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/buffer_head.h> /* for invalidate_bdev */
  40 #include <linux/poll.h>
  41 #include <linux/ctype.h>
  42 #include <linux/string.h>
  43 #include <linux/hdreg.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/random.h>
  46 #include <linux/reboot.h>
  47 #include <linux/file.h>
  48 #include <linux/compat.h>
  49 #include <linux/delay.h>
  50 #include <linux/raid/md_p.h>
  51 #include <linux/raid/md_u.h>
  52 #include <linux/slab.h>
  53 #include "md.h"
  54 #include "bitmap.h"
  55
  56 #define DEBUG 0
  57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  58
  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
  71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  72
  73 /*
  74  * Default number of read corrections we'll attempt on an rdev
  75  * before ejecting it from the array. We divide the read error
  76  * count by 2 for every hour elapsed between read errors.
  77  */
  78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  79 /*
  80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  81  * is 1000 KB/sec, so the extra system load does not show up that much.
  82  * Increase it if you want to have more _guaranteed_ speed. Note that
  83  * the RAID driver will use the maximum available bandwidth if the IO
  84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  85  * speed limit - in case reconstruction slows down your system despite
  86  * idle IO detection.
  87  *
  88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  89  * or /sys/block/mdX/md/sync_speed_{min,max}
  90  */
  91
  92 static int sysctl_speed_limit_min = 1000;
  93 static int sysctl_speed_limit_max = 200000;
  94 static inline int speed_min(mddev_t *mddev)
  95 {
  96         return mddev->sync_speed_min ?
  97                 mddev->sync_speed_min : sysctl_speed_limit_min;
  98 }
  99
 100 static inline int speed_max(mddev_t *mddev)
 101 {
 102         return mddev->sync_speed_max ?
 103                 mddev->sync_speed_max : sysctl_speed_limit_max;
 104 }
 105
 106 static struct ctl_table_header *raid_table_header;
 107
 108 static ctl_table raid_table[] = {
 109         {
 110                 .procname       = "speed_limit_min",
 111                 .data           = &sysctl_speed_limit_min,
 112                 .maxlen         = sizeof(int),
 113                 .mode           = S_IRUGO|S_IWUSR,
 114                 .proc_handler   = proc_dointvec,
 115         },
 116         {
 117                 .procname       = "speed_limit_max",
 118                 .data           = &sysctl_speed_limit_max,
 119                 .maxlen         = sizeof(int),
 120                 .mode           = S_IRUGO|S_IWUSR,
 121                 .proc_handler   = proc_dointvec,
 122         },
 123         { }
 124 };
 125
 126 static ctl_table raid_dir_table[] = {
 127         {
 128                 .procname       = "raid",
 129                 .maxlen         = 0,
 130                 .mode           = S_IRUGO|S_IXUGO,
 131                 .child          = raid_table,
 132         },
 133         { }
 134 };
 135
 136 static ctl_table raid_root_table[] = {
 137         {
 138                 .procname       = "dev",
 139                 .maxlen         = 0,
 140                 .mode           = 0555,
 141                 .child          = raid_dir_table,
 142         },
 143         {  }
 144 };
 145
 146 static const struct block_device_operations md_fops;
 147
 148 static int start_readonly;
 149
 150 /*
 151  * We have a system wide 'event count' that is incremented
 152  * on any 'interesting' event, and readers of /proc/mdstat
 153  * can use 'poll' or 'select' to find out when the event
 154  * count increases.
 155  *
 156  * Events are:
 157  *  start array, stop array, error, add device, remove device,
 158  *  start build, activate spare
 159  */
 160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 161 static atomic_t md_event_count;
 162 void md_new_event(mddev_t *mddev)
 163 {
 164         atomic_inc(&md_event_count);
 165         wake_up(&md_event_waiters);
 166 }
 167 EXPORT_SYMBOL_GPL(md_new_event);
 168
 169 /* Alternate version that can be called from interrupts
 170  * when calling sysfs_notify isn't needed.
 171  */
 172 static void md_new_event_inintr(mddev_t *mddev)
 173 {
 174         atomic_inc(&md_event_count);
 175         wake_up(&md_event_waiters);
 176 }
 177
 178 /*
 179  * Enables to iterate over all existing md arrays
 180  * all_mddevs_lock protects this list.
 181  */
 182 static LIST_HEAD(all_mddevs);
 183 static DEFINE_SPINLOCK(all_mddevs_lock);
 184
 185
 186 /*
 187  * iterates through all used mddevs in the system.
 188  * We take care to grab the all_mddevs_lock whenever navigating
 189  * the list, and to always hold a refcount when unlocked.
 190  * Any code which breaks out of this loop while own
 191  * a reference to the current mddev and must mddev_put it.
 192  */
 193 #define for_each_mddev(mddev,tmp)                                       \
 194                                                                         \
 195         for (({ spin_lock(&all_mddevs_lock);                            \
 196                 tmp = all_mddevs.next;                                  \
 197                 mddev = NULL;});                                        \
 198              ({ if (tmp != &all_mddevs)                                 \
 199                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 200                 spin_unlock(&all_mddevs_lock);                          \
 201                 if (mddev) mddev_put(mddev);                            \
 202                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 203                 tmp != &all_mddevs;});                                  \
 204              ({ spin_lock(&all_mddevs_lock);                            \
 205                 tmp = tmp->next;})                                      \
 206                 )
 207
 208
 209 /* Rather than calling directly into the personality make_request function,
 210  * IO requests come here first so that we can check if the device is
 211  * being suspended pending a reconfiguration.
 212  * We hold a refcount over the call to ->make_request.  By the time that
 213  * call has finished, the bio has been linked into some internal structure
 214  * and so is visible to ->quiesce(), so we don't need the refcount any more.
 215  */
 216 static int md_make_request(struct request_queue *q, struct bio *bio)
 217 {
 218         const int rw = bio_data_dir(bio);
 219         mddev_t *mddev = q->queuedata;
 220         int rv;
 221         int cpu;
 222
 223         if (mddev == NULL || mddev->pers == NULL) {
 224                 bio_io_error(bio);
 225                 return 0;
 226         }
 227         rcu_read_lock();
 228         if (mddev->suspended || mddev->barrier) {
 229                 DEFINE_WAIT(__wait);
 230                 for (;;) {
 231                         prepare_to_wait(&mddev->sb_wait, &__wait,
 232                                         TASK_UNINTERRUPTIBLE);
 233                         if (!mddev->suspended && !mddev->barrier)
 234                                 break;
 235                         rcu_read_unlock();
 236                         schedule();
 237                         rcu_read_lock();
 238                 }
 239                 finish_wait(&mddev->sb_wait, &__wait);
 240         }
 241         atomic_inc(&mddev->active_io);
 242         rcu_read_unlock();
 243
 244         rv = mddev->pers->make_request(mddev, bio);
 245
 246         cpu = part_stat_lock();
 247         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 248         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 249                       bio_sectors(bio));
 250         part_stat_unlock();
 251
 252         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 253                 wake_up(&mddev->sb_wait);
 254
 255         return rv;
 256 }
 257
 258 /* mddev_suspend makes sure no new requests are submitted
 259  * to the device, and that any requests that have been submitted
 260  * are completely handled.
