drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/kthread.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/sysctl.h>
  38 #include <linux/seq_file.h>
  39 #include <linux/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         unsigned int sectors;
 291
 292         if (mddev == NULL || mddev->pers == NULL
 293             || !mddev->ready) {
 294                 bio_io_error(bio);
 295                 return 0;
 296         }
 297         smp_rmb(); /* Ensure implications of  'active' are visible */
 298         rcu_read_lock();
 299         if (mddev->suspended) {
 300                 DEFINE_WAIT(__wait);
 301                 for (;;) {
 302                         prepare_to_wait(&mddev->sb_wait, &__wait,
 303                                         TASK_UNINTERRUPTIBLE);
 304                         if (!mddev->suspended)
 305                                 break;
 306                         rcu_read_unlock();
 307                         schedule();
 308                         rcu_read_lock();
 309                 }
 310                 finish_wait(&mddev->sb_wait, &__wait);
 311         }
 312         atomic_inc(&mddev->active_io);
 313         rcu_read_unlock();
 314
 315         /*
 316          * save the sectors now since our bio can
 317          * go away inside make_request
 318          */
 319         sectors = bio_sectors(bio);
 320         rv = mddev->pers->make_request(mddev, bio);
 321
 322         cpu = part_stat_lock();
 323         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 324         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
 325         part_stat_unlock();
 326
 327         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 328                 wake_up(&mddev->sb_wait);
 329
 330         return rv;
 331 }
 332
 333 /* mddev_suspend makes sure no new requests are submitted
 334  * to the device, and that any requests that have been submitted
 335  * are completely handled.
 336  * Once ->stop is called and completes, the module will be completely
 337  * unused.
 338  */
 339 void mddev_suspend(mddev_t *mddev)
 340 {
 341         BUG_ON(mddev->suspended);
 342         mddev->suspended = 1;
 343         synchronize_rcu();
 344         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 345         mddev->pers->quiesce(mddev, 1);
 346 }
 347 EXPORT_SYMBOL_GPL(mddev_suspend);
 348
 349 void mddev_resume(mddev_t *mddev)
 350 {
 351         mddev->suspended = 0;
 352         wake_up(&mddev->sb_wait);
 353         mddev->pers->quiesce(mddev, 0);
 354 }
 355 EXPORT_SYMBOL_GPL(mddev_resume);
 356
 357 int mddev_congested(mddev_t *mddev, int bits)
 358 {
 359         return mddev->suspended;
 360 }
 361 EXPORT_SYMBOL(mddev_congested);
 362
 363 /*
 364  * Generic flush handling for md
 365  */
 366
 367 static void md_end_flush(struct bio *bio, int err)
 368 {
 369         mdk_rdev_t *rdev = bio->bi_private;
 370         mddev_t *mddev = rdev->mddev;
 371
 372         rdev_dec_pending(rdev, mddev);
 373
 374         if (atomic_dec_and_test(&mddev->flush_pending)) {
 375                 /* The pre-request flush has finished */
 376                 queue_work(md_wq, &mddev->flush_work);
 377         }
 378         bio_put(bio);
 379 }
 380
 381 static void md_submit_flush_data(struct work_struct *ws);
 382
 383 static void submit_flushes(struct work_struct *ws)
 384 {
 385         mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 386         mdk_rdev_t *rdev;
 387
 388         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 389         atomic_set(&mddev->flush_pending, 1);
 390         rcu_read_lock();
 391         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
 392                 if (rdev->raid_disk >= 0 &&
 393                     !test_bit(Faulty, &rdev->flags)) {
 394                         /* Take two references, one is dropped
 395                          * when request finishes, one after
 396                          * we reclaim rcu_read_lock
 397                          */
 398                         struct bio *bi;
 399                         atomic_inc(&rdev->nr_pending);
 400                         atomic_inc(&rdev->nr_pending);
 401                         rcu_read_unlock();
 402                         bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
 403                         bi->bi_end_io = md_end_flush;
 404                         bi->bi_private = rdev;
 405                         bi->bi_bdev = rdev->bdev;
 406                         atomic_inc(&mddev->flush_pending);
 407                         submit_bio(WRITE_FLUSH, bi);
 408                         rcu_read_lock();
 409                         rdev_dec_pending(rdev, mddev);
 410                 }
 411         rcu_read_unlock();
 412         if (atomic_dec_and_test(&mddev->flush_pending))
 413                 queue_work(md_wq, &mddev->flush_work);
 414 }
 415
 416 static void md_submit_flush_data(struct work_struct *ws)
 417 {
 418         mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 419         struct bio *bio = mddev->flush_bio;
 420
 421         if (bio->bi_size == 0)
 422                 /* an empty barrier - all done */
 423                 bio_endio(bio, 0);
 424         else {
 425                 bio->bi_rw &= ~REQ_FLUSH;
 426                 if (mddev->pers->make_request(mddev, bio))
 427                         generic_make_request(bio);
 428         }
 429
 430         mddev->flush_bio = NULL;
 431         wake_up(&mddev->sb_wait);
 432 }
 433
 434 void md_flush_request(mddev_t *mddev, struct bio *bio)
 435 {
 436         spin_lock_irq(&mddev->write_lock);
 437         wait_event_lock_irq(mddev->sb_wait,
 438                             !mddev->flush_bio,
 439                             mddev->write_lock, /*nothing*/);
 440         mddev->flush_bio = bio;
 441         spin_unlock_irq(&mddev->write_lock);
 442
 443         INIT_WORK(&mddev->flush_work, submit_flushes);
 444         queue_work(md_wq, &mddev->flush_work);
 445 }
 446 EXPORT_SYMBOL(md_flush_request);
 447
 448 /* Support for plugging.
 449  * This mirrors the plugging support in request_queue, but does not
 450  * require having a whole queue
 451  */
 452 static void plugger_work(struct work_struct *work)
 453 {
 454         struct plug_handle *plug =
 455                 container_of(work, struct plug_handle, unplug_work);
 456         plug->unplug_fn(plug);
 457 }
 458 static void plugger_timeout(unsigned long data)
 459 {
 460         struct plug_handle *plug = (void *)data;
 461         kblockd_schedule_work(NULL, &plug->unplug_work);
 462 }
 463 void plugger_init(struct plug_handle *plug,
 464                   void (*unplug_fn)(struct plug_handle *))
 465 {
 466         plug->unplug_flag = 0;
 467         plug->unplug_fn = unplug_fn;
 468         init_timer(&plug->unplug_timer);
 469         plug->unplug_timer.function = plugger_timeout;
 470         plug->unplug_timer.data = (unsigned long)plug;
 471         INIT_WORK(&plug->unplug_work, plugger_work);
 472 }
 473 EXPORT_SYMBOL_GPL(plugger_init);
 474
 475 void plugger_set_plug(struct plug_handle *plug)
 476 {
 477         if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
 478                 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
 479 }
 480 EXPORT_SYMBOL_GPL(plugger_set_plug);
 481
 482 int plugger_remove_plug(struct plug_handle *plug)
 483 {
 484         if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
 485                 del_timer(&plug->unplug_timer);
 486                 return 1;
 487         } else
 488                 return 0;
 489 }
 490 EXPORT_SYMBOL_GPL(plugger_remove_plug);
 491
 492
 493 static inline mddev_t *mddev_get(mddev_t *mddev)
 494 {
 495         atomic_inc(&mddev->active);
 496         return mddev;
 497 }
 498
 499 static void mddev_delayed_delete(struct work_struct *ws);
 500
 501 static void mddev_put(mddev_t *mddev)
 502 {
 503         struct bio_set *bs = NULL;
 504
 505         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 506                 return;
 507         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 508             mddev->ctime == 0 && !mddev->hold_active) {
 509                 /* Array is not configured at all, and not held active,
 510                  * so destroy it */
 511                 list_del(&mddev->all_mddevs);
 512                 bs = mddev->bio_set;
 513                 mddev->bio_set = NULL;
 514                 if (mddev->gendisk) {
 515                         /* We did a probe so need to clean up.  Call
 516                          * queue_work inside the spinlock so that
 517                          * flush_workqueue() after mddev_find will
 518                          * succeed in waiting for the work to be done.
 519                          */
 520                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 521                         queue_work(md_misc_wq, &mddev->del_work);
 522                 } else
 523                         kfree(mddev);
 524         }
 525         spin_unlock(&all_mddevs_lock);
 526         if (bs)
 527                 bioset_free(bs);
 528 }
 529
 530 void mddev_init(mddev_t *mddev)
 531 {
 532         mutex_init(&mddev->open_mutex);
 533         mutex_init(&mddev->reconfig_mutex);
 534         mutex_init(&mddev->bitmap_info.mutex);
 535         INIT_LIST_HEAD(&mddev->disks);
 536         INIT_LIST_HEAD(&mddev->all_mddevs);
 537         init_timer(&mddev->safemode_timer);
 538         atomic_set(&mddev->active, 1);
 539         atomic_set(&mddev->openers, 0);
 540         atomic_set(&mddev->active_io, 0);
 541         spin_lock_init(&mddev->write_lock);
 542         atomic_set(&mddev->flush_pending, 0);
 543         init_waitqueue_head(&mddev->sb_wait);
 544         init_waitqueue_head(&mddev->recovery_wait);
 545         mddev->reshape_position = MaxSector;
 546         mddev->resync_min = 0;
 547         mddev->resync_max = MaxSector;
 548         mddev->level = LEVEL_NONE;
 549 }
 550 EXPORT_SYMBOL_GPL(mddev_init);
 551
 552 static mddev_t * mddev_find(dev_t unit)
 553 {
 554         mddev_t *mddev, *new = NULL;
 555
 556         if (unit && MAJOR(unit) != MD_MAJOR)
 557                 unit &= ~((1<<MdpMinorShift)-1);
 558
 559  retry:
 560         spin_lock(&all_mddevs_lock);
 561
 562         if (unit) {
 563                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 564                         if (mddev->unit == unit) {
 565                                 mddev_get(mddev);
 566                                 spin_unlock(&all_mddevs_lock);
 567                                 kfree(new);
 568                                 return mddev;
 569                         }
 570
 571                 if (new) {
 572                         list_add(&new->all_mddevs, &all_mddevs);
 573                         spin_unlock(&all_mddevs_lock);
 574                         new->hold_active = UNTIL_IOCTL;
 575                         return new;
 576                 }
 577         } else if (new) {
 578                 /* find an unused unit number */
 579                 static int next_minor = 512;
 580                 int start = next_minor;
 581                 int is_free = 0;
 582                 int dev = 0;
 583                 while (!is_free) {
 584                         dev = MKDEV(MD_MAJOR, next_minor);
 585                         next_minor++;
 586                         if (next_minor > MINORMASK)
 587                                 next_minor = 0;
 588                         if (next_minor == start) {
 589                                 /* Oh dear, all in use. */
 590                                 spin_unlock(&all_mddevs_lock);
 591                                 kfree(new);
 592                                 return NULL;
 593                         }
 594
 595                         is_free = 1;
 596                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 597                                 if (mddev->unit == dev) {
 598                                         is_free = 0;
 599                                         break;
 600                                 }
 601                 }
 602                 new->unit = dev;
 603                 new->md_minor = MINOR(dev);
 604                 new->hold_active = UNTIL_STOP;
 605                 list_add(&new->all_mddevs, &all_mddevs);
 606                 spin_unlock(&all_mddevs_lock);
 607                 return new;
 608         }
 609         spin_unlock(&all_mddevs_lock);
 610
 611         new = kzalloc(sizeof(*new), GFP_KERNEL);
 612         if (!new)
 613                 return NULL;
 614
 615         new->unit = unit;
 616         if (MAJOR(unit) == MD_MAJOR)
 617                 new->md_minor = MINOR(unit);
 618         else
 619                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 620
 621         mddev_init(new);
 622
 623         goto retry;
 624 }
 625
 626 static inline int mddev_lock(mddev_t * mddev)
 627 {
 628         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 629 }
 630
 631 static inline int mddev_is_locked(mddev_t *mddev)
 632 {
 633         return mutex_is_locked(&mddev->reconfig_mutex);
 634 }
 635
 636 static inline int mddev_trylock(mddev_t * mddev)
 637 {
 638         return mutex_trylock(&mddev->reconfig_mutex);
 639 }
 640
 641 static struct attribute_group md_redundancy_group;
 642
 643 static void mddev_unlock(mddev_t * mddev)
 644 {
 645         if (mddev->to_remove) {
 646                 /* These cannot be removed under reconfig_mutex as
 647                  * an access to the files will try to take reconfig_mutex
 648                  * while holding the file unremovable, which leads to
 649                  * a deadlock.
 650                  * So hold set sysfs_active while the remove in happeing,
 651                  * and anything else which might set ->to_remove or my
 652                  * otherwise change the sysfs namespace will fail with
 653                  * -EBUSY if sysfs_active is still set.
 654                  * We set sysfs_active under reconfig_mutex and elsewhere
 655                  * test it under the same mutex to ensure its correct value
 656                  * is seen.
 657                  */
 658                 struct attribute_group *to_remove = mddev->to_remove;
 659                 mddev->to_remove = NULL;
 660                 mddev->sysfs_active = 1;
 661                 mutex_unlock(&mddev->reconfig_mutex);
 662
 663                 if (mddev->kobj.sd) {
 664                         if (to_remove != &md_redundancy_group)
 665                                 sysfs_remove_group(&mddev->kobj, to_remove);
 666                         if (mddev->pers == NULL ||
 667                             mddev->pers->sync_request == NULL) {
 668                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 669                                 if (mddev->sysfs_action)
 670                                         sysfs_put(mddev->sysfs_action);
 671                                 mddev->sysfs_action = NULL;
 672                         }
 673                 }
 674                 mddev->sysfs_active = 0;
 675         } else
 676                 mutex_unlock(&mddev->reconfig_mutex);
 677
 678         md_wakeup_thread(mddev->thread);
 679 }
 680
 681 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 682 {
 683         mdk_rdev_t *rdev;
 684
 685         list_for_each_entry(rdev, &mddev->disks, same_set)
 686                 if (rdev->desc_nr == nr)
 687                         return rdev;
 688
 689         return NULL;
 690 }
 691
 692 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 693 {
 694         mdk_rdev_t *rdev;
 695
 696         list_for_each_entry(rdev, &mddev->disks, same_set)
 697                 if (rdev->bdev->bd_dev == dev)
 698                         return rdev;
 699
 700         return NULL;
 701 }
 702
 703 static struct mdk_personality *find_pers(int level, char *clevel)
 704 {
 705         struct mdk_personality *pers;
 706         list_for_each_entry(pers, &pers_list, list) {
 707                 if (level != LEVEL_NONE && pers->level == level)
 708                         return pers;
 709                 if (strcmp(pers->name, clevel)==0)
 710                         return pers;
 711         }
 712         return NULL;
 713 }
 714
 715 /* return the offset of the super block in 512byte sectors */
 716 static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
 717 {
 718         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 719         return MD_NEW_SIZE_SECTORS(num_sectors);
 720 }
 721
 722 static int alloc_disk_sb(mdk_rdev_t * rdev)
 723 {
 724         if (rdev->sb_page)
 725                 MD_BUG();
 726
 727         rdev->sb_page = alloc_page(GFP_KERNEL);
 728         if (!rdev->sb_page) {
 729                 printk(KERN_ALERT "md: out of memory.\n");
 730                 return -ENOMEM;
 731         }
 732
 733         return 0;
 734 }
 735
 736 static void free_disk_sb(mdk_rdev_t * rdev)
 737 {
 738         if (rdev->sb_page) {
 739                 put_page(rdev->sb_page);
 740                 rdev->sb_loaded = 0;
 741                 rdev->sb_page = NULL;
 742                 rdev->sb_start = 0;
 743                 rdev->sectors = 0;
 744         }
 745 }
 746
 747
 748 static void super_written(struct bio *bio, int error)
 749 {
 750         mdk_rdev_t *rdev = bio->bi_private;
 751         mddev_t *mddev = rdev->mddev;
 752
 753         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 754                 printk("md: super_written gets error=%d, uptodate=%d\n",
 755                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 756                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 757                 md_error(mddev, rdev);
 758         }
 759
 760         if (atomic_dec_and_test(&mddev->pending_writes))
 761                 wake_up(&mddev->sb_wait);
 762         bio_put(bio);
 763 }
 764
 765 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 766                    sector_t sector, int size, struct page *page)
 767 {
 768         /* write first size bytes of page to sector of rdev
 769          * Increment mddev->pending_writes before returning
 770          * and decrement it on completion, waking up sb_wait
 771          * if zero is reached.
 772          * If an error occurred, call md_error
 773          */
 774         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 775
 776         bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
 777         bio->bi_sector = sector;
 778         bio_add_page(bio, page, size, 0);
 779         bio->bi_private = rdev;
 780         bio->bi_end_io = super_written;
 781
 782         atomic_inc(&mddev->pending_writes);
 783         submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
 784                    bio);
 785 }
 786
 787 void md_super_wait(mddev_t *mddev)
 788 {
 789         /* wait for all superblock writes that were scheduled to complete */
 790         DEFINE_WAIT(wq);
 791         for(;;) {
 792                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 793                 if (atomic_read(&mddev->pending_writes)==0)
 794                         break;
 795                 schedule();
 796         }
 797         finish_wait(&mddev->sb_wait, &wq);
 798 }
 799
 800 static void bi_complete(struct bio *bio, int error)
 801 {
 802         complete((struct completion*)bio->bi_private);
 803 }
 804
 805 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
 806                  struct page *page, int rw, bool metadata_op)
 807 {
 808         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 809         struct completion event;
 810         int ret;
 811
 812         rw |= REQ_SYNC | REQ_UNPLUG;
 813
 814         bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
 815                 rdev->meta_bdev : rdev->bdev;
 816         if (metadata_op)
 817                 bio->bi_sector = sector + rdev->sb_start;
 818         else
 819                 bio->bi_sector = sector + rdev->data_offset;
 820         bio_add_page(bio, page, size, 0);
 821         init_completion(&event);
 822         bio->bi_private = &event;
 823         bio->bi_end_io = bi_complete;
 824         submit_bio(rw, bio);
 825         wait_for_completion(&event);
 826
 827         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 828         bio_put(bio);
 829         return ret;
 830 }
 831 EXPORT_SYMBOL_GPL(sync_page_io);
 832
 833 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 834 {
 835         char b[BDEVNAME_SIZE];
 836         if (!rdev->sb_page) {
 837                 MD_BUG();
 838                 return -EINVAL;
 839         }
 840         if (rdev->sb_loaded)
 841                 return 0;
 842
 843
 844         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
 845                 goto fail;
 846         rdev->sb_loaded = 1;
 847         return 0;
 848
 849 fail:
 850         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 851                 bdevname(rdev->bdev,b));
 852         return -EINVAL;
 853 }
 854
 855 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 856 {
 857         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 858                 sb1->set_uuid1 == sb2->set_uuid1 &&
 859                 sb1->set_uuid2 == sb2->set_uuid2 &&
 860                 sb1->set_uuid3 == sb2->set_uuid3;
 861 }
 862
 863 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 864 {
 865         int ret;
 866         mdp_super_t *tmp1, *tmp2;
 867
 868         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 869         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 870
 871         if (!tmp1 || !tmp2) {
 872                 ret = 0;
 873                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 874                 goto abort;
 875         }
 876
 877         *tmp1 = *sb1;
 878         *tmp2 = *sb2;
 879
 880         /*
 881          * nr_disks is not constant
 882          */
 883         tmp1->nr_disks = 0;
 884         tmp2->nr_disks = 0;
 885
 886         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 887 abort:
 888         kfree(tmp1);
 889         kfree(tmp2);
 890         return ret;
 891 }
 892
 893
 894 static u32 md_csum_fold(u32 csum)
 895 {
 896         csum = (csum & 0xffff) + (csum >> 16);
 897         return (csum & 0xffff) + (csum >> 16);
 898 }
 899
 900 static unsigned int calc_sb_csum(mdp_super_t * sb)
 901 {
 902         u64 newcsum = 0;
 903         u32 *sb32 = (u32*)sb;
 904         int i;
 905         unsigned int disk_csum, csum;
 906
 907         disk_csum = sb->sb_csum;
 908         sb->sb_csum = 0;
 909
 910         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 911                 newcsum += sb32[i];
 912         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 913
 914
 915 #ifdef CONFIG_ALPHA
 916         /* This used to use csum_partial, which was wrong for several
 917          * reasons including that different results are returned on
 918          * different architectures.  It isn't critical that we get exactly
 919          * the same return value as before (we always csum_fold before
 920          * testing, and that removes any differences).  However as we
 921          * know that csum_partial always returned a 16bit value on
 922          * alphas, do a fold to maximise conformity to previous behaviour.
 923          */
 924         sb->sb_csum = md_csum_fold(disk_csum);
 925 #else
 926         sb->sb_csum = disk_csum;
 927 #endif
 928         return csum;
 929 }
 930
 931
 932 /*
 933  * Handle superblock details.
 934  * We want to be able to handle multiple superblock formats
 935  * so we have a common interface to them all, and an array of
 936  * different handlers.
 937  * We rely on user-space to write the initial superblock, and support
 938  * reading and updating of superblocks.
 939  * Interface methods are:
 940  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 941  *      loads and validates a superblock on dev.
 942  *      if refdev != NULL, compare superblocks on both devices
 943  *    Return:
 944  *      0 - dev has a superblock that is compatible with refdev
 945  *      1 - dev has a superblock that is compatible and newer than refdev
 946  *          so dev should be used as the refdev in future
 947  *     -EINVAL superblock incompatible or invalid
 948  *     -othererror e.g. -EIO
 949  *
 950  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 951  *      Verify that dev is acceptable into mddev.
 952  *       The first time, mddev->raid_disks will be 0, and data from
 953  *       dev should be merged in.  Subsequent calls check that dev
 954  *       is new enough.  Return 0 or -EINVAL
 955  *
 956  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 957  *     Update the superblock for rdev with data in mddev
 958  *     This does not write to disc.
 959  *
 960  */
 961
 962 struct super_type  {
 963         char                *name;
 964         struct module       *owner;
 965         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 966                                           int minor_version);
 967         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 968         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 969         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 970                                                 sector_t num_sectors);
 971 };
 972
 973 /*
 974  * Check that the given mddev has no bitmap.
 975  *
 976  * This function is called from the run method of all personalities that do not
 977  * support bitmaps. It prints an error message and returns non-zero if mddev
 978  * has a bitmap. Otherwise, it returns 0.
 979  *
 980  */
 981 int md_check_no_bitmap(mddev_t *mddev)
 982 {
 983         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
 984                 return 0;
 985         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
 986                 mdname(mddev), mddev->pers->name);
 987         return 1;
 988 }
 989 EXPORT_SYMBOL(md_check_no_bitmap);
 990
 991 /*
 992  * load_super for 0.90.0
 993  */
 994 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 995 {
 996         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 997         mdp_super_t *sb;
 998         int ret;
 999
1000         /*
1001          * Calculate the position of the superblock (512byte sectors),
1002          * it's at the end of the disk.
1003          *
1004          * It also happens to be a multiple of 4Kb.
1005          */
1006         rdev->sb_start = calc_dev_sboffset(rdev);
1007
1008         ret = read_disk_sb(rdev, MD_SB_BYTES);
1009         if (ret) return ret;
1010
1011         ret = -EINVAL;
1012
1013         bdevname(rdev->bdev, b);
1014         sb = (mdp_super_t*)page_address(rdev->sb_page);
1015
1016         if (sb->md_magic != MD_SB_MAGIC) {
1017                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1018                        b);
1019                 goto abort;
1020         }
1021
1022         if (sb->major_version != 0 ||
1023             sb->minor_version < 90 ||
1024             sb->minor_version > 91) {
1025                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1026                         sb->major_version, sb->minor_version,
1027                         b);
1028                 goto abort;
1029         }
1030
1031         if (sb->raid_disks <= 0)
1032                 goto abort;
1033
1034         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1035                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1036                         b);
1037                 goto abort;
1038         }
1039
1040         rdev->preferred_minor = sb->md_minor;
1041         rdev->data_offset = 0;
1042         rdev->sb_size = MD_SB_BYTES;
1043
1044         if (sb->level == LEVEL_MULTIPATH)
1045                 rdev->desc_nr = -1;
1046         else
1047                 rdev->desc_nr = sb->this_disk.number;
1048
1049         if (!refdev) {
1050                 ret = 1;
1051         } else {
1052                 __u64 ev1, ev2;
1053                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1054                 if (!uuid_equal(refsb, sb)) {
1055                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
1056                                 b, bdevname(refdev->bdev,b2));
1057                         goto abort;
1058                 }
1059                 if (!sb_equal(refsb, sb)) {
1060                         printk(KERN_WARNING "md: %s has same UUID"
1061                                " but different superblock to %s\n",
1062                                b, bdevname(refdev->bdev, b2));
1063                         goto abort;
1064                 }
1065                 ev1 = md_event(sb);
1066                 ev2 = md_event(refsb);
1067                 if (ev1 > ev2)
1068                         ret = 1;
1069                 else
1070                         ret = 0;
1071         }
1072         rdev->sectors = rdev->sb_start;
1073
1074         if (rdev->sectors < sb->size * 2 && sb->level > 1)
1075                 /* "this cannot possibly happen" ... */
1076                 ret = -EINVAL;
1077
1078  abort:
1079         return ret;
1080 }
1081
1082 /*
1083  * validate_super for 0.90.0
1084  */
1085 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1086 {
1087         mdp_disk_t *desc;
1088         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1089         __u64 ev1 = md_event(sb);
1090
1091         rdev->raid_disk = -1;
1092         clear_bit(Faulty, &rdev->flags);
1093         clear_bit(In_sync, &rdev->flags);
1094         clear_bit(WriteMostly, &rdev->flags);
1095
1096         if (mddev->raid_disks == 0) {
1097                 mddev->major_version = 0;
1098                 mddev->minor_version = sb->minor_version;
1099                 mddev->patch_version = sb->patch_version;
1100                 mddev->external = 0;
1101                 mddev->chunk_sectors = sb->chunk_size >> 9;
1102                 mddev->ctime = sb->ctime;
1103                 mddev->utime = sb->utime;
1104                 mddev->level = sb->level;
1105                 mddev->clevel[0] = 0;
1106                 mddev->layout = sb->layout;
1107                 mddev->raid_disks = sb->raid_disks;
1108                 mddev->dev_sectors = sb->size * 2;
1109                 mddev->events = ev1;
1110                 mddev->bitmap_info.offset = 0;
1111                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1112
1113                 if (mddev->minor_version >= 91) {
1114                         mddev->reshape_position = sb->reshape_position;
1115                         mddev->delta_disks = sb->delta_disks;
1116                         mddev->new_level = sb->new_level;
1117                         mddev->new_layout = sb->new_layout;
1118                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1119                 } else {
1120                         mddev->reshape_position = MaxSector;
1121                         mddev->delta_disks = 0;
1122                         mddev->new_level = mddev->level;
1123                         mddev->new_layout = mddev->layout;
1124                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1125                 }
1126
1127                 if (sb->state & (1<<MD_SB_CLEAN))
1128                         mddev->recovery_cp = MaxSector;
1129                 else {
1130                         if (sb->events_hi == sb->cp_events_hi &&
1131                                 sb->events_lo == sb->cp_events_lo) {
1132                                 mddev->recovery_cp = sb->recovery_cp;
1133                         } else
1134                                 mddev->recovery_cp = 0;
1135                 }
1136
1137                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1138                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1139                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1140                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1141
1142                 mddev->max_disks = MD_SB_DISKS;
1143
1144                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1145                     mddev->bitmap_info.file == NULL)
1146                         mddev->bitmap_info.offset =
1147                                 mddev->bitmap_info.default_offset;
1148
1149         } else if (mddev->pers == NULL) {
1150                 /* Insist on good event counter while assembling, except
1151                  * for spares (which don't need an event count) */
1152                 ++ev1;
1153                 if (sb->disks[rdev->desc_nr].state & (
1154                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1155                         if (ev1 < mddev->events)
1156                                 return -EINVAL;
1157         } else if (mddev->bitmap) {
1158                 /* if adding to array with a bitmap, then we can accept an
1159                  * older device ... but not too old.
1160                  */
1161                 if (ev1 < mddev->bitmap->events_cleared)
1162                         return 0;
1163         } else {
1164                 if (ev1 < mddev->events)
1165                         /* just a hot-add of a new device, leave raid_disk at -1 */
1166                         return 0;
1167         }
1168
1169         if (mddev->level != LEVEL_MULTIPATH) {
1170                 desc = sb->disks + rdev->desc_nr;
1171
1172                 if (desc->state & (1<<MD_DISK_FAULTY))
1173                         set_bit(Faulty, &rdev->flags);
1174                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1175                             desc->raid_disk < mddev->raid_disks */) {
1176                         set_bit(In_sync, &rdev->flags);
1177                         rdev->raid_disk = desc->raid_disk;
1178                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1179                         /* active but not in sync implies recovery up to
1180                          * reshape position.  We don't know exactly where
1181                          * that is, so set to zero for now */
1182                         if (mddev->minor_version >= 91) {
1183                                 rdev->recovery_offset = 0;
1184                                 rdev->raid_disk = desc->raid_disk;
1185                         }
1186                 }
1187                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1188                         set_bit(WriteMostly, &rdev->flags);
1189         } else /* MULTIPATH are always insync */
1190                 set_bit(In_sync, &rdev->flags);
1191         return 0;
1192 }
1193
1194 /*
1195  * sync_super for 0.90.0
1196  */
1197 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1198 {
1199         mdp_super_t *sb;
1200         mdk_rdev_t *rdev2;
1201         int next_spare = mddev->raid_disks;
1202
1203
1204         /* make rdev->sb match mddev data..
