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 <[email protected]>
  11    - boot support for linear and striped mode by Harald Hoyer <[email protected]>
  12    - kerneld support by Boris Tobotras <[email protected]>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <[email protected]>
  15    - Devfs support by Richard Gooch <[email protected]>
  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 <[email protected]>.
  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/mutex.h>
  40 #include <linux/buffer_head.h> /* for invalidate_bdev */
  41 #include <linux/poll.h>
  42 #include <linux/ctype.h>
  43 #include <linux/string.h>
  44 #include <linux/hdreg.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/random.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 #define DEBUG 0
  58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  59
  60 #ifndef MODULE
  61 static void autostart_arrays(int part);
  62 #endif
  63
  64 static LIST_HEAD(pers_list);
  65 static DEFINE_SPINLOCK(pers_lock);
  66
  67 static void md_print_devices(void);
  68
  69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  70 static struct workqueue_struct *md_wq;
  71 static struct workqueue_struct *md_misc_wq;
  72
  73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  74
  75 /*
  76  * Default number of read corrections we'll attempt on an rdev
  77  * before ejecting it from the array. We divide the read error
  78  * count by 2 for every hour elapsed between read errors.
  79  */
  80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  81 /*
  82  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  83  * is 1000 KB/sec, so the extra system load does not show up that much.
  84  * Increase it if you want to have more _guaranteed_ speed. Note that
  85  * the RAID driver will use the maximum available bandwidth if the IO
  86  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  87  * speed limit - in case reconstruction slows down your system despite
  88  * idle IO detection.
  89  *
  90  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  91  * or /sys/block/mdX/md/sync_speed_{min,max}
  92  */
  93
  94 static int sysctl_speed_limit_min = 1000;
  95 static int sysctl_speed_limit_max = 200000;
  96 static inline int speed_min(mddev_t *mddev)
  97 {
  98         return mddev->sync_speed_min ?
  99                 mddev->sync_speed_min : sysctl_speed_limit_min;
 100 }
 101
 102 static inline int speed_max(mddev_t *mddev)
 103 {
 104         return mddev->sync_speed_max ?
 105                 mddev->sync_speed_max : sysctl_speed_limit_max;
 106 }
 107
 108 static struct ctl_table_header *raid_table_header;
 109
 110 static ctl_table raid_table[] = {
 111         {
 112                 .procname       = "speed_limit_min",
 113                 .data           = &sysctl_speed_limit_min,
 114                 .maxlen         = sizeof(int),
 115                 .mode           = S_IRUGO|S_IWUSR,
 116                 .proc_handler   = proc_dointvec,
 117         },
 118         {
 119                 .procname       = "speed_limit_max",
 120                 .data           = &sysctl_speed_limit_max,
 121                 .maxlen         = sizeof(int),
 122                 .mode           = S_IRUGO|S_IWUSR,
 123                 .proc_handler   = proc_dointvec,
 124         },
 125         { }
 126 };
 127
 128 static ctl_table raid_dir_table[] = {
 129         {
 130                 .procname       = "raid",
 131                 .maxlen         = 0,
 132                 .mode           = S_IRUGO|S_IXUGO,
 133                 .child          = raid_table,
 134         },
 135         { }
 136 };
 137
 138 static ctl_table raid_root_table[] = {
 139         {
 140                 .procname       = "dev",
 141                 .maxlen         = 0,
 142                 .mode           = 0555,
 143                 .child          = raid_dir_table,
 144         },
 145         {  }
 146 };
 147
 148 static const struct block_device_operations md_fops;
 149
 150 static int start_readonly;
 151
 152 /* bio_clone_mddev
 153  * like bio_clone, but with a local bio set
 154  */
 155
 156 static void mddev_bio_destructor(struct bio *bio)
 157 {
 158         mddev_t *mddev, **mddevp;
 159
 160         mddevp = (void*)bio;
 161         mddev = mddevp[-1];
 162
 163         bio_free(bio, mddev->bio_set);
 164 }
 165
 166 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 167                             mddev_t *mddev)
 168 {
 169         struct bio *b;
 170         mddev_t **mddevp;
 171
 172         if (!mddev || !mddev->bio_set)
 173                 return bio_alloc(gfp_mask, nr_iovecs);
 174
 175         b = bio_alloc_bioset(gfp_mask, nr_iovecs,
 176                              mddev->bio_set);
 177         if (!b)
 178                 return NULL;
 179         mddevp = (void*)b;
 180         mddevp[-1] = mddev;
 181         b->bi_destructor = mddev_bio_destructor;
 182         return b;
 183 }
 184 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 185
 186 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
 187                             mddev_t *mddev)
 188 {
 189         struct bio *b;
 190         mddev_t **mddevp;
 191
 192         if (!mddev || !mddev->bio_set)
 193                 return bio_clone(bio, gfp_mask);
 194
 195         b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
 196                              mddev->bio_set);
 197         if (!b)
 198                 return NULL;
 199         mddevp = (void*)b;
 200         mddevp[-1] = mddev;
 201         b->bi_destructor = mddev_bio_destructor;
 202         __bio_clone(b, bio);
 203         if (bio_integrity(bio)) {
 204                 int ret;
 205
 206                 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
 207
 208                 if (ret < 0) {
 209                         bio_put(b);
 210                         return NULL;
 211                 }
 212         }
 213
 214         return b;
 215 }
 216 EXPORT_SYMBOL_GPL(bio_clone_mddev);
 217
 218 /*
 219  * We have a system wide 'event count' that is incremented
 220  * on any 'interesting' event, and readers of /proc/mdstat
 221  * can use 'poll' or 'select' to find out when the event
 222  * count increases.
 223  *
 224  * Events are:
 225  *  start array, stop array, error, add device, remove device,
 226  *  start build, activate spare
 227  */
 228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 229 static atomic_t md_event_count;
 230 void md_new_event(mddev_t *mddev)
 231 {
 232         atomic_inc(&md_event_count);
 233         wake_up(&md_event_waiters);
 234 }
 235 EXPORT_SYMBOL_GPL(md_new_event);
 236
 237 /* Alternate version that can be called from interrupts
 238  * when calling sysfs_notify isn't needed.
 239  */
 240 static void md_new_event_inintr(mddev_t *mddev)
 241 {
 242         atomic_inc(&md_event_count);
 243         wake_up(&md_event_waiters);
 244 }
 245
 246 /*
 247  * Enables to iterate over all existing md arrays
 248  * all_mddevs_lock protects this list.
 249  */
 250 static LIST_HEAD(all_mddevs);
 251 static DEFINE_SPINLOCK(all_mddevs_lock);
 252
 253
 254 /*
 255  * iterates through all used mddevs in the system.
 256  * We take care to grab the all_mddevs_lock whenever navigating
 257  * the list, and to always hold a refcount when unlocked.
 258  * Any code which breaks out of this loop while own
 259  * a reference to the current mddev and must mddev_put it.
 260  */
 261 #define for_each_mddev(mddev,tmp)                                       \
 262                                                                         \
 263         for (({ spin_lock(&all_mddevs_lock);                            \
 264                 tmp = all_mddevs.next;                                  \
 265                 mddev = NULL;});                                        \
 266              ({ if (tmp != &all_mddevs)                                 \
 267                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 268                 spin_unlock(&all_mddevs_lock);                          \
 269                 if (mddev) mddev_put(mddev);                            \
 270                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 271                 tmp != &all_mddevs;});                                  \
 272              ({ spin_lock(&all_mddevs_lock);                            \
 273                 tmp = tmp->next;})                                      \
 274                 )
 275
 276
 277 /* Rather than calling directly into the personality make_request function,
 278  * IO requests come here first so that we can check if the device is
 279  * being suspended pending a reconfiguration.
 280  * We hold a refcount over the call to ->make_request.  By the time that
 281  * call has finished, the bio has been linked into some internal structure
 282  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 283  */
 284 static int md_make_request(struct request_queue *q, struct bio *bio)
 285 {
 286         const int rw = bio_data_dir(bio);
 287         mddev_t *mddev = q->queuedata;
 288         int rv;
 289         int cpu;
 290
 291         if (mddev == NULL || mddev->pers == NULL) {
 292                 bio_io_error(bio);
 293                 return 0;
 294         }
 295         rcu_read_lock();
 296         if (mddev->suspended) {
 297                 DEFINE_WAIT(__wait);
 298                 for (;;) {
 299                         prepare_to_wait(&mddev->sb_wait, &__wait,
 300                                         TASK_UNINTERRUPTIBLE);
 301                         if (!mddev->suspended)
 302                                 break;
 303                         rcu_read_unlock();
 304                         schedule();
 305                         rcu_read_lock();
 306                 }
 307                 finish_wait(&mddev->sb_wait, &__wait);
 308         }
 309         atomic_inc(&mddev->active_io);
 310         rcu_read_unlock();
 311
 312         rv = mddev->pers->make_request(mddev, bio);
 313
 314         cpu = part_stat_lock();
 315         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 316         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 317                       bio_sectors(bio));
 318         part_stat_unlock();
 319
 320         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 321                 wake_up(&mddev->sb_wait);
 322
 323         return rv;
 324 }
 325
 326 /* mddev_suspend makes sure no new requests are submitted
 327  * to the device, and that any requests that have been submitted
 328  * are completely handled.
 329  * Once ->stop is called and completes, the module will be completely
 330  * unused.
 331  */
 332 void mddev_suspend(mddev_t *mddev)
 333 {
 334         BUG_ON(mddev->suspended);
 335         mddev->suspended = 1;
 336         synchronize_rcu();
 337         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 338         mddev->pers->quiesce(mddev, 1);
 339 }
 340 EXPORT_SYMBOL_GPL(mddev_suspend);
 341
 342 void mddev_resume(mddev_t *mddev)
 343 {
 344         mddev->suspended = 0;
 345         wake_up(&mddev->sb_wait);
 346         mddev->pers->quiesce(mddev, 0);
 347 }
 348 EXPORT_SYMBOL_GPL(mddev_resume);
 349
 350 int mddev_congested(mddev_t *mddev, int bits)
 351 {
 352         return mddev->suspended;
 353 }
 354 EXPORT_SYMBOL(mddev_congested);
 355
 356 /*
 357  * Generic flush handling for md
 358  */
 359
 360 static void md_end_flush(struct bio *bio, int err)
 361 {
 362         mdk_rdev_t *rdev = bio->bi_private;
 363         mddev_t *mddev = rdev->mddev;
 364
 365         rdev_dec_pending(rdev, mddev);
 366
 367         if (atomic_dec_and_test(&mddev->flush_pending)) {
 368                 /* The pre-request flush has finished */
 369                 queue_work(md_wq, &mddev->flush_work);
 370         }
 371         bio_put(bio);
 372 }
 373
 374 static void md_submit_flush_data(struct work_struct *ws);
 375
 376 static void submit_flushes(struct work_struct *ws)
 377 {
 378         mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 379         mdk_rdev_t *rdev;
 380
 381         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 382         atomic_set(&mddev->flush_pending, 1);
 383         rcu_read_lock();
 384         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
 385                 if (rdev->raid_disk >= 0 &&
 386                     !test_bit(Faulty, &rdev->flags)) {
 387                         /* Take two references, one is dropped
 388                          * when request finishes, one after
 389                          * we reclaim rcu_read_lock
 390                          */
 391                         struct bio *bi;
 392                         atomic_inc(&rdev->nr_pending);
 393                         atomic_inc(&rdev->nr_pending);
 394                         rcu_read_unlock();
 395                         bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
 396                         bi->bi_end_io = md_end_flush;
 397                         bi->bi_private = rdev;
 398                         bi->bi_bdev = rdev->bdev;
 399                         atomic_inc(&mddev->flush_pending);
 400                         submit_bio(WRITE_FLUSH, bi);
 401                         rcu_read_lock();
 402                         rdev_dec_pending(rdev, mddev);
 403                 }
 404         rcu_read_unlock();
 405         if (atomic_dec_and_test(&mddev->flush_pending))
 406                 queue_work(md_wq, &mddev->flush_work);
 407 }
 408
 409 static void md_submit_flush_data(struct work_struct *ws)
 410 {
 411         mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 412         struct bio *bio = mddev->flush_bio;
 413
 414         if (bio->bi_size == 0)
 415                 /* an empty barrier - all done */
 416                 bio_endio(bio, 0);
 417         else {
 418                 bio->bi_rw &= ~REQ_FLUSH;
 419                 if (mddev->pers->make_request(mddev, bio))
 420                         generic_make_request(bio);
 421         }
 422
 423         mddev->flush_bio = NULL;
 424         wake_up(&mddev->sb_wait);
 425 }
 426
 427 void md_flush_request(mddev_t *mddev, struct bio *bio)
 428 {
 429         spin_lock_irq(&mddev->write_lock);
 430         wait_event_lock_irq(mddev->sb_wait,
 431                             !mddev->flush_bio,
 432                             mddev->write_lock, /*nothing*/);
 433         mddev->flush_bio = bio;
 434         spin_unlock_irq(&mddev->write_lock);
 435
 436         INIT_WORK(&mddev->flush_work, submit_flushes);
 437         queue_work(md_wq, &mddev->flush_work);
 438 }
 439 EXPORT_SYMBOL(md_flush_request);
 440
 441 /* Support for plugging.
 442  * This mirrors the plugging support in request_queue, but does not
 443  * require having a whole queue
 444  */
 445 static void plugger_work(struct work_struct *work)
 446 {
 447         struct plug_handle *plug =
 448                 container_of(work, struct plug_handle, unplug_work);
 449         plug->unplug_fn(plug);
 450 }
 451 static void plugger_timeout(unsigned long data)
 452 {
 453         struct plug_handle *plug = (void *)data;
 454         kblockd_schedule_work(NULL, &plug->unplug_work);
 455 }
 456 void plugger_init(struct plug_handle *plug,
 457                   void (*unplug_fn)(struct plug_handle *))
 458 {
 459         plug->unplug_flag = 0;
 460         plug->unplug_fn = unplug_fn;
 461         init_timer(&plug->unplug_timer);
 462         plug->unplug_timer.function = plugger_timeout;
 463         plug->unplug_timer.data = (unsigned long)plug;
 464         INIT_WORK(&plug->unplug_work, plugger_work);
 465 }
 466 EXPORT_SYMBOL_GPL(plugger_init);
 467
 468 void plugger_set_plug(struct plug_handle *plug)
 469 {
 470         if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
 471                 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
 472 }
 473 EXPORT_SYMBOL_GPL(plugger_set_plug);
 474
 475 int plugger_remove_plug(struct plug_handle *plug)
 476 {
 477         if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
 478                 del_timer(&plug->unplug_timer);
 479                 return 1;
 480         } else
 481                 return 0;
 482 }
 483 EXPORT_SYMBOL_GPL(plugger_remove_plug);
 484
 485
 486 static inline mddev_t *mddev_get(mddev_t *mddev)
 487 {
 488         atomic_inc(&mddev->active);
 489         return mddev;
 490 }
 491
 492 static void mddev_delayed_delete(struct work_struct *ws);
 493
 494 static void mddev_put(mddev_t *mddev)
 495 {
 496         struct bio_set *bs = NULL;
 497
 498         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 499                 return;
 500         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 501             mddev->ctime == 0 && !mddev->hold_active) {
 502                 /* Array is not configured at all, and not held active,
 503                  * so destroy it */
 504                 list_del(&mddev->all_mddevs);
 505                 bs = mddev->bio_set;
 506                 mddev->bio_set = NULL;
 507                 if (mddev->gendisk) {
 508                         /* We did a probe so need to clean up.  Call
 509                          * queue_work inside the spinlock so that
 510                          * flush_workqueue() after mddev_find will
 511                          * succeed in waiting for the work to be done.
 512                          */
 513                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 514                         queue_work(md_misc_wq, &mddev->del_work);
 515                 } else
 516                         kfree(mddev);
 517         }
 518         spin_unlock(&all_mddevs_lock);
 519         if (bs)
 520                 bioset_free(bs);
 521 }
 522
 523 void mddev_init(mddev_t *mddev)
 524 {
 525         mutex_init(&mddev->open_mutex);
 526         mutex_init(&mddev->reconfig_mutex);
 527         mutex_init(&mddev->bitmap_info.mutex);
 528         INIT_LIST_HEAD(&mddev->disks);
 529         INIT_LIST_HEAD(&mddev->all_mddevs);
 530         init_timer(&mddev->safemode_timer);
 531         atomic_set(&mddev->active, 1);
 532         atomic_set(&mddev->openers, 0);
 533         atomic_set(&mddev->active_io, 0);
 534         spin_lock_init(&mddev->write_lock);
 535         atomic_set(&mddev->flush_pending, 0);
 536         init_waitqueue_head(&mddev->sb_wait);
 537         init_waitqueue_head(&mddev->recovery_wait);
 538         mddev->reshape_position = MaxSector;
 539         mddev->resync_min = 0;
 540         mddev->resync_max = MaxSector;
 541         mddev->level = LEVEL_NONE;
 542 }
 543 EXPORT_SYMBOL_GPL(mddev_init);
 544
 545 static mddev_t * mddev_find(dev_t unit)
 546 {
 547         mddev_t *mddev, *new = NULL;
 548
 549  retry:
 550         spin_lock(&all_mddevs_lock);
 551
 552         if (unit) {
 553                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 554                         if (mddev->unit == unit) {
 555                                 mddev_get(mddev);
 556                                 spin_unlock(&all_mddevs_lock);
 557                                 kfree(new);
 558                                 return mddev;
 559                         }
 560
 561                 if (new) {
 562                         list_add(&new->all_mddevs, &all_mddevs);
 563                         spin_unlock(&all_mddevs_lock);
 564                         new->hold_active = UNTIL_IOCTL;
 565                         return new;
 566                 }
 567         } else if (new) {
 568                 /* find an unused unit number */
 569                 static int next_minor = 512;
 570                 int start = next_minor;
 571                 int is_free = 0;
 572                 int dev = 0;
 573                 while (!is_free) {
 574                         dev = MKDEV(MD_MAJOR, next_minor);
 575                         next_minor++;
 576                         if (next_minor > MINORMASK)
 577                                 next_minor = 0;
 578                         if (next_minor == start) {
 579                                 /* Oh dear, all in use. */
 580                                 spin_unlock(&all_mddevs_lock);
 581                                 kfree(new);
 582                                 return NULL;
 583                         }
 584
 585                         is_free = 1;
 586                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 587                                 if (mddev->unit == dev) {
 588                                         is_free = 0;
 589                                         break;
 590                                 }
 591                 }
 592                 new->unit = dev;
 593                 new->md_minor = MINOR(dev);
 594                 new->hold_active = UNTIL_STOP;
 595                 list_add(&new->all_mddevs, &all_mddevs);
 596                 spin_unlock(&all_mddevs_lock);
 597                 return new;
 598         }
 599         spin_unlock(&all_mddevs_lock);
 600
 601         new = kzalloc(sizeof(*new), GFP_KERNEL);
 602         if (!new)
 603                 return NULL;
 604
 605         new->unit = unit;
 606         if (MAJOR(unit) == MD_MAJOR)
 607                 new->md_minor = MINOR(unit);
 608         else
 609                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 610
 611         mddev_init(new);
 612
 613         goto retry;
 614 }
 615
 616 static inline int mddev_lock(mddev_t * mddev)
 617 {
 618         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 619 }
 620
 621 static inline int mddev_is_locked(mddev_t *mddev)
 622 {
 623         return mutex_is_locked(&mddev->reconfig_mutex);
 624 }
 625
 626 static inline int mddev_trylock(mddev_t * mddev)
 627 {
 628         return mutex_trylock(&mddev->reconfig_mutex);
 629 }
 630
 631 static struct attribute_group md_redundancy_group;
 632
 633 static void mddev_unlock(mddev_t * mddev)
 634 {
 635         if (mddev->to_remove) {
 636                 /* These cannot be removed under reconfig_mutex as
 637                  * an access to the files will try to take reconfig_mutex
 638                  * while holding the file unremovable, which leads to
 639                  * a deadlock.
 640                  * So hold set sysfs_active while the remove in happeing,
 641                  * and anything else which might set ->to_remove or my
 642                  * otherwise change the sysfs namespace will fail with
 643                  * -EBUSY if sysfs_active is still set.
 644                  * We set sysfs_active under reconfig_mutex and elsewhere
 645                  * test it under the same mutex to ensure its correct value
 646                  * is seen.
 647                  */
 648                 struct attribute_group *to_remove = mddev->to_remove;
 649                 mddev->to_remove = NULL;
 650                 mddev->sysfs_active = 1;
 651                 mutex_unlock(&mddev->reconfig_mutex);
 652
 653                 if (mddev->kobj.sd) {
 654                         if (to_remove != &md_redundancy_group)
 655                                 sysfs_remove_group(&mddev->kobj, to_remove);
 656                         if (mddev->pers == NULL ||
 657                             mddev->pers->sync_request == NULL) {
 658                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 659                                 if (mddev->sysfs_action)
 660                                         sysfs_put(mddev->sysfs_action);
 661                                 mddev->sysfs_action = NULL;
 662                         }
 663                 }
 664                 mddev->sysfs_active = 0;
 665         } else
 666                 mutex_unlock(&mddev->reconfig_mutex);
 667
 668         md_wakeup_thread(mddev->thread);
 669 }
 670
 671 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 672 {
 673         mdk_rdev_t *rdev;
 674
 675         list_for_each_entry(rdev, &mddev->disks, same_set)
 676                 if (rdev->desc_nr == nr)
 677                         return rdev;
 678
 679         return NULL;
 680 }
 681
 682 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 683 {
 684         mdk_rdev_t *rdev;
 685
 686         list_for_each_entry(rdev, &mddev->disks, same_set)
 687                 if (rdev->bdev->bd_dev == dev)
 688                         return rdev;
 689
 690         return NULL;
 691 }
 692
 693 static struct mdk_personality *find_pers(int level, char *clevel)
 694 {
 695         struct mdk_personality *pers;
 696         list_for_each_entry(pers, &pers_list, list) {
 697                 if (level != LEVEL_NONE && pers->level == level)
 698                         return pers;
 699                 if (strcmp(pers->name, clevel)==0)
 700                         return pers;
 701         }
 702         return NULL;
 703 }
 704
 705 /* return the offset of the super block in 512byte sectors */
 706 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 707 {
 708         sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
 709         return MD_NEW_SIZE_SECTORS(num_sectors);
 710 }
 711
 712 static int alloc_disk_sb(mdk_rdev_t * rdev)
 713 {
 714         if (rdev->sb_page)
 715                 MD_BUG();
 716
 717         rdev->sb_page = alloc_page(GFP_KERNEL);
 718         if (!rdev->sb_page) {
 719                 printk(KERN_ALERT "md: out of memory.\n");
 720                 return -ENOMEM;
 721         }
 722
 723         return 0;
 724 }
 725
 726 static void free_disk_sb(mdk_rdev_t * rdev)
 727 {
 728         if (rdev->sb_page) {
 729                 put_page(rdev->sb_page);
 730                 rdev->sb_loaded = 0;
 731                 rdev->sb_page = NULL;
 732                 rdev->sb_start = 0;
 733                 rdev->sectors = 0;
 734         }
 735 }
 736
 737
 738 static void super_written(struct bio *bio, int error)
 739 {
 740         mdk_rdev_t *rdev = bio->bi_private;
 741         mddev_t *mddev = rdev->mddev;
 742
 743         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 744                 printk("md: super_written gets error=%d, uptodate=%d\n",
 745                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 746                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 747                 md_error(mddev, rdev);
 748         }
 749
 750         if (atomic_dec_and_test(&mddev->pending_writes))
 751                 wake_up(&mddev->sb_wait);
 752         bio_put(bio);
 753 }
 754
 755 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 756                    sector_t sector, int size, struct page *page)
 757 {
 758         /* write first size bytes of page to sector of rdev
 759          * Increment mddev->pending_writes before returning
 760          * and decrement it on completion, waking up sb_wait
 761          * if zero is reached.
 762          * If an error occurred, call md_error
 763          */
 764         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 765
 766         bio->bi_bdev = rdev->bdev;
 767         bio->bi_sector = sector;
 768         bio_add_page(bio, page, size, 0);
 769         bio->bi_private = rdev;
 770         bio->bi_end_io = super_written;
 771
 772         atomic_inc(&mddev->pending_writes);
 773         submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
 774                    bio);
 775 }
 776
 777 void md_super_wait(mddev_t *mddev)
 778 {
 779         /* wait for all superblock writes that were scheduled to complete */
 780         DEFINE_WAIT(wq);
 781         for(;;) {
 782                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 783                 if (atomic_read(&mddev->pending_writes)==0)
 784                         break;
 785                 schedule();
 786         }
 787         finish_wait(&mddev->sb_wait, &wq);
 788 }
 789
 790 static void bi_complete(struct bio *bio, int error)
 791 {
 792         complete((struct completion*)bio->bi_private);
 793 }
 794
 795 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
 796                  struct page *page, int rw)
 797 {
 798         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 799         struct completion event;
 800         int ret;
 801
 802         rw |= REQ_SYNC | REQ_UNPLUG;
 803
 804         bio->bi_bdev = rdev->bdev;
 805         bio->bi_sector = sector;
 806         bio_add_page(bio, page, size, 0);
 807         init_completion(&event);
 808         bio->bi_private = &event;
 809         bio->bi_end_io = bi_complete;
 810         submit_bio(rw, bio);
 811         wait_for_completion(&event);
 812
 813         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 814         bio_put(bio);
 815         return ret;
 816 }
 817 EXPORT_SYMBOL_GPL(sync_page_io);
 818
 819 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 820 {
 821         char b[BDEVNAME_SIZE];
 822         if (!rdev->sb_page) {
 823                 MD_BUG();
 824                 return -EINVAL;
 825         }
 826         if (rdev->sb_loaded)
 827                 return 0;
 828
 829
 830         if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ))
 831                 goto fail;
 832         rdev->sb_loaded = 1;
 833         return 0;
 834
 835 fail:
 836         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 837                 bdevname(rdev->bdev,b));
 838         return -EINVAL;
 839 }
 840
 841 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 842 {
 843         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 844                 sb1->set_uuid1 == sb2->set_uuid1 &&
 845                 sb1->set_uuid2 == sb2->set_uuid2 &&
 846                 sb1->set_uuid3 == sb2->set_uuid3;
 847 }
 848
 849 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 850 {
 851         int ret;
 852         mdp_super_t *tmp1, *tmp2;
 853
 854         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 855         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 856
 857         if (!tmp1 || !tmp2) {
 858                 ret = 0;
 859                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 860                 goto abort;
 861         }
 862
 863         *tmp1 = *sb1;
 864         *tmp2 = *sb2;
 865
 866         /*
 867          * nr_disks is not constant
 868          */
 869         tmp1->nr_disks = 0;
 870         tmp2->nr_disks = 0;
 871
 872         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 873 abort:
 874         kfree(tmp1);
 875         kfree(tmp2);
 876         return ret;
 877 }
 878
 879
 880 static u32 md_csum_fold(u32 csum)
 881 {
 882         csum = (csum & 0xffff) + (csum >> 16);
 883         return (csum & 0xffff) + (csum >> 16);
 884 }
 885
 886 static unsigned int calc_sb_csum(mdp_super_t * sb)
 887 {
 888         u64 newcsum = 0;
 889         u32 *sb32 = (u32*)sb;
 890         int i;
 891         unsigned int disk_csum, csum;
 892
 893         disk_csum = sb->sb_csum;
 894         sb->sb_csum = 0;
 895
 896         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 897                 newcsum += sb32[i];
 898         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 899
 900
 901 #ifdef CONFIG_ALPHA
 902         /* This used to use csum_partial, which was wrong for several
 903          * reasons including that different results are returned on
 904          * different architectures.  It isn't critical that we get exactly
 905          * the same return value as before (we always csum_fold before
 906          * testing, and that removes any differences).  However as we
 907          * know that csum_partial always returned a 16bit value on
 908          * alphas, do a fold to maximise conformity to previous behaviour.
 909          */
 910         sb->sb_csum = md_csum_fold(disk_csum);
 911 #else
 912         sb->sb_csum = disk_csum;
 913 #endif
 914         return csum;
 915 }
 916
 917
 918 /*
 919  * Handle superblock details.
 920  * We want to be able to handle multiple superblock formats
 921  * so we have a common interface to them all, and an array of
 922  * different handlers.
 923  * We rely on user-space to write the initial superblock, and support
 924  * reading and updating of superblocks.
 925  * Interface methods are:
 926  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 927  *      loads and validates a superblock on dev.
 928  *      if refdev != NULL, compare superblocks on both devices
 929  *    Return:
 930  *      0 - dev has a superblock that is compatible with refdev
 931  *      1 - dev has a superblock that is compatible and newer than refdev
 932  *          so dev should be used as the refdev in future
 933  *     -EINVAL superblock incompatible or invalid
 934  *     -othererror e.g. -EIO
 935  *
 936  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 937  *      Verify that dev is acceptable into mddev.
 938  *       The first time, mddev->raid_disks will be 0, and data from
 939  *       dev should be merged in.  Subsequent calls check that dev
 940  *       is new enough.  Return 0 or -EINVAL
 941  *
 942  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 943  *     Update the superblock for rdev with data in mddev
 944  *     This does not write to disc.
 945  *
 946  */
 947
 948 struct super_type  {
 949         char                *name;
 950         struct module       *owner;
 951         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 952                                           int minor_version);
 953         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 954         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 955         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 956                                                 sector_t num_sectors);
 957 };
 958
 959 /*
 960  * Check that the given mddev has no bitmap.
