++ev1;
1693                 if (rdev->desc_nr >= 0 &&
1694                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1695                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1696                         if (ev1 < mddev->events)
1697                                 return -EINVAL;
1698         } else if (mddev->bitmap) {
1699                 /* If adding to array with a bitmap, then we can accept an
1700                  * older device, but not too old.
1701                  */
1702                 if (ev1 < mddev->bitmap->events_cleared)
1703                         return 0;
1704         } else {
1705                 if (ev1 < mddev->events)
1706                         /* just a hot-add of a new device, leave raid_disk at -1 */
1707                         return 0;
1708         }
1709         if (mddev->level != LEVEL_MULTIPATH) {
1710                 int role;
1711                 if (rdev->desc_nr < 0 ||
1712                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1713                         role = 0xffff;
1714                         rdev->desc_nr = -1;
1715                 } else
1716                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1717                 switch(role) {
1718                 case 0xffff: /* spare */
1719                         break;
1720                 case 0xfffe: /* faulty */
1721                         set_bit(Faulty, &rdev->flags);
1722                         break;
1723                 default:
1724                         if ((le32_to_cpu(sb->feature_map) &
1725                              MD_FEATURE_RECOVERY_OFFSET))
1726                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1727                         else
1728                                 set_bit(In_sync, &rdev->flags);
1729                         rdev->raid_disk = role;
1730                         break;
1731                 }
1732                 if (sb->devflags & WriteMostly1)
1733                         set_bit(WriteMostly, &rdev->flags);
1734                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1735                         set_bit(Replacement, &rdev->flags);
1736         } else /* MULTIPATH are always insync */
1737                 set_bit(In_sync, &rdev->flags);
1738
1739         return 0;
1740 }
1741
1742 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1743 {
1744         struct mdp_superblock_1 *sb;
1745         struct md_rdev *rdev2;
1746         int max_dev, i;
1747         /* make rdev->sb match mddev and rdev data. */
1748
1749         sb = page_address(rdev->sb_page);
1750
1751         sb->feature_map = 0;
1752         sb->pad0 = 0;
1753         sb->recovery_offset = cpu_to_le64(0);
1754         memset(sb->pad3, 0, sizeof(sb->pad3));
1755
1756         sb->utime = cpu_to_le64((__u64)mddev->utime);
1757         sb->events = cpu_to_le64(mddev->events);
1758         if (mddev->in_sync)
1759                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1760         else
1761                 sb->resync_offset = cpu_to_le64(0);
1762
1763         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1764
1765         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1766         sb->size = cpu_to_le64(mddev->dev_sectors);
1767         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1768         sb->level = cpu_to_le32(mddev->level);
1769         sb->layout = cpu_to_le32(mddev->layout);
1770
1771         if (test_bit(WriteMostly, &rdev->flags))
1772                 sb->devflags |= WriteMostly1;
1773         else
1774                 sb->devflags &= ~WriteMostly1;
1775         sb->data_offset = cpu_to_le64(rdev->data_offset);
1776         sb->data_size = cpu_to_le64(rdev->sectors);
1777
1778         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1779                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1780                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1781         }
1782
1783         if (rdev->raid_disk >= 0 &&
1784             !test_bit(In_sync, &rdev->flags)) {
1785                 sb->feature_map |=
1786                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1787                 sb->recovery_offset =
1788                         cpu_to_le64(rdev->recovery_offset);
1789         }
1790         if (test_bit(Replacement, &rdev->flags))
1791                 sb->feature_map |=
1792                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
1793
1794         if (mddev->reshape_position != MaxSector) {
1795                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1796                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1797                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1798                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1799                 sb->new_level = cpu_to_le32(mddev->new_level);
1800                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1801                 if (mddev->delta_disks == 0 &&
1802                     mddev->reshape_backwards)
1803                         sb->feature_map
1804                                 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1805                 if (rdev->new_data_offset != rdev->data_offset) {
1806                         sb->feature_map
1807                                 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1808                         sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1809                                                              - rdev->data_offset));
1810                 }
1811         }
1812
1813         if (rdev->badblocks.count == 0)
1814                 /* Nothing to do for bad blocks*/ ;
1815         else if (sb->bblog_offset == 0)
1816                 /* Cannot record bad blocks on this device */
1817                 md_error(mddev, rdev);
1818         else {
1819                 struct badblocks *bb = &rdev->badblocks;
1820                 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1821                 u64 *p = bb->page;
1822                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1823                 if (bb->changed) {
1824                         unsigned seq;
1825
1826 retry:
1827                         seq = read_seqbegin(&bb->lock);
1828
1829                         memset(bbp, 0xff, PAGE_SIZE);
1830
1831                         for (i = 0 ; i < bb->count ; i++) {
1832                                 u64 internal_bb = *p++;
1833                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1834                                                 | BB_LEN(internal_bb));
1835                                 *bbp++ = cpu_to_le64(store_bb);
1836                         }
1837                         bb->changed = 0;
1838                         if (read_seqretry(&bb->lock, seq))
1839                                 goto retry;
1840
1841                         bb->sector = (rdev->sb_start +
1842                                       (int)le32_to_cpu(sb->bblog_offset));
1843                         bb->size = le16_to_cpu(sb->bblog_size);
1844                 }
1845         }
1846
1847         max_dev = 0;
1848         rdev_for_each(rdev2, mddev)
1849                 if (rdev2->desc_nr+1 > max_dev)
1850                         max_dev = rdev2->desc_nr+1;
1851
1852         if (max_dev > le32_to_cpu(sb->max_dev)) {
1853                 int bmask;
1854                 sb->max_dev = cpu_to_le32(max_dev);
1855                 rdev->sb_size = max_dev * 2 + 256;
1856                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1857                 if (rdev->sb_size & bmask)
1858                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1859         } else
1860                 max_dev = le32_to_cpu(sb->max_dev);
1861
1862         for (i=0; i<max_dev;i++)
1863                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1864
1865         rdev_for_each(rdev2, mddev) {
1866                 i = rdev2->desc_nr;
1867                 if (test_bit(Faulty, &rdev2->flags))
1868                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1869                 else if (test_bit(In_sync, &rdev2->flags))
1870                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1871                 else if (rdev2->raid_disk >= 0)
1872                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1873                 else
1874                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1875         }
1876
1877         sb->sb_csum = calc_sb_1_csum(sb);
1878 }
1879
1880 static unsigned long long
1881 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1882 {
1883         struct mdp_superblock_1 *sb;
1884         sector_t max_sectors;
1885         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1886                 return 0; /* component must fit device */
1887         if (rdev->data_offset != rdev->new_data_offset)
1888                 return 0; /* too confusing */
1889         if (rdev->sb_start < rdev->data_offset) {
1890                 /* minor versions 1 and 2; superblock before data */
1891                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1892                 max_sectors -= rdev->data_offset;
1893                 if (!num_sectors || num_sectors > max_sectors)
1894                         num_sectors = max_sectors;
1895         } else if (rdev->mddev->bitmap_info.offset) {
1896                 /* minor version 0 with bitmap we can't move */
1897                 return 0;
1898         } else {
1899                 /* minor version 0; superblock after data */
1900                 sector_t sb_start;
1901                 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1902                 sb_start &= ~(sector_t)(4*2 - 1);
1903                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1904                 if (!num_sectors || num_sectors > max_sectors)
1905                         num_sectors = max_sectors;
1906                 rdev->sb_start = sb_start;
1907         }
1908         sb = page_address(rdev->sb_page);
1909         sb->data_size = cpu_to_le64(num_sectors);
1910         sb->super_offset = rdev->sb_start;
1911         sb->sb_csum = calc_sb_1_csum(sb);
1912         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1913                        rdev->sb_page);
1914         md_super_wait(rdev->mddev);
1915         return num_sectors;
1916
1917 }
1918
1919 static int
1920 super_1_allow_new_offset(struct md_rdev *rdev,
1921                          unsigned long long new_offset)
1922 {
1923         /* All necessary checks on new >= old have been done */
1924         struct bitmap *bitmap;
1925         if (new_offset >= rdev->data_offset)
1926                 return 1;
1927
1928         /* with 1.0 metadata, there is no metadata to tread on
1929          * so we can always move back */
1930         if (rdev->mddev->minor_version == 0)
1931                 return 1;
1932
1933         /* otherwise we must be sure not to step on
1934          * any metadata, so stay:
1935          * 36K beyond start of superblock
1936          * beyond end of badblocks
1937          * beyond write-intent bitmap
1938          */
1939         if (rdev->sb_start + (32+4)*2 > new_offset)
1940                 return 0;
1941         bitmap = rdev->mddev->bitmap;
1942         if (bitmap && !rdev->mddev->bitmap_info.file &&
1943             rdev->sb_start + rdev->mddev->bitmap_info.offset +
1944             bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1945                 return 0;
1946         if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1947                 return 0;
1948
1949         return 1;
1950 }
1951
1952 static struct super_type super_types[] = {
1953         [0] = {
1954                 .name   = "0.90.0",
1955                 .owner  = THIS_MODULE,
1956                 .load_super         = super_90_load,
1957                 .validate_super     = super_90_validate,
1958                 .sync_super         = super_90_sync,
1959                 .rdev_size_change   = super_90_rdev_size_change,
1960                 .allow_new_offset   = super_90_allow_new_offset,
1961         },
1962         [1] = {
1963                 .name   = "md-1",
1964                 .owner  = THIS_MODULE,
1965                 .load_super         = super_1_load,
1966                 .validate_super     = super_1_validate,
1967                 .sync_super         = super_1_sync,
1968                 .rdev_size_change   = super_1_rdev_size_change,
1969                 .allow_new_offset   = super_1_allow_new_offset,
1970         },
1971 };
1972
1973 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1974 {
1975         if (mddev->sync_super) {
1976                 mddev->sync_super(mddev, rdev);
1977                 return;
1978         }
1979
1980         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1981
1982         super_types[mddev->major_version].sync_super(mddev, rdev);
1983 }
1984
1985 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1986 {
1987         struct md_rdev *rdev, *rdev2;
1988
1989         rcu_read_lock();
1990         rdev_for_each_rcu(rdev, mddev1)
1991                 rdev_for_each_rcu(rdev2, mddev2)
1992                         if (rdev->bdev->bd_contains ==
1993                             rdev2->bdev->bd_contains) {
1994                                 rcu_read_unlock();
1995                                 return 1;
1996                         }
1997         rcu_read_unlock();
1998         return 0;
1999 }
2000
2001 static LIST_HEAD(pending_raid_disks);
2002
2003 /*
2004  * Try to register data integrity profile for an mddev
2005  *
2006  * This is called when an array is started and after a disk has been kicked
2007  * from the array. It only succeeds if all working and active component devices
2008  * are integrity capable with matching profiles.
2009  */
2010 int md_integrity_register(struct mddev *mddev)
2011 {
2012         struct md_rdev *rdev, *reference = NULL;
2013
2014         if (list_empty(&mddev->disks))
2015                 return 0; /* nothing to do */
2016         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2017                 return 0; /* shouldn't register, or already is */
2018         rdev_for_each(rdev, mddev) {
2019                 /* skip spares and non-functional disks */
2020                 if (test_bit(Faulty, &rdev->flags))
2021                         continue;
2022                 if (rdev->raid_disk < 0)
2023                         continue;
2024                 if (!reference) {
2025                         /* Use the first rdev as the reference */
2026                         reference = rdev;
2027                         continue;
2028                 }
2029                 /* does this rdev's profile match the reference profile? */
2030                 if (blk_integrity_compare(reference->bdev->bd_disk,
2031                                 rdev->bdev->bd_disk) < 0)
2032                         return -EINVAL;
2033         }
2034         if (!reference || !bdev_get_integrity(reference->bdev))
2035                 return 0;
2036         /*
2037          * All component devices are integrity capable and have matching
2038          * profiles, register the common profile for the md device.
2039          */
2040         if (blk_integrity_register(mddev->gendisk,
2041                         bdev_get_integrity(reference->bdev)) != 0) {
2042                 printk(KERN_ERR "md: failed to register integrity for %s\n",
2043                         mdname(mddev));
2044                 return -EINVAL;
2045         }
2046         printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2047         if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2048                 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2049                        mdname(mddev));
2050                 return -EINVAL;
2051         }
2052         return 0;
2053 }
2054 EXPORT_SYMBOL(md_integrity_register);
2055
2056 /* Disable data integrity if non-capable/non-matching disk is being added */
2057 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2058 {
2059         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
2060         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
2061
2062         if (!bi_mddev) /* nothing to do */
2063                 return;
2064         if (rdev->raid_disk < 0) /* skip spares */
2065                 return;
2066         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2067                                              rdev->bdev->bd_disk) >= 0)
2068                 return;
2069         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2070         blk_integrity_unregister(mddev->gendisk);
2071 }
2072 EXPORT_SYMBOL(md_integrity_add_rdev);
2073
2074 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2075 {
2076         char b[BDEVNAME_SIZE];
2077         struct kobject *ko;
2078         char *s;
2079         int err;
2080
2081         if (rdev->mddev) {
2082                 MD_BUG();
2083                 return -EINVAL;
2084         }
2085
2086         /* prevent duplicates */
2087         if (find_rdev(mddev, rdev->bdev->bd_dev))
2088                 return -EEXIST;
2089
2090         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2091         if (rdev->sectors && (mddev->dev_sectors == 0 ||
2092                         rdev->sectors < mddev->dev_sectors)) {
2093                 if (mddev->pers) {
2094                         /* Cannot change size, so fail
2095                          * If mddev->level <= 0, then we don't care
2096                          * about aligning sizes (e.g. linear)
2097                          */
2098                         if (mddev->level > 0)
2099                                 return -ENOSPC;
2100                 } else
2101                         mddev->dev_sectors = rdev->sectors;
2102         }
2103
2104         /* Verify rdev->desc_nr is unique.
2105          * If it is -1, assign a free number, else
2106          * check number is not in use
2107          */
2108         if (rdev->desc_nr < 0) {
2109                 int choice = 0;
2110                 if (mddev->pers) choice = mddev->raid_disks;
2111                 while (find_rdev_nr(mddev, choice))
2112                         choice++;
2113                 rdev->desc_nr = choice;
2114         } else {
2115                 if (find_rdev_nr(mddev, rdev->desc_nr))
2116                         return -EBUSY;
2117         }
2118         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2119                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2120                        mdname(mddev), mddev->max_disks);
2121                 return -EBUSY;
2122         }
2123         bdevname(rdev->bdev,b);
2124         while ( (s=strchr(b, '/')) != NULL)
2125                 *s = '!';
2126
2127         rdev->mddev = mddev;
2128         printk(KERN_INFO "md: bind<%s>\n", b);
2129
2130         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2131                 goto fail;
2132
2133         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2134         if (sysfs_create_link(&rdev->kobj, ko, "block"))
2135                 /* failure here is OK */;
2136         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2137
2138         list_add_rcu(&rdev->same_set, &mddev->disks);
2139         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2140
2141         /* May as well allow recovery to be retried once */
2142         mddev->recovery_disabled++;
2143
2144         return 0;
2145
2146  fail:
2147         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2148                b, mdname(mddev));
2149         return err;
2150 }
2151
2152 static void md_delayed_delete(struct work_struct *ws)
2153 {
2154         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2155         kobject_del(&rdev->kobj);
2156         kobject_put(&rdev->kobj);
2157 }
2158
2159 static void unbind_rdev_from_array(struct md_rdev * rdev)
2160 {
2161         char b[BDEVNAME_SIZE];
2162         if (!rdev->mddev) {
2163                 MD_BUG();
2164                 return;
2165         }
2166         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2167         list_del_rcu(&rdev->same_set);
2168         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2169         rdev->mddev = NULL;
2170         sysfs_remove_link(&rdev->kobj, "block");
2171         sysfs_put(rdev->sysfs_state);
2172         rdev->sysfs_state = NULL;
2173         rdev->badblocks.count = 0;
2174         /* We need to delay this, otherwise we can deadlock when
2175          * writing to 'remove' to "dev/state".  We also need
2176          * to delay it due to rcu usage.
2177          */
2178         synchronize_rcu();
2179         INIT_WORK(&rdev->del_work, md_delayed_delete);
2180         kobject_get(&rdev->kobj);
2181         queue_work(md_misc_wq, &rdev->del_work);
2182 }
2183
2184 /*
2185  * prevent the device from being mounted, repartitioned or
2186  * otherwise reused by a RAID array (or any other kernel
2187  * subsystem), by bd_claiming the device.
2188  */
2189 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2190 {
2191         int err = 0;
2192         struct block_device *bdev;
2193         char b[BDEVNAME_SIZE];
2194
2195         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2196                                  shared ? (struct md_rdev *)lock_rdev : rdev);
2197         if (IS_ERR(bdev)) {
2198                 printk(KERN_ERR "md: could not open %s.\n",
2199                         __bdevname(dev, b));
2200                 return PTR_ERR(bdev);
2201         }
2202         rdev->bdev = bdev;
2203         return err;
2204 }
2205
2206 static void unlock_rdev(struct md_rdev *rdev)
2207 {
2208         struct block_device *bdev = rdev->bdev;
2209         rdev->bdev = NULL;
2210         if (!bdev)
2211                 MD_BUG();
2212         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2213 }
2214
2215 void md_autodetect_dev(dev_t dev);
2216
2217 static void export_rdev(struct md_rdev * rdev)
2218 {
2219         char b[BDEVNAME_SIZE];
2220         printk(KERN_INFO "md: export_rdev(%s)\n",
2221                 bdevname(rdev->bdev,b));
2222         if (rdev->mddev)
2223                 MD_BUG();
2224         md_rdev_clear(rdev);
2225 #ifndef MODULE
2226         if (test_bit(AutoDetected, &rdev->flags))
2227                 md_autodetect_dev(rdev->bdev->bd_dev);
2228 #endif
2229         unlock_rdev(rdev);
2230         kobject_put(&rdev->kobj);
2231 }
2232
2233 static void kick_rdev_from_array(struct md_rdev * rdev)
2234 {
2235         unbind_rdev_from_array(rdev);
2236         export_rdev(rdev);
2237 }
2238
2239 static void export_array(struct mddev *mddev)
2240 {
2241         struct md_rdev *rdev, *tmp;
2242
2243         rdev_for_each_safe(rdev, tmp, mddev) {
2244                 if (!rdev->mddev) {
2245                         MD_BUG();
2246                         continue;
2247                 }
2248                 kick_rdev_from_array(rdev);
2249         }
2250         if (!list_empty(&mddev->disks))
2251                 MD_BUG();
2252         mddev->raid_disks = 0;
2253         mddev->major_version = 0;
2254 }
2255
2256 static void print_desc(mdp_disk_t *desc)
2257 {
2258         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2259                 desc->major,desc->minor,desc->raid_disk,desc->state);
2260 }
2261
2262 static void print_sb_90(mdp_super_t *sb)
2263 {
2264         int i;
2265
2266         printk(KERN_INFO
2267                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2268                 sb->major_version, sb->minor_version, sb->patch_version,
2269                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2270                 sb->ctime);
2271         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2272                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2273                 sb->md_minor, sb->layout, sb->chunk_size);
2274         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2275                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2276                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2277                 sb->failed_disks, sb->spare_disks,
2278                 sb->sb_csum, (unsigned long)sb->events_lo);
2279
2280         printk(KERN_INFO);
2281         for (i = 0; i < MD_SB_DISKS; i++) {
2282                 mdp_disk_t *desc;
2283
2284                 desc = sb->disks + i;
2285                 if (desc->number || desc->major || desc->minor ||
2286                     desc->raid_disk || (desc->state && (desc->state != 4))) {
2287                         printk("     D %2d: ", i);
2288                         print_desc(desc);
2289                 }
2290         }
2291         printk(KERN_INFO "md:     THIS: ");
2292         print_desc(&sb->this_disk);
2293 }
2294
2295 static void print_sb_1(struct mdp_superblock_1 *sb)
2296 {
2297         __u8 *uuid;
2298
2299         uuid = sb->set_uuid;
2300         printk(KERN_INFO
2301                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2302                "md:    Name: \"%s\" CT:%llu\n",
2303                 le32_to_cpu(sb->major_version),
2304                 le32_to_cpu(sb->feature_map),
2305                 uuid,
2306                 sb->set_name,
2307                 (unsigned long long)le64_to_cpu(sb->ctime)
2308                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2309
2310         uuid = sb->device_uuid;
2311         printk(KERN_INFO
2312                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2313                         " RO:%llu\n"
2314                "md:     Dev:%08x UUID: %pU\n"
2315                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2316                "md:         (MaxDev:%u) \n",
2317                 le32_to_cpu(sb->level),
2318                 (unsigned long long)le64_to_cpu(sb->size),
2319                 le32_to_cpu(sb->raid_disks),
2320                 le32_to_cpu(sb->layout),
2321                 le32_to_cpu(sb->chunksize),
2322                 (unsigned long long)le64_to_cpu(sb->data_offset),
2323                 (unsigned long long)le64_to_cpu(sb->data_size),
2324                 (unsigned long long)le64_to_cpu(sb->super_offset),
2325                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2326                 le32_to_cpu(sb->dev_number),
2327                 uuid,
2328                 sb->devflags,
2329                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2330                 (unsigned long long)le64_to_cpu(sb->events),
2331                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2332                 le32_to_cpu(sb->sb_csum),
2333                 le32_to_cpu(sb->max_dev)
2334                 );
2335 }
2336
2337 static void print_rdev(struct md_rdev *rdev, int major_version)
2338 {
2339         char b[BDEVNAME_SIZE];
2340         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2341                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2342                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2343                 rdev->desc_nr);
2344         if (rdev->sb_loaded) {
2345                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2346                 switch (major_version) {
2347                 case 0:
2348                         print_sb_90(page_address(rdev->sb_page));
2349                         break;
2350                 case 1:
2351                         print_sb_1(page_address(rdev->sb_page));
2352                         break;
2353                 }
2354         } else
2355                 printk(KERN_INFO "md: no rdev superblock!\n");
2356 }
2357
2358 static void md_print_devices(void)
2359 {
2360         struct list_head *tmp;
2361         struct md_rdev *rdev;
2362         struct mddev *mddev;
2363         char b[BDEVNAME_SIZE];
2364
2365         printk("\n");
2366         printk("md:     **********************************\n");
2367         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2368         printk("md:     **********************************\n");
2369         for_each_mddev(mddev, tmp) {
2370
2371                 if (mddev->bitmap)
2372                         bitmap_print_sb(mddev->bitmap);
2373                 else
2374                         printk("%s: ", mdname(mddev));
2375                 rdev_for_each(rdev, mddev)
2376                         printk("<%s>", bdevname(rdev->bdev,b));
2377                 printk("\n");
2378
2379                 rdev_for_each(rdev, mddev)
2380                         print_rdev(rdev, mddev->major_version);
2381         }
2382         printk("md:     **********************************\n");
2383         printk("\n");
2384 }
2385
2386
2387 static void sync_sbs(struct mddev * mddev, int nospares)
2388 {
2389         /* Update each superblock (in-memory image), but
2390          * if we are allowed to, skip spares which already
2391          * have the right event counter, or have one earlier
2392          * (which would mean they aren't being marked as dirty
2393          * with the rest of the array)
2394          */
2395         struct md_rdev *rdev;
2396         rdev_for_each(rdev, mddev) {
2397                 if (rdev->sb_events == mddev->events ||
2398                     (nospares &&
2399                      rdev->raid_disk < 0 &&
2400                      rdev->sb_events+1 == mddev->events)) {
2401                         /* Don't update this superblock */
2402                         rdev->sb_loaded = 2;
2403                 } else {
2404                         sync_super(mddev, rdev);
2405                         rdev->sb_loaded = 1;
2406                 }
2407         }
2408 }
2409
2410 static void md_update_sb(struct mddev * mddev, int force_change)
2411 {
2412         struct md_rdev *rdev;
2413         int sync_req;
2414         int nospares = 0;
2415         int any_badblocks_changed = 0;
2416
2417 repeat:
2418         /* First make sure individual recovery_offsets are correct */
2419         rdev_for_each(rdev, mddev) {
2420                 if (rdev->raid_disk >= 0 &&
2421                     mddev->delta_disks >= 0 &&
2422                     !test_bit(In_sync, &rdev->flags) &&
2423                     mddev->curr_resync_completed > rdev->recovery_offset)
2424                                 rdev->recovery_offset = mddev->curr_resync_completed;
2425
2426         }
2427         if (!mddev->persistent) {
2428                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2429                 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2430                 if (!mddev->external) {
2431                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2432                         rdev_for_each(rdev, mddev) {
2433                                 if (rdev->badblocks.changed) {
2434                                         rdev->badblocks.changed = 0;
2435                                         md_ack_all_badblocks(&rdev->badblocks);
2436                                         md_error(mddev, rdev);
2437                                 }
2438                                 clear_bit(Blocked, &rdev->flags);
2439                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2440                                 wake_up(&rdev->blocked_wait);
2441                         }
2442                 }
2443                 wake_up(&mddev->sb_wait);
2444                 return;
2445         }
2446
2447         spin_lock_irq(&mddev->write_lock);
2448
2449         mddev->utime = get_seconds();
2450
2451         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2452                 force_change = 1;
2453         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2454                 /* just a clean<-> dirty transition, possibly leave spares alone,
2455                  * though if events isn't the right even/odd, we will have to do
2456                  * spares after all
2457                  */
2458                 nospares = 1;
2459         if (force_change)
2460                 nospares = 0;
2461         if (mddev->degraded)
2462                 /* If the array is degraded, then skipping spares is both
2463                  * dangerous and fairly pointless.
2464                  * Dangerous because a device that was removed from the array
2465                  * might have a event_count that still looks up-to-date,
2466                  * so it can be re-added without a resync.
2467                  * Pointless because if there are any spares to skip,
2468                  * then a recovery will happen and soon that array won't
2469                  * be degraded any more and the spare can go back to sleep then.
2470                  */
2471                 nospares = 0;
2472
2473         sync_req = mddev->in_sync;
2474
2475         /* If this is just a dirty<->clean transition, and the array is clean
2476          * and 'events' is odd, we can roll back to the previous clean state */
2477         if (nospares
2478             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2479             && mddev->can_decrease_events
2480             && mddev->events != 1) {
2481                 mddev->events--;
2482                 mddev->can_decrease_events = 0;
2483         } else {
2484                 /* otherwise we have to go forward and ... */
2485                 mddev->events ++;
2486                 mddev->can_decrease_events = nospares;
2487         }
2488
2489         if (!mddev->events) {
2490                 /*
2491                  * oops, this 64-bit counter should never wrap.