 261  * Once ->stop is called and completes, the module will be completely
 262  * unused.
 263  */
 264 static void mddev_suspend(mddev_t *mddev)
 265 {
 266         BUG_ON(mddev->suspended);
 267         mddev->suspended = 1;
 268         synchronize_rcu();
 269         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 270         mddev->pers->quiesce(mddev, 1);
 271 }
 272
 273 static void mddev_resume(mddev_t *mddev)
 274 {
 275         mddev->suspended = 0;
 276         wake_up(&mddev->sb_wait);
 277         mddev->pers->quiesce(mddev, 0);
 278 }
 279
 280 int mddev_congested(mddev_t *mddev, int bits)
 281 {
 282         if (mddev->barrier)
 283                 return 1;
 284         return mddev->suspended;
 285 }
 286 EXPORT_SYMBOL(mddev_congested);
 287
 288 /*
 289  * Generic barrier handling for md
 290  */
 291
 292 #define POST_REQUEST_BARRIER ((void*)1)
 293
 294 static void md_end_barrier(struct bio *bio, int err)
 295 {
 296         mdk_rdev_t *rdev = bio->bi_private;
 297         mddev_t *mddev = rdev->mddev;
 298         if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
 299                 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
 300
 301         rdev_dec_pending(rdev, mddev);
 302
 303         if (atomic_dec_and_test(&mddev->flush_pending)) {
 304                 if (mddev->barrier == POST_REQUEST_BARRIER) {
 305                         /* This was a post-request barrier */
 306                         mddev->barrier = NULL;
 307                         wake_up(&mddev->sb_wait);
 308                 } else
 309                         /* The pre-request barrier has finished */
 310                         schedule_work(&mddev->barrier_work);
 311         }
 312         bio_put(bio);
 313 }
 314
 315 static void submit_barriers(mddev_t *mddev)
 316 {
 317         mdk_rdev_t *rdev;
 318
 319         rcu_read_lock();
 320         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
 321                 if (rdev->raid_disk >= 0 &&
 322                     !test_bit(Faulty, &rdev->flags)) {
 323                         /* Take two references, one is dropped
 324                          * when request finishes, one after
 325                          * we reclaim rcu_read_lock
 326                          */
 327                         struct bio *bi;
 328                         atomic_inc(&rdev->nr_pending);
 329                         atomic_inc(&rdev->nr_pending);
 330                         rcu_read_unlock();
 331                         bi = bio_alloc(GFP_KERNEL, 0);
 332                         bi->bi_end_io = md_end_barrier;
 333                         bi->bi_private = rdev;
 334                         bi->bi_bdev = rdev->bdev;
 335                         atomic_inc(&mddev->flush_pending);
 336                         submit_bio(WRITE_BARRIER, bi);
 337                         rcu_read_lock();
 338                         rdev_dec_pending(rdev, mddev);
 339                 }
 340         rcu_read_unlock();
 341 }
 342
 343 static void md_submit_barrier(struct work_struct *ws)
 344 {
 345         mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
 346         struct bio *bio = mddev->barrier;
 347
 348         atomic_set(&mddev->flush_pending, 1);
 349
 350         if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
 351                 bio_endio(bio, -EOPNOTSUPP);
 352         else if (bio->bi_size == 0)
 353                 /* an empty barrier - all done */
 354                 bio_endio(bio, 0);
 355         else {
 356                 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
 357                 if (mddev->pers->make_request(mddev, bio))
 358                         generic_make_request(bio);
 359                 mddev->barrier = POST_REQUEST_BARRIER;
 360                 submit_barriers(mddev);
 361         }
 362         if (atomic_dec_and_test(&mddev->flush_pending)) {
 363                 mddev->barrier = NULL;
 364                 wake_up(&mddev->sb_wait);
 365         }
 366 }
 367
 368 void md_barrier_request(mddev_t *mddev, struct bio *bio)
 369 {
 370         spin_lock_irq(&mddev->write_lock);
 371         wait_event_lock_irq(mddev->sb_wait,
 372                             !mddev->barrier,
 373                             mddev->write_lock, /*nothing*/);
 374         mddev->barrier = bio;
 375         spin_unlock_irq(&mddev->write_lock);
 376
 377         atomic_set(&mddev->flush_pending, 1);
 378         INIT_WORK(&mddev->barrier_work, md_submit_barrier);
 379
 380         submit_barriers(mddev);
 381
 382         if (atomic_dec_and_test(&mddev->flush_pending))
 383                 schedule_work(&mddev->barrier_work);
 384 }
 385 EXPORT_SYMBOL(md_barrier_request);
 386
 387 static inline mddev_t *mddev_get(mddev_t *mddev)
 388 {
 389         atomic_inc(&mddev->active);
 390         return mddev;
 391 }
 392
 393 static void mddev_delayed_delete(struct work_struct *ws);
 394
 395 static void mddev_put(mddev_t *mddev)
 396 {
 397         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 398                 return;
 399         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 400             mddev->ctime == 0 && !mddev->hold_active) {
 401                 /* Array is not configured at all, and not held active,
 402                  * so destroy it */
 403                 list_del(&mddev->all_mddevs);
 404                 if (mddev->gendisk) {
 405                         /* we did a probe so need to clean up.
 406                          * Call schedule_work inside the spinlock
 407                          * so that flush_scheduled_work() after
 408                          * mddev_find will succeed in waiting for the
 409                          * work to be done.