 961  *
 962  * This function is called from the run method of all personalities that do not
 963  * support bitmaps. It prints an error message and returns non-zero if mddev
 964  * has a bitmap. Otherwise, it returns 0.
 965  *
 966  */
 967 int md_check_no_bitmap(mddev_t *mddev)
 968 {
 969         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
 970                 return 0;
 971         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
 972                 mdname(mddev), mddev->pers->name);
 973         return 1;
 974 }
 975 EXPORT_SYMBOL(md_check_no_bitmap);
 976
 977 /*
 978  * load_super for 0.90.0
 979  */
 980 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 981 {
 982         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 983         mdp_super_t *sb;
 984         int ret;
 985
 986         /*
 987          * Calculate the position of the superblock (512byte sectors),
 988          * it's at the end of the disk.
 989          *
 990          * It also happens to be a multiple of 4Kb.
 991          */
 992         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 993
 994         ret = read_disk_sb(rdev, MD_SB_BYTES);
 995         if (ret) return ret;
 996
 997         ret = -EINVAL;
 998
 999         bdevname(rdev->bdev, b);
1000         sb = (mdp_super_t*)page_address(rdev->sb_page);
1001
1002         if (sb->md_magic != MD_SB_MAGIC) {
1003                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1004                        b);
1005                 goto abort;
1006         }
1007
1008         if (sb->major_version != 0 ||
1009             sb->minor_version < 90 ||
1010             sb->minor_version > 91) {
1011                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1012                         sb->major_version, sb->minor_version,
1013                         b);
1014                 goto abort;
1015         }
1016
1017         if (sb->raid_disks <= 0)
1018                 goto abort;
1019
1020         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1021                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1022                         b);
1023                 goto abort;
1024         }
1025
1026         rdev->preferred_minor = sb->md_minor;
1027         rdev->data_offset = 0;
1028         rdev->sb_size = MD_SB_BYTES;
1029
1030         if (sb->level == LEVEL_MULTIPATH)
1031                 rdev->desc_nr = -1;
1032         else
1033                 rdev->desc_nr = sb->this_disk.number;
1034
1035         if (!refdev) {
1036                 ret = 1;
1037         } else {
1038                 __u64 ev1, ev2;
1039                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1040                 if (!uuid_equal(refsb, sb)) {
1041                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
1042                                 b, bdevname(refdev->bdev,b2));
1043                         goto abort;
1044                 }
1045                 if (!sb_equal(refsb, sb)) {
1046                         printk(KERN_WARNING "md: %s has same UUID"
1047                                " but different superblock to %s\n",
1048                                b, bdevname(refdev->bdev, b2));
1049                         goto abort;
1050                 }
1051                 ev1 = md_event(sb);
1052                 ev2 = md_event(refsb);
1053                 if (ev1 > ev2)
1054                         ret = 1;
1055                 else
1056                         ret = 0;
1057         }
1058         rdev->sectors = rdev->sb_start;
1059
1060         if (rdev->sectors < sb->size * 2 && sb->level > 1)
1061                 /* "this cannot possibly happen" ... */
1062                 ret = -EINVAL;
1063
1064  abort:
1065         return ret;
1066 }
1067
1068 /*
1069  * validate_super for 0.90.0
1070  */
1071 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1072 {
1073         mdp_disk_t *desc;
1074         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1075         __u64 ev1 = md_event(sb);
1076
1077         rdev->raid_disk = -1;
1078         clear_bit(Faulty, &rdev->flags);
1079         clear_bit(In_sync, &rdev->flags);
1080         clear_bit(WriteMostly, &rdev->flags);
1081
1082         if (mddev->raid_disks == 0) {
1083                 mddev->major_version = 0;
1084                 mddev->minor_version = sb->minor_version;
1085                 mddev->patch_version = sb->patch_version;
1086                 mddev->external = 0;
1087                 mddev->chunk_sectors = sb->chunk_size >> 9;
1088                 mddev->ctime = sb->ctime;
1089                 mddev->utime = sb->utime;
1090                 mddev->level = sb->level;
1091                 mddev->clevel[0] = 0;
1092                 mddev->layout = sb->layout;
1093                 mddev->raid_disks = sb->raid_disks;
1094                 mddev->dev_sectors = sb->size * 2;
1095                 mddev->events = ev1;
1096                 mddev->bitmap_info.offset = 0;
1097                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1098
1099                 if (mddev->minor_version >= 91) {
1100                         mddev->reshape_position = sb->reshape_position;
1101                         mddev->delta_disks = sb->delta_disks;
1102                         mddev->new_level = sb->new_level;
1103                         mddev->new_layout = sb->new_layout;
1104                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1105                 } else {
1106                         mddev->reshape_position = MaxSector;
1107                         mddev->delta_disks = 0;
1108                         mddev->new_level = mddev->level;
1109                         mddev->new_layout = mddev->layout;
1110                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1111                 }
1112
1113                 if (sb->state & (1<<MD_SB_CLEAN))
1114                         mddev->recovery_cp = MaxSector;
1115                 else {
1116                         if (sb->events_hi == sb->cp_events_hi &&
1117                                 sb->events_lo == sb->cp_events_lo) {
1118                                 mddev->recovery_cp = sb->recovery_cp;
1119                         } else
1120                                 mddev->recovery_cp = 0;
1121                 }
1122
1123                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1124                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1125                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1126                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1127
1128                 mddev->max_disks = MD_SB_DISKS;
1129
1130                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1131                     mddev->bitmap_info.file == NULL)
1132                         mddev->bitmap_info.offset =
1133                                 mddev->bitmap_info.default_offset;
1134
1135         } else if (mddev->pers == NULL) {
1136                 /* Insist on good event counter while assembling, except
1137                  * for spares (which don't need an event count) */
1138                 ++ev1;
1139                 if (sb->disks[rdev->desc_nr].state & (
1140                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1141                         if (ev1 < mddev->events)
1142                                 return -EINVAL;
1143         } else if (mddev->bitmap) {
1144                 /* if adding to array with a bitmap, then we can accept an
1145                  * older device ... but not too old.
1146                  */
1147                 if (ev1 < mddev->bitmap->events_cleared)
1148                         return 0;
1149         } else {
1150                 if (ev1 < mddev->events)
1151                         /* just a hot-add of a new device, leave raid_disk at -1 */
1152                         return 0;
1153         }
1154
1155         if (mddev->level != LEVEL_MULTIPATH) {
1156                 desc = sb->disks + rdev->desc_nr;
1157
1158                 if (desc->state & (1<<MD_DISK_FAULTY))
1159                         set_bit(Faulty, &rdev->flags);
1160                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1161                             desc->raid_disk < mddev->raid_disks */) {
1162                         set_bit(In_sync, &rdev->flags);
1163                         rdev->raid_disk = desc->raid_disk;
1164                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1165                         /* active but not in sync implies recovery up to
1166                          * reshape position.  We don't know exactly where
1167                          * that is, so set to zero for now */
1168                         if (mddev->minor_version >= 91) {
1169                                 rdev->recovery_offset = 0;
1170                                 rdev->raid_disk = desc->raid_disk;
1171                         }
1172                 }
1173                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1174                         set_bit(WriteMostly, &rdev->flags);
1175         } else /* MULTIPATH are always insync */
1176                 set_bit(In_sync, &rdev->flags);
1177         return 0;
1178 }
1179
1180 /*
1181  * sync_super for 0.90.0
1182  */
1183 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1184 {
1185         mdp_super_t *sb;
1186         mdk_rdev_t *rdev2;
1187         int next_spare = mddev->raid_disks;
1188
1189
1190         /* make rdev->sb match mddev data..
1191          *
1192          * 1/ zero out disks
1193          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1194          * 3/ any empty disks < next_spare become removed
1195          *
1196          * disks[0] gets initialised to REMOVED because
1197          * we cannot be sure from other fields if it has
1198          * been initialised or not.
1199          */
1200         int i;
1201         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1202
1203         rdev->sb_size = MD_SB_BYTES;
1204
1205         sb = (mdp_super_t*)page_address(rdev->sb_page);
1206
1207         memset(sb, 0, sizeof(*sb));
1208
1209         sb->md_magic = MD_SB_MAGIC;
1210         sb->major_version = mddev->major_version;
1211         sb->patch_version = mddev->patch_version;
1212         sb->gvalid_words  = 0; /* ignored */
1213         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1214         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1215         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1216         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1217
1218         sb->ctime = mddev->ctime;
1219         sb->level = mddev->level;
1220         sb->size = mddev->dev_sectors / 2;
1221         sb->raid_disks = mddev->raid_disks;
1222         sb->md_minor = mddev->md_minor;
1223         sb->not_persistent = 0;
1224         sb->utime = mddev->utime;
1225         sb->state = 0;
1226         sb->events_hi = (mddev->events>>32);
1227         sb->events_lo = (u32)mddev->events;
1228
1229         if (mddev->reshape_position == MaxSector)
1230                 sb->minor_version = 90;
1231         else {
1232                 sb->minor_version = 91;
1233                 sb->reshape_position = mddev->reshape_position;
1234                 sb->new_level = mddev->new_level;
1235                 sb->delta_disks = mddev->delta_disks;
1236                 sb->new_layout = mddev->new_layout;
1237                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1238         }
1239         mddev->minor_version = sb->minor_version;
1240         if (mddev->in_sync)
1241         {
1242                 sb->recovery_cp = mddev->recovery_cp;
1243                 sb->cp_events_hi = (mddev->events>>32);
1244                 sb->cp_events_lo = (u32)mddev->events;
1245                 if (mddev->recovery_cp == MaxSector)
1246                         sb->state = (1<< MD_SB_CLEAN);
1247         } else
1248                 sb->recovery_cp = 0;
1249
1250         sb->layout = mddev->layout;
1251         sb->chunk_size = mddev->chunk_sectors << 9;
1252
1253         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1254                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1255
1256         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1257         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1258                 mdp_disk_t *d;
1259                 int desc_nr;
1260                 int is_active = test_bit(In_sync, &rdev2->flags);
1261
1262                 if (rdev2->raid_disk >= 0 &&
1263                     sb->minor_version >= 91)
1264                         /* we have nowhere to store the recovery_offset,
1265                          * but if it is not below the reshape_position,
1266                          * we can piggy-back on that.
1267                          */
1268                         is_active = 1;
1269                 if (rdev2->raid_disk < 0 ||
1270                     test_bit(Faulty, &rdev2->flags))
1271                         is_active = 0;
1272                 if (is_active)
1273                         desc_nr = rdev2->raid_disk;
1274                 else
1275                         desc_nr = next_spare++;
1276                 rdev2->desc_nr = desc_nr;
1277                 d = &sb->disks[rdev2->desc_nr];
1278                 nr_disks++;
1279                 d->number = rdev2->desc_nr;
1280                 d->major = MAJOR(rdev2->bdev->bd_dev);
1281                 d->minor = MINOR(rdev2->bdev->bd_dev);
1282                 if (is_active)
1283                         d->raid_disk = rdev2->raid_disk;
1284                 else
1285                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1286                 if (test_bit(Faulty, &rdev2->flags))
1287                         d->state = (1<<MD_DISK_FAULTY);
1288                 else if (is_active) {
1289                         d->state = (1<<MD_DISK_ACTIVE);
1290                         if (test_bit(In_sync, &rdev2->flags))
1291                                 d->state |= (1<<MD_DISK_SYNC);
1292                         active++;
1293                         working++;
1294                 } else {
1295                         d->state = 0;
1296                         spare++;
1297                         working++;
1298                 }
1299                 if (test_bit(WriteMostly, &rdev2->flags))
1300                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1301         }
1302         /* now set the "removed" and "faulty" bits on any missing devices */
1303         for (i=0 ; i < mddev->raid_disks ; i++) {
1304                 mdp_disk_t *d = &sb->disks[i];
1305                 if (d->state == 0 && d->number == 0) {
1306                         d->number = i;
1307                         d->raid_disk = i;
1308                         d->state = (1<<MD_DISK_REMOVED);
1309                         d->state |= (1<<MD_DISK_FAULTY);
1310                         failed++;
1311                 }
1312         }
1313         sb->nr_disks = nr_disks;
1314         sb->active_disks = active;
1315         sb->working_disks = working;
1316         sb->failed_disks = failed;
1317         sb->spare_disks = spare;
1318
1319         sb->this_disk = sb->disks[rdev->desc_nr];
1320         sb->sb_csum = calc_sb_csum(sb);
1321 }
1322
1323 /*
1324  * rdev_size_change for 0.90.0
1325  */
1326 static unsigned long long
1327 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1328 {
1329         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1330                 return 0; /* component must fit device */
1331         if (rdev->mddev->bitmap_info.offset)
1332                 return 0; /* can't move bitmap */
1333         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1334         if (!num_sectors || num_sectors > rdev->sb_start)
1335                 num_sectors = rdev->sb_start;
1336         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1337                        rdev->sb_page);
1338         md_super_wait(rdev->mddev);
1339         return num_sectors;
1340 }
1341
1342
1343 /*
1344  * version 1 superblock
1345  */
1346
1347 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1348 {
1349         __le32 disk_csum;
1350         u32 csum;
1351         unsigned long long newcsum;
1352         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1353         __le32 *isuper = (__le32*)sb;
1354         int i;
1355
1356         disk_csum = sb->sb_csum;
1357         sb->sb_csum = 0;
1358         newcsum = 0;
1359         for (i=0; size>=4; size -= 4 )
1360                 newcsum += le32_to_cpu(*isuper++);
1361
1362         if (size == 2)
1363                 newcsum += le16_to_cpu(*(__le16*) isuper);
1364
1365         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1366         sb->sb_csum = disk_csum;
1367         return cpu_to_le32(csum);
1368 }
1369
1370 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1371 {
1372         struct mdp_superblock_1 *sb;
1373         int ret;
1374         sector_t sb_start;
1375         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1376         int bmask;
1377
1378         /*
1379          * Calculate the position of the superblock in 512byte sectors.
1380          * It is always aligned to a 4K boundary and
1381          * depeding on minor_version, it can be:
1382          * 0: At least 8K, but less than 12K, from end of device
1383          * 1: At start of device
1384          * 2: 4K from start of device.
1385          */
1386         switch(minor_version) {
1387         case 0:
1388                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1389                 sb_start -= 8*2;
1390                 sb_start &= ~(sector_t)(4*2-1);
1391                 break;
1392         case 1:
1393                 sb_start = 0;
1394                 break;
1395         case 2:
1396                 sb_start = 8;
1397                 break;
1398         default:
1399                 return -EINVAL;
1400         }
1401         rdev->sb_start = sb_start;
1402
1403         /* superblock is rarely larger than 1K, but it can be larger,
1404          * and it is safe to read 4k, so we do that
1405          */
1406         ret = read_disk_sb(rdev, 4096);
1407         if (ret) return ret;
1408
1409
1410         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1411
1412         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413             sb->major_version != cpu_to_le32(1) ||
1414             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417                 return -EINVAL;
1418
1419         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420                 printk("md: invalid superblock checksum on %s\n",
1421                         bdevname(rdev->bdev,b));
1422                 return -EINVAL;
1423         }
1424         if (le64_to_cpu(sb->data_size) < 10) {
1425                 printk("md: data_size too small on %s\n",
1426                        bdevname(rdev->bdev,b));
1427                 return -EINVAL;
1428         }
1429
1430         rdev->preferred_minor = 0xffff;
1431         rdev->data_offset = le64_to_cpu(sb->data_offset);
1432         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1433
1434         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1435         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1436         if (rdev->sb_size & bmask)
1437                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1438
1439         if (minor_version
1440             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1441                 return -EINVAL;
1442
1443         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1444                 rdev->desc_nr = -1;
1445         else
1446                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1447
1448         if (!refdev) {
1449                 ret = 1;
1450         } else {
1451                 __u64 ev1, ev2;
1452                 struct mdp_superblock_1 *refsb =
1453                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1454
1455                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1456                     sb->level != refsb->level ||
1457                     sb->layout != refsb->layout ||
1458                     sb->chunksize != refsb->chunksize) {
1459                         printk(KERN_WARNING "md: %s has strangely different"
1460                                 " superblock to %s\n",
1461                                 bdevname(rdev->bdev,b),
1462                                 bdevname(refdev->bdev,b2));
1463                         return -EINVAL;
1464                 }
1465                 ev1 = le64_to_cpu(sb->events);
1466                 ev2 = le64_to_cpu(refsb->events);
1467
1468                 if (ev1 > ev2)
1469                         ret = 1;
1470                 else
1471                         ret = 0;
1472         }
1473         if (minor_version)
1474                 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1475                         le64_to_cpu(sb->data_offset);
1476         else
1477                 rdev->sectors = rdev->sb_start;
1478         if (rdev->sectors < le64_to_cpu(sb->data_size))
1479                 return -EINVAL;
1480         rdev->sectors = le64_to_cpu(sb->data_size);
1481         if (le64_to_cpu(sb->size) > rdev->sectors)
1482                 return -EINVAL;
1483         return ret;
1484 }
1485
1486 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1487 {
1488         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1489         __u64 ev1 = le64_to_cpu(sb->events);
1490
1491         rdev->raid_disk = -1;
1492         clear_bit(Faulty, &rdev->flags);
1493         clear_bit(In_sync, &rdev->flags);
1494         clear_bit(WriteMostly, &rdev->flags);
1495
1496         if (mddev->raid_disks == 0) {
1497                 mddev->major_version = 1;
1498                 mddev->patch_version = 0;
1499                 mddev->external = 0;
1500                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1501                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1502                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1503                 mddev->level = le32_to_cpu(sb->level);
1504                 mddev->clevel[0] = 0;
1505                 mddev->layout = le32_to_cpu(sb->layout);
1506                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1507                 mddev->dev_sectors = le64_to_cpu(sb->size);
1508                 mddev->events = ev1;
1509                 mddev->bitmap_info.offset = 0;
1510                 mddev->bitmap_info.default_offset = 1024 >> 9;
1511
1512                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1513                 memcpy(mddev->uuid, sb->set_uuid, 16);
1514
1515                 mddev->max_disks =  (4096-256)/2;
1516
1517                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1518                     mddev->bitmap_info.file == NULL )
1519                         mddev->bitmap_info.offset =
1520                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1521
1522                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1523                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1524                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1525                         mddev->new_level = le32_to_cpu(sb->new_level);
1526                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1527                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1528                 } else {
1529                         mddev->reshape_position = MaxSector;
1530                         mddev->delta_disks = 0;
1531                         mddev->new_level = mddev->level;
1532                         mddev->new_layout = mddev->layout;
1533                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1534                 }
1535
1536         } else if (mddev->pers == NULL) {
1537                 /* Insist of good event counter while assembling, except for
1538                  * spares (which don't need an event count) */
1539                 ++ev1;
1540                 if (rdev->desc_nr >= 0 &&
1541                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1542                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1543                         if (ev1 < mddev->events)
1544                                 return -EINVAL;
1545         } else if (mddev->bitmap) {
1546                 /* If adding to array with a bitmap, then we can accept an
1547                  * older device, but not too old.
1548                  */
1549                 if (ev1 < mddev->bitmap->events_cleared)
1550                         return 0;
1551         } else {
1552                 if (ev1 < mddev->events)
1553                         /* just a hot-add of a new device, leave raid_disk at -1 */
1554                         return 0;
1555         }
1556         if (mddev->level != LEVEL_MULTIPATH) {
1557                 int role;
1558                 if (rdev->desc_nr < 0 ||
1559                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1560                         role = 0xffff;
1561                         rdev->desc_nr = -1;
1562                 } else
1563                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1564                 switch(role) {
1565                 case 0xffff: /* spare */
1566                         break;
1567                 case 0xfffe: /* faulty */
1568                         set_bit(Faulty, &rdev->flags);
1569                         break;
1570                 default:
1571                         if ((le32_to_cpu(sb->feature_map) &
1572                              MD_FEATURE_RECOVERY_OFFSET))
1573                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1574                         else
1575                                 set_bit(In_sync, &rdev->flags);
1576                         rdev->raid_disk = role;
1577                         break;
1578                 }
1579                 if (sb->devflags & WriteMostly1)
1580                         set_bit(WriteMostly, &rdev->flags);
1581         } else /* MULTIPATH are always insync */
1582                 set_bit(In_sync, &rdev->flags);
1583
1584         return 0;
1585 }
1586
1587 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1588 {
1589         struct mdp_superblock_1 *sb;
1590         mdk_rdev_t *rdev2;
1591         int max_dev, i;
1592         /* make rdev->sb match mddev and rdev data. */
1593
1594         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1595
1596         sb->feature_map = 0;
1597         sb->pad0 = 0;
1598         sb->recovery_offset = cpu_to_le64(0);
1599         memset(sb->pad1, 0, sizeof(sb->pad1));
1600         memset(sb->pad2, 0, sizeof(sb->pad2));
1601         memset(sb->pad3, 0, sizeof(sb->pad3));
1602
1603         sb->utime = cpu_to_le64((__u64)mddev->utime);
1604         sb->events = cpu_to_le64(mddev->events);
1605         if (mddev->in_sync)
1606                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1607         else
1608                 sb->resync_offset = cpu_to_le64(0);
1609
1610         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1611
1612         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1613         sb->size = cpu_to_le64(mddev->dev_sectors);
1614         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1615         sb->level = cpu_to_le32(mddev->level);
1616         sb->layout = cpu_to_le32(mddev->layout);
1617
1618         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1619                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1620                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1621         }
1622
1623         if (rdev->raid_disk >= 0 &&
1624             !test_bit(In_sync, &rdev->flags)) {
1625                 sb->feature_map |=
1626                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1627                 sb->recovery_offset =
1628                         cpu_to_le64(rdev->recovery_offset);
1629         }
1630
1631         if (mddev->reshape_position != MaxSector) {
1632                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1633                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1634                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1635                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1636                 sb->new_level = cpu_to_le32(mddev->new_level);
1637                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1638         }
1639
1640         max_dev = 0;
1641         list_for_each_entry(rdev2, &mddev->disks, same_set)
1642                 if (rdev2->desc_nr+1 > max_dev)
1643                         max_dev = rdev2->desc_nr+1;
1644
1645         if (max_dev > le32_to_cpu(sb->max_dev)) {
1646                 int bmask;
1647                 sb->max_dev = cpu_to_le32(max_dev);
1648                 rdev->sb_size = max_dev * 2 + 256;
1649                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1650                 if (rdev->sb_size & bmask)
1651                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1652         } else
1653                 max_dev = le32_to_cpu(sb->max_dev);
1654
1655         for (i=0; i<max_dev;i++)
1656                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1657
1658         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1659                 i = rdev2->desc_nr;
1660                 if (test_bit(Faulty, &rdev2->flags))
1661                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1662                 else if (test_bit(In_sync, &rdev2->flags))
1663                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1664                 else if (rdev2->raid_disk >= 0)
1665                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1666                 else
1667                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1668         }
1669
1670         sb->sb_csum = calc_sb_1_csum(sb);
1671 }
1672
1673 static unsigned long long
1674 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1675 {
1676         struct mdp_superblock_1 *sb;
1677         sector_t max_sectors;
1678         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1679                 return 0; /* component must fit device */
1680         if (rdev->sb_start < rdev->data_offset) {
1681                 /* minor versions 1 and 2; superblock before data */
1682                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1683                 max_sectors -= rdev->data_offset;
1684                 if (!num_sectors || num_sectors > max_sectors)
1685                         num_sectors = max_sectors;
1686         } else if (rdev->mddev->bitmap_info.offset) {
1687                 /* minor version 0 with bitmap we can't move */
1688                 return 0;
1689         } else {
1690                 /* minor version 0; superblock after data */
1691                 sector_t sb_start;
1692                 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1693                 sb_start &= ~(sector_t)(4*2 - 1);
1694                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1695                 if (!num_sectors || num_sectors > max_sectors)
1696                         num_sectors = max_sectors;
1697                 rdev->sb_start = sb_start;
1698         }
1699         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1700         sb->data_size = cpu_to_le64(num_sectors);
1701         sb->super_offset = rdev->sb_start;
1702         sb->sb_csum = calc_sb_1_csum(sb);
1703         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1704                        rdev->sb_page);
1705         md_super_wait(rdev->mddev);
1706         return num_sectors;
1707 }
1708
1709 static struct super_type super_types[] = {
1710         [0] = {
1711                 .name   = "0.90.0",
1712                 .owner  = THIS_MODULE,
1713                 .load_super         = super_90_load,
1714                 .validate_super     = super_90_validate,
1715                 .sync_super         = super_90_sync,
1716                 .rdev_size_change   = super_90_rdev_size_change,
1717         },
1718         [1] = {
1719                 .name   = "md-1",
1720                 .owner  = THIS_MODULE,
1721                 .load_super         = super_1_load,
1722                 .validate_super     = super_1_validate,
1723                 .sync_super         = super_1_sync,
1724                 .rdev_size_change   = super_1_rdev_size_change,
1725         },
1726 };
1727
1728 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1729 {
1730         mdk_rdev_t *rdev, *rdev2;
1731
1732         rcu_read_lock();
1733         rdev_for_each_rcu(rdev, mddev1)
1734                 rdev_for_each_rcu(rdev2, mddev2)
1735                         if (rdev->bdev->bd_contains ==
1736                             rdev2->bdev->bd_contains) {
1737                                 rcu_read_unlock();
1738                                 return 1;
1739                         }
1740         rcu_read_unlock();
1741         return 0;
1742 }
1743
1744 static LIST_HEAD(pending_raid_disks);
1745
1746 /*
1747  * Try to register data integrity profile for an mddev
1748  *
1749  * This is called when an array is started and after a disk has been kicked
1750  * from the array. It only succeeds if all working and active component devices
1751  * are integrity capable with matching profiles.
1752  */
1753 int md_integrity_register(mddev_t *mddev)
1754 {
1755         mdk_rdev_t *rdev, *reference = NULL;
1756
1757         if (list_empty(&mddev->disks))
1758                 return 0; /* nothing to do */
1759         if (blk_get_integrity(mddev->gendisk))
1760                 return 0; /* already registered */
1761         list_for_each_entry(rdev, &mddev->disks, same_set) {
1762                 /* skip spares and non-functional disks */
1763                 if (test_bit(Faulty, &rdev->flags))
1764                         continue;
1765                 if (rdev->raid_disk < 0)
1766                         continue;
1767                 /*
1768                  * If at least one rdev is not integrity capable, we can not
1769                  * enable data integrity for the md device.
1770                  */
1771                 if (!bdev_get_integrity(rdev->bdev))
1772                         return -EINVAL;
1773                 if (!reference) {
1774                         /* Use the first rdev as the reference */
1775                         reference = rdev;
1776                         continue;
1777                 }
1778                 /* does this rdev's profile match the reference profile? */
1779                 if (blk_integrity_compare(reference->bdev->bd_disk,
1780                                 rdev->bdev->bd_disk) < 0)
1781                         return -EINVAL;
1782         }
1783         /*
1784          * All component devices are integrity capable and have matching
1785          * profiles, register the common profile for the md device.
1786          */
1787         if (blk_integrity_register(mddev->gendisk,
1788                         bdev_get_integrity(reference->bdev)) != 0) {
1789                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1790                         mdname(mddev));
1791                 return -EINVAL;
1792         }
1793         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1794                 mdname(mddev));
1795         return 0;
1796 }
1797 EXPORT_SYMBOL(md_integrity_register);
1798
1799 /* Disable data integrity if non-capable/non-matching disk is being added */
1800 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1801 {
1802         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1803         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1804
1805         if (!bi_mddev) /* nothing to do */
1806                 return;
1807         if (rdev->raid_disk < 0) /* skip spares */
1808                 return;
1809         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1810                                              rdev->bdev->bd_disk) >= 0)
1811                 return;
1812         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1813         blk_integrity_unregister(mddev->gendisk);
1814 }
1815 EXPORT_SYMBOL(md_integrity_add_rdev);
1816
1817 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1818 {
1819         char b[BDEVNAME_SIZE];
1820         struct kobject *ko;
1821         char *s;
1822         int err;
1823
1824         if (rdev->mddev) {
1825                 MD_BUG();
1826                 return -EINVAL;
1827         }
1828
1829         /* prevent duplicates */
1830         if (find_rdev(mddev, rdev->bdev->bd_dev))
1831                 return -EEXIST;
1832
1833         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1834         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1835                         rdev->sectors < mddev->dev_sectors)) {
1836                 if (mddev->pers) {
1837                         /* Cannot change size, so fail
1838                          * If mddev->level <= 0, then we don't care
1839                          * about aligning sizes (e.g. linear)
1840                          */
1841                         if (mddev->level > 0)
1842                                 return -ENOSPC;
1843                 } else
1844                         mddev->dev_sectors = rdev->sectors;
1845         }
1846
1847         /* Verify rdev->desc_nr is unique.