2492                  * Either we are in around ~1 trillion A.C., assuming
2493                  * 1 reboot per second, or we have a bug:
2494                  */
2495                 MD_BUG();
2496                 mddev->events --;
2497         }
2498
2499         rdev_for_each(rdev, mddev) {
2500                 if (rdev->badblocks.changed)
2501                         any_badblocks_changed++;
2502                 if (test_bit(Faulty, &rdev->flags))
2503                         set_bit(FaultRecorded, &rdev->flags);
2504         }
2505
2506         sync_sbs(mddev, nospares);
2507         spin_unlock_irq(&mddev->write_lock);
2508
2509         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2510                  mdname(mddev), mddev->in_sync);
2511
2512         bitmap_update_sb(mddev->bitmap);
2513         rdev_for_each(rdev, mddev) {
2514                 char b[BDEVNAME_SIZE];
2515
2516                 if (rdev->sb_loaded != 1)
2517                         continue; /* no noise on spare devices */
2518
2519                 if (!test_bit(Faulty, &rdev->flags) &&
2520                     rdev->saved_raid_disk == -1) {
2521                         md_super_write(mddev,rdev,
2522                                        rdev->sb_start, rdev->sb_size,
2523                                        rdev->sb_page);
2524                         pr_debug("md: (write) %s's sb offset: %llu\n",
2525                                  bdevname(rdev->bdev, b),
2526                                  (unsigned long long)rdev->sb_start);
2527                         rdev->sb_events = mddev->events;
2528                         if (rdev->badblocks.size) {
2529                                 md_super_write(mddev, rdev,
2530                                                rdev->badblocks.sector,
2531                                                rdev->badblocks.size << 9,
2532                                                rdev->bb_page);
2533                                 rdev->badblocks.size = 0;
2534                         }
2535
2536                 } else if (test_bit(Faulty, &rdev->flags))
2537                         pr_debug("md: %s (skipping faulty)\n",
2538                                  bdevname(rdev->bdev, b));
2539                 else
2540                         pr_debug("(skipping incremental s/r ");
2541
2542                 if (mddev->level == LEVEL_MULTIPATH)
2543                         /* only need to write one superblock... */
2544                         break;
2545         }
2546         md_super_wait(mddev);
2547         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2548
2549         spin_lock_irq(&mddev->write_lock);
2550         if (mddev->in_sync != sync_req ||
2551             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2552                 /* have to write it out again */
2553                 spin_unlock_irq(&mddev->write_lock);
2554                 goto repeat;
2555         }
2556         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2557         spin_unlock_irq(&mddev->write_lock);
2558         wake_up(&mddev->sb_wait);
2559         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2560                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2561
2562         rdev_for_each(rdev, mddev) {
2563                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2564                         clear_bit(Blocked, &rdev->flags);
2565
2566                 if (any_badblocks_changed)
2567                         md_ack_all_badblocks(&rdev->badblocks);
2568                 clear_bit(BlockedBadBlocks, &rdev->flags);
2569                 wake_up(&rdev->blocked_wait);
2570         }
2571 }
2572
2573 /* words written to sysfs files may, or may not, be \n terminated.
2574  * We want to accept with case. For this we use cmd_match.
2575  */
2576 static int cmd_match(const char *cmd, const char *str)
2577 {
2578         /* See if cmd, written into a sysfs file, matches
2579          * str.  They must either be the same, or cmd can
2580          * have a trailing newline
2581          */
2582         while (*cmd && *str && *cmd == *str) {
2583                 cmd++;
2584                 str++;
2585         }
2586         if (*cmd == '\n')
2587                 cmd++;
2588         if (*str || *cmd)
2589                 return 0;
2590         return 1;
2591 }
2592
2593 struct rdev_sysfs_entry {
2594         struct attribute attr;
2595         ssize_t (*show)(struct md_rdev *, char *);
2596         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2597 };
2598
2599 static ssize_t
2600 state_show(struct md_rdev *rdev, char *page)
2601 {
2602         char *sep = "";
2603         size_t len = 0;
2604
2605         if (test_bit(Faulty, &rdev->flags) ||
2606             rdev->badblocks.unacked_exist) {
2607                 len+= sprintf(page+len, "%sfaulty",sep);
2608                 sep = ",";
2609         }
2610         if (test_bit(In_sync, &rdev->flags)) {
2611                 len += sprintf(page+len, "%sin_sync",sep);
2612                 sep = ",";
2613         }
2614         if (test_bit(WriteMostly, &rdev->flags)) {
2615                 len += sprintf(page+len, "%swrite_mostly",sep);
2616                 sep = ",";
2617         }
2618         if (test_bit(Blocked, &rdev->flags) ||
2619             (rdev->badblocks.unacked_exist
2620              && !test_bit(Faulty, &rdev->flags))) {
2621                 len += sprintf(page+len, "%sblocked", sep);
2622                 sep = ",";
2623         }
2624         if (!test_bit(Faulty, &rdev->flags) &&
2625             !test_bit(In_sync, &rdev->flags)) {
2626                 len += sprintf(page+len, "%sspare", sep);
2627                 sep = ",";
2628         }
2629         if (test_bit(WriteErrorSeen, &rdev->flags)) {
2630                 len += sprintf(page+len, "%swrite_error", sep);
2631                 sep = ",";
2632         }
2633         if (test_bit(WantReplacement, &rdev->flags)) {
2634                 len += sprintf(page+len, "%swant_replacement", sep);
2635                 sep = ",";
2636         }
2637         if (test_bit(Replacement, &rdev->flags)) {
2638                 len += sprintf(page+len, "%sreplacement", sep);
2639                 sep = ",";
2640         }
2641
2642         return len+sprintf(page+len, "\n");
2643 }
2644
2645 static ssize_t
2646 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2647 {
2648         /* can write
2649          *  faulty  - simulates an error
2650          *  remove  - disconnects the device
2651          *  writemostly - sets write_mostly
2652          *  -writemostly - clears write_mostly
2653          *  blocked - sets the Blocked flags
2654          *  -blocked - clears the Blocked and possibly simulates an error
2655          *  insync - sets Insync providing device isn't active
2656          *  write_error - sets WriteErrorSeen
2657          *  -write_error - clears WriteErrorSeen
2658          */
2659         int err = -EINVAL;
2660         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2661                 md_error(rdev->mddev, rdev);
2662                 if (test_bit(Faulty, &rdev->flags))
2663                         err = 0;
2664                 else
2665                         err = -EBUSY;
2666         } else if (cmd_match(buf, "remove")) {
2667                 if (rdev->raid_disk >= 0)
2668                         err = -EBUSY;
2669                 else {
2670                         struct mddev *mddev = rdev->mddev;
2671                         kick_rdev_from_array(rdev);
2672                         if (mddev->pers)
2673                                 md_update_sb(mddev, 1);
2674                         md_new_event(mddev);
2675                         err = 0;
2676                 }
2677         } else if (cmd_match(buf, "writemostly")) {
2678                 set_bit(WriteMostly, &rdev->flags);
2679                 err = 0;
2680         } else if (cmd_match(buf, "-writemostly")) {
2681                 clear_bit(WriteMostly, &rdev->flags);
2682                 err = 0;
2683         } else if (cmd_match(buf, "blocked")) {
2684                 set_bit(Blocked, &rdev->flags);
2685                 err = 0;
2686         } else if (cmd_match(buf, "-blocked")) {
2687                 if (!test_bit(Faulty, &rdev->flags) &&
2688                     rdev->badblocks.unacked_exist) {
2689                         /* metadata handler doesn't understand badblocks,
2690                          * so we need to fail the device
2691                          */
2692                         md_error(rdev->mddev, rdev);
2693                 }
2694                 clear_bit(Blocked, &rdev->flags);
2695                 clear_bit(BlockedBadBlocks, &rdev->flags);
2696                 wake_up(&rdev->blocked_wait);
2697                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2698                 md_wakeup_thread(rdev->mddev->thread);
2699
2700                 err = 0;
2701         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2702                 set_bit(In_sync, &rdev->flags);
2703                 err = 0;
2704         } else if (cmd_match(buf, "write_error")) {
2705                 set_bit(WriteErrorSeen, &rdev->flags);
2706                 err = 0;
2707         } else if (cmd_match(buf, "-write_error")) {
2708                 clear_bit(WriteErrorSeen, &rdev->flags);
2709                 err = 0;
2710         } else if (cmd_match(buf, "want_replacement")) {
2711                 /* Any non-spare device that is not a replacement can
2712                  * become want_replacement at any time, but we then need to
2713                  * check if recovery is needed.
2714                  */
2715                 if (rdev->raid_disk >= 0 &&
2716                     !test_bit(Replacement, &rdev->flags))
2717                         set_bit(WantReplacement, &rdev->flags);
2718                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2719                 md_wakeup_thread(rdev->mddev->thread);
2720                 err = 0;
2721         } else if (cmd_match(buf, "-want_replacement")) {
2722                 /* Clearing 'want_replacement' is always allowed.
2723                  * Once replacements starts it is too late though.
2724                  */
2725                 err = 0;
2726                 clear_bit(WantReplacement, &rdev->flags);
2727         } else if (cmd_match(buf, "replacement")) {
2728                 /* Can only set a device as a replacement when array has not
2729                  * yet been started.  Once running, replacement is automatic
2730                  * from spares, or by assigning 'slot'.
2731                  */
2732                 if (rdev->mddev->pers)
2733                         err = -EBUSY;
2734                 else {
2735                         set_bit(Replacement, &rdev->flags);
2736                         err = 0;
2737                 }
2738         } else if (cmd_match(buf, "-replacement")) {
2739                 /* Similarly, can only clear Replacement before start */
2740                 if (rdev->mddev->pers)
2741                         err = -EBUSY;
2742                 else {
2743                         clear_bit(Replacement, &rdev->flags);
2744                         err = 0;
2745                 }
2746         }
2747         if (!err)
2748                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2749         return err ? err : len;
2750 }
2751 static struct rdev_sysfs_entry rdev_state =
2752 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2753
2754 static ssize_t
2755 errors_show(struct md_rdev *rdev, char *page)
2756 {
2757         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2758 }
2759
2760 static ssize_t
2761 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2762 {
2763         char *e;
2764         unsigned long n = simple_strtoul(buf, &e, 10);
2765         if (*buf && (*e == 0 || *e == '\n')) {
2766                 atomic_set(&rdev->corrected_errors, n);
2767                 return len;
2768         }
2769         return -EINVAL;
2770 }
2771 static struct rdev_sysfs_entry rdev_errors =
2772 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2773
2774 static ssize_t
2775 slot_show(struct md_rdev *rdev, char *page)
2776 {
2777         if (rdev->raid_disk < 0)
2778                 return sprintf(page, "none\n");
2779         else
2780                 return sprintf(page, "%d\n", rdev->raid_disk);
2781 }
2782
2783 static ssize_t
2784 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2785 {
2786         char *e;
2787         int err;
2788         int slot = simple_strtoul(buf, &e, 10);
2789         if (strncmp(buf, "none", 4)==0)
2790                 slot = -1;
2791         else if (e==buf || (*e && *e!= '\n'))
2792                 return -EINVAL;
2793         if (rdev->mddev->pers && slot == -1) {
2794                 /* Setting 'slot' on an active array requires also
2795                  * updating the 'rd%d' link, and communicating
2796                  * with the personality with ->hot_*_disk.
2797                  * For now we only support removing
2798                  * failed/spare devices.  This normally happens automatically,
2799                  * but not when the metadata is externally managed.
2800                  */
2801                 if (rdev->raid_disk == -1)
2802                         return -EEXIST;
2803                 /* personality does all needed checks */
2804                 if (rdev->mddev->pers->hot_remove_disk == NULL)
2805                         return -EINVAL;
2806                 err = rdev->mddev->pers->
2807                         hot_remove_disk(rdev->mddev, rdev);
2808                 if (err)
2809                         return err;
2810                 sysfs_unlink_rdev(rdev->mddev, rdev);
2811                 rdev->raid_disk = -1;
2812                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2813                 md_wakeup_thread(rdev->mddev->thread);
2814         } else if (rdev->mddev->pers) {
2815                 /* Activating a spare .. or possibly reactivating
2816                  * if we ever get bitmaps working here.
2817                  */
2818
2819                 if (rdev->raid_disk != -1)
2820                         return -EBUSY;
2821
2822                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2823                         return -EBUSY;
2824
2825                 if (rdev->mddev->pers->hot_add_disk == NULL)
2826                         return -EINVAL;
2827
2828                 if (slot >= rdev->mddev->raid_disks &&
2829                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2830                         return -ENOSPC;
2831
2832                 rdev->raid_disk = slot;
2833                 if (test_bit(In_sync, &rdev->flags))
2834                         rdev->saved_raid_disk = slot;
2835                 else
2836                         rdev->saved_raid_disk = -1;
2837                 clear_bit(In_sync, &rdev->flags);
2838                 err = rdev->mddev->pers->
2839                         hot_add_disk(rdev->mddev, rdev);
2840                 if (err) {
2841                         rdev->raid_disk = -1;
2842                         return err;
2843                 } else
2844                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2845                 if (sysfs_link_rdev(rdev->mddev, rdev))
2846                         /* failure here is OK */;
2847                 /* don't wakeup anyone, leave that to userspace. */
2848         } else {
2849                 if (slot >= rdev->mddev->raid_disks &&
2850                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2851                         return -ENOSPC;
2852                 rdev->raid_disk = slot;
2853                 /* assume it is working */
2854                 clear_bit(Faulty, &rdev->flags);
2855                 clear_bit(WriteMostly, &rdev->flags);
2856                 set_bit(In_sync, &rdev->flags);
2857                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2858         }
2859         return len;
2860 }
2861
2862
2863 static struct rdev_sysfs_entry rdev_slot =
2864 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2865
2866 static ssize_t
2867 offset_show(struct md_rdev *rdev, char *page)
2868 {
2869         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2870 }
2871
2872 static ssize_t
2873 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2874 {
2875         unsigned long long offset;
2876         if (strict_strtoull(buf, 10, &offset) < 0)
2877                 return -EINVAL;
2878         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2879                 return -EBUSY;
2880         if (rdev->sectors && rdev->mddev->external)
2881                 /* Must set offset before size, so overlap checks
2882                  * can be sane */
2883                 return -EBUSY;
2884         rdev->data_offset = offset;
2885         rdev->new_data_offset = offset;
2886         return len;
2887 }
2888
2889 static struct rdev_sysfs_entry rdev_offset =
2890 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2891
2892 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2893 {
2894         return sprintf(page, "%llu\n",
2895                        (unsigned long long)rdev->new_data_offset);
2896 }
2897
2898 static ssize_t new_offset_store(struct md_rdev *rdev,
2899                                 const char *buf, size_t len)
2900 {
2901         unsigned long long new_offset;
2902         struct mddev *mddev = rdev->mddev;
2903
2904         if (strict_strtoull(buf, 10, &new_offset) < 0)
2905                 return -EINVAL;
2906
2907         if (mddev->sync_thread)
2908                 return -EBUSY;
2909         if (new_offset == rdev->data_offset)
2910                 /* reset is always permitted */
2911                 ;
2912         else if (new_offset > rdev->data_offset) {
2913                 /* must not push array size beyond rdev_sectors */
2914                 if (new_offset - rdev->data_offset
2915                     + mddev->dev_sectors > rdev->sectors)
2916                                 return -E2BIG;
2917         }
2918         /* Metadata worries about other space details. */
2919
2920         /* decreasing the offset is inconsistent with a backwards
2921          * reshape.
2922          */
2923         if (new_offset < rdev->data_offset &&
2924             mddev->reshape_backwards)
2925                 return -EINVAL;
2926         /* Increasing offset is inconsistent with forwards
2927          * reshape.  reshape_direction should be set to
2928          * 'backwards' first.
2929          */
2930         if (new_offset > rdev->data_offset &&
2931             !mddev->reshape_backwards)
2932                 return -EINVAL;
2933
2934         if (mddev->pers && mddev->persistent &&
2935             !super_types[mddev->major_version]
2936             .allow_new_offset(rdev, new_offset))
2937                 return -E2BIG;
2938         rdev->new_data_offset = new_offset;
2939         if (new_offset > rdev->data_offset)
2940                 mddev->reshape_backwards = 1;
2941         else if (new_offset < rdev->data_offset)
2942                 mddev->reshape_backwards = 0;
2943
2944         return len;
2945 }
2946 static struct rdev_sysfs_entry rdev_new_offset =
2947 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2948
2949 static ssize_t
2950 rdev_size_show(struct md_rdev *rdev, char *page)
2951 {
2952         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2953 }
2954
2955 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2956 {
2957         /* check if two start/length pairs overlap */
2958         if (s1+l1 <= s2)
2959                 return 0;
2960         if (s2+l2 <= s1)
2961                 return 0;
2962         return 1;
2963 }
2964
2965 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2966 {
2967         unsigned long long blocks;
2968         sector_t new;
2969
2970         if (strict_strtoull(buf, 10, &blocks) < 0)
2971                 return -EINVAL;
2972
2973         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2974                 return -EINVAL; /* sector conversion overflow */
2975
2976         new = blocks * 2;
2977         if (new != blocks * 2)
2978                 return -EINVAL; /* unsigned long long to sector_t overflow */
2979
2980         *sectors = new;
2981         return 0;
2982 }
2983
2984 static ssize_t
2985 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2986 {
2987         struct mddev *my_mddev = rdev->mddev;
2988         sector_t oldsectors = rdev->sectors;
2989         sector_t sectors;
2990
2991         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2992                 return -EINVAL;
2993         if (rdev->data_offset != rdev->new_data_offset)
2994                 return -EINVAL; /* too confusing */
2995         if (my_mddev->pers && rdev->raid_disk >= 0) {
2996                 if (my_mddev->persistent) {
2997                         sectors = super_types[my_mddev->major_version].
2998                                 rdev_size_change(rdev, sectors);
2999                         if (!sectors)
3000                                 return -EBUSY;
3001                 } else if (!sectors)
3002                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3003                                 rdev->data_offset;
3004         }
3005         if (sectors < my_mddev->dev_sectors)
3006                 return -EINVAL; /* component must fit device */
3007
3008         rdev->sectors = sectors;
3009         if (sectors > oldsectors && my_mddev->external) {
3010                 /* need to check that all other rdevs with the same ->bdev
3011                  * do not overlap.  We need to unlock the mddev to avoid
3012                  * a deadlock.  We have already changed rdev->sectors, and if
3013                  * we have to change it back, we will have the lock again.
3014                  */
3015                 struct mddev *mddev;
3016                 int overlap = 0;
3017                 struct list_head *tmp;
3018
3019                 mddev_unlock(my_mddev);
3020                 for_each_mddev(mddev, tmp) {
3021                         struct md_rdev *rdev2;
3022
3023                         mddev_lock(mddev);
3024                         rdev_for_each(rdev2, mddev)
3025                                 if (rdev->bdev == rdev2->bdev &&
3026                                     rdev != rdev2 &&
3027                                     overlaps(rdev->data_offset, rdev->sectors,
3028                                              rdev2->data_offset,
3029                                              rdev2->sectors)) {
3030                                         overlap = 1;
3031                                         break;
3032                                 }
3033                         mddev_unlock(mddev);
3034                         if (overlap) {
3035                                 mddev_put(mddev);
3036                                 break;
3037                         }
3038                 }
3039                 mddev_lock(my_mddev);
3040                 if (overlap) {
3041                         /* Someone else could have slipped in a size
3042                          * change here, but doing so is just silly.
3043                          * We put oldsectors back because we *know* it is
3044                          * safe, and trust userspace not to race with
3045                          * itself
3046                          */
3047                         rdev->sectors = oldsectors;
3048                         return -EBUSY;
3049                 }
3050         }
3051         return len;
3052 }
3053
3054 static struct rdev_sysfs_entry rdev_size =
3055 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3056
3057
3058 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3059 {
3060         unsigned long long recovery_start = rdev->recovery_offset;
3061
3062         if (test_bit(In_sync, &rdev->flags) ||
3063             recovery_start == MaxSector)
3064                 return sprintf(page, "none\n");
3065
3066         return sprintf(page, "%llu\n", recovery_start);
3067 }
3068
3069 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3070 {
3071         unsigned long long recovery_start;
3072
3073         if (cmd_match(buf, "none"))
3074                 recovery_start = MaxSector;
3075         else if (strict_strtoull(buf, 10, &recovery_start))
3076                 return -EINVAL;
3077
3078         if (rdev->mddev->pers &&
3079             rdev->raid_disk >= 0)
3080                 return -EBUSY;
3081
3082         rdev->recovery_offset = recovery_start;
3083         if (recovery_start == MaxSector)
3084                 set_bit(In_sync, &rdev->flags);
3085         else
3086                 clear_bit(In_sync, &rdev->flags);
3087         return len;
3088 }
3089
3090 static struct rdev_sysfs_entry rdev_recovery_start =
3091 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3092
3093
3094 static ssize_t
3095 badblocks_show(struct badblocks *bb, char *page, int unack);
3096 static ssize_t
3097 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3098
3099 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3100 {
3101         return badblocks_show(&rdev->badblocks, page, 0);
3102 }
3103 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3104 {
3105         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3106         /* Maybe that ack was all we needed */
3107         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3108                 wake_up(&rdev->blocked_wait);
3109         return rv;
3110 }
3111 static struct rdev_sysfs_entry rdev_bad_blocks =
3112 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3113
3114
3115 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3116 {
3117         return badblocks_show(&rdev->badblocks, page, 1);
3118 }
3119 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3120 {
3121         return badblocks_store(&rdev->badblocks, page, len, 1);
3122 }
3123 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3124 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3125
3126 static struct attribute *rdev_default_attrs[] = {
3127         &rdev_state.attr,
3128         &rdev_errors.attr,
3129         &rdev_slot.attr,
3130         &rdev_offset.attr,
3131         &rdev_new_offset.attr,
3132         &rdev_size.attr,
3133         &rdev_recovery_start.attr,
3134         &rdev_bad_blocks.attr,
3135         &rdev_unack_bad_blocks.attr,
3136         NULL,
3137 };
3138 static ssize_t
3139 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3140 {
3141         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3142         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3143         struct mddev *mddev = rdev->mddev;
3144         ssize_t rv;
3145
3146         if (!entry->show)
3147                 return -EIO;
3148
3149         rv = mddev ? mddev_lock(mddev) : -EBUSY;
3150         if (!rv) {
3151                 if (rdev->mddev == NULL)
3152                         rv = -EBUSY;
3153                 else
3154                         rv = entry->show(rdev, page);
3155                 mddev_unlock(mddev);
3156         }
3157         return rv;
3158 }
3159
3160 static ssize_t
3161 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3162               const char *page, size_t length)
3163 {
3164         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3165         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3166         ssize_t rv;
3167         struct mddev *mddev = rdev->mddev;
3168
3169         if (!entry->store)
3170                 return -EIO;
3171         if (!capable(CAP_SYS_ADMIN))
3172                 return -EACCES;
3173         rv = mddev ? mddev_lock(mddev): -EBUSY;
3174         if (!rv) {
3175                 if (rdev->mddev == NULL)
3176                         rv = -EBUSY;
3177                 else
3178                         rv = entry->store(rdev, page, length);
3179                 mddev_unlock(mddev);
3180         }
3181         return rv;
3182 }
3183
3184 static void rdev_free(struct kobject *ko)
3185 {
3186         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3187         kfree(rdev);
3188 }
3189 static const struct sysfs_ops rdev_sysfs_ops = {
3190         .show           = rdev_attr_show,
3191         .store          = rdev_attr_store,
3192 };
3193 static struct kobj_type rdev_ktype = {
3194         .release        = rdev_free,
3195         .sysfs_ops      = &rdev_sysfs_ops,
3196         .default_attrs  = rdev_default_attrs,
3197 };
3198
3199 int md_rdev_init(struct md_rdev *rdev)
3200 {
3201         rdev->desc_nr = -1;
3202         rdev->saved_raid_disk = -1;
3203         rdev->raid_disk = -1;
3204         rdev->flags = 0;
3205         rdev->data_offset = 0;
3206         rdev->new_data_offset = 0;
3207         rdev->sb_events = 0;
3208         rdev->last_read_error.tv_sec  = 0;
3209         rdev->last_read_error.tv_nsec = 0;
3210         rdev->sb_loaded = 0;
3211         rdev->bb_page = NULL;
3212         atomic_set(&rdev->nr_pending, 0);
3213         atomic_set(&rdev->read_errors, 0);
3214         atomic_set(&rdev->corrected_errors, 0);
3215
3216         INIT_LIST_HEAD(&rdev->same_set);
3217         init_waitqueue_head(&rdev->blocked_wait);
3218
3219         /* Add space to store bad block list.
3220          * This reserves the space even on arrays where it cannot
3221          * be used - I wonder if that matters
3222          */
3223         rdev->badblocks.count = 0;
3224         rdev->badblocks.shift = 0;
3225         rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3226         seqlock_init(&rdev->badblocks.lock);
3227         if (rdev->badblocks.page == NULL)
3228                 return -ENOMEM;
3229
3230         return 0;
3231 }
3232 EXPORT_SYMBOL_GPL(md_rdev_init);
3233 /*
3234  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3235  *
3236  * mark the device faulty if:
3237  *
3238  *   - the device is nonexistent (zero size)
3239  *   - the device has no valid superblock
3240  *
3241  * a faulty rdev _never_ has rdev->sb set.
3242  */
3243 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3244 {
3245         char b[BDEVNAME_SIZE];
3246         int err;
3247         struct md_rdev *rdev;
3248         sector_t size;
3249
3250         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3251         if (!rdev) {
3252                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3253                 return ERR_PTR(-ENOMEM);
3254         }
3255
3256         err = md_rdev_init(rdev);
3257         if (err)
3258                 goto abort_free;
3259         err = alloc_disk_sb(rdev);
3260         if (err)
3261                 goto abort_free;
3262
3263         err = lock_rdev(rdev, newdev, super_format == -2);
3264         if (err)
3265                 goto abort_free;
3266
3267         kobject_init(&rdev->kobj, &rdev_ktype);
3268
3269         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3270         if (!size) {
3271                 printk(KERN_WARNING
3272                         "md: %s has zero or unknown size, marking faulty!\n",
3273                         bdevname(rdev->bdev,b));
3274                 err = -EINVAL;
3275                 goto abort_free;
3276         }
3277
3278         if (super_format >= 0) {
3279                 err = super_types[super_format].
3280                         load_super(rdev, NULL, super_minor);
3281                 if (err == -EINVAL) {
3282                         printk(KERN_WARNING
3283                                 "md: %s does not have a valid v%d.%d "
3284                                "superblock, not importing!\n",
3285                                 bdevname(rdev->bdev,b),
3286                                super_format, super_minor);
3287                         goto abort_free;
3288                 }
3289                 if (err < 0) {
3290                         printk(KERN_WARNING
3291                                 "md: could not read %s's sb, not importing!\n",
3292                                 bdevname(rdev->bdev,b));
3293                         goto abort_free;
3294                 }
3295         }
3296         if (super_format == -1)
3297                 /* hot-add for 0.90, or non-persistent: so no badblocks */
3298                 rdev->badblocks.shift = -1;
3299
3300         return rdev;
3301
3302 abort_free:
3303         if (rdev->bdev)
3304                 unlock_rdev(rdev);
3305         md_rdev_clear(rdev);
3306         kfree(rdev);
3307         return ERR_PTR(err);
3308 }
3309
3310 /*
3311  * Check a full RAID array for plausibility
3312  */
3313
3314
3315 static void analyze_sbs(struct mddev * mddev)
3316 {
3317         int i;
3318         struct md_rdev *rdev, *freshest, *tmp;
3319         char b[BDEVNAME_SIZE];
3320
3321         freshest = NULL;
3322         rdev_for_each_safe(rdev, tmp, mddev)
3323                 switch (super_types[mddev->major_version].
3324                         load_super(rdev, freshest, mddev->minor_version)) {
3325                 case 1:
3326                         freshest = rdev;
3327                         break;
3328                 case 0:
3329                         break;
3330                 default:
3331                         printk( KERN_ERR \
3332                                 "md: fatal superblock inconsistency in %s"
3333                                 " -- removing from array\n",
3334                                 bdevname(rdev->bdev,b));
3335                         kick_rdev_from_array(rdev);
3336                 }
3337
3338
3339         super_types[mddev->major_version].
3340                 validate_super(mddev, freshest);
3341
3342         i = 0;
3343         rdev_for_each_safe(rdev, tmp, mddev) {
3344                 if (mddev->max_disks &&
3345                     (rdev->desc_nr >= mddev->max_disks ||
3346                      i > mddev->max_disks)) {
3347                         printk(KERN_WARNING
3348                                "md: %s: %s: only %d devices permitted\n",
3349                                mdname(mddev), bdevname(rdev->bdev, b),
3350                                mddev->max_disks);
3351                         kick_rdev_from_array(rdev);
3352                         continue;
3353                 }
3354                 if (rdev != freshest)
3355                         if (super_types[mddev->major_version].
3356                             validate_super(mddev, rdev)) {
3357                                 printk(KERN_WARNING "md: kicking non-fresh %s"
3358                                         " from array!\n",
3359                                         bdevname(rdev->bdev,b));
3360                                 kick_rdev_from_array(rdev);
3361                                 continue;
3362                         }
3363                 if (mddev->level == LEVEL_MULTIPATH) {
3364                         rdev->desc_nr = i++;
3365                         rdev->raid_disk = rdev->desc_nr;
3366                         set_bit(In_sync, &rdev->flags);
3367                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3368                         rdev->raid_disk = -1;
3369                         clear_bit(In_sync, &rdev->flags);
3370                 }
3371         }
3372 }
3373
3374 /* Read a fixed-point number.