1205          *
1206          * 1/ zero out disks
1207          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1208          * 3/ any empty disks < next_spare become removed
1209          *
1210          * disks[0] gets initialised to REMOVED because
1211          * we cannot be sure from other fields if it has
1212          * been initialised or not.
1213          */
1214         int i;
1215         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1216
1217         rdev->sb_size = MD_SB_BYTES;
1218
1219         sb = (mdp_super_t*)page_address(rdev->sb_page);
1220
1221         memset(sb, 0, sizeof(*sb));
1222
1223         sb->md_magic = MD_SB_MAGIC;
1224         sb->major_version = mddev->major_version;
1225         sb->patch_version = mddev->patch_version;
1226         sb->gvalid_words  = 0; /* ignored */
1227         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1228         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1229         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1230         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1231
1232         sb->ctime = mddev->ctime;
1233         sb->level = mddev->level;
1234         sb->size = mddev->dev_sectors / 2;
1235         sb->raid_disks = mddev->raid_disks;
1236         sb->md_minor = mddev->md_minor;
1237         sb->not_persistent = 0;
1238         sb->utime = mddev->utime;
1239         sb->state = 0;
1240         sb->events_hi = (mddev->events>>32);
1241         sb->events_lo = (u32)mddev->events;
1242
1243         if (mddev->reshape_position == MaxSector)
1244                 sb->minor_version = 90;
1245         else {
1246                 sb->minor_version = 91;
1247                 sb->reshape_position = mddev->reshape_position;
1248                 sb->new_level = mddev->new_level;
1249                 sb->delta_disks = mddev->delta_disks;
1250                 sb->new_layout = mddev->new_layout;
1251                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1252         }
1253         mddev->minor_version = sb->minor_version;
1254         if (mddev->in_sync)
1255         {
1256                 sb->recovery_cp = mddev->recovery_cp;
1257                 sb->cp_events_hi = (mddev->events>>32);
1258                 sb->cp_events_lo = (u32)mddev->events;
1259                 if (mddev->recovery_cp == MaxSector)
1260                         sb->state = (1<< MD_SB_CLEAN);
1261         } else
1262                 sb->recovery_cp = 0;
1263
1264         sb->layout = mddev->layout;
1265         sb->chunk_size = mddev->chunk_sectors << 9;
1266
1267         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1268                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1269
1270         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1271         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1272                 mdp_disk_t *d;
1273                 int desc_nr;
1274                 int is_active = test_bit(In_sync, &rdev2->flags);
1275
1276                 if (rdev2->raid_disk >= 0 &&
1277                     sb->minor_version >= 91)
1278                         /* we have nowhere to store the recovery_offset,
1279                          * but if it is not below the reshape_position,
1280                          * we can piggy-back on that.
1281                          */
1282                         is_active = 1;
1283                 if (rdev2->raid_disk < 0 ||
1284                     test_bit(Faulty, &rdev2->flags))
1285                         is_active = 0;
1286                 if (is_active)
1287                         desc_nr = rdev2->raid_disk;
1288                 else
1289                         desc_nr = next_spare++;
1290                 rdev2->desc_nr = desc_nr;
1291                 d = &sb->disks[rdev2->desc_nr];
1292                 nr_disks++;
1293                 d->number = rdev2->desc_nr;
1294                 d->major = MAJOR(rdev2->bdev->bd_dev);
1295                 d->minor = MINOR(rdev2->bdev->bd_dev);
1296                 if (is_active)
1297                         d->raid_disk = rdev2->raid_disk;
1298                 else
1299                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1300                 if (test_bit(Faulty, &rdev2->flags))
1301                         d->state = (1<<MD_DISK_FAULTY);
1302                 else if (is_active) {
1303                         d->state = (1<<MD_DISK_ACTIVE);
1304                         if (test_bit(In_sync, &rdev2->flags))
1305                                 d->state |= (1<<MD_DISK_SYNC);
1306                         active++;
1307                         working++;
1308                 } else {
1309                         d->state = 0;
1310                         spare++;
1311                         working++;
1312                 }
1313                 if (test_bit(WriteMostly, &rdev2->flags))
1314                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1315         }
1316         /* now set the "removed" and "faulty" bits on any missing devices */
1317         for (i=0 ; i < mddev->raid_disks ; i++) {
1318                 mdp_disk_t *d = &sb->disks[i];
1319                 if (d->state == 0 && d->number == 0) {
1320                         d->number = i;
1321                         d->raid_disk = i;
1322                         d->state = (1<<MD_DISK_REMOVED);
1323                         d->state |= (1<<MD_DISK_FAULTY);
1324                         failed++;
1325                 }
1326         }
1327         sb->nr_disks = nr_disks;
1328         sb->active_disks = active;
1329         sb->working_disks = working;
1330         sb->failed_disks = failed;
1331         sb->spare_disks = spare;
1332
1333         sb->this_disk = sb->disks[rdev->desc_nr];
1334         sb->sb_csum = calc_sb_csum(sb);
1335 }
1336
1337 /*
1338  * rdev_size_change for 0.90.0
1339  */
1340 static unsigned long long
1341 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1342 {
1343         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1344                 return 0; /* component must fit device */
1345         if (rdev->mddev->bitmap_info.offset)
1346                 return 0; /* can't move bitmap */
1347         rdev->sb_start = calc_dev_sboffset(rdev);
1348         if (!num_sectors || num_sectors > rdev->sb_start)
1349                 num_sectors = rdev->sb_start;
1350         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1351                        rdev->sb_page);
1352         md_super_wait(rdev->mddev);
1353         return num_sectors;
1354 }
1355
1356
1357 /*
1358  * version 1 superblock
1359  */
1360
1361 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1362 {
1363         __le32 disk_csum;
1364         u32 csum;
1365         unsigned long long newcsum;
1366         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1367         __le32 *isuper = (__le32*)sb;
1368         int i;
1369
1370         disk_csum = sb->sb_csum;
1371         sb->sb_csum = 0;
1372         newcsum = 0;
1373         for (i=0; size>=4; size -= 4 )
1374                 newcsum += le32_to_cpu(*isuper++);
1375
1376         if (size == 2)
1377                 newcsum += le16_to_cpu(*(__le16*) isuper);
1378
1379         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1380         sb->sb_csum = disk_csum;
1381         return cpu_to_le32(csum);
1382 }
1383
1384 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1385 {
1386         struct mdp_superblock_1 *sb;
1387         int ret;
1388         sector_t sb_start;
1389         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1390         int bmask;
1391
1392         /*
1393          * Calculate the position of the superblock in 512byte sectors.
1394          * It is always aligned to a 4K boundary and
1395          * depeding on minor_version, it can be:
1396          * 0: At least 8K, but less than 12K, from end of device
1397          * 1: At start of device
1398          * 2: 4K from start of device.
1399          */
1400         switch(minor_version) {
1401         case 0:
1402                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1403                 sb_start -= 8*2;
1404                 sb_start &= ~(sector_t)(4*2-1);
1405                 break;
1406         case 1:
1407                 sb_start = 0;
1408                 break;
1409         case 2:
1410                 sb_start = 8;
1411                 break;
1412         default:
1413                 return -EINVAL;
1414         }
1415         rdev->sb_start = sb_start;
1416
1417         /* superblock is rarely larger than 1K, but it can be larger,
1418          * and it is safe to read 4k, so we do that
1419          */
1420         ret = read_disk_sb(rdev, 4096);
1421         if (ret) return ret;
1422
1423
1424         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1425
1426         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1427             sb->major_version != cpu_to_le32(1) ||
1428             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1429             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1430             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1431                 return -EINVAL;
1432
1433         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1434                 printk("md: invalid superblock checksum on %s\n",
1435                         bdevname(rdev->bdev,b));
1436                 return -EINVAL;
1437         }
1438         if (le64_to_cpu(sb->data_size) < 10) {
1439                 printk("md: data_size too small on %s\n",
1440                        bdevname(rdev->bdev,b));
1441                 return -EINVAL;
1442         }
1443
1444         rdev->preferred_minor = 0xffff;
1445         rdev->data_offset = le64_to_cpu(sb->data_offset);
1446         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1447
1448         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1449         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1450         if (rdev->sb_size & bmask)
1451                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1452
1453         if (minor_version
1454             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1455                 return -EINVAL;
1456
1457         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1458                 rdev->desc_nr = -1;
1459         else
1460                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1461
1462         if (!refdev) {
1463                 ret = 1;
1464         } else {
1465                 __u64 ev1, ev2;
1466                 struct mdp_superblock_1 *refsb =
1467                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1468
1469                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1470                     sb->level != refsb->level ||
1471                     sb->layout != refsb->layout ||
1472                     sb->chunksize != refsb->chunksize) {
1473                         printk(KERN_WARNING "md: %s has strangely different"
1474                                 " superblock to %s\n",
1475                                 bdevname(rdev->bdev,b),
1476                                 bdevname(refdev->bdev,b2));
1477                         return -EINVAL;
1478                 }
1479                 ev1 = le64_to_cpu(sb->events);
1480                 ev2 = le64_to_cpu(refsb->events);
1481
1482                 if (ev1 > ev2)
1483                         ret = 1;
1484                 else
1485                         ret = 0;
1486         }
1487         if (minor_version)
1488                 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1489                         le64_to_cpu(sb->data_offset);
1490         else
1491                 rdev->sectors = rdev->sb_start;
1492         if (rdev->sectors < le64_to_cpu(sb->data_size))
1493                 return -EINVAL;
1494         rdev->sectors = le64_to_cpu(sb->data_size);
1495         if (le64_to_cpu(sb->size) > rdev->sectors)
1496                 return -EINVAL;
1497         return ret;
1498 }
1499
1500 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1501 {
1502         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1503         __u64 ev1 = le64_to_cpu(sb->events);
1504
1505         rdev->raid_disk = -1;
1506         clear_bit(Faulty, &rdev->flags);
1507         clear_bit(In_sync, &rdev->flags);
1508         clear_bit(WriteMostly, &rdev->flags);
1509
1510         if (mddev->raid_disks == 0) {
1511                 mddev->major_version = 1;
1512                 mddev->patch_version = 0;
1513                 mddev->external = 0;
1514                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1515                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1516                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1517                 mddev->level = le32_to_cpu(sb->level);
1518                 mddev->clevel[0] = 0;
1519                 mddev->layout = le32_to_cpu(sb->layout);
1520                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1521                 mddev->dev_sectors = le64_to_cpu(sb->size);
1522                 mddev->events = ev1;
1523                 mddev->bitmap_info.offset = 0;
1524                 mddev->bitmap_info.default_offset = 1024 >> 9;
1525
1526                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1527                 memcpy(mddev->uuid, sb->set_uuid, 16);
1528
1529                 mddev->max_disks =  (4096-256)/2;
1530
1531                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1532                     mddev->bitmap_info.file == NULL )
1533                         mddev->bitmap_info.offset =
1534                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1535
1536                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1537                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1538                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1539                         mddev->new_level = le32_to_cpu(sb->new_level);
1540                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1541                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1542                 } else {
1543                         mddev->reshape_position = MaxSector;
1544                         mddev->delta_disks = 0;
1545                         mddev->new_level = mddev->level;
1546                         mddev->new_layout = mddev->layout;
1547                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1548                 }
1549
1550         } else if (mddev->pers == NULL) {
1551                 /* Insist of good event counter while assembling, except for
1552                  * spares (which don't need an event count) */
1553                 ++ev1;
1554                 if (rdev->desc_nr >= 0 &&
1555                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1556                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1557                         if (ev1 < mddev->events)
1558                                 return -EINVAL;
1559         } else if (mddev->bitmap) {
1560                 /* If adding to array with a bitmap, then we can accept an
1561                  * older device, but not too old.
1562                  */
1563                 if (ev1 < mddev->bitmap->events_cleared)
1564                         return 0;
1565         } else {
1566                 if (ev1 < mddev->events)
1567                         /* just a hot-add of a new device, leave raid_disk at -1 */
1568                         return 0;
1569         }
1570         if (mddev->level != LEVEL_MULTIPATH) {
1571                 int role;
1572                 if (rdev->desc_nr < 0 ||
1573                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1574                         role = 0xffff;
1575                         rdev->desc_nr = -1;
1576                 } else
1577                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1578                 switch(role) {
1579                 case 0xffff: /* spare */
1580                         break;
1581                 case 0xfffe: /* faulty */
1582                         set_bit(Faulty, &rdev->flags);
1583                         break;
1584                 default:
1585                         if ((le32_to_cpu(sb->feature_map) &
1586                              MD_FEATURE_RECOVERY_OFFSET))
1587                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1588                         else
1589                                 set_bit(In_sync, &rdev->flags);
1590                         rdev->raid_disk = role;
1591                         break;
1592                 }
1593                 if (sb->devflags & WriteMostly1)
1594                         set_bit(WriteMostly, &rdev->flags);
1595         } else /* MULTIPATH are always insync */
1596                 set_bit(In_sync, &rdev->flags);
1597
1598         return 0;
1599 }
1600
1601 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1602 {
1603         struct mdp_superblock_1 *sb;
1604         mdk_rdev_t *rdev2;
1605         int max_dev, i;
1606         /* make rdev->sb match mddev and rdev data. */
1607
1608         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1609
1610         sb->feature_map = 0;
1611         sb->pad0 = 0;
1612         sb->recovery_offset = cpu_to_le64(0);
1613         memset(sb->pad1, 0, sizeof(sb->pad1));
1614         memset(sb->pad2, 0, sizeof(sb->pad2));
1615         memset(sb->pad3, 0, sizeof(sb->pad3));
1616
1617         sb->utime = cpu_to_le64((__u64)mddev->utime);
1618         sb->events = cpu_to_le64(mddev->events);
1619         if (mddev->in_sync)
1620                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1621         else
1622                 sb->resync_offset = cpu_to_le64(0);
1623
1624         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1625
1626         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1627         sb->size = cpu_to_le64(mddev->dev_sectors);
1628         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1629         sb->level = cpu_to_le32(mddev->level);
1630         sb->layout = cpu_to_le32(mddev->layout);
1631
1632         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1633                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1634                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1635         }
1636
1637         if (rdev->raid_disk >= 0 &&
1638             !test_bit(In_sync, &rdev->flags)) {
1639                 sb->feature_map |=
1640                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1641                 sb->recovery_offset =
1642                         cpu_to_le64(rdev->recovery_offset);
1643         }
1644
1645         if (mddev->reshape_position != MaxSector) {
1646                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1647                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1648                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1649                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1650                 sb->new_level = cpu_to_le32(mddev->new_level);
1651                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1652         }
1653
1654         max_dev = 0;
1655         list_for_each_entry(rdev2, &mddev->disks, same_set)
1656                 if (rdev2->desc_nr+1 > max_dev)
1657                         max_dev = rdev2->desc_nr+1;
1658
1659         if (max_dev > le32_to_cpu(sb->max_dev)) {
1660                 int bmask;
1661                 sb->max_dev = cpu_to_le32(max_dev);
1662                 rdev->sb_size = max_dev * 2 + 256;
1663                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1664                 if (rdev->sb_size & bmask)
1665                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1666         } else
1667                 max_dev = le32_to_cpu(sb->max_dev);
1668
1669         for (i=0; i<max_dev;i++)
1670                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1671
1672         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1673                 i = rdev2->desc_nr;
1674                 if (test_bit(Faulty, &rdev2->flags))
1675                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1676                 else if (test_bit(In_sync, &rdev2->flags))
1677                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1678                 else if (rdev2->raid_disk >= 0)
1679                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1680                 else
1681                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1682         }
1683
1684         sb->sb_csum = calc_sb_1_csum(sb);
1685 }
1686
1687 static unsigned long long
1688 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1689 {
1690         struct mdp_superblock_1 *sb;
1691         sector_t max_sectors;
1692         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1693                 return 0; /* component must fit device */
1694         if (rdev->sb_start < rdev->data_offset) {
1695                 /* minor versions 1 and 2; superblock before data */
1696                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1697                 max_sectors -= rdev->data_offset;
1698                 if (!num_sectors || num_sectors > max_sectors)
1699                         num_sectors = max_sectors;
1700         } else if (rdev->mddev->bitmap_info.offset) {
1701                 /* minor version 0 with bitmap we can't move */
1702                 return 0;
1703         } else {
1704                 /* minor version 0; superblock after data */
1705                 sector_t sb_start;
1706                 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1707                 sb_start &= ~(sector_t)(4*2 - 1);
1708                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1709                 if (!num_sectors || num_sectors > max_sectors)
1710                         num_sectors = max_sectors;
1711                 rdev->sb_start = sb_start;
1712         }
1713         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1714         sb->data_size = cpu_to_le64(num_sectors);
1715         sb->super_offset = rdev->sb_start;
1716         sb->sb_csum = calc_sb_1_csum(sb);
1717         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1718                        rdev->sb_page);
1719         md_super_wait(rdev->mddev);
1720         return num_sectors;
1721 }
1722
1723 static struct super_type super_types[] = {
1724         [0] = {
1725                 .name   = "0.90.0",
1726                 .owner  = THIS_MODULE,
1727                 .load_super         = super_90_load,
1728                 .validate_super     = super_90_validate,
1729                 .sync_super         = super_90_sync,
1730                 .rdev_size_change   = super_90_rdev_size_change,
1731         },
1732         [1] = {
1733                 .name   = "md-1",
1734                 .owner  = THIS_MODULE,
1735                 .load_super         = super_1_load,
1736                 .validate_super     = super_1_validate,
1737                 .sync_super         = super_1_sync,
1738                 .rdev_size_change   = super_1_rdev_size_change,
1739         },
1740 };
1741
1742 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1743 {
1744         mdk_rdev_t *rdev, *rdev2;
1745
1746         rcu_read_lock();
1747         rdev_for_each_rcu(rdev, mddev1)
1748                 rdev_for_each_rcu(rdev2, mddev2)
1749                         if (rdev->bdev->bd_contains ==
1750                             rdev2->bdev->bd_contains) {
1751                                 rcu_read_unlock();
1752                                 return 1;
1753                         }
1754         rcu_read_unlock();
1755         return 0;
1756 }
1757
1758 static LIST_HEAD(pending_raid_disks);
1759
1760 /*
1761  * Try to register data integrity profile for an mddev
1762  *
1763  * This is called when an array is started and after a disk has been kicked
1764  * from the array. It only succeeds if all working and active component devices
1765  * are integrity capable with matching profiles.
1766  */
1767 int md_integrity_register(mddev_t *mddev)
1768 {
1769         mdk_rdev_t *rdev, *reference = NULL;
1770
1771         if (list_empty(&mddev->disks))
1772                 return 0; /* nothing to do */
1773         if (blk_get_integrity(mddev->gendisk))
1774                 return 0; /* already registered */
1775         list_for_each_entry(rdev, &mddev->disks, same_set) {
1776                 /* skip spares and non-functional disks */
1777                 if (test_bit(Faulty, &rdev->flags))
1778                         continue;
1779                 if (rdev->raid_disk < 0)
1780                         continue;
1781                 /*
1782                  * If at least one rdev is not integrity capable, we can not
1783                  * enable data integrity for the md device.
1784                  */
1785                 if (!bdev_get_integrity(rdev->bdev))
1786                         return -EINVAL;
1787                 if (!reference) {
1788                         /* Use the first rdev as the reference */
1789                         reference = rdev;
1790                         continue;
1791                 }
1792                 /* does this rdev's profile match the reference profile? */
1793                 if (blk_integrity_compare(reference->bdev->bd_disk,
1794                                 rdev->bdev->bd_disk) < 0)
1795                         return -EINVAL;
1796         }
1797         /*
1798          * All component devices are integrity capable and have matching
1799          * profiles, register the common profile for the md device.
1800          */
1801         if (blk_integrity_register(mddev->gendisk,
1802                         bdev_get_integrity(reference->bdev)) != 0) {
1803                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1804                         mdname(mddev));
1805                 return -EINVAL;
1806         }
1807         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1808                 mdname(mddev));
1809         return 0;
1810 }
1811 EXPORT_SYMBOL(md_integrity_register);
1812
1813 /* Disable data integrity if non-capable/non-matching disk is being added */
1814 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1815 {
1816         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1817         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1818
1819         if (!bi_mddev) /* nothing to do */
1820                 return;
1821         if (rdev->raid_disk < 0) /* skip spares */
1822                 return;
1823         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1824                                              rdev->bdev->bd_disk) >= 0)
1825                 return;
1826         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1827         blk_integrity_unregister(mddev->gendisk);
1828 }
1829 EXPORT_SYMBOL(md_integrity_add_rdev);
1830
1831 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1832 {
1833         char b[BDEVNAME_SIZE];
1834         struct kobject *ko;
1835         char *s;
1836         int err;
1837
1838         if (rdev->mddev) {
1839                 MD_BUG();
1840                 return -EINVAL;
1841         }
1842
1843         /* prevent duplicates */
1844         if (find_rdev(mddev, rdev->bdev->bd_dev))
1845                 return -EEXIST;
1846
1847         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1848         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1849                         rdev->sectors < mddev->dev_sectors)) {
1850                 if (mddev->pers) {
1851                         /* Cannot change size, so fail
1852                          * If mddev->level <= 0, then we don't care
1853                          * about aligning sizes (e.g. linear)
1854                          */
1855                         if (mddev->level > 0)
1856                                 return -ENOSPC;
1857                 } else
1858                         mddev->dev_sectors = rdev->sectors;
1859         }
1860
1861         /* Verify rdev->desc_nr is unique.
1862          * If it is -1, assign a free number, else
1863          * check number is not in use
1864          */
1865         if (rdev->desc_nr < 0) {
1866                 int choice = 0;
1867                 if (mddev->pers) choice = mddev->raid_disks;
1868                 while (find_rdev_nr(mddev, choice))
1869                         choice++;
1870                 rdev->desc_nr = choice;
1871         } else {
1872                 if (find_rdev_nr(mddev, rdev->desc_nr))
1873                         return -EBUSY;
1874         }
1875         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1876                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1877                        mdname(mddev), mddev->max_disks);
1878                 return -EBUSY;
1879         }
1880         bdevname(rdev->bdev,b);
1881         while ( (s=strchr(b, '/')) != NULL)
1882                 *s = '!';
1883
1884         rdev->mddev = mddev;
1885         printk(KERN_INFO "md: bind<%s>\n", b);
1886
1887         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1888                 goto fail;
1889
1890         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1891         if (sysfs_create_link(&rdev->kobj, ko, "block"))
1892                 /* failure here is OK */;
1893         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1894
1895         list_add_rcu(&rdev->same_set, &mddev->disks);
1896         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
1897
1898         /* May as well allow recovery to be retried once */
1899         mddev->recovery_disabled = 0;
1900
1901         return 0;
1902
1903  fail:
1904         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1905                b, mdname(mddev));
1906         return err;
1907 }
1908
1909 static void md_delayed_delete(struct work_struct *ws)
1910 {
1911         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1912         kobject_del(&rdev->kobj);
1913         kobject_put(&rdev->kobj);
1914 }
1915
1916 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1917 {
1918         char b[BDEVNAME_SIZE];
1919         if (!rdev->mddev) {
1920                 MD_BUG();
1921                 return;
1922         }
1923         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
1924         list_del_rcu(&rdev->same_set);
1925         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1926         rdev->mddev = NULL;
1927         sysfs_remove_link(&rdev->kobj, "block");
1928         sysfs_put(rdev->sysfs_state);
1929         rdev->sysfs_state = NULL;
1930         /* We need to delay this, otherwise we can deadlock when
1931          * writing to 'remove' to "dev/state".  We also need
1932          * to delay it due to rcu usage.
1933          */
1934         synchronize_rcu();
1935         INIT_WORK(&rdev->del_work, md_delayed_delete);
1936         kobject_get(&rdev->kobj);
1937         queue_work(md_misc_wq, &rdev->del_work);
1938 }
1939
1940 /*
1941  * prevent the device from being mounted, repartitioned or
1942  * otherwise reused by a RAID array (or any other kernel
1943  * subsystem), by bd_claiming the device.
1944  */
1945 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1946 {
1947         int err = 0;
1948         struct block_device *bdev;
1949         char b[BDEVNAME_SIZE];
1950
1951         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1952                                  shared ? (mdk_rdev_t *)lock_rdev : rdev);
1953         if (IS_ERR(bdev)) {
1954                 printk(KERN_ERR "md: could not open %s.\n",
1955                         __bdevname(dev, b));
1956                 return PTR_ERR(bdev);
1957         }
1958         rdev->bdev = bdev;
1959         return err;
1960 }
1961
1962 static void unlock_rdev(mdk_rdev_t *rdev)
1963 {
1964         struct block_device *bdev = rdev->bdev;
1965         rdev->bdev = NULL;
1966         if (!bdev)
1967                 MD_BUG();
1968         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1969 }
1970
1971 void md_autodetect_dev(dev_t dev);
1972
1973 static void export_rdev(mdk_rdev_t * rdev)
1974 {
1975         char b[BDEVNAME_SIZE];
1976         printk(KERN_INFO "md: export_rdev(%s)\n",
1977                 bdevname(rdev->bdev,b));
1978         if (rdev->mddev)
1979                 MD_BUG();
1980         free_disk_sb(rdev);
1981 #ifndef MODULE
1982         if (test_bit(AutoDetected, &rdev->flags))
1983                 md_autodetect_dev(rdev->bdev->bd_dev);
1984 #endif
1985         unlock_rdev(rdev);
1986         kobject_put(&rdev->kobj);
1987 }
1988
1989 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1990 {
1991         unbind_rdev_from_array(rdev);
1992         export_rdev(rdev);
1993 }
1994
1995 static void export_array(mddev_t *mddev)
1996 {
1997         mdk_rdev_t *rdev, *tmp;
1998
1999         rdev_for_each(rdev, tmp, mddev) {
2000                 if (!rdev->mddev) {
2001                         MD_BUG();
2002                         continue;
2003                 }
2004                 kick_rdev_from_array(rdev);
2005         }
2006         if (!list_empty(&mddev->disks))
2007                 MD_BUG();
2008         mddev->raid_disks = 0;
2009         mddev->major_version = 0;
2010 }
2011
2012 static void print_desc(mdp_disk_t *desc)
2013 {
2014         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2015                 desc->major,desc->minor,desc->raid_disk,desc->state);
2016 }
2017
2018 static void print_sb_90(mdp_super_t *sb)
2019 {
2020         int i;
2021
2022         printk(KERN_INFO
2023                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2024                 sb->major_version, sb->minor_version, sb->patch_version,
2025                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2026                 sb->ctime);
2027         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2028                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2029                 sb->md_minor, sb->layout, sb->chunk_size);
2030         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2031                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2032                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2033                 sb->failed_disks, sb->spare_disks,
2034                 sb->sb_csum, (unsigned long)sb->events_lo);
2035
2036         printk(KERN_INFO);
2037         for (i = 0; i < MD_SB_DISKS; i++) {
2038                 mdp_disk_t *desc;
2039
2040                 desc = sb->disks + i;
2041                 if (desc->number || desc->major || desc->minor ||
2042                     desc->raid_disk || (desc->state && (desc->state != 4))) {
2043                         printk("     D %2d: ", i);
2044                         print_desc(desc);
2045                 }
2046         }
2047         printk(KERN_INFO "md:     THIS: ");
2048         print_desc(&sb->this_disk);
2049 }
2050
2051 static void print_sb_1(struct mdp_superblock_1 *sb)
2052 {
2053         __u8 *uuid;
2054
2055         uuid = sb->set_uuid;
2056         printk(KERN_INFO
2057                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2058                "md:    Name: \"%s\" CT:%llu\n",
2059                 le32_to_cpu(sb->major_version),
2060                 le32_to_cpu(sb->feature_map),
2061                 uuid,
2062                 sb->set_name,
2063                 (unsigned long long)le64_to_cpu(sb->ctime)
2064                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2065
2066         uuid = sb->device_uuid;
2067         printk(KERN_INFO
2068                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2069                         " RO:%llu\n"
2070                "md:     Dev:%08x UUID: %pU\n"
2071                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2072                "md:         (MaxDev:%u) \n",
2073                 le32_to_cpu(sb->level),
2074                 (unsigned long long)le64_to_cpu(sb->size),
2075                 le32_to_cpu(sb->raid_disks),
2076                 le32_to_cpu(sb->layout),
2077                 le32_to_cpu(sb->chunksize),
2078                 (unsigned long long)le64_to_cpu(sb->data_offset),
2079                 (unsigned long long)le64_to_cpu(sb->data_size),
2080                 (unsigned long long)le64_to_cpu(sb->super_offset),
2081                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2082                 le32_to_cpu(sb->dev_number),
2083                 uuid,
2084                 sb->devflags,
2085                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2086                 (unsigned long long)le64_to_cpu(sb->events),
2087                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2088                 le32_to_cpu(sb->sb_csum),
2089                 le32_to_cpu(sb->max_dev)
2090                 );
2091 }
2092
2093 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2094 {
2095         char b[BDEVNAME_SIZE];
2096         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2097                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2098                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2099                 rdev->desc_nr);
2100         if (rdev->sb_loaded) {
2101                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2102                 switch (major_version) {
2103                 case 0:
2104                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2105                         break;
2106                 case 1:
2107                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2108                         break;
2109                 }
2110         } else
2111                 printk(KERN_INFO "md: no rdev superblock!\n");
2112 }
2113
2114 static void md_print_devices(void)
2115 {
2116         struct list_head *tmp;
2117         mdk_rdev_t *rdev;
2118         mddev_t *mddev;
2119         char b[BDEVNAME_SIZE];
2120
2121         printk("\n");
2122         printk("md:     **********************************\n");
2123         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2124         printk("md:     **********************************\n");
2125         for_each_mddev(mddev, tmp) {
2126
2127                 if (mddev->bitmap)
2128                         bitmap_print_sb(mddev->bitmap);
2129                 else
2130                         printk("%s: ", mdname(mddev));
2131                 list_for_each_entry(rdev, &mddev->disks, same_set)
2132                         printk("<%s>", bdevname(rdev->bdev,b));
2133                 printk("\n");
2134
2135                 list_for_each_entry(rdev, &mddev->disks, same_set)
2136                         print_rdev(rdev, mddev->major_version);
2137         }
2138         printk("md:     **********************************\n");
2139         printk("\n");
2140 }
2141
2142
2143 static void sync_sbs(mddev_t * mddev, int nospares)
2144 {
2145         /* Update each superblock (in-memory image), but
2146          * if we are allowed to, skip spares which already
2147          * have the right event counter, or have one earlier
2148          * (which would mean they aren't being marked as dirty
2149          * with the rest of the array)
2150          */
2151         mdk_rdev_t *rdev;
2152         list_for_each_entry(rdev, &mddev->disks, same_set) {
2153                 if (rdev->sb_events == mddev->events ||
2154                     (nospares &&
2155                      rdev->raid_disk < 0 &&
2156                      rdev->sb_events+1 == mddev->events)) {
2157                         /* Don't update this superblock */
2158                         rdev->sb_loaded = 2;
2159                 } else {
2160                         super_types[mddev->major_version].