1848          * If it is -1, assign a free number, else
1849          * check number is not in use
1850          */
1851         if (rdev->desc_nr < 0) {
1852                 int choice = 0;
1853                 if (mddev->pers) choice = mddev->raid_disks;
1854                 while (find_rdev_nr(mddev, choice))
1855                         choice++;
1856                 rdev->desc_nr = choice;
1857         } else {
1858                 if (find_rdev_nr(mddev, rdev->desc_nr))
1859                         return -EBUSY;
1860         }
1861         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1862                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1863                        mdname(mddev), mddev->max_disks);
1864                 return -EBUSY;
1865         }
1866         bdevname(rdev->bdev,b);
1867         while ( (s=strchr(b, '/')) != NULL)
1868                 *s = '!';
1869
1870         rdev->mddev = mddev;
1871         printk(KERN_INFO "md: bind<%s>\n", b);
1872
1873         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1874                 goto fail;
1875
1876         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1877         if (sysfs_create_link(&rdev->kobj, ko, "block"))
1878                 /* failure here is OK */;
1879         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1880
1881         list_add_rcu(&rdev->same_set, &mddev->disks);
1882         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
1883
1884         /* May as well allow recovery to be retried once */
1885         mddev->recovery_disabled = 0;
1886
1887         return 0;
1888
1889  fail:
1890         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1891                b, mdname(mddev));
1892         return err;
1893 }
1894
1895 static void md_delayed_delete(struct work_struct *ws)
1896 {
1897         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1898         kobject_del(&rdev->kobj);
1899         kobject_put(&rdev->kobj);
1900 }
1901
1902 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1903 {
1904         char b[BDEVNAME_SIZE];
1905         if (!rdev->mddev) {
1906                 MD_BUG();
1907                 return;
1908         }
1909         list_del_rcu(&rdev->same_set);
1910         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1911         rdev->mddev = NULL;
1912         sysfs_remove_link(&rdev->kobj, "block");
1913         sysfs_put(rdev->sysfs_state);
1914         rdev->sysfs_state = NULL;
1915         /* We need to delay this, otherwise we can deadlock when
1916          * writing to 'remove' to "dev/state".  We also need
1917          * to delay it due to rcu usage.
1918          */
1919         synchronize_rcu();
1920         INIT_WORK(&rdev->del_work, md_delayed_delete);
1921         kobject_get(&rdev->kobj);
1922         queue_work(md_misc_wq, &rdev->del_work);
1923 }
1924
1925 /*
1926  * prevent the device from being mounted, repartitioned or
1927  * otherwise reused by a RAID array (or any other kernel
1928  * subsystem), by bd_claiming the device.
1929  */
1930 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1931 {
1932         int err = 0;
1933         struct block_device *bdev;
1934         char b[BDEVNAME_SIZE];
1935
1936         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1937                                  shared ? (mdk_rdev_t *)lock_rdev : rdev);
1938         if (IS_ERR(bdev)) {
1939                 printk(KERN_ERR "md: could not open %s.\n",
1940                         __bdevname(dev, b));
1941                 return PTR_ERR(bdev);
1942         }
1943         if (!shared)
1944                 set_bit(AllReserved, &rdev->flags);
1945         rdev->bdev = bdev;
1946         return err;
1947 }
1948
1949 static void unlock_rdev(mdk_rdev_t *rdev)
1950 {
1951         struct block_device *bdev = rdev->bdev;
1952         rdev->bdev = NULL;
1953         if (!bdev)
1954                 MD_BUG();
1955         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1956 }
1957
1958 void md_autodetect_dev(dev_t dev);
1959
1960 static void export_rdev(mdk_rdev_t * rdev)
1961 {
1962         char b[BDEVNAME_SIZE];
1963         printk(KERN_INFO "md: export_rdev(%s)\n",
1964                 bdevname(rdev->bdev,b));
1965         if (rdev->mddev)
1966                 MD_BUG();
1967         free_disk_sb(rdev);
1968 #ifndef MODULE
1969         if (test_bit(AutoDetected, &rdev->flags))
1970                 md_autodetect_dev(rdev->bdev->bd_dev);
1971 #endif
1972         unlock_rdev(rdev);
1973         kobject_put(&rdev->kobj);
1974 }
1975
1976 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1977 {
1978         unbind_rdev_from_array(rdev);
1979         export_rdev(rdev);
1980 }
1981
1982 static void export_array(mddev_t *mddev)
1983 {
1984         mdk_rdev_t *rdev, *tmp;
1985
1986         rdev_for_each(rdev, tmp, mddev) {
1987                 if (!rdev->mddev) {
1988                         MD_BUG();
1989                         continue;
1990                 }
1991                 kick_rdev_from_array(rdev);
1992         }
1993         if (!list_empty(&mddev->disks))
1994                 MD_BUG();
1995         mddev->raid_disks = 0;
1996         mddev->major_version = 0;
1997 }
1998
1999 static void print_desc(mdp_disk_t *desc)
2000 {
2001         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2002                 desc->major,desc->minor,desc->raid_disk,desc->state);
2003 }
2004
2005 static void print_sb_90(mdp_super_t *sb)
2006 {
2007         int i;
2008
2009         printk(KERN_INFO
2010                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2011                 sb->major_version, sb->minor_version, sb->patch_version,
2012                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2013                 sb->ctime);
2014         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2015                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2016                 sb->md_minor, sb->layout, sb->chunk_size);
2017         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2018                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2019                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2020                 sb->failed_disks, sb->spare_disks,
2021                 sb->sb_csum, (unsigned long)sb->events_lo);
2022
2023         printk(KERN_INFO);
2024         for (i = 0; i < MD_SB_DISKS; i++) {
2025                 mdp_disk_t *desc;
2026
2027                 desc = sb->disks + i;
2028                 if (desc->number || desc->major || desc->minor ||
2029                     desc->raid_disk || (desc->state && (desc->state != 4))) {
2030                         printk("     D %2d: ", i);
2031                         print_desc(desc);
2032                 }
2033         }
2034         printk(KERN_INFO "md:     THIS: ");
2035         print_desc(&sb->this_disk);
2036 }
2037
2038 static void print_sb_1(struct mdp_superblock_1 *sb)
2039 {
2040         __u8 *uuid;
2041
2042         uuid = sb->set_uuid;
2043         printk(KERN_INFO
2044                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2045                "md:    Name: \"%s\" CT:%llu\n",
2046                 le32_to_cpu(sb->major_version),
2047                 le32_to_cpu(sb->feature_map),
2048                 uuid,
2049                 sb->set_name,
2050                 (unsigned long long)le64_to_cpu(sb->ctime)
2051                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2052
2053         uuid = sb->device_uuid;
2054         printk(KERN_INFO
2055                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2056                         " RO:%llu\n"
2057                "md:     Dev:%08x UUID: %pU\n"
2058                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2059                "md:         (MaxDev:%u) \n",
2060                 le32_to_cpu(sb->level),
2061                 (unsigned long long)le64_to_cpu(sb->size),
2062                 le32_to_cpu(sb->raid_disks),
2063                 le32_to_cpu(sb->layout),
2064                 le32_to_cpu(sb->chunksize),
2065                 (unsigned long long)le64_to_cpu(sb->data_offset),
2066                 (unsigned long long)le64_to_cpu(sb->data_size),
2067                 (unsigned long long)le64_to_cpu(sb->super_offset),
2068                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2069                 le32_to_cpu(sb->dev_number),
2070                 uuid,
2071                 sb->devflags,
2072                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2073                 (unsigned long long)le64_to_cpu(sb->events),
2074                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2075                 le32_to_cpu(sb->sb_csum),
2076                 le32_to_cpu(sb->max_dev)
2077                 );
2078 }
2079
2080 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2081 {
2082         char b[BDEVNAME_SIZE];
2083         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2084                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2085                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2086                 rdev->desc_nr);
2087         if (rdev->sb_loaded) {
2088                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2089                 switch (major_version) {
2090                 case 0:
2091                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2092                         break;
2093                 case 1:
2094                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2095                         break;
2096                 }
2097         } else
2098                 printk(KERN_INFO "md: no rdev superblock!\n");
2099 }
2100
2101 static void md_print_devices(void)
2102 {
2103         struct list_head *tmp;
2104         mdk_rdev_t *rdev;
2105         mddev_t *mddev;
2106         char b[BDEVNAME_SIZE];
2107
2108         printk("\n");
2109         printk("md:     **********************************\n");
2110         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2111         printk("md:     **********************************\n");
2112         for_each_mddev(mddev, tmp) {
2113
2114                 if (mddev->bitmap)
2115                         bitmap_print_sb(mddev->bitmap);
2116                 else
2117                         printk("%s: ", mdname(mddev));
2118                 list_for_each_entry(rdev, &mddev->disks, same_set)
2119                         printk("<%s>", bdevname(rdev->bdev,b));
2120                 printk("\n");
2121
2122                 list_for_each_entry(rdev, &mddev->disks, same_set)
2123                         print_rdev(rdev, mddev->major_version);
2124         }
2125         printk("md:     **********************************\n");
2126         printk("\n");
2127 }
2128
2129
2130 static void sync_sbs(mddev_t * mddev, int nospares)
2131 {
2132         /* Update each superblock (in-memory image), but
2133          * if we are allowed to, skip spares which already
2134          * have the right event counter, or have one earlier
2135          * (which would mean they aren't being marked as dirty
2136          * with the rest of the array)
2137          */
2138         mdk_rdev_t *rdev;
2139         list_for_each_entry(rdev, &mddev->disks, same_set) {
2140                 if (rdev->sb_events == mddev->events ||
2141                     (nospares &&
2142                      rdev->raid_disk < 0 &&
2143                      rdev->sb_events+1 == mddev->events)) {
2144                         /* Don't update this superblock */
2145                         rdev->sb_loaded = 2;
2146                 } else {
2147                         super_types[mddev->major_version].
2148                                 sync_super(mddev, rdev);
2149                         rdev->sb_loaded = 1;
2150                 }
2151         }
2152 }
2153
2154 static void md_update_sb(mddev_t * mddev, int force_change)
2155 {
2156         mdk_rdev_t *rdev;
2157         int sync_req;
2158         int nospares = 0;
2159
2160 repeat:
2161         /* First make sure individual recovery_offsets are correct */
2162         list_for_each_entry(rdev, &mddev->disks, same_set) {
2163                 if (rdev->raid_disk >= 0 &&
2164                     mddev->delta_disks >= 0 &&
2165                     !test_bit(In_sync, &rdev->flags) &&
2166                     mddev->curr_resync_completed > rdev->recovery_offset)
2167                                 rdev->recovery_offset = mddev->curr_resync_completed;
2168
2169         }
2170         if (!mddev->persistent) {
2171                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2172                 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2173                 if (!mddev->external)
2174                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2175                 wake_up(&mddev->sb_wait);
2176                 return;
2177         }
2178
2179         spin_lock_irq(&mddev->write_lock);
2180
2181         mddev->utime = get_seconds();
2182
2183         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2184                 force_change = 1;
2185         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2186                 /* just a clean<-> dirty transition, possibly leave spares alone,
2187                  * though if events isn't the right even/odd, we will have to do
2188                  * spares after all
2189                  */
2190                 nospares = 1;
2191         if (force_change)
2192                 nospares = 0;
2193         if (mddev->degraded)
2194                 /* If the array is degraded, then skipping spares is both
2195                  * dangerous and fairly pointless.
2196                  * Dangerous because a device that was removed from the array
2197                  * might have a event_count that still looks up-to-date,
2198                  * so it can be re-added without a resync.
2199                  * Pointless because if there are any spares to skip,
2200                  * then a recovery will happen and soon that array won't
2201                  * be degraded any more and the spare can go back to sleep then.
2202                  */
2203                 nospares = 0;
2204
2205         sync_req = mddev->in_sync;
2206
2207         /* If this is just a dirty<->clean transition, and the array is clean
2208          * and 'events' is odd, we can roll back to the previous clean state */
2209         if (nospares
2210             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2211             && mddev->can_decrease_events
2212             && mddev->events != 1) {
2213                 mddev->events--;
2214                 mddev->can_decrease_events = 0;
2215         } else {
2216                 /* otherwise we have to go forward and ... */
2217                 mddev->events ++;
2218                 mddev->can_decrease_events = nospares;
2219         }
2220
2221         if (!mddev->events) {
2222                 /*
2223                  * oops, this 64-bit counter should never wrap.
2224                  * Either we are in around ~1 trillion A.C., assuming
2225                  * 1 reboot per second, or we have a bug:
2226                  */
2227                 MD_BUG();
2228                 mddev->events --;
2229         }
2230         sync_sbs(mddev, nospares);
2231         spin_unlock_irq(&mddev->write_lock);
2232
2233         dprintk(KERN_INFO
2234                 "md: updating %s RAID superblock on device (in sync %d)\n",
2235                 mdname(mddev),mddev->in_sync);
2236
2237         bitmap_update_sb(mddev->bitmap);
2238         list_for_each_entry(rdev, &mddev->disks, same_set) {
2239                 char b[BDEVNAME_SIZE];
2240                 dprintk(KERN_INFO "md: ");
2241                 if (rdev->sb_loaded != 1)
2242                         continue; /* no noise on spare devices */
2243                 if (test_bit(Faulty, &rdev->flags))
2244                         dprintk("(skipping faulty ");
2245
2246                 dprintk("%s ", bdevname(rdev->bdev,b));
2247                 if (!test_bit(Faulty, &rdev->flags)) {
2248                         md_super_write(mddev,rdev,
2249                                        rdev->sb_start, rdev->sb_size,
2250                                        rdev->sb_page);
2251                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2252                                 bdevname(rdev->bdev,b),
2253                                 (unsigned long long)rdev->sb_start);
2254                         rdev->sb_events = mddev->events;
2255
2256                 } else
2257                         dprintk(")\n");
2258                 if (mddev->level == LEVEL_MULTIPATH)
2259                         /* only need to write one superblock... */
2260                         break;
2261         }
2262         md_super_wait(mddev);
2263         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2264
2265         spin_lock_irq(&mddev->write_lock);
2266         if (mddev->in_sync != sync_req ||
2267             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2268                 /* have to write it out again */
2269                 spin_unlock_irq(&mddev->write_lock);
2270                 goto repeat;
2271         }
2272         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2273         spin_unlock_irq(&mddev->write_lock);
2274         wake_up(&mddev->sb_wait);
2275         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2276                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2277
2278 }
2279
2280 /* words written to sysfs files may, or may not, be \n terminated.
2281  * We want to accept with case. For this we use cmd_match.
2282  */
2283 static int cmd_match(const char *cmd, const char *str)
2284 {
2285         /* See if cmd, written into a sysfs file, matches
2286          * str.  They must either be the same, or cmd can
2287          * have a trailing newline
2288          */
2289         while (*cmd && *str && *cmd == *str) {
2290                 cmd++;
2291                 str++;
2292         }
2293         if (*cmd == '\n')
2294                 cmd++;
2295         if (*str || *cmd)
2296                 return 0;
2297         return 1;
2298 }
2299
2300 struct rdev_sysfs_entry {
2301         struct attribute attr;
2302         ssize_t (*show)(mdk_rdev_t *, char *);
2303         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2304 };
2305
2306 static ssize_t
2307 state_show(mdk_rdev_t *rdev, char *page)
2308 {
2309         char *sep = "";
2310         size_t len = 0;
2311
2312         if (test_bit(Faulty, &rdev->flags)) {
2313                 len+= sprintf(page+len, "%sfaulty",sep);
2314                 sep = ",";
2315         }
2316         if (test_bit(In_sync, &rdev->flags)) {
2317                 len += sprintf(page+len, "%sin_sync",sep);
2318                 sep = ",";
2319         }
2320         if (test_bit(WriteMostly, &rdev->flags)) {
2321                 len += sprintf(page+len, "%swrite_mostly",sep);
2322                 sep = ",";
2323         }
2324         if (test_bit(Blocked, &rdev->flags)) {
2325                 len += sprintf(page+len, "%sblocked", sep);
2326                 sep = ",";
2327         }
2328         if (!test_bit(Faulty, &rdev->flags) &&
2329             !test_bit(In_sync, &rdev->flags)) {
2330                 len += sprintf(page+len, "%sspare", sep);
2331                 sep = ",";
2332         }
2333         return len+sprintf(page+len, "\n");
2334 }
2335
2336 static ssize_t
2337 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2338 {
2339         /* can write
2340          *  faulty  - simulates and error
2341          *  remove  - disconnects the device
2342          *  writemostly - sets write_mostly
2343          *  -writemostly - clears write_mostly
2344          *  blocked - sets the Blocked flag
2345          *  -blocked - clears the Blocked flag
2346          *  insync - sets Insync providing device isn't active
2347          */
2348         int err = -EINVAL;
2349         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2350                 md_error(rdev->mddev, rdev);
2351                 err = 0;
2352         } else if (cmd_match(buf, "remove")) {
2353                 if (rdev->raid_disk >= 0)
2354                         err = -EBUSY;
2355                 else {
2356                         mddev_t *mddev = rdev->mddev;
2357                         kick_rdev_from_array(rdev);
2358                         if (mddev->pers)
2359                                 md_update_sb(mddev, 1);
2360                         md_new_event(mddev);
2361                         err = 0;
2362                 }
2363         } else if (cmd_match(buf, "writemostly")) {
2364                 set_bit(WriteMostly, &rdev->flags);
2365                 err = 0;
2366         } else if (cmd_match(buf, "-writemostly")) {
2367                 clear_bit(WriteMostly, &rdev->flags);
2368                 err = 0;
2369         } else if (cmd_match(buf, "blocked")) {
2370                 set_bit(Blocked, &rdev->flags);
2371                 err = 0;
2372         } else if (cmd_match(buf, "-blocked")) {
2373                 clear_bit(Blocked, &rdev->flags);
2374                 wake_up(&rdev->blocked_wait);
2375                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2376                 md_wakeup_thread(rdev->mddev->thread);
2377
2378                 err = 0;
2379         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2380                 set_bit(In_sync, &rdev->flags);
2381                 err = 0;
2382         }
2383         if (!err)
2384                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2385         return err ? err : len;
2386 }
2387 static struct rdev_sysfs_entry rdev_state =
2388 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2389
2390 static ssize_t
2391 errors_show(mdk_rdev_t *rdev, char *page)
2392 {
2393         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2394 }
2395
2396 static ssize_t
2397 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2398 {
2399         char *e;
2400         unsigned long n = simple_strtoul(buf, &e, 10);
2401         if (*buf && (*e == 0 || *e == '\n')) {
2402                 atomic_set(&rdev->corrected_errors, n);
2403                 return len;
2404         }
2405         return -EINVAL;
2406 }
2407 static struct rdev_sysfs_entry rdev_errors =
2408 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2409
2410 static ssize_t
2411 slot_show(mdk_rdev_t *rdev, char *page)
2412 {
2413         if (rdev->raid_disk < 0)
2414                 return sprintf(page, "none\n");
2415         else
2416                 return sprintf(page, "%d\n", rdev->raid_disk);
2417 }
2418
2419 static ssize_t
2420 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2421 {
2422         char *e;
2423         int err;
2424         char nm[20];
2425         int slot = simple_strtoul(buf, &e, 10);
2426         if (strncmp(buf, "none", 4)==0)
2427                 slot = -1;
2428         else if (e==buf || (*e && *e!= '\n'))
2429                 return -EINVAL;
2430         if (rdev->mddev->pers && slot == -1) {
2431                 /* Setting 'slot' on an active array requires also
2432                  * updating the 'rd%d' link, and communicating
2433                  * with the personality with ->hot_*_disk.
2434                  * For now we only support removing
2435                  * failed/spare devices.  This normally happens automatically,
2436                  * but not when the metadata is externally managed.
2437                  */
2438                 if (rdev->raid_disk == -1)
2439                         return -EEXIST;
2440                 /* personality does all needed checks */
2441                 if (rdev->mddev->pers->hot_add_disk == NULL)
2442                         return -EINVAL;
2443                 err = rdev->mddev->pers->
2444                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2445                 if (err)
2446                         return err;
2447                 sprintf(nm, "rd%d", rdev->raid_disk);
2448                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2449                 rdev->raid_disk = -1;
2450                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2451                 md_wakeup_thread(rdev->mddev->thread);
2452         } else if (rdev->mddev->pers) {
2453                 mdk_rdev_t *rdev2;
2454                 /* Activating a spare .. or possibly reactivating
2455                  * if we ever get bitmaps working here.
2456                  */
2457
2458                 if (rdev->raid_disk != -1)
2459                         return -EBUSY;
2460
2461                 if (rdev->mddev->pers->hot_add_disk == NULL)
2462                         return -EINVAL;
2463
2464                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2465                         if (rdev2->raid_disk == slot)
2466                                 return -EEXIST;
2467
2468                 rdev->raid_disk = slot;
2469                 if (test_bit(In_sync, &rdev->flags))
2470                         rdev->saved_raid_disk = slot;
2471                 else
2472                         rdev->saved_raid_disk = -1;
2473                 err = rdev->mddev->pers->
2474                         hot_add_disk(rdev->mddev, rdev);
2475                 if (err) {
2476                         rdev->raid_disk = -1;
2477                         return err;
2478                 } else
2479                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2480                 sprintf(nm, "rd%d", rdev->raid_disk);
2481                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2482                         /* failure here is OK */;
2483                 /* don't wakeup anyone, leave that to userspace. */
2484         } else {
2485                 if (slot >= rdev->mddev->raid_disks)
2486                         return -ENOSPC;
2487                 rdev->raid_disk = slot;
2488                 /* assume it is working */
2489                 clear_bit(Faulty, &rdev->flags);
2490                 clear_bit(WriteMostly, &rdev->flags);
2491                 set_bit(In_sync, &rdev->flags);
2492                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2493         }
2494         return len;
2495 }
2496
2497
2498 static struct rdev_sysfs_entry rdev_slot =
2499 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2500
2501 static ssize_t
2502 offset_show(mdk_rdev_t *rdev, char *page)
2503 {
2504         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2505 }
2506
2507 static ssize_t
2508 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2509 {
2510         char *e;
2511         unsigned long long offset = simple_strtoull(buf, &e, 10);
2512         if (e==buf || (*e && *e != '\n'))
2513                 return -EINVAL;
2514         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2515                 return -EBUSY;
2516         if (rdev->sectors && rdev->mddev->external)
2517                 /* Must set offset before size, so overlap checks
2518                  * can be sane */
2519                 return -EBUSY;
2520         rdev->data_offset = offset;
2521         return len;
2522 }
2523
2524 static struct rdev_sysfs_entry rdev_offset =
2525 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2526
2527 static ssize_t
2528 rdev_size_show(mdk_rdev_t *rdev, char *page)
2529 {
2530         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2531 }
2532
2533 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2534 {
2535         /* check if two start/length pairs overlap */
2536         if (s1+l1 <= s2)
2537                 return 0;
2538         if (s2+l2 <= s1)
2539                 return 0;
2540         return 1;
2541 }
2542
2543 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2544 {
2545         unsigned long long blocks;
2546         sector_t new;
2547
2548         if (strict_strtoull(buf, 10, &blocks) < 0)
2549                 return -EINVAL;
2550
2551         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2552                 return -EINVAL; /* sector conversion overflow */
2553
2554         new = blocks * 2;
2555         if (new != blocks * 2)
2556                 return -EINVAL; /* unsigned long long to sector_t overflow */
2557
2558         *sectors = new;
2559         return 0;
2560 }
2561
2562 static ssize_t
2563 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2564 {
2565         mddev_t *my_mddev = rdev->mddev;
2566         sector_t oldsectors = rdev->sectors;
2567         sector_t sectors;
2568
2569         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2570                 return -EINVAL;
2571         if (my_mddev->pers && rdev->raid_disk >= 0) {
2572                 if (my_mddev->persistent) {
2573                         sectors = super_types[my_mddev->major_version].
2574                                 rdev_size_change(rdev, sectors);
2575                         if (!sectors)
2576                                 return -EBUSY;
2577                 } else if (!sectors)
2578                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2579                                 rdev->data_offset;
2580         }
2581         if (sectors < my_mddev->dev_sectors)
2582                 return -EINVAL; /* component must fit device */
2583
2584         rdev->sectors = sectors;
2585         if (sectors > oldsectors && my_mddev->external) {
2586                 /* need to check that all other rdevs with the same ->bdev
2587                  * do not overlap.  We need to unlock the mddev to avoid
2588                  * a deadlock.  We have already changed rdev->sectors, and if
2589                  * we have to change it back, we will have the lock again.
2590                  */
2591                 mddev_t *mddev;
2592                 int overlap = 0;
2593                 struct list_head *tmp;
2594
2595                 mddev_unlock(my_mddev);
2596                 for_each_mddev(mddev, tmp) {
2597                         mdk_rdev_t *rdev2;
2598
2599                         mddev_lock(mddev);
2600                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2601                                 if (test_bit(AllReserved, &rdev2->flags) ||
2602                                     (rdev->bdev == rdev2->bdev &&
2603                                      rdev != rdev2 &&
2604                                      overlaps(rdev->data_offset, rdev->sectors,
2605                                               rdev2->data_offset,
2606                                               rdev2->sectors))) {
2607                                         overlap = 1;
2608                                         break;
2609                                 }
2610                         mddev_unlock(mddev);
2611                         if (overlap) {
2612                                 mddev_put(mddev);
2613                                 break;
2614                         }
2615                 }
2616                 mddev_lock(my_mddev);
2617                 if (overlap) {
2618                         /* Someone else could have slipped in a size
2619                          * change here, but doing so is just silly.
2620                          * We put oldsectors back because we *know* it is
2621                          * safe, and trust userspace not to race with
2622                          * itself
2623                          */
2624                         rdev->sectors = oldsectors;
2625                         return -EBUSY;
2626                 }
2627         }
2628         return len;
2629 }
2630
2631 static struct rdev_sysfs_entry rdev_size =
2632 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2633
2634
2635 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2636 {
2637         unsigned long long recovery_start = rdev->recovery_offset;
2638
2639         if (test_bit(In_sync, &rdev->flags) ||
2640             recovery_start == MaxSector)
2641                 return sprintf(page, "none\n");
2642
2643         return sprintf(page, "%llu\n", recovery_start);
2644 }
2645
2646 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2647 {
2648         unsigned long long recovery_start;
2649
2650         if (cmd_match(buf, "none"))
2651                 recovery_start = MaxSector;
2652         else if (strict_strtoull(buf, 10, &recovery_start))
2653                 return -EINVAL;
2654
2655         if (rdev->mddev->pers &&
2656             rdev->raid_disk >= 0)
2657                 return -EBUSY;
2658
2659         rdev->recovery_offset = recovery_start;
2660         if (recovery_start == MaxSector)
2661                 set_bit(In_sync, &rdev->flags);
2662         else
2663                 clear_bit(In_sync, &rdev->flags);
2664         return len;
2665 }
2666
2667 static struct rdev_sysfs_entry rdev_recovery_start =
2668 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2669
2670 static struct attribute *rdev_default_attrs[] = {
2671         &rdev_state.attr,
2672         &rdev_errors.attr,
2673         &rdev_slot.attr,
2674         &rdev_offset.attr,
2675         &rdev_size.attr,
2676         &rdev_recovery_start.attr,
2677         NULL,
2678 };
2679 static ssize_t
2680 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2681 {
2682         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2683         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2684         mddev_t *mddev = rdev->mddev;
2685         ssize_t rv;
2686
2687         if (!entry->show)
2688                 return -EIO;
2689
2690         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2691         if (!rv) {
2692                 if (rdev->mddev == NULL)
2693                         rv = -EBUSY;
2694                 else
2695                         rv = entry->show(rdev, page);
2696                 mddev_unlock(mddev);
2697         }
2698         return rv;
2699 }
2700
2701 static ssize_t
2702 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2703               const char *page, size_t length)
2704 {
2705         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2706         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2707         ssize_t rv;
2708         mddev_t *mddev = rdev->mddev;
2709
2710         if (!entry->store)
2711                 return -EIO;
2712         if (!capable(CAP_SYS_ADMIN))
2713                 return -EACCES;
2714         rv = mddev ? mddev_lock(mddev): -EBUSY;
2715         if (!rv) {
2716                 if (rdev->mddev == NULL)
2717                         rv = -EBUSY;
2718                 else
2719                         rv = entry->store(rdev, page, length);
2720                 mddev_unlock(mddev);
2721         }
2722         return rv;
2723 }
2724
2725 static void rdev_free(struct kobject *ko)
2726 {
2727         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2728         kfree(rdev);
2729 }
2730 static const struct sysfs_ops rdev_sysfs_ops = {
2731         .show           = rdev_attr_show,
2732         .store          = rdev_attr_store,
2733 };
2734 static struct kobj_type rdev_ktype = {
2735         .release        = rdev_free,
2736         .sysfs_ops      = &rdev_sysfs_ops,
2737         .default_attrs  = rdev_default_attrs,
2738 };
2739
2740 void md_rdev_init(mdk_rdev_t *rdev)
2741 {
2742         rdev->desc_nr = -1;
2743         rdev->saved_raid_disk = -1;
2744         rdev->raid_disk = -1;
2745         rdev->flags = 0;
2746         rdev->data_offset = 0;
2747         rdev->sb_events = 0;
2748         rdev->last_read_error.tv_sec  = 0;
2749         rdev->last_read_error.tv_nsec = 0;
2750         atomic_set(&rdev->nr_pending, 0);
2751         atomic_set(&rdev->read_errors, 0);
2752         atomic_set(&rdev->corrected_errors, 0);
2753
2754         INIT_LIST_HEAD(&rdev->same_set);
2755         init_waitqueue_head(&rdev->blocked_wait);
2756 }
2757 EXPORT_SYMBOL_GPL(md_rdev_init);
2758 /*
2759  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2760  *
2761  * mark the device faulty if:
2762  *
2763  *   - the device is nonexistent (zero size)
2764  *   - the device has no valid superblock
2765  *
2766  * a faulty rdev _never_ has rdev->sb set.
2767  */
2768 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2769 {
2770         char b[BDEVNAME_SIZE];
2771         int err;
2772         mdk_rdev_t *rdev;
2773         sector_t size;
2774
2775         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2776         if (!rdev) {
2777                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2778                 return ERR_PTR(-ENOMEM);
2779         }
2780
2781         md_rdev_init(rdev);
2782         if ((err = alloc_disk_sb(rdev)))
2783                 goto abort_free;
2784
2785         err = lock_rdev(rdev, newdev, super_format == -2);
2786         if (err)
2787                 goto abort_free;
2788
2789         kobject_init(&rdev->kobj, &rdev_ktype);
2790
2791         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2792         if (!size) {
2793                 printk(KERN_WARNING
2794                         "md: %s has zero or unknown size, marking faulty!\n",
2795                         bdevname(rdev->bdev,b));
2796                 err = -EINVAL;
2797                 goto abort_free;
2798         }
2799
2800         if (super_format >= 0) {
2801                 err = super_types[super_format].