3375  * Numbers in sysfs attributes should be in "standard" units where
3376  * possible, so time should be in seconds.
3377  * However we internally use a a much smaller unit such as
3378  * milliseconds or jiffies.
3379  * This function takes a decimal number with a possible fractional
3380  * component, and produces an integer which is the result of
3381  * multiplying that number by 10^'scale'.
3382  * all without any floating-point arithmetic.
3383  */
3384 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3385 {
3386         unsigned long result = 0;
3387         long decimals = -1;
3388         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3389                 if (*cp == '.')
3390                         decimals = 0;
3391                 else if (decimals < scale) {
3392                         unsigned int value;
3393                         value = *cp - '0';
3394                         result = result * 10 + value;
3395                         if (decimals >= 0)
3396                                 decimals++;
3397                 }
3398                 cp++;
3399         }
3400         if (*cp == '\n')
3401                 cp++;
3402         if (*cp)
3403                 return -EINVAL;
3404         if (decimals < 0)
3405                 decimals = 0;
3406         while (decimals < scale) {
3407                 result *= 10;
3408                 decimals ++;
3409         }
3410         *res = result;
3411         return 0;
3412 }
3413
3414
3415 static void md_safemode_timeout(unsigned long data);
3416
3417 static ssize_t
3418 safe_delay_show(struct mddev *mddev, char *page)
3419 {
3420         int msec = (mddev->safemode_delay*1000)/HZ;
3421         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3422 }
3423 static ssize_t
3424 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3425 {
3426         unsigned long msec;
3427
3428         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3429                 return -EINVAL;
3430         if (msec == 0)
3431                 mddev->safemode_delay = 0;
3432         else {
3433                 unsigned long old_delay = mddev->safemode_delay;
3434                 mddev->safemode_delay = (msec*HZ)/1000;
3435                 if (mddev->safemode_delay == 0)
3436                         mddev->safemode_delay = 1;
3437                 if (mddev->safemode_delay < old_delay)
3438                         md_safemode_timeout((unsigned long)mddev);
3439         }
3440         return len;
3441 }
3442 static struct md_sysfs_entry md_safe_delay =
3443 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3444
3445 static ssize_t
3446 level_show(struct mddev *mddev, char *page)
3447 {
3448         struct md_personality *p = mddev->pers;
3449         if (p)
3450                 return sprintf(page, "%s\n", p->name);
3451         else if (mddev->clevel[0])
3452                 return sprintf(page, "%s\n", mddev->clevel);
3453         else if (mddev->level != LEVEL_NONE)
3454                 return sprintf(page, "%d\n", mddev->level);
3455         else
3456                 return 0;
3457 }
3458
3459 static ssize_t
3460 level_store(struct mddev *mddev, const char *buf, size_t len)
3461 {
3462         char clevel[16];
3463         ssize_t rv = len;
3464         struct md_personality *pers;
3465         long level;
3466         void *priv;
3467         struct md_rdev *rdev;
3468
3469         if (mddev->pers == NULL) {
3470                 if (len == 0)
3471                         return 0;
3472                 if (len >= sizeof(mddev->clevel))
3473                         return -ENOSPC;
3474                 strncpy(mddev->clevel, buf, len);
3475                 if (mddev->clevel[len-1] == '\n')
3476                         len--;
3477                 mddev->clevel[len] = 0;
3478                 mddev->level = LEVEL_NONE;
3479                 return rv;
3480         }
3481
3482         /* request to change the personality.  Need to ensure:
3483          *  - array is not engaged in resync/recovery/reshape
3484          *  - old personality can be suspended
3485          *  - new personality will access other array.
3486          */
3487
3488         if (mddev->sync_thread ||
3489             mddev->reshape_position != MaxSector ||
3490             mddev->sysfs_active)
3491                 return -EBUSY;
3492
3493         if (!mddev->pers->quiesce) {
3494                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3495                        mdname(mddev), mddev->pers->name);
3496                 return -EINVAL;
3497         }
3498
3499         /* Now find the new personality */
3500         if (len == 0 || len >= sizeof(clevel))
3501                 return -EINVAL;
3502         strncpy(clevel, buf, len);
3503         if (clevel[len-1] == '\n')
3504                 len--;
3505         clevel[len] = 0;
3506         if (strict_strtol(clevel, 10, &level))
3507                 level = LEVEL_NONE;
3508
3509         if (request_module("md-%s", clevel) != 0)
3510                 request_module("md-level-%s", clevel);
3511         spin_lock(&pers_lock);
3512         pers = find_pers(level, clevel);
3513         if (!pers || !try_module_get(pers->owner)) {
3514                 spin_unlock(&pers_lock);
3515                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3516                 return -EINVAL;
3517         }
3518         spin_unlock(&pers_lock);
3519
3520         if (pers == mddev->pers) {
3521                 /* Nothing to do! */
3522                 module_put(pers->owner);
3523                 return rv;
3524         }
3525         if (!pers->takeover) {
3526                 module_put(pers->owner);
3527                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3528                        mdname(mddev), clevel);
3529                 return -EINVAL;
3530         }
3531
3532         rdev_for_each(rdev, mddev)
3533                 rdev->new_raid_disk = rdev->raid_disk;
3534
3535         /* ->takeover must set new_* and/or delta_disks
3536          * if it succeeds, and may set them when it fails.
3537          */
3538         priv = pers->takeover(mddev);
3539         if (IS_ERR(priv)) {
3540                 mddev->new_level = mddev->level;
3541                 mddev->new_layout = mddev->layout;
3542                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3543                 mddev->raid_disks -= mddev->delta_disks;
3544                 mddev->delta_disks = 0;
3545                 mddev->reshape_backwards = 0;
3546                 module_put(pers->owner);
3547                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3548                        mdname(mddev), clevel);
3549                 return PTR_ERR(priv);
3550         }
3551
3552         /* Looks like we have a winner */
3553         mddev_suspend(mddev);
3554         mddev->pers->stop(mddev);
3555
3556         if (mddev->pers->sync_request == NULL &&
3557             pers->sync_request != NULL) {
3558                 /* need to add the md_redundancy_group */
3559                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3560                         printk(KERN_WARNING
3561                                "md: cannot register extra attributes for %s\n",
3562                                mdname(mddev));
3563                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3564         }
3565         if (mddev->pers->sync_request != NULL &&
3566             pers->sync_request == NULL) {
3567                 /* need to remove the md_redundancy_group */
3568                 if (mddev->to_remove == NULL)
3569                         mddev->to_remove = &md_redundancy_group;
3570         }
3571
3572         if (mddev->pers->sync_request == NULL &&
3573             mddev->external) {
3574                 /* We are converting from a no-redundancy array
3575                  * to a redundancy array and metadata is managed
3576                  * externally so we need to be sure that writes
3577                  * won't block due to a need to transition
3578                  *      clean->dirty
3579                  * until external management is started.
3580                  */
3581                 mddev->in_sync = 0;
3582                 mddev->safemode_delay = 0;
3583                 mddev->safemode = 0;
3584         }
3585
3586         rdev_for_each(rdev, mddev) {
3587                 if (rdev->raid_disk < 0)
3588                         continue;
3589                 if (rdev->new_raid_disk >= mddev->raid_disks)
3590                         rdev->new_raid_disk = -1;
3591                 if (rdev->new_raid_disk == rdev->raid_disk)
3592                         continue;
3593                 sysfs_unlink_rdev(mddev, rdev);
3594         }
3595         rdev_for_each(rdev, mddev) {
3596                 if (rdev->raid_disk < 0)
3597                         continue;
3598                 if (rdev->new_raid_disk == rdev->raid_disk)
3599                         continue;
3600                 rdev->raid_disk = rdev->new_raid_disk;
3601                 if (rdev->raid_disk < 0)
3602                         clear_bit(In_sync, &rdev->flags);
3603                 else {
3604                         if (sysfs_link_rdev(mddev, rdev))
3605                                 printk(KERN_WARNING "md: cannot register rd%d"
3606                                        " for %s after level change\n",
3607                                        rdev->raid_disk, mdname(mddev));
3608                 }
3609         }
3610
3611         module_put(mddev->pers->owner);
3612         mddev->pers = pers;
3613         mddev->private = priv;
3614         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3615         mddev->level = mddev->new_level;
3616         mddev->layout = mddev->new_layout;
3617         mddev->chunk_sectors = mddev->new_chunk_sectors;
3618         mddev->delta_disks = 0;
3619         mddev->reshape_backwards = 0;
3620         mddev->degraded = 0;
3621         if (mddev->pers->sync_request == NULL) {
3622                 /* this is now an array without redundancy, so
3623                  * it must always be in_sync
3624                  */
3625                 mddev->in_sync = 1;
3626                 del_timer_sync(&mddev->safemode_timer);
3627         }
3628         pers->run(mddev);
3629         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3630         mddev_resume(mddev);
3631         sysfs_notify(&mddev->kobj, NULL, "level");
3632         md_new_event(mddev);
3633         return rv;
3634 }
3635
3636 static struct md_sysfs_entry md_level =
3637 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3638
3639
3640 static ssize_t
3641 layout_show(struct mddev *mddev, char *page)
3642 {
3643         /* just a number, not meaningful for all levels */
3644         if (mddev->reshape_position != MaxSector &&
3645             mddev->layout != mddev->new_layout)
3646                 return sprintf(page, "%d (%d)\n",
3647                                mddev->new_layout, mddev->layout);
3648         return sprintf(page, "%d\n", mddev->layout);
3649 }
3650
3651 static ssize_t
3652 layout_store(struct mddev *mddev, const char *buf, size_t len)
3653 {
3654         char *e;
3655         unsigned long n = simple_strtoul(buf, &e, 10);
3656
3657         if (!*buf || (*e && *e != '\n'))
3658                 return -EINVAL;
3659
3660         if (mddev->pers) {
3661                 int err;
3662                 if (mddev->pers->check_reshape == NULL)
3663                         return -EBUSY;
3664                 mddev->new_layout = n;
3665                 err = mddev->pers->check_reshape(mddev);
3666                 if (err) {
3667                         mddev->new_layout = mddev->layout;
3668                         return err;
3669                 }
3670         } else {
3671                 mddev->new_layout = n;
3672                 if (mddev->reshape_position == MaxSector)
3673                         mddev->layout = n;
3674         }
3675         return len;
3676 }
3677 static struct md_sysfs_entry md_layout =
3678 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3679
3680
3681 static ssize_t
3682 raid_disks_show(struct mddev *mddev, char *page)
3683 {
3684         if (mddev->raid_disks == 0)
3685                 return 0;
3686         if (mddev->reshape_position != MaxSector &&
3687             mddev->delta_disks != 0)
3688                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3689                                mddev->raid_disks - mddev->delta_disks);
3690         return sprintf(page, "%d\n", mddev->raid_disks);
3691 }
3692
3693 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3694
3695 static ssize_t
3696 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3697 {
3698         char *e;
3699         int rv = 0;
3700         unsigned long n = simple_strtoul(buf, &e, 10);
3701
3702         if (!*buf || (*e && *e != '\n'))
3703                 return -EINVAL;
3704
3705         if (mddev->pers)
3706                 rv = update_raid_disks(mddev, n);
3707         else if (mddev->reshape_position != MaxSector) {
3708                 struct md_rdev *rdev;
3709                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3710
3711                 rdev_for_each(rdev, mddev) {
3712                         if (olddisks < n &&
3713                             rdev->data_offset < rdev->new_data_offset)
3714                                 return -EINVAL;
3715                         if (olddisks > n &&
3716                             rdev->data_offset > rdev->new_data_offset)
3717                                 return -EINVAL;
3718                 }
3719                 mddev->delta_disks = n - olddisks;
3720                 mddev->raid_disks = n;
3721                 mddev->reshape_backwards = (mddev->delta_disks < 0);
3722         } else
3723                 mddev->raid_disks = n;
3724         return rv ? rv : len;
3725 }
3726 static struct md_sysfs_entry md_raid_disks =
3727 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3728
3729 static ssize_t
3730 chunk_size_show(struct mddev *mddev, char *page)
3731 {
3732         if (mddev->reshape_position != MaxSector &&
3733             mddev->chunk_sectors != mddev->new_chunk_sectors)
3734                 return sprintf(page, "%d (%d)\n",
3735                                mddev->new_chunk_sectors << 9,
3736                                mddev->chunk_sectors << 9);
3737         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3738 }
3739
3740 static ssize_t
3741 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3742 {
3743         char *e;
3744         unsigned long n = simple_strtoul(buf, &e, 10);
3745
3746         if (!*buf || (*e && *e != '\n'))
3747                 return -EINVAL;
3748
3749         if (mddev->pers) {
3750                 int err;
3751                 if (mddev->pers->check_reshape == NULL)
3752                         return -EBUSY;
3753                 mddev->new_chunk_sectors = n >> 9;
3754                 err = mddev->pers->check_reshape(mddev);
3755                 if (err) {
3756                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3757                         return err;
3758                 }
3759         } else {
3760                 mddev->new_chunk_sectors = n >> 9;
3761                 if (mddev->reshape_position == MaxSector)
3762                         mddev->chunk_sectors = n >> 9;
3763         }
3764         return len;
3765 }
3766 static struct md_sysfs_entry md_chunk_size =
3767 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3768
3769 static ssize_t
3770 resync_start_show(struct mddev *mddev, char *page)
3771 {
3772         if (mddev->recovery_cp == MaxSector)
3773                 return sprintf(page, "none\n");
3774         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3775 }
3776
3777 static ssize_t
3778 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3779 {
3780         char *e;
3781         unsigned long long n = simple_strtoull(buf, &e, 10);
3782
3783         if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3784                 return -EBUSY;
3785         if (cmd_match(buf, "none"))
3786                 n = MaxSector;
3787         else if (!*buf || (*e && *e != '\n'))
3788                 return -EINVAL;
3789
3790         mddev->recovery_cp = n;
3791         return len;
3792 }
3793 static struct md_sysfs_entry md_resync_start =
3794 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3795
3796 /*
3797  * The array state can be:
3798  *
3799  * clear
3800  *     No devices, no size, no level
3801  *     Equivalent to STOP_ARRAY ioctl
3802  * inactive
3803  *     May have some settings, but array is not active
3804  *        all IO results in error
3805  *     When written, doesn't tear down array, but just stops it
3806  * suspended (not supported yet)
3807  *     All IO requests will block. The array can be reconfigured.
3808  *     Writing this, if accepted, will block until array is quiescent
3809  * readonly
3810  *     no resync can happen.  no superblocks get written.
3811  *     write requests fail
3812  * read-auto
3813  *     like readonly, but behaves like 'clean' on a write request.
3814  *
3815  * clean - no pending writes, but otherwise active.
3816  *     When written to inactive array, starts without resync
3817  *     If a write request arrives then
3818  *       if metadata is known, mark 'dirty' and switch to 'active'.
3819  *       if not known, block and switch to write-pending
3820  *     If written to an active array that has pending writes, then fails.
3821  * active
3822  *     fully active: IO and resync can be happening.
3823  *     When written to inactive array, starts with resync
3824  *
3825  * write-pending
3826  *     clean, but writes are blocked waiting for 'active' to be written.
3827  *
3828  * active-idle
3829  *     like active, but no writes have been seen for a while (100msec).
3830  *
3831  */
3832 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3833                    write_pending, active_idle, bad_word};
3834 static char *array_states[] = {
3835         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3836         "write-pending", "active-idle", NULL };
3837
3838 static int match_word(const char *word, char **list)
3839 {
3840         int n;
3841         for (n=0; list[n]; n++)
3842                 if (cmd_match(word, list[n]))
3843                         break;
3844         return n;
3845 }
3846
3847 static ssize_t
3848 array_state_show(struct mddev *mddev, char *page)
3849 {
3850         enum array_state st = inactive;
3851
3852         if (mddev->pers)
3853                 switch(mddev->ro) {
3854                 case 1:
3855                         st = readonly;
3856                         break;
3857                 case 2:
3858                         st = read_auto;
3859                         break;
3860                 case 0:
3861                         if (mddev->in_sync)
3862                                 st = clean;
3863                         else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3864                                 st = write_pending;
3865                         else if (mddev->safemode)
3866                                 st = active_idle;
3867                         else
3868                                 st = active;
3869                 }
3870         else {
3871                 if (list_empty(&mddev->disks) &&
3872                     mddev->raid_disks == 0 &&
3873                     mddev->dev_sectors == 0)
3874                         st = clear;
3875                 else
3876                         st = inactive;
3877         }
3878         return sprintf(page, "%s\n", array_states[st]);
3879 }
3880
3881 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3882 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3883 static int do_md_run(struct mddev * mddev);
3884 static int restart_array(struct mddev *mddev);
3885
3886 static ssize_t
3887 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3888 {
3889         int err = -EINVAL;
3890         enum array_state st = match_word(buf, array_states);
3891         switch(st) {
3892         case bad_word:
3893                 break;
3894         case clear:
3895                 /* stopping an active array */
3896                 err = do_md_stop(mddev, 0, NULL);
3897                 break;
3898         case inactive:
3899                 /* stopping an active array */
3900                 if (mddev->pers)
3901                         err = do_md_stop(mddev, 2, NULL);
3902                 else
3903                         err = 0; /* already inactive */
3904                 break;
3905         case suspended:
3906                 break; /* not supported yet */
3907         case readonly:
3908                 if (mddev->pers)
3909                         err = md_set_readonly(mddev, NULL);
3910                 else {
3911                         mddev->ro = 1;
3912                         set_disk_ro(mddev->gendisk, 1);
3913                         err = do_md_run(mddev);
3914                 }
3915                 break;
3916         case read_auto:
3917                 if (mddev->pers) {
3918                         if (mddev->ro == 0)
3919                                 err = md_set_readonly(mddev, NULL);
3920                         else if (mddev->ro == 1)
3921                                 err = restart_array(mddev);
3922                         if (err == 0) {
3923                                 mddev->ro = 2;
3924                                 set_disk_ro(mddev->gendisk, 0);
3925                         }
3926                 } else {
3927                         mddev->ro = 2;
3928                         err = do_md_run(mddev);
3929                 }
3930                 break;
3931         case clean:
3932                 if (mddev->pers) {
3933                         restart_array(mddev);
3934                         spin_lock_irq(&mddev->write_lock);
3935                         if (atomic_read(&mddev->writes_pending) == 0) {
3936                                 if (mddev->in_sync == 0) {
3937                                         mddev->in_sync = 1;
3938                                         if (mddev->safemode == 1)
3939                                                 mddev->safemode = 0;
3940                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3941                                 }
3942                                 err = 0;
3943                         } else
3944                                 err = -EBUSY;
3945                         spin_unlock_irq(&mddev->write_lock);
3946                 } else
3947                         err = -EINVAL;
3948                 break;
3949         case active:
3950                 if (mddev->pers) {
3951                         restart_array(mddev);
3952                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3953                         wake_up(&mddev->sb_wait);
3954                         err = 0;
3955                 } else {
3956                         mddev->ro = 0;
3957                         set_disk_ro(mddev->gendisk, 0);
3958                         err = do_md_run(mddev);
3959                 }
3960                 break;
3961         case write_pending:
3962         case active_idle:
3963                 /* these cannot be set */
3964                 break;
3965         }
3966         if (err)
3967                 return err;
3968         else {
3969                 if (mddev->hold_active == UNTIL_IOCTL)
3970                         mddev->hold_active = 0;
3971                 sysfs_notify_dirent_safe(mddev->sysfs_state);
3972                 return len;
3973         }
3974 }
3975 static struct md_sysfs_entry md_array_state =
3976 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3977
3978 static ssize_t
3979 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3980         return sprintf(page, "%d\n",
3981                        atomic_read(&mddev->max_corr_read_errors));
3982 }
3983
3984 static ssize_t
3985 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3986 {
3987         char *e;
3988         unsigned long n = simple_strtoul(buf, &e, 10);
3989
3990         if (*buf && (*e == 0 || *e == '\n')) {
3991                 atomic_set(&mddev->max_corr_read_errors, n);
3992                 return len;
3993         }
3994         return -EINVAL;
3995 }
3996
3997 static struct md_sysfs_entry max_corr_read_errors =
3998 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3999         max_corrected_read_errors_store);
4000
4001 static ssize_t
4002 null_show(struct mddev *mddev, char *page)
4003 {
4004         return -EINVAL;
4005 }
4006
4007 static ssize_t
4008 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4009 {
4010         /* buf must be %d:%d\n? giving major and minor numbers */
4011         /* The new device is added to the array.
4012          * If the array has a persistent superblock, we read the
4013          * superblock to initialise info and check validity.
4014          * Otherwise, only checking done is that in bind_rdev_to_array,
4015          * which mainly checks size.
4016          */
4017         char *e;
4018         int major = simple_strtoul(buf, &e, 10);
4019         int minor;
4020         dev_t dev;
4021         struct md_rdev *rdev;
4022         int err;
4023
4024         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4025                 return -EINVAL;
4026         minor = simple_strtoul(e+1, &e, 10);
4027         if (*e && *e != '\n')
4028                 return -EINVAL;
4029         dev = MKDEV(major, minor);
4030         if (major != MAJOR(dev) ||
4031             minor != MINOR(dev))
4032                 return -EOVERFLOW;
4033
4034
4035         if (mddev->persistent) {
4036                 rdev = md_import_device(dev, mddev->major_version,
4037                                         mddev->minor_version);
4038                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4039                         struct md_rdev *rdev0
4040                                 = list_entry(mddev->disks.next,
4041                                              struct md_rdev, same_set);
4042                         err = super_types[mddev->major_version]
4043                                 .load_super(rdev, rdev0, mddev->minor_version);
4044                         if (err < 0)
4045                                 goto out;
4046                 }
4047         } else if (mddev->external)
4048                 rdev = md_import_device(dev, -2, -1);
4049         else
4050                 rdev = md_import_device(dev, -1, -1);
4051
4052         if (IS_ERR(rdev))
4053                 return PTR_ERR(rdev);
4054         err = bind_rdev_to_array(rdev, mddev);
4055  out:
4056         if (err)
4057                 export_rdev(rdev);
4058         return err ? err : len;
4059 }
4060
4061 static struct md_sysfs_entry md_new_device =
4062 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4063
4064 static ssize_t
4065 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4066 {
4067         char *end;
4068         unsigned long chunk, end_chunk;
4069
4070         if (!mddev->bitmap)
4071                 goto out;
4072         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4073         while (*buf) {
4074                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4075                 if (buf == end) break;
4076                 if (*end == '-') { /* range */
4077                         buf = end + 1;
4078                         end_chunk = simple_strtoul(buf, &end, 0);
4079                         if (buf == end) break;
4080                 }
4081                 if (*end && !isspace(*end)) break;
4082                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4083                 buf = skip_spaces(end);
4084         }
4085         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4086 out:
4087         return len;
4088 }
4089
4090 static struct md_sysfs_entry md_bitmap =
4091 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4092
4093 static ssize_t
4094 size_show(struct mddev *mddev, char *page)
4095 {
4096         return sprintf(page, "%llu\n",
4097                 (unsigned long long)mddev->dev_sectors / 2);
4098 }
4099
4100 static int update_size(struct mddev *mddev, sector_t num_sectors);
4101
4102 static ssize_t
4103 size_store(struct mddev *mddev, const char *buf, size_t len)
4104 {
4105         /* If array is inactive, we can reduce the component size, but
4106          * not increase it (except from 0).
4107          * If array is active, we can try an on-line resize
4108          */
4109         sector_t sectors;
4110         int err = strict_blocks_to_sectors(buf, &sectors);
4111
4112         if (err < 0)
4113                 return err;
4114         if (mddev->pers) {
4115                 err = update_size(mddev, sectors);
4116                 md_update_sb(mddev, 1);
4117         } else {
4118                 if (mddev->dev_sectors == 0 ||
4119                     mddev->dev_sectors > sectors)
4120                         mddev->dev_sectors = sectors;
4121                 else
4122                         err = -ENOSPC;
4123         }
4124         return err ? err : len;
4125 }
4126
4127 static struct md_sysfs_entry md_size =
4128 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4129
4130
4131 /* Metdata version.
4132  * This is one of
4133  *   'none' for arrays with no metadata (good luck...)
4134  *   'external' for arrays with externally managed metadata,
4135  * or N.M for internally known formats
4136  */
4137 static ssize_t
4138 metadata_show(struct mddev *mddev, char *page)
4139 {
4140         if (mddev->persistent)
4141                 return sprintf(page, "%d.%d\n",
4142                                mddev->major_version, mddev->minor_version);
4143         else if (mddev->external)
4144                 return sprintf(page, "external:%s\n", mddev->metadata_type);
4145         else
4146                 return sprintf(page, "none\n");
4147 }
4148
4149 static ssize_t
4150 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4151 {
4152         int major, minor;
4153         char *e;
4154         /* Changing the details of 'external' metadata is
4155          * always permitted.  Otherwise there must be
4156          * no devices attached to the array.
4157          */
4158         if (mddev->external && strncmp(buf, "external:", 9) == 0)
4159                 ;
4160         else if (!list_empty(&mddev->disks))
4161                 return -EBUSY;
4162
4163         if (cmd_match(buf, "none")) {
4164                 mddev->persistent = 0;
4165                 mddev->external = 0;
4166                 mddev->major_version = 0;
4167                 mddev->minor_version = 90;
4168                 return len;
4169         }
4170         if (strncmp(buf, "external:", 9) == 0) {
4171                 size_t namelen = len-9;
4172                 if (namelen >= sizeof(mddev->metadata_type))
4173                         namelen = sizeof(mddev->metadata_type)-1;
4174                 strncpy(mddev->metadata_type, buf+9, namelen);
4175                 mddev->metadata_type[namelen] = 0;
4176                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4177                         mddev->metadata_type[--namelen] = 0;
4178                 mddev->persistent = 0;
4179                 mddev->external = 1;
4180                 mddev->major_version = 0;
4181                 mddev->minor_version = 90;
4182                 return len;
4183         }
4184         major = simple_strtoul(buf, &e, 10);
4185         if (e==buf || *e != '.')