2161                                 sync_super(mddev, rdev);
2162                         rdev->sb_loaded = 1;
2163                 }
2164         }
2165 }
2166
2167 static void md_update_sb(mddev_t * mddev, int force_change)
2168 {
2169         mdk_rdev_t *rdev;
2170         int sync_req;
2171         int nospares = 0;
2172
2173 repeat:
2174         /* First make sure individual recovery_offsets are correct */
2175         list_for_each_entry(rdev, &mddev->disks, same_set) {
2176                 if (rdev->raid_disk >= 0 &&
2177                     mddev->delta_disks >= 0 &&
2178                     !test_bit(In_sync, &rdev->flags) &&
2179                     mddev->curr_resync_completed > rdev->recovery_offset)
2180                                 rdev->recovery_offset = mddev->curr_resync_completed;
2181
2182         }
2183         if (!mddev->persistent) {
2184                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2185                 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2186                 if (!mddev->external)
2187                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2188                 wake_up(&mddev->sb_wait);
2189                 return;
2190         }
2191
2192         spin_lock_irq(&mddev->write_lock);
2193
2194         mddev->utime = get_seconds();
2195
2196         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2197                 force_change = 1;
2198         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2199                 /* just a clean<-> dirty transition, possibly leave spares alone,
2200                  * though if events isn't the right even/odd, we will have to do
2201                  * spares after all
2202                  */
2203                 nospares = 1;
2204         if (force_change)
2205                 nospares = 0;
2206         if (mddev->degraded)
2207                 /* If the array is degraded, then skipping spares is both
2208                  * dangerous and fairly pointless.
2209                  * Dangerous because a device that was removed from the array
2210                  * might have a event_count that still looks up-to-date,
2211                  * so it can be re-added without a resync.
2212                  * Pointless because if there are any spares to skip,
2213                  * then a recovery will happen and soon that array won't
2214                  * be degraded any more and the spare can go back to sleep then.
2215                  */
2216                 nospares = 0;
2217
2218         sync_req = mddev->in_sync;
2219
2220         /* If this is just a dirty<->clean transition, and the array is clean
2221          * and 'events' is odd, we can roll back to the previous clean state */
2222         if (nospares
2223             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2224             && mddev->can_decrease_events
2225             && mddev->events != 1) {
2226                 mddev->events--;
2227                 mddev->can_decrease_events = 0;
2228         } else {
2229                 /* otherwise we have to go forward and ... */
2230                 mddev->events ++;
2231                 mddev->can_decrease_events = nospares;
2232         }
2233
2234         if (!mddev->events) {
2235                 /*
2236                  * oops, this 64-bit counter should never wrap.
2237                  * Either we are in around ~1 trillion A.C., assuming
2238                  * 1 reboot per second, or we have a bug:
2239                  */
2240                 MD_BUG();
2241                 mddev->events --;
2242         }
2243         sync_sbs(mddev, nospares);
2244         spin_unlock_irq(&mddev->write_lock);
2245
2246         dprintk(KERN_INFO
2247                 "md: updating %s RAID superblock on device (in sync %d)\n",
2248                 mdname(mddev),mddev->in_sync);
2249
2250         bitmap_update_sb(mddev->bitmap);
2251         list_for_each_entry(rdev, &mddev->disks, same_set) {
2252                 char b[BDEVNAME_SIZE];
2253                 dprintk(KERN_INFO "md: ");
2254                 if (rdev->sb_loaded != 1)
2255                         continue; /* no noise on spare devices */
2256                 if (test_bit(Faulty, &rdev->flags))
2257                         dprintk("(skipping faulty ");
2258
2259                 dprintk("%s ", bdevname(rdev->bdev,b));
2260                 if (!test_bit(Faulty, &rdev->flags)) {
2261                         md_super_write(mddev,rdev,
2262                                        rdev->sb_start, rdev->sb_size,
2263                                        rdev->sb_page);
2264                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2265                                 bdevname(rdev->bdev,b),
2266                                 (unsigned long long)rdev->sb_start);
2267                         rdev->sb_events = mddev->events;
2268
2269                 } else
2270                         dprintk(")\n");
2271                 if (mddev->level == LEVEL_MULTIPATH)
2272                         /* only need to write one superblock... */
2273                         break;
2274         }
2275         md_super_wait(mddev);
2276         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2277
2278         spin_lock_irq(&mddev->write_lock);
2279         if (mddev->in_sync != sync_req ||
2280             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2281                 /* have to write it out again */
2282                 spin_unlock_irq(&mddev->write_lock);
2283                 goto repeat;
2284         }
2285         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2286         spin_unlock_irq(&mddev->write_lock);
2287         wake_up(&mddev->sb_wait);
2288         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2289                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2290
2291 }
2292
2293 /* words written to sysfs files may, or may not, be \n terminated.
2294  * We want to accept with case. For this we use cmd_match.
2295  */
2296 static int cmd_match(const char *cmd, const char *str)
2297 {
2298         /* See if cmd, written into a sysfs file, matches
2299          * str.  They must either be the same, or cmd can
2300          * have a trailing newline
2301          */
2302         while (*cmd && *str && *cmd == *str) {
2303                 cmd++;
2304                 str++;
2305         }
2306         if (*cmd == '\n')
2307                 cmd++;
2308         if (*str || *cmd)
2309                 return 0;
2310         return 1;
2311 }
2312
2313 struct rdev_sysfs_entry {
2314         struct attribute attr;
2315         ssize_t (*show)(mdk_rdev_t *, char *);
2316         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2317 };
2318
2319 static ssize_t
2320 state_show(mdk_rdev_t *rdev, char *page)
2321 {
2322         char *sep = "";
2323         size_t len = 0;
2324
2325         if (test_bit(Faulty, &rdev->flags)) {
2326                 len+= sprintf(page+len, "%sfaulty",sep);
2327                 sep = ",";
2328         }
2329         if (test_bit(In_sync, &rdev->flags)) {
2330                 len += sprintf(page+len, "%sin_sync",sep);
2331                 sep = ",";
2332         }
2333         if (test_bit(WriteMostly, &rdev->flags)) {
2334                 len += sprintf(page+len, "%swrite_mostly",sep);
2335                 sep = ",";
2336         }
2337         if (test_bit(Blocked, &rdev->flags)) {
2338                 len += sprintf(page+len, "%sblocked", sep);
2339                 sep = ",";
2340         }
2341         if (!test_bit(Faulty, &rdev->flags) &&
2342             !test_bit(In_sync, &rdev->flags)) {
2343                 len += sprintf(page+len, "%sspare", sep);
2344                 sep = ",";
2345         }
2346         return len+sprintf(page+len, "\n");
2347 }
2348
2349 static ssize_t
2350 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2351 {
2352         /* can write
2353          *  faulty  - simulates and error
2354          *  remove  - disconnects the device
2355          *  writemostly - sets write_mostly
2356          *  -writemostly - clears write_mostly
2357          *  blocked - sets the Blocked flag
2358          *  -blocked - clears the Blocked flag
2359          *  insync - sets Insync providing device isn't active
2360          */
2361         int err = -EINVAL;
2362         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2363                 md_error(rdev->mddev, rdev);
2364                 err = 0;
2365         } else if (cmd_match(buf, "remove")) {
2366                 if (rdev->raid_disk >= 0)
2367                         err = -EBUSY;
2368                 else {
2369                         mddev_t *mddev = rdev->mddev;
2370                         kick_rdev_from_array(rdev);
2371                         if (mddev->pers)
2372                                 md_update_sb(mddev, 1);
2373                         md_new_event(mddev);
2374                         err = 0;
2375                 }
2376         } else if (cmd_match(buf, "writemostly")) {
2377                 set_bit(WriteMostly, &rdev->flags);
2378                 err = 0;
2379         } else if (cmd_match(buf, "-writemostly")) {
2380                 clear_bit(WriteMostly, &rdev->flags);
2381                 err = 0;
2382         } else if (cmd_match(buf, "blocked")) {
2383                 set_bit(Blocked, &rdev->flags);
2384                 err = 0;
2385         } else if (cmd_match(buf, "-blocked")) {
2386                 clear_bit(Blocked, &rdev->flags);
2387                 wake_up(&rdev->blocked_wait);
2388                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2389                 md_wakeup_thread(rdev->mddev->thread);
2390
2391                 err = 0;
2392         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2393                 set_bit(In_sync, &rdev->flags);
2394                 err = 0;
2395         }
2396         if (!err)
2397                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2398         return err ? err : len;
2399 }
2400 static struct rdev_sysfs_entry rdev_state =
2401 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2402
2403 static ssize_t
2404 errors_show(mdk_rdev_t *rdev, char *page)
2405 {
2406         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2407 }
2408
2409 static ssize_t
2410 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2411 {
2412         char *e;
2413         unsigned long n = simple_strtoul(buf, &e, 10);
2414         if (*buf && (*e == 0 || *e == '\n')) {
2415                 atomic_set(&rdev->corrected_errors, n);
2416                 return len;
2417         }
2418         return -EINVAL;
2419 }
2420 static struct rdev_sysfs_entry rdev_errors =
2421 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2422
2423 static ssize_t
2424 slot_show(mdk_rdev_t *rdev, char *page)
2425 {
2426         if (rdev->raid_disk < 0)
2427                 return sprintf(page, "none\n");
2428         else
2429                 return sprintf(page, "%d\n", rdev->raid_disk);
2430 }
2431
2432 static ssize_t
2433 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2434 {
2435         char *e;
2436         int err;
2437         char nm[20];
2438         int slot = simple_strtoul(buf, &e, 10);
2439         if (strncmp(buf, "none", 4)==0)
2440                 slot = -1;
2441         else if (e==buf || (*e && *e!= '\n'))
2442                 return -EINVAL;
2443         if (rdev->mddev->pers && slot == -1) {
2444                 /* Setting 'slot' on an active array requires also
2445                  * updating the 'rd%d' link, and communicating
2446                  * with the personality with ->hot_*_disk.
2447                  * For now we only support removing
2448                  * failed/spare devices.  This normally happens automatically,
2449                  * but not when the metadata is externally managed.
2450                  */
2451                 if (rdev->raid_disk == -1)
2452                         return -EEXIST;
2453                 /* personality does all needed checks */
2454                 if (rdev->mddev->pers->hot_add_disk == NULL)
2455                         return -EINVAL;
2456                 err = rdev->mddev->pers->
2457                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2458                 if (err)
2459                         return err;
2460                 sprintf(nm, "rd%d", rdev->raid_disk);
2461                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2462                 rdev->raid_disk = -1;
2463                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2464                 md_wakeup_thread(rdev->mddev->thread);
2465         } else if (rdev->mddev->pers) {
2466                 mdk_rdev_t *rdev2;
2467                 /* Activating a spare .. or possibly reactivating
2468                  * if we ever get bitmaps working here.
2469                  */
2470
2471                 if (rdev->raid_disk != -1)
2472                         return -EBUSY;
2473
2474                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2475                         return -EBUSY;
2476
2477                 if (rdev->mddev->pers->hot_add_disk == NULL)
2478                         return -EINVAL;
2479
2480                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2481                         if (rdev2->raid_disk == slot)
2482                                 return -EEXIST;
2483
2484                 if (slot >= rdev->mddev->raid_disks &&
2485                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2486                         return -ENOSPC;
2487
2488                 rdev->raid_disk = slot;
2489                 if (test_bit(In_sync, &rdev->flags))
2490                         rdev->saved_raid_disk = slot;
2491                 else
2492                         rdev->saved_raid_disk = -1;
2493                 err = rdev->mddev->pers->
2494                         hot_add_disk(rdev->mddev, rdev);
2495                 if (err) {
2496                         rdev->raid_disk = -1;
2497                         return err;
2498                 } else
2499                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2500                 sprintf(nm, "rd%d", rdev->raid_disk);
2501                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2502                         /* failure here is OK */;
2503                 /* don't wakeup anyone, leave that to userspace. */
2504         } else {
2505                 if (slot >= rdev->mddev->raid_disks &&
2506                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2507                         return -ENOSPC;
2508                 rdev->raid_disk = slot;
2509                 /* assume it is working */
2510                 clear_bit(Faulty, &rdev->flags);
2511                 clear_bit(WriteMostly, &rdev->flags);
2512                 set_bit(In_sync, &rdev->flags);
2513                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2514         }
2515         return len;
2516 }
2517
2518
2519 static struct rdev_sysfs_entry rdev_slot =
2520 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2521
2522 static ssize_t
2523 offset_show(mdk_rdev_t *rdev, char *page)
2524 {
2525         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2526 }
2527
2528 static ssize_t
2529 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2530 {
2531         char *e;
2532         unsigned long long offset = simple_strtoull(buf, &e, 10);
2533         if (e==buf || (*e && *e != '\n'))
2534                 return -EINVAL;
2535         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2536                 return -EBUSY;
2537         if (rdev->sectors && rdev->mddev->external)
2538                 /* Must set offset before size, so overlap checks
2539                  * can be sane */
2540                 return -EBUSY;
2541         rdev->data_offset = offset;
2542         return len;
2543 }
2544
2545 static struct rdev_sysfs_entry rdev_offset =
2546 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2547
2548 static ssize_t
2549 rdev_size_show(mdk_rdev_t *rdev, char *page)
2550 {
2551         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2552 }
2553
2554 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2555 {
2556         /* check if two start/length pairs overlap */
2557         if (s1+l1 <= s2)
2558                 return 0;
2559         if (s2+l2 <= s1)
2560                 return 0;
2561         return 1;
2562 }
2563
2564 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2565 {
2566         unsigned long long blocks;
2567         sector_t new;
2568
2569         if (strict_strtoull(buf, 10, &blocks) < 0)
2570                 return -EINVAL;
2571
2572         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2573                 return -EINVAL; /* sector conversion overflow */
2574
2575         new = blocks * 2;
2576         if (new != blocks * 2)
2577                 return -EINVAL; /* unsigned long long to sector_t overflow */
2578
2579         *sectors = new;
2580         return 0;
2581 }
2582
2583 static ssize_t
2584 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2585 {
2586         mddev_t *my_mddev = rdev->mddev;
2587         sector_t oldsectors = rdev->sectors;
2588         sector_t sectors;
2589
2590         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2591                 return -EINVAL;
2592         if (my_mddev->pers && rdev->raid_disk >= 0) {
2593                 if (my_mddev->persistent) {
2594                         sectors = super_types[my_mddev->major_version].
2595                                 rdev_size_change(rdev, sectors);
2596                         if (!sectors)
2597                                 return -EBUSY;
2598                 } else if (!sectors)
2599                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2600                                 rdev->data_offset;
2601         }
2602         if (sectors < my_mddev->dev_sectors)
2603                 return -EINVAL; /* component must fit device */
2604
2605         rdev->sectors = sectors;
2606         if (sectors > oldsectors && my_mddev->external) {
2607                 /* need to check that all other rdevs with the same ->bdev
2608                  * do not overlap.  We need to unlock the mddev to avoid
2609                  * a deadlock.  We have already changed rdev->sectors, and if
2610                  * we have to change it back, we will have the lock again.
2611                  */
2612                 mddev_t *mddev;
2613                 int overlap = 0;
2614                 struct list_head *tmp;
2615
2616                 mddev_unlock(my_mddev);
2617                 for_each_mddev(mddev, tmp) {
2618                         mdk_rdev_t *rdev2;
2619
2620                         mddev_lock(mddev);
2621                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2622                                 if (rdev->bdev == rdev2->bdev &&
2623                                     rdev != rdev2 &&
2624                                     overlaps(rdev->data_offset, rdev->sectors,
2625                                              rdev2->data_offset,
2626                                              rdev2->sectors)) {
2627                                         overlap = 1;
2628                                         break;
2629                                 }
2630                         mddev_unlock(mddev);
2631                         if (overlap) {
2632                                 mddev_put(mddev);
2633                                 break;
2634                         }
2635                 }
2636                 mddev_lock(my_mddev);
2637                 if (overlap) {
2638                         /* Someone else could have slipped in a size
2639                          * change here, but doing so is just silly.
2640                          * We put oldsectors back because we *know* it is
2641                          * safe, and trust userspace not to race with
2642                          * itself
2643                          */
2644                         rdev->sectors = oldsectors;
2645                         return -EBUSY;
2646                 }
2647         }
2648         return len;
2649 }
2650
2651 static struct rdev_sysfs_entry rdev_size =
2652 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2653
2654
2655 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2656 {
2657         unsigned long long recovery_start = rdev->recovery_offset;
2658
2659         if (test_bit(In_sync, &rdev->flags) ||
2660             recovery_start == MaxSector)
2661                 return sprintf(page, "none\n");
2662
2663         return sprintf(page, "%llu\n", recovery_start);
2664 }
2665
2666 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2667 {
2668         unsigned long long recovery_start;
2669
2670         if (cmd_match(buf, "none"))
2671                 recovery_start = MaxSector;
2672         else if (strict_strtoull(buf, 10, &recovery_start))
2673                 return -EINVAL;
2674
2675         if (rdev->mddev->pers &&
2676             rdev->raid_disk >= 0)
2677                 return -EBUSY;
2678
2679         rdev->recovery_offset = recovery_start;
2680         if (recovery_start == MaxSector)
2681                 set_bit(In_sync, &rdev->flags);
2682         else
2683                 clear_bit(In_sync, &rdev->flags);
2684         return len;
2685 }
2686
2687 static struct rdev_sysfs_entry rdev_recovery_start =
2688 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2689
2690 static struct attribute *rdev_default_attrs[] = {
2691         &rdev_state.attr,
2692         &rdev_errors.attr,
2693         &rdev_slot.attr,
2694         &rdev_offset.attr,
2695         &rdev_size.attr,
2696         &rdev_recovery_start.attr,
2697         NULL,
2698 };
2699 static ssize_t
2700 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2701 {
2702         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2703         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2704         mddev_t *mddev = rdev->mddev;
2705         ssize_t rv;
2706
2707         if (!entry->show)
2708                 return -EIO;
2709
2710         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2711         if (!rv) {
2712                 if (rdev->mddev == NULL)
2713                         rv = -EBUSY;
2714                 else
2715                         rv = entry->show(rdev, page);
2716                 mddev_unlock(mddev);
2717         }
2718         return rv;
2719 }
2720
2721 static ssize_t
2722 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2723               const char *page, size_t length)
2724 {
2725         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2726         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2727         ssize_t rv;
2728         mddev_t *mddev = rdev->mddev;
2729
2730         if (!entry->store)
2731                 return -EIO;
2732         if (!capable(CAP_SYS_ADMIN))
2733                 return -EACCES;
2734         rv = mddev ? mddev_lock(mddev): -EBUSY;
2735         if (!rv) {
2736                 if (rdev->mddev == NULL)
2737                         rv = -EBUSY;
2738                 else
2739                         rv = entry->store(rdev, page, length);
2740                 mddev_unlock(mddev);
2741         }
2742         return rv;
2743 }
2744
2745 static void rdev_free(struct kobject *ko)
2746 {
2747         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2748         kfree(rdev);
2749 }
2750 static const struct sysfs_ops rdev_sysfs_ops = {
2751         .show           = rdev_attr_show,
2752         .store          = rdev_attr_store,
2753 };
2754 static struct kobj_type rdev_ktype = {
2755         .release        = rdev_free,
2756         .sysfs_ops      = &rdev_sysfs_ops,
2757         .default_attrs  = rdev_default_attrs,
2758 };
2759
2760 void md_rdev_init(mdk_rdev_t *rdev)
2761 {
2762         rdev->desc_nr = -1;
2763         rdev->saved_raid_disk = -1;
2764         rdev->raid_disk = -1;
2765         rdev->flags = 0;
2766         rdev->data_offset = 0;
2767         rdev->sb_events = 0;
2768         rdev->last_read_error.tv_sec  = 0;
2769         rdev->last_read_error.tv_nsec = 0;
2770         atomic_set(&rdev->nr_pending, 0);
2771         atomic_set(&rdev->read_errors, 0);
2772         atomic_set(&rdev->corrected_errors, 0);
2773
2774         INIT_LIST_HEAD(&rdev->same_set);
2775         init_waitqueue_head(&rdev->blocked_wait);
2776 }
2777 EXPORT_SYMBOL_GPL(md_rdev_init);
2778 /*
2779  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2780  *
2781  * mark the device faulty if:
2782  *
2783  *   - the device is nonexistent (zero size)
2784  *   - the device has no valid superblock
2785  *
2786  * a faulty rdev _never_ has rdev->sb set.
2787  */
2788 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2789 {
2790         char b[BDEVNAME_SIZE];
2791         int err;
2792         mdk_rdev_t *rdev;
2793         sector_t size;
2794
2795         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2796         if (!rdev) {
2797                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2798                 return ERR_PTR(-ENOMEM);
2799         }
2800
2801         md_rdev_init(rdev);
2802         if ((err = alloc_disk_sb(rdev)))
2803                 goto abort_free;
2804
2805         err = lock_rdev(rdev, newdev, super_format == -2);
2806         if (err)
2807                 goto abort_free;
2808
2809         kobject_init(&rdev->kobj, &rdev_ktype);
2810
2811         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2812         if (!size) {
2813                 printk(KERN_WARNING
2814                         "md: %s has zero or unknown size, marking faulty!\n",
2815                         bdevname(rdev->bdev,b));
2816                 err = -EINVAL;
2817                 goto abort_free;
2818         }
2819
2820         if (super_format >= 0) {
2821                 err = super_types[super_format].
2822                         load_super(rdev, NULL, super_minor);
2823                 if (err == -EINVAL) {
2824                         printk(KERN_WARNING
2825                                 "md: %s does not have a valid v%d.%d "
2826                                "superblock, not importing!\n",
2827                                 bdevname(rdev->bdev,b),
2828                                super_format, super_minor);
2829                         goto abort_free;
2830                 }
2831                 if (err < 0) {
2832                         printk(KERN_WARNING
2833                                 "md: could not read %s's sb, not importing!\n",
2834                                 bdevname(rdev->bdev,b));
2835                         goto abort_free;
2836                 }
2837         }
2838
2839         return rdev;
2840
2841 abort_free:
2842         if (rdev->sb_page) {
2843                 if (rdev->bdev)
2844                         unlock_rdev(rdev);
2845                 free_disk_sb(rdev);
2846         }
2847         kfree(rdev);
2848         return ERR_PTR(err);
2849 }
2850
2851 /*
2852  * Check a full RAID array for plausibility
2853  */
2854
2855
2856 static void analyze_sbs(mddev_t * mddev)
2857 {
2858         int i;
2859         mdk_rdev_t *rdev, *freshest, *tmp;
2860         char b[BDEVNAME_SIZE];
2861
2862         freshest = NULL;
2863         rdev_for_each(rdev, tmp, mddev)
2864                 switch (super_types[mddev->major_version].
2865                         load_super(rdev, freshest, mddev->minor_version)) {
2866                 case 1:
2867                         freshest = rdev;
2868                         break;
2869                 case 0:
2870                         break;
2871                 default:
2872                         printk( KERN_ERR \
2873                                 "md: fatal superblock inconsistency in %s"
2874                                 " -- removing from array\n",
2875                                 bdevname(rdev->bdev,b));
2876                         kick_rdev_from_array(rdev);
2877                 }
2878
2879
2880         super_types[mddev->major_version].
2881                 validate_super(mddev, freshest);
2882
2883         i = 0;
2884         rdev_for_each(rdev, tmp, mddev) {
2885                 if (mddev->max_disks &&
2886                     (rdev->desc_nr >= mddev->max_disks ||
2887                      i > mddev->max_disks)) {
2888                         printk(KERN_WARNING
2889                                "md: %s: %s: only %d devices permitted\n",
2890                                mdname(mddev), bdevname(rdev->bdev, b),
2891                                mddev->max_disks);
2892                         kick_rdev_from_array(rdev);
2893                         continue;
2894                 }
2895                 if (rdev != freshest)
2896                         if (super_types[mddev->major_version].
2897                             validate_super(mddev, rdev)) {
2898                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2899                                         " from array!\n",
2900                                         bdevname(rdev->bdev,b));
2901                                 kick_rdev_from_array(rdev);
2902                                 continue;
2903                         }
2904                 if (mddev->level == LEVEL_MULTIPATH) {
2905                         rdev->desc_nr = i++;
2906                         rdev->raid_disk = rdev->desc_nr;
2907                         set_bit(In_sync, &rdev->flags);
2908                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2909                         rdev->raid_disk = -1;
2910                         clear_bit(In_sync, &rdev->flags);
2911                 }
2912         }
2913 }
2914
2915 /* Read a fixed-point number.
2916  * Numbers in sysfs attributes should be in "standard" units where
2917  * possible, so time should be in seconds.
2918  * However we internally use a a much smaller unit such as
2919  * milliseconds or jiffies.
2920  * This function takes a decimal number with a possible fractional
2921  * component, and produces an integer which is the result of
2922  * multiplying that number by 10^'scale'.
2923  * all without any floating-point arithmetic.
2924  */
2925 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2926 {
2927         unsigned long result = 0;
2928         long decimals = -1;
2929         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2930                 if (*cp == '.')