 410                          */
 411                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 412                         schedule_work(&mddev->del_work);
 413                 } else
 414                         kfree(mddev);
 415         }
 416         spin_unlock(&all_mddevs_lock);
 417 }
 418
 419 static void mddev_init(mddev_t *mddev)
 420 {
 421         mutex_init(&mddev->open_mutex);
 422         mutex_init(&mddev->reconfig_mutex);
 423         mutex_init(&mddev->bitmap_info.mutex);
 424         INIT_LIST_HEAD(&mddev->disks);
 425         INIT_LIST_HEAD(&mddev->all_mddevs);
 426         init_timer(&mddev->safemode_timer);
 427         atomic_set(&mddev->active, 1);
 428         atomic_set(&mddev->openers, 0);
 429         atomic_set(&mddev->active_io, 0);
 430         spin_lock_init(&mddev->write_lock);
 431         atomic_set(&mddev->flush_pending, 0);
 432         init_waitqueue_head(&mddev->sb_wait);
 433         init_waitqueue_head(&mddev->recovery_wait);
 434         mddev->reshape_position = MaxSector;
 435         mddev->resync_min = 0;
 436         mddev->resync_max = MaxSector;
 437         mddev->level = LEVEL_NONE;
 438 }
 439
 440 static mddev_t * mddev_find(dev_t unit)
 441 {
 442         mddev_t *mddev, *new = NULL;
 443
 444  retry:
 445         spin_lock(&all_mddevs_lock);
 446
 447         if (unit) {
 448                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 449                         if (mddev->unit == unit) {
 450                                 mddev_get(mddev);
 451                                 spin_unlock(&all_mddevs_lock);
 452                                 kfree(new);
 453                                 return mddev;
 454                         }
 455
 456                 if (new) {
 457                         list_add(&new->all_mddevs, &all_mddevs);
 458                         spin_unlock(&all_mddevs_lock);
 459                         new->hold_active = UNTIL_IOCTL;
 460                         return new;
 461                 }
 462         } else if (new) {
 463                 /* find an unused unit number */
 464                 static int next_minor = 512;
 465                 int start = next_minor;
 466                 int is_free = 0;
 467                 int dev = 0;
 468                 while (!is_free) {
 469                         dev = MKDEV(MD_MAJOR, next_minor);
 470                         next_minor++;
 471                         if (next_minor > MINORMASK)
 472                                 next_minor = 0;
 473                         if (next_minor == start) {
 474                                 /* Oh dear, all in use. */
 475                                 spin_unlock(&all_mddevs_lock);
 476                                 kfree(new);
 477                                 return NULL;
 478                         }
 479
 480                         is_free = 1;
 481                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 482                                 if (mddev->unit == dev) {
 483                                         is_free = 0;
 484                                         break;
 485                                 }
 486                 }
 487                 new->unit = dev;
 488                 new->md_minor = MINOR(dev);
 489                 new->hold_active = UNTIL_STOP;
 490                 list_add(&new->all_mddevs, &all_mddevs);
 491                 spin_unlock(&all_mddevs_lock);
 492                 return new;
 493         }
 494         spin_unlock(&all_mddevs_lock);
 495
 496         new = kzalloc(sizeof(*new), GFP_KERNEL);
 497         if (!new)
 498                 return NULL;
 499
 500         new->unit = unit;
 501         if (MAJOR(unit) == MD_MAJOR)
 502                 new->md_minor = MINOR(unit);
 503         else
 504                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 505
 506         mddev_init(new);
 507
 508         goto retry;
 509 }
 510
 511 static inline int mddev_lock(mddev_t * mddev)
 512 {
 513         return mutex_lock_interruptible(&mddev->reconfig_mutex);
 514 }
 515
 516 static inline int mddev_is_locked(mddev_t *mddev)
 517 {
 518         return mutex_is_locked(&mddev->reconfig_mutex);
 519 }
 520
 521 static inline int mddev_trylock(mddev_t * mddev)
 522 {
 523         return mutex_trylock(&mddev->reconfig_mutex);
 524 }
 525
 526 static struct attribute_group md_redundancy_group;
 527
 528 static void mddev_unlock(mddev_t * mddev)
 529 {
 530         if (mddev->to_remove) {
 531                 /* These cannot be removed under reconfig_mutex as
 532                  * an access to the files will try to take reconfig_mutex
 533                  * while holding the file unremovable, which leads to
 534                  * a deadlock.
 535                  * So hold open_mutex instead - we are allowed to take
 536                  * it while holding reconfig_mutex, and md_run can
 537                  * use it to wait for the remove to complete.
 538                  */
 539                 struct attribute_group *to_remove = mddev->to_remove;
 540                 mddev->to_remove = NULL;
 541                 mutex_lock(&mddev->open_mutex);
 542                 mutex_unlock(&mddev->reconfig_mutex);
 543
 544                 if (to_remove != &md_redundancy_group)
 545                         sysfs_remove_group(&mddev->kobj, to_remove);
 546                 if (mddev->pers == NULL ||
 547                     mddev->pers->sync_request == NULL) {
 548                         sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 549                         if (mddev->sysfs_action)
 550                                 sysfs_put(mddev->sysfs_action);
 551                         mddev->sysfs_action = NULL;
 552                 }
 553                 mutex_unlock(&mddev->open_mutex);
 554         } else
 555                 mutex_unlock(&mddev->reconfig_mutex);
 556
 557         md_wakeup_thread(mddev->thread);
 558 }
 559
 560 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 561 {
 562         mdk_rdev_t *rdev;
 563
 564         list_for_each_entry(rdev, &mddev->disks, same_set)
 565                 if (rdev->desc_nr == nr)
 566                         return rdev;
 567
 568         return NULL;
 569 }
 570
 571 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 572 {
 573         mdk_rdev_t *rdev;
 574
 575         list_for_each_entry(rdev, &mddev->disks, same_set)
 576                 if (rdev->bdev->bd_dev == dev)
 577                         return rdev;
 578
 579         return NULL;
 580 }
 581
 582 static struct mdk_personality *find_pers(int level, char *clevel)
 583 {
 584         struct mdk_personality *pers;
 585         list_for_each_entry(pers, &pers_list, list) {
 586                 if (level != LEVEL_NONE && pers->level == level)
 587                         return pers;
 588                 if (strcmp(pers->name, clevel)==0)
 589                         return pers;
 590         }
 591         return NULL;
 592 }
 593
 594 /* return the offset of the super block in 512byte sectors */
 595 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 596 {
 597         sector_t num_sectors = bdev->bd_inode->i_size / 512;
 598         return MD_NEW_SIZE_SECTORS(num_sectors);
 599 }
 600
 601 static int alloc_disk_sb(mdk_rdev_t * rdev)
 602 {
 603         if (rdev->sb_page)
 604                 MD_BUG();
 605
 606         rdev->sb_page = alloc_page(GFP_KERNEL);
 607         if (!rdev->sb_page) {
 608                 printk(KERN_ALERT "md: out of memory.\n");
 609                 return -ENOMEM;
 610         }
 611
 612         return 0;
 613 }
 614
 615 static void free_disk_sb(mdk_rdev_t * rdev)
 616 {
 617         if (rdev->sb_page) {
 618                 put_page(rdev->sb_page);
 619                 rdev->sb_loaded = 0;
 620                 rdev->sb_page = NULL;
 621                 rdev->sb_start = 0;
 622                 rdev->sectors = 0;
 623         }
 624 }
 625
 626
 627 static void super_written(struct bio *bio, int error)
 628 {
 629         mdk_rdev_t *rdev = bio->bi_private;
 630         mddev_t *mddev = rdev->mddev;
 631
 632         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 633                 printk("md: super_written gets error=%d, uptodate=%d\n",
 634                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 635                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 636                 md_error(mddev, rdev);
 637         }
 638
 639         if (atomic_dec_and_test(&mddev->pending_writes))
 640                 wake_up(&mddev->sb_wait);
 641         bio_put(bio);
 642 }
 643
 644 static void super_written_barrier(struct bio *bio, int error)
 645 {
 646         struct bio *bio2 = bio->bi_private;
 647         mdk_rdev_t *rdev = bio2->bi_private;
 648         mddev_t *mddev = rdev->mddev;
 649
 650         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 651             error == -EOPNOTSUPP) {
 652                 unsigned long flags;
 653                 /* barriers don't appear to be supported :-( */
 654                 set_bit(BarriersNotsupp, &rdev->flags);
 655                 mddev->barriers_work = 0;
 656                 spin_lock_irqsave(&mddev->write_lock, flags);
 657                 bio2->bi_next = mddev->biolist;
 658                 mddev->biolist = bio2;
 659                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 660                 wake_up(&mddev->sb_wait);
 661                 bio_put(bio);
 662         } else {
 663                 bio_put(bio2);
 664                 bio->bi_private = rdev;
 665                 super_written(bio, error);
 666         }
 667 }
 668
 669 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 670                    sector_t sector, int size, struct page *page)
 671 {
 672         /* write first size bytes of page to sector of rdev
 673          * Increment mddev->pending_writes before returning
 674          * and decrement it on completion, waking up sb_wait
 675          * if zero is reached.