4186                 return -EINVAL;
4187         buf = e+1;
4188         minor = simple_strtoul(buf, &e, 10);
4189         if (e==buf || (*e && *e != '\n') )
4190                 return -EINVAL;
4191         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4192                 return -ENOENT;
4193         mddev->major_version = major;
4194         mddev->minor_version = minor;
4195         mddev->persistent = 1;
4196         mddev->external = 0;
4197         return len;
4198 }
4199
4200 static struct md_sysfs_entry md_metadata =
4201 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4202
4203 static ssize_t
4204 action_show(struct mddev *mddev, char *page)
4205 {
4206         char *type = "idle";
4207         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4208                 type = "frozen";
4209         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4210             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4211                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4212                         type = "reshape";
4213                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4214                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4215                                 type = "resync";
4216                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4217                                 type = "check";
4218                         else
4219                                 type = "repair";
4220                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4221                         type = "recover";
4222         }
4223         return sprintf(page, "%s\n", type);
4224 }
4225
4226 static void reap_sync_thread(struct mddev *mddev);
4227
4228 static ssize_t
4229 action_store(struct mddev *mddev, const char *page, size_t len)
4230 {
4231         if (!mddev->pers || !mddev->pers->sync_request)
4232                 return -EINVAL;
4233
4234         if (cmd_match(page, "frozen"))
4235                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4236         else
4237                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4238
4239         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4240                 if (mddev->sync_thread) {
4241                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4242                         reap_sync_thread(mddev);
4243                 }
4244         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4245                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4246                 return -EBUSY;
4247         else if (cmd_match(page, "resync"))
4248                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4249         else if (cmd_match(page, "recover")) {
4250                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4251                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4252         } else if (cmd_match(page, "reshape")) {
4253                 int err;
4254                 if (mddev->pers->start_reshape == NULL)
4255                         return -EINVAL;
4256                 err = mddev->pers->start_reshape(mddev);
4257                 if (err)
4258                         return err;
4259                 sysfs_notify(&mddev->kobj, NULL, "degraded");
4260         } else {
4261                 if (cmd_match(page, "check"))
4262                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4263                 else if (!cmd_match(page, "repair"))
4264                         return -EINVAL;
4265                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4266                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4267         }
4268         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4269         md_wakeup_thread(mddev->thread);
4270         sysfs_notify_dirent_safe(mddev->sysfs_action);
4271         return len;
4272 }
4273
4274 static ssize_t
4275 mismatch_cnt_show(struct mddev *mddev, char *page)
4276 {
4277         return sprintf(page, "%llu\n",
4278                        (unsigned long long) mddev->resync_mismatches);
4279 }
4280
4281 static struct md_sysfs_entry md_scan_mode =
4282 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4283
4284
4285 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4286
4287 static ssize_t
4288 sync_min_show(struct mddev *mddev, char *page)
4289 {
4290         return sprintf(page, "%d (%s)\n", speed_min(mddev),
4291                        mddev->sync_speed_min ? "local": "system");
4292 }
4293
4294 static ssize_t
4295 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4296 {
4297         int min;
4298         char *e;
4299         if (strncmp(buf, "system", 6)==0) {
4300                 mddev->sync_speed_min = 0;
4301                 return len;
4302         }
4303         min = simple_strtoul(buf, &e, 10);
4304         if (buf == e || (*e && *e != '\n') || min <= 0)
4305                 return -EINVAL;
4306         mddev->sync_speed_min = min;
4307         return len;
4308 }
4309
4310 static struct md_sysfs_entry md_sync_min =
4311 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4312
4313 static ssize_t
4314 sync_max_show(struct mddev *mddev, char *page)
4315 {
4316         return sprintf(page, "%d (%s)\n", speed_max(mddev),
4317                        mddev->sync_speed_max ? "local": "system");
4318 }
4319
4320 static ssize_t
4321 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4322 {
4323         int max;
4324         char *e;
4325         if (strncmp(buf, "system", 6)==0) {
4326                 mddev->sync_speed_max = 0;
4327                 return len;
4328         }
4329         max = simple_strtoul(buf, &e, 10);
4330         if (buf == e || (*e && *e != '\n') || max <= 0)
4331                 return -EINVAL;
4332         mddev->sync_speed_max = max;
4333         return len;
4334 }
4335
4336 static struct md_sysfs_entry md_sync_max =
4337 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4338
4339 static ssize_t
4340 degraded_show(struct mddev *mddev, char *page)
4341 {
4342         return sprintf(page, "%d\n", mddev->degraded);
4343 }
4344 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4345
4346 static ssize_t
4347 sync_force_parallel_show(struct mddev *mddev, char *page)
4348 {
4349         return sprintf(page, "%d\n", mddev->parallel_resync);
4350 }
4351
4352 static ssize_t
4353 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4354 {
4355         long n;
4356
4357         if (strict_strtol(buf, 10, &n))
4358                 return -EINVAL;
4359
4360         if (n != 0 && n != 1)
4361                 return -EINVAL;
4362
4363         mddev->parallel_resync = n;
4364
4365         if (mddev->sync_thread)
4366                 wake_up(&resync_wait);
4367
4368         return len;
4369 }
4370
4371 /* force parallel resync, even with shared block devices */
4372 static struct md_sysfs_entry md_sync_force_parallel =
4373 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4374        sync_force_parallel_show, sync_force_parallel_store);
4375
4376 static ssize_t
4377 sync_speed_show(struct mddev *mddev, char *page)
4378 {
4379         unsigned long resync, dt, db;
4380         if (mddev->curr_resync == 0)
4381                 return sprintf(page, "none\n");
4382         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4383         dt = (jiffies - mddev->resync_mark) / HZ;
4384         if (!dt) dt++;
4385         db = resync - mddev->resync_mark_cnt;
4386         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4387 }
4388
4389 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4390
4391 static ssize_t
4392 sync_completed_show(struct mddev *mddev, char *page)
4393 {
4394         unsigned long long max_sectors, resync;
4395
4396         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4397                 return sprintf(page, "none\n");
4398
4399         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4400             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4401                 max_sectors = mddev->resync_max_sectors;
4402         else
4403                 max_sectors = mddev->dev_sectors;
4404
4405         resync = mddev->curr_resync_completed;
4406         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4407 }
4408
4409 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4410
4411 static ssize_t
4412 min_sync_show(struct mddev *mddev, char *page)
4413 {
4414         return sprintf(page, "%llu\n",
4415                        (unsigned long long)mddev->resync_min);
4416 }
4417 static ssize_t
4418 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4419 {
4420         unsigned long long min;
4421         if (strict_strtoull(buf, 10, &min))
4422                 return -EINVAL;
4423         if (min > mddev->resync_max)
4424                 return -EINVAL;
4425         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4426                 return -EBUSY;
4427
4428         /* Must be a multiple of chunk_size */
4429         if (mddev->chunk_sectors) {
4430                 sector_t temp = min;
4431                 if (sector_div(temp, mddev->chunk_sectors))
4432                         return -EINVAL;
4433         }
4434         mddev->resync_min = min;
4435
4436         return len;
4437 }
4438
4439 static struct md_sysfs_entry md_min_sync =
4440 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4441
4442 static ssize_t
4443 max_sync_show(struct mddev *mddev, char *page)
4444 {
4445         if (mddev->resync_max == MaxSector)
4446                 return sprintf(page, "max\n");
4447         else
4448                 return sprintf(page, "%llu\n",
4449                                (unsigned long long)mddev->resync_max);
4450 }
4451 static ssize_t
4452 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4453 {
4454         if (strncmp(buf, "max", 3) == 0)
4455                 mddev->resync_max = MaxSector;
4456         else {
4457                 unsigned long long max;
4458                 if (strict_strtoull(buf, 10, &max))
4459                         return -EINVAL;
4460                 if (max < mddev->resync_min)
4461                         return -EINVAL;
4462                 if (max < mddev->resync_max &&
4463                     mddev->ro == 0 &&
4464                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4465                         return -EBUSY;
4466
4467                 /* Must be a multiple of chunk_size */
4468                 if (mddev->chunk_sectors) {
4469                         sector_t temp = max;
4470                         if (sector_div(temp, mddev->chunk_sectors))
4471                                 return -EINVAL;
4472                 }
4473                 mddev->resync_max = max;
4474         }
4475         wake_up(&mddev->recovery_wait);
4476         return len;
4477 }
4478
4479 static struct md_sysfs_entry md_max_sync =
4480 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4481
4482 static ssize_t
4483 suspend_lo_show(struct mddev *mddev, char *page)
4484 {
4485         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4486 }
4487
4488 static ssize_t
4489 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4490 {
4491         char *e;
4492         unsigned long long new = simple_strtoull(buf, &e, 10);
4493         unsigned long long old = mddev->suspend_lo;
4494
4495         if (mddev->pers == NULL ||
4496             mddev->pers->quiesce == NULL)
4497                 return -EINVAL;
4498         if (buf == e || (*e && *e != '\n'))
4499                 return -EINVAL;
4500
4501         mddev->suspend_lo = new;
4502         if (new >= old)
4503                 /* Shrinking suspended region */
4504                 mddev->pers->quiesce(mddev, 2);
4505         else {
4506                 /* Expanding suspended region - need to wait */
4507                 mddev->pers->quiesce(mddev, 1);
4508                 mddev->pers->quiesce(mddev, 0);
4509         }
4510         return len;
4511 }
4512 static struct md_sysfs_entry md_suspend_lo =
4513 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4514
4515
4516 static ssize_t
4517 suspend_hi_show(struct mddev *mddev, char *page)
4518 {
4519         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4520 }
4521
4522 static ssize_t
4523 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4524 {
4525         char *e;
4526         unsigned long long new = simple_strtoull(buf, &e, 10);
4527         unsigned long long old = mddev->suspend_hi;
4528
4529         if (mddev->pers == NULL ||
4530             mddev->pers->quiesce == NULL)
4531                 return -EINVAL;
4532         if (buf == e || (*e && *e != '\n'))
4533                 return -EINVAL;
4534
4535         mddev->suspend_hi = new;
4536         if (new <= old)
4537                 /* Shrinking suspended region */
4538                 mddev->pers->quiesce(mddev, 2);
4539         else {
4540                 /* Expanding suspended region - need to wait */
4541                 mddev->pers->quiesce(mddev, 1);
4542                 mddev->pers->quiesce(mddev, 0);
4543         }
4544         return len;
4545 }
4546 static struct md_sysfs_entry md_suspend_hi =
4547 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4548
4549 static ssize_t
4550 reshape_position_show(struct mddev *mddev, char *page)
4551 {
4552         if (mddev->reshape_position != MaxSector)
4553                 return sprintf(page, "%llu\n",
4554                                (unsigned long long)mddev->reshape_position);
4555         strcpy(page, "none\n");
4556         return 5;
4557 }
4558
4559 static ssize_t
4560 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4561 {
4562         struct md_rdev *rdev;
4563         char *e;
4564         unsigned long long new = simple_strtoull(buf, &e, 10);
4565         if (mddev->pers)
4566                 return -EBUSY;
4567         if (buf == e || (*e && *e != '\n'))
4568                 return -EINVAL;
4569         mddev->reshape_position = new;
4570         mddev->delta_disks = 0;
4571         mddev->reshape_backwards = 0;
4572         mddev->new_level = mddev->level;
4573         mddev->new_layout = mddev->layout;
4574         mddev->new_chunk_sectors = mddev->chunk_sectors;
4575         rdev_for_each(rdev, mddev)
4576                 rdev->new_data_offset = rdev->data_offset;
4577         return len;
4578 }
4579
4580 static struct md_sysfs_entry md_reshape_position =
4581 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4582        reshape_position_store);
4583
4584 static ssize_t
4585 reshape_direction_show(struct mddev *mddev, char *page)
4586 {
4587         return sprintf(page, "%s\n",
4588                        mddev->reshape_backwards ? "backwards" : "forwards");
4589 }
4590
4591 static ssize_t
4592 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4593 {
4594         int backwards = 0;
4595         if (cmd_match(buf, "forwards"))
4596                 backwards = 0;
4597         else if (cmd_match(buf, "backwards"))
4598                 backwards = 1;
4599         else
4600                 return -EINVAL;
4601         if (mddev->reshape_backwards == backwards)
4602                 return len;
4603
4604         /* check if we are allowed to change */
4605         if (mddev->delta_disks)
4606                 return -EBUSY;
4607
4608         if (mddev->persistent &&
4609             mddev->major_version == 0)
4610                 return -EINVAL;
4611
4612         mddev->reshape_backwards = backwards;
4613         return len;
4614 }
4615
4616 static struct md_sysfs_entry md_reshape_direction =
4617 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4618        reshape_direction_store);
4619
4620 static ssize_t
4621 array_size_show(struct mddev *mddev, char *page)
4622 {
4623         if (mddev->external_size)
4624                 return sprintf(page, "%llu\n",
4625                                (unsigned long long)mddev->array_sectors/2);
4626         else
4627                 return sprintf(page, "default\n");
4628 }
4629
4630 static ssize_t
4631 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4632 {
4633         sector_t sectors;
4634
4635         if (strncmp(buf, "default", 7) == 0) {
4636                 if (mddev->pers)
4637                         sectors = mddev->pers->size(mddev, 0, 0);
4638                 else
4639                         sectors = mddev->array_sectors;
4640
4641                 mddev->external_size = 0;
4642         } else {
4643                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4644                         return -EINVAL;
4645                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4646                         return -E2BIG;
4647
4648                 mddev->external_size = 1;
4649         }
4650
4651         mddev->array_sectors = sectors;
4652         if (mddev->pers) {
4653                 set_capacity(mddev->gendisk, mddev->array_sectors);
4654                 revalidate_disk(mddev->gendisk);
4655         }
4656         return len;
4657 }
4658
4659 static struct md_sysfs_entry md_array_size =
4660 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4661        array_size_store);
4662
4663 static struct attribute *md_default_attrs[] = {
4664         &md_level.attr,
4665         &md_layout.attr,
4666         &md_raid_disks.attr,
4667         &md_chunk_size.attr,
4668         &md_size.attr,
4669         &md_resync_start.attr,
4670         &md_metadata.attr,
4671         &md_new_device.attr,
4672         &md_safe_delay.attr,
4673         &md_array_state.attr,
4674         &md_reshape_position.attr,
4675         &md_reshape_direction.attr,
4676         &md_array_size.attr,
4677         &max_corr_read_errors.attr,
4678         NULL,
4679 };
4680
4681 static struct attribute *md_redundancy_attrs[] = {
4682         &md_scan_mode.attr,
4683         &md_mismatches.attr,
4684         &md_sync_min.attr,
4685         &md_sync_max.attr,
4686         &md_sync_speed.attr,
4687         &md_sync_force_parallel.attr,
4688         &md_sync_completed.attr,
4689         &md_min_sync.attr,
4690         &md_max_sync.attr,
4691         &md_suspend_lo.attr,
4692         &md_suspend_hi.attr,
4693         &md_bitmap.attr,
4694         &md_degraded.attr,
4695         NULL,
4696 };
4697 static struct attribute_group md_redundancy_group = {
4698         .name = NULL,
4699         .attrs = md_redundancy_attrs,
4700 };
4701
4702
4703 static ssize_t
4704 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4705 {
4706         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4707         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4708         ssize_t rv;
4709
4710         if (!entry->show)
4711                 return -EIO;
4712         spin_lock(&all_mddevs_lock);
4713         if (list_empty(&mddev->all_mddevs)) {
4714                 spin_unlock(&all_mddevs_lock);
4715                 return -EBUSY;
4716         }
4717         mddev_get(mddev);
4718         spin_unlock(&all_mddevs_lock);
4719
4720         rv = mddev_lock(mddev);
4721         if (!rv) {
4722                 rv = entry->show(mddev, page);
4723                 mddev_unlock(mddev);
4724         }
4725         mddev_put(mddev);
4726         return rv;
4727 }
4728
4729 static ssize_t
4730 md_attr_store(struct kobject *kobj, struct attribute *attr,
4731               const char *page, size_t length)
4732 {
4733         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4734         struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4735         ssize_t rv;
4736
4737         if (!entry->store)
4738                 return -EIO;
4739         if (!capable(CAP_SYS_ADMIN))
4740                 return -EACCES;
4741         spin_lock(&all_mddevs_lock);
4742         if (list_empty(&mddev->all_mddevs)) {
4743                 spin_unlock(&all_mddevs_lock);
4744                 return -EBUSY;
4745         }
4746         mddev_get(mddev);
4747         spin_unlock(&all_mddevs_lock);
4748         rv = mddev_lock(mddev);
4749         if (!rv) {
4750                 rv = entry->store(mddev, page, length);
4751                 mddev_unlock(mddev);
4752         }
4753         mddev_put(mddev);
4754         return rv;
4755 }
4756
4757 static void md_free(struct kobject *ko)
4758 {
4759         struct mddev *mddev = container_of(ko, struct mddev, kobj);
4760
4761         if (mddev->sysfs_state)
4762                 sysfs_put(mddev->sysfs_state);
4763
4764         if (mddev->gendisk) {
4765                 del_gendisk(mddev->gendisk);
4766                 put_disk(mddev->gendisk);
4767         }
4768         if (mddev->queue)
4769                 blk_cleanup_queue(mddev->queue);
4770
4771         kfree(mddev);
4772 }
4773
4774 static const struct sysfs_ops md_sysfs_ops = {
4775         .show   = md_attr_show,
4776         .store  = md_attr_store,
4777 };
4778 static struct kobj_type md_ktype = {
4779         .release        = md_free,
4780         .sysfs_ops      = &md_sysfs_ops,
4781         .default_attrs  = md_default_attrs,
4782 };
4783
4784 int mdp_major = 0;
4785
4786 static void mddev_delayed_delete(struct work_struct *ws)
4787 {
4788         struct mddev *mddev = container_of(ws, struct mddev, del_work);
4789
4790         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4791         kobject_del(&mddev->kobj);
4792         kobject_put(&mddev->kobj);
4793 }
4794
4795 static int md_alloc(dev_t dev, char *name)
4796 {
4797         static DEFINE_MUTEX(disks_mutex);
4798         struct mddev *mddev = mddev_find(dev);
4799         struct gendisk *disk;
4800         int partitioned;
4801         int shift;
4802         int unit;
4803         int error;
4804
4805         if (!mddev)
4806                 return -ENODEV;
4807
4808         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4809         shift = partitioned ? MdpMinorShift : 0;
4810         unit = MINOR(mddev->unit) >> shift;
4811
4812         /* wait for any previous instance of this device to be
4813          * completely removed (mddev_delayed_delete).
4814          */
4815         flush_workqueue(md_misc_wq);
4816
4817         mutex_lock(&disks_mutex);
4818         error = -EEXIST;
4819         if (mddev->gendisk)
4820                 goto abort;
4821
4822         if (name) {
4823                 /* Need to ensure that 'name' is not a duplicate.
4824                  */
4825                 struct mddev *mddev2;
4826                 spin_lock(&all_mddevs_lock);
4827
4828                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4829                         if (mddev2->gendisk &&
4830                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4831                                 spin_unlock(&all_mddevs_lock);
4832                                 goto abort;
4833                         }
4834                 spin_unlock(&all_mddevs_lock);
4835         }
4836
4837         error = -ENOMEM;
4838         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4839         if (!mddev->queue)
4840                 goto abort;
4841         mddev->queue->queuedata = mddev;
4842
4843         blk_queue_make_request(mddev->queue, md_make_request);
4844         blk_set_stacking_limits(&mddev->queue->limits);
4845
4846         disk = alloc_disk(1 << shift);
4847         if (!disk) {
4848                 blk_cleanup_queue(mddev->queue);
4849                 mddev->queue = NULL;
4850                 goto abort;
4851         }
4852         disk->major = MAJOR(mddev->unit);
4853         disk->first_minor = unit << shift;
4854         if (name)
4855                 strcpy(disk->disk_name, name);
4856         else if (partitioned)
4857                 sprintf(disk->disk_name, "md_d%d", unit);
4858         else
4859                 sprintf(disk->disk_name, "md%d", unit);
4860         disk->fops = &md_fops;
4861         disk->private_data = mddev;
4862         disk->queue = mddev->queue;
4863         blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4864         /* Allow extended partitions.  This makes the
4865          * 'mdp' device redundant, but we can't really
4866          * remove it now.
4867          */
4868         disk->flags |= GENHD_FL_EXT_DEVT;
4869         mddev->gendisk = disk;
4870         /* As soon as we call add_disk(), another thread could get
4871          * through to md_open, so make sure it doesn't get too far
4872          */
4873         mutex_lock(&mddev->open_mutex);
4874         add_disk(disk);
4875
4876         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4877                                      &disk_to_dev(disk)->kobj, "%s", "md");
4878         if (error) {
4879                 /* This isn't possible, but as kobject_init_and_add is marked
4880                  * __must_check, we must do something with the result
4881                  */
4882                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4883                        disk->disk_name);
4884                 error = 0;
4885         }
4886         if (mddev->kobj.sd &&
4887             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4888                 printk(KERN_DEBUG "pointless warning\n");
4889         mutex_unlock(&mddev->open_mutex);
4890  abort:
4891         mutex_unlock(&disks_mutex);
4892         if (!error && mddev->kobj.sd) {
4893                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4894                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4895         }
4896         mddev_put(mddev);
4897         return error;
4898 }
4899
4900 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4901 {
4902         md_alloc(dev, NULL);
4903         return NULL;
4904 }
4905
4906 static int add_named_array(const char *val, struct kernel_param *kp)
4907 {
4908         /* val must be "md_*" where * is not all digits.
4909          * We allocate an array with a large free minor number, and
4910          * set the name to val.  val must not already be an active name.
4911          */
4912         int len = strlen(val);
4913         char buf[DISK_NAME_LEN];
4914
4915         while (len && val[len-1] == '\n')
4916                 len--;
4917         if (len >= DISK_NAME_LEN)
4918                 return -E2BIG;
4919         strlcpy(buf, val, len+1);
4920         if (strncmp(buf, "md_", 3) != 0)
4921                 return -EINVAL;
4922         return md_alloc(0, buf);
4923 }
4924
4925 static void md_safemode_timeout(unsigned long data)
4926 {
4927         struct mddev *mddev = (struct mddev *) data;
4928
4929         if (!atomic_read(&mddev->writes_pending)) {
4930                 mddev->safemode = 1;
4931                 if (mddev->external)
4932                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4933         }
4934         md_wakeup_thread(mddev->thread);
4935 }
4936
4937 static int start_dirty_degraded;
4938
4939 int md_run(struct mddev *mddev)
4940 {
4941         int err;
4942         struct md_rdev *rdev;
4943         struct md_personality *pers;
4944
4945         if (list_empty(&mddev->disks))
4946                 /* cannot run an array with no devices.. */
4947                 return -EINVAL;
4948
4949         if (mddev->pers)
4950                 return -EBUSY;
4951         /* Cannot run until previous stop completes properly */
4952         if (mddev->sysfs_active)
4953                 return -EBUSY;
4954
4955         /*
4956          * Analyze all RAID superblock(s)
4957          */
4958         if (!mddev->raid_disks) {
4959                 if (!mddev->persistent)
4960                         return -EINVAL;
4961                 analyze_sbs(mddev);
4962         }
4963
4964         if (mddev->level != LEVEL_NONE)
4965                 request_module("md-level-%d", mddev->level);
4966         else if (mddev->clevel[0])
4967                 request_module("md-%s", mddev->clevel);
4968
4969         /*
4970          * Drop all container device buffers, from now on
4971          * the only valid external interface is through the md
4972          * device.
4973          */
4974         rdev_for_each(rdev, mddev) {
4975                 if (test_bit(Faulty, &rdev->flags))
4976                         continue;
4977                 sync_blockdev(rdev->bdev);
4978                 invalidate_bdev(rdev->bdev);
4979
4980                 /* perform some consistency tests on the device.
4981                  * We don't want the data to overlap the metadata,
4982                  * Internal Bitmap issues have been handled elsewhere.
4983                  */
4984                 if (rdev->meta_bdev) {
4985                         /* Nothing to check */;
4986                 } else if (rdev->data_offset < rdev->sb_start) {
4987                         if (mddev->dev_sectors &&
4988                             rdev->data_offset + mddev->dev_sectors
4989                             > rdev->sb_start) {
4990                                 printk("md: %s: data overlaps metadata\n",
4991                                        mdname(mddev));
4992                                 return -EINVAL;
4993                         }
4994                 } else {
4995                         if (rdev->sb_start + rdev->sb_size/512
4996                             > rdev->data_offset) {
4997                                 printk("md: %s: metadata overlaps data\n",
4998                                        mdname(mddev));
4999                                 return -EINVAL;
5000                         }
5001                 }
5002                 sysfs_notify_dirent_safe(rdev->sysfs_state);
5003         }
5004
5005         if (mddev->bio_set == NULL)
5006                 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
5007                                                sizeof(struct mddev *));
5008
5009         spin_lock(&pers_lock);
5010         pers = find_pers(mddev->level, mddev->clevel);
5011         if (!pers || !try_module_get(pers->owner)) {
5012                 spin_unlock(&pers_lock);
5013                 if (mddev->level != LEVEL_NONE)
5014                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5015                                mddev->level);
5016                 else
5017                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5018                                mddev->clevel);
5019                 return -EINVAL;
5020         }
5021         mddev->pers = pers;
5022         spin_unlock(&pers_lock);
5023         if (mddev->level != pers->level) {
5024                 mddev->level = pers->level;
5025                 mddev->new_level = pers->level;
5026         }
5027         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5028
5029         if (mddev->reshape_position != MaxSector &&
5030             pers->start_reshape == NULL) {
5031                 /* This personality cannot handle reshaping... */
5032                 mddev->pers = NULL;
5033                 module_put(pers->owner);
5034                 return -EINVAL;
5035         }
5036
5037         if (pers->sync_request) {
5038                 /* Warn if this is a potentially silly
5039                  * configuration.