2931                         decimals = 0;
2932                 else if (decimals < scale) {
2933                         unsigned int value;
2934                         value = *cp - '0';
2935                         result = result * 10 + value;
2936                         if (decimals >= 0)
2937                                 decimals++;
2938                 }
2939                 cp++;
2940         }
2941         if (*cp == '\n')
2942                 cp++;
2943         if (*cp)
2944                 return -EINVAL;
2945         if (decimals < 0)
2946                 decimals = 0;
2947         while (decimals < scale) {
2948                 result *= 10;
2949                 decimals ++;
2950         }
2951         *res = result;
2952         return 0;
2953 }
2954
2955
2956 static void md_safemode_timeout(unsigned long data);
2957
2958 static ssize_t
2959 safe_delay_show(mddev_t *mddev, char *page)
2960 {
2961         int msec = (mddev->safemode_delay*1000)/HZ;
2962         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2963 }
2964 static ssize_t
2965 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2966 {
2967         unsigned long msec;
2968
2969         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2970                 return -EINVAL;
2971         if (msec == 0)
2972                 mddev->safemode_delay = 0;
2973         else {
2974                 unsigned long old_delay = mddev->safemode_delay;
2975                 mddev->safemode_delay = (msec*HZ)/1000;
2976                 if (mddev->safemode_delay == 0)
2977                         mddev->safemode_delay = 1;
2978                 if (mddev->safemode_delay < old_delay)
2979                         md_safemode_timeout((unsigned long)mddev);
2980         }
2981         return len;
2982 }
2983 static struct md_sysfs_entry md_safe_delay =
2984 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2985
2986 static ssize_t
2987 level_show(mddev_t *mddev, char *page)
2988 {
2989         struct mdk_personality *p = mddev->pers;
2990         if (p)
2991                 return sprintf(page, "%s\n", p->name);
2992         else if (mddev->clevel[0])
2993                 return sprintf(page, "%s\n", mddev->clevel);
2994         else if (mddev->level != LEVEL_NONE)
2995                 return sprintf(page, "%d\n", mddev->level);
2996         else
2997                 return 0;
2998 }
2999
3000 static ssize_t
3001 level_store(mddev_t *mddev, const char *buf, size_t len)
3002 {
3003         char clevel[16];
3004         ssize_t rv = len;
3005         struct mdk_personality *pers;
3006         long level;
3007         void *priv;
3008         mdk_rdev_t *rdev;
3009
3010         if (mddev->pers == NULL) {
3011                 if (len == 0)
3012                         return 0;
3013                 if (len >= sizeof(mddev->clevel))
3014                         return -ENOSPC;
3015                 strncpy(mddev->clevel, buf, len);
3016                 if (mddev->clevel[len-1] == '\n')
3017                         len--;
3018                 mddev->clevel[len] = 0;
3019                 mddev->level = LEVEL_NONE;
3020                 return rv;
3021         }
3022
3023         /* request to change the personality.  Need to ensure:
3024          *  - array is not engaged in resync/recovery/reshape
3025          *  - old personality can be suspended
3026          *  - new personality will access other array.
3027          */
3028
3029         if (mddev->sync_thread ||
3030             mddev->reshape_position != MaxSector ||
3031             mddev->sysfs_active)
3032                 return -EBUSY;
3033
3034         if (!mddev->pers->quiesce) {
3035                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3036                        mdname(mddev), mddev->pers->name);
3037                 return -EINVAL;
3038         }
3039
3040         /* Now find the new personality */
3041         if (len == 0 || len >= sizeof(clevel))
3042                 return -EINVAL;
3043         strncpy(clevel, buf, len);
3044         if (clevel[len-1] == '\n')
3045                 len--;
3046         clevel[len] = 0;
3047         if (strict_strtol(clevel, 10, &level))
3048                 level = LEVEL_NONE;
3049
3050         if (request_module("md-%s", clevel) != 0)
3051                 request_module("md-level-%s", clevel);
3052         spin_lock(&pers_lock);
3053         pers = find_pers(level, clevel);
3054         if (!pers || !try_module_get(pers->owner)) {
3055                 spin_unlock(&pers_lock);
3056                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3057                 return -EINVAL;
3058         }
3059         spin_unlock(&pers_lock);
3060
3061         if (pers == mddev->pers) {
3062                 /* Nothing to do! */
3063                 module_put(pers->owner);
3064                 return rv;
3065         }
3066         if (!pers->takeover) {
3067                 module_put(pers->owner);
3068                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3069                        mdname(mddev), clevel);
3070                 return -EINVAL;
3071         }
3072
3073         list_for_each_entry(rdev, &mddev->disks, same_set)
3074                 rdev->new_raid_disk = rdev->raid_disk;
3075
3076         /* ->takeover must set new_* and/or delta_disks
3077          * if it succeeds, and may set them when it fails.
3078          */
3079         priv = pers->takeover(mddev);
3080         if (IS_ERR(priv)) {
3081                 mddev->new_level = mddev->level;
3082                 mddev->new_layout = mddev->layout;
3083                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3084                 mddev->raid_disks -= mddev->delta_disks;
3085                 mddev->delta_disks = 0;
3086                 module_put(pers->owner);
3087                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3088                        mdname(mddev), clevel);
3089                 return PTR_ERR(priv);
3090         }
3091
3092         /* Looks like we have a winner */
3093         mddev_suspend(mddev);
3094         mddev->pers->stop(mddev);
3095
3096         if (mddev->pers->sync_request == NULL &&
3097             pers->sync_request != NULL) {
3098                 /* need to add the md_redundancy_group */
3099                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3100                         printk(KERN_WARNING
3101                                "md: cannot register extra attributes for %s\n",
3102                                mdname(mddev));
3103                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3104         }
3105         if (mddev->pers->sync_request != NULL &&
3106             pers->sync_request == NULL) {
3107                 /* need to remove the md_redundancy_group */
3108                 if (mddev->to_remove == NULL)
3109                         mddev->to_remove = &md_redundancy_group;
3110         }
3111
3112         if (mddev->pers->sync_request == NULL &&
3113             mddev->external) {
3114                 /* We are converting from a no-redundancy array
3115                  * to a redundancy array and metadata is managed
3116                  * externally so we need to be sure that writes
3117                  * won't block due to a need to transition
3118                  *      clean->dirty
3119                  * until external management is started.
3120                  */
3121                 mddev->in_sync = 0;
3122                 mddev->safemode_delay = 0;
3123                 mddev->safemode = 0;
3124         }
3125
3126         list_for_each_entry(rdev, &mddev->disks, same_set) {
3127                 char nm[20];
3128                 if (rdev->raid_disk < 0)
3129                         continue;
3130                 if (rdev->new_raid_disk >= mddev->raid_disks)
3131                         rdev->new_raid_disk = -1;
3132                 if (rdev->new_raid_disk == rdev->raid_disk)
3133                         continue;
3134                 sprintf(nm, "rd%d", rdev->raid_disk);
3135                 sysfs_remove_link(&mddev->kobj, nm);
3136         }
3137         list_for_each_entry(rdev, &mddev->disks, same_set) {
3138                 if (rdev->raid_disk < 0)
3139                         continue;
3140                 if (rdev->new_raid_disk == rdev->raid_disk)
3141                         continue;
3142                 rdev->raid_disk = rdev->new_raid_disk;
3143                 if (rdev->raid_disk < 0)
3144                         clear_bit(In_sync, &rdev->flags);
3145                 else {
3146                         char nm[20];
3147                         sprintf(nm, "rd%d", rdev->raid_disk);
3148                         if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3149                                 printk("md: cannot register %s for %s after level change\n",
3150                                        nm, mdname(mddev));
3151                 }
3152         }
3153
3154         module_put(mddev->pers->owner);
3155         mddev->pers = pers;
3156         mddev->private = priv;
3157         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3158         mddev->level = mddev->new_level;
3159         mddev->layout = mddev->new_layout;
3160         mddev->chunk_sectors = mddev->new_chunk_sectors;
3161         mddev->delta_disks = 0;
3162         if (mddev->pers->sync_request == NULL) {
3163                 /* this is now an array without redundancy, so
3164                  * it must always be in_sync
3165                  */
3166                 mddev->in_sync = 1;
3167                 del_timer_sync(&mddev->safemode_timer);
3168         }
3169         pers->run(mddev);
3170         mddev_resume(mddev);
3171         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3172         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3173         md_wakeup_thread(mddev->thread);
3174         sysfs_notify(&mddev->kobj, NULL, "level");
3175         md_new_event(mddev);
3176         return rv;
3177 }
3178
3179 static struct md_sysfs_entry md_level =
3180 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3181
3182
3183 static ssize_t
3184 layout_show(mddev_t *mddev, char *page)
3185 {
3186         /* just a number, not meaningful for all levels */
3187         if (mddev->reshape_position != MaxSector &&
3188             mddev->layout != mddev->new_layout)
3189                 return sprintf(page, "%d (%d)\n",
3190                                mddev->new_layout, mddev->layout);
3191         return sprintf(page, "%d\n", mddev->layout);
3192 }
3193
3194 static ssize_t
3195 layout_store(mddev_t *mddev, const char *buf, size_t len)
3196 {
3197         char *e;
3198         unsigned long n = simple_strtoul(buf, &e, 10);
3199
3200         if (!*buf || (*e && *e != '\n'))
3201                 return -EINVAL;
3202
3203         if (mddev->pers) {
3204                 int err;
3205                 if (mddev->pers->check_reshape == NULL)
3206                         return -EBUSY;
3207                 mddev->new_layout = n;
3208                 err = mddev->pers->check_reshape(mddev);
3209                 if (err) {
3210                         mddev->new_layout = mddev->layout;
3211                         return err;
3212                 }
3213         } else {
3214                 mddev->new_layout = n;
3215                 if (mddev->reshape_position == MaxSector)
3216                         mddev->layout = n;
3217         }
3218         return len;
3219 }
3220 static struct md_sysfs_entry md_layout =
3221 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3222
3223
3224 static ssize_t
3225 raid_disks_show(mddev_t *mddev, char *page)
3226 {
3227         if (mddev->raid_disks == 0)
3228                 return 0;
3229         if (mddev->reshape_position != MaxSector &&
3230             mddev->delta_disks != 0)
3231                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3232                                mddev->raid_disks - mddev->delta_disks);
3233         return sprintf(page, "%d\n", mddev->raid_disks);
3234 }
3235
3236 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3237
3238 static ssize_t
3239 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3240 {
3241         char *e;
3242         int rv = 0;
3243         unsigned long n = simple_strtoul(buf, &e, 10);
3244
3245         if (!*buf || (*e && *e != '\n'))
3246                 return -EINVAL;
3247
3248         if (mddev->pers)
3249                 rv = update_raid_disks(mddev, n);
3250         else if (mddev->reshape_position != MaxSector) {
3251                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3252                 mddev->delta_disks = n - olddisks;
3253                 mddev->raid_disks = n;
3254         } else
3255                 mddev->raid_disks = n;
3256         return rv ? rv : len;
3257 }
3258 static struct md_sysfs_entry md_raid_disks =
3259 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3260
3261 static ssize_t
3262 chunk_size_show(mddev_t *mddev, char *page)
3263 {
3264         if (mddev->reshape_position != MaxSector &&
3265             mddev->chunk_sectors != mddev->new_chunk_sectors)
3266                 return sprintf(page, "%d (%d)\n",
3267                                mddev->new_chunk_sectors << 9,
3268                                mddev->chunk_sectors << 9);
3269         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3270 }
3271
3272 static ssize_t
3273 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3274 {
3275         char *e;
3276         unsigned long n = simple_strtoul(buf, &e, 10);
3277
3278         if (!*buf || (*e && *e != '\n'))
3279                 return -EINVAL;
3280
3281         if (mddev->pers) {
3282                 int err;
3283                 if (mddev->pers->check_reshape == NULL)
3284                         return -EBUSY;
3285                 mddev->new_chunk_sectors = n >> 9;
3286                 err = mddev->pers->check_reshape(mddev);
3287                 if (err) {
3288                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3289                         return err;
3290                 }
3291         } else {
3292                 mddev->new_chunk_sectors = n >> 9;
3293                 if (mddev->reshape_position == MaxSector)
3294                         mddev->chunk_sectors = n >> 9;
3295         }
3296         return len;
3297 }
3298 static struct md_sysfs_entry md_chunk_size =
3299 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3300
3301 static ssize_t
3302 resync_start_show(mddev_t *mddev, char *page)
3303 {
3304         if (mddev->recovery_cp == MaxSector)
3305                 return sprintf(page, "none\n");
3306         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3307 }
3308
3309 static ssize_t
3310 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3311 {
3312         char *e;
3313         unsigned long long n = simple_strtoull(buf, &e, 10);
3314
3315         if (mddev->pers)
3316                 return -EBUSY;
3317         if (cmd_match(buf, "none"))
3318                 n = MaxSector;
3319         else if (!*buf || (*e && *e != '\n'))
3320                 return -EINVAL;
3321
3322         mddev->recovery_cp = n;
3323         return len;
3324 }
3325 static struct md_sysfs_entry md_resync_start =
3326 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3327
3328 /*
3329  * The array state can be:
3330  *
3331  * clear
3332  *     No devices, no size, no level
3333  *     Equivalent to STOP_ARRAY ioctl
3334  * inactive
3335  *     May have some settings, but array is not active
3336  *        all IO results in error
3337  *     When written, doesn't tear down array, but just stops it
3338  * suspended (not supported yet)
3339  *     All IO requests will block. The array can be reconfigured.
3340  *     Writing this, if accepted, will block until array is quiescent
3341  * readonly
3342  *     no resync can happen.  no superblocks get written.
3343  *     write requests fail
3344  * read-auto
3345  *     like readonly, but behaves like 'clean' on a write request.
3346  *
3347  * clean - no pending writes, but otherwise active.
3348  *     When written to inactive array, starts without resync
3349  *     If a write request arrives then
3350  *       if metadata is known, mark 'dirty' and switch to 'active'.
3351  *       if not known, block and switch to write-pending
3352  *     If written to an active array that has pending writes, then fails.
3353  * active
3354  *     fully active: IO and resync can be happening.
3355  *     When written to inactive array, starts with resync
3356  *
3357  * write-pending
3358  *     clean, but writes are blocked waiting for 'active' to be written.
3359  *
3360  * active-idle
3361  *     like active, but no writes have been seen for a while (100msec).
3362  *
3363  */
3364 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3365                    write_pending, active_idle, bad_word};
3366 static char *array_states[] = {
3367         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3368         "write-pending", "active-idle", NULL };
3369
3370 static int match_word(const char *word, char **list)
3371 {
3372         int n;
3373         for (n=0; list[n]; n++)
3374                 if (cmd_match(word, list[n]))
3375                         break;
3376         return n;
3377 }
3378
3379 static ssize_t
3380 array_state_show(mddev_t *mddev, char *page)
3381 {
3382         enum array_state st = inactive;
3383
3384         if (mddev->pers)
3385                 switch(mddev->ro) {
3386                 case 1:
3387                         st = readonly;
3388                         break;
3389                 case 2:
3390                         st = read_auto;
3391                         break;
3392                 case 0:
3393                         if (mddev->in_sync)
3394                                 st = clean;
3395                         else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3396                                 st = write_pending;
3397                         else if (mddev->safemode)
3398                                 st = active_idle;
3399                         else
3400                                 st = active;
3401                 }
3402         else {
3403                 if (list_empty(&mddev->disks) &&
3404                     mddev->raid_disks == 0 &&
3405                     mddev->dev_sectors == 0)
3406                         st = clear;
3407                 else
3408                         st = inactive;
3409         }
3410         return sprintf(page, "%s\n", array_states[st]);
3411 }
3412
3413 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3414 static int md_set_readonly(mddev_t * mddev, int is_open);
3415 static int do_md_run(mddev_t * mddev);
3416 static int restart_array(mddev_t *mddev);
3417
3418 static ssize_t
3419 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3420 {
3421         int err = -EINVAL;
3422         enum array_state st = match_word(buf, array_states);
3423         switch(st) {
3424         case bad_word:
3425                 break;
3426         case clear:
3427                 /* stopping an active array */
3428                 if (atomic_read(&mddev->openers) > 0)
3429                         return -EBUSY;
3430                 err = do_md_stop(mddev, 0, 0);
3431                 break;
3432         case inactive:
3433                 /* stopping an active array */
3434                 if (mddev->pers) {
3435                         if (atomic_read(&mddev->openers) > 0)
3436                                 return -EBUSY;
3437                         err = do_md_stop(mddev, 2, 0);
3438                 } else
3439                         err = 0; /* already inactive */
3440                 break;
3441         case suspended:
3442                 break; /* not supported yet */
3443         case readonly:
3444                 if (mddev->pers)
3445                         err = md_set_readonly(mddev, 0);
3446                 else {
3447                         mddev->ro = 1;
3448                         set_disk_ro(mddev->gendisk, 1);
3449                         err = do_md_run(mddev);
3450                 }
3451                 break;
3452         case read_auto:
3453                 if (mddev->pers) {
3454                         if (mddev->ro == 0)
3455                                 err = md_set_readonly(mddev, 0);
3456                         else if (mddev->ro == 1)
3457                                 err = restart_array(mddev);
3458                         if (err == 0) {
3459                                 mddev->ro = 2;
3460                                 set_disk_ro(mddev->gendisk, 0);
3461                         }
3462                 } else {
3463                         mddev->ro = 2;
3464                         err = do_md_run(mddev);
3465                 }
3466                 break;
3467         case clean:
3468                 if (mddev->pers) {
3469                         restart_array(mddev);
3470                         spin_lock_irq(&mddev->write_lock);
3471                         if (atomic_read(&mddev->writes_pending) == 0) {
3472                                 if (mddev->in_sync == 0) {
3473                                         mddev->in_sync = 1;
3474                                         if (mddev->safemode == 1)
3475                                                 mddev->safemode = 0;
3476                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3477                                 }
3478                                 err = 0;
3479                         } else
3480                                 err = -EBUSY;
3481                         spin_unlock_irq(&mddev->write_lock);
3482                 } else
3483                         err = -EINVAL;
3484                 break;
3485         case active:
3486                 if (mddev->pers) {
3487                         restart_array(mddev);
3488                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3489                         wake_up(&mddev->sb_wait);
3490                         err = 0;
3491                 } else {
3492                         mddev->ro = 0;
3493                         set_disk_ro(mddev->gendisk, 0);
3494                         err = do_md_run(mddev);
3495                 }
3496                 break;
3497         case write_pending:
3498         case active_idle:
3499                 /* these cannot be set */
3500                 break;
3501         }
3502         if (err)
3503                 return err;
3504         else {
3505                 sysfs_notify_dirent_safe(mddev->sysfs_state);
3506                 return len;
3507         }
3508 }
3509 static struct md_sysfs_entry md_array_state =
3510 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3511
3512 static ssize_t
3513 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3514         return sprintf(page, "%d\n",
3515                        atomic_read(&mddev->max_corr_read_errors));
3516 }
3517
3518 static ssize_t
3519 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3520 {
3521         char *e;
3522         unsigned long n = simple_strtoul(buf, &e, 10);
3523
3524         if (*buf && (*e == 0 || *e == '\n')) {
3525                 atomic_set(&mddev->max_corr_read_errors, n);
3526                 return len;
3527         }
3528         return -EINVAL;
3529 }
3530
3531 static struct md_sysfs_entry max_corr_read_errors =
3532 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3533         max_corrected_read_errors_store);
3534
3535 static ssize_t
3536 null_show(mddev_t *mddev, char *page)
3537 {
3538         return -EINVAL;
3539 }
3540
3541 static ssize_t
3542 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3543 {
3544         /* buf must be %d:%d\n? giving major and minor numbers */
3545         /* The new device is added to the array.
3546          * If the array has a persistent superblock, we read the
3547          * superblock to initialise info and check validity.
3548          * Otherwise, only checking done is that in bind_rdev_to_array,
3549          * which mainly checks size.
3550          */
3551         char *e;
3552         int major = simple_strtoul(buf, &e, 10);
3553         int minor;
3554         dev_t dev;
3555         mdk_rdev_t *rdev;
3556         int err;
3557
3558         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3559                 return -EINVAL;
3560         minor = simple_strtoul(e+1, &e, 10);
3561         if (*e && *e != '\n')
3562                 return -EINVAL;
3563         dev = MKDEV(major, minor);
3564         if (major != MAJOR(dev) ||
3565             minor != MINOR(dev))
3566                 return -EOVERFLOW;
3567
3568
3569         if (mddev->persistent) {
3570                 rdev = md_import_device(dev, mddev->major_version,
3571                                         mddev->minor_version);
3572                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3573                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3574                                                        mdk_rdev_t, same_set);
3575                         err = super_types[mddev->major_version]
3576                                 .load_super(rdev, rdev0, mddev->minor_version);
3577                         if (err < 0)
3578                                 goto out;
3579                 }
3580         } else if (mddev->external)
3581                 rdev = md_import_device(dev, -2, -1);
3582         else
3583                 rdev = md_import_device(dev, -1, -1);
3584
3585         if (IS_ERR(rdev))
3586                 return PTR_ERR(rdev);
3587         err = bind_rdev_to_array(rdev, mddev);
3588  out:
3589         if (err)
3590                 export_rdev(rdev);
3591         return err ? err : len;
3592 }
3593
3594 static struct md_sysfs_entry md_new_device =
3595 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3596
3597 static ssize_t
3598 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3599 {
3600         char *end;
3601         unsigned long chunk, end_chunk;
3602
3603         if (!mddev->bitmap)
3604                 goto out;
3605         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3606         while (*buf) {
3607                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3608                 if (buf == end) break;
3609                 if (*end == '-') { /* range */
3610                         buf = end + 1;
3611                         end_chunk = simple_strtoul(buf, &end, 0);
3612                         if (buf == end) break;
3613                 }
3614                 if (*end && !isspace(*end)) break;
3615                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3616                 buf = skip_spaces(end);
3617         }
3618         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3619 out:
3620         return len;
3621 }
3622
3623 static struct md_sysfs_entry md_bitmap =
3624 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3625
3626 static ssize_t
3627 size_show(mddev_t *mddev, char *page)
3628 {
3629         return sprintf(page, "%llu\n",
3630                 (unsigned long long)mddev->dev_sectors / 2);
3631 }
3632
3633 static int update_size(mddev_t *mddev, sector_t num_sectors);
3634
3635 static ssize_t
3636 size_store(mddev_t *mddev, const char *buf, size_t len)
3637 {
3638         /* If array is inactive, we can reduce the component size, but
3639          * not increase it (except from 0).
3640          * If array is active, we can try an on-line resize
3641          */
3642         sector_t sectors;
3643         int err = strict_blocks_to_sectors(buf, &sectors);
3644
3645         if (err < 0)
3646                 return err;
3647         if (mddev->pers) {
3648                 err = update_size(mddev, sectors);
3649                 md_update_sb(mddev, 1);
3650         } else {
3651                 if (mddev->dev_sectors == 0 ||
3652                     mddev->dev_sectors > sectors)
3653                         mddev->dev_sectors = sectors;
3654                 else
3655                         err = -ENOSPC;
3656         }
3657         return err ? err : len;
3658 }
3659
3660 static struct md_sysfs_entry md_size =
3661 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3662
3663
3664 /* Metdata version.
3665  * This is one of
3666  *   'none' for arrays with no metadata (good luck...)
3667  *   'external' for arrays with externally managed metadata,
3668  * or N.M for internally known formats
3669  */
3670 static ssize_t
3671 metadata_show(mddev_t *mddev, char *page)
3672 {
3673         if (mddev->persistent)
3674                 return sprintf(page, "%d.%d\n",
3675                                mddev->major_version, mddev->minor_version);
3676         else if (mddev->external)
3677                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3678         else
3679                 return sprintf(page, "none\n");
3680 }
3681
3682 static ssize_t
3683 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3684 {
3685         int major, minor;
3686         char *e;
3687         /* Changing the details of 'external' metadata is
3688          * always permitted.  Otherwise there must be
3689          * no devices attached to the array.
3690          */
3691         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3692                 ;
3693         else if (!list_empty(&mddev->disks))
3694                 return -EBUSY;
3695
3696         if (cmd_match(buf, "none")) {
3697                 mddev->persistent = 0;
3698                 mddev->external = 0;
3699                 mddev->major_version = 0;
3700                 mddev->minor_version = 90;
3701                 return len;
3702         }
3703         if (strncmp(buf, "external:", 9) == 0) {
3704                 size_t namelen = len-9;
3705                 if (namelen >= sizeof(mddev->metadata_type))
3706                         namelen = sizeof(mddev->metadata_type)-1;
3707                 strncpy(mddev->metadata_type, buf+9, namelen);
3708                 mddev->metadata_type[namelen] = 0;
3709                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3710                         mddev->metadata_type[--namelen] = 0;
3711                 mddev->persistent = 0;
3712                 mddev->external = 1;
3713                 mddev->major_version = 0;
3714                 mddev->minor_version = 90;
3715                 return len;
3716         }
3717         major = simple_strtoul(buf, &e, 10);
3718         if (e==buf || *e != '.')