 676          * If an error occurred, call md_error
 677          *
 678          * As we might need to resubmit the request if BIO_RW_BARRIER
 679          * causes ENOTSUPP, we allocate a spare bio...
 680          */
 681         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 682         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
 683
 684         bio->bi_bdev = rdev->bdev;
 685         bio->bi_sector = sector;
 686         bio_add_page(bio, page, size, 0);
 687         bio->bi_private = rdev;
 688         bio->bi_end_io = super_written;
 689         bio->bi_rw = rw;
 690
 691         atomic_inc(&mddev->pending_writes);
 692         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 693                 struct bio *rbio;
 694                 rw |= (1<<BIO_RW_BARRIER);
 695                 rbio = bio_clone(bio, GFP_NOIO);
 696                 rbio->bi_private = bio;
 697                 rbio->bi_end_io = super_written_barrier;
 698                 submit_bio(rw, rbio);
 699         } else
 700                 submit_bio(rw, bio);
 701 }
 702
 703 void md_super_wait(mddev_t *mddev)
 704 {
 705         /* wait for all superblock writes that were scheduled to complete.
 706          * if any had to be retried (due to BARRIER problems), retry them
 707          */
 708         DEFINE_WAIT(wq);
 709         for(;;) {
 710                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 711                 if (atomic_read(&mddev->pending_writes)==0)
 712                         break;
 713                 while (mddev->biolist) {
 714                         struct bio *bio;
 715                         spin_lock_irq(&mddev->write_lock);
 716                         bio = mddev->biolist;
 717                         mddev->biolist = bio->bi_next ;
 718                         bio->bi_next = NULL;
 719                         spin_unlock_irq(&mddev->write_lock);
 720                         submit_bio(bio->bi_rw, bio);
 721                 }
 722                 schedule();
 723         }
 724         finish_wait(&mddev->sb_wait, &wq);
 725 }
 726
 727 static void bi_complete(struct bio *bio, int error)
 728 {
 729         complete((struct completion*)bio->bi_private);
 730 }
 731
 732 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 733                    struct page *page, int rw)
 734 {
 735         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 736         struct completion event;
 737         int ret;
 738
 739         rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
 740
 741         bio->bi_bdev = bdev;
 742         bio->bi_sector = sector;
 743         bio_add_page(bio, page, size, 0);
 744         init_completion(&event);
 745         bio->bi_private = &event;
 746         bio->bi_end_io = bi_complete;
 747         submit_bio(rw, bio);
 748         wait_for_completion(&event);
 749
 750         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 751         bio_put(bio);
 752         return ret;
 753 }
 754 EXPORT_SYMBOL_GPL(sync_page_io);
 755
 756 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 757 {
 758         char b[BDEVNAME_SIZE];
 759         if (!rdev->sb_page) {
 760                 MD_BUG();
 761                 return -EINVAL;
 762         }
 763         if (rdev->sb_loaded)
 764                 return 0;
 765
 766
 767         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
 768                 goto fail;
 769         rdev->sb_loaded = 1;
 770         return 0;
 771
 772 fail:
 773         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 774                 bdevname(rdev->bdev,b));
 775         return -EINVAL;
 776 }
 777
 778 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 779 {
 780         return  sb1->set_uuid0 == sb2->set_uuid0 &&
 781                 sb1->set_uuid1 == sb2->set_uuid1 &&
 782                 sb1->set_uuid2 == sb2->set_uuid2 &&
 783                 sb1->set_uuid3 == sb2->set_uuid3;
 784 }
 785
 786 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 787 {
 788         int ret;
 789         mdp_super_t *tmp1, *tmp2;
 790
 791         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 792         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 793
 794         if (!tmp1 || !tmp2) {
 795                 ret = 0;
 796                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 797                 goto abort;
 798         }
 799
 800         *tmp1 = *sb1;
 801         *tmp2 = *sb2;
 802
 803         /*
 804          * nr_disks is not constant
 805          */
 806         tmp1->nr_disks = 0;
 807         tmp2->nr_disks = 0;
 808
 809         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 810 abort:
 811         kfree(tmp1);
 812         kfree(tmp2);
 813         return ret;
 814 }
 815
 816
 817 static u32 md_csum_fold(u32 csum)
 818 {
 819         csum = (csum & 0xffff) + (csum >> 16);
 820         return (csum & 0xffff) + (csum >> 16);
 821 }
 822
 823 static unsigned int calc_sb_csum(mdp_super_t * sb)
 824 {
 825         u64 newcsum = 0;
 826         u32 *sb32 = (u32*)sb;
 827         int i;
 828         unsigned int disk_csum, csum;
 829
 830         disk_csum = sb->sb_csum;
 831         sb->sb_csum = 0;
 832
 833         for (i = 0; i < MD_SB_BYTES/4 ; i++)
 834                 newcsum += sb32[i];
 835         csum = (newcsum & 0xffffffff) + (newcsum>>32);
 836
 837
 838 #ifdef CONFIG_ALPHA
 839         /* This used to use csum_partial, which was wrong for several
 840          * reasons including that different results are returned on
 841          * different architectures.  It isn't critical that we get exactly
 842          * the same return value as before (we always csum_fold before
 843          * testing, and that removes any differences).  However as we
 844          * know that csum_partial always returned a 16bit value on
 845          * alphas, do a fold to maximise conformity to previous behaviour.