5040                  */
5041                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5042                 struct md_rdev *rdev2;
5043                 int warned = 0;
5044
5045                 rdev_for_each(rdev, mddev)
5046                         rdev_for_each(rdev2, mddev) {
5047                                 if (rdev < rdev2 &&
5048                                     rdev->bdev->bd_contains ==
5049                                     rdev2->bdev->bd_contains) {
5050                                         printk(KERN_WARNING
5051                                                "%s: WARNING: %s appears to be"
5052                                                " on the same physical disk as"
5053                                                " %s.\n",
5054                                                mdname(mddev),
5055                                                bdevname(rdev->bdev,b),
5056                                                bdevname(rdev2->bdev,b2));
5057                                         warned = 1;
5058                                 }
5059                         }
5060
5061                 if (warned)
5062                         printk(KERN_WARNING
5063                                "True protection against single-disk"
5064                                " failure might be compromised.\n");
5065         }
5066
5067         mddev->recovery = 0;
5068         /* may be over-ridden by personality */
5069         mddev->resync_max_sectors = mddev->dev_sectors;
5070
5071         mddev->ok_start_degraded = start_dirty_degraded;
5072
5073         if (start_readonly && mddev->ro == 0)
5074                 mddev->ro = 2; /* read-only, but switch on first write */
5075
5076         err = mddev->pers->run(mddev);
5077         if (err)
5078                 printk(KERN_ERR "md: pers->run() failed ...\n");
5079         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5080                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5081                           " but 'external_size' not in effect?\n", __func__);
5082                 printk(KERN_ERR
5083                        "md: invalid array_size %llu > default size %llu\n",
5084                        (unsigned long long)mddev->array_sectors / 2,
5085                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5086                 err = -EINVAL;
5087                 mddev->pers->stop(mddev);
5088         }
5089         if (err == 0 && mddev->pers->sync_request &&
5090             (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5091                 err = bitmap_create(mddev);
5092                 if (err) {
5093                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5094                                mdname(mddev), err);
5095                         mddev->pers->stop(mddev);
5096                 }
5097         }
5098         if (err) {
5099                 module_put(mddev->pers->owner);
5100                 mddev->pers = NULL;
5101                 bitmap_destroy(mddev);
5102                 return err;
5103         }
5104         if (mddev->pers->sync_request) {
5105                 if (mddev->kobj.sd &&
5106                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5107                         printk(KERN_WARNING
5108                                "md: cannot register extra attributes for %s\n",
5109                                mdname(mddev));
5110                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5111         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5112                 mddev->ro = 0;
5113
5114         atomic_set(&mddev->writes_pending,0);
5115         atomic_set(&mddev->max_corr_read_errors,
5116                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5117         mddev->safemode = 0;
5118         mddev->safemode_timer.function = md_safemode_timeout;
5119         mddev->safemode_timer.data = (unsigned long) mddev;
5120         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5121         mddev->in_sync = 1;
5122         smp_wmb();
5123         mddev->ready = 1;
5124         rdev_for_each(rdev, mddev)
5125                 if (rdev->raid_disk >= 0)
5126                         if (sysfs_link_rdev(mddev, rdev))
5127                                 /* failure here is OK */;
5128
5129         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5130
5131         if (mddev->flags)
5132                 md_update_sb(mddev, 0);
5133
5134         md_new_event(mddev);
5135         sysfs_notify_dirent_safe(mddev->sysfs_state);
5136         sysfs_notify_dirent_safe(mddev->sysfs_action);
5137         sysfs_notify(&mddev->kobj, NULL, "degraded");
5138         return 0;
5139 }
5140 EXPORT_SYMBOL_GPL(md_run);
5141
5142 static int do_md_run(struct mddev *mddev)
5143 {
5144         int err;
5145
5146         err = md_run(mddev);
5147         if (err)
5148                 goto out;
5149         err = bitmap_load(mddev);
5150         if (err) {
5151                 bitmap_destroy(mddev);
5152                 goto out;
5153         }
5154
5155         md_wakeup_thread(mddev->thread);
5156         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5157
5158         set_capacity(mddev->gendisk, mddev->array_sectors);
5159         revalidate_disk(mddev->gendisk);
5160         mddev->changed = 1;
5161         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5162 out:
5163         return err;
5164 }
5165
5166 static int restart_array(struct mddev *mddev)
5167 {
5168         struct gendisk *disk = mddev->gendisk;
5169
5170         /* Complain if it has no devices */
5171         if (list_empty(&mddev->disks))
5172                 return -ENXIO;
5173         if (!mddev->pers)
5174                 return -EINVAL;
5175         if (!mddev->ro)
5176                 return -EBUSY;
5177         mddev->safemode = 0;
5178         mddev->ro = 0;
5179         set_disk_ro(disk, 0);
5180         printk(KERN_INFO "md: %s switched to read-write mode.\n",
5181                 mdname(mddev));
5182         /* Kick recovery or resync if necessary */
5183         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5184         md_wakeup_thread(mddev->thread);
5185         md_wakeup_thread(mddev->sync_thread);
5186         sysfs_notify_dirent_safe(mddev->sysfs_state);
5187         return 0;
5188 }
5189
5190 /* similar to deny_write_access, but accounts for our holding a reference
5191  * to the file ourselves */
5192 static int deny_bitmap_write_access(struct file * file)
5193 {
5194         struct inode *inode = file->f_mapping->host;
5195
5196         spin_lock(&inode->i_lock);
5197         if (atomic_read(&inode->i_writecount) > 1) {
5198                 spin_unlock(&inode->i_lock);
5199                 return -ETXTBSY;
5200         }
5201         atomic_set(&inode->i_writecount, -1);
5202         spin_unlock(&inode->i_lock);
5203
5204         return 0;
5205 }
5206
5207 void restore_bitmap_write_access(struct file *file)
5208 {
5209         struct inode *inode = file->f_mapping->host;
5210
5211         spin_lock(&inode->i_lock);
5212         atomic_set(&inode->i_writecount, 1);
5213         spin_unlock(&inode->i_lock);
5214 }
5215
5216 static void md_clean(struct mddev *mddev)
5217 {
5218         mddev->array_sectors = 0;
5219         mddev->external_size = 0;
5220         mddev->dev_sectors = 0;
5221         mddev->raid_disks = 0;
5222         mddev->recovery_cp = 0;
5223         mddev->resync_min = 0;
5224         mddev->resync_max = MaxSector;
5225         mddev->reshape_position = MaxSector;
5226         mddev->external = 0;
5227         mddev->persistent = 0;
5228         mddev->level = LEVEL_NONE;
5229         mddev->clevel[0] = 0;
5230         mddev->flags = 0;
5231         mddev->ro = 0;
5232         mddev->metadata_type[0] = 0;
5233         mddev->chunk_sectors = 0;
5234         mddev->ctime = mddev->utime = 0;
5235         mddev->layout = 0;
5236         mddev->max_disks = 0;
5237         mddev->events = 0;
5238         mddev->can_decrease_events = 0;
5239         mddev->delta_disks = 0;
5240         mddev->reshape_backwards = 0;
5241         mddev->new_level = LEVEL_NONE;
5242         mddev->new_layout = 0;
5243         mddev->new_chunk_sectors = 0;
5244         mddev->curr_resync = 0;
5245         mddev->resync_mismatches = 0;
5246         mddev->suspend_lo = mddev->suspend_hi = 0;
5247         mddev->sync_speed_min = mddev->sync_speed_max = 0;
5248         mddev->recovery = 0;
5249         mddev->in_sync = 0;
5250         mddev->changed = 0;
5251         mddev->degraded = 0;
5252         mddev->safemode = 0;
5253         mddev->merge_check_needed = 0;
5254         mddev->bitmap_info.offset = 0;
5255         mddev->bitmap_info.default_offset = 0;
5256         mddev->bitmap_info.default_space = 0;
5257         mddev->bitmap_info.chunksize = 0;
5258         mddev->bitmap_info.daemon_sleep = 0;
5259         mddev->bitmap_info.max_write_behind = 0;
5260 }
5261
5262 static void __md_stop_writes(struct mddev *mddev)
5263 {
5264         if (mddev->sync_thread) {
5265                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5266                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5267                 reap_sync_thread(mddev);
5268         }
5269
5270         del_timer_sync(&mddev->safemode_timer);
5271
5272         bitmap_flush(mddev);
5273         md_super_wait(mddev);
5274
5275         if (!mddev->in_sync || mddev->flags) {
5276                 /* mark array as shutdown cleanly */
5277                 mddev->in_sync = 1;
5278                 md_update_sb(mddev, 1);
5279         }
5280 }
5281
5282 void md_stop_writes(struct mddev *mddev)
5283 {
5284         mddev_lock(mddev);
5285         __md_stop_writes(mddev);
5286         mddev_unlock(mddev);
5287 }
5288 EXPORT_SYMBOL_GPL(md_stop_writes);
5289
5290 void md_stop(struct mddev *mddev)
5291 {
5292         mddev->ready = 0;
5293         mddev->pers->stop(mddev);
5294         if (mddev->pers->sync_request && mddev->to_remove == NULL)
5295                 mddev->to_remove = &md_redundancy_group;
5296         module_put(mddev->pers->owner);
5297         mddev->pers = NULL;
5298         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5299 }
5300 EXPORT_SYMBOL_GPL(md_stop);
5301
5302 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5303 {
5304         int err = 0;
5305         mutex_lock(&mddev->open_mutex);
5306         if (atomic_read(&mddev->openers) > !!bdev) {
5307                 printk("md: %s still in use.\n",mdname(mddev));
5308                 err = -EBUSY;
5309                 goto out;
5310         }
5311         if (bdev)
5312                 sync_blockdev(bdev);
5313         if (mddev->pers) {
5314                 __md_stop_writes(mddev);
5315
5316                 err  = -ENXIO;
5317                 if (mddev->ro==1)
5318                         goto out;
5319                 mddev->ro = 1;
5320                 set_disk_ro(mddev->gendisk, 1);
5321                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5322                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5323                 err = 0;
5324         }
5325 out:
5326         mutex_unlock(&mddev->open_mutex);
5327         return err;
5328 }
5329
5330 /* mode:
5331  *   0 - completely stop and dis-assemble array
5332  *   2 - stop but do not disassemble array
5333  */
5334 static int do_md_stop(struct mddev * mddev, int mode,
5335                       struct block_device *bdev)
5336 {
5337         struct gendisk *disk = mddev->gendisk;
5338         struct md_rdev *rdev;
5339
5340         mutex_lock(&mddev->open_mutex);
5341         if (atomic_read(&mddev->openers) > !!bdev ||
5342             mddev->sysfs_active) {
5343                 printk("md: %s still in use.\n",mdname(mddev));
5344                 mutex_unlock(&mddev->open_mutex);
5345                 return -EBUSY;
5346         }
5347         if (bdev)
5348                 /* It is possible IO was issued on some other
5349                  * open file which was closed before we took ->open_mutex.
5350                  * As that was not the last close __blkdev_put will not
5351                  * have called sync_blockdev, so we must.
5352                  */
5353                 sync_blockdev(bdev);
5354
5355         if (mddev->pers) {
5356                 if (mddev->ro)
5357                         set_disk_ro(disk, 0);
5358
5359                 __md_stop_writes(mddev);
5360                 md_stop(mddev);
5361                 mddev->queue->merge_bvec_fn = NULL;
5362                 mddev->queue->backing_dev_info.congested_fn = NULL;
5363
5364                 /* tell userspace to handle 'inactive' */
5365                 sysfs_notify_dirent_safe(mddev->sysfs_state);
5366
5367                 rdev_for_each(rdev, mddev)
5368                         if (rdev->raid_disk >= 0)
5369                                 sysfs_unlink_rdev(mddev, rdev);
5370
5371                 set_capacity(disk, 0);
5372                 mutex_unlock(&mddev->open_mutex);
5373                 mddev->changed = 1;
5374                 revalidate_disk(disk);
5375
5376                 if (mddev->ro)
5377                         mddev->ro = 0;
5378         } else
5379                 mutex_unlock(&mddev->open_mutex);
5380         /*
5381          * Free resources if final stop
5382          */
5383         if (mode == 0) {
5384                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5385
5386                 bitmap_destroy(mddev);
5387                 if (mddev->bitmap_info.file) {
5388                         restore_bitmap_write_access(mddev->bitmap_info.file);
5389                         fput(mddev->bitmap_info.file);
5390                         mddev->bitmap_info.file = NULL;
5391                 }
5392                 mddev->bitmap_info.offset = 0;
5393
5394                 export_array(mddev);
5395
5396                 md_clean(mddev);
5397                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5398                 if (mddev->hold_active == UNTIL_STOP)
5399                         mddev->hold_active = 0;
5400         }
5401         blk_integrity_unregister(disk);
5402         md_new_event(mddev);
5403         sysfs_notify_dirent_safe(mddev->sysfs_state);
5404         return 0;
5405 }
5406
5407 #ifndef MODULE
5408 static void autorun_array(struct mddev *mddev)
5409 {
5410         struct md_rdev *rdev;
5411         int err;
5412
5413         if (list_empty(&mddev->disks))
5414                 return;
5415
5416         printk(KERN_INFO "md: running: ");
5417
5418         rdev_for_each(rdev, mddev) {
5419                 char b[BDEVNAME_SIZE];
5420                 printk("<%s>", bdevname(rdev->bdev,b));
5421         }
5422         printk("\n");
5423
5424         err = do_md_run(mddev);
5425         if (err) {
5426                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5427                 do_md_stop(mddev, 0, NULL);
5428         }
5429 }
5430
5431 /*
5432  * lets try to run arrays based on all disks that have arrived
5433  * until now. (those are in pending_raid_disks)
5434  *
5435  * the method: pick the first pending disk, collect all disks with
5436  * the same UUID, remove all from the pending list and put them into
5437  * the 'same_array' list. Then order this list based on superblock
5438  * update time (freshest comes first), kick out 'old' disks and
5439  * compare superblocks. If everything's fine then run it.
5440  *
5441  * If "unit" is allocated, then bump its reference count
5442  */
5443 static void autorun_devices(int part)
5444 {
5445         struct md_rdev *rdev0, *rdev, *tmp;
5446         struct mddev *mddev;
5447         char b[BDEVNAME_SIZE];
5448
5449         printk(KERN_INFO "md: autorun ...\n");
5450         while (!list_empty(&pending_raid_disks)) {
5451                 int unit;
5452                 dev_t dev;
5453                 LIST_HEAD(candidates);
5454                 rdev0 = list_entry(pending_raid_disks.next,
5455                                          struct md_rdev, same_set);
5456
5457                 printk(KERN_INFO "md: considering %s ...\n",
5458                         bdevname(rdev0->bdev,b));
5459                 INIT_LIST_HEAD(&candidates);
5460                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5461                         if (super_90_load(rdev, rdev0, 0) >= 0) {
5462                                 printk(KERN_INFO "md:  adding %s ...\n",
5463                                         bdevname(rdev->bdev,b));
5464                                 list_move(&rdev->same_set, &candidates);
5465                         }
5466                 /*
5467                  * now we have a set of devices, with all of them having
5468                  * mostly sane superblocks. It's time to allocate the
5469                  * mddev.
5470                  */
5471                 if (part) {
5472                         dev = MKDEV(mdp_major,
5473                                     rdev0->preferred_minor << MdpMinorShift);
5474                         unit = MINOR(dev) >> MdpMinorShift;
5475                 } else {
5476                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5477                         unit = MINOR(dev);
5478                 }
5479                 if (rdev0->preferred_minor != unit) {
5480                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5481                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5482                         break;
5483                 }
5484
5485                 md_probe(dev, NULL, NULL);
5486                 mddev = mddev_find(dev);
5487                 if (!mddev || !mddev->gendisk) {
5488                         if (mddev)
5489                                 mddev_put(mddev);
5490                         printk(KERN_ERR
5491                                 "md: cannot allocate memory for md drive.\n");
5492                         break;
5493                 }
5494                 if (mddev_lock(mddev))
5495                         printk(KERN_WARNING "md: %s locked, cannot run\n",
5496                                mdname(mddev));
5497                 else if (mddev->raid_disks || mddev->major_version
5498                          || !list_empty(&mddev->disks)) {
5499                         printk(KERN_WARNING
5500                                 "md: %s already running, cannot run %s\n",
5501                                 mdname(mddev), bdevname(rdev0->bdev,b));
5502                         mddev_unlock(mddev);
5503                 } else {
5504                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
5505                         mddev->persistent = 1;
5506                         rdev_for_each_list(rdev, tmp, &candidates) {
5507                                 list_del_init(&rdev->same_set);
5508                                 if (bind_rdev_to_array(rdev, mddev))
5509                                         export_rdev(rdev);
5510                         }
5511                         autorun_array(mddev);
5512                         mddev_unlock(mddev);
5513                 }
5514                 /* on success, candidates will be empty, on error
5515                  * it won't...
5516                  */
5517                 rdev_for_each_list(rdev, tmp, &candidates) {
5518                         list_del_init(&rdev->same_set);
5519                         export_rdev(rdev);
5520                 }
5521                 mddev_put(mddev);
5522         }
5523         printk(KERN_INFO "md: ... autorun DONE.\n");
5524 }
5525 #endif /* !MODULE */
5526
5527 static int get_version(void __user * arg)
5528 {
5529         mdu_version_t ver;
5530
5531         ver.major = MD_MAJOR_VERSION;
5532         ver.minor = MD_MINOR_VERSION;
5533         ver.patchlevel = MD_PATCHLEVEL_VERSION;
5534
5535         if (copy_to_user(arg, &ver, sizeof(ver)))
5536                 return -EFAULT;
5537
5538         return 0;
5539 }
5540
5541 static int get_array_info(struct mddev * mddev, void __user * arg)
5542 {
5543         mdu_array_info_t info;
5544         int nr,working,insync,failed,spare;
5545         struct md_rdev *rdev;
5546
5547         nr=working=insync=failed=spare=0;
5548         rdev_for_each(rdev, mddev) {
5549                 nr++;
5550                 if (test_bit(Faulty, &rdev->flags))
5551                         failed++;
5552                 else {
5553                         working++;
5554                         if (test_bit(In_sync, &rdev->flags))
5555                                 insync++;
5556                         else
5557                                 spare++;
5558                 }
5559         }
5560
5561         info.major_version = mddev->major_version;
5562         info.minor_version = mddev->minor_version;
5563         info.patch_version = MD_PATCHLEVEL_VERSION;
5564         info.ctime         = mddev->ctime;
5565         info.level         = mddev->level;
5566         info.size          = mddev->dev_sectors / 2;
5567         if (info.size != mddev->dev_sectors / 2) /* overflow */
5568                 info.size = -1;
5569         info.nr_disks      = nr;
5570         info.raid_disks    = mddev->raid_disks;
5571         info.md_minor      = mddev->md_minor;
5572         info.not_persistent= !mddev->persistent;
5573
5574         info.utime         = mddev->utime;
5575         info.state         = 0;
5576         if (mddev->in_sync)
5577                 info.state = (1<<MD_SB_CLEAN);
5578         if (mddev->bitmap && mddev->bitmap_info.offset)
5579                 info.state = (1<<MD_SB_BITMAP_PRESENT);
5580         info.active_disks  = insync;
5581         info.working_disks = working;
5582         info.failed_disks  = failed;
5583         info.spare_disks   = spare;
5584
5585         info.layout        = mddev->layout;
5586         info.chunk_size    = mddev->chunk_sectors << 9;
5587
5588         if (copy_to_user(arg, &info, sizeof(info)))
5589                 return -EFAULT;
5590
5591         return 0;
5592 }
5593
5594 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5595 {
5596         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5597         char *ptr, *buf = NULL;
5598         int err = -ENOMEM;
5599
5600         if (md_allow_write(mddev))
5601                 file = kmalloc(sizeof(*file), GFP_NOIO);
5602         else
5603                 file = kmalloc(sizeof(*file), GFP_KERNEL);
5604
5605         if (!file)
5606                 goto out;
5607
5608         /* bitmap disabled, zero the first byte and copy out */
5609         if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5610                 file->pathname[0] = '\0';
5611                 goto copy_out;
5612         }
5613
5614         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5615         if (!buf)
5616                 goto out;
5617
5618         ptr = d_path(&mddev->bitmap->storage.file->f_path,
5619                      buf, sizeof(file->pathname));
5620         if (IS_ERR(ptr))
5621                 goto out;
5622
5623         strcpy(file->pathname, ptr);
5624
5625 copy_out:
5626         err = 0;
5627         if (copy_to_user(arg, file, sizeof(*file)))
5628                 err = -EFAULT;
5629 out:
5630         kfree(buf);
5631         kfree(file);
5632         return err;
5633 }
5634
5635 static int get_disk_info(struct mddev * mddev, void __user * arg)
5636 {
5637         mdu_disk_info_t info;
5638         struct md_rdev *rdev;
5639
5640         if (copy_from_user(&info, arg, sizeof(info)))
5641                 return -EFAULT;
5642
5643         rdev = find_rdev_nr(mddev, info.number);
5644         if (rdev) {
5645                 info.major = MAJOR(rdev->bdev->bd_dev);
5646                 info.minor = MINOR(rdev->bdev->bd_dev);
5647                 info.raid_disk = rdev->raid_disk;
5648                 info.state = 0;
5649                 if (test_bit(Faulty, &rdev->flags))
5650                         info.state |= (1<<MD_DISK_FAULTY);
5651                 else if (test_bit(In_sync, &rdev->flags)) {
5652                         info.state |= (1<<MD_DISK_ACTIVE);
5653                         info.state |= (1<<MD_DISK_SYNC);
5654                 }
5655                 if (test_bit(WriteMostly, &rdev->flags))
5656                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5657         } else {
5658                 info.major = info.minor = 0;
5659                 info.raid_disk = -1;
5660                 info.state = (1<<MD_DISK_REMOVED);
5661         }
5662
5663         if (copy_to_user(arg, &info, sizeof(info)))
5664                 return -EFAULT;
5665
5666         return 0;
5667 }
5668
5669 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5670 {
5671         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5672         struct md_rdev *rdev;
5673         dev_t dev = MKDEV(info->major,info->minor);
5674
5675         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5676                 return -EOVERFLOW;
5677
5678         if (!mddev->raid_disks) {
5679                 int err;
5680                 /* expecting a device which has a superblock */
5681                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5682                 if (IS_ERR(rdev)) {
5683                         printk(KERN_WARNING
5684                                 "md: md_import_device returned %ld\n",
5685                                 PTR_ERR(rdev));
5686                         return PTR_ERR(rdev);
5687                 }
5688                 if (!list_empty(&mddev->disks)) {
5689                         struct md_rdev *rdev0
5690                                 = list_entry(mddev->disks.next,
5691                                              struct md_rdev, same_set);
5692                         err = super_types[mddev->major_version]
5693                                 .load_super(rdev, rdev0, mddev->minor_version);
5694                         if (err < 0) {
5695                                 printk(KERN_WARNING
5696                                         "md: %s has different UUID to %s\n",
5697                                         bdevname(rdev->bdev,b),
5698                                         bdevname(rdev0->bdev,b2));
5699                                 export_rdev(rdev);
5700                                 return -EINVAL;
5701                         }
5702                 }
5703                 err = bind_rdev_to_array(rdev, mddev);
5704                 if (err)
5705                         export_rdev(rdev);
5706                 return err;
5707         }
5708
5709         /*
5710          * add_new_disk can be used once the array is assembled
5711          * to add "hot spares".  They must already have a superblock
5712          * written
5713          */
5714         if (mddev->pers) {
5715                 int err;
5716                 if (!mddev->pers->hot_add_disk) {
5717                         printk(KERN_WARNING
5718                                 "%s: personality does not support diskops!\n",
5719                                mdname(mddev));
5720                         return -EINVAL;
5721                 }
5722                 if (mddev->persistent)
5723                         rdev = md_import_device(dev, mddev->major_version,
5724                                                 mddev->minor_version);
5725                 else
5726                         rdev = md_import_device(dev, -1, -1);
5727                 if (IS_ERR(rdev)) {
5728                         printk(KERN_WARNING
5729                                 "md: md_import_device returned %ld\n",
5730                                 PTR_ERR(rdev));
5731                         return PTR_ERR(rdev);
5732                 }
5733                 /* set saved_raid_disk if appropriate */
5734                 if (!mddev->persistent) {
5735                         if (info->state & (1<<MD_DISK_SYNC)  &&
5736                             info->raid_disk < mddev->raid_disks) {
5737                                 rdev->raid_disk = info->raid_disk;
5738                                 set_bit(In_sync, &rdev->flags);
5739                         } else
5740                                 rdev->raid_disk = -1;
5741                 } else
5742                         super_types[mddev->major_version].
5743                                 validate_super(mddev, rdev);
5744                 if ((info->state & (1<<MD_DISK_SYNC)) &&
5745                      rdev->raid_disk != info->raid_disk) {
5746                         /* This was a hot-add request, but events doesn't
5747                          * match, so reject it.
5748                          */
5749                         export_rdev(rdev);
5750                         return -EINVAL;
5751                 }
5752
5753                 if (test_bit(In_sync, &rdev->flags))
5754                         rdev->saved_raid_disk = rdev->raid_disk;
5755                 else
5756                         rdev->saved_raid_disk = -1;
5757
5758                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5759                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5760                         set_bit(WriteMostly, &rdev->flags);
5761                 else
5762                         clear_bit(WriteMostly, &rdev->flags);
5763
5764                 rdev->raid_disk = -1;
5765                 err = bind_rdev_to_array(rdev, mddev);
5766                 if (!err && !mddev->pers->hot_remove_disk) {
5767                         /* If there is hot_add_disk but no hot_remove_disk
5768                          * then added disks for geometry changes,
5769                          * and should be added immediately.
5770                          */
5771                         super_types[mddev->major_version].
5772                                 validate_super(mddev, rdev);
5773                         err = mddev->pers->hot_add_disk(mddev, rdev);
5774                         if (err)
5775                                 unbind_rdev_from_array(rdev);
5776                 }
5777                 if (err)
5778                         export_rdev(rdev);
5779                 else
5780                         sysfs_notify_dirent_safe(rdev->sysfs_state);
5781
5782                 md_update_sb(mddev, 1);
5783                 if (mddev->degraded)
5784                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5785                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5786                 if (!err)
5787                         md_new_event(mddev);
5788                 md_wakeup_thread(mddev->thread);
5789                 return err;
5790         }
5791
5792         /* otherwise, add_new_disk is only allowed
5793          * for major_version==0 superblocks
5794          */
5795         if (mddev->major_version != 0) {
5796                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5797                        mdname(mddev));
5798                 return -EINVAL;
5799         }
5800
5801         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5802                 int err;
5803                 rdev = md_import_device(dev, -1, 0);
5804                 if (IS_ERR(rdev)) {
5805                         printk(KERN_WARNING
5806                                 "md: error, md_import_device() returned %ld\n",
5807                                 PTR_ERR(rdev));
5808                         return PTR_ERR(rdev);
5809                 }
5810                 rdev->desc_nr = info->number;
5811                 if (info->raid_disk < mddev->raid_disks)
5812                         rdev->raid_disk = info->raid_disk;
5813                 else
5814                         rdev->raid_disk = -1;
5815
5816                 if (rdev->raid_disk < mddev->raid_disks)
5817                         if (info->state & (1<<MD_DISK_SYNC))
5818                                 set_bit(In_sync, &rdev->flags);
5819
5820                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5821                         set_bit(WriteMostly, &rdev->flags);
5822
5823                 if (!mddev->persistent) {
5824                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5825                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5826                 } else
5827                         rdev->sb_start = calc_dev_sboffset(rdev);
5828                 rdev->sectors = rdev->sb_start;
5829
5830                 err = bind_rdev_to_array(rdev, mddev);
5831                 if (err) {
5832                         export_rdev(rdev);
5833                         return err;
5834                 }
5835         }
5836
5837         return 0;
5838 }
5839
5840 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5841 {
5842         char b[BDEVNAME_SIZE];
5843         struct md_rdev *rdev;
5844
5845         rdev = find_rdev(mddev, dev);
5846         if (!rdev)
5847                 return -ENXIO;
5848
5849         if (rdev->raid_disk >= 0)
5850                 goto busy;
5851
5852         kick_rdev_from_array(rdev);
5853         md_update_sb(mddev, 1);
5854         md_new_event(mddev);
5855
5856         return 0;
5857 busy:
5858         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5859                 bdevname(rdev->bdev,b), mdname(mddev));
5860         return -EBUSY;
5861 }
5862
5863 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5864 {
5865         char b[BDEVNAME_SIZE];
5866         int err;
5867         struct md_rdev *rdev;
5868
5869         if (!mddev->pers)
5870                 return -ENODEV;
5871
5872         if (mddev->major_version != 0) {
5873                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5874                         " version-0 superblocks.\n",
5875                         mdname(mddev));
5876                 return -EINVAL;
5877         }
5878         if (!mddev->pers->hot_add_disk) {
5879                 printk(KERN_WARNING
5880                         "%s: personality does not support diskops!\n",
5881                         mdname(mddev));
5882                 return -EINVAL;
5883         }
5884
5885         rdev = md_import_device(dev, -1, 0);
5886         if (IS_ERR(rdev)) {
5887                 printk(KERN_WARNING
5888                         "md: error, md_import_device() returned %ld\n",
5889                         PTR_ERR(rdev));
5890                 return -EINVAL;
5891         }
5892
5893         if (mddev->persistent)
5894                 rdev->sb_start = calc_dev_sboffset(rdev);
5895         else
5896                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5897
5898         rdev->sectors = rdev->sb_start;
5899
5900         if (test_bit(Faulty, &rdev->flags)) {
5901                 printk(KERN_WARNING
5902                         "md: can not hot-add faulty %s disk to %s!\n",
5903                         bdevname(rdev->bdev,b), mdname(mddev));
5904                 err = -EINVAL;
5905                 goto abort_export;
5906         }
5907         clear_bit(In_sync, &rdev->flags);
5908         rdev->desc_nr = -1;
5909         rdev->saved_raid_disk = -1;
5910         err = bind_rdev_to_array(rdev, mddev);
5911         if (err)
5912                 goto abort_export;
5913
5914         /*
5915          * The rest should better be atomic, we can have disk failures
5916          * noticed in interrupt contexts ...
5917          */
5918
5919         rdev->raid_disk = -1;
5920
5921         md_update_sb(mddev, 1);
5922
5923         /*
5924          * Kick recovery, maybe this spare has to be added to the
5925          * array immediately.
5926          */
5927         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5928         md_wakeup_thread(mddev->thread);
5929         md_new_event(mddev);
5930         return 0;
5931
5932 abort_export:
5933         export_rdev(rdev);
5934         return err;
5935 }
5936
5937 static int set_bitmap_file(struct mddev *mddev, int fd)
5938 {
5939         int err;
5940
5941         if (mddev->pers) {
5942                 if (!mddev->pers->quiesce)
5943                         return -EBUSY;
5944                 if (mddev->recovery || mddev->sync_thread)
5945                         return -EBUSY;
5946                 /* we should be able to change the bitmap.. */
5947         }
5948
5949
5950         if (fd >= 0) {
5951                 if (mddev->bitmap)
5952                         return -EEXIST; /* cannot add when bitmap is present */
5953                 mddev->bitmap_info.file = fget(fd);
5954
5955                 if (mddev->bitmap_info.file == NULL) {
5956                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5957                                mdname(mddev));
5958                         return -EBADF;
5959                 }
5960
5961                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5962                 if (err) {
5963                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5964                                mdname(mddev));
5965                         fput(mddev->bitmap_info.file);
5966                         mddev->bitmap_info.file = NULL;
5967                         return err;
5968                 }
5969                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5970         } else if (mddev->bitmap == NULL)
5971                 return -ENOENT; /* cannot remove what isn't there */
5972         err = 0;
5973         if (mddev->pers) {
5974                 mddev->pers->quiesce(mddev, 1);
5975                 if (fd >= 0) {
5976                         err = bitmap_create(mddev);
5977                         if (!err)
5978                                 err = bitmap_load(mddev);
5979                 }
5980                 if (fd < 0 || err) {
5981                         bitmap_destroy(mddev);
5982                         fd = -1; /* make sure to put the file */
5983                 }
5984                 mddev->pers->quiesce(mddev, 0);
5985         }
5986         if (fd < 0) {
5987                 if (mddev->bitmap_info.file) {
5988                         restore_bitmap_write_access(mddev->bitmap_info.file);
5989                         fput(mddev->bitmap_info.file);
5990                 }
5991                 mddev->bitmap_info.file = NULL;
5992         }
5993
5994         return err;
5995 }
5996
5997 /*
5998  * set_array_info is used two different ways
5999  * The original usage is when creating a new array.