2802                         load_super(rdev, NULL, super_minor);
2803                 if (err == -EINVAL) {
2804                         printk(KERN_WARNING
2805                                 "md: %s does not have a valid v%d.%d "
2806                                "superblock, not importing!\n",
2807                                 bdevname(rdev->bdev,b),
2808                                super_format, super_minor);
2809                         goto abort_free;
2810                 }
2811                 if (err < 0) {
2812                         printk(KERN_WARNING
2813                                 "md: could not read %s's sb, not importing!\n",
2814                                 bdevname(rdev->bdev,b));
2815                         goto abort_free;
2816                 }
2817         }
2818
2819         return rdev;
2820
2821 abort_free:
2822         if (rdev->sb_page) {
2823                 if (rdev->bdev)
2824                         unlock_rdev(rdev);
2825                 free_disk_sb(rdev);
2826         }
2827         kfree(rdev);
2828         return ERR_PTR(err);
2829 }
2830
2831 /*
2832  * Check a full RAID array for plausibility
2833  */
2834
2835
2836 static void analyze_sbs(mddev_t * mddev)
2837 {
2838         int i;
2839         mdk_rdev_t *rdev, *freshest, *tmp;
2840         char b[BDEVNAME_SIZE];
2841
2842         freshest = NULL;
2843         rdev_for_each(rdev, tmp, mddev)
2844                 switch (super_types[mddev->major_version].
2845                         load_super(rdev, freshest, mddev->minor_version)) {
2846                 case 1:
2847                         freshest = rdev;
2848                         break;
2849                 case 0:
2850                         break;
2851                 default:
2852                         printk( KERN_ERR \
2853                                 "md: fatal superblock inconsistency in %s"
2854                                 " -- removing from array\n",
2855                                 bdevname(rdev->bdev,b));
2856                         kick_rdev_from_array(rdev);
2857                 }
2858
2859
2860         super_types[mddev->major_version].
2861                 validate_super(mddev, freshest);
2862
2863         i = 0;
2864         rdev_for_each(rdev, tmp, mddev) {
2865                 if (mddev->max_disks &&
2866                     (rdev->desc_nr >= mddev->max_disks ||
2867                      i > mddev->max_disks)) {
2868                         printk(KERN_WARNING
2869                                "md: %s: %s: only %d devices permitted\n",
2870                                mdname(mddev), bdevname(rdev->bdev, b),
2871                                mddev->max_disks);
2872                         kick_rdev_from_array(rdev);
2873                         continue;
2874                 }
2875                 if (rdev != freshest)
2876                         if (super_types[mddev->major_version].
2877                             validate_super(mddev, rdev)) {
2878                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2879                                         " from array!\n",
2880                                         bdevname(rdev->bdev,b));
2881                                 kick_rdev_from_array(rdev);
2882                                 continue;
2883                         }
2884                 if (mddev->level == LEVEL_MULTIPATH) {
2885                         rdev->desc_nr = i++;
2886                         rdev->raid_disk = rdev->desc_nr;
2887                         set_bit(In_sync, &rdev->flags);
2888                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2889                         rdev->raid_disk = -1;
2890                         clear_bit(In_sync, &rdev->flags);
2891                 }
2892         }
2893 }
2894
2895 /* Read a fixed-point number.
2896  * Numbers in sysfs attributes should be in "standard" units where
2897  * possible, so time should be in seconds.
2898  * However we internally use a a much smaller unit such as
2899  * milliseconds or jiffies.
2900  * This function takes a decimal number with a possible fractional
2901  * component, and produces an integer which is the result of
2902  * multiplying that number by 10^'scale'.
2903  * all without any floating-point arithmetic.
2904  */
2905 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2906 {
2907         unsigned long result = 0;
2908         long decimals = -1;
2909         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2910                 if (*cp == '.')
2911                         decimals = 0;
2912                 else if (decimals < scale) {
2913                         unsigned int value;
2914                         value = *cp - '0';
2915                         result = result * 10 + value;
2916                         if (decimals >= 0)
2917                                 decimals++;
2918                 }
2919                 cp++;
2920         }
2921         if (*cp == '\n')
2922                 cp++;
2923         if (*cp)
2924                 return -EINVAL;
2925         if (decimals < 0)
2926                 decimals = 0;
2927         while (decimals < scale) {
2928                 result *= 10;
2929                 decimals ++;
2930         }
2931         *res = result;
2932         return 0;
2933 }
2934
2935
2936 static void md_safemode_timeout(unsigned long data);
2937
2938 static ssize_t
2939 safe_delay_show(mddev_t *mddev, char *page)
2940 {
2941         int msec = (mddev->safemode_delay*1000)/HZ;
2942         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2943 }
2944 static ssize_t
2945 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2946 {
2947         unsigned long msec;
2948
2949         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2950                 return -EINVAL;
2951         if (msec == 0)
2952                 mddev->safemode_delay = 0;
2953         else {
2954                 unsigned long old_delay = mddev->safemode_delay;
2955                 mddev->safemode_delay = (msec*HZ)/1000;
2956                 if (mddev->safemode_delay == 0)
2957                         mddev->safemode_delay = 1;
2958                 if (mddev->safemode_delay < old_delay)
2959                         md_safemode_timeout((unsigned long)mddev);
2960         }
2961         return len;
2962 }
2963 static struct md_sysfs_entry md_safe_delay =
2964 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2965
2966 static ssize_t
2967 level_show(mddev_t *mddev, char *page)
2968 {
2969         struct mdk_personality *p = mddev->pers;
2970         if (p)
2971                 return sprintf(page, "%s\n", p->name);
2972         else if (mddev->clevel[0])
2973                 return sprintf(page, "%s\n", mddev->clevel);
2974         else if (mddev->level != LEVEL_NONE)
2975                 return sprintf(page, "%d\n", mddev->level);
2976         else
2977                 return 0;
2978 }
2979
2980 static ssize_t
2981 level_store(mddev_t *mddev, const char *buf, size_t len)
2982 {
2983         char clevel[16];
2984         ssize_t rv = len;
2985         struct mdk_personality *pers;
2986         long level;
2987         void *priv;
2988         mdk_rdev_t *rdev;
2989
2990         if (mddev->pers == NULL) {
2991                 if (len == 0)
2992                         return 0;
2993                 if (len >= sizeof(mddev->clevel))
2994                         return -ENOSPC;
2995                 strncpy(mddev->clevel, buf, len);
2996                 if (mddev->clevel[len-1] == '\n')
2997                         len--;
2998                 mddev->clevel[len] = 0;
2999                 mddev->level = LEVEL_NONE;
3000                 return rv;
3001         }
3002
3003         /* request to change the personality.  Need to ensure:
3004          *  - array is not engaged in resync/recovery/reshape
3005          *  - old personality can be suspended
3006          *  - new personality will access other array.
3007          */
3008
3009         if (mddev->sync_thread ||
3010             mddev->reshape_position != MaxSector ||
3011             mddev->sysfs_active)
3012                 return -EBUSY;
3013
3014         if (!mddev->pers->quiesce) {
3015                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3016                        mdname(mddev), mddev->pers->name);
3017                 return -EINVAL;
3018         }
3019
3020         /* Now find the new personality */
3021         if (len == 0 || len >= sizeof(clevel))
3022                 return -EINVAL;
3023         strncpy(clevel, buf, len);
3024         if (clevel[len-1] == '\n')
3025                 len--;
3026         clevel[len] = 0;
3027         if (strict_strtol(clevel, 10, &level))
3028                 level = LEVEL_NONE;
3029
3030         if (request_module("md-%s", clevel) != 0)
3031                 request_module("md-level-%s", clevel);
3032         spin_lock(&pers_lock);
3033         pers = find_pers(level, clevel);
3034         if (!pers || !try_module_get(pers->owner)) {
3035                 spin_unlock(&pers_lock);
3036                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3037                 return -EINVAL;
3038         }
3039         spin_unlock(&pers_lock);
3040
3041         if (pers == mddev->pers) {
3042                 /* Nothing to do! */
3043                 module_put(pers->owner);
3044                 return rv;
3045         }
3046         if (!pers->takeover) {
3047                 module_put(pers->owner);
3048                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3049                        mdname(mddev), clevel);
3050                 return -EINVAL;
3051         }
3052
3053         list_for_each_entry(rdev, &mddev->disks, same_set)
3054                 rdev->new_raid_disk = rdev->raid_disk;
3055
3056         /* ->takeover must set new_* and/or delta_disks
3057          * if it succeeds, and may set them when it fails.
3058          */
3059         priv = pers->takeover(mddev);
3060         if (IS_ERR(priv)) {
3061                 mddev->new_level = mddev->level;
3062                 mddev->new_layout = mddev->layout;
3063                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3064                 mddev->raid_disks -= mddev->delta_disks;
3065                 mddev->delta_disks = 0;
3066                 module_put(pers->owner);
3067                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3068                        mdname(mddev), clevel);
3069                 return PTR_ERR(priv);
3070         }
3071
3072         /* Looks like we have a winner */
3073         mddev_suspend(mddev);
3074         mddev->pers->stop(mddev);
3075
3076         if (mddev->pers->sync_request == NULL &&
3077             pers->sync_request != NULL) {
3078                 /* need to add the md_redundancy_group */
3079                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3080                         printk(KERN_WARNING
3081                                "md: cannot register extra attributes for %s\n",
3082                                mdname(mddev));
3083                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3084         }
3085         if (mddev->pers->sync_request != NULL &&
3086             pers->sync_request == NULL) {
3087                 /* need to remove the md_redundancy_group */
3088                 if (mddev->to_remove == NULL)
3089                         mddev->to_remove = &md_redundancy_group;
3090         }
3091
3092         if (mddev->pers->sync_request == NULL &&
3093             mddev->external) {
3094                 /* We are converting from a no-redundancy array
3095                  * to a redundancy array and metadata is managed
3096                  * externally so we need to be sure that writes
3097                  * won't block due to a need to transition
3098                  *      clean->dirty
3099                  * until external management is started.
3100                  */
3101                 mddev->in_sync = 0;
3102                 mddev->safemode_delay = 0;
3103                 mddev->safemode = 0;
3104         }
3105
3106         list_for_each_entry(rdev, &mddev->disks, same_set) {
3107                 char nm[20];
3108                 if (rdev->raid_disk < 0)
3109                         continue;
3110                 if (rdev->new_raid_disk > mddev->raid_disks)
3111                         rdev->new_raid_disk = -1;
3112                 if (rdev->new_raid_disk == rdev->raid_disk)
3113                         continue;
3114                 sprintf(nm, "rd%d", rdev->raid_disk);
3115                 sysfs_remove_link(&mddev->kobj, nm);
3116         }
3117         list_for_each_entry(rdev, &mddev->disks, same_set) {
3118                 if (rdev->raid_disk < 0)
3119                         continue;
3120                 if (rdev->new_raid_disk == rdev->raid_disk)
3121                         continue;
3122                 rdev->raid_disk = rdev->new_raid_disk;
3123                 if (rdev->raid_disk < 0)
3124                         clear_bit(In_sync, &rdev->flags);
3125                 else {
3126                         char nm[20];
3127                         sprintf(nm, "rd%d", rdev->raid_disk);
3128                         if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3129                                 printk("md: cannot register %s for %s after level change\n",
3130                                        nm, mdname(mddev));
3131                 }
3132         }
3133
3134         module_put(mddev->pers->owner);
3135         mddev->pers = pers;
3136         mddev->private = priv;
3137         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3138         mddev->level = mddev->new_level;
3139         mddev->layout = mddev->new_layout;
3140         mddev->chunk_sectors = mddev->new_chunk_sectors;
3141         mddev->delta_disks = 0;
3142         if (mddev->pers->sync_request == NULL) {
3143                 /* this is now an array without redundancy, so
3144                  * it must always be in_sync
3145                  */
3146                 mddev->in_sync = 1;
3147                 del_timer_sync(&mddev->safemode_timer);
3148         }
3149         pers->run(mddev);
3150         mddev_resume(mddev);
3151         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3152         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3153         md_wakeup_thread(mddev->thread);
3154         sysfs_notify(&mddev->kobj, NULL, "level");
3155         md_new_event(mddev);
3156         return rv;
3157 }
3158
3159 static struct md_sysfs_entry md_level =
3160 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3161
3162
3163 static ssize_t
3164 layout_show(mddev_t *mddev, char *page)
3165 {
3166         /* just a number, not meaningful for all levels */
3167         if (mddev->reshape_position != MaxSector &&
3168             mddev->layout != mddev->new_layout)
3169                 return sprintf(page, "%d (%d)\n",
3170                                mddev->new_layout, mddev->layout);
3171         return sprintf(page, "%d\n", mddev->layout);
3172 }
3173
3174 static ssize_t
3175 layout_store(mddev_t *mddev, const char *buf, size_t len)
3176 {
3177         char *e;
3178         unsigned long n = simple_strtoul(buf, &e, 10);
3179
3180         if (!*buf || (*e && *e != '\n'))
3181                 return -EINVAL;
3182
3183         if (mddev->pers) {
3184                 int err;
3185                 if (mddev->pers->check_reshape == NULL)
3186                         return -EBUSY;
3187                 mddev->new_layout = n;
3188                 err = mddev->pers->check_reshape(mddev);
3189                 if (err) {
3190                         mddev->new_layout = mddev->layout;
3191                         return err;
3192                 }
3193         } else {
3194                 mddev->new_layout = n;
3195                 if (mddev->reshape_position == MaxSector)
3196                         mddev->layout = n;
3197         }
3198         return len;
3199 }
3200 static struct md_sysfs_entry md_layout =
3201 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3202
3203
3204 static ssize_t
3205 raid_disks_show(mddev_t *mddev, char *page)
3206 {
3207         if (mddev->raid_disks == 0)
3208                 return 0;
3209         if (mddev->reshape_position != MaxSector &&
3210             mddev->delta_disks != 0)
3211                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3212                                mddev->raid_disks - mddev->delta_disks);
3213         return sprintf(page, "%d\n", mddev->raid_disks);
3214 }
3215
3216 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3217
3218 static ssize_t
3219 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3220 {
3221         char *e;
3222         int rv = 0;
3223         unsigned long n = simple_strtoul(buf, &e, 10);
3224
3225         if (!*buf || (*e && *e != '\n'))
3226                 return -EINVAL;
3227
3228         if (mddev->pers)
3229                 rv = update_raid_disks(mddev, n);
3230         else if (mddev->reshape_position != MaxSector) {
3231                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3232                 mddev->delta_disks = n - olddisks;
3233                 mddev->raid_disks = n;
3234         } else
3235                 mddev->raid_disks = n;
3236         return rv ? rv : len;
3237 }
3238 static struct md_sysfs_entry md_raid_disks =
3239 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3240
3241 static ssize_t
3242 chunk_size_show(mddev_t *mddev, char *page)
3243 {
3244         if (mddev->reshape_position != MaxSector &&
3245             mddev->chunk_sectors != mddev->new_chunk_sectors)
3246                 return sprintf(page, "%d (%d)\n",
3247                                mddev->new_chunk_sectors << 9,
3248                                mddev->chunk_sectors << 9);
3249         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3250 }
3251
3252 static ssize_t
3253 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3254 {
3255         char *e;
3256         unsigned long n = simple_strtoul(buf, &e, 10);
3257
3258         if (!*buf || (*e && *e != '\n'))
3259                 return -EINVAL;
3260
3261         if (mddev->pers) {
3262                 int err;
3263                 if (mddev->pers->check_reshape == NULL)
3264                         return -EBUSY;
3265                 mddev->new_chunk_sectors = n >> 9;
3266                 err = mddev->pers->check_reshape(mddev);
3267                 if (err) {
3268                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3269                         return err;
3270                 }
3271         } else {
3272                 mddev->new_chunk_sectors = n >> 9;
3273                 if (mddev->reshape_position == MaxSector)
3274                         mddev->chunk_sectors = n >> 9;
3275         }
3276         return len;
3277 }
3278 static struct md_sysfs_entry md_chunk_size =
3279 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3280
3281 static ssize_t
3282 resync_start_show(mddev_t *mddev, char *page)
3283 {
3284         if (mddev->recovery_cp == MaxSector)
3285                 return sprintf(page, "none\n");
3286         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3287 }
3288
3289 static ssize_t
3290 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3291 {
3292         char *e;
3293         unsigned long long n = simple_strtoull(buf, &e, 10);
3294
3295         if (mddev->pers)
3296                 return -EBUSY;
3297         if (cmd_match(buf, "none"))
3298                 n = MaxSector;
3299         else if (!*buf || (*e && *e != '\n'))
3300                 return -EINVAL;
3301
3302         mddev->recovery_cp = n;
3303         return len;
3304 }
3305 static struct md_sysfs_entry md_resync_start =
3306 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3307
3308 /*
3309  * The array state can be:
3310  *
3311  * clear
3312  *     No devices, no size, no level
3313  *     Equivalent to STOP_ARRAY ioctl
3314  * inactive
3315  *     May have some settings, but array is not active
3316  *        all IO results in error
3317  *     When written, doesn't tear down array, but just stops it
3318  * suspended (not supported yet)
3319  *     All IO requests will block. The array can be reconfigured.
3320  *     Writing this, if accepted, will block until array is quiescent
3321  * readonly
3322  *     no resync can happen.  no superblocks get written.
3323  *     write requests fail
3324  * read-auto
3325  *     like readonly, but behaves like 'clean' on a write request.
3326  *
3327  * clean - no pending writes, but otherwise active.
3328  *     When written to inactive array, starts without resync
3329  *     If a write request arrives then
3330  *       if metadata is known, mark 'dirty' and switch to 'active'.
3331  *       if not known, block and switch to write-pending
3332  *     If written to an active array that has pending writes, then fails.
3333  * active
3334  *     fully active: IO and resync can be happening.
3335  *     When written to inactive array, starts with resync
3336  *
3337  * write-pending
3338  *     clean, but writes are blocked waiting for 'active' to be written.
3339  *
3340  * active-idle
3341  *     like active, but no writes have been seen for a while (100msec).
3342  *
3343  */
3344 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3345                    write_pending, active_idle, bad_word};
3346 static char *array_states[] = {
3347         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3348         "write-pending", "active-idle", NULL };
3349
3350 static int match_word(const char *word, char **list)
3351 {
3352         int n;
3353         for (n=0; list[n]; n++)
3354                 if (cmd_match(word, list[n]))
3355                         break;
3356         return n;
3357 }
3358
3359 static ssize_t
3360 array_state_show(mddev_t *mddev, char *page)
3361 {
3362         enum array_state st = inactive;
3363
3364         if (mddev->pers)
3365                 switch(mddev->ro) {
3366                 case 1:
3367                         st = readonly;
3368                         break;
3369                 case 2:
3370                         st = read_auto;
3371                         break;
3372                 case 0:
3373                         if (mddev->in_sync)
3374                                 st = clean;
3375                         else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3376                                 st = write_pending;
3377                         else if (mddev->safemode)
3378                                 st = active_idle;
3379                         else
3380                                 st = active;
3381                 }
3382         else {
3383                 if (list_empty(&mddev->disks) &&
3384                     mddev->raid_disks == 0 &&
3385                     mddev->dev_sectors == 0)
3386                         st = clear;
3387                 else
3388                         st = inactive;
3389         }
3390         return sprintf(page, "%s\n", array_states[st]);
3391 }
3392
3393 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3394 static int md_set_readonly(mddev_t * mddev, int is_open);
3395 static int do_md_run(mddev_t * mddev);
3396 static int restart_array(mddev_t *mddev);
3397
3398 static ssize_t
3399 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3400 {
3401         int err = -EINVAL;
3402         enum array_state st = match_word(buf, array_states);
3403         switch(st) {
3404         case bad_word:
3405                 break;
3406         case clear:
3407                 /* stopping an active array */
3408                 if (atomic_read(&mddev->openers) > 0)
3409                         return -EBUSY;
3410                 err = do_md_stop(mddev, 0, 0);
3411                 break;
3412         case inactive:
3413                 /* stopping an active array */
3414                 if (mddev->pers) {
3415                         if (atomic_read(&mddev->openers) > 0)
3416                                 return -EBUSY;
3417                         err = do_md_stop(mddev, 2, 0);
3418                 } else
3419                         err = 0; /* already inactive */
3420                 break;
3421         case suspended:
3422                 break; /* not supported yet */
3423         case readonly:
3424                 if (mddev->pers)
3425                         err = md_set_readonly(mddev, 0);
3426                 else {
3427                         mddev->ro = 1;
3428                         set_disk_ro(mddev->gendisk, 1);
3429                         err = do_md_run(mddev);
3430                 }
3431                 break;
3432         case read_auto:
3433                 if (mddev->pers) {
3434                         if (mddev->ro == 0)
3435                                 err = md_set_readonly(mddev, 0);
3436                         else if (mddev->ro == 1)
3437                                 err = restart_array(mddev);
3438                         if (err == 0) {
3439                                 mddev->ro = 2;
3440                                 set_disk_ro(mddev->gendisk, 0);
3441                         }
3442                 } else {
3443                         mddev->ro = 2;
3444                         err = do_md_run(mddev);
3445                 }
3446                 break;
3447         case clean:
3448                 if (mddev->pers) {
3449                         restart_array(mddev);
3450                         spin_lock_irq(&mddev->write_lock);
3451                         if (atomic_read(&mddev->writes_pending) == 0) {
3452                                 if (mddev->in_sync == 0) {
3453                                         mddev->in_sync = 1;
3454                                         if (mddev->safemode == 1)
3455                                                 mddev->safemode = 0;
3456                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3457                                 }
3458                                 err = 0;
3459                         } else
3460                                 err = -EBUSY;
3461                         spin_unlock_irq(&mddev->write_lock);
3462                 } else
3463                         err = -EINVAL;
3464                 break;
3465         case active:
3466                 if (mddev->pers) {
3467                         restart_array(mddev);
3468                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3469                         wake_up(&mddev->sb_wait);
3470                         err = 0;
3471                 } else {
3472                         mddev->ro = 0;
3473                         set_disk_ro(mddev->gendisk, 0);
3474                         err = do_md_run(mddev);
3475                 }
3476                 break;
3477         case write_pending:
3478         case active_idle:
3479                 /* these cannot be set */
3480                 break;
3481         }
3482         if (err)
3483                 return err;
3484         else {
3485                 sysfs_notify_dirent_safe(mddev->sysfs_state);
3486                 return len;
3487         }
3488 }
3489 static struct md_sysfs_entry md_array_state =
3490 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3491
3492 static ssize_t
3493 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3494         return sprintf(page, "%d\n",
3495                        atomic_read(&mddev->max_corr_read_errors));
3496 }
3497
3498 static ssize_t
3499 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3500 {
3501         char *e;
3502         unsigned long n = simple_strtoul(buf, &e, 10);
3503
3504         if (*buf && (*e == 0 || *e == '\n')) {
3505                 atomic_set(&mddev->max_corr_read_errors, n);
3506                 return len;
3507         }
3508         return -EINVAL;
3509 }
3510
3511 static struct md_sysfs_entry max_corr_read_errors =
3512 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3513         max_corrected_read_errors_store);
3514
3515 static ssize_t
3516 null_show(mddev_t *mddev, char *page)
3517 {
3518         return -EINVAL;
3519 }
3520
3521 static ssize_t
3522 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3523 {
3524         /* buf must be %d:%d\n? giving major and minor numbers */
3525         /* The new device is added to the array.
3526          * If the array has a persistent superblock, we read the
3527          * superblock to initialise info and check validity.
3528          * Otherwise, only checking done is that in bind_rdev_to_array,
3529          * which mainly checks size.
3530          */
3531         char *e;
3532         int major = simple_strtoul(buf, &e, 10);
3533         int minor;
3534         dev_t dev;
3535         mdk_rdev_t *rdev;
3536         int err;
3537
3538         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3539                 return -EINVAL;
3540         minor = simple_strtoul(e+1, &e, 10);
3541         if (*e && *e != '\n')
3542                 return -EINVAL;
3543         dev = MKDEV(major, minor);
3544         if (major != MAJOR(dev) ||
3545             minor != MINOR(dev))
3546                 return -EOVERFLOW;
3547
3548
3549         if (mddev->persistent) {
3550                 rdev = md_import_device(dev, mddev->major_version,
3551                                         mddev->minor_version);
3552                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3553                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3554                                                        mdk_rdev_t, same_set);
3555                         err = super_types[mddev->major_version]
3556                                 .load_super(rdev, rdev0, mddev->minor_version);
3557                         if (err < 0)
3558                                 goto out;
3559                 }
3560         } else if (mddev->external)
3561                 rdev = md_import_device(dev, -2, -1);
3562         else
3563                 rdev = md_import_device(dev, -1, -1);
3564
3565         if (IS_ERR(rdev))
3566                 return PTR_ERR(rdev);
3567         err = bind_rdev_to_array(rdev, mddev);
3568  out:
3569         if (err)
3570                 export_rdev(rdev);
3571         return err ? err : len;
3572 }
3573
3574 static struct md_sysfs_entry md_new_device =
3575 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3576
3577 static ssize_t
3578 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3579 {
3580         char *end;
3581         unsigned long chunk, end_chunk;
3582
3583         if (!mddev->bitmap)
3584                 goto out;
3585         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3586         while (*buf) {
3587                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3588                 if (buf == end) break;
3589                 if (*end == '-') { /* range */
3590                         buf = end + 1;
3591                         end_chunk = simple_strtoul(buf, &end, 0);
3592                         if (buf == end) break;
3593                 }
3594                 if (*end && !isspace(*end)) break;
3595                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3596                 buf = skip_spaces(end);
3597         }
3598         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3599 out:
3600         return len;
3601 }
3602
3603 static struct md_sysfs_entry md_bitmap =
3604 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3605
3606 static ssize_t
3607 size_show(mddev_t *mddev, char *page)
3608 {
3609         return sprintf(page, "%llu\n",
3610                 (unsigned long long)mddev->dev_sectors / 2);
3611 }
3612
3613 static int update_size(mddev_t *mddev, sector_t num_sectors);
3614
3615 static ssize_t
3616 size_store(mddev_t *mddev, const char *buf, size_t len)
3617 {
3618         /* If array is inactive, we can reduce the component size, but
3619          * not increase it (except from 0).
3620          * If array is active, we can try an on-line resize
3621          */
3622         sector_t sectors;
3623         int err = strict_blocks_to_sectors(buf, &sectors);
3624
3625         if (err < 0)
3626                 return err;
3627         if (mddev->pers) {
3628                 err = update_size(mddev, sectors);
3629                 md_update_sb(mddev, 1);
3630         } else {
3631                 if (mddev->dev_sectors == 0 ||
3632                     mddev->dev_sectors > sectors)
3633                         mddev->dev_sectors = sectors;
3634                 else
3635                         err = -ENOSPC;
3636         }
3637         return err ? err : len;
3638 }
3639
3640 static struct md_sysfs_entry md_size =
3641 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3642
3643
3644 /* Metdata version.
3645  * This is one of
3646  *   'none' for arrays with no metadata (good luck...)
3647  *   'external' for arrays with externally managed metadata,
3648  * or N.M for internally known formats
3649  */
3650 static ssize_t
3651 metadata_show(mddev_t *mddev, char *page)
3652 {
3653         if (mddev->persistent)
3654                 return sprintf(page, "%d.%d\n",
3655                                mddev->major_version, mddev->minor_version);
3656         else if (mddev->external)
3657                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3658         else
3659                 return sprintf(page, "none\n");
3660 }
3661
3662 static ssize_t
3663 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3664 {
3665         int major, minor;
3666         char *e;
3667         /* Changing the details of 'external' metadata is
3668          * always permitted.  Otherwise there must be
3669          * no devices attached to the array.
3670          */
3671         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3672                 ;
3673         else if (!list_empty(&mddev->disks))
3674                 return -EBUSY;
3675
3676         if (cmd_match(buf, "none")) {
3677                 mddev->persistent = 0;
3678                 mddev->external = 0;
3679                 mddev->major_version = 0;
3680                 mddev->minor_version = 90;
3681                 return len;
3682         }
3683         if (strncmp(buf, "external:", 9) == 0) {
3684                 size_t namelen = len-9;
3685                 if (namelen >= sizeof(mddev->metadata_type))
3686                         namelen = sizeof(mddev->metadata_type)-1;
3687                 strncpy(mddev->metadata_type, buf+9, namelen);
3688                 mddev->metadata_type[namelen] = 0;
3689                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3690                         mddev->metadata_type[--namelen] = 0;
3691                 mddev->persistent = 0;
3692                 mddev->external = 1;
3693                 mddev->major_version = 0;
3694                 mddev->minor_version = 90;
3695                 return len;
3696         }
3697         major = simple_strtoul(buf, &e, 10);
3698         if (e==buf || *e != '.')