 846          */
 847         sb->sb_csum = md_csum_fold(disk_csum);
 848 #else
 849         sb->sb_csum = disk_csum;
 850 #endif
 851         return csum;
 852 }
 853
 854
 855 /*
 856  * Handle superblock details.
 857  * We want to be able to handle multiple superblock formats
 858  * so we have a common interface to them all, and an array of
 859  * different handlers.
 860  * We rely on user-space to write the initial superblock, and support
 861  * reading and updating of superblocks.
 862  * Interface methods are:
 863  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 864  *      loads and validates a superblock on dev.
 865  *      if refdev != NULL, compare superblocks on both devices
 866  *    Return:
 867  *      0 - dev has a superblock that is compatible with refdev
 868  *      1 - dev has a superblock that is compatible and newer than refdev
 869  *          so dev should be used as the refdev in future
 870  *     -EINVAL superblock incompatible or invalid
 871  *     -othererror e.g. -EIO
 872  *
 873  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 874  *      Verify that dev is acceptable into mddev.
 875  *       The first time, mddev->raid_disks will be 0, and data from
 876  *       dev should be merged in.  Subsequent calls check that dev
 877  *       is new enough.  Return 0 or -EINVAL
 878  *
 879  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 880  *     Update the superblock for rdev with data in mddev
 881  *     This does not write to disc.
 882  *
 883  */
 884
 885 struct super_type  {
 886         char                *name;
 887         struct module       *owner;
 888         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 889                                           int minor_version);
 890         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 891         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 892         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 893                                                 sector_t num_sectors);
 894 };
 895
 896 /*
 897  * Check that the given mddev has no bitmap.
 898  *
 899  * This function is called from the run method of all personalities that do not
 900  * support bitmaps. It prints an error message and returns non-zero if mddev
 901  * has a bitmap. Otherwise, it returns 0.
 902  *
 903  */
 904 int md_check_no_bitmap(mddev_t *mddev)
 905 {
 906         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
 907                 return 0;
 908         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
 909                 mdname(mddev), mddev->pers->name);
 910         return 1;
 911 }
 912 EXPORT_SYMBOL(md_check_no_bitmap);
 913
 914 /*
 915  * load_super for 0.90.0
 916  */
 917 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 918 {
 919         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 920         mdp_super_t *sb;
 921         int ret;
 922
 923         /*
 924          * Calculate the position of the superblock (512byte sectors),
 925          * it's at the end of the disk.
 926          *
 927          * It also happens to be a multiple of 4Kb.
 928          */
 929         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 930
 931         ret = read_disk_sb(rdev, MD_SB_BYTES);
 932         if (ret) return ret;
 933
 934         ret = -EINVAL;
 935
 936         bdevname(rdev->bdev, b);
 937         sb = (mdp_super_t*)page_address(rdev->sb_page);
 938
 939         if (sb->md_magic != MD_SB_MAGIC) {
 940                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 941                        b);
 942                 goto abort;
 943         }
 944
 945         if (sb->major_version != 0 ||
 946             sb->minor_version < 90 ||
 947             sb->minor_version > 91) {
 948                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 949                         sb->major_version, sb->minor_version,
 950                         b);
 951                 goto abort;
 952         }
 953
 954         if (sb->raid_disks <= 0)
 955                 goto abort;
 956
 957         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
 958                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 959                         b);
 960                 goto abort;
 961         }
 962
 963         rdev->preferred_minor = sb->md_minor;
 964         rdev->data_offset = 0;
 965         rdev->sb_size = MD_SB_BYTES;
 966
 967         if (sb->level == LEVEL_MULTIPATH)
 968                 rdev->desc_nr = -1;
 969         else
 970                 rdev->desc_nr = sb->this_disk.number;
 971
 972         if (!refdev) {
 973                 ret = 1;
 974         } else {
 975                 __u64 ev1, ev2;
 976                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 977                 if (!uuid_equal(refsb, sb)) {
 978                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 979                                 b, bdevname(refdev->bdev,b2));
 980                         goto abort;
 981                 }
 982                 if (!sb_equal(refsb, sb)) {
 983                         printk(KERN_WARNING "md: %s has same UUID"
 984                                " but different superblock to %s\n",
 985                                b, bdevname(refdev->bdev, b2));
 986                         goto abort;
 987                 }
 988                 ev1 = md_event(sb);
 989                 ev2 = md_event(refsb);
 990                 if (ev1 > ev2)
 991                         ret = 1;
 992                 else
 993                         ret = 0;
 994         }
 995         rdev->sectors = rdev->sb_start;
 996
 997         if (rdev->sectors < sb->size * 2 && sb->level > 1)
 998                 /* "this cannot possibly happen" ... */
 999                 ret = -EINVAL;
1000
1001  abort:
1002         return ret;
1003 }
1004
1005 /*
1006  * validate_super for 0.90.0
1007  */
1008 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1009 {
1010         mdp_disk_t *desc;
1011         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1012         __u64 ev1 = md_event(sb);
1013
1014         rdev->raid_disk = -1;
1015         clear_bit(Faulty, &rdev->flags);
1016         clear_bit(In_sync, &rdev->flags);
1017         clear_bit(WriteMostly, &rdev->flags);
1018         clear_bit(BarriersNotsupp, &rdev->flags);
1019
1020         if (mddev->raid_disks == 0) {
1021                 mddev->major_version = 0;
1022                 mddev->minor_version = sb->minor_version;
1023                 mddev->patch_version = sb->patch_version;
1024                 mddev->external = 0;
1025                 mddev->chunk_sectors = sb->chunk_size >> 9;
1026                 mddev->ctime = sb->ctime;
1027                 mddev->utime = sb->utime;
1028                 mddev->level = sb->level;
1029                 mddev->clevel[0] = 0;
1030                 mddev->layout = sb->layout;
1031                 mddev->raid_disks = sb->raid_disks;
1032                 mddev->dev_sectors = sb->size * 2;
1033                 mddev->events = ev1;
1034                 mddev->bitmap_info.offset = 0;
1035                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1036
1037                 if (mddev->minor_version >= 91) {
1038                         mddev->reshape_position = sb->reshape_position;
1039                         mddev->delta_disks = sb->delta_disks;
1040                         mddev->new_level = sb->new_level;
1041                         mddev->new_layout = sb->new_layout;
1042                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1043                 } else {
1044                         mddev->reshape_position = MaxSector;
1045                         mddev->delta_disks = 0;
1046                         mddev->new_level = mddev->level;
1047                         mddev->new_layout = mddev->layout;
1048                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1049                 }
1050
1051                 if (sb->state & (1<<MD_SB_CLEAN))
1052                         mddev->recovery_cp = MaxSector;
1053                 else {
1054                         if (sb->events_hi == sb->cp_events_hi &&
1055                                 sb->events_lo == sb->cp_events_lo) {
1056                                 mddev->recovery_cp = sb->recovery_cp;
1057                         } else
1058                                 mddev->recovery_cp = 0;
1059                 }
1060
1061                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1062                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1063                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1064                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1065
1066                 mddev->max_disks = MD_SB_DISKS;
1067
1068                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1069                     mddev->bitmap_info.file == NULL)
1070                         mddev->bitmap_info.offset =
1071                                 mddev->bitmap_info.default_offset;
1072
1073         } else if (mddev->pers == NULL) {
1074                 /* Insist on good event counter while assembling, except
1075                  * for spares (which don't need an event count) */
1076                 ++ev1;
1077                 if (sb->disks[rdev->desc_nr].state & (
1078                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1079                         if (ev1 < mddev->events)
1080                                 return -EINVAL;
1081         } else if (mddev->bitmap) {
1082                 /* if adding to array with a bitmap, then we can accept an
1083                  * older device ... but not too old.