3719                 return -EINVAL;
3720         buf = e+1;
3721         minor = simple_strtoul(buf, &e, 10);
3722         if (e==buf || (*e && *e != '\n') )
3723                 return -EINVAL;
3724         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3725                 return -ENOENT;
3726         mddev->major_version = major;
3727         mddev->minor_version = minor;
3728         mddev->persistent = 1;
3729         mddev->external = 0;
3730         return len;
3731 }
3732
3733 static struct md_sysfs_entry md_metadata =
3734 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3735
3736 static ssize_t
3737 action_show(mddev_t *mddev, char *page)
3738 {
3739         char *type = "idle";
3740         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3741                 type = "frozen";
3742         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3743             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3744                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3745                         type = "reshape";
3746                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3747                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3748                                 type = "resync";
3749                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3750                                 type = "check";
3751                         else
3752                                 type = "repair";
3753                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3754                         type = "recover";
3755         }
3756         return sprintf(page, "%s\n", type);
3757 }
3758
3759 static void reap_sync_thread(mddev_t *mddev);
3760
3761 static ssize_t
3762 action_store(mddev_t *mddev, const char *page, size_t len)
3763 {
3764         if (!mddev->pers || !mddev->pers->sync_request)
3765                 return -EINVAL;
3766
3767         if (cmd_match(page, "frozen"))
3768                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3769         else
3770                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3771
3772         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3773                 if (mddev->sync_thread) {
3774                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3775                         reap_sync_thread(mddev);
3776                 }
3777         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3778                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3779                 return -EBUSY;
3780         else if (cmd_match(page, "resync"))
3781                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3782         else if (cmd_match(page, "recover")) {
3783                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3784                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3785         } else if (cmd_match(page, "reshape")) {
3786                 int err;
3787                 if (mddev->pers->start_reshape == NULL)
3788                         return -EINVAL;
3789                 err = mddev->pers->start_reshape(mddev);
3790                 if (err)
3791                         return err;
3792                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3793         } else {
3794                 if (cmd_match(page, "check"))
3795                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3796                 else if (!cmd_match(page, "repair"))
3797                         return -EINVAL;
3798                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3799                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3800         }
3801         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3802         md_wakeup_thread(mddev->thread);
3803         sysfs_notify_dirent_safe(mddev->sysfs_action);
3804         return len;
3805 }
3806
3807 static ssize_t
3808 mismatch_cnt_show(mddev_t *mddev, char *page)
3809 {
3810         return sprintf(page, "%llu\n",
3811                        (unsigned long long) mddev->resync_mismatches);
3812 }
3813
3814 static struct md_sysfs_entry md_scan_mode =
3815 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3816
3817
3818 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3819
3820 static ssize_t
3821 sync_min_show(mddev_t *mddev, char *page)
3822 {
3823         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3824                        mddev->sync_speed_min ? "local": "system");
3825 }
3826
3827 static ssize_t
3828 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3829 {
3830         int min;
3831         char *e;
3832         if (strncmp(buf, "system", 6)==0) {
3833                 mddev->sync_speed_min = 0;
3834                 return len;
3835         }
3836         min = simple_strtoul(buf, &e, 10);
3837         if (buf == e || (*e && *e != '\n') || min <= 0)
3838                 return -EINVAL;
3839         mddev->sync_speed_min = min;
3840         return len;
3841 }
3842
3843 static struct md_sysfs_entry md_sync_min =
3844 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3845
3846 static ssize_t
3847 sync_max_show(mddev_t *mddev, char *page)
3848 {
3849         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3850                        mddev->sync_speed_max ? "local": "system");
3851 }
3852
3853 static ssize_t
3854 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3855 {
3856         int max;
3857         char *e;
3858         if (strncmp(buf, "system", 6)==0) {
3859                 mddev->sync_speed_max = 0;
3860                 return len;
3861         }
3862         max = simple_strtoul(buf, &e, 10);
3863         if (buf == e || (*e && *e != '\n') || max <= 0)
3864                 return -EINVAL;
3865         mddev->sync_speed_max = max;
3866         return len;
3867 }
3868
3869 static struct md_sysfs_entry md_sync_max =
3870 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3871
3872 static ssize_t
3873 degraded_show(mddev_t *mddev, char *page)
3874 {
3875         return sprintf(page, "%d\n", mddev->degraded);
3876 }
3877 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3878
3879 static ssize_t
3880 sync_force_parallel_show(mddev_t *mddev, char *page)
3881 {
3882         return sprintf(page, "%d\n", mddev->parallel_resync);
3883 }
3884
3885 static ssize_t
3886 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3887 {
3888         long n;
3889
3890         if (strict_strtol(buf, 10, &n))
3891                 return -EINVAL;
3892
3893         if (n != 0 && n != 1)
3894                 return -EINVAL;
3895
3896         mddev->parallel_resync = n;
3897
3898         if (mddev->sync_thread)
3899                 wake_up(&resync_wait);
3900
3901         return len;
3902 }
3903
3904 /* force parallel resync, even with shared block devices */
3905 static struct md_sysfs_entry md_sync_force_parallel =
3906 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3907        sync_force_parallel_show, sync_force_parallel_store);
3908
3909 static ssize_t
3910 sync_speed_show(mddev_t *mddev, char *page)
3911 {
3912         unsigned long resync, dt, db;
3913         if (mddev->curr_resync == 0)
3914                 return sprintf(page, "none\n");
3915         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3916         dt = (jiffies - mddev->resync_mark) / HZ;
3917         if (!dt) dt++;
3918         db = resync - mddev->resync_mark_cnt;
3919         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3920 }
3921
3922 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3923
3924 static ssize_t
3925 sync_completed_show(mddev_t *mddev, char *page)
3926 {
3927         unsigned long long max_sectors, resync;
3928
3929         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3930                 return sprintf(page, "none\n");
3931
3932         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3933                 max_sectors = mddev->resync_max_sectors;
3934         else
3935                 max_sectors = mddev->dev_sectors;
3936
3937         resync = mddev->curr_resync_completed;
3938         return sprintf(page, "%llu / %llu\n", resync, max_sectors);
3939 }
3940
3941 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3942
3943 static ssize_t
3944 min_sync_show(mddev_t *mddev, char *page)
3945 {
3946         return sprintf(page, "%llu\n",
3947                        (unsigned long long)mddev->resync_min);
3948 }
3949 static ssize_t
3950 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3951 {
3952         unsigned long long min;
3953         if (strict_strtoull(buf, 10, &min))
3954                 return -EINVAL;
3955         if (min > mddev->resync_max)
3956                 return -EINVAL;
3957         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3958                 return -EBUSY;
3959
3960         /* Must be a multiple of chunk_size */
3961         if (mddev->chunk_sectors) {
3962                 sector_t temp = min;
3963                 if (sector_div(temp, mddev->chunk_sectors))
3964                         return -EINVAL;
3965         }
3966         mddev->resync_min = min;
3967
3968         return len;
3969 }
3970
3971 static struct md_sysfs_entry md_min_sync =
3972 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3973
3974 static ssize_t
3975 max_sync_show(mddev_t *mddev, char *page)
3976 {
3977         if (mddev->resync_max == MaxSector)
3978                 return sprintf(page, "max\n");
3979         else
3980                 return sprintf(page, "%llu\n",
3981                                (unsigned long long)mddev->resync_max);
3982 }
3983 static ssize_t
3984 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3985 {
3986         if (strncmp(buf, "max", 3) == 0)
3987                 mddev->resync_max = MaxSector;
3988         else {
3989                 unsigned long long max;
3990                 if (strict_strtoull(buf, 10, &max))
3991                         return -EINVAL;
3992                 if (max < mddev->resync_min)
3993                         return -EINVAL;
3994                 if (max < mddev->resync_max &&
3995                     mddev->ro == 0 &&
3996                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3997                         return -EBUSY;
3998
3999                 /* Must be a multiple of chunk_size */
4000                 if (mddev->chunk_sectors) {
4001                         sector_t temp = max;
4002                         if (sector_div(temp, mddev->chunk_sectors))
4003                                 return -EINVAL;
4004                 }
4005                 mddev->resync_max = max;
4006         }
4007         wake_up(&mddev->recovery_wait);
4008         return len;
4009 }
4010
4011 static struct md_sysfs_entry md_max_sync =
4012 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4013
4014 static ssize_t
4015 suspend_lo_show(mddev_t *mddev, char *page)
4016 {
4017         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4018 }
4019
4020 static ssize_t
4021 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4022 {
4023         char *e;
4024         unsigned long long new = simple_strtoull(buf, &e, 10);
4025         unsigned long long old = mddev->suspend_lo;
4026
4027         if (mddev->pers == NULL ||
4028             mddev->pers->quiesce == NULL)
4029                 return -EINVAL;
4030         if (buf == e || (*e && *e != '\n'))
4031                 return -EINVAL;
4032
4033         mddev->suspend_lo = new;
4034         if (new >= old)
4035                 /* Shrinking suspended region */
4036                 mddev->pers->quiesce(mddev, 2);
4037         else {
4038                 /* Expanding suspended region - need to wait */
4039                 mddev->pers->quiesce(mddev, 1);
4040                 mddev->pers->quiesce(mddev, 0);
4041         }
4042         return len;
4043 }
4044 static struct md_sysfs_entry md_suspend_lo =
4045 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4046
4047
4048 static ssize_t
4049 suspend_hi_show(mddev_t *mddev, char *page)
4050 {
4051         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4052 }
4053
4054 static ssize_t
4055 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4056 {
4057         char *e;
4058         unsigned long long new = simple_strtoull(buf, &e, 10);
4059         unsigned long long old = mddev->suspend_hi;
4060
4061         if (mddev->pers == NULL ||
4062             mddev->pers->quiesce == NULL)
4063                 return -EINVAL;
4064         if (buf == e || (*e && *e != '\n'))
4065                 return -EINVAL;
4066
4067         mddev->suspend_hi = new;
4068         if (new <= old)
4069                 /* Shrinking suspended region */
4070                 mddev->pers->quiesce(mddev, 2);
4071         else {
4072                 /* Expanding suspended region - need to wait */
4073                 mddev->pers->quiesce(mddev, 1);
4074                 mddev->pers->quiesce(mddev, 0);
4075         }
4076         return len;
4077 }
4078 static struct md_sysfs_entry md_suspend_hi =
4079 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4080
4081 static ssize_t
4082 reshape_position_show(mddev_t *mddev, char *page)
4083 {
4084         if (mddev->reshape_position != MaxSector)
4085                 return sprintf(page, "%llu\n",
4086                                (unsigned long long)mddev->reshape_position);
4087         strcpy(page, "none\n");
4088         return 5;
4089 }
4090
4091 static ssize_t
4092 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4093 {
4094         char *e;
4095         unsigned long long new = simple_strtoull(buf, &e, 10);
4096         if (mddev->pers)
4097                 return -EBUSY;
4098         if (buf == e || (*e && *e != '\n'))
4099                 return -EINVAL;
4100         mddev->reshape_position = new;
4101         mddev->delta_disks = 0;
4102         mddev->new_level = mddev->level;
4103         mddev->new_layout = mddev->layout;
4104         mddev->new_chunk_sectors = mddev->chunk_sectors;
4105         return len;
4106 }
4107
4108 static struct md_sysfs_entry md_reshape_position =
4109 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4110        reshape_position_store);
4111
4112 static ssize_t
4113 array_size_show(mddev_t *mddev, char *page)
4114 {
4115         if (mddev->external_size)
4116                 return sprintf(page, "%llu\n",
4117                                (unsigned long long)mddev->array_sectors/2);
4118         else
4119                 return sprintf(page, "default\n");
4120 }
4121
4122 static ssize_t
4123 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4124 {
4125         sector_t sectors;
4126
4127         if (strncmp(buf, "default", 7) == 0) {
4128                 if (mddev->pers)
4129                         sectors = mddev->pers->size(mddev, 0, 0);
4130                 else
4131                         sectors = mddev->array_sectors;
4132
4133                 mddev->external_size = 0;
4134         } else {
4135                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4136                         return -EINVAL;
4137                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4138                         return -E2BIG;
4139
4140                 mddev->external_size = 1;
4141         }
4142
4143         mddev->array_sectors = sectors;
4144         if (mddev->pers) {
4145                 set_capacity(mddev->gendisk, mddev->array_sectors);
4146                 revalidate_disk(mddev->gendisk);
4147         }
4148         return len;
4149 }
4150
4151 static struct md_sysfs_entry md_array_size =
4152 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4153        array_size_store);
4154
4155 static struct attribute *md_default_attrs[] = {
4156         &md_level.attr,
4157         &md_layout.attr,
4158         &md_raid_disks.attr,
4159         &md_chunk_size.attr,
4160         &md_size.attr,
4161         &md_resync_start.attr,
4162         &md_metadata.attr,
4163         &md_new_device.attr,
4164         &md_safe_delay.attr,
4165         &md_array_state.attr,
4166         &md_reshape_position.attr,
4167         &md_array_size.attr,
4168         &max_corr_read_errors.attr,
4169         NULL,
4170 };
4171
4172 static struct attribute *md_redundancy_attrs[] = {
4173         &md_scan_mode.attr,
4174         &md_mismatches.attr,
4175         &md_sync_min.attr,
4176         &md_sync_max.attr,
4177         &md_sync_speed.attr,
4178         &md_sync_force_parallel.attr,
4179         &md_sync_completed.attr,
4180         &md_min_sync.attr,
4181         &md_max_sync.attr,
4182         &md_suspend_lo.attr,
4183         &md_suspend_hi.attr,
4184         &md_bitmap.attr,
4185         &md_degraded.attr,
4186         NULL,
4187 };
4188 static struct attribute_group md_redundancy_group = {
4189         .name = NULL,
4190         .attrs = md_redundancy_attrs,
4191 };
4192
4193
4194 static ssize_t
4195 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4196 {
4197         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4198         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4199         ssize_t rv;
4200
4201         if (!entry->show)
4202                 return -EIO;
4203         rv = mddev_lock(mddev);
4204         if (!rv) {
4205                 rv = entry->show(mddev, page);
4206                 mddev_unlock(mddev);
4207         }
4208         return rv;
4209 }
4210
4211 static ssize_t
4212 md_attr_store(struct kobject *kobj, struct attribute *attr,
4213               const char *page, size_t length)
4214 {
4215         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4216         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4217         ssize_t rv;
4218
4219         if (!entry->store)
4220                 return -EIO;
4221         if (!capable(CAP_SYS_ADMIN))
4222                 return -EACCES;
4223         rv = mddev_lock(mddev);
4224         if (mddev->hold_active == UNTIL_IOCTL)
4225                 mddev->hold_active = 0;
4226         if (!rv) {
4227                 rv = entry->store(mddev, page, length);
4228                 mddev_unlock(mddev);
4229         }
4230         return rv;
4231 }
4232
4233 static void md_free(struct kobject *ko)
4234 {
4235         mddev_t *mddev = container_of(ko, mddev_t, kobj);
4236
4237         if (mddev->sysfs_state)
4238                 sysfs_put(mddev->sysfs_state);
4239
4240         if (mddev->gendisk) {
4241                 del_gendisk(mddev->gendisk);
4242                 put_disk(mddev->gendisk);
4243         }
4244         if (mddev->queue)
4245                 blk_cleanup_queue(mddev->queue);
4246
4247         kfree(mddev);
4248 }
4249
4250 static const struct sysfs_ops md_sysfs_ops = {
4251         .show   = md_attr_show,
4252         .store  = md_attr_store,
4253 };
4254 static struct kobj_type md_ktype = {
4255         .release        = md_free,
4256         .sysfs_ops      = &md_sysfs_ops,
4257         .default_attrs  = md_default_attrs,
4258 };
4259
4260 int mdp_major = 0;
4261
4262 static void mddev_delayed_delete(struct work_struct *ws)
4263 {
4264         mddev_t *mddev = container_of(ws, mddev_t, del_work);
4265
4266         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4267         kobject_del(&mddev->kobj);
4268         kobject_put(&mddev->kobj);
4269 }
4270
4271 static int md_alloc(dev_t dev, char *name)
4272 {
4273         static DEFINE_MUTEX(disks_mutex);
4274         mddev_t *mddev = mddev_find(dev);
4275         struct gendisk *disk;
4276         int partitioned;
4277         int shift;
4278         int unit;
4279         int error;
4280
4281         if (!mddev)
4282                 return -ENODEV;
4283
4284         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4285         shift = partitioned ? MdpMinorShift : 0;
4286         unit = MINOR(mddev->unit) >> shift;
4287
4288         /* wait for any previous instance of this device to be
4289          * completely removed (mddev_delayed_delete).
4290          */
4291         flush_workqueue(md_misc_wq);
4292
4293         mutex_lock(&disks_mutex);
4294         error = -EEXIST;
4295         if (mddev->gendisk)
4296                 goto abort;
4297
4298         if (name) {
4299                 /* Need to ensure that 'name' is not a duplicate.
4300                  */
4301                 mddev_t *mddev2;
4302                 spin_lock(&all_mddevs_lock);
4303
4304                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4305                         if (mddev2->gendisk &&
4306                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4307                                 spin_unlock(&all_mddevs_lock);
4308                                 goto abort;
4309                         }
4310                 spin_unlock(&all_mddevs_lock);
4311         }
4312
4313         error = -ENOMEM;
4314         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4315         if (!mddev->queue)
4316                 goto abort;
4317         mddev->queue->queuedata = mddev;
4318
4319         blk_queue_make_request(mddev->queue, md_make_request);
4320
4321         disk = alloc_disk(1 << shift);
4322         if (!disk) {
4323                 blk_cleanup_queue(mddev->queue);
4324                 mddev->queue = NULL;
4325                 goto abort;
4326         }
4327         disk->major = MAJOR(mddev->unit);
4328         disk->first_minor = unit << shift;
4329         if (name)
4330                 strcpy(disk->disk_name, name);
4331         else if (partitioned)
4332                 sprintf(disk->disk_name, "md_d%d", unit);
4333         else
4334                 sprintf(disk->disk_name, "md%d", unit);
4335         disk->fops = &md_fops;
4336         disk->private_data = mddev;
4337         disk->queue = mddev->queue;
4338         /* Allow extended partitions.  This makes the
4339          * 'mdp' device redundant, but we can't really
4340          * remove it now.
4341          */
4342         disk->flags |= GENHD_FL_EXT_DEVT;
4343         add_disk(disk);
4344         mddev->gendisk = disk;
4345         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4346                                      &disk_to_dev(disk)->kobj, "%s", "md");
4347         if (error) {
4348                 /* This isn't possible, but as kobject_init_and_add is marked
4349                  * __must_check, we must do something with the result
4350                  */
4351                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4352                        disk->disk_name);
4353                 error = 0;
4354         }
4355         if (mddev->kobj.sd &&
4356             sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4357                 printk(KERN_DEBUG "pointless warning\n");
4358
4359         blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4360  abort:
4361         mutex_unlock(&disks_mutex);
4362         if (!error && mddev->kobj.sd) {
4363                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4364                 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4365         }
4366         mddev_put(mddev);
4367         return error;
4368 }
4369
4370 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4371 {
4372         md_alloc(dev, NULL);
4373         return NULL;
4374 }
4375
4376 static int add_named_array(const char *val, struct kernel_param *kp)
4377 {
4378         /* val must be "md_*" where * is not all digits.
4379          * We allocate an array with a large free minor number, and
4380          * set the name to val.  val must not already be an active name.
4381          */
4382         int len = strlen(val);
4383         char buf[DISK_NAME_LEN];
4384
4385         while (len && val[len-1] == '\n')
4386                 len--;
4387         if (len >= DISK_NAME_LEN)
4388                 return -E2BIG;
4389         strlcpy(buf, val, len+1);
4390         if (strncmp(buf, "md_", 3) != 0)
4391                 return -EINVAL;
4392         return md_alloc(0, buf);
4393 }
4394
4395 static void md_safemode_timeout(unsigned long data)
4396 {
4397         mddev_t *mddev = (mddev_t *) data;
4398
4399         if (!atomic_read(&mddev->writes_pending)) {
4400                 mddev->safemode = 1;
4401                 if (mddev->external)
4402                         sysfs_notify_dirent_safe(mddev->sysfs_state);
4403         }
4404         md_wakeup_thread(mddev->thread);
4405 }
4406
4407 static int start_dirty_degraded;
4408
4409 int md_run(mddev_t *mddev)
4410 {
4411         int err;
4412         mdk_rdev_t *rdev;
4413         struct mdk_personality *pers;
4414
4415         if (list_empty(&mddev->disks))
4416                 /* cannot run an array with no devices.. */
4417                 return -EINVAL;
4418
4419         if (mddev->pers)
4420                 return -EBUSY;
4421         /* Cannot run until previous stop completes properly */
4422         if (mddev->sysfs_active)
4423                 return -EBUSY;
4424
4425         /*
4426          * Analyze all RAID superblock(s)
4427          */
4428         if (!mddev->raid_disks) {
4429                 if (!mddev->persistent)
4430                         return -EINVAL;
4431                 analyze_sbs(mddev);
4432         }
4433
4434         if (mddev->level != LEVEL_NONE)
4435                 request_module("md-level-%d", mddev->level);
4436         else if (mddev->clevel[0])
4437                 request_module("md-%s", mddev->clevel);
4438
4439         /*
4440          * Drop all container device buffers, from now on
4441          * the only valid external interface is through the md
4442          * device.
4443          */
4444         list_for_each_entry(rdev, &mddev->disks, same_set) {
4445                 if (test_bit(Faulty, &rdev->flags))
4446                         continue;
4447                 sync_blockdev(rdev->bdev);
4448                 invalidate_bdev(rdev->bdev);
4449
4450                 /* perform some consistency tests on the device.
4451                  * We don't want the data to overlap the metadata,
4452                  * Internal Bitmap issues have been handled elsewhere.
4453                  */
4454                 if (rdev->meta_bdev) {
4455                         /* Nothing to check */;
4456                 } else if (rdev->data_offset < rdev->sb_start) {
4457                         if (mddev->dev_sectors &&
4458                             rdev->data_offset + mddev->dev_sectors
4459                             > rdev->sb_start) {
4460                                 printk("md: %s: data overlaps metadata\n",
4461                                        mdname(mddev));
4462                                 return -EINVAL;
4463                         }
4464                 } else {
4465                         if (rdev->sb_start + rdev->sb_size/512
4466                             > rdev->data_offset) {
4467                                 printk("md: %s: metadata overlaps data\n",
4468                                        mdname(mddev));
4469                                 return -EINVAL;
4470                         }
4471                 }
4472                 sysfs_notify_dirent_safe(rdev->sysfs_state);
4473         }
4474
4475         if (mddev->bio_set == NULL)
4476                 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4477
4478         spin_lock(&pers_lock);
4479         pers = find_pers(mddev->level, mddev->clevel);
4480         if (!pers || !try_module_get(pers->owner)) {
4481                 spin_unlock(&pers_lock);
4482                 if (mddev->level != LEVEL_NONE)
4483                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4484                                mddev->level);
4485                 else
4486                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4487                                mddev->clevel);
4488                 return -EINVAL;
4489         }
4490         mddev->pers = pers;
4491         spin_unlock(&pers_lock);
4492         if (mddev->level != pers->level) {
4493                 mddev->level = pers->level;
4494                 mddev->new_level = pers->level;
4495         }
4496         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4497
4498         if (mddev->reshape_position != MaxSector &&
4499             pers->start_reshape == NULL) {
4500                 /* This personality cannot handle reshaping... */
4501                 mddev->pers = NULL;
4502                 module_put(pers->owner);
4503                 return -EINVAL;
4504         }
4505
4506         if (pers->sync_request) {
4507                 /* Warn if this is a potentially silly
4508                  * configuration.
4509                  */
4510                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4511                 mdk_rdev_t *rdev2;
4512                 int warned = 0;
4513
4514                 list_for_each_entry(rdev, &mddev->disks, same_set)
4515                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4516                                 if (rdev < rdev2 &&
4517                                     rdev->bdev->bd_contains ==
4518                                     rdev2->bdev->bd_contains) {
4519                                         printk(KERN_WARNING
4520                                                "%s: WARNING: %s appears to be"
4521                                                " on the same physical disk as"
4522                                                " %s.\n",
4523                                                mdname(mddev),
4524                                                bdevname(rdev->bdev,b),
4525                                                bdevname(rdev2->bdev,b2));
4526                                         warned = 1;
4527                                 }
4528                         }
4529
4530                 if (warned)
4531                         printk(KERN_WARNING
4532                                "True protection against single-disk"
4533                                " failure might be compromised.\n");
4534         }
4535
4536         mddev->recovery = 0;
4537         /* may be over-ridden by personality */
4538         mddev->resync_max_sectors = mddev->dev_sectors;
4539
4540         mddev->ok_start_degraded = start_dirty_degraded;
4541
4542         if (start_readonly && mddev->ro == 0)
4543                 mddev->ro = 2; /* read-only, but switch on first write */
4544
4545         err = mddev->pers->run(mddev);
4546         if (err)
4547                 printk(KERN_ERR "md: pers->run() failed ...\n");
4548         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4549                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4550                           " but 'external_size' not in effect?\n", __func__);
4551                 printk(KERN_ERR
4552                        "md: invalid array_size %llu > default size %llu\n",
4553                        (unsigned long long)mddev->array_sectors / 2,
4554                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4555                 err = -EINVAL;
4556                 mddev->pers->stop(mddev);
4557         }
4558         if (err == 0 && mddev->pers->sync_request) {
4559                 err = bitmap_create(mddev);
4560                 if (err) {
4561                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4562                                mdname(mddev), err);
4563                         mddev->pers->stop(mddev);
4564                 }
4565         }
4566         if (err) {
4567                 module_put(mddev->pers->owner);
4568                 mddev->pers = NULL;
4569                 bitmap_destroy(mddev);
4570                 return err;
4571         }
4572         if (mddev->pers->sync_request) {
4573                 if (mddev->kobj.sd &&
4574                     sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4575                         printk(KERN_WARNING
4576                                "md: cannot register extra attributes for %s\n",
4577                                mdname(mddev));
4578                 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4579         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4580                 mddev->ro = 0;
4581
4582         atomic_set(&mddev->writes_pending,0);
4583         atomic_set(&mddev->max_corr_read_errors,
4584                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4585         mddev->safemode = 0;
4586         mddev->safemode_timer.function = md_safemode_timeout;
4587         mddev->safemode_timer.data = (unsigned long) mddev;
4588         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4589         mddev->in_sync = 1;
4590         smp_wmb();
4591         mddev->ready = 1;
4592         list_for_each_entry(rdev, &mddev->disks, same_set)
4593                 if (rdev->raid_disk >= 0) {
4594                         char nm[20];
4595                         sprintf(nm, "rd%d", rdev->raid_disk);
4596                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4597                                 /* failure here is OK */;
4598                 }
4599
4600         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4601
4602         if (mddev->flags)
4603                 md_update_sb(mddev, 0);
4604
4605         md_wakeup_thread(mddev->thread);
4606         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4607
4608         md_new_event(mddev);
4609         sysfs_notify_dirent_safe(mddev->sysfs_state);
4610         sysfs_notify_dirent_safe(mddev->sysfs_action);
4611         sysfs_notify(&mddev->kobj, NULL, "degraded");
4612         return 0;
4613 }
4614 EXPORT_SYMBOL_GPL(md_run);
4615
4616 static int do_md_run(mddev_t *mddev)
4617 {
4618         int err;
4619
4620         err = md_run(mddev);
4621         if (err)
4622                 goto out;
4623         err = bitmap_load(mddev);
4624         if (err) {
4625                 bitmap_destroy(mddev);
4626                 goto out;
4627         }
4628         set_capacity(mddev->gendisk, mddev->array_sectors);
4629         revalidate_disk(mddev->gendisk);
4630         mddev->changed = 1;
4631         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4632 out:
4633         return err;
4634 }
4635
4636 static int restart_array(mddev_t *mddev)
4637 {
4638         struct gendisk *disk = mddev->gendisk;
4639
4640         /* Complain if it has no devices */
4641         if (list_empty(&mddev->disks))
4642                 return -ENXIO;
4643         if (!mddev->pers)
4644                 return -EINVAL;
4645         if (!mddev->ro)
4646                 return -EBUSY;
4647         mddev->safemode = 0;
4648         mddev->ro = 0;
4649         set_disk_ro(disk, 0);
4650         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4651                 mdname(mddev));
4652         /* Kick recovery or resync if necessary */
4653         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4654         md_wakeup_thread(mddev->thread);
4655         md_wakeup_thread(mddev->sync_thread);
4656         sysfs_notify_dirent_safe(mddev->sysfs_state);
4657         return 0;
4658 }
4659
4660 /* similar to deny_write_access, but accounts for our holding a reference
4661  * to the file ourselves */
4662 static int deny_bitmap_write_access(struct file * file)
4663 {
4664         struct inode *inode = file->f_mapping->host;
4665
4666         spin_lock(&inode->i_lock);
4667         if (atomic_read(&inode->i_writecount) > 1) {
4668                 spin_unlock(&inode->i_lock);
4669                 return -ETXTBSY;
4670         }
4671         atomic_set(&inode->i_writecount, -1);
4672         spin_unlock(&inode->i_lock);
4673
4674         return 0;
4675 }
4676
4677 void restore_bitmap_write_access(struct file *file)
4678 {
4679         struct inode *inode = file->f_mapping->host;
4680
4681         spin_lock(&inode->i_lock);
4682         atomic_set(&inode->i_writecount, 1);
4683         spin_unlock(&inode->i_lock);
4684 }
4685
4686 static void md_clean(mddev_t *mddev)
4687 {
4688         mddev->array_sectors = 0;
4689         mddev->external_size = 0;
4690         mddev->dev_sectors = 0;
4691         mddev->raid_disks = 0;
4692         mddev->recovery_cp = 0;
4693         mddev->resync_min = 0;
4694         mddev->resync_max = MaxSector;
4695         mddev->reshape_position = MaxSector;
4696         mddev->external = 0;
4697         mddev->persistent = 0;
4698         mddev->level = LEVEL_NONE;
4699         mddev->clevel[0] = 0;
4700         mddev->flags = 0;
4701         mddev->ro = 0;
4702         mddev->metadata_type[0] = 0;
4703         mddev->chunk_sectors = 0;
4704         mddev->ctime = mddev->utime = 0;
4705         mddev->layout = 0;
4706         mddev->max_disks = 0;
4707         mddev->events = 0;
4708         mddev->can_decrease_events = 0;
4709         mddev->delta_disks = 0;
4710         mddev->new_level = LEVEL_NONE;
4711         mddev->new_layout = 0;
4712         mddev->new_chunk_sectors = 0;
4713         mddev->curr_resync = 0;
4714         mddev->resync_mismatches = 0;
4715         mddev->suspend_lo = mddev->suspend_hi = 0;
4716         mddev->sync_speed_min = mddev->sync_speed_max = 0;
4717         mddev->recovery = 0;
4718         mddev->in_sync = 0;
4719         mddev->changed = 0;
4720         mddev->degraded = 0;
4721         mddev->safemode = 0;
4722         mddev->bitmap_info.offset = 0;
4723         mddev->bitmap_info.default_offset = 0;
4724         mddev->bitmap_info.chunksize = 0;
4725         mddev->bitmap_info.daemon_sleep = 0;
4726         mddev->bitmap_info.max_write_behind = 0;
4727         mddev->plug = NULL;
4728 }
4729
4730 static void __md_stop_writes(mddev_t *mddev)
4731 {
4732         if (mddev->sync_thread) {
4733                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4734                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4735                 reap_sync_thread(mddev);
4736         }
4737
4738         del_timer_sync(&mddev->safemode_timer);
4739
4740         bitmap_flush(mddev);
4741         md_super_wait(mddev);
4742
4743         if (!mddev->in_sync || mddev->flags) {
4744                 /* mark array as shutdown cleanly */
4745                 mddev->in_sync = 1;
4746                 md_update_sb(mddev, 1);
4747         }
4748 }
4749
4750 void md_stop_writes(mddev_t *mddev)
4751 {
4752         mddev_lock(mddev);
4753         __md_stop_writes(mddev);
4754         mddev_unlock(mddev);
4755 }
4756 EXPORT_SYMBOL_GPL(md_stop_writes);
4757
4758 void md_stop(mddev_t *mddev)
4759 {
4760         mddev->ready = 0;
4761         mddev->pers->stop(mddev);
4762         if (mddev->pers->sync_request && mddev->to_remove == NULL)
4763                 mddev->to_remove = &md_redundancy_group;
4764         module_put(mddev->pers->owner);
4765         mddev->pers = NULL;
4766         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4767 }
4768 EXPORT_SYMBOL_GPL(md_stop);
4769
4770 static int md_set_readonly(mddev_t *mddev, int is_open)
4771 {
4772         int err = 0;
4773         mutex_lock(&mddev->open_mutex);
4774         if (atomic_read(&mddev->openers) > is_open) {
4775                 printk("md: %s still in use.\n",mdname(mddev));
4776                 err = -EBUSY;
4777                 goto out;
4778         }
4779         if (mddev->pers) {
4780                 __md_stop_writes(mddev);
4781
4782                 err  = -ENXIO;
4783                 if (mddev->ro==1)
4784                         goto out;
4785                 mddev->ro = 1;
4786                 set_disk_ro(mddev->gendisk, 1);
4787                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4788                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4789                 err = 0;
4790         }
4791 out:
4792         mutex_unlock(&mddev->open_mutex);
4793         return err;
4794 }
4795
4796 /* mode:
4797  *   0 - completely stop and dis-assemble array
4798  *   2 - stop but do not disassemble array
4799  */
4800 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4801 {
4802         struct gendisk *disk = mddev->gendisk;
4803         mdk_rdev_t *rdev;
4804
4805         mutex_lock(&mddev->open_mutex);
4806         if (atomic_read(&mddev->openers) > is_open ||
4807             mddev->sysfs_active) {
4808                 printk("md: %s still in use.\n",mdname(mddev));
4809                 mutex_unlock(&mddev->open_mutex);
4810                 return -EBUSY;
4811         }
4812
4813         if (mddev->pers) {
4814                 if (mddev->ro)
4815                         set_disk_ro(disk, 0);
4816
4817                 __md_stop_writes(mddev);
4818                 md_stop(mddev);
4819                 mddev->queue->merge_bvec_fn = NULL;
4820                 mddev->queue->unplug_fn = NULL;
4821                 mddev->queue->backing_dev_info.congested_fn = NULL;
4822
4823                 /* tell userspace to handle 'inactive' */
4824                 sysfs_notify_dirent_safe(mddev->sysfs_state);
4825
4826                 list_for_each_entry(rdev, &mddev->disks, same_set)
4827                         if (rdev->raid_disk >= 0) {
4828                                 char nm[20];
4829                                 sprintf(nm, "rd%d", rdev->raid_disk);
4830                                 sysfs_remove_link(&mddev->kobj, nm);
4831                         }
4832
4833                 set_capacity(disk, 0);
4834                 mutex_unlock(&mddev->open_mutex);
4835                 mddev->changed = 1;
4836                 revalidate_disk(disk);
4837
4838                 if (mddev->ro)
4839                         mddev->ro = 0;
4840         } else
4841                 mutex_unlock(&mddev->open_mutex);
4842         /*
4843          * Free resources if final stop
4844          */
4845         if (mode == 0) {
4846                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4847
4848                 bitmap_destroy(mddev);
4849                 if (mddev->bitmap_info.file) {
4850                         restore_bitmap_write_access(mddev->bitmap_info.file);
4851                         fput(mddev->bitmap_info.file);
4852                         mddev->bitmap_info.file = NULL;
4853                 }
4854                 mddev->bitmap_info.offset = 0;
4855
4856                 export_array(mddev);
4857
4858                 md_clean(mddev);
4859                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4860                 if (mddev->hold_active == UNTIL_STOP)
4861                         mddev->hold_active = 0;
4862         }
4863         blk_integrity_unregister(disk);
4864         md_new_event(mddev);
4865         sysfs_notify_dirent_safe(mddev->sysfs_state);
4866         return 0;
4867 }
4868
4869 #ifndef MODULE
4870 static void autorun_array(mddev_t *mddev)
4871 {
4872         mdk_rdev_t *rdev;
4873         int err;
4874
4875         if (list_empty(&mddev->disks))
4876                 return;
4877
4878         printk(KERN_INFO "md: running: ");
4879
4880         list_for_each_entry(rdev, &mddev->disks, same_set) {
4881                 char b[BDEVNAME_SIZE];
4882                 printk("<%s>", bdevname(rdev->bdev,b));
4883         }
4884         printk("\n");
4885
4886         err = do_md_run(mddev);
4887         if (err) {
4888                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4889                 do_md_stop(mddev, 0, 0);
4890         }
4891 }
4892
4893 /*
4894  * lets try to run arrays based on all disks that have arrived
4895  * until now. (those are in pending_raid_disks)
4896  *
4897  * the method: pick the first pending disk, collect all disks with
4898  * the same UUID, remove all from the pending list and put them into
4899  * the 'same_array' list. Then order this list based on superblock
4900  * update time (freshest comes first), kick out 'old' disks and
4901  * compare superblocks. If everything's fine then run it.