3699                 return -EINVAL;
3700         buf = e+1;
3701         minor = simple_strtoul(buf, &e, 10);
3702         if (e==buf || (*e && *e != '\n') )
3703                 return -EINVAL;
3704         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3705                 return -ENOENT;
3706         mddev->major_version = major;
3707         mddev->minor_version = minor;
3708         mddev->persistent = 1;
3709         mddev->external = 0;
3710         return len;
3711 }
3712
3713 static struct md_sysfs_entry md_metadata =
3714 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3715
3716 static ssize_t
3717 action_show(mddev_t *mddev, char *page)
3718 {
3719         char *type = "idle";
3720         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3721                 type = "frozen";
3722         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3723             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3724                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3725                         type = "reshape";
3726                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3727                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3728                                 type = "resync";
3729                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3730                                 type = "check";
3731                         else
3732                                 type = "repair";
3733                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3734                         type = "recover";
3735         }
3736         return sprintf(page, "%s\n", type);
3737 }
3738
3739 static ssize_t
3740 action_store(mddev_t *mddev, const char *page, size_t len)
3741 {
3742         if (!mddev->pers || !mddev->pers->sync_request)
3743                 return -EINVAL;
3744
3745         if (cmd_match(page, "frozen"))
3746                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3747         else
3748                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3749
3750         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3751                 if (mddev->sync_thread) {
3752                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3753                         md_unregister_thread(mddev->sync_thread);
3754                         mddev->sync_thread = NULL;
3755                         mddev->recovery = 0;
3756                 }
3757         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3758                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3759                 return -EBUSY;
3760         else if (cmd_match(page, "resync"))
3761                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3762         else if (cmd_match(page, "recover")) {
3763                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3764                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3765         } else if (cmd_match(page, "reshape")) {
3766                 int err;
3767                 if (mddev->pers->start_reshape == NULL)
3768                         return -EINVAL;
3769                 err = mddev->pers->start_reshape(mddev);
3770                 if (err)
3771                         return err;
3772                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3773         } else {
3774                 if (cmd_match(page, "check"))
3775                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3776                 else if (!cmd_match(page, "repair"))
3777                         return -EINVAL;
3778                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3779                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3780         }
3781         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3782         md_wakeup_thread(mddev->thread);
3783         sysfs_notify_dirent_safe(mddev->sysfs_action);
3784         return len;
3785 }
3786
3787 static ssize_t
3788 mismatch_cnt_show(mddev_t *mddev, char *page)
3789 {
3790         return sprintf(page, "%llu\n",
3791                        (unsigned long long) mddev->resync_mismatches);
3792 }
3793
3794 static struct md_sysfs_entry md_scan_mode =
3795 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3796
3797
3798 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3799
3800 static ssize_t
3801 sync_min_show(mddev_t *mddev, char *page)
3802 {
3803         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3804                        mddev->sync_speed_min ? "local": "system");
3805 }
3806
3807 static ssize_t
3808 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3809 {
3810         int min;
3811         char *e;
3812         if (strncmp(buf, "system", 6)==0) {
3813                 mddev->sync_speed_min = 0;
3814                 return len;
3815         }
3816         min = simple_strtoul(buf, &e, 10);
3817         if (buf == e || (*e && *e != '\n') || min <= 0)
3818                 return -EINVAL;
3819         mddev->sync_speed_min = min;
3820         return len;
3821 }
3822
3823 static struct md_sysfs_entry md_sync_min =
3824 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3825
3826 static ssize_t
3827 sync_max_show(mddev_t *mddev, char *page)
3828 {
3829         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3830                        mddev->sync_speed_max ? "local": "system");
3831 }
3832
3833 static ssize_t
3834 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3835 {
3836         int max;
3837         char *e;
3838         if (strncmp(buf, "system", 6)==0) {
3839                 mddev->sync_speed_max = 0;
3840                 return len;
3841         }
3842         max = simple_strtoul(buf, &e, 10);
3843         if (buf == e || (*e && *e != '\n') || max <= 0)
3844                 return -EINVAL;
3845         mddev->sync_speed_max = max;
3846         return len;
3847 }
3848
3849 static struct md_sysfs_entry md_sync_max =
3850 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3851
3852 static ssize_t
3853 degraded_show(mddev_t *mddev, char *page)
3854 {
3855         return sprintf(page, "%d\n", mddev->degraded);
3856 }
3857 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3858
3859 static ssize_t
3860 sync_force_parallel_show(mddev_t *mddev, char *page)
3861 {
3862         return sprintf(page, "%d\n", mddev->parallel_resync);
3863 }
3864
3865 static ssize_t
3866 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3867 {
3868         long n;
3869
3870         if (strict_strtol(buf, 10, &n))
3871                 return -EINVAL;
3872
3873         if (n != 0 && n != 1)
3874                 return -EINVAL;
3875
3876         mddev->parallel_resync = n;
3877
3878         if (mddev->sync_thread)
3879                 wake_up(&resync_wait);
3880
3881         return len;
3882 }
3883
3884 /* force parallel resync, even with shared block devices */
3885 static struct md_sysfs_entry md_sync_force_parallel =
3886 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3887        sync_force_parallel_show, sync_force_parallel_store);
3888
3889 static ssize_t
3890 sync_speed_show(mddev_t *mddev, char *page)
3891 {
3892         unsigned long resync, dt, db;
3893         if (mddev->curr_resync == 0)
3894                 return sprintf(page, "none\n");
3895         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3896         dt = (jiffies - mddev->resync_mark) / HZ;
3897         if (!dt) dt++;
3898         db = resync - mddev->resync_mark_cnt;
3899         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3900 }
3901
3902 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3903
3904 static ssize_t
3905 sync_completed_show(mddev_t *mddev, char *page)
3906 {
3907         unsigned long max_sectors, resync;
3908
3909         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3910                 return sprintf(page, "none\n");
3911
3912         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3913                 max_sectors = mddev->resync_max_sectors;
3914         else
3915                 max_sectors = mddev->dev_sectors;
3916
3917         resync = mddev->curr_resync_completed;
3918         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3919 }
3920
3921 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3922
3923 static ssize_t
3924 min_sync_show(mddev_t *mddev, char *page)
3925 {
3926         return sprintf(page, "%llu\n",
3927                        (unsigned long long)mddev->resync_min);
3928 }
3929 static ssize_t
3930 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3931 {
3932         unsigned long long min;
3933         if (strict_strtoull(buf, 10, &min))
3934                 return -EINVAL;
3935         if (min > mddev->resync_max)
3936                 return -EINVAL;
3937         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3938                 return -EBUSY;
3939
3940         /* Must be a multiple of chunk_size */
3941         if (mddev->chunk_sectors) {
3942                 sector_t temp = min;
3943                 if (sector_div(temp, mddev->chunk_sectors))
3944                         return -EINVAL;
3945         }
3946         mddev->resync_min = min;
3947
3948         return len;
3949 }
3950
3951 static struct md_sysfs_entry md_min_sync =
3952 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3953
3954 static ssize_t
3955 max_sync_show(mddev_t *mddev, char *page)
3956 {
3957         if (mddev->resync_max == MaxSector)
3958                 return sprintf(page, "max\n");
3959         else
3960                 return sprintf(page, "%llu\n",
3961                                (unsigned long long)mddev->resync_max);
3962 }
3963 static ssize_t
3964 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3965 {
3966         if (strncmp(buf, "max", 3) == 0)
3967                 mddev->resync_max = MaxSector;
3968         else {
3969                 unsigned long long max;
3970                 if (strict_strtoull(buf, 10, &max))
3971                         return -EINVAL;
3972                 if (max < mddev->resync_min)
3973                         return -EINVAL;
3974                 if (max < mddev->resync_max &&
3975                     mddev->ro == 0 &&
3976                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3977                         return -EBUSY;
3978
3979                 /* Must be a multiple of chunk_size */
3980                 if (mddev->chunk_sectors) {
3981                         sector_t temp = max;
3982                         if (sector_div(temp, mddev->chunk_sectors))
3983                                 return -EINVAL;
3984                 }
3985                 mddev->resync_max = max;
3986         }
3987         wake_up(&mddev->recovery_wait);
3988         return len;
3989 }
3990
3991 static struct md_sysfs_entry md_max_sync =
3992 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3993
3994 static ssize_t
3995 suspend_lo_show(mddev_t *mddev, char *page)
3996 {
3997         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3998 }
3999
4000 static ssize_t
4001 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4002 {
4003         char *e;
4004         unsigned long long new = simple_strtoull(buf, &e, 10);
4005
4006         if (mddev->pers == NULL ||
4007             mddev->pers->quiesce == NULL)
4008                 return -EINVAL;
4009         if (buf == e || (*e && *e != '\n'))
4010                 return -EINVAL;
4011         if (new >= mddev->suspend_hi ||
4012             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4013                 mddev->suspend_lo = new;
4014                 mddev->pers->quiesce(mddev, 2);
4015                 return len;
4016         } else
4017                 return -EINVAL;
4018 }
4019 static struct md_sysfs_entry md_suspend_lo =
4020 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4021
4022
4023 static ssize_t
4024 suspend_hi_show(mddev_t *mddev, char *page)
4025 {
4026         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4027 }
4028
4029 static ssize_t
4030 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4031 {
4032         char *e;
4033         unsigned long long new = simple_strtoull(buf, &e, 10);
4034
4035         if (mddev->pers == NULL ||
4036             mddev->pers->quiesce == NULL)
4037                 return -EINVAL;
4038         if (buf == e || (*e && *e != '\n'))
4039                 return -EINVAL;
4040         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4041             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4042                 mddev->suspend_hi = new;
4043                 mddev->pers->quiesce(mddev, 1);
4044                 mddev->pers->quiesce(mddev, 0);
4045                 return len;
4046         } else
4047                 return -EINVAL;
4048 }
4049 static struct md_sysfs_entry md_suspend_hi =
4050 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4051
4052 static ssize_t
4053 reshape_position_show(mddev_t *mddev, char *page)
4054 {
4055         if (mddev->reshape_position != MaxSector)
4056                 return sprintf(page, "%llu\n",
4057                                (unsigned long long)mddev->reshape_position);
4058         strcpy(page, "none\n");
4059         return 5;
4060 }
4061
4062 static ssize_t
4063 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4064 {
4065         char *e;
4066         unsigned long long new = simple_strtoull(buf, &e, 10);
4067         if (mddev->pers)
4068                 return -EBUSY;
4069         if (buf == e || (*e && *e != '\n'))
4070                 return -EINVAL;
4071         mddev->reshape_position = new;
4072         mddev->delta_disks = 0;
4073         mddev->new_level = mddev->level;
4074         mddev->new_layout = mddev->layout;
4075         mddev->new_chunk_sectors = mddev->chunk_sectors;
4076         return len;
4077 }
4078
4079 static struct md_sysfs_entry md_reshape_position =
4080 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4081        reshape_position_store);
4082
4083 static ssize_t
4084 array_size_show(mddev_t *mddev, char *page)
4085 {
4086         if (mddev->external_size)
4087                 return sprintf(page, "%llu\n",
4088                                (unsigned long long)mddev->array_sectors/2);
4089         else
4090                 return sprintf(page, "default\n");
4091 }
4092
4093 static ssize_t
4094 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4095 {
4096         sector_t sectors;
4097
4098         if (strncmp(buf, "default", 7) == 0) {
4099                 if (mddev->pers)
4100                         sectors = mddev->pers->size(mddev, 0, 0);
4101                 else
4102                         sectors = mddev->array_sectors;
4103
4104                 mddev->external_size = 0;
4105         } else {
4106                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4107                         return -EINVAL;
4108                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4109                         return -E2BIG;
4110
4111                 mddev->external_size = 1;
4112         }
4113
4114         mddev->array_sectors = sectors;
4115         set_capacity(mddev->gendisk, mddev->array_sectors);
4116         if (mddev->pers)
4117                 revalidate_disk(mddev->gendisk);
4118
4119         return len;
4120 }
4121
4122 static struct md_sysfs_entry md_array_size =
4123 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4124        array_size_store);
4125
4126 static struct attribute *md_default_attrs[] = {
4127         &md_level.attr,
4128         &md_layout.attr,
4129         &md_raid_disks.attr,
4130         &md_chunk_size.attr,
4131         &md_size.attr,
4132         &md_resync_start.attr,
4133         &md_metadata.attr,
4134         &md_new_device.attr,
4135         &md_safe_delay.attr,
4136         &md_array_state.attr,
4137         &md_reshape_position.attr,
4138         &md_array_size.attr,
4139         &max_corr_read_errors.attr,
4140         NULL,
4141 };
4142
4143 static struct attribute *md_redundancy_attrs[] = {
4144         &md_scan_mode.attr,
4145         &md_mismatches.attr,
4146         &md_sync_min.attr,
4147         &md_sync_max.attr,
4148         &md_sync_speed.attr,
4149         &md_sync_force_parallel.attr,
4150         &md_sync_completed.attr,
4151         &md_min_sync.attr,
4152         &md_max_sync.attr,
4153         &md_suspend_lo.attr,
4154         &md_suspend_hi.attr,
4155         &md_bitmap.attr,
4156         &md_degraded.attr,
4157         NULL,
4158 };
4159 static struct attribute_group md_redundancy_group = {
4160         .name = NULL,
4161         .attrs = md_redundancy_attrs,
4162 };
4163
4164
4165 static ssize_t
4166 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4167 {
4168         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4169         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4170         ssize_t rv;
4171
4172         if (!entry->show)
4173                 return -EIO;
4174         rv = mddev_lock(mddev);
4175         if (!rv) {
4176                 rv = entry->show(mddev, page);
4177                 mddev_unlock(mddev);
4178         }
4179         return rv;
4180 }
4181
4182 static ssize_t
4183 md_attr_store(struct kobject *kobj, struct attribute *attr,
4184               const char *page, size_t length)
4185 {
4186         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4187         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4188         ssize_t rv;
4189
4190         if (!entry->store)
4191                 return -EIO;
4192         if (!capable(CAP_SYS_ADMIN))
4193                 return -EACCES;
4194         rv = mddev_lock(mddev);
4195         if (mddev->hold_active == UNTIL_IOCTL)
4196                 mddev->hold_active = 0;
4197         if (!rv) {
4198                 rv = entry->store(mddev, page, length);
4199                 mddev_unlock(mddev);
4200         }
4201         return rv;
4202 }
4203
4204 static void md_free(struct kobject *ko)
4205 {
4206         mddev_t *mddev = container_of(ko, mddev_t, kobj);
4207
4208         if (mddev->sysfs_state)
4209                 sysfs_put(mddev->sysfs_state);
4210
4211         if (mddev->gendisk) {
4212                 del_gendisk(mddev->gendisk);
4213                 put_disk(mddev->gendisk);
4214         }
4215         if (mddev->queue)
4216                 blk_cleanup_queue(mddev->queue);
4217
4218         kfree(mddev);
4219 }
4220
4221 static const struct sysfs_ops md_sysfs_ops = {
4222         .show   = md_attr_show,
4223         .store  = md_attr_store,
4224 };
4225 static struct kobj_type md_ktype = {
4226         .release        = md_free,
4227         .sysfs_ops      = &md_sysfs_ops,
4228         .default_attrs  = md_default_attrs,
4229 };
4230
4231 int mdp_major = 0;
4232
4233 static void mddev_delayed_delete(struct work_struct *ws)
4234 {
4235         mddev_t *mddev = container_of(ws, mddev_t, del_work);
4236
4237         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4238         kobject_del(&mddev->kobj);
4239         kobject_put(&mddev->kobj);
4240 }
4241
4242 static int md_alloc(dev_t dev, char *name)
4243 {
4244         static DEFINE_MUTEX(disks_mutex);
4245         mddev_t *mddev = mddev_find(dev);
4246         struct gendisk *disk;
4247         int partitioned;
4248         int shift;
4249         int unit;
4250         int error;
4251
4252         if (!mddev)
4253                 return -ENODEV;
4254
4255         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4256         shift = partitioned ? MdpMinorShift : 0;
4257         unit = MINOR(mddev->unit) >> shift;
4258
4259         /* wait for any previous instance of this device to be
4260          * completely removed (mddev_delayed_delete).
4261          */
4262         flush_workqueue(md_misc_wq);
4263
4264         mutex_lock(&disks_mutex);
4265         error = -EEXIST;
4266         if (mddev->gendisk)
4267                 goto abort;
4268
4269         if (name) {
4270                 /* Need to ensure that 'name' is not a duplicate.
4271                  */
4272                 mddev_t *mddev2;
4273                 spin_lock(&all_mddevs_lock);
4274
4275                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4276                         if (mddev2->gendisk &&
4277                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4278                                 spin_unlock(&all_mddevs_lock);
4279                                 goto abort;
4280                         }
4281                 spin_unlock(&all_mddevs_lock);
4282         }
4283
4284         error = -ENOMEM;
4285         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4286         if (!mddev->queue)
4287                 goto abort;
4288         mddev->queue->queuedata = mddev;
4289
4290         blk_queue_make_request(mddev->queue, md_make_request);
4291
4292         disk = alloc_disk(1 << shift);
4293         if (!disk) {
4294                 blk_cleanup_queue(mddev->queue);
4295                 mddev->queue = NULL;
4296                 goto abort;
4297         }
4298         disk->major = MAJOR(mddev->unit);
4299         disk->first_minor = unit << shift;
4300         if (name)
4301                 strcpy(disk->disk_name, name);
4302         else if (partitioned)
4303                 sprintf(disk->disk_name, "md_d%d", unit);
4304         else
4305                 sprintf(disk->disk_name, "md%d", unit);
4306         disk->fops = &md_fops;
4307         disk->private_data = mddev;
4308         disk->queue = mddev->queue;
4309         /* Allow extended partitions.  This makes the
4310          * 'mdp' device redundant, but we can't really
4311          * remove it now.
4312          */
4313         disk->flags |= GENHD_FL_EXT_DEVT;
4314         add_disk(disk);
4315         mddev->gendisk = disk;
4316         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4317                                      &disk_to_dev(disk)->kobj, "%s", "md");
4318         if (error) {
4319                 /* This isn't possible, but as kobject_init_and_add is marked
4320                  * __must_check, we must do something with the result
4321                  */
4322                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4323                        disk->disk_name);
4324                 error = 0;
4325         }
4326         if (mddev->kobj.sd &&
4327             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4328                 printk(KERN_DEBUG "pointless warning\n");
4329
4330         blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4331  abort:
4332         mutex_unlock(&disks_mutex);
4333         if (!error && mddev->kobj.sd) {
4334                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4335                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4336         }
4337         mddev_put(mddev);
4338         return error;
4339 }
4340
4341 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4342 {
4343         md_alloc(dev, NULL);
4344         return NULL;
4345 }
4346
4347 static int add_named_array(const char *val, struct kernel_param *kp)
4348 {
4349         /* val must be "md_*" where * is not all digits.
4350          * We allocate an array with a large free minor number, and
4351          * set the name to val.  val must not already be an active name.
4352          */
4353         int len = strlen(val);
4354         char buf[DISK_NAME_LEN];
4355
4356         while (len && val[len-1] == '\n')
4357                 len--;
4358         if (len >= DISK_NAME_LEN)
4359                 return -E2BIG;
4360         strlcpy(buf, val, len+1);
4361         if (strncmp(buf, "md_", 3) != 0)
4362                 return -EINVAL;
4363         return md_alloc(0, buf);
4364 }
4365
4366 static void md_safemode_timeout(unsigned long data)
4367 {
4368         mddev_t *mddev = (mddev_t *) data;
4369
4370         if (!atomic_read(&mddev->writes_pending)) {
4371                 mddev->safemode = 1;
4372                 if (mddev->external)
4373                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4374         }
4375         md_wakeup_thread(mddev->thread);
4376 }
4377
4378 static int start_dirty_degraded;
4379
4380 int md_run(mddev_t *mddev)
4381 {
4382         int err;
4383         mdk_rdev_t *rdev;
4384         struct mdk_personality *pers;
4385
4386         if (list_empty(&mddev->disks))
4387                 /* cannot run an array with no devices.. */
4388                 return -EINVAL;
4389
4390         if (mddev->pers)
4391                 return -EBUSY;
4392         /* Cannot run until previous stop completes properly */
4393         if (mddev->sysfs_active)
4394                 return -EBUSY;
4395
4396         /*
4397          * Analyze all RAID superblock(s)
4398          */
4399         if (!mddev->raid_disks) {
4400                 if (!mddev->persistent)
4401                         return -EINVAL;
4402                 analyze_sbs(mddev);
4403         }
4404
4405         if (mddev->level != LEVEL_NONE)
4406                 request_module("md-level-%d", mddev->level);
4407         else if (mddev->clevel[0])
4408                 request_module("md-%s", mddev->clevel);
4409
4410         /*
4411          * Drop all container device buffers, from now on
4412          * the only valid external interface is through the md
4413          * device.
4414          */
4415         list_for_each_entry(rdev, &mddev->disks, same_set) {
4416                 if (test_bit(Faulty, &rdev->flags))
4417                         continue;
4418                 sync_blockdev(rdev->bdev);
4419                 invalidate_bdev(rdev->bdev);
4420
4421                 /* perform some consistency tests on the device.
4422                  * We don't want the data to overlap the metadata,
4423                  * Internal Bitmap issues have been handled elsewhere.
4424                  */
4425                 if (rdev->data_offset < rdev->sb_start) {
4426                         if (mddev->dev_sectors &&
4427                             rdev->data_offset + mddev->dev_sectors
4428                             > rdev->sb_start) {
4429                                 printk("md: %s: data overlaps metadata\n",
4430                                        mdname(mddev));
4431                                 return -EINVAL;
4432                         }
4433                 } else {
4434                         if (rdev->sb_start + rdev->sb_size/512
4435                             > rdev->data_offset) {
4436                                 printk("md: %s: metadata overlaps data\n",
4437                                        mdname(mddev));
4438                                 return -EINVAL;
4439                         }
4440                 }
4441                 sysfs_notify_dirent_safe(rdev->sysfs_state);
4442         }
4443
4444         if (mddev->bio_set == NULL)
4445                 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4446
4447         spin_lock(&pers_lock);
4448         pers = find_pers(mddev->level, mddev->clevel);
4449         if (!pers || !try_module_get(pers->owner)) {
4450                 spin_unlock(&pers_lock);
4451                 if (mddev->level != LEVEL_NONE)
4452                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4453                                mddev->level);
4454                 else
4455                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4456                                mddev->clevel);
4457                 return -EINVAL;
4458         }
4459         mddev->pers = pers;
4460         spin_unlock(&pers_lock);
4461         if (mddev->level != pers->level) {
4462                 mddev->level = pers->level;
4463                 mddev->new_level = pers->level;
4464         }
4465         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4466
4467         if (mddev->reshape_position != MaxSector &&
4468             pers->start_reshape == NULL) {
4469                 /* This personality cannot handle reshaping... */
4470                 mddev->pers = NULL;
4471                 module_put(pers->owner);
4472                 return -EINVAL;
4473         }
4474
4475         if (pers->sync_request) {
4476                 /* Warn if this is a potentially silly
4477                  * configuration.
4478                  */
4479                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4480                 mdk_rdev_t *rdev2;
4481                 int warned = 0;
4482
4483                 list_for_each_entry(rdev, &mddev->disks, same_set)
4484                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4485                                 if (rdev < rdev2 &&
4486                                     rdev->bdev->bd_contains ==
4487                                     rdev2->bdev->bd_contains) {
4488                                         printk(KERN_WARNING
4489                                                "%s: WARNING: %s appears to be"
4490                                                " on the same physical disk as"
4491                                                " %s.\n",
4492                                                mdname(mddev),
4493                                                bdevname(rdev->bdev,b),
4494                                                bdevname(rdev2->bdev,b2));
4495                                         warned = 1;
4496                                 }
4497                         }
4498
4499                 if (warned)
4500                         printk(KERN_WARNING
4501                                "True protection against single-disk"
4502                                " failure might be compromised.\n");
4503         }
4504
4505         mddev->recovery = 0;
4506         /* may be over-ridden by personality */
4507         mddev->resync_max_sectors = mddev->dev_sectors;
4508
4509         mddev->ok_start_degraded = start_dirty_degraded;
4510
4511         if (start_readonly && mddev->ro == 0)
4512                 mddev->ro = 2; /* read-only, but switch on first write */
4513
4514         err = mddev->pers->run(mddev);
4515         if (err)
4516                 printk(KERN_ERR "md: pers->run() failed ...\n");
4517         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4518                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4519                           " but 'external_size' not in effect?\n", __func__);
4520                 printk(KERN_ERR
4521                        "md: invalid array_size %llu > default size %llu\n",
4522                        (unsigned long long)mddev->array_sectors / 2,
4523                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4524                 err = -EINVAL;
4525                 mddev->pers->stop(mddev);
4526         }
4527         if (err == 0 && mddev->pers->sync_request) {
4528                 err = bitmap_create(mddev);
4529                 if (err) {
4530                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4531                                mdname(mddev), err);
4532                         mddev->pers->stop(mddev);
4533                 }
4534         }
4535         if (err) {
4536                 module_put(mddev->pers->owner);
4537                 mddev->pers = NULL;
4538                 bitmap_destroy(mddev);
4539                 return err;
4540         }
4541         if (mddev->pers->sync_request) {
4542                 if (mddev->kobj.sd &&
4543                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4544                         printk(KERN_WARNING
4545                                "md: cannot register extra attributes for %s\n",
4546                                mdname(mddev));
4547                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4548         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4549                 mddev->ro = 0;
4550
4551         atomic_set(&mddev->writes_pending,0);
4552         atomic_set(&mddev->max_corr_read_errors,
4553                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4554         mddev->safemode = 0;
4555         mddev->safemode_timer.function = md_safemode_timeout;
4556         mddev->safemode_timer.data = (unsigned long) mddev;
4557         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4558         mddev->in_sync = 1;
4559
4560         list_for_each_entry(rdev, &mddev->disks, same_set)
4561                 if (rdev->raid_disk >= 0) {
4562                         char nm[20];
4563                         sprintf(nm, "rd%d", rdev->raid_disk);
4564                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4565                                 /* failure here is OK */;
4566                 }
4567
4568         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4569
4570         if (mddev->flags)
4571                 md_update_sb(mddev, 0);
4572
4573         md_wakeup_thread(mddev->thread);
4574         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4575
4576         md_new_event(mddev);
4577         sysfs_notify_dirent_safe(mddev->sysfs_state);
4578         sysfs_notify_dirent_safe(mddev->sysfs_action);
4579         sysfs_notify(&mddev->kobj, NULL, "degraded");
4580         return 0;
4581 }
4582 EXPORT_SYMBOL_GPL(md_run);
4583
4584 static int do_md_run(mddev_t *mddev)
4585 {
4586         int err;
4587
4588         err = md_run(mddev);
4589         if (err)
4590                 goto out;
4591         err = bitmap_load(mddev);
4592         if (err) {
4593                 bitmap_destroy(mddev);
4594                 goto out;
4595         }
4596         set_capacity(mddev->gendisk, mddev->array_sectors);
4597         revalidate_disk(mddev->gendisk);
4598         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4599 out:
4600         return err;
4601 }
4602
4603 static int restart_array(mddev_t *mddev)
4604 {
4605         struct gendisk *disk = mddev->gendisk;
4606
4607         /* Complain if it has no devices */
4608         if (list_empty(&mddev->disks))
4609                 return -ENXIO;
4610         if (!mddev->pers)
4611                 return -EINVAL;
4612         if (!mddev->ro)
4613                 return -EBUSY;
4614         mddev->safemode = 0;
4615         mddev->ro = 0;
4616         set_disk_ro(disk, 0);
4617         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4618                 mdname(mddev));
4619         /* Kick recovery or resync if necessary */
4620         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4621         md_wakeup_thread(mddev->thread);
4622         md_wakeup_thread(mddev->sync_thread);
4623         sysfs_notify_dirent_safe(mddev->sysfs_state);
4624         return 0;
4625 }
4626
4627 /* similar to deny_write_access, but accounts for our holding a reference
4628  * to the file ourselves */
4629 static int deny_bitmap_write_access(struct file * file)
4630 {
4631         struct inode *inode = file->f_mapping->host;
4632
4633         spin_lock(&inode->i_lock);
4634         if (atomic_read(&inode->i_writecount) > 1) {
4635                 spin_unlock(&inode->i_lock);
4636                 return -ETXTBSY;
4637         }
4638         atomic_set(&inode->i_writecount, -1);
4639         spin_unlock(&inode->i_lock);
4640
4641         return 0;
4642 }
4643
4644 void restore_bitmap_write_access(struct file *file)
4645 {
4646         struct inode *inode = file->f_mapping->host;
4647
4648         spin_lock(&inode->i_lock);
4649         atomic_set(&inode->i_writecount, 1);
4650         spin_unlock(&inode->i_lock);
4651 }
4652
4653 static void md_clean(mddev_t *mddev)
4654 {
4655         mddev->array_sectors = 0;
4656         mddev->external_size = 0;
4657         mddev->dev_sectors = 0;
4658         mddev->raid_disks = 0;
4659         mddev->recovery_cp = 0;
4660         mddev->resync_min = 0;
4661         mddev->resync_max = MaxSector;
4662         mddev->reshape_position = MaxSector;
4663         mddev->external = 0;
4664         mddev->persistent = 0;
4665         mddev->level = LEVEL_NONE;
4666         mddev->clevel[0] = 0;
4667         mddev->flags = 0;
4668         mddev->ro = 0;
4669         mddev->metadata_type[0] = 0;
4670         mddev->chunk_sectors = 0;
4671         mddev->ctime = mddev->utime = 0;
4672         mddev->layout = 0;
4673         mddev->max_disks = 0;
4674         mddev->events = 0;
4675         mddev->can_decrease_events = 0;
4676         mddev->delta_disks = 0;
4677         mddev->new_level = LEVEL_NONE;
4678         mddev->new_layout = 0;
4679         mddev->new_chunk_sectors = 0;
4680         mddev->curr_resync = 0;
4681         mddev->resync_mismatches = 0;
4682         mddev->suspend_lo = mddev->suspend_hi = 0;
4683         mddev->sync_speed_min = mddev->sync_speed_max = 0;
4684         mddev->recovery = 0;
4685         mddev->in_sync = 0;
4686         mddev->degraded = 0;
4687         mddev->safemode = 0;
4688         mddev->bitmap_info.offset = 0;
4689         mddev->bitmap_info.default_offset = 0;
4690         mddev->bitmap_info.chunksize = 0;
4691         mddev->bitmap_info.daemon_sleep = 0;
4692         mddev->bitmap_info.max_write_behind = 0;
4693         mddev->plug = NULL;
4694 }
4695
4696 void md_stop_writes(mddev_t *mddev)
4697 {
4698         if (mddev->sync_thread) {
4699                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4700                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4701                 md_unregister_thread(mddev->sync_thread);
4702                 mddev->sync_thread = NULL;
4703         }
4704
4705         del_timer_sync(&mddev->safemode_timer);
4706
4707         bitmap_flush(mddev);
4708         md_super_wait(mddev);
4709
4710         if (!mddev->in_sync || mddev->flags) {
4711                 /* mark array as shutdown cleanly */
4712                 mddev->in_sync = 1;
4713                 md_update_sb(mddev, 1);
4714         }
4715 }
4716 EXPORT_SYMBOL_GPL(md_stop_writes);
4717
4718 void md_stop(mddev_t *mddev)
4719 {
4720         mddev->pers->stop(mddev);
4721         if (mddev->pers->sync_request && mddev->to_remove == NULL)
4722                 mddev->to_remove = &md_redundancy_group;
4723         module_put(mddev->pers->owner);
4724         mddev->pers = NULL;
4725         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4726 }
4727 EXPORT_SYMBOL_GPL(md_stop);
4728
4729 static int md_set_readonly(mddev_t *mddev, int is_open)
4730 {
4731         int err = 0;
4732         mutex_lock(&mddev->open_mutex);
4733         if (atomic_read(&mddev->openers) > is_open) {
4734                 printk("md: %s still in use.\n",mdname(mddev));
4735                 err = -EBUSY;
4736                 goto out;
4737         }
4738         if (mddev->pers) {
4739                 md_stop_writes(mddev);
4740
4741                 err  = -ENXIO;
4742                 if (mddev->ro==1)
4743                         goto out;
4744                 mddev->ro = 1;
4745                 set_disk_ro(mddev->gendisk, 1);
4746                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4747                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4748                 err = 0;
4749         }
4750 out:
4751         mutex_unlock(&mddev->open_mutex);
4752         return err;
4753 }
4754
4755 /* mode:
4756  *   0 - completely stop and dis-assemble array
4757  *   2 - stop but do not disassemble array
4758  */
4759 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4760 {
4761         struct gendisk *disk = mddev->gendisk;
4762         mdk_rdev_t *rdev;
4763
4764         mutex_lock(&mddev->open_mutex);
4765         if (atomic_read(&mddev->openers) > is_open ||
4766             mddev->sysfs_active) {
4767                 printk("md: %s still in use.\n",mdname(mddev));
4768                 mutex_unlock(&mddev->open_mutex);
4769                 return -EBUSY;
4770         }
4771
4772         if (mddev->pers) {
4773                 if (mddev->ro)
4774                         set_disk_ro(disk, 0);
4775
4776                 md_stop_writes(mddev);
4777                 md_stop(mddev);
4778                 mddev->queue->merge_bvec_fn = NULL;
4779                 mddev->queue->unplug_fn = NULL;
4780                 mddev->queue->backing_dev_info.congested_fn = NULL;
4781
4782                 /* tell userspace to handle 'inactive' */
4783                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4784
4785                 list_for_each_entry(rdev, &mddev->disks, same_set)
4786                         if (rdev->raid_disk >= 0) {
4787                                 char nm[20];
4788                                 sprintf(nm, "rd%d", rdev->raid_disk);
4789                                 sysfs_remove_link(&mddev->kobj, nm);
4790                         }
4791
4792                 set_capacity(disk, 0);
4793                 mutex_unlock(&mddev->open_mutex);
4794                 revalidate_disk(disk);
4795
4796                 if (mddev->ro)
4797                         mddev->ro = 0;
4798         } else
4799                 mutex_unlock(&mddev->open_mutex);
4800         /*
4801          * Free resources if final stop
4802          */
4803         if (mode == 0) {
4804                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4805
4806                 bitmap_destroy(mddev);
4807                 if (mddev->bitmap_info.file) {
4808                         restore_bitmap_write_access(mddev->bitmap_info.file);
4809                         fput(mddev->bitmap_info.file);
4810                         mddev->bitmap_info.file = NULL;
4811                 }
4812                 mddev->bitmap_info.offset = 0;
4813
4814                 export_array(mddev);
4815
4816                 md_clean(mddev);
4817                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4818                 if (mddev->hold_active == UNTIL_STOP)
4819                         mddev->hold_active = 0;
4820         }
4821         blk_integrity_unregister(disk);
4822         md_new_event(mddev);
4823         sysfs_notify_dirent_safe(mddev->sysfs_state);
4824         return 0;
4825 }
4826
4827 #ifndef MODULE
4828 static void autorun_array(mddev_t *mddev)
4829 {
4830         mdk_rdev_t *rdev;
4831         int err;
4832
4833         if (list_empty(&mddev->disks))
4834                 return;
4835
4836         printk(KERN_INFO "md: running: ");
4837
4838         list_for_each_entry(rdev, &mddev->disks, same_set) {
4839                 char b[BDEVNAME_SIZE];
4840                 printk("<%s>", bdevname(rdev->bdev,b));
4841         }
4842         printk("\n");
4843
4844         err = do_md_run(mddev);
4845         if (err) {
4846                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4847                 do_md_stop(mddev, 0, 0);
4848         }
4849 }
4850
4851 /*
4852  * lets try to run arrays based on all disks that have arrived
4853  * until now. (those are in pending_raid_disks)
4854  *
4855  * the method: pick the first pending disk, collect all disks with
4856  * the same UUID, remove all from the pending list and put them into
4857  * the 'same_array' list. Then order this list based on superblock
4858  * update time (freshest comes first), kick out 'old' disks and
4859  * compare superblocks. If everything's fine then run it.