1084                  */
1085                 if (ev1 < mddev->bitmap->events_cleared)
1086                         return 0;
1087         } else {
1088                 if (ev1 < mddev->events)
1089                         /* just a hot-add of a new device, leave raid_disk at -1 */
1090                         return 0;
1091         }
1092
1093         if (mddev->level != LEVEL_MULTIPATH) {
1094                 desc = sb->disks + rdev->desc_nr;
1095
1096                 if (desc->state & (1<<MD_DISK_FAULTY))
1097                         set_bit(Faulty, &rdev->flags);
1098                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1099                             desc->raid_disk < mddev->raid_disks */) {
1100                         set_bit(In_sync, &rdev->flags);
1101                         rdev->raid_disk = desc->raid_disk;
1102                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1103                         /* active but not in sync implies recovery up to
1104                          * reshape position.  We don't know exactly where
1105                          * that is, so set to zero for now */
1106                         if (mddev->minor_version >= 91) {
1107                                 rdev->recovery_offset = 0;
1108                                 rdev->raid_disk = desc->raid_disk;
1109                         }
1110                 }
1111                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1112                         set_bit(WriteMostly, &rdev->flags);
1113         } else /* MULTIPATH are always insync */
1114                 set_bit(In_sync, &rdev->flags);
1115         return 0;
1116 }
1117
1118 /*
1119  * sync_super for 0.90.0
1120  */
1121 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1122 {
1123         mdp_super_t *sb;
1124         mdk_rdev_t *rdev2;
1125         int next_spare = mddev->raid_disks;
1126
1127
1128         /* make rdev->sb match mddev data..
1129          *
1130          * 1/ zero out disks
1131          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1132          * 3/ any empty disks < next_spare become removed
1133          *
1134          * disks[0] gets initialised to REMOVED because
1135          * we cannot be sure from other fields if it has
1136          * been initialised or not.
1137          */
1138         int i;
1139         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1140
1141         rdev->sb_size = MD_SB_BYTES;
1142
1143         sb = (mdp_super_t*)page_address(rdev->sb_page);
1144
1145         memset(sb, 0, sizeof(*sb));
1146
1147         sb->md_magic = MD_SB_MAGIC;
1148         sb->major_version = mddev->major_version;
1149         sb->patch_version = mddev->patch_version;
1150         sb->gvalid_words  = 0; /* ignored */
1151         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1152         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1153         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1154         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1155
1156         sb->ctime = mddev->ctime;
1157         sb->level = mddev->level;
1158         sb->size = mddev->dev_sectors / 2;
1159         sb->raid_disks = mddev->raid_disks;
1160         sb->md_minor = mddev->md_minor;
1161         sb->not_persistent = 0;
1162         sb->utime = mddev->utime;
1163         sb->state = 0;
1164         sb->events_hi = (mddev->events>>32);
1165         sb->events_lo = (u32)mddev->events;
1166
1167         if (mddev->reshape_position == MaxSector)
1168                 sb->minor_version = 90;
1169         else {
1170                 sb->minor_version = 91;
1171                 sb->reshape_position = mddev->reshape_position;
1172                 sb->new_level = mddev->new_level;
1173                 sb->delta_disks = mddev->delta_disks;
1174                 sb->new_layout = mddev->new_layout;
1175                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1176         }
1177         mddev->minor_version = sb->minor_version;
1178         if (mddev->in_sync)
1179         {
1180                 sb->recovery_cp = mddev->recovery_cp;
1181                 sb->cp_events_hi = (mddev->events>>32);
1182                 sb->cp_events_lo = (u32)mddev->events;
1183                 if (mddev->recovery_cp == MaxSector)
1184                         sb->state = (1<< MD_SB_CLEAN);
1185         } else
1186                 sb->recovery_cp = 0;
1187
1188         sb->layout = mddev->layout;
1189         sb->chunk_size = mddev->chunk_sectors << 9;
1190
1191         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1192                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1193
1194         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1195         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1196                 mdp_disk_t *d;
1197                 int desc_nr;
1198                 int is_active = test_bit(In_sync, &rdev2->flags);
1199
1200                 if (rdev2->raid_disk >= 0 &&
1201                     sb->minor_version >= 91)
1202                         /* we have nowhere to store the recovery_offset,
1203                          * but if it is not below the reshape_position,
1204                          * we can piggy-back on that.