4902  *
4903  * If "unit" is allocated, then bump its reference count
4904  */
4905 static void autorun_devices(int part)
4906 {
4907         mdk_rdev_t *rdev0, *rdev, *tmp;
4908         mddev_t *mddev;
4909         char b[BDEVNAME_SIZE];
4910
4911         printk(KERN_INFO "md: autorun ...\n");
4912         while (!list_empty(&pending_raid_disks)) {
4913                 int unit;
4914                 dev_t dev;
4915                 LIST_HEAD(candidates);
4916                 rdev0 = list_entry(pending_raid_disks.next,
4917                                          mdk_rdev_t, same_set);
4918
4919                 printk(KERN_INFO "md: considering %s ...\n",
4920                         bdevname(rdev0->bdev,b));
4921                 INIT_LIST_HEAD(&candidates);
4922                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4923                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4924                                 printk(KERN_INFO "md:  adding %s ...\n",
4925                                         bdevname(rdev->bdev,b));
4926                                 list_move(&rdev->same_set, &candidates);
4927                         }
4928                 /*
4929                  * now we have a set of devices, with all of them having
4930                  * mostly sane superblocks. It's time to allocate the
4931                  * mddev.
4932                  */
4933                 if (part) {
4934                         dev = MKDEV(mdp_major,
4935                                     rdev0->preferred_minor << MdpMinorShift);
4936                         unit = MINOR(dev) >> MdpMinorShift;
4937                 } else {
4938                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4939                         unit = MINOR(dev);
4940                 }
4941                 if (rdev0->preferred_minor != unit) {
4942                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4943                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4944                         break;
4945                 }
4946
4947                 md_probe(dev, NULL, NULL);
4948                 mddev = mddev_find(dev);
4949                 if (!mddev || !mddev->gendisk) {
4950                         if (mddev)
4951                                 mddev_put(mddev);
4952                         printk(KERN_ERR
4953                                 "md: cannot allocate memory for md drive.\n");
4954                         break;
4955                 }
4956                 if (mddev_lock(mddev))
4957                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4958                                mdname(mddev));
4959                 else if (mddev->raid_disks || mddev->major_version
4960                          || !list_empty(&mddev->disks)) {
4961                         printk(KERN_WARNING
4962                                 "md: %s already running, cannot run %s\n",
4963                                 mdname(mddev), bdevname(rdev0->bdev,b));
4964                         mddev_unlock(mddev);
4965                 } else {
4966                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4967                         mddev->persistent = 1;
4968                         rdev_for_each_list(rdev, tmp, &candidates) {
4969                                 list_del_init(&rdev->same_set);
4970                                 if (bind_rdev_to_array(rdev, mddev))
4971                                         export_rdev(rdev);
4972                         }
4973                         autorun_array(mddev);
4974                         mddev_unlock(mddev);
4975                 }
4976                 /* on success, candidates will be empty, on error
4977                  * it won't...
4978                  */
4979                 rdev_for_each_list(rdev, tmp, &candidates) {
4980                         list_del_init(&rdev->same_set);
4981                         export_rdev(rdev);
4982                 }
4983                 mddev_put(mddev);
4984         }
4985         printk(KERN_INFO "md: ... autorun DONE.\n");
4986 }
4987 #endif /* !MODULE */
4988
4989 static int get_version(void __user * arg)
4990 {
4991         mdu_version_t ver;
4992
4993         ver.major = MD_MAJOR_VERSION;
4994         ver.minor = MD_MINOR_VERSION;
4995         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4996
4997         if (copy_to_user(arg, &ver, sizeof(ver)))
4998                 return -EFAULT;
4999
5000         return 0;
5001 }
5002
5003 static int get_array_info(mddev_t * mddev, void __user * arg)
5004 {
5005         mdu_array_info_t info;
5006         int nr,working,insync,failed,spare;
5007         mdk_rdev_t *rdev;
5008
5009         nr=working=insync=failed=spare=0;
5010         list_for_each_entry(rdev, &mddev->disks, same_set) {
5011                 nr++;
5012                 if (test_bit(Faulty, &rdev->flags))
5013                         failed++;
5014                 else {
5015                         working++;
5016                         if (test_bit(In_sync, &rdev->flags))
5017                                 insync++;
5018                         else
5019                                 spare++;
5020                 }
5021         }
5022
5023         info.major_version = mddev->major_version;
5024         info.minor_version = mddev->minor_version;
5025         info.patch_version = MD_PATCHLEVEL_VERSION;
5026         info.ctime         = mddev->ctime;
5027         info.level         = mddev->level;
5028         info.size          = mddev->dev_sectors / 2;
5029         if (info.size != mddev->dev_sectors / 2) /* overflow */
5030                 info.size = -1;
5031         info.nr_disks      = nr;
5032         info.raid_disks    = mddev->raid_disks;
5033         info.md_minor      = mddev->md_minor;
5034         info.not_persistent= !mddev->persistent;
5035
5036         info.utime         = mddev->utime;
5037         info.state         = 0;
5038         if (mddev->in_sync)
5039                 info.state = (1<<MD_SB_CLEAN);
5040         if (mddev->bitmap && mddev->bitmap_info.offset)
5041                 info.state = (1<<MD_SB_BITMAP_PRESENT);
5042         info.active_disks  = insync;
5043         info.working_disks = working;
5044         info.failed_disks  = failed;
5045         info.spare_disks   = spare;
5046
5047         info.layout        = mddev->layout;
5048         info.chunk_size    = mddev->chunk_sectors << 9;
5049
5050         if (copy_to_user(arg, &info, sizeof(info)))
5051                 return -EFAULT;
5052
5053         return 0;
5054 }
5055
5056 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5057 {
5058         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5059         char *ptr, *buf = NULL;
5060         int err = -ENOMEM;
5061
5062         if (md_allow_write(mddev))
5063                 file = kmalloc(sizeof(*file), GFP_NOIO);
5064         else
5065                 file = kmalloc(sizeof(*file), GFP_KERNEL);
5066
5067         if (!file)
5068                 goto out;
5069
5070         /* bitmap disabled, zero the first byte and copy out */
5071         if (!mddev->bitmap || !mddev->bitmap->file) {
5072                 file->pathname[0] = '\0';
5073                 goto copy_out;
5074         }
5075
5076         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5077         if (!buf)
5078                 goto out;
5079
5080         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5081         if (IS_ERR(ptr))
5082                 goto out;
5083
5084         strcpy(file->pathname, ptr);
5085
5086 copy_out:
5087         err = 0;
5088         if (copy_to_user(arg, file, sizeof(*file)))
5089                 err = -EFAULT;
5090 out:
5091         kfree(buf);
5092         kfree(file);
5093         return err;
5094 }
5095
5096 static int get_disk_info(mddev_t * mddev, void __user * arg)
5097 {
5098         mdu_disk_info_t info;
5099         mdk_rdev_t *rdev;
5100
5101         if (copy_from_user(&info, arg, sizeof(info)))
5102                 return -EFAULT;
5103
5104         rdev = find_rdev_nr(mddev, info.number);
5105         if (rdev) {
5106                 info.major = MAJOR(rdev->bdev->bd_dev);
5107                 info.minor = MINOR(rdev->bdev->bd_dev);
5108                 info.raid_disk = rdev->raid_disk;
5109                 info.state = 0;
5110                 if (test_bit(Faulty, &rdev->flags))
5111                         info.state |= (1<<MD_DISK_FAULTY);
5112                 else if (test_bit(In_sync, &rdev->flags)) {
5113                         info.state |= (1<<MD_DISK_ACTIVE);
5114                         info.state |= (1<<MD_DISK_SYNC);
5115                 }
5116                 if (test_bit(WriteMostly, &rdev->flags))
5117                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5118         } else {
5119                 info.major = info.minor = 0;
5120                 info.raid_disk = -1;
5121                 info.state = (1<<MD_DISK_REMOVED);
5122         }
5123
5124         if (copy_to_user(arg, &info, sizeof(info)))
5125                 return -EFAULT;
5126
5127         return 0;
5128 }
5129
5130 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5131 {
5132         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5133         mdk_rdev_t *rdev;
5134         dev_t dev = MKDEV(info->major,info->minor);
5135
5136         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5137                 return -EOVERFLOW;
5138
5139         if (!mddev->raid_disks) {
5140                 int err;
5141                 /* expecting a device which has a superblock */
5142                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5143                 if (IS_ERR(rdev)) {
5144                         printk(KERN_WARNING
5145                                 "md: md_import_device returned %ld\n",
5146                                 PTR_ERR(rdev));
5147                         return PTR_ERR(rdev);
5148                 }
5149                 if (!list_empty(&mddev->disks)) {
5150                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5151                                                         mdk_rdev_t, same_set);
5152                         err = super_types[mddev->major_version]
5153                                 .load_super(rdev, rdev0, mddev->minor_version);
5154                         if (err < 0) {
5155                                 printk(KERN_WARNING
5156                                         "md: %s has different UUID to %s\n",
5157                                         bdevname(rdev->bdev,b),
5158                                         bdevname(rdev0->bdev,b2));
5159                                 export_rdev(rdev);
5160                                 return -EINVAL;
5161                         }
5162                 }
5163                 err = bind_rdev_to_array(rdev, mddev);
5164                 if (err)
5165                         export_rdev(rdev);
5166                 return err;
5167         }
5168
5169         /*
5170          * add_new_disk can be used once the array is assembled
5171          * to add "hot spares".  They must already have a superblock
5172          * written
5173          */
5174         if (mddev->pers) {
5175                 int err;
5176                 if (!mddev->pers->hot_add_disk) {
5177                         printk(KERN_WARNING
5178                                 "%s: personality does not support diskops!\n",
5179                                mdname(mddev));
5180                         return -EINVAL;
5181                 }
5182                 if (mddev->persistent)
5183                         rdev = md_import_device(dev, mddev->major_version,
5184                                                 mddev->minor_version);
5185                 else
5186                         rdev = md_import_device(dev, -1, -1);
5187                 if (IS_ERR(rdev)) {
5188                         printk(KERN_WARNING
5189                                 "md: md_import_device returned %ld\n",
5190                                 PTR_ERR(rdev));
5191                         return PTR_ERR(rdev);
5192                 }
5193                 /* set saved_raid_disk if appropriate */
5194                 if (!mddev->persistent) {
5195                         if (info->state & (1<<MD_DISK_SYNC)  &&
5196                             info->raid_disk < mddev->raid_disks) {
5197                                 rdev->raid_disk = info->raid_disk;
5198                                 set_bit(In_sync, &rdev->flags);
5199                         } else
5200                                 rdev->raid_disk = -1;
5201                 } else
5202                         super_types[mddev->major_version].
5203                                 validate_super(mddev, rdev);
5204                 if (test_bit(In_sync, &rdev->flags))
5205                         rdev->saved_raid_disk = rdev->raid_disk;
5206                 else
5207                         rdev->saved_raid_disk = -1;
5208
5209                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5210                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5211                         set_bit(WriteMostly, &rdev->flags);
5212                 else
5213                         clear_bit(WriteMostly, &rdev->flags);
5214
5215                 rdev->raid_disk = -1;
5216                 err = bind_rdev_to_array(rdev, mddev);
5217                 if (!err && !mddev->pers->hot_remove_disk) {
5218                         /* If there is hot_add_disk but no hot_remove_disk
5219                          * then added disks for geometry changes,
5220                          * and should be added immediately.
5221                          */
5222                         super_types[mddev->major_version].
5223                                 validate_super(mddev, rdev);
5224                         err = mddev->pers->hot_add_disk(mddev, rdev);
5225                         if (err)
5226                                 unbind_rdev_from_array(rdev);
5227                 }
5228                 if (err)
5229                         export_rdev(rdev);
5230                 else
5231                         sysfs_notify_dirent_safe(rdev->sysfs_state);
5232
5233                 md_update_sb(mddev, 1);
5234                 if (mddev->degraded)
5235                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5236                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5237                 md_wakeup_thread(mddev->thread);
5238                 return err;
5239         }
5240
5241         /* otherwise, add_new_disk is only allowed
5242          * for major_version==0 superblocks
5243          */
5244         if (mddev->major_version != 0) {
5245                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5246                        mdname(mddev));
5247                 return -EINVAL;
5248         }
5249
5250         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5251                 int err;
5252                 rdev = md_import_device(dev, -1, 0);
5253                 if (IS_ERR(rdev)) {
5254                         printk(KERN_WARNING
5255                                 "md: error, md_import_device() returned %ld\n",
5256                                 PTR_ERR(rdev));
5257                         return PTR_ERR(rdev);
5258                 }
5259                 rdev->desc_nr = info->number;
5260                 if (info->raid_disk < mddev->raid_disks)
5261                         rdev->raid_disk = info->raid_disk;
5262                 else
5263                         rdev->raid_disk = -1;
5264
5265                 if (rdev->raid_disk < mddev->raid_disks)
5266                         if (info->state & (1<<MD_DISK_SYNC))
5267                                 set_bit(In_sync, &rdev->flags);
5268
5269                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5270                         set_bit(WriteMostly, &rdev->flags);
5271
5272                 if (!mddev->persistent) {
5273                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5274                         rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5275                 } else
5276                         rdev->sb_start = calc_dev_sboffset(rdev);
5277                 rdev->sectors = rdev->sb_start;
5278
5279                 err = bind_rdev_to_array(rdev, mddev);
5280                 if (err) {
5281                         export_rdev(rdev);
5282                         return err;
5283                 }
5284         }
5285
5286         return 0;
5287 }
5288
5289 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5290 {
5291         char b[BDEVNAME_SIZE];
5292         mdk_rdev_t *rdev;
5293
5294         rdev = find_rdev(mddev, dev);
5295         if (!rdev)
5296                 return -ENXIO;
5297
5298         if (rdev->raid_disk >= 0)
5299                 goto busy;
5300
5301         kick_rdev_from_array(rdev);
5302         md_update_sb(mddev, 1);
5303         md_new_event(mddev);
5304
5305         return 0;
5306 busy:
5307         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5308                 bdevname(rdev->bdev,b), mdname(mddev));
5309         return -EBUSY;
5310 }
5311
5312 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5313 {
5314         char b[BDEVNAME_SIZE];
5315         int err;
5316         mdk_rdev_t *rdev;
5317
5318         if (!mddev->pers)
5319                 return -ENODEV;
5320
5321         if (mddev->major_version != 0) {
5322                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5323                         " version-0 superblocks.\n",
5324                         mdname(mddev));
5325                 return -EINVAL;
5326         }
5327         if (!mddev->pers->hot_add_disk) {
5328                 printk(KERN_WARNING
5329                         "%s: personality does not support diskops!\n",
5330                         mdname(mddev));
5331                 return -EINVAL;
5332         }
5333
5334         rdev = md_import_device(dev, -1, 0);
5335         if (IS_ERR(rdev)) {
5336                 printk(KERN_WARNING
5337                         "md: error, md_import_device() returned %ld\n",
5338                         PTR_ERR(rdev));
5339                 return -EINVAL;
5340         }
5341
5342         if (mddev->persistent)
5343                 rdev->sb_start = calc_dev_sboffset(rdev);
5344         else
5345                 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5346
5347         rdev->sectors = rdev->sb_start;
5348
5349         if (test_bit(Faulty, &rdev->flags)) {
5350                 printk(KERN_WARNING
5351                         "md: can not hot-add faulty %s disk to %s!\n",
5352                         bdevname(rdev->bdev,b), mdname(mddev));
5353                 err = -EINVAL;
5354                 goto abort_export;
5355         }
5356         clear_bit(In_sync, &rdev->flags);
5357         rdev->desc_nr = -1;
5358         rdev->saved_raid_disk = -1;
5359         err = bind_rdev_to_array(rdev, mddev);
5360         if (err)
5361                 goto abort_export;
5362
5363         /*
5364          * The rest should better be atomic, we can have disk failures
5365          * noticed in interrupt contexts ...
5366          */
5367
5368         rdev->raid_disk = -1;
5369
5370         md_update_sb(mddev, 1);
5371
5372         /*
5373          * Kick recovery, maybe this spare has to be added to the
5374          * array immediately.
5375          */
5376         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5377         md_wakeup_thread(mddev->thread);
5378         md_new_event(mddev);
5379         return 0;
5380
5381 abort_export:
5382         export_rdev(rdev);
5383         return err;
5384 }
5385
5386 static int set_bitmap_file(mddev_t *mddev, int fd)
5387 {
5388         int err;
5389
5390         if (mddev->pers) {
5391                 if (!mddev->pers->quiesce)
5392                         return -EBUSY;
5393                 if (mddev->recovery || mddev->sync_thread)
5394                         return -EBUSY;
5395                 /* we should be able to change the bitmap.. */
5396         }
5397
5398
5399         if (fd >= 0) {
5400                 if (mddev->bitmap)
5401                         return -EEXIST; /* cannot add when bitmap is present */
5402                 mddev->bitmap_info.file = fget(fd);
5403
5404                 if (mddev->bitmap_info.file == NULL) {
5405                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5406                                mdname(mddev));
5407                         return -EBADF;
5408                 }
5409
5410                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5411                 if (err) {
5412                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5413                                mdname(mddev));
5414                         fput(mddev->bitmap_info.file);
5415                         mddev->bitmap_info.file = NULL;
5416                         return err;
5417                 }
5418                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5419         } else if (mddev->bitmap == NULL)
5420                 return -ENOENT; /* cannot remove what isn't there */
5421         err = 0;
5422         if (mddev->pers) {
5423                 mddev->pers->quiesce(mddev, 1);
5424                 if (fd >= 0) {
5425                         err = bitmap_create(mddev);
5426                         if (!err)
5427                                 err = bitmap_load(mddev);
5428                 }
5429                 if (fd < 0 || err) {
5430                         bitmap_destroy(mddev);
5431                         fd = -1; /* make sure to put the file */
5432                 }
5433                 mddev->pers->quiesce(mddev, 0);
5434         }
5435         if (fd < 0) {
5436                 if (mddev->bitmap_info.file) {
5437                         restore_bitmap_write_access(mddev->bitmap_info.file);
5438                         fput(mddev->bitmap_info.file);
5439                 }
5440                 mddev->bitmap_info.file = NULL;
5441         }
5442
5443         return err;
5444 }
5445
5446 /*
5447  * set_array_info is used two different ways
5448  * The original usage is when creating a new array.
5449  * In this usage, raid_disks is > 0 and it together with
5450  *  level, size, not_persistent,layout,chunksize determine the
5451  *  shape of the array.
5452  *  This will always create an array with a type-0.90.0 superblock.
5453  * The newer usage is when assembling an array.
5454  *  In this case raid_disks will be 0, and the major_version field is
5455  *  use to determine which style super-blocks are to be found on the devices.
5456  *  The minor and patch _version numbers are also kept incase the
5457  *  super_block handler wishes to interpret them.
5458  */
5459 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5460 {
5461
5462         if (info->raid_disks == 0) {
5463                 /* just setting version number for superblock loading */
5464                 if (info->major_version < 0 ||
5465                     info->major_version >= ARRAY_SIZE(super_types) ||
5466                     super_types[info->major_version].name == NULL) {
5467                         /* maybe try to auto-load a module? */
5468                         printk(KERN_INFO
5469                                 "md: superblock version %d not known\n",
5470                                 info->major_version);
5471                         return -EINVAL;
5472                 }
5473                 mddev->major_version = info->major_version;
5474                 mddev->minor_version = info->minor_version;
5475                 mddev->patch_version = info->patch_version;
5476                 mddev->persistent = !info->not_persistent;
5477                 /* ensure mddev_put doesn't delete this now that there
5478                  * is some minimal configuration.
5479                  */
5480                 mddev->ctime         = get_seconds();
5481                 return 0;
5482         }
5483         mddev->major_version = MD_MAJOR_VERSION;
5484         mddev->minor_version = MD_MINOR_VERSION;
5485         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5486         mddev->ctime         = get_seconds();
5487
5488         mddev->level         = info->level;
5489         mddev->clevel[0]     = 0;
5490         mddev->dev_sectors   = 2 * (sector_t)info->size;
5491         mddev->raid_disks    = info->raid_disks;
5492         /* don't set md_minor, it is determined by which /dev/md* was
5493          * openned
5494          */
5495         if (info->state & (1<<MD_SB_CLEAN))
5496                 mddev->recovery_cp = MaxSector;
5497         else
5498                 mddev->recovery_cp = 0;
5499         mddev->persistent    = ! info->not_persistent;
5500         mddev->external      = 0;
5501
5502         mddev->layout        = info->layout;
5503         mddev->chunk_sectors = info->chunk_size >> 9;
5504
5505         mddev->max_disks     = MD_SB_DISKS;
5506
5507         if (mddev->persistent)
5508                 mddev->flags         = 0;
5509         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5510
5511         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5512         mddev->bitmap_info.offset = 0;
5513
5514         mddev->reshape_position = MaxSector;
5515
5516         /*
5517          * Generate a 128 bit UUID
5518          */
5519         get_random_bytes(mddev->uuid, 16);
5520
5521         mddev->new_level = mddev->level;
5522         mddev->new_chunk_sectors = mddev->chunk_sectors;
5523         mddev->new_layout = mddev->layout;
5524         mddev->delta_disks = 0;
5525
5526         return 0;
5527 }
5528
5529 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5530 {
5531         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5532
5533         if (mddev->external_size)
5534                 return;
5535
5536         mddev->array_sectors = array_sectors;
5537 }
5538 EXPORT_SYMBOL(md_set_array_sectors);
5539
5540 static int update_size(mddev_t *mddev, sector_t num_sectors)
5541 {
5542         mdk_rdev_t *rdev;
5543         int rv;
5544         int fit = (num_sectors == 0);
5545
5546         if (mddev->pers->resize == NULL)
5547                 return -EINVAL;
5548         /* The "num_sectors" is the number of sectors of each device that
5549          * is used.  This can only make sense for arrays with redundancy.
5550          * linear and raid0 always use whatever space is available. We can only
5551          * consider changing this number if no resync or reconstruction is
5552          * happening, and if the new size is acceptable. It must fit before the
5553          * sb_start or, if that is <data_offset, it must fit before the size
5554          * of each device.  If num_sectors is zero, we find the largest size
5555          * that fits.
5556          */
5557         if (mddev->sync_thread)
5558                 return -EBUSY;
5559         if (mddev->bitmap)
5560                 /* Sorry, cannot grow a bitmap yet, just remove it,
5561                  * grow, and re-add.
5562                  */
5563                 return -EBUSY;
5564         list_for_each_entry(rdev, &mddev->disks, same_set) {
5565                 sector_t avail = rdev->sectors;
5566
5567                 if (fit && (num_sectors == 0 || num_sectors > avail))
5568                         num_sectors = avail;
5569                 if (avail < num_sectors)
5570                         return -ENOSPC;
5571         }
5572         rv = mddev->pers->resize(mddev, num_sectors);
5573         if (!rv)
5574                 revalidate_disk(mddev->gendisk);
5575         return rv;
5576 }
5577
5578 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5579 {
5580         int rv;
5581         /* change the number of raid disks */
5582         if (mddev->pers->check_reshape == NULL)
5583                 return -EINVAL;
5584         if (raid_disks <= 0 ||
5585             (mddev->max_disks && raid_disks >= mddev->max_disks))
5586                 return -EINVAL;
5587         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5588                 return -EBUSY;
5589         mddev->delta_disks = raid_disks - mddev->raid_disks;
5590
5591         rv = mddev->pers->check_reshape(mddev);
5592         if (rv < 0)
5593                 mddev->delta_disks = 0;
5594         return rv;
5595 }
5596
5597
5598 /*
5599  * update_array_info is used to change the configuration of an
5600  * on-line array.
5601  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5602  * fields in the info are checked against the array.
5603  * Any differences that cannot be handled will cause an error.
5604  * Normally, only one change can be managed at a time.
5605  */
5606 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5607 {
5608         int rv = 0;
5609         int cnt = 0;
5610         int state = 0;
5611
5612         /* calculate expected state,ignoring low bits */
5613         if (mddev->bitmap && mddev->bitmap_info.offset)
5614                 state |= (1 << MD_SB_BITMAP_PRESENT);
5615
5616         if (mddev->major_version != info->major_version ||
5617             mddev->minor_version != info->minor_version ||
5618 /*          mddev->patch_version != info->patch_version || */
5619             mddev->ctime         != info->ctime         ||
5620             mddev->level         != info->level         ||
5621 /*          mddev->layout        != info->layout        || */
5622             !mddev->persistent   != info->not_persistent||
5623             mddev->chunk_sectors != info->chunk_size >> 9 ||
5624             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5625             ((state^info->state) & 0xfffffe00)
5626                 )
5627                 return -EINVAL;
5628         /* Check there is only one change */
5629         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5630                 cnt++;
5631         if (mddev->raid_disks != info->raid_disks)
5632                 cnt++;
5633         if (mddev->layout != info->layout)
5634                 cnt++;
5635         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5636                 cnt++;
5637         if (cnt == 0)
5638                 return 0;
5639         if (cnt > 1)
5640                 return -EINVAL;
5641
5642         if (mddev->layout != info->layout) {
5643                 /* Change layout
5644                  * we don't need to do anything at the md level, the
5645                  * personality will take care of it all.