4860  *
4861  * If "unit" is allocated, then bump its reference count
4862  */
4863 static void autorun_devices(int part)
4864 {
4865         mdk_rdev_t *rdev0, *rdev, *tmp;
4866         mddev_t *mddev;
4867         char b[BDEVNAME_SIZE];
4868
4869         printk(KERN_INFO "md: autorun ...\n");
4870         while (!list_empty(&pending_raid_disks)) {
4871                 int unit;
4872                 dev_t dev;
4873                 LIST_HEAD(candidates);
4874                 rdev0 = list_entry(pending_raid_disks.next,
4875                                          mdk_rdev_t, same_set);
4876
4877                 printk(KERN_INFO "md: considering %s ...\n",
4878                         bdevname(rdev0->bdev,b));
4879                 INIT_LIST_HEAD(&candidates);
4880                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4881                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4882                                 printk(KERN_INFO "md:  adding %s ...\n",
4883                                         bdevname(rdev->bdev,b));
4884                                 list_move(&rdev->same_set, &candidates);
4885                         }
4886                 /*
4887                  * now we have a set of devices, with all of them having
4888                  * mostly sane superblocks. It's time to allocate the
4889                  * mddev.
4890                  */
4891                 if (part) {
4892                         dev = MKDEV(mdp_major,
4893                                     rdev0->preferred_minor << MdpMinorShift);
4894                         unit = MINOR(dev) >> MdpMinorShift;
4895                 } else {
4896                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4897                         unit = MINOR(dev);
4898                 }
4899                 if (rdev0->preferred_minor != unit) {
4900                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4901                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4902                         break;
4903                 }
4904
4905                 md_probe(dev, NULL, NULL);
4906                 mddev = mddev_find(dev);
4907                 if (!mddev || !mddev->gendisk) {
4908                         if (mddev)
4909                                 mddev_put(mddev);
4910                         printk(KERN_ERR
4911                                 "md: cannot allocate memory for md drive.\n");
4912                         break;
4913                 }
4914                 if (mddev_lock(mddev))
4915                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4916                                mdname(mddev));
4917                 else if (mddev->raid_disks || mddev->major_version
4918                          || !list_empty(&mddev->disks)) {
4919                         printk(KERN_WARNING
4920                                 "md: %s already running, cannot run %s\n",
4921                                 mdname(mddev), bdevname(rdev0->bdev,b));
4922                         mddev_unlock(mddev);
4923                 } else {
4924                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4925                         mddev->persistent = 1;
4926                         rdev_for_each_list(rdev, tmp, &candidates) {
4927                                 list_del_init(&rdev->same_set);
4928                                 if (bind_rdev_to_array(rdev, mddev))
4929                                         export_rdev(rdev);
4930                         }
4931                         autorun_array(mddev);
4932                         mddev_unlock(mddev);
4933                 }
4934                 /* on success, candidates will be empty, on error
4935                  * it won't...
4936                  */
4937                 rdev_for_each_list(rdev, tmp, &candidates) {
4938                         list_del_init(&rdev->same_set);
4939                         export_rdev(rdev);
4940                 }
4941                 mddev_put(mddev);
4942         }
4943         printk(KERN_INFO "md: ... autorun DONE.\n");
4944 }
4945 #endif /* !MODULE */
4946
4947 static int get_version(void __user * arg)
4948 {
4949         mdu_version_t ver;
4950
4951         ver.major = MD_MAJOR_VERSION;
4952         ver.minor = MD_MINOR_VERSION;
4953         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4954
4955         if (copy_to_user(arg, &ver, sizeof(ver)))
4956                 return -EFAULT;
4957
4958         return 0;
4959 }
4960
4961 static int get_array_info(mddev_t * mddev, void __user * arg)
4962 {
4963         mdu_array_info_t info;
4964         int nr,working,insync,failed,spare;
4965         mdk_rdev_t *rdev;
4966
4967         nr=working=insync=failed=spare=0;
4968         list_for_each_entry(rdev, &mddev->disks, same_set) {
4969                 nr++;
4970                 if (test_bit(Faulty, &rdev->flags))
4971                         failed++;
4972                 else {
4973                         working++;
4974                         if (test_bit(In_sync, &rdev->flags))
4975                                 insync++;
4976                         else
4977                                 spare++;
4978                 }
4979         }
4980
4981         info.major_version = mddev->major_version;
4982         info.minor_version = mddev->minor_version;
4983         info.patch_version = MD_PATCHLEVEL_VERSION;
4984         info.ctime         = mddev->ctime;
4985         info.level         = mddev->level;
4986         info.size          = mddev->dev_sectors / 2;
4987         if (info.size != mddev->dev_sectors / 2) /* overflow */
4988                 info.size = -1;
4989         info.nr_disks      = nr;
4990         info.raid_disks    = mddev->raid_disks;
4991         info.md_minor      = mddev->md_minor;
4992         info.not_persistent= !mddev->persistent;
4993
4994         info.utime         = mddev->utime;
4995         info.state         = 0;
4996         if (mddev->in_sync)
4997                 info.state = (1<<MD_SB_CLEAN);
4998         if (mddev->bitmap && mddev->bitmap_info.offset)
4999                 info.state = (1<<MD_SB_BITMAP_PRESENT);
5000         info.active_disks  = insync;
5001         info.working_disks = working;
5002         info.failed_disks  = failed;
5003         info.spare_disks   = spare;
5004
5005         info.layout        = mddev->layout;
5006         info.chunk_size    = mddev->chunk_sectors << 9;
5007
5008         if (copy_to_user(arg, &info, sizeof(info)))
5009                 return -EFAULT;
5010
5011         return 0;
5012 }
5013
5014 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5015 {
5016         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5017         char *ptr, *buf = NULL;
5018         int err = -ENOMEM;
5019
5020         if (md_allow_write(mddev))
5021                 file = kmalloc(sizeof(*file), GFP_NOIO);
5022         else
5023                 file = kmalloc(sizeof(*file), GFP_KERNEL);
5024
5025         if (!file)
5026                 goto out;
5027
5028         /* bitmap disabled, zero the first byte and copy out */
5029         if (!mddev->bitmap || !mddev->bitmap->file) {
5030                 file->pathname[0] = '\0';
5031                 goto copy_out;
5032         }
5033
5034         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5035         if (!buf)
5036                 goto out;
5037
5038         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5039         if (IS_ERR(ptr))
5040                 goto out;
5041
5042         strcpy(file->pathname, ptr);
5043
5044 copy_out:
5045         err = 0;
5046         if (copy_to_user(arg, file, sizeof(*file)))
5047                 err = -EFAULT;
5048 out:
5049         kfree(buf);
5050         kfree(file);
5051         return err;
5052 }
5053
5054 static int get_disk_info(mddev_t * mddev, void __user * arg)
5055 {
5056         mdu_disk_info_t info;
5057         mdk_rdev_t *rdev;
5058
5059         if (copy_from_user(&info, arg, sizeof(info)))
5060                 return -EFAULT;
5061
5062         rdev = find_rdev_nr(mddev, info.number);
5063         if (rdev) {
5064                 info.major = MAJOR(rdev->bdev->bd_dev);
5065                 info.minor = MINOR(rdev->bdev->bd_dev);
5066                 info.raid_disk = rdev->raid_disk;
5067                 info.state = 0;
5068                 if (test_bit(Faulty, &rdev->flags))
5069                         info.state |= (1<<MD_DISK_FAULTY);
5070                 else if (test_bit(In_sync, &rdev->flags)) {
5071                         info.state |= (1<<MD_DISK_ACTIVE);
5072                         info.state |= (1<<MD_DISK_SYNC);
5073                 }
5074                 if (test_bit(WriteMostly, &rdev->flags))
5075                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5076         } else {
5077                 info.major = info.minor = 0;
5078                 info.raid_disk = -1;
5079                 info.state = (1<<MD_DISK_REMOVED);
5080         }
5081
5082         if (copy_to_user(arg, &info, sizeof(info)))
5083                 return -EFAULT;
5084
5085         return 0;
5086 }
5087
5088 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5089 {
5090         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5091         mdk_rdev_t *rdev;
5092         dev_t dev = MKDEV(info->major,info->minor);
5093
5094         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5095                 return -EOVERFLOW;
5096
5097         if (!mddev->raid_disks) {
5098                 int err;
5099                 /* expecting a device which has a superblock */
5100                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5101                 if (IS_ERR(rdev)) {
5102                         printk(KERN_WARNING
5103                                 "md: md_import_device returned %ld\n",
5104                                 PTR_ERR(rdev));
5105                         return PTR_ERR(rdev);
5106                 }
5107                 if (!list_empty(&mddev->disks)) {
5108                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5109                                                         mdk_rdev_t, same_set);
5110                         err = super_types[mddev->major_version]
5111                                 .load_super(rdev, rdev0, mddev->minor_version);
5112                         if (err < 0) {
5113                                 printk(KERN_WARNING
5114                                         "md: %s has different UUID to %s\n",
5115                                         bdevname(rdev->bdev,b),
5116                                         bdevname(rdev0->bdev,b2));
5117                                 export_rdev(rdev);
5118                                 return -EINVAL;
5119                         }
5120                 }
5121                 err = bind_rdev_to_array(rdev, mddev);
5122                 if (err)
5123                         export_rdev(rdev);
5124                 return err;
5125         }
5126
5127         /*
5128          * add_new_disk can be used once the array is assembled
5129          * to add "hot spares".  They must already have a superblock
5130          * written
5131          */
5132         if (mddev->pers) {
5133                 int err;
5134                 if (!mddev->pers->hot_add_disk) {
5135                         printk(KERN_WARNING
5136                                 "%s: personality does not support diskops!\n",
5137                                mdname(mddev));
5138                         return -EINVAL;
5139                 }
5140                 if (mddev->persistent)
5141                         rdev = md_import_device(dev, mddev->major_version,
5142                                                 mddev->minor_version);
5143                 else
5144                         rdev = md_import_device(dev, -1, -1);
5145                 if (IS_ERR(rdev)) {
5146                         printk(KERN_WARNING
5147                                 "md: md_import_device returned %ld\n",
5148                                 PTR_ERR(rdev));
5149                         return PTR_ERR(rdev);
5150                 }
5151                 /* set saved_raid_disk if appropriate */
5152                 if (!mddev->persistent) {
5153                         if (info->state & (1<<MD_DISK_SYNC)  &&
5154                             info->raid_disk < mddev->raid_disks)
5155                                 rdev->raid_disk = info->raid_disk;
5156                         else
5157                                 rdev->raid_disk = -1;
5158                 } else
5159                         super_types[mddev->major_version].
5160                                 validate_super(mddev, rdev);
5161                 if (test_bit(In_sync, &rdev->flags))
5162                         rdev->saved_raid_disk = rdev->raid_disk;
5163                 else
5164                         rdev->saved_raid_disk = -1;
5165
5166                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5167                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5168                         set_bit(WriteMostly, &rdev->flags);
5169                 else
5170                         clear_bit(WriteMostly, &rdev->flags);
5171
5172                 rdev->raid_disk = -1;
5173                 err = bind_rdev_to_array(rdev, mddev);
5174                 if (!err && !mddev->pers->hot_remove_disk) {
5175                         /* If there is hot_add_disk but no hot_remove_disk
5176                          * then added disks for geometry changes,
5177                          * and should be added immediately.
5178                          */
5179                         super_types[mddev->major_version].
5180                                 validate_super(mddev, rdev);
5181                         err = mddev->pers->hot_add_disk(mddev, rdev);
5182                         if (err)
5183                                 unbind_rdev_from_array(rdev);
5184                 }
5185                 if (err)
5186                         export_rdev(rdev);
5187                 else
5188                         sysfs_notify_dirent_safe(rdev->sysfs_state);
5189
5190                 md_update_sb(mddev, 1);
5191                 if (mddev->degraded)
5192                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5193                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5194                 md_wakeup_thread(mddev->thread);
5195                 return err;
5196         }
5197
5198         /* otherwise, add_new_disk is only allowed
5199          * for major_version==0 superblocks
5200          */
5201         if (mddev->major_version != 0) {
5202                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5203                        mdname(mddev));
5204                 return -EINVAL;
5205         }
5206
5207         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5208                 int err;
5209                 rdev = md_import_device(dev, -1, 0);
5210                 if (IS_ERR(rdev)) {
5211                         printk(KERN_WARNING
5212                                 "md: error, md_import_device() returned %ld\n",
5213                                 PTR_ERR(rdev));
5214                         return PTR_ERR(rdev);
5215                 }
5216                 rdev->desc_nr = info->number;
5217                 if (info->raid_disk < mddev->raid_disks)
5218                         rdev->raid_disk = info->raid_disk;
5219                 else
5220                         rdev->raid_disk = -1;
5221
5222                 if (rdev->raid_disk < mddev->raid_disks)
5223                         if (info->state & (1<<MD_DISK_SYNC))
5224                                 set_bit(In_sync, &rdev->flags);
5225
5226                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5227                         set_bit(WriteMostly, &rdev->flags);
5228
5229                 if (!mddev->persistent) {
5230                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5231                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5232                 } else
5233                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5234                 rdev->sectors = rdev->sb_start;
5235
5236                 err = bind_rdev_to_array(rdev, mddev);
5237                 if (err) {
5238                         export_rdev(rdev);
5239                         return err;
5240                 }
5241         }
5242
5243         return 0;
5244 }
5245
5246 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5247 {
5248         char b[BDEVNAME_SIZE];
5249         mdk_rdev_t *rdev;
5250
5251         rdev = find_rdev(mddev, dev);
5252         if (!rdev)
5253                 return -ENXIO;
5254
5255         if (rdev->raid_disk >= 0)
5256                 goto busy;
5257
5258         kick_rdev_from_array(rdev);
5259         md_update_sb(mddev, 1);
5260         md_new_event(mddev);
5261
5262         return 0;
5263 busy:
5264         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5265                 bdevname(rdev->bdev,b), mdname(mddev));
5266         return -EBUSY;
5267 }
5268
5269 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5270 {
5271         char b[BDEVNAME_SIZE];
5272         int err;
5273         mdk_rdev_t *rdev;
5274
5275         if (!mddev->pers)
5276                 return -ENODEV;
5277
5278         if (mddev->major_version != 0) {
5279                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5280                         " version-0 superblocks.\n",
5281                         mdname(mddev));
5282                 return -EINVAL;
5283         }
5284         if (!mddev->pers->hot_add_disk) {
5285                 printk(KERN_WARNING
5286                         "%s: personality does not support diskops!\n",
5287                         mdname(mddev));
5288                 return -EINVAL;
5289         }
5290
5291         rdev = md_import_device(dev, -1, 0);
5292         if (IS_ERR(rdev)) {
5293                 printk(KERN_WARNING
5294                         "md: error, md_import_device() returned %ld\n",
5295                         PTR_ERR(rdev));
5296                 return -EINVAL;
5297         }
5298
5299         if (mddev->persistent)
5300                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5301         else
5302                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5303
5304         rdev->sectors = rdev->sb_start;
5305
5306         if (test_bit(Faulty, &rdev->flags)) {
5307                 printk(KERN_WARNING
5308                         "md: can not hot-add faulty %s disk to %s!\n",
5309                         bdevname(rdev->bdev,b), mdname(mddev));
5310                 err = -EINVAL;
5311                 goto abort_export;
5312         }
5313         clear_bit(In_sync, &rdev->flags);
5314         rdev->desc_nr = -1;
5315         rdev->saved_raid_disk = -1;
5316         err = bind_rdev_to_array(rdev, mddev);
5317         if (err)
5318                 goto abort_export;
5319
5320         /*
5321          * The rest should better be atomic, we can have disk failures
5322          * noticed in interrupt contexts ...
5323          */
5324
5325         rdev->raid_disk = -1;
5326
5327         md_update_sb(mddev, 1);
5328
5329         /*
5330          * Kick recovery, maybe this spare has to be added to the
5331          * array immediately.
5332          */
5333         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5334         md_wakeup_thread(mddev->thread);
5335         md_new_event(mddev);
5336         return 0;
5337
5338 abort_export:
5339         export_rdev(rdev);
5340         return err;
5341 }
5342
5343 static int set_bitmap_file(mddev_t *mddev, int fd)
5344 {
5345         int err;
5346
5347         if (mddev->pers) {
5348                 if (!mddev->pers->quiesce)
5349                         return -EBUSY;
5350                 if (mddev->recovery || mddev->sync_thread)
5351                         return -EBUSY;
5352                 /* we should be able to change the bitmap.. */
5353         }
5354
5355
5356         if (fd >= 0) {
5357                 if (mddev->bitmap)
5358                         return -EEXIST; /* cannot add when bitmap is present */
5359                 mddev->bitmap_info.file = fget(fd);
5360
5361                 if (mddev->bitmap_info.file == NULL) {
5362                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5363                                mdname(mddev));
5364                         return -EBADF;
5365                 }
5366
5367                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5368                 if (err) {
5369                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5370                                mdname(mddev));
5371                         fput(mddev->bitmap_info.file);
5372                         mddev->bitmap_info.file = NULL;
5373                         return err;
5374                 }
5375                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5376         } else if (mddev->bitmap == NULL)
5377                 return -ENOENT; /* cannot remove what isn't there */
5378         err = 0;
5379         if (mddev->pers) {
5380                 mddev->pers->quiesce(mddev, 1);
5381                 if (fd >= 0) {
5382                         err = bitmap_create(mddev);
5383                         if (!err)
5384                                 err = bitmap_load(mddev);
5385                 }
5386                 if (fd < 0 || err) {
5387                         bitmap_destroy(mddev);
5388                         fd = -1; /* make sure to put the file */
5389                 }
5390                 mddev->pers->quiesce(mddev, 0);
5391         }
5392         if (fd < 0) {
5393                 if (mddev->bitmap_info.file) {
5394                         restore_bitmap_write_access(mddev->bitmap_info.file);
5395                         fput(mddev->bitmap_info.file);
5396                 }
5397                 mddev->bitmap_info.file = NULL;
5398         }
5399
5400         return err;
5401 }
5402
5403 /*
5404  * set_array_info is used two different ways
5405  * The original usage is when creating a new array.
5406  * In this usage, raid_disks is > 0 and it together with
5407  *  level, size, not_persistent,layout,chunksize determine the
5408  *  shape of the array.
5409  *  This will always create an array with a type-0.90.0 superblock.
5410  * The newer usage is when assembling an array.
5411  *  In this case raid_disks will be 0, and the major_version field is
5412  *  use to determine which style super-blocks are to be found on the devices.
5413  *  The minor and patch _version numbers are also kept incase the
5414  *  super_block handler wishes to interpret them.
5415  */
5416 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5417 {
5418
5419         if (info->raid_disks == 0) {
5420                 /* just setting version number for superblock loading */
5421                 if (info->major_version < 0 ||
5422                     info->major_version >= ARRAY_SIZE(super_types) ||
5423                     super_types[info->major_version].name == NULL) {
5424                         /* maybe try to auto-load a module? */
5425                         printk(KERN_INFO
5426                                 "md: superblock version %d not known\n",
5427                                 info->major_version);
5428                         return -EINVAL;
5429                 }
5430                 mddev->major_version = info->major_version;
5431                 mddev->minor_version = info->minor_version;
5432                 mddev->patch_version = info->patch_version;
5433                 mddev->persistent = !info->not_persistent;
5434                 /* ensure mddev_put doesn't delete this now that there
5435                  * is some minimal configuration.
5436                  */
5437                 mddev->ctime         = get_seconds();
5438                 return 0;
5439         }
5440         mddev->major_version = MD_MAJOR_VERSION;
5441         mddev->minor_version = MD_MINOR_VERSION;
5442         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5443         mddev->ctime         = get_seconds();
5444
5445         mddev->level         = info->level;
5446         mddev->clevel[0]     = 0;
5447         mddev->dev_sectors   = 2 * (sector_t)info->size;
5448         mddev->raid_disks    = info->raid_disks;
5449         /* don't set md_minor, it is determined by which /dev/md* was
5450          * openned
5451          */
5452         if (info->state & (1<<MD_SB_CLEAN))
5453                 mddev->recovery_cp = MaxSector;
5454         else
5455                 mddev->recovery_cp = 0;
5456         mddev->persistent    = ! info->not_persistent;
5457         mddev->external      = 0;
5458
5459         mddev->layout        = info->layout;
5460         mddev->chunk_sectors = info->chunk_size >> 9;
5461
5462         mddev->max_disks     = MD_SB_DISKS;
5463
5464         if (mddev->persistent)
5465                 mddev->flags         = 0;
5466         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5467
5468         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5469         mddev->bitmap_info.offset = 0;
5470
5471         mddev->reshape_position = MaxSector;
5472
5473         /*
5474          * Generate a 128 bit UUID
5475          */
5476         get_random_bytes(mddev->uuid, 16);
5477
5478         mddev->new_level = mddev->level;
5479         mddev->new_chunk_sectors = mddev->chunk_sectors;
5480         mddev->new_layout = mddev->layout;
5481         mddev->delta_disks = 0;
5482
5483         return 0;
5484 }
5485
5486 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5487 {
5488         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5489
5490         if (mddev->external_size)
5491                 return;
5492
5493         mddev->array_sectors = array_sectors;
5494 }
5495 EXPORT_SYMBOL(md_set_array_sectors);
5496
5497 static int update_size(mddev_t *mddev, sector_t num_sectors)
5498 {
5499         mdk_rdev_t *rdev;
5500         int rv;
5501         int fit = (num_sectors == 0);
5502
5503         if (mddev->pers->resize == NULL)
5504                 return -EINVAL;
5505         /* The "num_sectors" is the number of sectors of each device that
5506          * is used.  This can only make sense for arrays with redundancy.
5507          * linear and raid0 always use whatever space is available. We can only
5508          * consider changing this number if no resync or reconstruction is
5509          * happening, and if the new size is acceptable. It must fit before the
5510          * sb_start or, if that is <data_offset, it must fit before the size
5511          * of each device.  If num_sectors is zero, we find the largest size
5512          * that fits.
5513
5514          */
5515         if (mddev->sync_thread)
5516                 return -EBUSY;
5517         if (mddev->bitmap)
5518                 /* Sorry, cannot grow a bitmap yet, just remove it,
5519                  * grow, and re-add.
5520                  */
5521                 return -EBUSY;
5522         list_for_each_entry(rdev, &mddev->disks, same_set) {
5523                 sector_t avail = rdev->sectors;
5524
5525                 if (fit && (num_sectors == 0 || num_sectors > avail))
5526                         num_sectors = avail;
5527                 if (avail < num_sectors)
5528                         return -ENOSPC;
5529         }
5530         rv = mddev->pers->resize(mddev, num_sectors);
5531         if (!rv)
5532                 revalidate_disk(mddev->gendisk);
5533         return rv;
5534 }
5535
5536 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5537 {
5538         int rv;
5539         /* change the number of raid disks */
5540         if (mddev->pers->check_reshape == NULL)
5541                 return -EINVAL;
5542         if (raid_disks <= 0 ||
5543             (mddev->max_disks && raid_disks >= mddev->max_disks))
5544                 return -EINVAL;
5545         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5546                 return -EBUSY;
5547         mddev->delta_disks = raid_disks - mddev->raid_disks;
5548
5549         rv = mddev->pers->check_reshape(mddev);
5550         return rv;
5551 }
5552
5553
5554 /*
5555  * update_array_info is used to change the configuration of an
5556  * on-line array.
5557  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5558  * fields in the info are checked against the array.
5559  * Any differences that cannot be handled will cause an error.