1205                          */
1206                         is_active = 1;
1207                 if (rdev2->raid_disk < 0 ||
1208                     test_bit(Faulty, &rdev2->flags))
1209                         is_active = 0;
1210                 if (is_active)
1211                         desc_nr = rdev2->raid_disk;
1212                 else
1213                         desc_nr = next_spare++;
1214                 rdev2->desc_nr = desc_nr;
1215                 d = &sb->disks[rdev2->desc_nr];
1216                 nr_disks++;
1217                 d->number = rdev2->desc_nr;
1218                 d->major = MAJOR(rdev2->bdev->bd_dev);
1219                 d->minor = MINOR(rdev2->bdev->bd_dev);
1220                 if (is_active)
1221                         d->raid_disk = rdev2->raid_disk;
1222                 else
1223                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1224                 if (test_bit(Faulty, &rdev2->flags))
1225                         d->state = (1<<MD_DISK_FAULTY);
1226                 else if (is_active) {
1227                         d->state = (1<<MD_DISK_ACTIVE);
1228                         if (test_bit(In_sync, &rdev2->flags))
1229                                 d->state |= (1<<MD_DISK_SYNC);
1230                         active++;
1231                         working++;
1232                 } else {
1233                         d->state = 0;
1234                         spare++;
1235                         working++;
1236                 }
1237                 if (test_bit(WriteMostly, &rdev2->flags))
1238                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1239         }
1240         /* now set the "removed" and "faulty" bits on any missing devices */
1241         for (i=0 ; i < mddev->raid_disks ; i++) {
1242                 mdp_disk_t *d = &sb->disks[i];
1243                 if (d->state == 0 && d->number == 0) {
1244                         d->number = i;
1245                         d->raid_disk = i;
1246                         d->state = (1<<MD_DISK_REMOVED);
1247                         d->state |= (1<<MD_DISK_FAULTY);
1248                         failed++;
1249                 }
1250         }
1251         sb->nr_disks = nr_disks;
1252         sb->active_disks = active;
1253         sb->working_disks = working;
1254         sb->failed_disks = failed;
1255         sb->spare_disks = spare;
1256
1257         sb->this_disk = sb->disks[rdev->desc_nr];
1258         sb->sb_csum = calc_sb_csum(sb);
1259 }
1260
1261 /*
1262  * rdev_size_change for 0.90.0
1263  */
1264 static unsigned long long
1265 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1266 {
1267         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1268                 return 0; /* component must fit device */
1269         if (rdev->mddev->bitmap_info.offset)
1270                 return 0; /* can't move bitmap */
1271         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1272         if (!num_sectors || num_sectors > rdev->sb_start)
1273                 num_sectors = rdev->sb_start;
1274         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1275                        rdev->sb_page);
1276         md_super_wait(rdev->mddev);
1277         return num_sectors / 2; /* kB for sysfs */
1278 }
1279
1280
1281 /*
1282  * version 1 superblock
1283  */
1284
1285 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1286 {
1287         __le32 disk_csum;
1288         u32 csum;
1289         unsigned long long newcsum;
1290         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1291         __le32 *isuper = (__le32*)sb;
1292         int i;
1293
1294         disk_csum = sb->sb_csum;
1295         sb->sb_csum = 0;
1296         newcsum = 0;
1297         for (i=0; size>=4; size -= 4 )
1298                 newcsum += le32_to_cpu(*isuper++);
1299
1300         if (size == 2)
1301                 newcsum += le16_to_cpu(*(__le16*) isuper);
1302
1303         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1304         sb->sb_csum = disk_csum;
1305         return cpu_to_le32(csum);
1306 }
1307
1308 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1309 {
1310         struct mdp_superblock_1 *sb;
1311         int ret;
1312         sector_t sb_start;
1313         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1314         int bmask;
1315
1316         /*
1317          * Calculate the position of the superblock in 512byte sectors.
1318          * It is always aligned to a 4K boundary and
1319          * depeding on minor_version, it can be:
1320          * 0: At least 8K, but less than 12K, from end of device
1321          * 1: At start of device
1322          * 2: 4K from start of device.
1323          */
1324         switch(minor_version) {
1325         case 0:
1326                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1327                 sb_start -= 8*2;
1328                 sb_start &= ~(sector_t)(4*2-1);
1329                 break;
1330         case 1:
1331                 sb_start = 0;
1332                 break;
1333         case 2:
1334                 sb_start = 8;
1335                 break;
1336         default:
1337                 return -EINVAL;
1338         }
1339         rdev->sb_start = sb_start;
1340
1341         /* superblock is rarely larger than 1K, but it can be larger,
1342          * and it is safe to read 4k, so we do that
1343          */
1344         ret = read_disk_sb(rdev, 4096);
1345         if (ret) return ret;
1346
1347
1348         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1349
1350         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1351             sb->major_version != cpu_to_le32(1) ||
1352             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1353             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1354             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1355                 return -EINVAL;
1356
1357         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1358                 printk("md: invalid superblock checksum on %s\n",
1359                         bdevname(rdev->bdev,b));
1360                 return -EINVAL;
1361         }
1362         if (le64_to_cpu(sb->data_size) < 10) {
1363                 printk("md: data_size too small on %s\n",
1364                        bdevname(rdev->bdev,b));
1365                 return -EINVAL;
1366         }
1367
1368         rdev->preferred_minor = 0xffff;
1369         rdev->data_offset = le64_to_cpu(sb->data_offset);
1370         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1371
1372         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1373         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1374         if (rdev->sb_size & bmask)
1375                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1376
1377         if (minor_version
1378             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1379                 return -EINVAL;
1380
1381         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1382                 rdev->desc_nr = -1;
1383         else
1384                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1385
1386         if (!refdev) {
1387                 ret = 1;
1388         } else {
1389                 __u64 ev1, ev2;
1390                 struct mdp_superblock_1 *refsb =
1391                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1392
1393                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1394                     sb->level != refsb->level ||
1395                     sb->layout != refsb->layout ||
1396                     sb->chunksize != refsb->chunksize) {
1397                         printk(KERN_WARNING "md: %s has strangely different"
1398                                 " superblock to %s\n",
1399                                 bdevname(rdev->bdev,b),
1400                                 bdevname(refdev->bdev,b2));
1401                         return -EINVAL;
1402                 }
1403                 ev1 = le64_to_cpu(sb->events);
1404                 ev2 = le64_to_cpu(refsb->events);
1405
1406                 if (ev1 > ev2)
1407                         ret = 1;
1408                 else
1409                         ret = 0;
1410         }
1411         if (minor_version)
1412                 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1413                         le64_to_cpu(sb->data_offset);
1414         else
1415                 rdev->sectors = rdev->sb_start;
1416         if (rdev->sectors < le64_to_cpu(sb->data_size))
1417                 return -EINVAL;
1418         rdev->sectors = le64_to_cpu(sb->data_size);
1419         if (le64_to_cpu(sb->size) > rdev->sectors)
1420                 return -EINVAL;
1421         return ret;
1422 }
1423
1424 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1425 {
1426         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1427         __u64 ev1 = le64_to_cpu(sb->events);
1428
1429         rdev->raid_disk = -1;
1430         clear_bit(Faulty, &rdev->flags);
1431         clear_bit(In_sync, &rdev->flags);
1432         clear_bit(WriteMostly, &rdev->flags);
1433         clear_bit(BarriersNotsupp, &rdev->flags);
1434
1435         if (mddev->raid_disks == 0) {
1436                 mddev->major_version = 1;
1437                 mddev->patch_version = 0;
1438                 mddev->external = 0;
1439                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1440                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1441                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1442                 mddev->level = le32_to_cpu(sb->level);
1443                 mddev->clevel[0] = 0;
1444                 mddev->layout = le32_to_cpu(sb->layout);
1445                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1446                 mddev->dev_sectors = le64_to_cpu(sb->size);
1447                 mddev->events = ev1;
1448                 mddev->bitmap_info.offset = 0;
1449                 mddev->bitmap_info.default_offset = 1024 >> 9;
1450
1451                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1452                 memcpy(mddev->uuid, sb->set_uuid, 16);
1453
1454                 mddev->max_disks =  (4096-256)/2;
1455
1456                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1457                     mddev->bitmap_info.file == NULL )
1458                         mddev->bitmap_info.offset =
1459                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1460
1461                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1462                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1463                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1464                         mddev->new_level = le32_to_cpu(sb->new_level);
1465                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1466                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1467                 } else {
1468                         mddev->reshape_position = MaxSector;
1469                         mddev->delta_disks = 0;
1470                         mddev->new_level = mddev->level;
1471                         mddev->new_layout = mddev->layout;
1472                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1473                 }
1474
1475         } else if (mddev->pers == NULL) {
1476                 /* Insist of good event counter while assembling, except for
1477                  * spares (which don't need an event count) */
1478                 ++ev1;
1479                 if (rdev->desc_nr >= 0 &&
1480                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1481                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1482                         if (ev1 < mddev->events)
1483                                 return -EINVAL;
1484         } else if (mddev->bitmap) {
1485                 /* If adding to array with a bitmap, then we can accept an
1486                  * older device, but not too old.