5646                  */
5647                 if (mddev->pers->check_reshape == NULL)
5648                         return -EINVAL;
5649                 else {
5650                         mddev->new_layout = info->layout;
5651                         rv = mddev->pers->check_reshape(mddev);
5652                         if (rv)
5653                                 mddev->new_layout = mddev->layout;
5654                         return rv;
5655                 }
5656         }
5657         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5658                 rv = update_size(mddev, (sector_t)info->size * 2);
5659
5660         if (mddev->raid_disks    != info->raid_disks)
5661                 rv = update_raid_disks(mddev, info->raid_disks);
5662
5663         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5664                 if (mddev->pers->quiesce == NULL)
5665                         return -EINVAL;
5666                 if (mddev->recovery || mddev->sync_thread)
5667                         return -EBUSY;
5668                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5669                         /* add the bitmap */
5670                         if (mddev->bitmap)
5671                                 return -EEXIST;
5672                         if (mddev->bitmap_info.default_offset == 0)
5673                                 return -EINVAL;
5674                         mddev->bitmap_info.offset =
5675                                 mddev->bitmap_info.default_offset;
5676                         mddev->pers->quiesce(mddev, 1);
5677                         rv = bitmap_create(mddev);
5678                         if (!rv)
5679                                 rv = bitmap_load(mddev);
5680                         if (rv)
5681                                 bitmap_destroy(mddev);
5682                         mddev->pers->quiesce(mddev, 0);
5683                 } else {
5684                         /* remove the bitmap */
5685                         if (!mddev->bitmap)
5686                                 return -ENOENT;
5687                         if (mddev->bitmap->file)
5688                                 return -EINVAL;
5689                         mddev->pers->quiesce(mddev, 1);
5690                         bitmap_destroy(mddev);
5691                         mddev->pers->quiesce(mddev, 0);
5692                         mddev->bitmap_info.offset = 0;
5693                 }
5694         }
5695         md_update_sb(mddev, 1);
5696         return rv;
5697 }
5698
5699 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5700 {
5701         mdk_rdev_t *rdev;
5702
5703         if (mddev->pers == NULL)
5704                 return -ENODEV;
5705
5706         rdev = find_rdev(mddev, dev);
5707         if (!rdev)
5708                 return -ENODEV;
5709
5710         md_error(mddev, rdev);
5711         return 0;
5712 }
5713
5714 /*
5715  * We have a problem here : there is no easy way to give a CHS
5716  * virtual geometry. We currently pretend that we have a 2 heads
5717  * 4 sectors (with a BIG number of cylinders...). This drives
5718  * dosfs just mad... ;-)
5719  */
5720 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5721 {
5722         mddev_t *mddev = bdev->bd_disk->private_data;
5723
5724         geo->heads = 2;
5725         geo->sectors = 4;
5726         geo->cylinders = mddev->array_sectors / 8;
5727         return 0;
5728 }
5729
5730 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5731                         unsigned int cmd, unsigned long arg)
5732 {
5733         int err = 0;
5734         void __user *argp = (void __user *)arg;
5735         mddev_t *mddev = NULL;
5736         int ro;
5737
5738         if (!capable(CAP_SYS_ADMIN))
5739                 return -EACCES;
5740
5741         /*
5742          * Commands dealing with the RAID driver but not any
5743          * particular array:
5744          */
5745         switch (cmd)
5746         {
5747                 case RAID_VERSION:
5748                         err = get_version(argp);
5749                         goto done;
5750
5751                 case PRINT_RAID_DEBUG:
5752                         err = 0;
5753                         md_print_devices();
5754                         goto done;
5755
5756 #ifndef MODULE
5757                 case RAID_AUTORUN:
5758                         err = 0;
5759                         autostart_arrays(arg);
5760                         goto done;
5761 #endif
5762                 default:;
5763         }
5764
5765         /*
5766          * Commands creating/starting a new array:
5767          */
5768
5769         mddev = bdev->bd_disk->private_data;
5770
5771         if (!mddev) {
5772                 BUG();
5773                 goto abort;
5774         }
5775
5776         err = mddev_lock(mddev);
5777         if (err) {
5778                 printk(KERN_INFO
5779                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5780                         err, cmd);
5781                 goto abort;
5782         }
5783
5784         switch (cmd)
5785         {
5786                 case SET_ARRAY_INFO:
5787                         {
5788                                 mdu_array_info_t info;
5789                                 if (!arg)
5790                                         memset(&info, 0, sizeof(info));
5791                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5792                                         err = -EFAULT;
5793                                         goto abort_unlock;
5794                                 }
5795                                 if (mddev->pers) {
5796                                         err = update_array_info(mddev, &info);
5797                                         if (err) {
5798                                                 printk(KERN_WARNING "md: couldn't update"
5799                                                        " array info. %d\n", err);
5800                                                 goto abort_unlock;
5801                                         }
5802                                         goto done_unlock;
5803                                 }
5804                                 if (!list_empty(&mddev->disks)) {
5805                                         printk(KERN_WARNING
5806                                                "md: array %s already has disks!\n",
5807                                                mdname(mddev));
5808                                         err = -EBUSY;
5809                                         goto abort_unlock;
5810                                 }
5811                                 if (mddev->raid_disks) {
5812                                         printk(KERN_WARNING
5813                                                "md: array %s already initialised!\n",
5814                                                mdname(mddev));
5815                                         err = -EBUSY;
5816                                         goto abort_unlock;
5817                                 }
5818                                 err = set_array_info(mddev, &info);
5819                                 if (err) {
5820                                         printk(KERN_WARNING "md: couldn't set"
5821                                                " array info. %d\n", err);
5822                                         goto abort_unlock;
5823                                 }
5824                         }
5825                         goto done_unlock;
5826
5827                 default:;
5828         }
5829
5830         /*
5831          * Commands querying/configuring an existing array:
5832          */
5833         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5834          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5835         if ((!mddev->raid_disks && !mddev->external)
5836             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5837             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5838             && cmd != GET_BITMAP_FILE) {
5839                 err = -ENODEV;
5840                 goto abort_unlock;
5841         }
5842
5843         /*
5844          * Commands even a read-only array can execute:
5845          */
5846         switch (cmd)
5847         {
5848                 case GET_ARRAY_INFO:
5849                         err = get_array_info(mddev, argp);
5850                         goto done_unlock;
5851
5852                 case GET_BITMAP_FILE:
5853                         err = get_bitmap_file(mddev, argp);
5854                         goto done_unlock;
5855
5856                 case GET_DISK_INFO:
5857                         err = get_disk_info(mddev, argp);
5858                         goto done_unlock;
5859
5860                 case RESTART_ARRAY_RW:
5861                         err = restart_array(mddev);
5862                         goto done_unlock;
5863
5864                 case STOP_ARRAY:
5865                         err = do_md_stop(mddev, 0, 1);
5866                         goto done_unlock;
5867
5868                 case STOP_ARRAY_RO:
5869                         err = md_set_readonly(mddev, 1);
5870                         goto done_unlock;
5871
5872                 case BLKROSET:
5873                         if (get_user(ro, (int __user *)(arg))) {
5874                                 err = -EFAULT;
5875                                 goto done_unlock;
5876                         }
5877                         err = -EINVAL;
5878
5879                         /* if the bdev is going readonly the value of mddev->ro
5880                          * does not matter, no writes are coming
5881                          */
5882                         if (ro)
5883                                 goto done_unlock;
5884
5885                         /* are we are already prepared for writes? */
5886                         if (mddev->ro != 1)
5887                                 goto done_unlock;
5888
5889                         /* transitioning to readauto need only happen for
5890                          * arrays that call md_write_start
5891                          */
5892                         if (mddev->pers) {
5893                                 err = restart_array(mddev);
5894                                 if (err == 0) {
5895                                         mddev->ro = 2;
5896                                         set_disk_ro(mddev->gendisk, 0);
5897                                 }
5898                         }
5899                         goto done_unlock;
5900         }
5901
5902         /*
5903          * The remaining ioctls are changing the state of the
5904          * superblock, so we do not allow them on read-only arrays.
5905          * However non-MD ioctls (e.g. get-size) will still come through
5906          * here and hit the 'default' below, so only disallow
5907          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5908          */
5909         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5910                 if (mddev->ro == 2) {
5911                         mddev->ro = 0;
5912                         sysfs_notify_dirent_safe(mddev->sysfs_state);
5913                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5914                         md_wakeup_thread(mddev->thread);
5915                 } else {
5916                         err = -EROFS;
5917                         goto abort_unlock;
5918                 }
5919         }
5920
5921         switch (cmd)
5922         {
5923                 case ADD_NEW_DISK:
5924                 {
5925                         mdu_disk_info_t info;
5926                         if (copy_from_user(&info, argp, sizeof(info)))
5927                                 err = -EFAULT;
5928                         else
5929                                 err = add_new_disk(mddev, &info);
5930                         goto done_unlock;
5931                 }
5932
5933                 case HOT_REMOVE_DISK:
5934                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5935                         goto done_unlock;
5936
5937                 case HOT_ADD_DISK:
5938                         err = hot_add_disk(mddev, new_decode_dev(arg));
5939                         goto done_unlock;
5940
5941                 case SET_DISK_FAULTY:
5942                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5943                         goto done_unlock;
5944
5945                 case RUN_ARRAY:
5946                         err = do_md_run(mddev);
5947                         goto done_unlock;
5948
5949                 case SET_BITMAP_FILE:
5950                         err = set_bitmap_file(mddev, (int)arg);
5951                         goto done_unlock;
5952
5953                 default:
5954                         err = -EINVAL;
5955                         goto abort_unlock;
5956         }
5957
5958 done_unlock:
5959 abort_unlock:
5960         if (mddev->hold_active == UNTIL_IOCTL &&
5961             err != -EINVAL)
5962                 mddev->hold_active = 0;
5963         mddev_unlock(mddev);
5964
5965         return err;
5966 done:
5967         if (err)
5968                 MD_BUG();
5969 abort:
5970         return err;
5971 }
5972 #ifdef CONFIG_COMPAT
5973 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5974                     unsigned int cmd, unsigned long arg)
5975 {
5976         switch (cmd) {
5977         case HOT_REMOVE_DISK:
5978         case HOT_ADD_DISK:
5979         case SET_DISK_FAULTY:
5980         case SET_BITMAP_FILE:
5981                 /* These take in integer arg, do not convert */
5982                 break;
5983         default:
5984                 arg = (unsigned long)compat_ptr(arg);
5985                 break;
5986         }
5987
5988         return md_ioctl(bdev, mode, cmd, arg);
5989 }
5990 #endif /* CONFIG_COMPAT */
5991
5992 static int md_open(struct block_device *bdev, fmode_t mode)
5993 {
5994         /*
5995          * Succeed if we can lock the mddev, which confirms that
5996          * it isn't being stopped right now.
5997          */
5998         mddev_t *mddev = mddev_find(bdev->bd_dev);
5999         int err;
6000
6001         if (mddev->gendisk != bdev->bd_disk) {
6002                 /* we are racing with mddev_put which is discarding this
6003                  * bd_disk.
6004                  */
6005                 mddev_put(mddev);
6006                 /* Wait until bdev->bd_disk is definitely gone */
6007                 flush_workqueue(md_misc_wq);
6008                 /* Then retry the open from the top */
6009                 return -ERESTARTSYS;
6010         }
6011         BUG_ON(mddev != bdev->bd_disk->private_data);
6012
6013         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6014                 goto out;
6015
6016         err = 0;
6017         atomic_inc(&mddev->openers);
6018         mutex_unlock(&mddev->open_mutex);
6019
6020         check_disk_change(bdev);
6021  out:
6022         return err;
6023 }
6024
6025 static int md_release(struct gendisk *disk, fmode_t mode)
6026 {
6027         mddev_t *mddev = disk->private_data;
6028
6029         BUG_ON(!mddev);
6030         atomic_dec(&mddev->openers);
6031         mddev_put(mddev);
6032
6033         return 0;
6034 }
6035
6036 static int md_media_changed(struct gendisk *disk)
6037 {
6038         mddev_t *mddev = disk->private_data;
6039
6040         return mddev->changed;
6041 }
6042
6043 static int md_revalidate(struct gendisk *disk)
6044 {
6045         mddev_t *mddev = disk->private_data;
6046
6047         mddev->changed = 0;
6048         return 0;
6049 }
6050 static const struct block_device_operations md_fops =
6051 {
6052         .owner          = THIS_MODULE,
6053         .open           = md_open,
6054         .release        = md_release,
6055         .ioctl          = md_ioctl,
6056 #ifdef CONFIG_COMPAT
6057         .compat_ioctl   = md_compat_ioctl,
6058 #endif
6059         .getgeo         = md_getgeo,
6060         .media_changed  = md_media_changed,
6061         .revalidate_disk= md_revalidate,
6062 };
6063
6064 static int md_thread(void * arg)
6065 {
6066         mdk_thread_t *thread = arg;
6067
6068         /*
6069          * md_thread is a 'system-thread', it's priority should be very
6070          * high. We avoid resource deadlocks individually in each
6071          * raid personality. (RAID5 does preallocation) We also use RR and
6072          * the very same RT priority as kswapd, thus we will never get
6073          * into a priority inversion deadlock.
6074          *
6075          * we definitely have to have equal or higher priority than
6076          * bdflush, otherwise bdflush will deadlock if there are too
6077          * many dirty RAID5 blocks.
6078          */
6079
6080         allow_signal(SIGKILL);
6081         while (!kthread_should_stop()) {
6082
6083                 /* We need to wait INTERRUPTIBLE so that
6084                  * we don't add to the load-average.
6085                  * That means we need to be sure no signals are
6086                  * pending
6087                  */
6088                 if (signal_pending(current))
6089                         flush_signals(current);
6090
6091                 wait_event_interruptible_timeout
6092                         (thread->wqueue,
6093                          test_bit(THREAD_WAKEUP, &thread->flags)
6094                          || kthread_should_stop(),
6095                          thread->timeout);
6096
6097                 clear_bit(THREAD_WAKEUP, &thread->flags);
6098                 if (!kthread_should_stop())
6099                         thread->run(thread->mddev);
6100         }
6101
6102         return 0;
6103 }
6104
6105 void md_wakeup_thread(mdk_thread_t *thread)
6106 {
6107         if (thread) {
6108                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6109                 set_bit(THREAD_WAKEUP, &thread->flags);
6110                 wake_up(&thread->wqueue);
6111         }
6112 }
6113
6114 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6115                                  const char *name)
6116 {
6117         mdk_thread_t *thread;
6118
6119         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6120         if (!thread)
6121                 return NULL;
6122
6123         init_waitqueue_head(&thread->wqueue);
6124
6125         thread->run = run;
6126         thread->mddev = mddev;
6127         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6128         thread->tsk = kthread_run(md_thread, thread,
6129                                   "%s_%s",
6130                                   mdname(thread->mddev),
6131                                   name ?: mddev->pers->name);
6132         if (IS_ERR(thread->tsk)) {
6133                 kfree(thread);
6134                 return NULL;
6135         }
6136         return thread;
6137 }
6138
6139 void md_unregister_thread(mdk_thread_t *thread)
6140 {
6141         if (!thread)
6142                 return;
6143         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6144
6145         kthread_stop(thread->tsk);
6146         kfree(thread);
6147 }
6148
6149 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6150 {
6151         if (!mddev) {
6152                 MD_BUG();
6153                 return;
6154         }
6155
6156         if (!rdev || test_bit(Faulty, &rdev->flags))
6157                 return;
6158
6159         if (mddev->external)
6160                 set_bit(Blocked, &rdev->flags);
6161 /*
6162         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6163                 mdname(mddev),
6164                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6165                 __builtin_return_address(0),__builtin_return_address(1),
6166                 __builtin_return_address(2),__builtin_return_address(3));
6167 */
6168         if (!mddev->pers)
6169                 return;
6170         if (!mddev->pers->error_handler)
6171                 return;
6172         mddev->pers->error_handler(mddev,rdev);
6173         if (mddev->degraded)
6174                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6175         sysfs_notify_dirent_safe(rdev->sysfs_state);
6176         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6177         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6178         md_wakeup_thread(mddev->thread);
6179         if (mddev->event_work.func)
6180                 queue_work(md_misc_wq, &mddev->event_work);
6181         md_new_event_inintr(mddev);
6182 }
6183
6184 /* seq_file implementation /proc/mdstat */
6185
6186 static void status_unused(struct seq_file *seq)
6187 {
6188         int i = 0;
6189         mdk_rdev_t *rdev;
6190
6191         seq_printf(seq, "unused devices: ");
6192
6193         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6194                 char b[BDEVNAME_SIZE];
6195                 i++;
6196                 seq_printf(seq, "%s ",
6197                               bdevname(rdev->bdev,b));
6198         }
6199         if (!i)
6200                 seq_printf(seq, "<none>");
6201
6202         seq_printf(seq, "\n");
6203 }
6204
6205
6206 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6207 {
6208         sector_t max_sectors, resync, res;
6209         unsigned long dt, db;
6210         sector_t rt;
6211         int scale;
6212         unsigned int per_milli;
6213
6214         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6215
6216         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6217                 max_sectors = mddev->resync_max_sectors;
6218         else
6219                 max_sectors = mddev->dev_sectors;
6220
6221         /*
6222          * Should not happen.
6223          */
6224         if (!max_sectors) {
6225                 MD_BUG();
6226                 return;
6227         }
6228         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6229          * in a sector_t, and (max_sectors>>scale) will fit in a
6230          * u32, as those are the requirements for sector_div.
6231          * Thus 'scale' must be at least 10
6232          */
6233         scale = 10;
6234         if (sizeof(sector_t) > sizeof(unsigned long)) {
6235                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6236                         scale++;
6237         }
6238         res = (resync>>scale)*1000;
6239         sector_div(res, (u32)((max_sectors>>scale)+1));
6240
6241         per_milli = res;
6242         {
6243                 int i, x = per_milli/50, y = 20-x;
6244                 seq_printf(seq, "[");
6245                 for (i = 0; i < x; i++)
6246                         seq_printf(seq, "=");
6247                 seq_printf(seq, ">");
6248                 for (i = 0; i < y; i++)
6249                         seq_printf(seq, ".");
6250                 seq_printf(seq, "] ");
6251         }
6252         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6253                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6254                     "reshape" :
6255                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6256                      "check" :
6257                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6258                       "resync" : "recovery"))),
6259                    per_milli/10, per_milli % 10,
6260                    (unsigned long long) resync/2,
6261                    (unsigned long long) max_sectors/2);
6262
6263         /*
6264          * dt: time from mark until now
6265          * db: blocks written from mark until now
6266          * rt: remaining time
6267          *
6268          * rt is a sector_t, so could be 32bit or 64bit.
6269          * So we divide before multiply in case it is 32bit and close
6270          * to the limit.
6271          * We scale the divisor (db) by 32 to avoid loosing precision
6272          * near the end of resync when the number of remaining sectors
6273          * is close to 'db'.
6274          * We then divide rt by 32 after multiplying by db to compensate.
6275          * The '+1' avoids division by zero if db is very small.
6276          */
6277         dt = ((jiffies - mddev->resync_mark) / HZ);
6278         if (!dt) dt++;
6279         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6280                 - mddev->resync_mark_cnt;
6281
6282         rt = max_sectors - resync;    /* number of remaining sectors */
6283         sector_div(rt, db/32+1);
6284         rt *= dt;
6285         rt >>= 5;
6286
6287         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6288                    ((unsigned long)rt % 60)/6);
6289
6290         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6291 }
6292
6293 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6294 {
6295         struct list_head *tmp;
6296         loff_t l = *pos;
6297         mddev_t *mddev;
6298
6299         if (l >= 0x10000)
6300                 return NULL;
6301         if (!l--)
6302                 /* header */
6303                 return (void*)1;
6304
6305         spin_lock(&all_mddevs_lock);
6306         list_for_each(tmp,&all_mddevs)
6307                 if (!l--) {
6308                         mddev = list_entry(tmp, mddev_t, all_mddevs);
6309                         mddev_get(mddev);
6310                         spin_unlock(&all_mddevs_lock);
6311                         return mddev;
6312                 }
6313         spin_unlock(&all_mddevs_lock);
6314         if (!l--)
6315                 return (void*)2;/* tail */
6316         return NULL;
6317 }
6318
6319 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6320 {
6321         struct list_head *tmp;
6322         mddev_t *next_mddev, *mddev = v;
6323
6324         ++*pos;
6325         if (v == (void*)2)
6326                 return NULL;
6327
6328         spin_lock(&all_mddevs_lock);
6329         if (v == (void*)1)
6330                 tmp = all_mddevs.next;
6331         else
6332                 tmp = mddev->all_mddevs.next;
6333         if (tmp != &all_mddevs)
6334                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6335         else {
6336                 next_mddev = (void*)2;
6337                 *pos = 0x10000;
6338         }
6339         spin_unlock(&all_mddevs_lock);
6340
6341         if (v != (void*)1)
6342                 mddev_put(mddev);
6343         return next_mddev;
6344
6345 }
6346
6347 static void md_seq_stop(struct seq_file *seq, void *v)
6348 {
6349         mddev_t *mddev = v;
6350
6351         if (mddev && v != (void*)1 && v != (void*)2)
6352                 mddev_put(mddev);
6353 }
6354
6355 struct mdstat_info {
6356         int event;
6357 };
6358
6359 static int md_seq_show(struct seq_file *seq, void *v)
6360 {
6361         mddev_t *mddev = v;
6362         sector_t sectors;
6363         mdk_rdev_t *rdev;
6364         struct mdstat_info *mi = seq->private;
6365         struct bitmap *bitmap;
6366
6367         if (v == (void*)1) {
6368                 struct mdk_personality *pers;
6369                 seq_printf(seq, "Personalities : ");
6370                 spin_lock(&pers_lock);
6371                 list_for_each_entry(pers, &pers_list, list)
6372                         seq_printf(seq, "[%s] ", pers->name);
6373
6374                 spin_unlock(&pers_lock);
6375                 seq_printf(seq, "\n");
6376                 mi->event = atomic_read(&md_event_count);
6377                 return 0;
6378         }
6379         if (v == (void*)2) {
6380                 status_unused(seq);
6381                 return 0;
6382         }
6383
6384         if (mddev_lock(mddev) < 0)
6385                 return -EINTR;
6386
6387         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6388                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6389                                                 mddev->pers ? "" : "in");
6390                 if (mddev->pers) {
6391                         if (mddev->ro==1)
6392                                 seq_printf(seq, " (read-only)");
6393                         if (mddev->ro==2)
6394                                 seq_printf(seq, " (auto-read-only)");
6395                         seq_printf(seq, " %s", mddev->pers->name);
6396                 }
6397
6398                 sectors = 0;
6399                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6400                         char b[BDEVNAME_SIZE];
6401                         seq_printf(seq, " %s[%d]",
6402                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6403                         if (test_bit(WriteMostly, &rdev->flags))
6404                                 seq_printf(seq, "(W)");
6405                         if (test_bit(Faulty, &rdev->flags)) {
6406                                 seq_printf(seq, "(F)");
6407                                 continue;
6408                         } else if (rdev->raid_disk < 0)
6409                                 seq_printf(seq, "(S)"); /* spare */
6410                         sectors += rdev->sectors;
6411                 }
6412
6413                 if (!list_empty(&mddev->disks)) {
6414                         if (mddev->pers)
6415                                 seq_printf(seq, "\n      %llu blocks",
6416                                            (unsigned long long)
6417                                            mddev->array_sectors / 2);
6418                         else
6419                                 seq_printf(seq, "\n      %llu blocks",
6420                                            (unsigned long long)sectors / 2);
6421                 }
6422                 if (mddev->persistent) {
6423                         if (mddev->major_version != 0 ||
6424                             mddev->minor_version != 90) {
6425                                 seq_printf(seq," super %d.%d",
6426                                            mddev->major_version,
6427                                            mddev->minor_version);
6428                         }
6429                 } else if (mddev->external)
6430                         seq_printf(seq, " super external:%s",
6431                                    mddev->metadata_type);
6432                 else
6433                         seq_printf(seq, " super non-persistent");
6434
6435                 if (mddev->pers) {
6436                         mddev->pers->status(seq, mddev);
6437                         seq_printf(seq, "\n      ");
6438                         if (mddev->pers->sync_request) {
6439                                 if (mddev->curr_resync > 2) {
6440                                         status_resync(seq, mddev);
6441                                         seq_printf(seq, "\n      ");
6442                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6443                                         seq_printf(seq, "\tresync=DELAYED\n      ");
6444                                 else if (mddev->recovery_cp < MaxSector)
6445                                         seq_printf(seq, "\tresync=PENDING\n      ");
6446                         }
6447                 } else
6448                         seq_printf(seq, "\n       ");
6449
6450                 if ((bitmap = mddev->bitmap)) {
6451                         unsigned long chunk_kb;
6452                         unsigned long flags;
6453                         spin_lock_irqsave(&bitmap->lock, flags);
6454                         chunk_kb = mddev->bitmap_info.chunksize >> 10;
6455                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6456                                 "%lu%s chunk",
6457                                 bitmap->pages - bitmap->missing_pages,
6458                                 bitmap->pages,
6459                                 (bitmap->pages - bitmap->missing_pages)
6460                                         << (PAGE_SHIFT - 10),
6461                                 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6462                                 chunk_kb ? "KB" : "B");
6463                         if (bitmap->file) {
6464                                 seq_printf(seq, ", file: ");
6465                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
6466                         }
6467
6468                         seq_printf(seq, "\n");
6469                         spin_unlock_irqrestore(&bitmap->lock, flags);
6470                 }
6471
6472                 seq_printf(seq, "\n");
6473         }
6474         mddev_unlock(mddev);
6475
6476         return 0;
6477 }
6478
6479 static const struct seq_operations md_seq_ops = {
6480         .start  = md_seq_start,
6481         .next   = md_seq_next,
6482         .stop   = md_seq_stop,
6483         .show   = md_seq_show,
6484 };
6485
6486 static int md_seq_open(struct inode *inode, struct file *file)
6487 {
6488         int error;
6489         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6490         if (mi == NULL)
6491                 return -ENOMEM;
6492
6493         error = seq_open(file, &md_seq_ops);
6494         if (error)
6495                 kfree(mi);
6496         else {
6497                 struct seq_file *p = file->private_data;
6498                 p->private = mi;
6499                 mi->event = atomic_read(&md_event_count);
6500         }
6501         return error;
6502 }
6503
6504 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6505 {
6506         struct seq_file *m = filp->private_data;
6507         struct mdstat_info *mi = m->private;
6508         int mask;
6509
6510         poll_wait(filp, &md_event_waiters, wait);
6511
6512         /* always allow read */
6513         mask = POLLIN | POLLRDNORM;
6514
6515         if (mi->event != atomic_read(&md_event_count))
6516                 mask |= POLLERR | POLLPRI;
6517         return mask;
6518 }
6519
6520 static const struct file_operations md_seq_fops = {
6521         .owner          = THIS_MODULE,
6522         .open           = md_seq_open,
6523         .read           = seq_read,
6524         .llseek         = seq_lseek,
6525         .release        = seq_release_private,
6526         .poll           = mdstat_poll,
6527 };
6528
6529 int register_md_personality(struct mdk_personality *p)
6530 {
6531         spin_lock(&pers_lock);
6532         list_add_tail(&p->list, &pers_list);
6533         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6534         spin_unlock(&pers_lock);
6535         return 0;
6536 }
6537
6538 int unregister_md_personality(struct mdk_personality *p)
6539 {
6540         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6541         spin_lock(&pers_lock);
6542         list_del_init(&p->list);
6543         spin_unlock(&pers_lock);
6544         return 0;
6545 }
6546
6547 static int is_mddev_idle(mddev_t *mddev, int init)
6548 {
6549         mdk_rdev_t * rdev;
6550         int idle;
6551         int curr_events;
6552
6553         idle = 1;
6554         rcu_read_lock();
6555         rdev_for_each_rcu(rdev, mddev) {
6556                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6557                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6558                               (int)part_stat_read(&disk->part0, sectors[1]) -
6559                               atomic_read(&disk->sync_io);
6560                 /* sync IO will cause sync_io to increase before the disk_stats
6561                  * as sync_io is counted when a request starts, and
6562                  * disk_stats is counted when it completes.
6563                  * So resync activity will cause curr_events to be smaller than
6564                  * when there was no such activity.
6565                  * non-sync IO will cause disk_stat to increase without
6566                  * increasing sync_io so curr_events will (eventually)
6567                  * be larger than it was before.  Once it becomes
6568                  * substantially larger, the test below will cause
6569                  * the array to appear non-idle, and resync will slow
6570                  * down.
6571                  * If there is a lot of outstanding resync activity when
6572                  * we set last_event to curr_events, then all that activity
6573                  * completing might cause the array to appear non-idle
6574                  * and resync will be slowed down even though there might
6575                  * not have been non-resync activity.  This will only
6576                  * happen once though.  'last_events' will soon reflect
6577                  * the state where there is little or no outstanding
6578                  * resync requests, and further resync activity will
6579                  * always make curr_events less than last_events.
6580                  *
6581                  */
6582                 if (init || curr_events - rdev->last_events > 64) {
6583                         rdev->last_events = curr_events;
6584                         idle = 0;
6585                 }
6586         }
6587         rcu_read_unlock();
6588         return idle;
6589 }
6590
6591 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6592 {
6593         /* another "blocks" (512byte) blocks have been synced */
6594         atomic_sub(blocks, &mddev->recovery_active);
6595         wake_up(&mddev->recovery_wait);
6596         if (!ok) {
6597                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6598                 md_wakeup_thread(mddev->thread);
6599                 // stop recovery, signal do_sync ....
6600         }
6601 }
6602
6603
6604 /* md_write_start(mddev, bi)
6605  * If we need to update some array metadata (e.g. 'active' flag
6606  * in superblock) before writing, schedule a superblock update
6607  * and wait for it to complete.