6000  * In this usage, raid_disks is > 0 and it together with
6001  *  level, size, not_persistent,layout,chunksize determine the
6002  *  shape of the array.
6003  *  This will always create an array with a type-0.90.0 superblock.
6004  * The newer usage is when assembling an array.
6005  *  In this case raid_disks will be 0, and the major_version field is
6006  *  use to determine which style super-blocks are to be found on the devices.
6007  *  The minor and patch _version numbers are also kept incase the
6008  *  super_block handler wishes to interpret them.
6009  */
6010 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6011 {
6012
6013         if (info->raid_disks == 0) {
6014                 /* just setting version number for superblock loading */
6015                 if (info->major_version < 0 ||
6016                     info->major_version >= ARRAY_SIZE(super_types) ||
6017                     super_types[info->major_version].name == NULL) {
6018                         /* maybe try to auto-load a module? */
6019                         printk(KERN_INFO
6020                                 "md: superblock version %d not known\n",
6021                                 info->major_version);
6022                         return -EINVAL;
6023                 }
6024                 mddev->major_version = info->major_version;
6025                 mddev->minor_version = info->minor_version;
6026                 mddev->patch_version = info->patch_version;
6027                 mddev->persistent = !info->not_persistent;
6028                 /* ensure mddev_put doesn't delete this now that there
6029                  * is some minimal configuration.
6030                  */
6031                 mddev->ctime         = get_seconds();
6032                 return 0;
6033         }
6034         mddev->major_version = MD_MAJOR_VERSION;
6035         mddev->minor_version = MD_MINOR_VERSION;
6036         mddev->patch_version = MD_PATCHLEVEL_VERSION;
6037         mddev->ctime         = get_seconds();
6038
6039         mddev->level         = info->level;
6040         mddev->clevel[0]     = 0;
6041         mddev->dev_sectors   = 2 * (sector_t)info->size;
6042         mddev->raid_disks    = info->raid_disks;
6043         /* don't set md_minor, it is determined by which /dev/md* was
6044          * openned
6045          */
6046         if (info->state & (1<<MD_SB_CLEAN))
6047                 mddev->recovery_cp = MaxSector;
6048         else
6049                 mddev->recovery_cp = 0;
6050         mddev->persistent    = ! info->not_persistent;
6051         mddev->external      = 0;
6052
6053         mddev->layout        = info->layout;
6054         mddev->chunk_sectors = info->chunk_size >> 9;
6055
6056         mddev->max_disks     = MD_SB_DISKS;
6057
6058         if (mddev->persistent)
6059                 mddev->flags         = 0;
6060         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6061
6062         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6063         mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6064         mddev->bitmap_info.offset = 0;
6065
6066         mddev->reshape_position = MaxSector;
6067
6068         /*
6069          * Generate a 128 bit UUID
6070          */
6071         get_random_bytes(mddev->uuid, 16);
6072
6073         mddev->new_level = mddev->level;
6074         mddev->new_chunk_sectors = mddev->chunk_sectors;
6075         mddev->new_layout = mddev->layout;
6076         mddev->delta_disks = 0;
6077         mddev->reshape_backwards = 0;
6078
6079         return 0;
6080 }
6081
6082 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6083 {
6084         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6085
6086         if (mddev->external_size)
6087                 return;
6088
6089         mddev->array_sectors = array_sectors;
6090 }
6091 EXPORT_SYMBOL(md_set_array_sectors);
6092
6093 static int update_size(struct mddev *mddev, sector_t num_sectors)
6094 {
6095         struct md_rdev *rdev;
6096         int rv;
6097         int fit = (num_sectors == 0);
6098
6099         if (mddev->pers->resize == NULL)
6100                 return -EINVAL;
6101         /* The "num_sectors" is the number of sectors of each device that
6102          * is used.  This can only make sense for arrays with redundancy.
6103          * linear and raid0 always use whatever space is available. We can only
6104          * consider changing this number if no resync or reconstruction is
6105          * happening, and if the new size is acceptable. It must fit before the
6106          * sb_start or, if that is <data_offset, it must fit before the size
6107          * of each device.  If num_sectors is zero, we find the largest size
6108          * that fits.
6109          */
6110         if (mddev->sync_thread)
6111                 return -EBUSY;
6112
6113         rdev_for_each(rdev, mddev) {
6114                 sector_t avail = rdev->sectors;
6115
6116                 if (fit && (num_sectors == 0 || num_sectors > avail))
6117                         num_sectors = avail;
6118                 if (avail < num_sectors)
6119                         return -ENOSPC;
6120         }
6121         rv = mddev->pers->resize(mddev, num_sectors);
6122         if (!rv)
6123                 revalidate_disk(mddev->gendisk);
6124         return rv;
6125 }
6126
6127 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6128 {
6129         int rv;
6130         struct md_rdev *rdev;
6131         /* change the number of raid disks */
6132         if (mddev->pers->check_reshape == NULL)
6133                 return -EINVAL;
6134         if (raid_disks <= 0 ||
6135             (mddev->max_disks && raid_disks >= mddev->max_disks))
6136                 return -EINVAL;
6137         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6138                 return -EBUSY;
6139
6140         rdev_for_each(rdev, mddev) {
6141                 if (mddev->raid_disks < raid_disks &&
6142                     rdev->data_offset < rdev->new_data_offset)
6143                         return -EINVAL;
6144                 if (mddev->raid_disks > raid_disks &&
6145                     rdev->data_offset > rdev->new_data_offset)
6146                         return -EINVAL;
6147         }
6148
6149         mddev->delta_disks = raid_disks - mddev->raid_disks;
6150         if (mddev->delta_disks < 0)
6151                 mddev->reshape_backwards = 1;
6152         else if (mddev->delta_disks > 0)
6153                 mddev->reshape_backwards = 0;
6154
6155         rv = mddev->pers->check_reshape(mddev);
6156         if (rv < 0) {
6157                 mddev->delta_disks = 0;
6158                 mddev->reshape_backwards = 0;
6159         }
6160         return rv;
6161 }
6162
6163
6164 /*
6165  * update_array_info is used to change the configuration of an
6166  * on-line array.
6167  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6168  * fields in the info are checked against the array.
6169  * Any differences that cannot be handled will cause an error.
6170  * Normally, only one change can be managed at a time.
6171  */
6172 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6173 {
6174         int rv = 0;
6175         int cnt = 0;
6176         int state = 0;
6177
6178         /* calculate expected state,ignoring low bits */
6179         if (mddev->bitmap && mddev->bitmap_info.offset)
6180                 state |= (1 << MD_SB_BITMAP_PRESENT);
6181
6182         if (mddev->major_version != info->major_version ||
6183             mddev->minor_version != info->minor_version ||
6184 /*          mddev->patch_version != info->patch_version || */
6185             mddev->ctime         != info->ctime         ||
6186             mddev->level         != info->level         ||
6187 /*          mddev->layout        != info->layout        || */
6188             !mddev->persistent   != info->not_persistent||
6189             mddev->chunk_sectors != info->chunk_size >> 9 ||
6190             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6191             ((state^info->state) & 0xfffffe00)
6192                 )
6193                 return -EINVAL;
6194         /* Check there is only one change */
6195         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6196                 cnt++;
6197         if (mddev->raid_disks != info->raid_disks)
6198                 cnt++;
6199         if (mddev->layout != info->layout)
6200                 cnt++;
6201         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6202                 cnt++;
6203         if (cnt == 0)
6204                 return 0;
6205         if (cnt > 1)
6206                 return -EINVAL;
6207
6208         if (mddev->layout != info->layout) {
6209                 /* Change layout
6210                  * we don't need to do anything at the md level, the
6211                  * personality will take care of it all.
6212                  */
6213                 if (mddev->pers->check_reshape == NULL)
6214                         return -EINVAL;
6215                 else {
6216                         mddev->new_layout = info->layout;
6217                         rv = mddev->pers->check_reshape(mddev);
6218                         if (rv)
6219                                 mddev->new_layout = mddev->layout;
6220                         return rv;
6221                 }
6222         }
6223         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6224                 rv = update_size(mddev, (sector_t)info->size * 2);
6225
6226         if (mddev->raid_disks    != info->raid_disks)
6227                 rv = update_raid_disks(mddev, info->raid_disks);
6228
6229         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6230                 if (mddev->pers->quiesce == NULL)
6231                         return -EINVAL;
6232                 if (mddev->recovery || mddev->sync_thread)
6233                         return -EBUSY;
6234                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6235                         /* add the bitmap */
6236                         if (mddev->bitmap)
6237                                 return -EEXIST;
6238                         if (mddev->bitmap_info.default_offset == 0)
6239                                 return -EINVAL;
6240                         mddev->bitmap_info.offset =
6241                                 mddev->bitmap_info.default_offset;
6242                         mddev->bitmap_info.space =
6243                                 mddev->bitmap_info.default_space;
6244                         mddev->pers->quiesce(mddev, 1);
6245                         rv = bitmap_create(mddev);
6246                         if (!rv)
6247                                 rv = bitmap_load(mddev);
6248                         if (rv)
6249                                 bitmap_destroy(mddev);
6250                         mddev->pers->quiesce(mddev, 0);
6251                 } else {
6252                         /* remove the bitmap */
6253                         if (!mddev->bitmap)
6254                                 return -ENOENT;
6255                         if (mddev->bitmap->storage.file)
6256                                 return -EINVAL;
6257                         mddev->pers->quiesce(mddev, 1);
6258                         bitmap_destroy(mddev);
6259                         mddev->pers->quiesce(mddev, 0);
6260                         mddev->bitmap_info.offset = 0;
6261                 }
6262         }
6263         md_update_sb(mddev, 1);
6264         return rv;
6265 }
6266
6267 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6268 {
6269         struct md_rdev *rdev;
6270
6271         if (mddev->pers == NULL)
6272                 return -ENODEV;
6273
6274         rdev = find_rdev(mddev, dev);
6275         if (!rdev)
6276                 return -ENODEV;
6277
6278         md_error(mddev, rdev);
6279         if (!test_bit(Faulty, &rdev->flags))
6280                 return -EBUSY;
6281         return 0;
6282 }
6283
6284 /*
6285  * We have a problem here : there is no easy way to give a CHS
6286  * virtual geometry. We currently pretend that we have a 2 heads
6287  * 4 sectors (with a BIG number of cylinders...). This drives
6288  * dosfs just mad... ;-)
6289  */
6290 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6291 {
6292         struct mddev *mddev = bdev->bd_disk->private_data;
6293
6294         geo->heads = 2;
6295         geo->sectors = 4;
6296         geo->cylinders = mddev->array_sectors / 8;
6297         return 0;
6298 }
6299
6300 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6301                         unsigned int cmd, unsigned long arg)
6302 {
6303         int err = 0;
6304         void __user *argp = (void __user *)arg;
6305         struct mddev *mddev = NULL;
6306         int ro;
6307
6308         switch (cmd) {
6309         case RAID_VERSION:
6310         case GET_ARRAY_INFO:
6311         case GET_DISK_INFO:
6312                 break;
6313         default:
6314                 if (!capable(CAP_SYS_ADMIN))
6315                         return -EACCES;
6316         }
6317
6318         /*
6319          * Commands dealing with the RAID driver but not any
6320          * particular array:
6321          */
6322         switch (cmd)
6323         {
6324                 case RAID_VERSION:
6325                         err = get_version(argp);
6326                         goto done;
6327
6328                 case PRINT_RAID_DEBUG:
6329                         err = 0;
6330                         md_print_devices();
6331                         goto done;
6332
6333 #ifndef MODULE
6334                 case RAID_AUTORUN:
6335                         err = 0;
6336                         autostart_arrays(arg);
6337                         goto done;
6338 #endif
6339                 default:;
6340         }
6341
6342         /*
6343          * Commands creating/starting a new array:
6344          */
6345
6346         mddev = bdev->bd_disk->private_data;
6347
6348         if (!mddev) {
6349                 BUG();
6350                 goto abort;
6351         }
6352
6353         err = mddev_lock(mddev);
6354         if (err) {
6355                 printk(KERN_INFO
6356                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
6357                         err, cmd);
6358                 goto abort;
6359         }
6360
6361         switch (cmd)
6362         {
6363                 case SET_ARRAY_INFO:
6364                         {
6365                                 mdu_array_info_t info;
6366                                 if (!arg)
6367                                         memset(&info, 0, sizeof(info));
6368                                 else if (copy_from_user(&info, argp, sizeof(info))) {
6369                                         err = -EFAULT;
6370                                         goto abort_unlock;
6371                                 }
6372                                 if (mddev->pers) {
6373                                         err = update_array_info(mddev, &info);
6374                                         if (err) {
6375                                                 printk(KERN_WARNING "md: couldn't update"
6376                                                        " array info. %d\n", err);
6377                                                 goto abort_unlock;
6378                                         }
6379                                         goto done_unlock;
6380                                 }
6381                                 if (!list_empty(&mddev->disks)) {
6382                                         printk(KERN_WARNING
6383                                                "md: array %s already has disks!\n",
6384                                                mdname(mddev));
6385                                         err = -EBUSY;
6386                                         goto abort_unlock;
6387                                 }
6388                                 if (mddev->raid_disks) {
6389                                         printk(KERN_WARNING
6390                                                "md: array %s already initialised!\n",
6391                                                mdname(mddev));
6392                                         err = -EBUSY;
6393                                         goto abort_unlock;
6394                                 }
6395                                 err = set_array_info(mddev, &info);
6396                                 if (err) {
6397                                         printk(KERN_WARNING "md: couldn't set"
6398                                                " array info. %d\n", err);
6399                                         goto abort_unlock;
6400                                 }
6401                         }
6402                         goto done_unlock;
6403
6404                 default:;
6405         }
6406
6407         /*
6408          * Commands querying/configuring an existing array:
6409          */
6410         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6411          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6412         if ((!mddev->raid_disks && !mddev->external)
6413             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6414             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6415             && cmd != GET_BITMAP_FILE) {
6416                 err = -ENODEV;
6417                 goto abort_unlock;
6418         }
6419
6420         /*
6421          * Commands even a read-only array can execute:
6422          */
6423         switch (cmd)
6424         {
6425                 case GET_ARRAY_INFO:
6426                         err = get_array_info(mddev, argp);
6427                         goto done_unlock;
6428
6429                 case GET_BITMAP_FILE:
6430                         err = get_bitmap_file(mddev, argp);
6431                         goto done_unlock;
6432
6433                 case GET_DISK_INFO:
6434                         err = get_disk_info(mddev, argp);
6435                         goto done_unlock;
6436
6437                 case RESTART_ARRAY_RW:
6438                         err = restart_array(mddev);
6439                         goto done_unlock;
6440
6441                 case STOP_ARRAY:
6442                         err = do_md_stop(mddev, 0, bdev);
6443                         goto done_unlock;
6444
6445                 case STOP_ARRAY_RO:
6446                         err = md_set_readonly(mddev, bdev);
6447                         goto done_unlock;
6448
6449                 case BLKROSET:
6450                         if (get_user(ro, (int __user *)(arg))) {
6451                                 err = -EFAULT;
6452                                 goto done_unlock;
6453                         }
6454                         err = -EINVAL;
6455
6456                         /* if the bdev is going readonly the value of mddev->ro
6457                          * does not matter, no writes are coming
6458                          */
6459                         if (ro)
6460                                 goto done_unlock;
6461
6462                         /* are we are already prepared for writes? */
6463                         if (mddev->ro != 1)
6464                                 goto done_unlock;
6465
6466                         /* transitioning to readauto need only happen for
6467                          * arrays that call md_write_start
6468                          */
6469                         if (mddev->pers) {
6470                                 err = restart_array(mddev);
6471                                 if (err == 0) {
6472                                         mddev->ro = 2;
6473                                         set_disk_ro(mddev->gendisk, 0);
6474                                 }
6475                         }
6476                         goto done_unlock;
6477         }
6478
6479         /*
6480          * The remaining ioctls are changing the state of the
6481          * superblock, so we do not allow them on read-only arrays.
6482          * However non-MD ioctls (e.g. get-size) will still come through
6483          * here and hit the 'default' below, so only disallow
6484          * 'md' ioctls, and switch to rw mode if started auto-readonly.
6485          */
6486         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6487                 if (mddev->ro == 2) {
6488                         mddev->ro = 0;
6489                         sysfs_notify_dirent_safe(mddev->sysfs_state);
6490                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6491                         md_wakeup_thread(mddev->thread);
6492                 } else {
6493                         err = -EROFS;
6494                         goto abort_unlock;
6495                 }
6496         }
6497
6498         switch (cmd)
6499         {
6500                 case ADD_NEW_DISK:
6501                 {
6502                         mdu_disk_info_t info;
6503                         if (copy_from_user(&info, argp, sizeof(info)))
6504                                 err = -EFAULT;
6505                         else
6506                                 err = add_new_disk(mddev, &info);
6507                         goto done_unlock;
6508                 }
6509
6510                 case HOT_REMOVE_DISK:
6511                         err = hot_remove_disk(mddev, new_decode_dev(arg));
6512                         goto done_unlock;
6513
6514                 case HOT_ADD_DISK:
6515                         err = hot_add_disk(mddev, new_decode_dev(arg));
6516                         goto done_unlock;
6517
6518                 case SET_DISK_FAULTY:
6519                         err = set_disk_faulty(mddev, new_decode_dev(arg));
6520                         goto done_unlock;
6521
6522                 case RUN_ARRAY:
6523                         err = do_md_run(mddev);
6524                         goto done_unlock;
6525
6526                 case SET_BITMAP_FILE:
6527                         err = set_bitmap_file(mddev, (int)arg);
6528                         goto done_unlock;
6529
6530                 default:
6531                         err = -EINVAL;
6532                         goto abort_unlock;
6533         }
6534
6535 done_unlock:
6536 abort_unlock:
6537         if (mddev->hold_active == UNTIL_IOCTL &&
6538             err != -EINVAL)
6539                 mddev->hold_active = 0;
6540         mddev_unlock(mddev);
6541
6542         return err;
6543 done:
6544         if (err)
6545                 MD_BUG();
6546 abort:
6547         return err;
6548 }
6549 #ifdef CONFIG_COMPAT
6550 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6551                     unsigned int cmd, unsigned long arg)
6552 {
6553         switch (cmd) {
6554         case HOT_REMOVE_DISK:
6555         case HOT_ADD_DISK:
6556         case SET_DISK_FAULTY:
6557         case SET_BITMAP_FILE:
6558                 /* These take in integer arg, do not convert */
6559                 break;
6560         default:
6561                 arg = (unsigned long)compat_ptr(arg);
6562                 break;
6563         }
6564
6565         return md_ioctl(bdev, mode, cmd, arg);
6566 }
6567 #endif /* CONFIG_COMPAT */
6568
6569 static int md_open(struct block_device *bdev, fmode_t mode)
6570 {
6571         /*
6572          * Succeed if we can lock the mddev, which confirms that
6573          * it isn't being stopped right now.
6574          */
6575         struct mddev *mddev = mddev_find(bdev->bd_dev);
6576         int err;
6577
6578         if (!mddev)
6579                 return -ENODEV;
6580
6581         if (mddev->gendisk != bdev->bd_disk) {
6582                 /* we are racing with mddev_put which is discarding this
6583                  * bd_disk.
6584                  */
6585                 mddev_put(mddev);
6586                 /* Wait until bdev->bd_disk is definitely gone */
6587                 flush_workqueue(md_misc_wq);
6588                 /* Then retry the open from the top */
6589                 return -ERESTARTSYS;
6590         }
6591         BUG_ON(mddev != bdev->bd_disk->private_data);
6592
6593         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6594                 goto out;
6595
6596         err = 0;
6597         atomic_inc(&mddev->openers);
6598         mutex_unlock(&mddev->open_mutex);
6599
6600         check_disk_change(bdev);
6601  out:
6602         return err;
6603 }
6604
6605 static int md_release(struct gendisk *disk, fmode_t mode)
6606 {
6607         struct mddev *mddev = disk->private_data;
6608
6609         BUG_ON(!mddev);
6610         atomic_dec(&mddev->openers);
6611         mddev_put(mddev);
6612
6613         return 0;
6614 }
6615
6616 static int md_media_changed(struct gendisk *disk)
6617 {
6618         struct mddev *mddev = disk->private_data;
6619
6620         return mddev->changed;
6621 }
6622
6623 static int md_revalidate(struct gendisk *disk)
6624 {
6625         struct mddev *mddev = disk->private_data;
6626
6627         mddev->changed = 0;
6628         return 0;
6629 }
6630 static const struct block_device_operations md_fops =
6631 {
6632         .owner          = THIS_MODULE,
6633         .open           = md_open,
6634         .release        = md_release,
6635         .ioctl          = md_ioctl,
6636 #ifdef CONFIG_COMPAT
6637         .compat_ioctl   = md_compat_ioctl,
6638 #endif
6639         .getgeo         = md_getgeo,
6640         .media_changed  = md_media_changed,
6641         .revalidate_disk= md_revalidate,
6642 };
6643
6644 static int md_thread(void * arg)
6645 {
6646         struct md_thread *thread = arg;
6647
6648         /*
6649          * md_thread is a 'system-thread', it's priority should be very
6650          * high. We avoid resource deadlocks individually in each
6651          * raid personality. (RAID5 does preallocation) We also use RR and
6652          * the very same RT priority as kswapd, thus we will never get
6653          * into a priority inversion deadlock.
6654          *
6655          * we definitely have to have equal or higher priority than
6656          * bdflush, otherwise bdflush will deadlock if there are too
6657          * many dirty RAID5 blocks.
6658          */
6659
6660         allow_signal(SIGKILL);
6661         while (!kthread_should_stop()) {
6662
6663                 /* We need to wait INTERRUPTIBLE so that
6664                  * we don't add to the load-average.
6665                  * That means we need to be sure no signals are
6666                  * pending
6667                  */
6668                 if (signal_pending(current))
6669                         flush_signals(current);
6670
6671                 wait_event_interruptible_timeout
6672                         (thread->wqueue,
6673                          test_bit(THREAD_WAKEUP, &thread->flags)
6674                          || kthread_should_stop(),
6675                          thread->timeout);
6676
6677                 clear_bit(THREAD_WAKEUP, &thread->flags);
6678                 if (!kthread_should_stop())
6679                         thread->run(thread->mddev);
6680         }
6681
6682         return 0;
6683 }
6684
6685 void md_wakeup_thread(struct md_thread *thread)
6686 {
6687         if (thread) {
6688                 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6689                 set_bit(THREAD_WAKEUP, &thread->flags);
6690                 wake_up(&thread->wqueue);
6691         }
6692 }
6693
6694 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
6695                                  const char *name)
6696 {
6697         struct md_thread *thread;
6698
6699         thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6700         if (!thread)
6701                 return NULL;
6702
6703         init_waitqueue_head(&thread->wqueue);
6704
6705         thread->run = run;
6706         thread->mddev = mddev;
6707         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6708         thread->tsk = kthread_run(md_thread, thread,
6709                                   "%s_%s",
6710                                   mdname(thread->mddev),
6711                                   name);
6712         if (IS_ERR(thread->tsk)) {
6713                 kfree(thread);
6714                 return NULL;
6715         }
6716         return thread;
6717 }
6718
6719 void md_unregister_thread(struct md_thread **threadp)
6720 {
6721         struct md_thread *thread = *threadp;
6722         if (!thread)
6723                 return;
6724         pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6725         /* Locking ensures that mddev_unlock does not wake_up a
6726          * non-existent thread
6727          */
6728         spin_lock(&pers_lock);
6729         *threadp = NULL;
6730         spin_unlock(&pers_lock);
6731
6732         kthread_stop(thread->tsk);
6733         kfree(thread);
6734 }
6735
6736 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6737 {
6738         if (!mddev) {
6739                 MD_BUG();
6740                 return;
6741         }
6742
6743         if (!rdev || test_bit(Faulty, &rdev->flags))
6744                 return;
6745
6746         if (!mddev->pers || !mddev->pers->error_handler)
6747                 return;
6748         mddev->pers->error_handler(mddev,rdev);
6749         if (mddev->degraded)
6750                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6751         sysfs_notify_dirent_safe(rdev->sysfs_state);
6752         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6753         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6754         md_wakeup_thread(mddev->thread);
6755         if (mddev->event_work.func)
6756                 queue_work(md_misc_wq, &mddev->event_work);
6757         md_new_event_inintr(mddev);
6758 }
6759
6760 /* seq_file implementation /proc/mdstat */
6761
6762 static void status_unused(struct seq_file *seq)
6763 {
6764         int i = 0;
6765         struct md_rdev *rdev;
6766
6767         seq_printf(seq, "unused devices: ");
6768
6769         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6770                 char b[BDEVNAME_SIZE];
6771                 i++;
6772                 seq_printf(seq, "%s ",
6773                               bdevname(rdev->bdev,b));
6774         }
6775         if (!i)
6776                 seq_printf(seq, "<none>");
6777
6778         seq_printf(seq, "\n");
6779 }
6780
6781
6782 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6783 {
6784         sector_t max_sectors, resync, res;
6785         unsigned long dt, db;
6786         sector_t rt;
6787         int scale;
6788         unsigned int per_milli;
6789
6790         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6791
6792         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6793             test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6794                 max_sectors = mddev->resync_max_sectors;
6795         else
6796                 max_sectors = mddev->dev_sectors;
6797
6798         /*
6799          * Should not happen.
6800          */
6801         if (!max_sectors) {
6802                 MD_BUG();
6803                 return;
6804         }
6805         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6806          * in a sector_t, and (max_sectors>>scale) will fit in a
6807          * u32, as those are the requirements for sector_div.
6808          * Thus 'scale' must be at least 10
6809          */
6810         scale = 10;
6811         if (sizeof(sector_t) > sizeof(unsigned long)) {
6812                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6813                         scale++;
6814         }
6815         res = (resync>>scale)*1000;
6816         sector_div(res, (u32)((max_sectors>>scale)+1));
6817
6818         per_milli = res;
6819         {
6820                 int i, x = per_milli/50, y = 20-x;
6821                 seq_printf(seq, "[");
6822                 for (i = 0; i < x; i++)
6823                         seq_printf(seq, "=");
6824                 seq_printf(seq, ">");
6825                 for (i = 0; i < y; i++)
6826                         seq_printf(seq, ".");
6827                 seq_printf(seq, "] ");
6828         }
6829         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6830                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6831                     "reshape" :
6832                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6833                      "check" :
6834                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6835                       "resync" : "recovery"))),
6836                    per_milli/10, per_milli % 10,
6837                    (unsigned long long) resync/2,
6838                    (unsigned long long) max_sectors/2);
6839
6840         /*
6841          * dt: time from mark until now
6842          * db: blocks written from mark until now
6843          * rt: remaining time
6844          *
6845          * rt is a sector_t, so could be 32bit or 64bit.
6846          * So we divide before multiply in case it is 32bit and close
6847          * to the limit.
6848          * We scale the divisor (db) by 32 to avoid losing precision
6849          * near the end of resync when the number of remaining sectors
6850          * is close to 'db'.