5560  * Normally, only one change can be managed at a time.
5561  */
5562 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5563 {
5564         int rv = 0;
5565         int cnt = 0;
5566         int state = 0;
5567
5568         /* calculate expected state,ignoring low bits */
5569         if (mddev->bitmap && mddev->bitmap_info.offset)
5570                 state |= (1 << MD_SB_BITMAP_PRESENT);
5571
5572         if (mddev->major_version != info->major_version ||
5573             mddev->minor_version != info->minor_version ||
5574 /*          mddev->patch_version != info->patch_version || */
5575             mddev->ctime         != info->ctime         ||
5576             mddev->level         != info->level         ||
5577 /*          mddev->layout        != info->layout        || */
5578             !mddev->persistent   != info->not_persistent||
5579             mddev->chunk_sectors != info->chunk_size >> 9 ||
5580             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5581             ((state^info->state) & 0xfffffe00)
5582                 )
5583                 return -EINVAL;
5584         /* Check there is only one change */
5585         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5586                 cnt++;
5587         if (mddev->raid_disks != info->raid_disks)
5588                 cnt++;
5589         if (mddev->layout != info->layout)
5590                 cnt++;
5591         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5592                 cnt++;
5593         if (cnt == 0)
5594                 return 0;
5595         if (cnt > 1)
5596                 return -EINVAL;
5597
5598         if (mddev->layout != info->layout) {
5599                 /* Change layout
5600                  * we don't need to do anything at the md level, the
5601                  * personality will take care of it all.
5602                  */
5603                 if (mddev->pers->check_reshape == NULL)
5604                         return -EINVAL;
5605                 else {
5606                         mddev->new_layout = info->layout;
5607                         rv = mddev->pers->check_reshape(mddev);
5608                         if (rv)
5609                                 mddev->new_layout = mddev->layout;
5610                         return rv;
5611                 }
5612         }
5613         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5614                 rv = update_size(mddev, (sector_t)info->size * 2);
5615
5616         if (mddev->raid_disks    != info->raid_disks)
5617                 rv = update_raid_disks(mddev, info->raid_disks);
5618
5619         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5620                 if (mddev->pers->quiesce == NULL)
5621                         return -EINVAL;
5622                 if (mddev->recovery || mddev->sync_thread)
5623                         return -EBUSY;
5624                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5625                         /* add the bitmap */
5626                         if (mddev->bitmap)
5627                                 return -EEXIST;
5628                         if (mddev->bitmap_info.default_offset == 0)
5629                                 return -EINVAL;
5630                         mddev->bitmap_info.offset =
5631                                 mddev->bitmap_info.default_offset;
5632                         mddev->pers->quiesce(mddev, 1);
5633                         rv = bitmap_create(mddev);
5634                         if (!rv)
5635                                 rv = bitmap_load(mddev);
5636                         if (rv)
5637                                 bitmap_destroy(mddev);
5638                         mddev->pers->quiesce(mddev, 0);
5639                 } else {
5640                         /* remove the bitmap */
5641                         if (!mddev->bitmap)
5642                                 return -ENOENT;
5643                         if (mddev->bitmap->file)
5644                                 return -EINVAL;
5645                         mddev->pers->quiesce(mddev, 1);
5646                         bitmap_destroy(mddev);
5647                         mddev->pers->quiesce(mddev, 0);
5648                         mddev->bitmap_info.offset = 0;
5649                 }
5650         }
5651         md_update_sb(mddev, 1);
5652         return rv;
5653 }
5654
5655 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5656 {
5657         mdk_rdev_t *rdev;
5658
5659         if (mddev->pers == NULL)
5660                 return -ENODEV;
5661
5662         rdev = find_rdev(mddev, dev);
5663         if (!rdev)
5664                 return -ENODEV;
5665
5666         md_error(mddev, rdev);
5667         return 0;
5668 }
5669
5670 /*
5671  * We have a problem here : there is no easy way to give a CHS
5672  * virtual geometry. We currently pretend that we have a 2 heads
5673  * 4 sectors (with a BIG number of cylinders...). This drives
5674  * dosfs just mad... ;-)
5675  */
5676 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5677 {
5678         mddev_t *mddev = bdev->bd_disk->private_data;
5679
5680         geo->heads = 2;
5681         geo->sectors = 4;
5682         geo->cylinders = mddev->array_sectors / 8;
5683         return 0;
5684 }
5685
5686 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5687                         unsigned int cmd, unsigned long arg)
5688 {
5689         int err = 0;
5690         void __user *argp = (void __user *)arg;
5691         mddev_t *mddev = NULL;
5692         int ro;
5693
5694         if (!capable(CAP_SYS_ADMIN))
5695                 return -EACCES;
5696
5697         /*
5698          * Commands dealing with the RAID driver but not any
5699          * particular array:
5700          */
5701         switch (cmd)
5702         {
5703                 case RAID_VERSION:
5704                         err = get_version(argp);
5705                         goto done;
5706
5707                 case PRINT_RAID_DEBUG:
5708                         err = 0;
5709                         md_print_devices();
5710                         goto done;
5711
5712 #ifndef MODULE
5713                 case RAID_AUTORUN:
5714                         err = 0;
5715                         autostart_arrays(arg);
5716                         goto done;
5717 #endif
5718                 default:;
5719         }
5720
5721         /*
5722          * Commands creating/starting a new array:
5723          */
5724
5725         mddev = bdev->bd_disk->private_data;
5726
5727         if (!mddev) {
5728                 BUG();
5729                 goto abort;
5730         }
5731
5732         err = mddev_lock(mddev);
5733         if (err) {
5734                 printk(KERN_INFO
5735                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5736                         err, cmd);
5737                 goto abort;
5738         }
5739
5740         switch (cmd)
5741         {
5742                 case SET_ARRAY_INFO:
5743                         {
5744                                 mdu_array_info_t info;
5745                                 if (!arg)
5746                                         memset(&info, 0, sizeof(info));
5747                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5748                                         err = -EFAULT;
5749                                         goto abort_unlock;
5750                                 }
5751                                 if (mddev->pers) {
5752                                         err = update_array_info(mddev, &info);
5753                                         if (err) {
5754                                                 printk(KERN_WARNING "md: couldn't update"
5755                                                        " array info. %d\n", err);
5756                                                 goto abort_unlock;
5757                                         }
5758                                         goto done_unlock;
5759                                 }
5760                                 if (!list_empty(&mddev->disks)) {
5761                                         printk(KERN_WARNING
5762                                                "md: array %s already has disks!\n",
5763                                                mdname(mddev));
5764                                         err = -EBUSY;
5765                                         goto abort_unlock;
5766                                 }
5767                                 if (mddev->raid_disks) {
5768                                         printk(KERN_WARNING
5769                                                "md: array %s already initialised!\n",
5770                                                mdname(mddev));
5771                                         err = -EBUSY;
5772                                         goto abort_unlock;
5773                                 }
5774                                 err = set_array_info(mddev, &info);
5775                                 if (err) {
5776                                         printk(KERN_WARNING "md: couldn't set"
5777                                                " array info. %d\n", err);
5778                                         goto abort_unlock;
5779                                 }
5780                         }
5781                         goto done_unlock;
5782
5783                 default:;
5784         }
5785
5786         /*
5787          * Commands querying/configuring an existing array:
5788          */
5789         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5790          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5791         if ((!mddev->raid_disks && !mddev->external)
5792             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5793             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5794             && cmd != GET_BITMAP_FILE) {
5795                 err = -ENODEV;
5796                 goto abort_unlock;
5797         }
5798
5799         /*
5800          * Commands even a read-only array can execute:
5801          */
5802         switch (cmd)
5803         {
5804                 case GET_ARRAY_INFO:
5805                         err = get_array_info(mddev, argp);
5806                         goto done_unlock;
5807
5808                 case GET_BITMAP_FILE:
5809                         err = get_bitmap_file(mddev, argp);
5810                         goto done_unlock;
5811
5812                 case GET_DISK_INFO:
5813                         err = get_disk_info(mddev, argp);
5814                         goto done_unlock;
5815
5816                 case RESTART_ARRAY_RW:
5817                         err = restart_array(mddev);
5818                         goto done_unlock;
5819
5820                 case STOP_ARRAY:
5821                         err = do_md_stop(mddev, 0, 1);
5822                         goto done_unlock;
5823
5824                 case STOP_ARRAY_RO:
5825                         err = md_set_readonly(mddev, 1);
5826                         goto done_unlock;
5827
5828                 case BLKROSET:
5829                         if (get_user(ro, (int __user *)(arg))) {
5830                                 err = -EFAULT;
5831                                 goto done_unlock;
5832                         }
5833                         err = -EINVAL;
5834
5835                         /* if the bdev is going readonly the value of mddev->ro
5836                          * does not matter, no writes are coming
5837                          */
5838                         if (ro)
5839                                 goto done_unlock;
5840
5841                         /* are we are already prepared for writes? */
5842                         if (mddev->ro != 1)
5843                                 goto done_unlock;
5844
5845                         /* transitioning to readauto need only happen for
5846                          * arrays that call md_write_start
5847                          */
5848                         if (mddev->pers) {
5849                                 err = restart_array(mddev);
5850                                 if (err == 0) {
5851                                         mddev->ro = 2;
5852                                         set_disk_ro(mddev->gendisk, 0);
5853                                 }
5854                         }
5855                         goto done_unlock;
5856         }
5857
5858         /*
5859          * The remaining ioctls are changing the state of the
5860          * superblock, so we do not allow them on read-only arrays.
5861          * However non-MD ioctls (e.g. get-size) will still come through
5862          * here and hit the 'default' below, so only disallow
5863          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5864          */
5865         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5866                 if (mddev->ro == 2) {
5867                         mddev->ro = 0;
5868                         sysfs_notify_dirent_safe(mddev->sysfs_state);
5869                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5870                         md_wakeup_thread(mddev->thread);
5871                 } else {
5872                         err = -EROFS;
5873                         goto abort_unlock;
5874                 }
5875         }
5876
5877         switch (cmd)
5878         {
5879                 case ADD_NEW_DISK:
5880                 {
5881                         mdu_disk_info_t info;
5882                         if (copy_from_user(&info, argp, sizeof(info)))
5883                                 err = -EFAULT;
5884                         else
5885                                 err = add_new_disk(mddev, &info);
5886                         goto done_unlock;
5887                 }
5888
5889                 case HOT_REMOVE_DISK:
5890                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5891                         goto done_unlock;
5892
5893                 case HOT_ADD_DISK:
5894                         err = hot_add_disk(mddev, new_decode_dev(arg));
5895                         goto done_unlock;
5896
5897                 case SET_DISK_FAULTY:
5898                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5899                         goto done_unlock;
5900
5901                 case RUN_ARRAY:
5902                         err = do_md_run(mddev);
5903                         goto done_unlock;
5904
5905                 case SET_BITMAP_FILE:
5906                         err = set_bitmap_file(mddev, (int)arg);
5907                         goto done_unlock;
5908
5909                 default:
5910                         err = -EINVAL;
5911                         goto abort_unlock;
5912         }
5913
5914 done_unlock:
5915 abort_unlock:
5916         if (mddev->hold_active == UNTIL_IOCTL &&
5917             err != -EINVAL)
5918                 mddev->hold_active = 0;
5919         mddev_unlock(mddev);
5920
5921         return err;
5922 done:
5923         if (err)
5924                 MD_BUG();
5925 abort:
5926         return err;
5927 }
5928 #ifdef CONFIG_COMPAT
5929 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5930                     unsigned int cmd, unsigned long arg)
5931 {
5932         switch (cmd) {
5933         case HOT_REMOVE_DISK:
5934         case HOT_ADD_DISK:
5935         case SET_DISK_FAULTY:
5936         case SET_BITMAP_FILE:
5937                 /* These take in integer arg, do not convert */
5938                 break;
5939         default:
5940                 arg = (unsigned long)compat_ptr(arg);
5941                 break;
5942         }
5943
5944         return md_ioctl(bdev, mode, cmd, arg);
5945 }
5946 #endif /* CONFIG_COMPAT */
5947
5948 static int md_open(struct block_device *bdev, fmode_t mode)
5949 {
5950         /*
5951          * Succeed if we can lock the mddev, which confirms that
5952          * it isn't being stopped right now.
5953          */
5954         mddev_t *mddev = mddev_find(bdev->bd_dev);
5955         int err;
5956
5957         if (mddev->gendisk != bdev->bd_disk) {
5958                 /* we are racing with mddev_put which is discarding this
5959                  * bd_disk.
5960                  */
5961                 mddev_put(mddev);
5962                 /* Wait until bdev->bd_disk is definitely gone */
5963                 flush_workqueue(md_misc_wq);
5964                 /* Then retry the open from the top */
5965                 return -ERESTARTSYS;
5966         }
5967         BUG_ON(mddev != bdev->bd_disk->private_data);
5968
5969         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5970                 goto out;
5971
5972         err = 0;
5973         atomic_inc(&mddev->openers);
5974         mutex_unlock(&mddev->open_mutex);
5975
5976         check_disk_size_change(mddev->gendisk, bdev);
5977  out:
5978         return err;
5979 }
5980
5981 static int md_release(struct gendisk *disk, fmode_t mode)
5982 {
5983         mddev_t *mddev = disk->private_data;
5984
5985         BUG_ON(!mddev);
5986         atomic_dec(&mddev->openers);
5987         mddev_put(mddev);
5988
5989         return 0;
5990 }
5991 static const struct block_device_operations md_fops =
5992 {
5993         .owner          = THIS_MODULE,
5994         .open           = md_open,
5995         .release        = md_release,
5996         .ioctl          = md_ioctl,
5997 #ifdef CONFIG_COMPAT
5998         .compat_ioctl   = md_compat_ioctl,
5999 #endif
6000         .getgeo         = md_getgeo,
6001 };
6002
6003 static int md_thread(void * arg)
6004 {
6005         mdk_thread_t *thread = arg;
6006
6007         /*
6008          * md_thread is a 'system-thread', it's priority should be very
6009          * high. We avoid resource deadlocks individually in each
6010          * raid personality. (RAID5 does preallocation) We also use RR and
6011          * the very same RT priority as kswapd, thus we will never get
6012          * into a priority inversion deadlock.
6013          *
6014          * we definitely have to have equal or higher priority than
6015          * bdflush, otherwise bdflush will deadlock if there are too
6016          * many dirty RAID5 blocks.
6017          */
6018
6019         allow_signal(SIGKILL);
6020         while (!kthread_should_stop()) {
6021
6022                 /* We need to wait INTERRUPTIBLE so that
6023                  * we don't add to the load-average.
6024                  * That means we need to be sure no signals are
6025                  * pending
6026                  */
6027                 if (signal_pending(current))
6028                         flush_signals(current);
6029
6030                 wait_event_interruptible_timeout
6031                         (thread->wqueue,
6032                          test_bit(THREAD_WAKEUP, &thread->flags)
6033                          || kthread_should_stop(),
6034                          thread->timeout);
6035
6036                 if (test_and_clear_bit(THREAD_WAKEUP, &thread->flags))
6037                         thread->run(thread->mddev);
6038         }
6039
6040         return 0;
6041 }
6042
6043 void md_wakeup_thread(mdk_thread_t *thread)
6044 {
6045         if (thread) {
6046                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6047                 set_bit(THREAD_WAKEUP, &thread->flags);
6048                 wake_up(&thread->wqueue);
6049         }
6050 }
6051
6052 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6053                                  const char *name)
6054 {
6055         mdk_thread_t *thread;
6056
6057         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6058         if (!thread)
6059                 return NULL;
6060
6061         init_waitqueue_head(&thread->wqueue);
6062
6063         thread->run = run;
6064         thread->mddev = mddev;
6065         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6066         thread->tsk = kthread_run(md_thread, thread,
6067                                   "%s_%s",
6068                                   mdname(thread->mddev),
6069                                   name ?: mddev->pers->name);
6070         if (IS_ERR(thread->tsk)) {
6071                 kfree(thread);
6072                 return NULL;
6073         }
6074         return thread;
6075 }
6076
6077 void md_unregister_thread(mdk_thread_t *thread)
6078 {
6079         if (!thread)
6080                 return;
6081         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6082
6083         kthread_stop(thread->tsk);
6084         kfree(thread);
6085 }
6086
6087 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6088 {
6089         if (!mddev) {
6090                 MD_BUG();
6091                 return;
6092         }
6093
6094         if (!rdev || test_bit(Faulty, &rdev->flags))
6095                 return;
6096
6097         if (mddev->external)
6098                 set_bit(Blocked, &rdev->flags);
6099 /*
6100         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6101                 mdname(mddev),
6102                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6103                 __builtin_return_address(0),__builtin_return_address(1),
6104                 __builtin_return_address(2),__builtin_return_address(3));
6105 */
6106         if (!mddev->pers)
6107                 return;
6108         if (!mddev->pers->error_handler)
6109                 return;
6110         mddev->pers->error_handler(mddev,rdev);
6111         if (mddev->degraded)
6112                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6113         sysfs_notify_dirent_safe(rdev->sysfs_state);
6114         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6115         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6116         md_wakeup_thread(mddev->thread);
6117         if (mddev->event_work.func)
6118                 queue_work(md_misc_wq, &mddev->event_work);
6119         md_new_event_inintr(mddev);
6120 }
6121
6122 /* seq_file implementation /proc/mdstat */
6123
6124 static void status_unused(struct seq_file *seq)
6125 {
6126         int i = 0;
6127         mdk_rdev_t *rdev;
6128
6129         seq_printf(seq, "unused devices: ");
6130
6131         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6132                 char b[BDEVNAME_SIZE];
6133                 i++;
6134                 seq_printf(seq, "%s ",
6135                               bdevname(rdev->bdev,b));
6136         }
6137         if (!i)
6138                 seq_printf(seq, "<none>");
6139
6140         seq_printf(seq, "\n");
6141 }
6142
6143
6144 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6145 {
6146         sector_t max_sectors, resync, res;
6147         unsigned long dt, db;
6148         sector_t rt;
6149         int scale;
6150         unsigned int per_milli;
6151
6152         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6153
6154         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6155                 max_sectors = mddev->resync_max_sectors;
6156         else
6157                 max_sectors = mddev->dev_sectors;
6158
6159         /*
6160          * Should not happen.
6161          */
6162         if (!max_sectors) {
6163                 MD_BUG();
6164                 return;
6165         }
6166         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6167          * in a sector_t, and (max_sectors>>scale) will fit in a
6168          * u32, as those are the requirements for sector_div.
6169          * Thus 'scale' must be at least 10
6170          */
6171         scale = 10;
6172         if (sizeof(sector_t) > sizeof(unsigned long)) {
6173                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6174                         scale++;
6175         }
6176         res = (resync>>scale)*1000;
6177         sector_div(res, (u32)((max_sectors>>scale)+1));
6178
6179         per_milli = res;
6180         {
6181                 int i, x = per_milli/50, y = 20-x;
6182                 seq_printf(seq, "[");
6183                 for (i = 0; i < x; i++)
6184                         seq_printf(seq, "=");
6185                 seq_printf(seq, ">");
6186                 for (i = 0; i < y; i++)
6187                         seq_printf(seq, ".");
6188                 seq_printf(seq, "] ");
6189         }
6190         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6191                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6192                     "reshape" :
6193                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6194                      "check" :
6195                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6196                       "resync" : "recovery"))),
6197                    per_milli/10, per_milli % 10,
6198                    (unsigned long long) resync/2,
6199                    (unsigned long long) max_sectors/2);
6200
6201         /*
6202          * dt: time from mark until now
6203          * db: blocks written from mark until now
6204          * rt: remaining time
6205          *
6206          * rt is a sector_t, so could be 32bit or 64bit.
6207          * So we divide before multiply in case it is 32bit and close
6208          * to the limit.
6209          * We scale the divisor (db) by 32 to avoid loosing precision
6210          * near the end of resync when the number of remaining sectors
6211          * is close to 'db'.
6212          * We then divide rt by 32 after multiplying by db to compensate.
6213          * The '+1' avoids division by zero if db is very small.
6214          */
6215         dt = ((jiffies - mddev->resync_mark) / HZ);
6216         if (!dt) dt++;
6217         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6218                 - mddev->resync_mark_cnt;
6219
6220         rt = max_sectors - resync;    /* number of remaining sectors */
6221         sector_div(rt, db/32+1);
6222         rt *= dt;
6223         rt >>= 5;
6224
6225         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6226                    ((unsigned long)rt % 60)/6);
6227
6228         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6229 }
6230
6231 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6232 {
6233         struct list_head *tmp;
6234         loff_t l = *pos;
6235         mddev_t *mddev;
6236
6237         if (l >= 0x10000)
6238                 return NULL;
6239         if (!l--)
6240                 /* header */
6241                 return (void*)1;
6242
6243         spin_lock(&all_mddevs_lock);
6244         list_for_each(tmp,&all_mddevs)
6245                 if (!l--) {
6246                         mddev = list_entry(tmp, mddev_t, all_mddevs);
6247                         mddev_get(mddev);
6248                         spin_unlock(&all_mddevs_lock);
6249                         return mddev;
6250                 }
6251         spin_unlock(&all_mddevs_lock);
6252         if (!l--)
6253                 return (void*)2;/* tail */
6254         return NULL;
6255 }
6256
6257 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6258 {
6259         struct list_head *tmp;
6260         mddev_t *next_mddev, *mddev = v;
6261
6262         ++*pos;
6263         if (v == (void*)2)
6264                 return NULL;
6265
6266         spin_lock(&all_mddevs_lock);
6267         if (v == (void*)1)
6268                 tmp = all_mddevs.next;
6269         else
6270                 tmp = mddev->all_mddevs.next;
6271         if (tmp != &all_mddevs)
6272                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6273         else {
6274                 next_mddev = (void*)2;
6275                 *pos = 0x10000;
6276         }
6277         spin_unlock(&all_mddevs_lock);
6278
6279         if (v != (void*)1)
6280                 mddev_put(mddev);
6281         return next_mddev;
6282
6283 }
6284
6285 static void md_seq_stop(struct seq_file *seq, void *v)
6286 {
6287         mddev_t *mddev = v;
6288
6289         if (mddev && v != (void*)1 && v != (void*)2)
6290                 mddev_put(mddev);
6291 }
6292
6293 struct mdstat_info {
6294         int event;
6295 };
6296
6297 static int md_seq_show(struct seq_file *seq, void *v)
6298 {
6299         mddev_t *mddev = v;
6300         sector_t sectors;
6301         mdk_rdev_t *rdev;
6302         struct mdstat_info *mi = seq->private;
6303         struct bitmap *bitmap;
6304
6305         if (v == (void*)1) {
6306                 struct mdk_personality *pers;
6307                 seq_printf(seq, "Personalities : ");
6308                 spin_lock(&pers_lock);
6309                 list_for_each_entry(pers, &pers_list, list)
6310                         seq_printf(seq, "[%s] ", pers->name);
6311
6312                 spin_unlock(&pers_lock);
6313                 seq_printf(seq, "\n");
6314                 mi->event = atomic_read(&md_event_count);
6315                 return 0;
6316         }
6317         if (v == (void*)2) {
6318                 status_unused(seq);
6319                 return 0;
6320         }
6321
6322         if (mddev_lock(mddev) < 0)
6323                 return -EINTR;
6324
6325         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6326                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6327                                                 mddev->pers ? "" : "in");
6328                 if (mddev->pers) {
6329                         if (mddev->ro==1)
6330                                 seq_printf(seq, " (read-only)");
6331                         if (mddev->ro==2)
6332                                 seq_printf(seq, " (auto-read-only)");
6333                         seq_printf(seq, " %s", mddev->pers->name);
6334                 }
6335
6336                 sectors = 0;
6337                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6338                         char b[BDEVNAME_SIZE];
6339                         seq_printf(seq, " %s[%d]",
6340                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6341                         if (test_bit(WriteMostly, &rdev->flags))
6342                                 seq_printf(seq, "(W)");
6343                         if (test_bit(Faulty, &rdev->flags)) {
6344                                 seq_printf(seq, "(F)");
6345                                 continue;
6346                         } else if (rdev->raid_disk < 0)
6347                                 seq_printf(seq, "(S)"); /* spare */
6348                         sectors += rdev->sectors;
6349                 }
6350
6351                 if (!list_empty(&mddev->disks)) {
6352                         if (mddev->pers)
6353                                 seq_printf(seq, "\n      %llu blocks",
6354                                            (unsigned long long)
6355                                            mddev->array_sectors / 2);
6356                         else
6357                                 seq_printf(seq, "\n      %llu blocks",
6358                                            (unsigned long long)sectors / 2);
6359                 }
6360                 if (mddev->persistent) {
6361                         if (mddev->major_version != 0 ||
6362                             mddev->minor_version != 90) {
6363                                 seq_printf(seq," super %d.%d",
6364                                            mddev->major_version,
6365                                            mddev->minor_version);
6366                         }
6367                 } else if (mddev->external)
6368                         seq_printf(seq, " super external:%s",
6369                                    mddev->metadata_type);
6370                 else
6371                         seq_printf(seq, " super non-persistent");
6372
6373                 if (mddev->pers) {
6374                         mddev->pers->status(seq, mddev);
6375                         seq_printf(seq, "\n      ");
6376                         if (mddev->pers->sync_request) {
6377                                 if (mddev->curr_resync > 2) {
6378                                         status_resync(seq, mddev);
6379                                         seq_printf(seq, "\n      ");
6380                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6381                                         seq_printf(seq, "\tresync=DELAYED\n      ");
6382                                 else if (mddev->recovery_cp < MaxSector)
6383                                         seq_printf(seq, "\tresync=PENDING\n      ");
6384                         }
6385                 } else
6386                         seq_printf(seq, "\n       ");
6387
6388                 if ((bitmap = mddev->bitmap)) {
6389                         unsigned long chunk_kb;
6390                         unsigned long flags;
6391                         spin_lock_irqsave(&bitmap->lock, flags);
6392                         chunk_kb = mddev->bitmap_info.chunksize >> 10;
6393                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6394                                 "%lu%s chunk",
6395                                 bitmap->pages - bitmap->missing_pages,
6396                                 bitmap->pages,
6397                                 (bitmap->pages - bitmap->missing_pages)
6398                                         << (PAGE_SHIFT - 10),
6399                                 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6400                                 chunk_kb ? "KB" : "B");
6401                         if (bitmap->file) {
6402                                 seq_printf(seq, ", file: ");
6403                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
6404                         }
6405
6406                         seq_printf(seq, "\n");
6407                         spin_unlock_irqrestore(&bitmap->lock, flags);
6408                 }
6409
6410                 seq_printf(seq, "\n");
6411         }
6412         mddev_unlock(mddev);
6413
6414         return 0;
6415 }
6416
6417 static const struct seq_operations md_seq_ops = {
6418         .start  = md_seq_start,
6419         .next   = md_seq_next,
6420         .stop   = md_seq_stop,
6421         .show   = md_seq_show,
6422 };
6423
6424 static int md_seq_open(struct inode *inode, struct file *file)
6425 {
6426         int error;
6427         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6428         if (mi == NULL)
6429                 return -ENOMEM;
6430
6431         error = seq_open(file, &md_seq_ops);
6432         if (error)
6433                 kfree(mi);
6434         else {
6435                 struct seq_file *p = file->private_data;
6436                 p->private = mi;
6437                 mi->event = atomic_read(&md_event_count);
6438         }
6439         return error;
6440 }
6441
6442 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6443 {
6444         struct seq_file *m = filp->private_data;
6445         struct mdstat_info *mi = m->private;
6446         int mask;
6447
6448         poll_wait(filp, &md_event_waiters, wait);
6449
6450         /* always allow read */
6451         mask = POLLIN | POLLRDNORM;
6452
6453         if (mi->event != atomic_read(&md_event_count))
6454                 mask |= POLLERR | POLLPRI;
6455         return mask;
6456 }
6457
6458 static const struct file_operations md_seq_fops = {
6459         .owner          = THIS_MODULE,
6460         .open           = md_seq_open,
6461         .read           = seq_read,
6462         .llseek         = seq_lseek,
6463         .release        = seq_release_private,
6464         .poll           = mdstat_poll,
6465 };
6466
6467 int register_md_personality(struct mdk_personality *p)
6468 {
6469         spin_lock(&pers_lock);
6470         list_add_tail(&p->list, &pers_list);
6471         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6472         spin_unlock(&pers_lock);
6473         return 0;
6474 }
6475
6476 int unregister_md_personality(struct mdk_personality *p)
6477 {
6478         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6479         spin_lock(&pers_lock);
6480         list_del_init(&p->list);
6481         spin_unlock(&pers_lock);
6482         return 0;
6483 }
6484
6485 static int is_mddev_idle(mddev_t *mddev, int init)
6486 {
6487         mdk_rdev_t * rdev;
6488         int idle;
6489         int curr_events;
6490
6491         idle = 1;
6492         rcu_read_lock();
6493         rdev_for_each_rcu(rdev, mddev) {
6494                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6495                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6496                               (int)part_stat_read(&disk->part0, sectors[1]) -
6497                               atomic_read(&disk->sync_io);
6498                 /* sync IO will cause sync_io to increase before the disk_stats
6499                  * as sync_io is counted when a request starts, and
6500                  * disk_stats is counted when it completes.
6501                  * So resync activity will cause curr_events to be smaller than
6502                  * when there was no such activity.
6503                  * non-sync IO will cause disk_stat to increase without
6504                  * increasing sync_io so curr_events will (eventually)
6505                  * be larger than it was before.  Once it becomes
6506                  * substantially larger, the test below will cause
6507                  * the array to appear non-idle, and resync will slow
6508                  * down.