6608  */
6609 void md_write_start(mddev_t *mddev, struct bio *bi)
6610 {
6611         int did_change = 0;
6612         if (bio_data_dir(bi) != WRITE)
6613                 return;
6614
6615         BUG_ON(mddev->ro == 1);
6616         if (mddev->ro == 2) {
6617                 /* need to switch to read/write */
6618                 mddev->ro = 0;
6619                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6620                 md_wakeup_thread(mddev->thread);
6621                 md_wakeup_thread(mddev->sync_thread);
6622                 did_change = 1;
6623         }
6624         atomic_inc(&mddev->writes_pending);
6625         if (mddev->safemode == 1)
6626                 mddev->safemode = 0;
6627         if (mddev->in_sync) {
6628                 spin_lock_irq(&mddev->write_lock);
6629                 if (mddev->in_sync) {
6630                         mddev->in_sync = 0;
6631                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6632                         set_bit(MD_CHANGE_PENDING, &mddev->flags);
6633                         md_wakeup_thread(mddev->thread);
6634                         did_change = 1;
6635                 }
6636                 spin_unlock_irq(&mddev->write_lock);
6637         }
6638         if (did_change)
6639                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6640         wait_event(mddev->sb_wait,
6641                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6642 }
6643
6644 void md_write_end(mddev_t *mddev)
6645 {
6646         if (atomic_dec_and_test(&mddev->writes_pending)) {
6647                 if (mddev->safemode == 2)
6648                         md_wakeup_thread(mddev->thread);
6649                 else if (mddev->safemode_delay)
6650                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6651         }
6652 }
6653
6654 /* md_allow_write(mddev)
6655  * Calling this ensures that the array is marked 'active' so that writes
6656  * may proceed without blocking.  It is important to call this before
6657  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6658  * Must be called with mddev_lock held.
6659  *
6660  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6661  * is dropped, so return -EAGAIN after notifying userspace.
6662  */
6663 int md_allow_write(mddev_t *mddev)
6664 {
6665         if (!mddev->pers)
6666                 return 0;
6667         if (mddev->ro)
6668                 return 0;
6669         if (!mddev->pers->sync_request)
6670                 return 0;
6671
6672         spin_lock_irq(&mddev->write_lock);
6673         if (mddev->in_sync) {
6674                 mddev->in_sync = 0;
6675                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6676                 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6677                 if (mddev->safemode_delay &&
6678                     mddev->safemode == 0)
6679                         mddev->safemode = 1;
6680                 spin_unlock_irq(&mddev->write_lock);
6681                 md_update_sb(mddev, 0);
6682                 sysfs_notify_dirent_safe(mddev->sysfs_state);
6683         } else
6684                 spin_unlock_irq(&mddev->write_lock);
6685
6686         if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6687                 return -EAGAIN;
6688         else
6689                 return 0;
6690 }
6691 EXPORT_SYMBOL_GPL(md_allow_write);
6692
6693 void md_unplug(mddev_t *mddev)
6694 {
6695         if (mddev->queue)
6696                 blk_unplug(mddev->queue);
6697         if (mddev->plug)
6698                 mddev->plug->unplug_fn(mddev->plug);
6699 }
6700
6701 #define SYNC_MARKS      10
6702 #define SYNC_MARK_STEP  (3*HZ)
6703 void md_do_sync(mddev_t *mddev)
6704 {
6705         mddev_t *mddev2;
6706         unsigned int currspeed = 0,
6707                  window;
6708         sector_t max_sectors,j, io_sectors;
6709         unsigned long mark[SYNC_MARKS];
6710         sector_t mark_cnt[SYNC_MARKS];
6711         int last_mark,m;
6712         struct list_head *tmp;
6713         sector_t last_check;
6714         int skipped = 0;
6715         mdk_rdev_t *rdev;
6716         char *desc;
6717
6718         /* just incase thread restarts... */
6719         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6720                 return;
6721         if (mddev->ro) /* never try to sync a read-only array */
6722                 return;
6723
6724         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6725                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6726                         desc = "data-check";
6727                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6728                         desc = "requested-resync";
6729                 else
6730                         desc = "resync";
6731         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6732                 desc = "reshape";
6733         else
6734                 desc = "recovery";
6735
6736         /* we overload curr_resync somewhat here.
6737          * 0 == not engaged in resync at all
6738          * 2 == checking that there is no conflict with another sync
6739          * 1 == like 2, but have yielded to allow conflicting resync to
6740          *              commense
6741          * other == active in resync - this many blocks
6742          *
6743          * Before starting a resync we must have set curr_resync to
6744          * 2, and then checked that every "conflicting" array has curr_resync
6745          * less than ours.  When we find one that is the same or higher
6746          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6747          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6748          * This will mean we have to start checking from the beginning again.
6749          *
6750          */
6751
6752         do {
6753                 mddev->curr_resync = 2;
6754
6755         try_again:
6756                 if (kthread_should_stop())
6757                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6758
6759                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6760                         goto skip;
6761                 for_each_mddev(mddev2, tmp) {
6762                         if (mddev2 == mddev)
6763                                 continue;
6764                         if (!mddev->parallel_resync
6765                         &&  mddev2->curr_resync
6766                         &&  match_mddev_units(mddev, mddev2)) {
6767                                 DEFINE_WAIT(wq);
6768                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6769                                         /* arbitrarily yield */
6770                                         mddev->curr_resync = 1;
6771                                         wake_up(&resync_wait);
6772                                 }
6773                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6774                                         /* no need to wait here, we can wait the next
6775                                          * time 'round when curr_resync == 2
6776                                          */
6777                                         continue;
6778                                 /* We need to wait 'interruptible' so as not to
6779                                  * contribute to the load average, and not to
6780                                  * be caught by 'softlockup'
6781                                  */
6782                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6783                                 if (!kthread_should_stop() &&
6784                                     mddev2->curr_resync >= mddev->curr_resync) {
6785                                         printk(KERN_INFO "md: delaying %s of %s"
6786                                                " until %s has finished (they"
6787                                                " share one or more physical units)\n",
6788                                                desc, mdname(mddev), mdname(mddev2));
6789                                         mddev_put(mddev2);
6790                                         if (signal_pending(current))
6791                                                 flush_signals(current);
6792                                         schedule();
6793                                         finish_wait(&resync_wait, &wq);
6794                                         goto try_again;
6795                                 }
6796                                 finish_wait(&resync_wait, &wq);
6797                         }
6798                 }
6799         } while (mddev->curr_resync < 2);
6800
6801         j = 0;
6802         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6803                 /* resync follows the size requested by the personality,
6804                  * which defaults to physical size, but can be virtual size
6805                  */
6806                 max_sectors = mddev->resync_max_sectors;
6807                 mddev->resync_mismatches = 0;
6808                 /* we don't use the checkpoint if there's a bitmap */
6809                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6810                         j = mddev->resync_min;
6811                 else if (!mddev->bitmap)
6812                         j = mddev->recovery_cp;
6813
6814         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6815                 max_sectors = mddev->dev_sectors;
6816         else {
6817                 /* recovery follows the physical size of devices */
6818                 max_sectors = mddev->dev_sectors;
6819                 j = MaxSector;
6820                 rcu_read_lock();
6821                 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6822                         if (rdev->raid_disk >= 0 &&
6823                             !test_bit(Faulty, &rdev->flags) &&
6824                             !test_bit(In_sync, &rdev->flags) &&
6825                             rdev->recovery_offset < j)
6826                                 j = rdev->recovery_offset;
6827                 rcu_read_unlock();
6828         }
6829
6830         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6831         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6832                 " %d KB/sec/disk.\n", speed_min(mddev));
6833         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6834                "(but not more than %d KB/sec) for %s.\n",
6835                speed_max(mddev), desc);
6836
6837         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6838
6839         io_sectors = 0;
6840         for (m = 0; m < SYNC_MARKS; m++) {
6841                 mark[m] = jiffies;
6842                 mark_cnt[m] = io_sectors;
6843         }
6844         last_mark = 0;
6845         mddev->resync_mark = mark[last_mark];
6846         mddev->resync_mark_cnt = mark_cnt[last_mark];
6847
6848         /*
6849          * Tune reconstruction:
6850          */
6851         window = 32*(PAGE_SIZE/512);
6852         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6853                 window/2,(unsigned long long) max_sectors/2);
6854
6855         atomic_set(&mddev->recovery_active, 0);
6856         last_check = 0;
6857
6858         if (j>2) {
6859                 printk(KERN_INFO
6860                        "md: resuming %s of %s from checkpoint.\n",
6861                        desc, mdname(mddev));
6862                 mddev->curr_resync = j;
6863         }
6864         mddev->curr_resync_completed = j;
6865
6866         while (j < max_sectors) {
6867                 sector_t sectors;
6868
6869                 skipped = 0;
6870
6871                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6872                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6873                       (mddev->curr_resync - mddev->curr_resync_completed)
6874                       > (max_sectors >> 4)) ||
6875                      (j - mddev->curr_resync_completed)*2
6876                      >= mddev->resync_max - mddev->curr_resync_completed
6877                             )) {
6878                         /* time to update curr_resync_completed */
6879                         md_unplug(mddev);
6880                         wait_event(mddev->recovery_wait,
6881                                    atomic_read(&mddev->recovery_active) == 0);
6882                         mddev->curr_resync_completed = j;
6883                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6884                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6885                 }
6886
6887                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6888                         /* As this condition is controlled by user-space,
6889                          * we can block indefinitely, so use '_interruptible'
6890                          * to avoid triggering warnings.
6891                          */
6892                         flush_signals(current); /* just in case */
6893                         wait_event_interruptible(mddev->recovery_wait,
6894                                                  mddev->resync_max > j
6895                                                  || kthread_should_stop());
6896                 }
6897
6898                 if (kthread_should_stop())
6899                         goto interrupted;
6900
6901                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6902                                                   currspeed < speed_min(mddev));
6903                 if (sectors == 0) {
6904                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6905                         goto out;
6906                 }
6907
6908                 if (!skipped) { /* actual IO requested */
6909                         io_sectors += sectors;
6910                         atomic_add(sectors, &mddev->recovery_active);
6911                 }
6912
6913                 j += sectors;
6914                 if (j>1) mddev->curr_resync = j;
6915                 mddev->curr_mark_cnt = io_sectors;
6916                 if (last_check == 0)
6917                         /* this is the earliers that rebuilt will be
6918                          * visible in /proc/mdstat
6919                          */
6920                         md_new_event(mddev);
6921
6922                 if (last_check + window > io_sectors || j == max_sectors)
6923                         continue;
6924
6925                 last_check = io_sectors;
6926
6927                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6928                         break;
6929
6930         repeat:
6931                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6932                         /* step marks */
6933                         int next = (last_mark+1) % SYNC_MARKS;
6934
6935                         mddev->resync_mark = mark[next];
6936                         mddev->resync_mark_cnt = mark_cnt[next];
6937                         mark[next] = jiffies;
6938                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6939                         last_mark = next;
6940                 }
6941
6942
6943                 if (kthread_should_stop())
6944                         goto interrupted;
6945
6946
6947                 /*
6948                  * this loop exits only if either when we are slower than
6949                  * the 'hard' speed limit, or the system was IO-idle for
6950                  * a jiffy.
6951                  * the system might be non-idle CPU-wise, but we only care
6952                  * about not overloading the IO subsystem. (things like an
6953                  * e2fsck being done on the RAID array should execute fast)
6954                  */
6955                 md_unplug(mddev);
6956                 cond_resched();
6957
6958                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6959                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6960
6961                 if (currspeed > speed_min(mddev)) {
6962                         if ((currspeed > speed_max(mddev)) ||
6963                                         !is_mddev_idle(mddev, 0)) {
6964                                 msleep(500);
6965                                 goto repeat;
6966                         }
6967                 }
6968         }
6969         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6970         /*
6971          * this also signals 'finished resyncing' to md_stop
6972          */
6973  out:
6974         md_unplug(mddev);
6975
6976         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6977
6978         /* tell personality that we are finished */
6979         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6980
6981         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6982             mddev->curr_resync > 2) {
6983                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6984                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6985                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6986                                         printk(KERN_INFO
6987                                                "md: checkpointing %s of %s.\n",
6988                                                desc, mdname(mddev));
6989                                         mddev->recovery_cp = mddev->curr_resync;
6990                                 }
6991                         } else
6992                                 mddev->recovery_cp = MaxSector;
6993                 } else {
6994                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6995                                 mddev->curr_resync = MaxSector;
6996                         rcu_read_lock();
6997                         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6998                                 if (rdev->raid_disk >= 0 &&
6999                                     mddev->delta_disks >= 0 &&
7000                                     !test_bit(Faulty, &rdev->flags) &&
7001                                     !test_bit(In_sync, &rdev->flags) &&
7002                                     rdev->recovery_offset < mddev->curr_resync)
7003                                         rdev->recovery_offset = mddev->curr_resync;
7004                         rcu_read_unlock();
7005                 }
7006         }
7007         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7008
7009  skip:
7010         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7011                 /* We completed so min/max setting can be forgotten if used. */
7012                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7013                         mddev->resync_min = 0;
7014                 mddev->resync_max = MaxSector;
7015         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7016                 mddev->resync_min = mddev->curr_resync_completed;
7017         mddev->curr_resync = 0;
7018         wake_up(&resync_wait);
7019         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7020         md_wakeup_thread(mddev->thread);
7021         return;
7022
7023  interrupted:
7024         /*
7025          * got a signal, exit.
7026          */
7027         printk(KERN_INFO
7028                "md: md_do_sync() got signal ... exiting\n");
7029         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7030         goto out;
7031
7032 }
7033 EXPORT_SYMBOL_GPL(md_do_sync);
7034
7035
7036 static int remove_and_add_spares(mddev_t *mddev)
7037 {
7038         mdk_rdev_t *rdev;
7039         int spares = 0;
7040
7041         mddev->curr_resync_completed = 0;
7042
7043         list_for_each_entry(rdev, &mddev->disks, same_set)
7044                 if (rdev->raid_disk >= 0 &&
7045                     !test_bit(Blocked, &rdev->flags) &&
7046                     (test_bit(Faulty, &rdev->flags) ||
7047                      ! test_bit(In_sync, &rdev->flags)) &&
7048                     atomic_read(&rdev->nr_pending)==0) {
7049                         if (mddev->pers->hot_remove_disk(
7050                                     mddev, rdev->raid_disk)==0) {
7051                                 char nm[20];
7052                                 sprintf(nm,"rd%d", rdev->raid_disk);
7053                                 sysfs_remove_link(&mddev->kobj, nm);
7054                                 rdev->raid_disk = -1;
7055                         }
7056                 }
7057
7058         if (mddev->degraded && !mddev->recovery_disabled) {
7059                 list_for_each_entry(rdev, &mddev->disks, same_set) {
7060                         if (rdev->raid_disk >= 0 &&
7061                             !test_bit(In_sync, &rdev->flags) &&
7062                             !test_bit(Blocked, &rdev->flags))
7063                                 spares++;
7064                         if (rdev->raid_disk < 0
7065                             && !test_bit(Faulty, &rdev->flags)) {
7066                                 rdev->recovery_offset = 0;
7067                                 if (mddev->pers->
7068                                     hot_add_disk(mddev, rdev) == 0) {
7069                                         char nm[20];
7070                                         sprintf(nm, "rd%d", rdev->raid_disk);
7071                                         if (sysfs_create_link(&mddev->kobj,
7072                                                               &rdev->kobj, nm))
7073                                                 /* failure here is OK */;
7074                                         spares++;
7075                                         md_new_event(mddev);
7076                                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
7077                                 } else
7078                                         break;
7079                         }
7080                 }
7081         }
7082         return spares;
7083 }
7084
7085 static void reap_sync_thread(mddev_t *mddev)
7086 {
7087         mdk_rdev_t *rdev;
7088
7089         /* resync has finished, collect result */
7090         md_unregister_thread(mddev->sync_thread);
7091         mddev->sync_thread = NULL;
7092         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7093             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7094                 /* success...*/
7095                 /* activate any spares */
7096                 if (mddev->pers->spare_active(mddev))
7097                         sysfs_notify(&mddev->kobj, NULL,
7098                                      "degraded");
7099         }
7100         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7101             mddev->pers->finish_reshape)
7102                 mddev->pers->finish_reshape(mddev);
7103         md_update_sb(mddev, 1);
7104
7105         /* if array is no-longer degraded, then any saved_raid_disk
7106          * information must be scrapped
7107          */
7108         if (!mddev->degraded)
7109                 list_for_each_entry(rdev, &mddev->disks, same_set)
7110                         rdev->saved_raid_disk = -1;
7111
7112         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7113         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7114         clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7115         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7116         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7117         /* flag recovery needed just to double check */
7118         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7119         sysfs_notify_dirent_safe(mddev->sysfs_action);
7120         md_new_event(mddev);
7121 }
7122
7123 /*
7124  * This routine is regularly called by all per-raid-array threads to
7125  * deal with generic issues like resync and super-block update.
7126  * Raid personalities that don't have a thread (linear/raid0) do not
7127  * need this as they never do any recovery or update the superblock.
7128  *
7129  * It does not do any resync itself, but rather "forks" off other threads
7130  * to do that as needed.
7131  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7132  * "->recovery" and create a thread at ->sync_thread.
7133  * When the thread finishes it sets MD_RECOVERY_DONE
7134  * and wakeups up this thread which will reap the thread and finish up.
7135  * This thread also removes any faulty devices (with nr_pending == 0).
7136  *
7137  * The overall approach is:
7138  *  1/ if the superblock needs updating, update it.
7139  *  2/ If a recovery thread is running, don't do anything else.
7140  *  3/ If recovery has finished, clean up, possibly marking spares active.
7141  *  4/ If there are any faulty devices, remove them.
7142  *  5/ If array is degraded, try to add spares devices
7143  *  6/ If array has spares or is not in-sync, start a resync thread.
7144  */
7145 void md_check_recovery(mddev_t *mddev)
7146 {
7147         if (mddev->bitmap)
7148                 bitmap_daemon_work(mddev);
7149
7150         if (mddev->ro)
7151                 return;
7152
7153         if (signal_pending(current)) {
7154                 if (mddev->pers->sync_request && !mddev->external) {
7155                         printk(KERN_INFO "md: %s in immediate safe mode\n",
7156                                mdname(mddev));
7157                         mddev->safemode = 2;
7158                 }
7159                 flush_signals(current);
7160         }
7161
7162         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7163                 return;
7164         if ( ! (
7165                 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7166                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7167                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7168                 (mddev->external == 0 && mddev->safemode == 1) ||
7169                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7170                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7171                 ))
7172                 return;
7173
7174         if (mddev_trylock(mddev)) {
7175                 int spares = 0;
7176
7177                 if (mddev->ro) {
7178                         /* Only thing we do on a ro array is remove
7179                          * failed devices.
7180                          */
7181                         mdk_rdev_t *rdev;
7182                         list_for_each_entry(rdev, &mddev->disks, same_set)
7183                                 if (rdev->raid_disk >= 0 &&
7184                                     !test_bit(Blocked, &rdev->flags) &&
7185                                     test_bit(Faulty, &rdev->flags) &&
7186                                     atomic_read(&rdev->nr_pending)==0) {
7187                                         if (mddev->pers->hot_remove_disk(
7188                                                     mddev, rdev->raid_disk)==0) {
7189                                                 char nm[20];
7190                                                 sprintf(nm,"rd%d", rdev->raid_disk);
7191                                                 sysfs_remove_link(&mddev->kobj, nm);
7192                                                 rdev->raid_disk = -1;
7193                                         }
7194                                 }
7195                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7196                         goto unlock;
7197                 }
7198
7199                 if (!mddev->external) {
7200                         int did_change = 0;
7201                         spin_lock_irq(&mddev->write_lock);
7202                         if (mddev->safemode &&
7203                             !atomic_read(&mddev->writes_pending) &&
7204                             !mddev->in_sync &&
7205                             mddev->recovery_cp == MaxSector) {
7206                                 mddev->in_sync = 1;
7207                                 did_change = 1;
7208                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7209                         }
7210                         if (mddev->safemode == 1)
7211                                 mddev->safemode = 0;
7212                         spin_unlock_irq(&mddev->write_lock);
7213                         if (did_change)
7214                                 sysfs_notify_dirent_safe(mddev->sysfs_state);
7215                 }
7216
7217                 if (mddev->flags)
7218                         md_update_sb(mddev, 0);
7219
7220                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7221                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7222                         /* resync/recovery still happening */
7223                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7224                         goto unlock;
7225                 }
7226                 if (mddev->sync_thread) {
7227                         reap_sync_thread(mddev);
7228                         goto unlock;
7229                 }
7230                 /* Set RUNNING before clearing NEEDED to avoid
7231                  * any transients in the value of "sync_action".
7232                  */
7233                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7234                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7235                 /* Clear some bits that don't mean anything, but
7236                  * might be left set
7237                  */
7238                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7239                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7240
7241                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7242                         goto unlock;
7243                 /* no recovery is running.
7244                  * remove any failed drives, then
7245                  * add spares if possible.
7246                  * Spare are also removed and re-added, to allow
7247                  * the personality to fail the re-add.
7248                  */
7249
7250                 if (mddev->reshape_position != MaxSector) {
7251                         if (mddev->pers->check_reshape == NULL ||
7252                             mddev->pers->check_reshape(mddev) != 0)
7253                                 /* Cannot proceed */
7254                                 goto unlock;
7255                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7256                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7257                 } else if ((spares = remove_and_add_spares(mddev))) {
7258                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7259                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7260                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7261                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7262                 } else if (mddev->recovery_cp < MaxSector) {
7263                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7264                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7265                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7266                         /* nothing to be done ... */
7267                         goto unlock;
7268
7269                 if (mddev->pers->sync_request) {
7270                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7271                                 /* We are adding a device or devices to an array
7272                                  * which has the bitmap stored on all devices.
7273                                  * So make sure all bitmap pages get written
7274                                  */
7275                                 bitmap_write_all(mddev->bitmap);
7276                         }
7277                         mddev->sync_thread = md_register_thread(md_do_sync,
7278                                                                 mddev,
7279                                                                 "resync");
7280                         if (!mddev->sync_thread) {
7281                                 printk(KERN_ERR "%s: could not start resync"
7282                                         " thread...\n",
7283                                         mdname(mddev));
7284                                 /* leave the spares where they are, it shouldn't hurt */
7285                                 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7286                                 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7287                                 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7288                                 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7289                                 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7290                         } else
7291                                 md_wakeup_thread(mddev->sync_thread);
7292                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7293                         md_new_event(mddev);
7294                 }
7295         unlock:
7296                 if (!mddev->sync_thread) {
7297                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7298                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7299                                                &mddev->recovery))
7300                                 if (mddev->sysfs_action)
7301                                         sysfs_notify_dirent_safe(mddev->sysfs_action);
7302                 }
7303                 mddev_unlock(mddev);
7304         }
7305 }
7306
7307 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7308 {
7309         sysfs_notify_dirent_safe(rdev->sysfs_state);
7310         wait_event_timeout(rdev->blocked_wait,
7311                            !test_bit(Blocked, &rdev->flags),
7312                            msecs_to_jiffies(5000));
7313         rdev_dec_pending(rdev, mddev);
7314 }
7315 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7316
7317 static int md_notify_reboot(struct notifier_block *this,
7318                             unsigned long code, void *x)
7319 {
7320         struct list_head *tmp;
7321         mddev_t *mddev;
7322
7323         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7324
7325                 printk(KERN_INFO "md: stopping all md devices.\n");
7326
7327                 for_each_mddev(mddev, tmp)
7328                         if (mddev_trylock(mddev)) {
7329                                 /* Force a switch to readonly even array
7330                                  * appears to still be in use.  Hence
7331                                  * the '100'.
7332                                  */
7333                                 md_set_readonly(mddev, 100);
7334                                 mddev_unlock(mddev);
7335                         }
7336                 /*
7337                  * certain more exotic SCSI devices are known to be
7338                  * volatile wrt too early system reboots. While the
7339                  * right place to handle this issue is the given
7340                  * driver, we do want to have a safe RAID driver ...
7341                  */
7342                 mdelay(1000*1);
7343         }
7344         return NOTIFY_DONE;
7345 }
7346
7347 static struct notifier_block md_notifier = {
7348         .notifier_call  = md_notify_reboot,
7349         .next           = NULL,
7350         .priority       = INT_MAX, /* before any real devices */
7351 };
7352
7353 static void md_geninit(void)
7354 {
7355         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7356
7357         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7358 }
7359
7360 static int __init md_init(void)
7361 {
7362         int ret = -ENOMEM;
7363
7364         md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7365         if (!md_wq)
7366                 goto err_wq;
7367
7368         md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7369         if (!md_misc_wq)
7370                 goto err_misc_wq;
7371
7372         if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7373                 goto err_md;
7374
7375         if ((ret = register_blkdev(0, "mdp")) < 0)
7376                 goto err_mdp;
7377         mdp_major = ret;
7378
7379         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7380                             md_probe, NULL, NULL);
7381         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7382                             md_probe, NULL, NULL);
7383
7384         register_reboot_notifier(&md_notifier);
7385         raid_table_header = register_sysctl_table(raid_root_table);
7386
7387         md_geninit();
7388         return 0;
7389
7390 err_mdp:
7391         unregister_blkdev(MD_MAJOR, "md");
7392 err_md:
7393         destroy_workqueue(md_misc_wq);
7394 err_misc_wq:
7395         destroy_workqueue(md_wq);
7396 err_wq:
7397         return ret;
7398 }
7399
7400 #ifndef MODULE
7401
7402 /*
7403  * Searches all registered partitions for autorun RAID arrays
7404  * at boot time.
7405  */
7406
7407 static LIST_HEAD(all_detected_devices);
7408 struct detected_devices_node {
7409         struct list_head list;
7410         dev_t dev;
7411 };
7412
7413 void md_autodetect_dev(dev_t dev)
7414 {
7415         struct detected_devices_node *node_detected_dev;
7416
7417         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7418         if (node_detected_dev) {
7419                 node_detected_dev->dev = dev;
7420                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7421         } else {
7422                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7423                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7424         }
7425 }
7426
7427
7428 static void autostart_arrays(int part)
7429 {
7430         mdk_rdev_t *rdev;
7431         struct detected_devices_node *node_detected_dev;
7432         dev_t dev;
7433         int i_scanned, i_passed;
7434
7435         i_scanned = 0;
7436         i_passed = 0;
7437
7438         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7439
7440         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7441                 i_scanned++;
7442                 node_detected_dev = list_entry(all_detected_devices.next,
7443                                         struct detected_devices_node, list);
7444                 list_del(&node_detected_dev->list);
7445                 dev = node_detected_dev->dev;
7446                 kfree(node_detected_dev);
7447                 rdev = md_import_device(dev,0, 90);
7448                 if (IS_ERR(rdev))
7449                         continue;
7450
7451                 if (test_bit(Faulty, &rdev->flags)) {
7452                         MD_BUG();
7453                         continue;
7454                 }
7455                 set_bit(AutoDetected, &rdev->flags);
7456                 list_add(&rdev->same_set, &pending_raid_disks);
7457                 i_passed++;
7458         }
7459
7460         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7461                                                 i_scanned, i_passed);
7462
7463         autorun_devices(part);
7464 }
7465
7466 #endif /* !MODULE */
7467
7468 static __exit void md_exit(void)
7469 {
7470         mddev_t *mddev;
7471         struct list_head *tmp;
7472
7473         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7474         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7475
7476         unregister_blkdev(MD_MAJOR,"md");
7477         unregister_blkdev(mdp_major, "mdp");
7478         unregister_reboot_notifier(&md_notifier);
7479         unregister_sysctl_table(raid_table_header);
7480         remove_proc_entry("mdstat", NULL);
7481         for_each_mddev(mddev, tmp) {
7482                 export_array(mddev);
7483                 mddev->hold_active = 0;
7484         }
7485         destroy_workqueue(md_misc_wq);
7486         destroy_workqueue(md_wq);
7487 }
7488
7489 subsys_initcall(md_init);
7490 module_exit(md_exit)
7491
7492 static int get_ro(char *buffer, struct kernel_param *kp)
7493 {
7494         return sprintf(buffer, "%d", start_readonly);
7495 }
7496 static int set_ro(const char *val, struct kernel_param *kp)
7497 {
7498         char *e;
7499         int num = simple_strtoul(val, &e, 10);
7500         if (*val && (*e == '\0' || *e == '\n')) {
7501                 start_readonly = num;
7502                 return 0;
7503         }
7504         return -EINVAL;
7505 }
7506
7507 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7508 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7509
7510 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7511
7512 EXPORT_SYMBOL(register_md_personality);
7513 EXPORT_SYMBOL(unregister_md_personality);
7514 EXPORT_SYMBOL(md_error);
7515 EXPORT_SYMBOL(md_done_sync);
7516 EXPORT_SYMBOL(md_write_start);
7517 EXPORT_SYMBOL(md_write_end);
7518 EXPORT_SYMBOL(md_register_thread);
7519 EXPORT_SYMBOL(md_unregister_thread);
7520 EXPORT_SYMBOL(md_wakeup_thread);
7521 EXPORT_SYMBOL(md_check_recovery);
7522 MODULE_LICENSE("GPL");
7523 MODULE_DESCRIPTION("MD RAID framework");
7524 MODULE_ALIAS("md");
7525 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);