6851          * We then divide rt by 32 after multiplying by db to compensate.
6852          * The '+1' avoids division by zero if db is very small.
6853          */
6854         dt = ((jiffies - mddev->resync_mark) / HZ);
6855         if (!dt) dt++;
6856         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6857                 - mddev->resync_mark_cnt;
6858
6859         rt = max_sectors - resync;    /* number of remaining sectors */
6860         sector_div(rt, db/32+1);
6861         rt *= dt;
6862         rt >>= 5;
6863
6864         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6865                    ((unsigned long)rt % 60)/6);
6866
6867         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6868 }
6869
6870 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6871 {
6872         struct list_head *tmp;
6873         loff_t l = *pos;
6874         struct mddev *mddev;
6875
6876         if (l >= 0x10000)
6877                 return NULL;
6878         if (!l--)
6879                 /* header */
6880                 return (void*)1;
6881
6882         spin_lock(&all_mddevs_lock);
6883         list_for_each(tmp,&all_mddevs)
6884                 if (!l--) {
6885                         mddev = list_entry(tmp, struct mddev, all_mddevs);
6886                         mddev_get(mddev);
6887                         spin_unlock(&all_mddevs_lock);
6888                         return mddev;
6889                 }
6890         spin_unlock(&all_mddevs_lock);
6891         if (!l--)
6892                 return (void*)2;/* tail */
6893         return NULL;
6894 }
6895
6896 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6897 {
6898         struct list_head *tmp;
6899         struct mddev *next_mddev, *mddev = v;
6900
6901         ++*pos;
6902         if (v == (void*)2)
6903                 return NULL;
6904
6905         spin_lock(&all_mddevs_lock);
6906         if (v == (void*)1)
6907                 tmp = all_mddevs.next;
6908         else
6909                 tmp = mddev->all_mddevs.next;
6910         if (tmp != &all_mddevs)
6911                 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6912         else {
6913                 next_mddev = (void*)2;
6914                 *pos = 0x10000;
6915         }
6916         spin_unlock(&all_mddevs_lock);
6917
6918         if (v != (void*)1)
6919                 mddev_put(mddev);
6920         return next_mddev;
6921
6922 }
6923
6924 static void md_seq_stop(struct seq_file *seq, void *v)
6925 {
6926         struct mddev *mddev = v;
6927
6928         if (mddev && v != (void*)1 && v != (void*)2)
6929                 mddev_put(mddev);
6930 }
6931
6932 static int md_seq_show(struct seq_file *seq, void *v)
6933 {
6934         struct mddev *mddev = v;
6935         sector_t sectors;
6936         struct md_rdev *rdev;
6937
6938         if (v == (void*)1) {
6939                 struct md_personality *pers;
6940                 seq_printf(seq, "Personalities : ");
6941                 spin_lock(&pers_lock);
6942                 list_for_each_entry(pers, &pers_list, list)
6943                         seq_printf(seq, "[%s] ", pers->name);
6944
6945                 spin_unlock(&pers_lock);
6946                 seq_printf(seq, "\n");
6947                 seq->poll_event = atomic_read(&md_event_count);
6948                 return 0;
6949         }
6950         if (v == (void*)2) {
6951                 status_unused(seq);
6952                 return 0;
6953         }
6954
6955         if (mddev_lock(mddev) < 0)
6956                 return -EINTR;
6957
6958         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6959                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6960                                                 mddev->pers ? "" : "in");
6961                 if (mddev->pers) {
6962                         if (mddev->ro==1)
6963                                 seq_printf(seq, " (read-only)");
6964                         if (mddev->ro==2)
6965                                 seq_printf(seq, " (auto-read-only)");
6966                         seq_printf(seq, " %s", mddev->pers->name);
6967                 }
6968
6969                 sectors = 0;
6970                 rdev_for_each(rdev, mddev) {
6971                         char b[BDEVNAME_SIZE];
6972                         seq_printf(seq, " %s[%d]",
6973                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6974                         if (test_bit(WriteMostly, &rdev->flags))
6975                                 seq_printf(seq, "(W)");
6976                         if (test_bit(Faulty, &rdev->flags)) {
6977                                 seq_printf(seq, "(F)");
6978                                 continue;
6979                         }
6980                         if (rdev->raid_disk < 0)
6981                                 seq_printf(seq, "(S)"); /* spare */
6982                         if (test_bit(Replacement, &rdev->flags))
6983                                 seq_printf(seq, "(R)");
6984                         sectors += rdev->sectors;
6985                 }
6986
6987                 if (!list_empty(&mddev->disks)) {
6988                         if (mddev->pers)
6989                                 seq_printf(seq, "\n      %llu blocks",
6990                                            (unsigned long long)
6991                                            mddev->array_sectors / 2);
6992                         else
6993                                 seq_printf(seq, "\n      %llu blocks",
6994                                            (unsigned long long)sectors / 2);
6995                 }
6996                 if (mddev->persistent) {
6997                         if (mddev->major_version != 0 ||
6998                             mddev->minor_version != 90) {
6999                                 seq_printf(seq," super %d.%d",
7000                                            mddev->major_version,
7001                                            mddev->minor_version);
7002                         }
7003                 } else if (mddev->external)
7004                         seq_printf(seq, " super external:%s",
7005                                    mddev->metadata_type);
7006                 else
7007                         seq_printf(seq, " super non-persistent");
7008
7009                 if (mddev->pers) {
7010                         mddev->pers->status(seq, mddev);
7011                         seq_printf(seq, "\n      ");
7012                         if (mddev->pers->sync_request) {
7013                                 if (mddev->curr_resync > 2) {
7014                                         status_resync(seq, mddev);
7015                                         seq_printf(seq, "\n      ");
7016                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
7017                                         seq_printf(seq, "\tresync=DELAYED\n      ");
7018                                 else if (mddev->recovery_cp < MaxSector)
7019                                         seq_printf(seq, "\tresync=PENDING\n      ");
7020                         }
7021                 } else
7022                         seq_printf(seq, "\n       ");
7023
7024                 bitmap_status(seq, mddev->bitmap);
7025
7026                 seq_printf(seq, "\n");
7027         }
7028         mddev_unlock(mddev);
7029
7030         return 0;
7031 }
7032
7033 static const struct seq_operations md_seq_ops = {
7034         .start  = md_seq_start,
7035         .next   = md_seq_next,
7036         .stop   = md_seq_stop,
7037         .show   = md_seq_show,
7038 };
7039
7040 static int md_seq_open(struct inode *inode, struct file *file)
7041 {
7042         struct seq_file *seq;
7043         int error;
7044
7045         error = seq_open(file, &md_seq_ops);
7046         if (error)
7047                 return error;
7048
7049         seq = file->private_data;
7050         seq->poll_event = atomic_read(&md_event_count);
7051         return error;
7052 }
7053
7054 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7055 {
7056         struct seq_file *seq = filp->private_data;
7057         int mask;
7058
7059         poll_wait(filp, &md_event_waiters, wait);
7060
7061         /* always allow read */
7062         mask = POLLIN | POLLRDNORM;
7063
7064         if (seq->poll_event != atomic_read(&md_event_count))
7065                 mask |= POLLERR | POLLPRI;
7066         return mask;
7067 }
7068
7069 static const struct file_operations md_seq_fops = {
7070         .owner          = THIS_MODULE,
7071         .open           = md_seq_open,
7072         .read           = seq_read,
7073         .llseek         = seq_lseek,
7074         .release        = seq_release_private,
7075         .poll           = mdstat_poll,
7076 };
7077
7078 int register_md_personality(struct md_personality *p)
7079 {
7080         spin_lock(&pers_lock);
7081         list_add_tail(&p->list, &pers_list);
7082         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7083         spin_unlock(&pers_lock);
7084         return 0;
7085 }
7086
7087 int unregister_md_personality(struct md_personality *p)
7088 {
7089         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7090         spin_lock(&pers_lock);
7091         list_del_init(&p->list);
7092         spin_unlock(&pers_lock);
7093         return 0;
7094 }
7095
7096 static int is_mddev_idle(struct mddev *mddev, int init)
7097 {
7098         struct md_rdev * rdev;
7099         int idle;
7100         int curr_events;
7101
7102         idle = 1;
7103         rcu_read_lock();
7104         rdev_for_each_rcu(rdev, mddev) {
7105                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7106                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7107                               (int)part_stat_read(&disk->part0, sectors[1]) -
7108                               atomic_read(&disk->sync_io);
7109                 /* sync IO will cause sync_io to increase before the disk_stats
7110                  * as sync_io is counted when a request starts, and
7111                  * disk_stats is counted when it completes.
7112                  * So resync activity will cause curr_events to be smaller than
7113                  * when there was no such activity.
7114                  * non-sync IO will cause disk_stat to increase without
7115                  * increasing sync_io so curr_events will (eventually)
7116                  * be larger than it was before.  Once it becomes
7117                  * substantially larger, the test below will cause
7118                  * the array to appear non-idle, and resync will slow
7119                  * down.
7120                  * If there is a lot of outstanding resync activity when
7121                  * we set last_event to curr_events, then all that activity
7122                  * completing might cause the array to appear non-idle
7123                  * and resync will be slowed down even though there might
7124                  * not have been non-resync activity.  This will only
7125                  * happen once though.  'last_events' will soon reflect
7126                  * the state where there is little or no outstanding
7127                  * resync requests, and further resync activity will
7128                  * always make curr_events less than last_events.
7129                  *
7130                  */
7131                 if (init || curr_events - rdev->last_events > 64) {
7132                         rdev->last_events = curr_events;
7133                         idle = 0;
7134                 }
7135         }
7136         rcu_read_unlock();
7137         return idle;
7138 }
7139
7140 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7141 {
7142         /* another "blocks" (512byte) blocks have been synced */
7143         atomic_sub(blocks, &mddev->recovery_active);
7144         wake_up(&mddev->recovery_wait);
7145         if (!ok) {
7146                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7147                 md_wakeup_thread(mddev->thread);
7148                 // stop recovery, signal do_sync ....
7149         }
7150 }
7151
7152
7153 /* md_write_start(mddev, bi)
7154  * If we need to update some array metadata (e.g. 'active' flag
7155  * in superblock) before writing, schedule a superblock update
7156  * and wait for it to complete.
7157  */
7158 void md_write_start(struct mddev *mddev, struct bio *bi)
7159 {
7160         int did_change = 0;
7161         if (bio_data_dir(bi) != WRITE)
7162                 return;
7163
7164         BUG_ON(mddev->ro == 1);
7165         if (mddev->ro == 2) {
7166                 /* need to switch to read/write */
7167                 mddev->ro = 0;
7168                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7169                 md_wakeup_thread(mddev->thread);
7170                 md_wakeup_thread(mddev->sync_thread);
7171                 did_change = 1;
7172         }
7173         atomic_inc(&mddev->writes_pending);
7174         if (mddev->safemode == 1)
7175                 mddev->safemode = 0;
7176         if (mddev->in_sync) {
7177                 spin_lock_irq(&mddev->write_lock);
7178                 if (mddev->in_sync) {
7179                         mddev->in_sync = 0;
7180                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7181                         set_bit(MD_CHANGE_PENDING, &mddev->flags);
7182                         md_wakeup_thread(mddev->thread);
7183                         did_change = 1;
7184                 }
7185                 spin_unlock_irq(&mddev->write_lock);
7186         }
7187         if (did_change)
7188                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7189         wait_event(mddev->sb_wait,
7190                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7191 }
7192
7193 void md_write_end(struct mddev *mddev)
7194 {
7195         if (atomic_dec_and_test(&mddev->writes_pending)) {
7196                 if (mddev->safemode == 2)
7197                         md_wakeup_thread(mddev->thread);
7198                 else if (mddev->safemode_delay)
7199                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7200         }
7201 }
7202
7203 /* md_allow_write(mddev)
7204  * Calling this ensures that the array is marked 'active' so that writes
7205  * may proceed without blocking.  It is important to call this before
7206  * attempting a GFP_KERNEL allocation while holding the mddev lock.
7207  * Must be called with mddev_lock held.
7208  *
7209  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7210  * is dropped, so return -EAGAIN after notifying userspace.
7211  */
7212 int md_allow_write(struct mddev *mddev)
7213 {
7214         if (!mddev->pers)
7215                 return 0;
7216         if (mddev->ro)
7217                 return 0;
7218         if (!mddev->pers->sync_request)
7219                 return 0;
7220
7221         spin_lock_irq(&mddev->write_lock);
7222         if (mddev->in_sync) {
7223                 mddev->in_sync = 0;
7224                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7225                 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7226                 if (mddev->safemode_delay &&
7227                     mddev->safemode == 0)
7228                         mddev->safemode = 1;
7229                 spin_unlock_irq(&mddev->write_lock);
7230                 md_update_sb(mddev, 0);
7231                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7232         } else
7233                 spin_unlock_irq(&mddev->write_lock);
7234
7235         if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7236                 return -EAGAIN;
7237         else
7238                 return 0;
7239 }
7240 EXPORT_SYMBOL_GPL(md_allow_write);
7241
7242 #define SYNC_MARKS      10
7243 #define SYNC_MARK_STEP  (3*HZ)
7244 void md_do_sync(struct mddev *mddev)
7245 {
7246         struct mddev *mddev2;
7247         unsigned int currspeed = 0,
7248                  window;
7249         sector_t max_sectors,j, io_sectors;
7250         unsigned long mark[SYNC_MARKS];
7251         sector_t mark_cnt[SYNC_MARKS];
7252         int last_mark,m;
7253         struct list_head *tmp;
7254         sector_t last_check;
7255         int skipped = 0;
7256         struct md_rdev *rdev;
7257         char *desc;
7258         struct blk_plug plug;
7259
7260         /* just incase thread restarts... */
7261         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7262                 return;
7263         if (mddev->ro) /* never try to sync a read-only array */
7264                 return;
7265
7266         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7267                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7268                         desc = "data-check";
7269                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7270                         desc = "requested-resync";
7271                 else
7272                         desc = "resync";
7273         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7274                 desc = "reshape";
7275         else
7276                 desc = "recovery";
7277
7278         /* we overload curr_resync somewhat here.
7279          * 0 == not engaged in resync at all
7280          * 2 == checking that there is no conflict with another sync
7281          * 1 == like 2, but have yielded to allow conflicting resync to
7282          *              commense
7283          * other == active in resync - this many blocks
7284          *
7285          * Before starting a resync we must have set curr_resync to
7286          * 2, and then checked that every "conflicting" array has curr_resync
7287          * less than ours.  When we find one that is the same or higher
7288          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
7289          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7290          * This will mean we have to start checking from the beginning again.
7291          *
7292          */
7293
7294         do {
7295                 mddev->curr_resync = 2;
7296
7297         try_again:
7298                 if (kthread_should_stop())
7299                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7300
7301                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7302                         goto skip;
7303                 for_each_mddev(mddev2, tmp) {
7304                         if (mddev2 == mddev)
7305                                 continue;
7306                         if (!mddev->parallel_resync
7307                         &&  mddev2->curr_resync
7308                         &&  match_mddev_units(mddev, mddev2)) {
7309                                 DEFINE_WAIT(wq);
7310                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
7311                                         /* arbitrarily yield */
7312                                         mddev->curr_resync = 1;
7313                                         wake_up(&resync_wait);
7314                                 }
7315                                 if (mddev > mddev2 && mddev->curr_resync == 1)
7316                                         /* no need to wait here, we can wait the next
7317                                          * time 'round when curr_resync == 2
7318                                          */
7319                                         continue;
7320                                 /* We need to wait 'interruptible' so as not to
7321                                  * contribute to the load average, and not to
7322                                  * be caught by 'softlockup'
7323                                  */
7324                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7325                                 if (!kthread_should_stop() &&
7326                                     mddev2->curr_resync >= mddev->curr_resync) {
7327                                         printk(KERN_INFO "md: delaying %s of %s"
7328                                                " until %s has finished (they"
7329                                                " share one or more physical units)\n",
7330                                                desc, mdname(mddev), mdname(mddev2));
7331                                         mddev_put(mddev2);
7332                                         if (signal_pending(current))
7333                                                 flush_signals(current);
7334                                         schedule();
7335                                         finish_wait(&resync_wait, &wq);
7336                                         goto try_again;
7337                                 }
7338                                 finish_wait(&resync_wait, &wq);
7339                         }
7340                 }
7341         } while (mddev->curr_resync < 2);
7342
7343         j = 0;
7344         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7345                 /* resync follows the size requested by the personality,
7346                  * which defaults to physical size, but can be virtual size
7347                  */
7348                 max_sectors = mddev->resync_max_sectors;
7349                 mddev->resync_mismatches = 0;
7350                 /* we don't use the checkpoint if there's a bitmap */
7351                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7352                         j = mddev->resync_min;
7353                 else if (!mddev->bitmap)
7354                         j = mddev->recovery_cp;
7355
7356         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7357                 max_sectors = mddev->resync_max_sectors;
7358         else {
7359                 /* recovery follows the physical size of devices */
7360                 max_sectors = mddev->dev_sectors;
7361                 j = MaxSector;
7362                 rcu_read_lock();
7363                 rdev_for_each_rcu(rdev, mddev)
7364                         if (rdev->raid_disk >= 0 &&
7365                             !test_bit(Faulty, &rdev->flags) &&
7366                             !test_bit(In_sync, &rdev->flags) &&
7367                             rdev->recovery_offset < j)
7368                                 j = rdev->recovery_offset;
7369                 rcu_read_unlock();
7370         }
7371
7372         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7373         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
7374                 " %d KB/sec/disk.\n", speed_min(mddev));
7375         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7376                "(but not more than %d KB/sec) for %s.\n",
7377                speed_max(mddev), desc);
7378
7379         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7380
7381         io_sectors = 0;
7382         for (m = 0; m < SYNC_MARKS; m++) {
7383                 mark[m] = jiffies;
7384                 mark_cnt[m] = io_sectors;
7385         }
7386         last_mark = 0;
7387         mddev->resync_mark = mark[last_mark];
7388         mddev->resync_mark_cnt = mark_cnt[last_mark];
7389
7390         /*
7391          * Tune reconstruction:
7392          */
7393         window = 32*(PAGE_SIZE/512);
7394         printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7395                 window/2, (unsigned long long)max_sectors/2);
7396
7397         atomic_set(&mddev->recovery_active, 0);
7398         last_check = 0;
7399
7400         if (j>2) {
7401                 printk(KERN_INFO
7402                        "md: resuming %s of %s from checkpoint.\n",
7403                        desc, mdname(mddev));
7404                 mddev->curr_resync = j;
7405         }
7406         mddev->curr_resync_completed = j;
7407
7408         blk_start_plug(&plug);
7409         while (j < max_sectors) {
7410                 sector_t sectors;
7411
7412                 skipped = 0;
7413
7414                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7415                     ((mddev->curr_resync > mddev->curr_resync_completed &&
7416                       (mddev->curr_resync - mddev->curr_resync_completed)
7417                       > (max_sectors >> 4)) ||
7418                      (j - mddev->curr_resync_completed)*2
7419                      >= mddev->resync_max - mddev->curr_resync_completed
7420                             )) {
7421                         /* time to update curr_resync_completed */
7422                         wait_event(mddev->recovery_wait,
7423                                    atomic_read(&mddev->recovery_active) == 0);
7424                         mddev->curr_resync_completed = j;
7425                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7426                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7427                 }
7428
7429                 while (j >= mddev->resync_max && !kthread_should_stop()) {
7430                         /* As this condition is controlled by user-space,
7431                          * we can block indefinitely, so use '_interruptible'
7432                          * to avoid triggering warnings.
7433                          */
7434                         flush_signals(current); /* just in case */
7435                         wait_event_interruptible(mddev->recovery_wait,
7436                                                  mddev->resync_max > j
7437                                                  || kthread_should_stop());
7438                 }
7439
7440                 if (kthread_should_stop())
7441                         goto interrupted;
7442
7443                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7444                                                   currspeed < speed_min(mddev));
7445                 if (sectors == 0) {
7446                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7447                         goto out;
7448                 }
7449
7450                 if (!skipped) { /* actual IO requested */
7451                         io_sectors += sectors;
7452                         atomic_add(sectors, &mddev->recovery_active);
7453                 }
7454
7455                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7456                         break;
7457
7458                 j += sectors;
7459                 if (j>1) mddev->curr_resync = j;
7460                 mddev->curr_mark_cnt = io_sectors;
7461                 if (last_check == 0)
7462                         /* this is the earliest that rebuild will be
7463                          * visible in /proc/mdstat
7464                          */
7465                         md_new_event(mddev);
7466
7467                 if (last_check + window > io_sectors || j == max_sectors)
7468                         continue;
7469
7470                 last_check = io_sectors;
7471         repeat:
7472                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7473                         /* step marks */
7474                         int next = (last_mark+1) % SYNC_MARKS;
7475
7476                         mddev->resync_mark = mark[next];
7477                         mddev->resync_mark_cnt = mark_cnt[next];
7478                         mark[next] = jiffies;
7479                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7480                         last_mark = next;
7481                 }
7482
7483
7484                 if (kthread_should_stop())
7485                         goto interrupted;
7486
7487
7488                 /*
7489                  * this loop exits only if either when we are slower than
7490                  * the 'hard' speed limit, or the system was IO-idle for
7491                  * a jiffy.
7492                  * the system might be non-idle CPU-wise, but we only care
7493                  * about not overloading the IO subsystem. (things like an
7494                  * e2fsck being done on the RAID array should execute fast)
7495                  */
7496                 cond_resched();
7497
7498                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7499                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
7500
7501                 if (currspeed > speed_min(mddev)) {
7502                         if ((currspeed > speed_max(mddev)) ||
7503                                         !is_mddev_idle(mddev, 0)) {
7504                                 msleep(500);
7505                                 goto repeat;
7506                         }
7507                 }
7508         }
7509         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7510         /*
7511          * this also signals 'finished resyncing' to md_stop
7512          */
7513  out:
7514         blk_finish_plug(&plug);
7515         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7516
7517         /* tell personality that we are finished */
7518         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7519
7520         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7521             mddev->curr_resync > 2) {
7522                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7523                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7524                                 if (mddev->curr_resync >= mddev->recovery_cp) {
7525                                         printk(KERN_INFO
7526                                                "md: checkpointing %s of %s.\n",
7527                                                desc, mdname(mddev));
7528                                         mddev->recovery_cp =
7529                                                 mddev->curr_resync_completed;
7530                                 }
7531                         } else
7532                                 mddev->recovery_cp = MaxSector;
7533                 } else {
7534                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7535                                 mddev->curr_resync = MaxSector;
7536                         rcu_read_lock();
7537                         rdev_for_each_rcu(rdev, mddev)
7538                                 if (rdev->raid_disk >= 0 &&
7539                                     mddev->delta_disks >= 0 &&
7540                                     !test_bit(Faulty, &rdev->flags) &&
7541                                     !test_bit(In_sync, &rdev->flags) &&
7542                                     rdev->recovery_offset < mddev->curr_resync)
7543                                         rdev->recovery_offset = mddev->curr_resync;
7544                         rcu_read_unlock();
7545                 }
7546         }
7547  skip:
7548         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7549
7550         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7551                 /* We completed so min/max setting can be forgotten if used. */
7552                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7553                         mddev->resync_min = 0;
7554                 mddev->resync_max = MaxSector;
7555         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7556                 mddev->resync_min = mddev->curr_resync_completed;
7557         mddev->curr_resync = 0;
7558         wake_up(&resync_wait);
7559         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7560         md_wakeup_thread(mddev->thread);
7561         return;
7562
7563  interrupted:
7564         /*
7565          * got a signal, exit.
7566          */
7567         printk(KERN_INFO
7568                "md: md_do_sync() got signal ... exiting\n");
7569         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7570         goto out;
7571
7572 }
7573 EXPORT_SYMBOL_GPL(md_do_sync);
7574
7575 static int remove_and_add_spares(struct mddev *mddev)
7576 {
7577         struct md_rdev *rdev;
7578         int spares = 0;
7579         int removed = 0;
7580
7581         mddev->curr_resync_completed = 0;
7582
7583         rdev_for_each(rdev, mddev)
7584                 if (rdev->raid_disk >= 0 &&
7585                     !test_bit(Blocked, &rdev->flags) &&
7586                     (test_bit(Faulty, &rdev->flags) ||
7587                      ! test_bit(In_sync, &rdev->flags)) &&
7588                     atomic_read(&rdev->nr_pending)==0) {
7589                         if (mddev->pers->hot_remove_disk(
7590                                     mddev, rdev) == 0) {
7591                                 sysfs_unlink_rdev(mddev, rdev);
7592                                 rdev->raid_disk = -1;
7593                                 removed++;
7594                         }
7595                 }
7596         if (removed)
7597                 sysfs_notify(&mddev->kobj, NULL,
7598                              "degraded");
7599
7600
7601         rdev_for_each(rdev, mddev) {
7602                 if (rdev->raid_disk >= 0 &&
7603                     !test_bit(In_sync, &rdev->flags) &&
7604                     !test_bit(Faulty, &rdev->flags))
7605                         spares++;
7606                 if (rdev->raid_disk < 0
7607                     && !test_bit(Faulty, &rdev->flags)) {
7608                         rdev->recovery_offset = 0;
7609                         if (mddev->pers->
7610                             hot_add_disk(mddev, rdev) == 0) {
7611                                 if (sysfs_link_rdev(mddev, rdev))
7612                                         /* failure here is OK */;
7613                                 spares++;
7614                                 md_new_event(mddev);
7615                                 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7616                         }
7617                 }
7618         }
7619         return spares;
7620 }
7621
7622 static void reap_sync_thread(struct mddev *mddev)
7623 {
7624         struct md_rdev *rdev;
7625
7626         /* resync has finished, collect result */
7627         md_unregister_thread(&mddev->sync_thread);
7628         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7629             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7630                 /* success...*/
7631                 /* activate any spares */
7632                 if (mddev->pers->spare_active(mddev))
7633                         sysfs_notify(&mddev->kobj, NULL,
7634                                      "degraded");
7635         }
7636         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7637             mddev->pers->finish_reshape)
7638                 mddev->pers->finish_reshape(mddev);
7639
7640         /* If array is no-longer degraded, then any saved_raid_disk
7641          * information must be scrapped.  Also if any device is now
7642          * In_sync we must scrape the saved_raid_disk for that device
7643          * do the superblock for an incrementally recovered device
7644          * written out.
7645          */
7646         rdev_for_each(rdev, mddev)
7647                 if (!mddev->degraded ||
7648                     test_bit(In_sync, &rdev->flags))
7649                         rdev->saved_raid_disk = -1;
7650
7651         md_update_sb(mddev, 1);
7652         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7653         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7654         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7655         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7656         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7657         /* flag recovery needed just to double check */
7658         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7659         sysfs_notify_dirent_safe(mddev->sysfs_action);
7660         md_new_event(mddev);
7661         if (mddev->event_work.func)
7662                 queue_work(md_misc_wq, &mddev->event_work);
7663 }
7664
7665 /*
7666  * This routine is regularly called by all per-raid-array threads to
7667  * deal with generic issues like resync and super-block update.
7668  * Raid personalities that don't have a thread (linear/raid0) do not
7669  * need this as they never do any recovery or update the superblock.
7670  *
7671  * It does not do any resync itself, but rather "forks" off other threads
7672  * to do that as needed.
7673  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7674  * "->recovery" and create a thread at ->sync_thread.
7675  * When the thread finishes it sets MD_RECOVERY_DONE
7676  * and wakeups up this thread which will reap the thread and finish up.
7677  * This thread also removes any faulty devices (with nr_pending == 0).
7678  *
7679  * The overall approach is:
7680  *  1/ if the superblock needs updating, update it.
7681  *  2/ If a recovery thread is running, don't do anything else.
7682  *  3/ If recovery has finished, clean up, possibly marking spares active.
7683  *  4/ If there are any faulty devices, remove them.
7684  *  5/ If array is degraded, try to add spares devices
7685  *  6/ If array has spares or is not in-sync, start a resync thread.