6509                  * If there is a lot of outstanding resync activity when
6510                  * we set last_event to curr_events, then all that activity
6511                  * completing might cause the array to appear non-idle
6512                  * and resync will be slowed down even though there might
6513                  * not have been non-resync activity.  This will only
6514                  * happen once though.  'last_events' will soon reflect
6515                  * the state where there is little or no outstanding
6516                  * resync requests, and further resync activity will
6517                  * always make curr_events less than last_events.
6518                  *
6519                  */
6520                 if (init || curr_events - rdev->last_events > 64) {
6521                         rdev->last_events = curr_events;
6522                         idle = 0;
6523                 }
6524         }
6525         rcu_read_unlock();
6526         return idle;
6527 }
6528
6529 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6530 {
6531         /* another "blocks" (512byte) blocks have been synced */
6532         atomic_sub(blocks, &mddev->recovery_active);
6533         wake_up(&mddev->recovery_wait);
6534         if (!ok) {
6535                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6536                 md_wakeup_thread(mddev->thread);
6537                 // stop recovery, signal do_sync ....
6538         }
6539 }
6540
6541
6542 /* md_write_start(mddev, bi)
6543  * If we need to update some array metadata (e.g. 'active' flag
6544  * in superblock) before writing, schedule a superblock update
6545  * and wait for it to complete.
6546  */
6547 void md_write_start(mddev_t *mddev, struct bio *bi)
6548 {
6549         int did_change = 0;
6550         if (bio_data_dir(bi) != WRITE)
6551                 return;
6552
6553         BUG_ON(mddev->ro == 1);
6554         if (mddev->ro == 2) {
6555                 /* need to switch to read/write */
6556                 mddev->ro = 0;
6557                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6558                 md_wakeup_thread(mddev->thread);
6559                 md_wakeup_thread(mddev->sync_thread);
6560                 did_change = 1;
6561         }
6562         atomic_inc(&mddev->writes_pending);
6563         if (mddev->safemode == 1)
6564                 mddev->safemode = 0;
6565         if (mddev->in_sync) {
6566                 spin_lock_irq(&mddev->write_lock);
6567                 if (mddev->in_sync) {
6568                         mddev->in_sync = 0;
6569                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6570                         set_bit(MD_CHANGE_PENDING, &mddev->flags);
6571                         md_wakeup_thread(mddev->thread);
6572                         did_change = 1;
6573                 }
6574                 spin_unlock_irq(&mddev->write_lock);
6575         }
6576         if (did_change)
6577                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6578         wait_event(mddev->sb_wait,
6579                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6580 }
6581
6582 void md_write_end(mddev_t *mddev)
6583 {
6584         if (atomic_dec_and_test(&mddev->writes_pending)) {
6585                 if (mddev->safemode == 2)
6586                         md_wakeup_thread(mddev->thread);
6587                 else if (mddev->safemode_delay)
6588                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6589         }
6590 }
6591
6592 /* md_allow_write(mddev)
6593  * Calling this ensures that the array is marked 'active' so that writes
6594  * may proceed without blocking.  It is important to call this before
6595  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6596  * Must be called with mddev_lock held.
6597  *
6598  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6599  * is dropped, so return -EAGAIN after notifying userspace.
6600  */
6601 int md_allow_write(mddev_t *mddev)
6602 {
6603         if (!mddev->pers)
6604                 return 0;
6605         if (mddev->ro)
6606                 return 0;
6607         if (!mddev->pers->sync_request)
6608                 return 0;
6609
6610         spin_lock_irq(&mddev->write_lock);
6611         if (mddev->in_sync) {
6612                 mddev->in_sync = 0;
6613                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6614                 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6615                 if (mddev->safemode_delay &&
6616                     mddev->safemode == 0)
6617                         mddev->safemode = 1;
6618                 spin_unlock_irq(&mddev->write_lock);
6619                 md_update_sb(mddev, 0);
6620                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6621         } else
6622                 spin_unlock_irq(&mddev->write_lock);
6623
6624         if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6625                 return -EAGAIN;
6626         else
6627                 return 0;
6628 }
6629 EXPORT_SYMBOL_GPL(md_allow_write);
6630
6631 void md_unplug(mddev_t *mddev)
6632 {
6633         if (mddev->queue)
6634                 blk_unplug(mddev->queue);
6635         if (mddev->plug)
6636                 mddev->plug->unplug_fn(mddev->plug);
6637 }
6638
6639 #define SYNC_MARKS      10
6640 #define SYNC_MARK_STEP  (3*HZ)
6641 void md_do_sync(mddev_t *mddev)
6642 {
6643         mddev_t *mddev2;
6644         unsigned int currspeed = 0,
6645                  window;
6646         sector_t max_sectors,j, io_sectors;
6647         unsigned long mark[SYNC_MARKS];
6648         sector_t mark_cnt[SYNC_MARKS];
6649         int last_mark,m;
6650         struct list_head *tmp;
6651         sector_t last_check;
6652         int skipped = 0;
6653         mdk_rdev_t *rdev;
6654         char *desc;
6655
6656         /* just incase thread restarts... */
6657         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6658                 return;
6659         if (mddev->ro) /* never try to sync a read-only array */
6660                 return;
6661
6662         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6663                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6664                         desc = "data-check";
6665                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6666                         desc = "requested-resync";
6667                 else
6668                         desc = "resync";
6669         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6670                 desc = "reshape";
6671         else
6672                 desc = "recovery";
6673
6674         /* we overload curr_resync somewhat here.
6675          * 0 == not engaged in resync at all
6676          * 2 == checking that there is no conflict with another sync
6677          * 1 == like 2, but have yielded to allow conflicting resync to
6678          *              commense
6679          * other == active in resync - this many blocks
6680          *
6681          * Before starting a resync we must have set curr_resync to
6682          * 2, and then checked that every "conflicting" array has curr_resync
6683          * less than ours.  When we find one that is the same or higher
6684          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6685          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6686          * This will mean we have to start checking from the beginning again.
6687          *
6688          */
6689
6690         do {
6691                 mddev->curr_resync = 2;
6692
6693         try_again:
6694                 if (kthread_should_stop())
6695                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6696
6697                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6698                         goto skip;
6699                 for_each_mddev(mddev2, tmp) {
6700                         if (mddev2 == mddev)
6701                                 continue;
6702                         if (!mddev->parallel_resync
6703                         &&  mddev2->curr_resync
6704                         &&  match_mddev_units(mddev, mddev2)) {
6705                                 DEFINE_WAIT(wq);
6706                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6707                                         /* arbitrarily yield */
6708                                         mddev->curr_resync = 1;
6709                                         wake_up(&resync_wait);
6710                                 }
6711                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6712                                         /* no need to wait here, we can wait the next
6713                                          * time 'round when curr_resync == 2
6714                                          */
6715                                         continue;
6716                                 /* We need to wait 'interruptible' so as not to
6717                                  * contribute to the load average, and not to
6718                                  * be caught by 'softlockup'
6719                                  */
6720                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6721                                 if (!kthread_should_stop() &&
6722                                     mddev2->curr_resync >= mddev->curr_resync) {
6723                                         printk(KERN_INFO "md: delaying %s of %s"
6724                                                " until %s has finished (they"
6725                                                " share one or more physical units)\n",
6726                                                desc, mdname(mddev), mdname(mddev2));
6727                                         mddev_put(mddev2);
6728                                         if (signal_pending(current))
6729                                                 flush_signals(current);
6730                                         schedule();
6731                                         finish_wait(&resync_wait, &wq);
6732                                         goto try_again;
6733                                 }
6734                                 finish_wait(&resync_wait, &wq);
6735                         }
6736                 }
6737         } while (mddev->curr_resync < 2);
6738
6739         j = 0;
6740         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6741                 /* resync follows the size requested by the personality,
6742                  * which defaults to physical size, but can be virtual size
6743                  */
6744                 max_sectors = mddev->resync_max_sectors;
6745                 mddev->resync_mismatches = 0;
6746                 /* we don't use the checkpoint if there's a bitmap */
6747                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6748                         j = mddev->resync_min;
6749                 else if (!mddev->bitmap)
6750                         j = mddev->recovery_cp;
6751
6752         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6753                 max_sectors = mddev->dev_sectors;
6754         else {
6755                 /* recovery follows the physical size of devices */
6756                 max_sectors = mddev->dev_sectors;
6757                 j = MaxSector;
6758                 rcu_read_lock();
6759                 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6760                         if (rdev->raid_disk >= 0 &&
6761                             !test_bit(Faulty, &rdev->flags) &&
6762                             !test_bit(In_sync, &rdev->flags) &&
6763                             rdev->recovery_offset < j)
6764                                 j = rdev->recovery_offset;
6765                 rcu_read_unlock();
6766         }
6767
6768         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6769         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6770                 " %d KB/sec/disk.\n", speed_min(mddev));
6771         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6772                "(but not more than %d KB/sec) for %s.\n",
6773                speed_max(mddev), desc);
6774
6775         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6776
6777         io_sectors = 0;
6778         for (m = 0; m < SYNC_MARKS; m++) {
6779                 mark[m] = jiffies;
6780                 mark_cnt[m] = io_sectors;
6781         }
6782         last_mark = 0;
6783         mddev->resync_mark = mark[last_mark];
6784         mddev->resync_mark_cnt = mark_cnt[last_mark];
6785
6786         /*
6787          * Tune reconstruction:
6788          */
6789         window = 32*(PAGE_SIZE/512);
6790         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6791                 window/2,(unsigned long long) max_sectors/2);
6792
6793         atomic_set(&mddev->recovery_active, 0);
6794         last_check = 0;
6795
6796         if (j>2) {
6797                 printk(KERN_INFO
6798                        "md: resuming %s of %s from checkpoint.\n",
6799                        desc, mdname(mddev));
6800                 mddev->curr_resync = j;
6801         }
6802         mddev->curr_resync_completed = mddev->curr_resync;
6803
6804         while (j < max_sectors) {
6805                 sector_t sectors;
6806
6807                 skipped = 0;
6808
6809                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6810                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6811                       (mddev->curr_resync - mddev->curr_resync_completed)
6812                       > (max_sectors >> 4)) ||
6813                      (j - mddev->curr_resync_completed)*2
6814                      >= mddev->resync_max - mddev->curr_resync_completed
6815                             )) {
6816                         /* time to update curr_resync_completed */
6817                         md_unplug(mddev);
6818                         wait_event(mddev->recovery_wait,
6819                                    atomic_read(&mddev->recovery_active) == 0);
6820                         mddev->curr_resync_completed =
6821                                 mddev->curr_resync;
6822                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6823                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6824                 }
6825
6826                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6827                         /* As this condition is controlled by user-space,
6828                          * we can block indefinitely, so use '_interruptible'
6829                          * to avoid triggering warnings.
6830                          */
6831                         flush_signals(current); /* just in case */
6832                         wait_event_interruptible(mddev->recovery_wait,
6833                                                  mddev->resync_max > j
6834                                                  || kthread_should_stop());
6835                 }
6836
6837                 if (kthread_should_stop())
6838                         goto interrupted;
6839
6840                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6841                                                   currspeed < speed_min(mddev));
6842                 if (sectors == 0) {
6843                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6844                         goto out;
6845                 }
6846
6847                 if (!skipped) { /* actual IO requested */
6848                         io_sectors += sectors;
6849                         atomic_add(sectors, &mddev->recovery_active);
6850                 }
6851
6852                 j += sectors;
6853                 if (j>1) mddev->curr_resync = j;
6854                 mddev->curr_mark_cnt = io_sectors;
6855                 if (last_check == 0)
6856                         /* this is the earliers that rebuilt will be
6857                          * visible in /proc/mdstat
6858                          */
6859                         md_new_event(mddev);
6860
6861                 if (last_check + window > io_sectors || j == max_sectors)
6862                         continue;
6863
6864                 last_check = io_sectors;
6865
6866                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6867                         break;
6868
6869         repeat:
6870                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6871                         /* step marks */
6872                         int next = (last_mark+1) % SYNC_MARKS;
6873
6874                         mddev->resync_mark = mark[next];
6875                         mddev->resync_mark_cnt = mark_cnt[next];
6876                         mark[next] = jiffies;
6877                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6878                         last_mark = next;
6879                 }
6880
6881
6882                 if (kthread_should_stop())
6883                         goto interrupted;
6884
6885
6886                 /*
6887                  * this loop exits only if either when we are slower than
6888                  * the 'hard' speed limit, or the system was IO-idle for
6889                  * a jiffy.
6890                  * the system might be non-idle CPU-wise, but we only care
6891                  * about not overloading the IO subsystem. (things like an
6892                  * e2fsck being done on the RAID array should execute fast)
6893                  */
6894                 md_unplug(mddev);
6895                 cond_resched();
6896
6897                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6898                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6899
6900                 if (currspeed > speed_min(mddev)) {
6901                         if ((currspeed > speed_max(mddev)) ||
6902                                         !is_mddev_idle(mddev, 0)) {
6903                                 msleep(500);
6904                                 goto repeat;
6905                         }
6906                 }
6907         }
6908         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6909         /*
6910          * this also signals 'finished resyncing' to md_stop
6911          */
6912  out:
6913         md_unplug(mddev);
6914
6915         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6916
6917         /* tell personality that we are finished */
6918         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6919
6920         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6921             mddev->curr_resync > 2) {
6922                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6923                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6924                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6925                                         printk(KERN_INFO
6926                                                "md: checkpointing %s of %s.\n",
6927                                                desc, mdname(mddev));
6928                                         mddev->recovery_cp = mddev->curr_resync;
6929                                 }
6930                         } else
6931                                 mddev->recovery_cp = MaxSector;
6932                 } else {
6933                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6934                                 mddev->curr_resync = MaxSector;
6935                         rcu_read_lock();
6936                         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6937                                 if (rdev->raid_disk >= 0 &&
6938                                     mddev->delta_disks >= 0 &&
6939                                     !test_bit(Faulty, &rdev->flags) &&
6940                                     !test_bit(In_sync, &rdev->flags) &&
6941                                     rdev->recovery_offset < mddev->curr_resync)
6942                                         rdev->recovery_offset = mddev->curr_resync;
6943                         rcu_read_unlock();
6944                 }
6945         }
6946         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6947
6948  skip:
6949         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6950                 /* We completed so min/max setting can be forgotten if used. */
6951                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6952                         mddev->resync_min = 0;
6953                 mddev->resync_max = MaxSector;
6954         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6955                 mddev->resync_min = mddev->curr_resync_completed;
6956         mddev->curr_resync = 0;
6957         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6958                 mddev->curr_resync_completed = 0;
6959         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6960         wake_up(&resync_wait);
6961         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6962         md_wakeup_thread(mddev->thread);
6963         return;
6964
6965  interrupted:
6966         /*
6967          * got a signal, exit.
6968          */
6969         printk(KERN_INFO
6970                "md: md_do_sync() got signal ... exiting\n");
6971         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6972         goto out;
6973
6974 }
6975 EXPORT_SYMBOL_GPL(md_do_sync);
6976
6977
6978 static int remove_and_add_spares(mddev_t *mddev)
6979 {
6980         mdk_rdev_t *rdev;
6981         int spares = 0;
6982
6983         mddev->curr_resync_completed = 0;
6984
6985         list_for_each_entry(rdev, &mddev->disks, same_set)
6986                 if (rdev->raid_disk >= 0 &&
6987                     !test_bit(Blocked, &rdev->flags) &&
6988                     (test_bit(Faulty, &rdev->flags) ||
6989                      ! test_bit(In_sync, &rdev->flags)) &&
6990                     atomic_read(&rdev->nr_pending)==0) {
6991                         if (mddev->pers->hot_remove_disk(
6992                                     mddev, rdev->raid_disk)==0) {
6993                                 char nm[20];
6994                                 sprintf(nm,"rd%d", rdev->raid_disk);
6995                                 sysfs_remove_link(&mddev->kobj, nm);
6996                                 rdev->raid_disk = -1;
6997                         }
6998                 }
6999
7000         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7001                 list_for_each_entry(rdev, &mddev->disks, same_set) {
7002                         if (rdev->raid_disk >= 0 &&
7003                             !test_bit(In_sync, &rdev->flags) &&
7004                             !test_bit(Blocked, &rdev->flags))
7005                                 spares++;
7006                         if (rdev->raid_disk < 0
7007                             && !test_bit(Faulty, &rdev->flags)) {
7008                                 rdev->recovery_offset = 0;
7009                                 if (mddev->pers->
7010                                     hot_add_disk(mddev, rdev) == 0) {
7011                                         char nm[20];
7012                                         sprintf(nm, "rd%d", rdev->raid_disk);
7013                                         if (sysfs_create_link(&mddev->kobj,
7014                                                               &rdev->kobj, nm))
7015                                                 /* failure here is OK */;
7016                                         spares++;
7017                                         md_new_event(mddev);
7018                                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7019                                 } else
7020                                         break;
7021                         }
7022                 }
7023         }
7024         return spares;
7025 }
7026 /*
7027  * This routine is regularly called by all per-raid-array threads to
7028  * deal with generic issues like resync and super-block update.
7029  * Raid personalities that don't have a thread (linear/raid0) do not
7030  * need this as they never do any recovery or update the superblock.
7031  *
7032  * It does not do any resync itself, but rather "forks" off other threads
7033  * to do that as needed.
7034  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7035  * "->recovery" and create a thread at ->sync_thread.
7036  * When the thread finishes it sets MD_RECOVERY_DONE
7037  * and wakeups up this thread which will reap the thread and finish up.
7038  * This thread also removes any faulty devices (with nr_pending == 0).
7039  *
7040  * The overall approach is:
7041  *  1/ if the superblock needs updating, update it.
7042  *  2/ If a recovery thread is running, don't do anything else.
7043  *  3/ If recovery has finished, clean up, possibly marking spares active.
7044  *  4/ If there are any faulty devices, remove them.
7045  *  5/ If array is degraded, try to add spares devices
7046  *  6/ If array has spares or is not in-sync, start a resync thread.
7047  */
7048 void md_check_recovery(mddev_t *mddev)
7049 {
7050         mdk_rdev_t *rdev;
7051
7052
7053         if (mddev->bitmap)
7054                 bitmap_daemon_work(mddev);
7055
7056         if (mddev->ro)
7057                 return;
7058
7059         if (signal_pending(current)) {
7060                 if (mddev->pers->sync_request && !mddev->external) {
7061                         printk(KERN_INFO "md: %s in immediate safe mode\n",
7062                                mdname(mddev));
7063                         mddev->safemode = 2;
7064                 }
7065                 flush_signals(current);
7066         }
7067
7068         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7069                 return;
7070         if ( ! (
7071                 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7072                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7073                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7074                 (mddev->external == 0 && mddev->safemode == 1) ||
7075                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7076                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7077                 ))
7078                 return;
7079
7080         if (mddev_trylock(mddev)) {
7081                 int spares = 0;
7082
7083                 if (mddev->ro) {
7084                         /* Only thing we do on a ro array is remove
7085                          * failed devices.
7086                          */
7087                         remove_and_add_spares(mddev);
7088                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7089                         goto unlock;
7090                 }
7091
7092                 if (!mddev->external) {
7093                         int did_change = 0;
7094                         spin_lock_irq(&mddev->write_lock);
7095                         if (mddev->safemode &&
7096                             !atomic_read(&mddev->writes_pending) &&
7097                             !mddev->in_sync &&
7098                             mddev->recovery_cp == MaxSector) {
7099                                 mddev->in_sync = 1;
7100                                 did_change = 1;
7101                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7102                         }
7103                         if (mddev->safemode == 1)
7104                                 mddev->safemode = 0;
7105                         spin_unlock_irq(&mddev->write_lock);
7106                         if (did_change)
7107                                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7108                 }
7109
7110                 if (mddev->flags)
7111                         md_update_sb(mddev, 0);
7112
7113                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7114                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7115                         /* resync/recovery still happening */
7116                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7117                         goto unlock;
7118                 }
7119                 if (mddev->sync_thread) {
7120                         /* resync has finished, collect result */
7121                         md_unregister_thread(mddev->sync_thread);
7122                         mddev->sync_thread = NULL;
7123                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7124                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7125                                 /* success...*/
7126                                 /* activate any spares */
7127                                 if (mddev->pers->spare_active(mddev))
7128                                         sysfs_notify(&mddev->kobj, NULL,
7129                                                      "degraded");
7130                         }
7131                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7132                             mddev->pers->finish_reshape)
7133                                 mddev->pers->finish_reshape(mddev);
7134                         md_update_sb(mddev, 1);
7135
7136                         /* if array is no-longer degraded, then any saved_raid_disk
7137                          * information must be scrapped
7138                          */
7139                         if (!mddev->degraded)
7140                                 list_for_each_entry(rdev, &mddev->disks, same_set)
7141                                         rdev->saved_raid_disk = -1;
7142
7143                         mddev->recovery = 0;
7144                         /* flag recovery needed just to double check */
7145                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7146                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7147                         md_new_event(mddev);
7148                         goto unlock;
7149                 }
7150                 /* Set RUNNING before clearing NEEDED to avoid
7151                  * any transients in the value of "sync_action".
7152                  */
7153                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7154                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7155                 /* Clear some bits that don't mean anything, but
7156                  * might be left set
7157                  */
7158                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7159                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7160
7161                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7162                         goto unlock;
7163                 /* no recovery is running.
7164                  * remove any failed drives, then
7165                  * add spares if possible.
7166                  * Spare are also removed and re-added, to allow
7167                  * the personality to fail the re-add.
7168                  */
7169
7170                 if (mddev->reshape_position != MaxSector) {
7171                         if (mddev->pers->check_reshape == NULL ||
7172                             mddev->pers->check_reshape(mddev) != 0)
7173                                 /* Cannot proceed */
7174                                 goto unlock;
7175                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7176                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7177                 } else if ((spares = remove_and_add_spares(mddev))) {
7178                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7179                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7180                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7181                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7182                 } else if (mddev->recovery_cp < MaxSector) {
7183                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7184                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7185                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7186                         /* nothing to be done ... */
7187                         goto unlock;
7188
7189                 if (mddev->pers->sync_request) {
7190                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7191                                 /* We are adding a device or devices to an array
7192                                  * which has the bitmap stored on all devices.
7193                                  * So make sure all bitmap pages get written
7194                                  */
7195                                 bitmap_write_all(mddev->bitmap);
7196                         }
7197                         mddev->sync_thread = md_register_thread(md_do_sync,
7198                                                                 mddev,
7199                                                                 "resync");
7200                         if (!mddev->sync_thread) {
7201                                 printk(KERN_ERR "%s: could not start resync"
7202                                         " thread...\n",
7203                                         mdname(mddev));
7204                                 /* leave the spares where they are, it shouldn't hurt */
7205                                 mddev->recovery = 0;
7206                         } else
7207                                 md_wakeup_thread(mddev->sync_thread);
7208                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7209                         md_new_event(mddev);
7210                 }
7211         unlock:
7212                 if (!mddev->sync_thread) {
7213                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7214                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7215                                                &mddev->recovery))
7216                                 if (mddev->sysfs_action)
7217                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7218                 }
7219                 mddev_unlock(mddev);
7220         }
7221 }
7222
7223 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7224 {
7225         sysfs_notify_dirent_safe(rdev->sysfs_state);
7226         wait_event_timeout(rdev->blocked_wait,
7227                            !test_bit(Blocked, &rdev->flags),
7228                            msecs_to_jiffies(5000));
7229         rdev_dec_pending(rdev, mddev);
7230 }
7231 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7232
7233 static int md_notify_reboot(struct notifier_block *this,
7234                             unsigned long code, void *x)
7235 {
7236         struct list_head *tmp;
7237         mddev_t *mddev;
7238
7239         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7240
7241                 printk(KERN_INFO "md: stopping all md devices.\n");
7242
7243                 for_each_mddev(mddev, tmp)
7244                         if (mddev_trylock(mddev)) {
7245                                 /* Force a switch to readonly even array
7246                                  * appears to still be in use.  Hence
7247                                  * the '100'.
7248                                  */
7249                                 md_set_readonly(mddev, 100);
7250                                 mddev_unlock(mddev);
7251                         }
7252                 /*
7253                  * certain more exotic SCSI devices are known to be
7254                  * volatile wrt too early system reboots. While the
7255                  * right place to handle this issue is the given
7256                  * driver, we do want to have a safe RAID driver ...
7257                  */
7258                 mdelay(1000*1);
7259         }
7260         return NOTIFY_DONE;
7261 }
7262
7263 static struct notifier_block md_notifier = {
7264         .notifier_call  = md_notify_reboot,
7265         .next           = NULL,
7266         .priority       = INT_MAX, /* before any real devices */
7267 };
7268
7269 static void md_geninit(void)
7270 {
7271         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7272
7273         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7274 }
7275
7276 static int __init md_init(void)
7277 {
7278         int ret = -ENOMEM;
7279
7280         md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7281         if (!md_wq)
7282                 goto err_wq;
7283
7284         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7285         if (!md_misc_wq)
7286                 goto err_misc_wq;
7287
7288         if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7289                 goto err_md;
7290
7291         if ((ret = register_blkdev(0, "mdp")) < 0)
7292                 goto err_mdp;
7293         mdp_major = ret;
7294
7295         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7296                             md_probe, NULL, NULL);
7297         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7298                             md_probe, NULL, NULL);
7299
7300         register_reboot_notifier(&md_notifier);
7301         raid_table_header = register_sysctl_table(raid_root_table);
7302
7303         md_geninit();
7304         return 0;
7305
7306 err_mdp:
7307         unregister_blkdev(MD_MAJOR, "md");
7308 err_md:
7309         destroy_workqueue(md_misc_wq);
7310 err_misc_wq:
7311         destroy_workqueue(md_wq);
7312 err_wq:
7313         return ret;
7314 }
7315
7316 #ifndef MODULE
7317
7318 /*
7319  * Searches all registered partitions for autorun RAID arrays
7320  * at boot time.
7321  */
7322
7323 static LIST_HEAD(all_detected_devices);
7324 struct detected_devices_node {
7325         struct list_head list;
7326         dev_t dev;
7327 };
7328
7329 void md_autodetect_dev(dev_t dev)
7330 {
7331         struct detected_devices_node *node_detected_dev;
7332
7333         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7334         if (node_detected_dev) {
7335                 node_detected_dev->dev = dev;
7336                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7337         } else {
7338                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7339                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7340         }
7341 }
7342
7343
7344 static void autostart_arrays(int part)
7345 {
7346         mdk_rdev_t *rdev;
7347         struct detected_devices_node *node_detected_dev;
7348         dev_t dev;
7349         int i_scanned, i_passed;
7350
7351         i_scanned = 0;
7352         i_passed = 0;
7353
7354         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7355
7356         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7357                 i_scanned++;
7358                 node_detected_dev = list_entry(all_detected_devices.next,
7359                                         struct detected_devices_node, list);
7360                 list_del(&node_detected_dev->list);
7361                 dev = node_detected_dev->dev;
7362                 kfree(node_detected_dev);
7363                 rdev = md_import_device(dev,0, 90);
7364                 if (IS_ERR(rdev))
7365                         continue;
7366
7367                 if (test_bit(Faulty, &rdev->flags)) {
7368                         MD_BUG();
7369                         continue;
7370                 }
7371                 set_bit(AutoDetected, &rdev->flags);
7372                 list_add(&rdev->same_set, &pending_raid_disks);
7373                 i_passed++;
7374         }
7375
7376         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7377                                                 i_scanned, i_passed);
7378
7379         autorun_devices(part);
7380 }
7381
7382 #endif /* !MODULE */
7383
7384 static __exit void md_exit(void)
7385 {
7386         mddev_t *mddev;
7387         struct list_head *tmp;
7388
7389         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7390         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7391
7392         unregister_blkdev(MD_MAJOR,"md");
7393         unregister_blkdev(mdp_major, "mdp");
7394         unregister_reboot_notifier(&md_notifier);
7395         unregister_sysctl_table(raid_table_header);
7396         remove_proc_entry("mdstat", NULL);
7397         for_each_mddev(mddev, tmp) {
7398                 export_array(mddev);
7399                 mddev->hold_active = 0;
7400         }
7401         destroy_workqueue(md_misc_wq);
7402         destroy_workqueue(md_wq);
7403 }
7404
7405 subsys_initcall(md_init);
7406 module_exit(md_exit)
7407
7408 static int get_ro(char *buffer, struct kernel_param *kp)
7409 {
7410         return sprintf(buffer, "%d", start_readonly);
7411 }
7412 static int set_ro(const char *val, struct kernel_param *kp)
7413 {
7414         char *e;
7415         int num = simple_strtoul(val, &e, 10);
7416         if (*val && (*e == '\0' || *e == '\n')) {
7417                 start_readonly = num;
7418                 return 0;
7419         }
7420         return -EINVAL;
7421 }
7422
7423 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7424 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7425
7426 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7427
7428 EXPORT_SYMBOL(register_md_personality);
7429 EXPORT_SYMBOL(unregister_md_personality);
7430 EXPORT_SYMBOL(md_error);
7431 EXPORT_SYMBOL(md_done_sync);
7432 EXPORT_SYMBOL(md_write_start);
7433 EXPORT_SYMBOL(md_write_end);
7434 EXPORT_SYMBOL(md_register_thread);
7435 EXPORT_SYMBOL(md_unregister_thread);
7436 EXPORT_SYMBOL(md_wakeup_thread);
7437 EXPORT_SYMBOL(md_check_recovery);
7438 MODULE_LICENSE("GPL");
7439 MODULE_DESCRIPTION("MD RAID framework");
7440 MODULE_ALIAS("md");
7441 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);