7686  */
7687 void md_check_recovery(struct mddev *mddev)
7688 {
7689         if (mddev->suspended)
7690                 return;
7691
7692         if (mddev->bitmap)
7693                 bitmap_daemon_work(mddev);
7694
7695         if (signal_pending(current)) {
7696                 if (mddev->pers->sync_request && !mddev->external) {
7697                         printk(KERN_INFO "md: %s in immediate safe mode\n",
7698                                mdname(mddev));
7699                         mddev->safemode = 2;
7700                 }
7701                 flush_signals(current);
7702         }
7703
7704         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7705                 return;
7706         if ( ! (
7707                 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7708                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7709                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7710                 (mddev->external == 0 && mddev->safemode == 1) ||
7711                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7712                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7713                 ))
7714                 return;
7715
7716         if (mddev_trylock(mddev)) {
7717                 int spares = 0;
7718
7719                 if (mddev->ro) {
7720                         /* Only thing we do on a ro array is remove
7721                          * failed devices.
7722                          */
7723                         struct md_rdev *rdev;
7724                         rdev_for_each(rdev, mddev)
7725                                 if (rdev->raid_disk >= 0 &&
7726                                     !test_bit(Blocked, &rdev->flags) &&
7727                                     test_bit(Faulty, &rdev->flags) &&
7728                                     atomic_read(&rdev->nr_pending)==0) {
7729                                         if (mddev->pers->hot_remove_disk(
7730                                                     mddev, rdev) == 0) {
7731                                                 sysfs_unlink_rdev(mddev, rdev);
7732                                                 rdev->raid_disk = -1;
7733                                         }
7734                                 }
7735                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7736                         goto unlock;
7737                 }
7738
7739                 if (!mddev->external) {
7740                         int did_change = 0;
7741                         spin_lock_irq(&mddev->write_lock);
7742                         if (mddev->safemode &&
7743                             !atomic_read(&mddev->writes_pending) &&
7744                             !mddev->in_sync &&
7745                             mddev->recovery_cp == MaxSector) {
7746                                 mddev->in_sync = 1;
7747                                 did_change = 1;
7748                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7749                         }
7750                         if (mddev->safemode == 1)
7751                                 mddev->safemode = 0;
7752                         spin_unlock_irq(&mddev->write_lock);
7753                         if (did_change)
7754                                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7755                 }
7756
7757                 if (mddev->flags)
7758                         md_update_sb(mddev, 0);
7759
7760                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7761                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7762                         /* resync/recovery still happening */
7763                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7764                         goto unlock;
7765                 }
7766                 if (mddev->sync_thread) {
7767                         reap_sync_thread(mddev);
7768                         goto unlock;
7769                 }
7770                 /* Set RUNNING before clearing NEEDED to avoid
7771                  * any transients in the value of "sync_action".
7772                  */
7773                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7774                 /* Clear some bits that don't mean anything, but
7775                  * might be left set
7776                  */
7777                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7778                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7779
7780                 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7781                     test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7782                         goto unlock;
7783                 /* no recovery is running.
7784                  * remove any failed drives, then
7785                  * add spares if possible.
7786                  * Spare are also removed and re-added, to allow
7787                  * the personality to fail the re-add.
7788                  */
7789
7790                 if (mddev->reshape_position != MaxSector) {
7791                         if (mddev->pers->check_reshape == NULL ||
7792                             mddev->pers->check_reshape(mddev) != 0)
7793                                 /* Cannot proceed */
7794                                 goto unlock;
7795                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7796                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7797                 } else if ((spares = remove_and_add_spares(mddev))) {
7798                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7799                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7800                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7801                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7802                 } else if (mddev->recovery_cp < MaxSector) {
7803                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7804                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7805                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7806                         /* nothing to be done ... */
7807                         goto unlock;
7808
7809                 if (mddev->pers->sync_request) {
7810                         if (spares) {
7811                                 /* We are adding a device or devices to an array
7812                                  * which has the bitmap stored on all devices.
7813                                  * So make sure all bitmap pages get written
7814                                  */
7815                                 bitmap_write_all(mddev->bitmap);
7816                         }
7817                         mddev->sync_thread = md_register_thread(md_do_sync,
7818                                                                 mddev,
7819                                                                 "resync");
7820                         if (!mddev->sync_thread) {
7821                                 printk(KERN_ERR "%s: could not start resync"
7822                                         " thread...\n",
7823                                         mdname(mddev));
7824                                 /* leave the spares where they are, it shouldn't hurt */
7825                                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7826                                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7827                                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7828                                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7829                                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7830                         } else
7831                                 md_wakeup_thread(mddev->sync_thread);
7832                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7833                         md_new_event(mddev);
7834                 }
7835         unlock:
7836                 if (!mddev->sync_thread) {
7837                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7838                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7839                                                &mddev->recovery))
7840                                 if (mddev->sysfs_action)
7841                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7842                 }
7843                 mddev_unlock(mddev);
7844         }
7845 }
7846
7847 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7848 {
7849         sysfs_notify_dirent_safe(rdev->sysfs_state);
7850         wait_event_timeout(rdev->blocked_wait,
7851                            !test_bit(Blocked, &rdev->flags) &&
7852                            !test_bit(BlockedBadBlocks, &rdev->flags),
7853                            msecs_to_jiffies(5000));
7854         rdev_dec_pending(rdev, mddev);
7855 }
7856 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7857
7858 void md_finish_reshape(struct mddev *mddev)
7859 {
7860         /* called be personality module when reshape completes. */
7861         struct md_rdev *rdev;
7862
7863         rdev_for_each(rdev, mddev) {
7864                 if (rdev->data_offset > rdev->new_data_offset)
7865                         rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7866                 else
7867                         rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7868                 rdev->data_offset = rdev->new_data_offset;
7869         }
7870 }
7871 EXPORT_SYMBOL(md_finish_reshape);
7872
7873 /* Bad block management.
7874  * We can record which blocks on each device are 'bad' and so just
7875  * fail those blocks, or that stripe, rather than the whole device.
7876  * Entries in the bad-block table are 64bits wide.  This comprises:
7877  * Length of bad-range, in sectors: 0-511 for lengths 1-512
7878  * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7879  *  A 'shift' can be set so that larger blocks are tracked and
7880  *  consequently larger devices can be covered.
7881  * 'Acknowledged' flag - 1 bit. - the most significant bit.
7882  *
7883  * Locking of the bad-block table uses a seqlock so md_is_badblock
7884  * might need to retry if it is very unlucky.
7885  * We will sometimes want to check for bad blocks in a bi_end_io function,
7886  * so we use the write_seqlock_irq variant.
7887  *
7888  * When looking for a bad block we specify a range and want to
7889  * know if any block in the range is bad.  So we binary-search
7890  * to the last range that starts at-or-before the given endpoint,
7891  * (or "before the sector after the target range")
7892  * then see if it ends after the given start.
7893  * We return
7894  *  0 if there are no known bad blocks in the range
7895  *  1 if there are known bad block which are all acknowledged
7896  * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7897  * plus the start/length of the first bad section we overlap.
7898  */
7899 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7900                    sector_t *first_bad, int *bad_sectors)
7901 {
7902         int hi;
7903         int lo = 0;
7904         u64 *p = bb->page;
7905         int rv = 0;
7906         sector_t target = s + sectors;
7907         unsigned seq;
7908
7909         if (bb->shift > 0) {
7910                 /* round the start down, and the end up */
7911                 s >>= bb->shift;
7912                 target += (1<<bb->shift) - 1;
7913                 target >>= bb->shift;
7914                 sectors = target - s;
7915         }
7916         /* 'target' is now the first block after the bad range */
7917
7918 retry:
7919         seq = read_seqbegin(&bb->lock);
7920
7921         hi = bb->count;
7922
7923         /* Binary search between lo and hi for 'target'
7924          * i.e. for the last range that starts before 'target'
7925          */
7926         /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7927          * are known not to be the last range before target.
7928          * VARIANT: hi-lo is the number of possible
7929          * ranges, and decreases until it reaches 1
7930          */
7931         while (hi - lo > 1) {
7932                 int mid = (lo + hi) / 2;
7933                 sector_t a = BB_OFFSET(p[mid]);
7934                 if (a < target)
7935                         /* This could still be the one, earlier ranges
7936                          * could not. */
7937                         lo = mid;
7938                 else
7939                         /* This and later ranges are definitely out. */
7940                         hi = mid;
7941         }
7942         /* 'lo' might be the last that started before target, but 'hi' isn't */
7943         if (hi > lo) {
7944                 /* need to check all range that end after 's' to see if
7945                  * any are unacknowledged.
7946                  */
7947                 while (lo >= 0 &&
7948                        BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7949                         if (BB_OFFSET(p[lo]) < target) {
7950                                 /* starts before the end, and finishes after
7951                                  * the start, so they must overlap
7952                                  */
7953                                 if (rv != -1 && BB_ACK(p[lo]))
7954                                         rv = 1;
7955                                 else
7956                                         rv = -1;
7957                                 *first_bad = BB_OFFSET(p[lo]);
7958                                 *bad_sectors = BB_LEN(p[lo]);
7959                         }
7960                         lo--;
7961                 }
7962         }
7963
7964         if (read_seqretry(&bb->lock, seq))
7965                 goto retry;
7966
7967         return rv;
7968 }
7969 EXPORT_SYMBOL_GPL(md_is_badblock);
7970
7971 /*
7972  * Add a range of bad blocks to the table.
7973  * This might extend the table, or might contract it
7974  * if two adjacent ranges can be merged.
7975  * We binary-search to find the 'insertion' point, then
7976  * decide how best to handle it.
7977  */
7978 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7979                             int acknowledged)
7980 {
7981         u64 *p;
7982         int lo, hi;
7983         int rv = 1;
7984
7985         if (bb->shift < 0)
7986                 /* badblocks are disabled */
7987                 return 0;
7988
7989         if (bb->shift) {
7990                 /* round the start down, and the end up */
7991                 sector_t next = s + sectors;
7992                 s >>= bb->shift;
7993                 next += (1<<bb->shift) - 1;
7994                 next >>= bb->shift;
7995                 sectors = next - s;
7996         }
7997
7998         write_seqlock_irq(&bb->lock);
7999
8000         p = bb->page;
8001         lo = 0;
8002         hi = bb->count;
8003         /* Find the last range that starts at-or-before 's' */
8004         while (hi - lo > 1) {
8005                 int mid = (lo + hi) / 2;
8006                 sector_t a = BB_OFFSET(p[mid]);
8007                 if (a <= s)
8008                         lo = mid;
8009                 else
8010                         hi = mid;
8011         }
8012         if (hi > lo && BB_OFFSET(p[lo]) > s)
8013                 hi = lo;
8014
8015         if (hi > lo) {
8016                 /* we found a range that might merge with the start
8017                  * of our new range
8018                  */
8019                 sector_t a = BB_OFFSET(p[lo]);
8020                 sector_t e = a + BB_LEN(p[lo]);
8021                 int ack = BB_ACK(p[lo]);
8022                 if (e >= s) {
8023                         /* Yes, we can merge with a previous range */
8024                         if (s == a && s + sectors >= e)
8025                                 /* new range covers old */
8026                                 ack = acknowledged;
8027                         else
8028                                 ack = ack && acknowledged;
8029
8030                         if (e < s + sectors)
8031                                 e = s + sectors;
8032                         if (e - a <= BB_MAX_LEN) {
8033                                 p[lo] = BB_MAKE(a, e-a, ack);
8034                                 s = e;
8035                         } else {
8036                                 /* does not all fit in one range,
8037                                  * make p[lo] maximal
8038                                  */
8039                                 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8040                                         p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8041                                 s = a + BB_MAX_LEN;
8042                         }
8043                         sectors = e - s;
8044                 }
8045         }
8046         if (sectors && hi < bb->count) {
8047                 /* 'hi' points to the first range that starts after 's'.
8048                  * Maybe we can merge with the start of that range */
8049                 sector_t a = BB_OFFSET(p[hi]);
8050                 sector_t e = a + BB_LEN(p[hi]);
8051                 int ack = BB_ACK(p[hi]);
8052                 if (a <= s + sectors) {
8053                         /* merging is possible */
8054                         if (e <= s + sectors) {
8055                                 /* full overlap */
8056                                 e = s + sectors;
8057                                 ack = acknowledged;
8058                         } else
8059                                 ack = ack && acknowledged;
8060
8061                         a = s;
8062                         if (e - a <= BB_MAX_LEN) {
8063                                 p[hi] = BB_MAKE(a, e-a, ack);
8064                                 s = e;
8065                         } else {
8066                                 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8067                                 s = a + BB_MAX_LEN;
8068                         }
8069                         sectors = e - s;
8070                         lo = hi;
8071                         hi++;
8072                 }
8073         }
8074         if (sectors == 0 && hi < bb->count) {
8075                 /* we might be able to combine lo and hi */
8076                 /* Note: 's' is at the end of 'lo' */
8077                 sector_t a = BB_OFFSET(p[hi]);
8078                 int lolen = BB_LEN(p[lo]);
8079                 int hilen = BB_LEN(p[hi]);
8080                 int newlen = lolen + hilen - (s - a);
8081                 if (s >= a && newlen < BB_MAX_LEN) {
8082                         /* yes, we can combine them */
8083                         int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8084                         p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8085                         memmove(p + hi, p + hi + 1,
8086                                 (bb->count - hi - 1) * 8);
8087                         bb->count--;
8088                 }
8089         }
8090         while (sectors) {
8091                 /* didn't merge (it all).
8092                  * Need to add a range just before 'hi' */
8093                 if (bb->count >= MD_MAX_BADBLOCKS) {
8094                         /* No room for more */
8095                         rv = 0;
8096                         break;
8097                 } else {
8098                         int this_sectors = sectors;
8099                         memmove(p + hi + 1, p + hi,
8100                                 (bb->count - hi) * 8);
8101                         bb->count++;
8102
8103                         if (this_sectors > BB_MAX_LEN)
8104                                 this_sectors = BB_MAX_LEN;
8105                         p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8106                         sectors -= this_sectors;
8107                         s += this_sectors;
8108                 }
8109         }
8110
8111         bb->changed = 1;
8112         if (!acknowledged)
8113                 bb->unacked_exist = 1;
8114         write_sequnlock_irq(&bb->lock);
8115
8116         return rv;
8117 }
8118
8119 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8120                        int is_new)
8121 {
8122         int rv;
8123         if (is_new)
8124                 s += rdev->new_data_offset;
8125         else
8126                 s += rdev->data_offset;
8127         rv = md_set_badblocks(&rdev->badblocks,
8128                               s, sectors, 0);
8129         if (rv) {
8130                 /* Make sure they get written out promptly */
8131                 sysfs_notify_dirent_safe(rdev->sysfs_state);
8132                 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8133                 md_wakeup_thread(rdev->mddev->thread);
8134         }
8135         return rv;
8136 }
8137 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8138
8139 /*
8140  * Remove a range of bad blocks from the table.
8141  * This may involve extending the table if we spilt a region,
8142  * but it must not fail.  So if the table becomes full, we just
8143  * drop the remove request.
8144  */
8145 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8146 {
8147         u64 *p;
8148         int lo, hi;
8149         sector_t target = s + sectors;
8150         int rv = 0;
8151
8152         if (bb->shift > 0) {
8153                 /* When clearing we round the start up and the end down.
8154                  * This should not matter as the shift should align with
8155                  * the block size and no rounding should ever be needed.
8156                  * However it is better the think a block is bad when it
8157                  * isn't than to think a block is not bad when it is.
8158                  */
8159                 s += (1<<bb->shift) - 1;
8160                 s >>= bb->shift;
8161                 target >>= bb->shift;
8162                 sectors = target - s;
8163         }
8164
8165         write_seqlock_irq(&bb->lock);
8166
8167         p = bb->page;
8168         lo = 0;
8169         hi = bb->count;
8170         /* Find the last range that starts before 'target' */
8171         while (hi - lo > 1) {
8172                 int mid = (lo + hi) / 2;
8173                 sector_t a = BB_OFFSET(p[mid]);
8174                 if (a < target)
8175                         lo = mid;
8176                 else
8177                         hi = mid;
8178         }
8179         if (hi > lo) {
8180                 /* p[lo] is the last range that could overlap the
8181                  * current range.  Earlier ranges could also overlap,
8182                  * but only this one can overlap the end of the range.
8183                  */
8184                 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8185                         /* Partial overlap, leave the tail of this range */
8186                         int ack = BB_ACK(p[lo]);
8187                         sector_t a = BB_OFFSET(p[lo]);
8188                         sector_t end = a + BB_LEN(p[lo]);
8189
8190                         if (a < s) {
8191                                 /* we need to split this range */
8192                                 if (bb->count >= MD_MAX_BADBLOCKS) {
8193                                         rv = 0;
8194                                         goto out;
8195                                 }
8196                                 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8197                                 bb->count++;
8198                                 p[lo] = BB_MAKE(a, s-a, ack);
8199                                 lo++;
8200                         }
8201                         p[lo] = BB_MAKE(target, end - target, ack);
8202                         /* there is no longer an overlap */
8203                         hi = lo;
8204                         lo--;
8205                 }
8206                 while (lo >= 0 &&
8207                        BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8208                         /* This range does overlap */
8209                         if (BB_OFFSET(p[lo]) < s) {
8210                                 /* Keep the early parts of this range. */
8211                                 int ack = BB_ACK(p[lo]);
8212                                 sector_t start = BB_OFFSET(p[lo]);
8213                                 p[lo] = BB_MAKE(start, s - start, ack);
8214                                 /* now low doesn't overlap, so.. */
8215                                 break;
8216                         }
8217                         lo--;
8218                 }
8219                 /* 'lo' is strictly before, 'hi' is strictly after,
8220                  * anything between needs to be discarded
8221                  */
8222                 if (hi - lo > 1) {
8223                         memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8224                         bb->count -= (hi - lo - 1);
8225                 }
8226         }
8227
8228         bb->changed = 1;
8229 out:
8230         write_sequnlock_irq(&bb->lock);
8231         return rv;
8232 }
8233
8234 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8235                          int is_new)
8236 {
8237         if (is_new)
8238                 s += rdev->new_data_offset;
8239         else
8240                 s += rdev->data_offset;
8241         return md_clear_badblocks(&rdev->badblocks,
8242                                   s, sectors);
8243 }
8244 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8245
8246 /*
8247  * Acknowledge all bad blocks in a list.
8248  * This only succeeds if ->changed is clear.  It is used by
8249  * in-kernel metadata updates
8250  */
8251 void md_ack_all_badblocks(struct badblocks *bb)
8252 {
8253         if (bb->page == NULL || bb->changed)
8254                 /* no point even trying */
8255                 return;
8256         write_seqlock_irq(&bb->lock);
8257
8258         if (bb->changed == 0 && bb->unacked_exist) {
8259                 u64 *p = bb->page;
8260                 int i;
8261                 for (i = 0; i < bb->count ; i++) {
8262                         if (!BB_ACK(p[i])) {
8263                                 sector_t start = BB_OFFSET(p[i]);
8264                                 int len = BB_LEN(p[i]);
8265                                 p[i] = BB_MAKE(start, len, 1);
8266                         }
8267                 }
8268                 bb->unacked_exist = 0;
8269         }
8270         write_sequnlock_irq(&bb->lock);
8271 }
8272 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8273
8274 /* sysfs access to bad-blocks list.
8275  * We present two files.
8276  * 'bad-blocks' lists sector numbers and lengths of ranges that
8277  *    are recorded as bad.  The list is truncated to fit within
8278  *    the one-page limit of sysfs.
8279  *    Writing "sector length" to this file adds an acknowledged
8280  *    bad block list.
8281  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8282  *    been acknowledged.  Writing to this file adds bad blocks
8283  *    without acknowledging them.  This is largely for testing.
8284  */
8285
8286 static ssize_t
8287 badblocks_show(struct badblocks *bb, char *page, int unack)
8288 {
8289         size_t len;
8290         int i;
8291         u64 *p = bb->page;
8292         unsigned seq;
8293
8294         if (bb->shift < 0)
8295                 return 0;
8296
8297 retry:
8298         seq = read_seqbegin(&bb->lock);
8299
8300         len = 0;
8301         i = 0;
8302
8303         while (len < PAGE_SIZE && i < bb->count) {
8304                 sector_t s = BB_OFFSET(p[i]);
8305                 unsigned int length = BB_LEN(p[i]);
8306                 int ack = BB_ACK(p[i]);
8307                 i++;
8308
8309                 if (unack && ack)
8310                         continue;
8311
8312                 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8313                                 (unsigned long long)s << bb->shift,
8314                                 length << bb->shift);
8315         }
8316         if (unack && len == 0)
8317                 bb->unacked_exist = 0;
8318
8319         if (read_seqretry(&bb->lock, seq))
8320                 goto retry;
8321
8322         return len;
8323 }
8324
8325 #define DO_DEBUG 1
8326
8327 static ssize_t
8328 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8329 {
8330         unsigned long long sector;
8331         int length;
8332         char newline;
8333 #ifdef DO_DEBUG
8334         /* Allow clearing via sysfs *only* for testing/debugging.
8335          * Normally only a successful write may clear a badblock
8336          */
8337         int clear = 0;
8338         if (page[0] == '-') {
8339                 clear = 1;
8340                 page++;
8341         }
8342 #endif /* DO_DEBUG */
8343
8344         switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8345         case 3:
8346                 if (newline != '\n')
8347                         return -EINVAL;
8348         case 2:
8349                 if (length <= 0)
8350                         return -EINVAL;
8351                 break;
8352         default:
8353                 return -EINVAL;
8354         }
8355
8356 #ifdef DO_DEBUG
8357         if (clear) {
8358                 md_clear_badblocks(bb, sector, length);
8359                 return len;
8360         }
8361 #endif /* DO_DEBUG */
8362         if (md_set_badblocks(bb, sector, length, !unack))
8363                 return len;
8364         else
8365                 return -ENOSPC;
8366 }
8367
8368 static int md_notify_reboot(struct notifier_block *this,
8369                             unsigned long code, void *x)
8370 {
8371         struct list_head *tmp;
8372         struct mddev *mddev;
8373         int need_delay = 0;
8374
8375         for_each_mddev(mddev, tmp) {
8376                 if (mddev_trylock(mddev)) {
8377                         if (mddev->pers)
8378                                 __md_stop_writes(mddev);
8379                         mddev->safemode = 2;
8380                         mddev_unlock(mddev);
8381                 }
8382                 need_delay = 1;
8383         }
8384         /*
8385          * certain more exotic SCSI devices are known to be
8386          * volatile wrt too early system reboots. While the
8387          * right place to handle this issue is the given
8388          * driver, we do want to have a safe RAID driver ...
8389          */
8390         if (need_delay)
8391                 mdelay(1000*1);
8392
8393         return NOTIFY_DONE;
8394 }
8395
8396 static struct notifier_block md_notifier = {
8397         .notifier_call  = md_notify_reboot,
8398         .next           = NULL,
8399         .priority       = INT_MAX, /* before any real devices */
8400 };
8401
8402 static void md_geninit(void)
8403 {
8404         pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8405
8406         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8407 }
8408
8409 static int __init md_init(void)
8410 {
8411         int ret = -ENOMEM;
8412
8413         md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8414         if (!md_wq)
8415                 goto err_wq;
8416
8417         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8418         if (!md_misc_wq)
8419                 goto err_misc_wq;
8420
8421         if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8422                 goto err_md;
8423
8424         if ((ret = register_blkdev(0, "mdp")) < 0)
8425                 goto err_mdp;
8426         mdp_major = ret;
8427
8428         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8429                             md_probe, NULL, NULL);
8430         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8431                             md_probe, NULL, NULL);
8432
8433         register_reboot_notifier(&md_notifier);
8434         raid_table_header = register_sysctl_table(raid_root_table);
8435
8436         md_geninit();
8437         return 0;
8438
8439 err_mdp:
8440         unregister_blkdev(MD_MAJOR, "md");
8441 err_md:
8442         destroy_workqueue(md_misc_wq);
8443 err_misc_wq:
8444         destroy_workqueue(md_wq);
8445 err_wq:
8446         return ret;
8447 }
8448
8449 #ifndef MODULE
8450
8451 /*
8452  * Searches all registered partitions for autorun RAID arrays
8453  * at boot time.
8454  */
8455
8456 static LIST_HEAD(all_detected_devices);
8457 struct detected_devices_node {
8458         struct list_head list;
8459         dev_t dev;
8460 };
8461
8462 void md_autodetect_dev(dev_t dev)
8463 {
8464         struct detected_devices_node *node_detected_dev;
8465
8466         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8467         if (node_detected_dev) {
8468                 node_detected_dev->dev = dev;
8469                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8470         } else {
8471                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8472                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8473         }
8474 }
8475
8476
8477 static void autostart_arrays(int part)
8478 {
8479         struct md_rdev *rdev;
8480         struct detected_devices_node *node_detected_dev;
8481         dev_t dev;
8482         int i_scanned, i_passed;
8483
8484         i_scanned = 0;
8485         i_passed = 0;
8486
8487         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8488
8489         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8490                 i_scanned++;
8491                 node_detected_dev = list_entry(all_detected_devices.next,
8492                                         struct detected_devices_node, list);
8493                 list_del(&node_detected_dev->list);
8494                 dev = node_detected_dev->dev;
8495                 kfree(node_detected_dev);
8496                 rdev = md_import_device(dev,0, 90);
8497                 if (IS_ERR(rdev))
8498                         continue;
8499
8500                 if (test_bit(Faulty, &rdev->flags)) {
8501                         MD_BUG();
8502                         continue;
8503                 }
8504                 set_bit(AutoDetected, &rdev->flags);
8505                 list_add(&rdev->same_set, &pending_raid_disks);
8506                 i_passed++;
8507         }
8508
8509         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8510                                                 i_scanned, i_passed);
8511
8512         autorun_devices(part);
8513 }
8514
8515 #endif /* !MODULE */
8516
8517 static __exit void md_exit(void)
8518 {
8519         struct mddev *mddev;
8520         struct list_head *tmp;
8521
8522         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8523         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8524
8525         unregister_blkdev(MD_MAJOR,"md");
8526         unregister_blkdev(mdp_major, "mdp");
8527         unregister_reboot_notifier(&md_notifier);
8528         unregister_sysctl_table(raid_table_header);
8529         remove_proc_entry("mdstat", NULL);
8530         for_each_mddev(mddev, tmp) {
8531                 export_array(mddev);
8532                 mddev->hold_active = 0;
8533         }
8534         destroy_workqueue(md_misc_wq);
8535         destroy_workqueue(md_wq);
8536 }
8537
8538 subsys_initcall(md_init);
8539 module_exit(md_exit)
8540
8541 static int get_ro(char *buffer, struct kernel_param *kp)
8542 {
8543         return sprintf(buffer, "%d", start_readonly);
8544 }
8545 static int set_ro(const char *val, struct kernel_param *kp)
8546 {
8547         char *e;
8548         int num = simple_strtoul(val, &e, 10);
8549         if (*val && (*e == '\0' || *e == '\n')) {
8550                 start_readonly = num;
8551                 return 0;
8552         }
8553         return -EINVAL;
8554 }
8555
8556 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8557 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8558
8559 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8560
8561 EXPORT_SYMBOL(register_md_personality);
8562 EXPORT_SYMBOL(unregister_md_personality);
8563 EXPORT_SYMBOL(md_error);
8564 EXPORT_SYMBOL(md_done_sync);
8565 EXPORT_SYMBOL(md_write_start);
8566 EXPORT_SYMBOL(md_write_end);
8567 EXPORT_SYMBOL(md_register_thread);
8568 EXPORT_SYMBOL(md_unregister_thread);
8569 EXPORT_SYMBOL(md_wakeup_thread);
8570 EXPORT_SYMBOL(md_check_recovery);
8571 MODULE_LICENSE("GPL");
8572 MODULE_DESCRIPTION("MD RAID framework");
8573 MODULE_ALIAS("md");
8574 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);