1487                  */
1488                 if (ev1 < mddev->bitmap->events_cleared)
1489                         return 0;
1490         } else {
1491                 if (ev1 < mddev->events)
1492                         /* just a hot-add of a new device, leave raid_disk at -1 */
1493                         return 0;
1494         }
1495         if (mddev->level != LEVEL_MULTIPATH) {
1496                 int role;
1497                 if (rdev->desc_nr < 0 ||
1498                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1499                         role = 0xffff;
1500                         rdev->desc_nr = -1;
1501                 } else
1502                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1503                 switch(role) {
1504                 case 0xffff: /* spare */
1505                         break;
1506                 case 0xfffe: /* faulty */
1507                         set_bit(Faulty, &rdev->flags);
1508                         break;
1509                 default:
1510                         if ((le32_to_cpu(sb->feature_map) &
1511                              MD_FEATURE_RECOVERY_OFFSET))
1512                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1513                         else
1514                                 set_bit(In_sync, &rdev->flags);
1515                         rdev->raid_disk = role;
1516                         break;
1517                 }
1518                 if (sb->devflags & WriteMostly1)
1519                         set_bit(WriteMostly, &rdev->flags);
1520         } else /* MULTIPATH are always insync */
1521                 set_bit(In_sync, &rdev->flags);
1522
1523         return 0;
1524 }
1525
1526 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1527 {
1528         struct mdp_superblock_1 *sb;
1529         mdk_rdev_t *rdev2;
1530         int max_dev, i;
1531         /* make rdev->sb match mddev and rdev data. */
1532
1533         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1534
1535         sb->feature_map = 0;
1536         sb->pad0 = 0;
1537         sb->recovery_offset = cpu_to_le64(0);
1538         memset(sb->pad1, 0, sizeof(sb->pad1));
1539         memset(sb->pad2, 0, sizeof(sb->pad2));
1540         memset(sb->pad3, 0, sizeof(sb->pad3));
1541
1542         sb->utime = cpu_to_le64((__u64)mddev->utime);
1543         sb->events = cpu_to_le64(mddev->events);
1544         if (mddev->in_sync)
1545                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1546         else
1547                 sb->resync_offset = cpu_to_le64(0);
1548
1549         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1550
1551         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1552         sb->size = cpu_to_le64(mddev->dev_sectors);
1553         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1554         sb->level = cpu_to_le32(mddev->level);
1555         sb->layout = cpu_to_le32(mddev->layout);
1556
1557         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1558                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1559                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1560         }
1561
1562         if (rdev->raid_disk >= 0 &&
1563             !test_bit(In_sync, &rdev->flags)) {
1564                 sb->feature_map |=
1565                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1566                 sb->recovery_offset =
1567                         cpu_to_le64(rdev->recovery_offset);
1568         }
1569
1570         if (mddev->reshape_position != MaxSector) {
1571                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1572                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1573                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1574                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1575                 sb->new_level = cpu_to_le32(mddev->new_level);
1576                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1577         }
1578
1579         max_dev = 0;
1580         list_for_each_entry(rdev2, &mddev->disks, same_set)
1581                 if (rdev2->desc_nr+1 > max_dev)
1582                         max_dev = rdev2->desc_nr+1;
1583
1584         if (max_dev > le32_to_cpu(sb->max_dev)) {
1585                 int bmask;
1586                 sb->max_dev = cpu_to_le32(max_dev);
1587                 rdev->sb_size = max_dev * 2 + 256;
1588                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1589                 if (rdev->sb_size & bmask)
1590                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1591         }
1592         for (i=0; i<max_dev;i++)
1593                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1594
1595         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1596                 i = rdev2->desc_nr;
1597                 if (test_bit(Faulty, &rdev2->flags))
1598                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1599                 else if (test_bit(In_sync, &rdev2->flags))
1600                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1601                 else if (rdev2->raid_disk >= 0)
1602                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1603                 else
1604                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1605         }
1606
1607         sb->sb_csum = calc_sb_1_csum(sb);
1608 }
1609
1610 static unsigned long long
1611 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1612 {
1613         struct mdp_superblock_1 *sb;
1614         sector_t max_sectors;
1615         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1616                 return 0; /* component must fit device */
1617         if (rdev->sb_start < rdev->data_offset) {
1618                 /* minor versions 1 and 2; superblock before data */
1619                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1620                 max_sectors -= rdev->data_offset;
1621                 if (!num_sectors || num_sectors > max_sectors)
1622                         num_sectors = max_sectors;
1623         } else if (rdev->mddev->bitmap_info.offset) {
1624                 /* minor version 0 with bitmap we can't move */
1625                 return 0;
1626         } else {
1627                 /* minor version 0; superblock after data */
1628                 sector_t sb_start;
1629                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1630                 sb_start &= ~(sector_t)(4*2 - 1);
1631                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1632                 if (!num_sectors || num_sectors > max_sectors)
1633                         num_sectors = max_sectors;
1634                 rdev->sb_start = sb_start;
1635         }
1636         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1637         sb->data_size = cpu_to_le64(num_sectors);
1638         sb->super_offset = rdev->sb_start;
1639         sb->sb_csum = calc_sb_1_csum(sb);
1640         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1641                        rdev->sb_page);
1642         md_super_wait(rdev->mddev);
1643         return num_sectors / 2; /* kB for sysfs */
1644 }
1645
1646 static struct super_type super_types[] = {
1647         [0] = {
1648                 .name   = "0.90.0",
1649                 .owner  = THIS_MODULE,
1650                 .load_super         = super_90_load,
1651                 .validate_super     = super_90_validate,
1652                 .sync_super         = super_90_sync,
1653                 .rdev_size_change   = super_90_rdev_size_change,
1654         },
1655         [1] = {
1656                 .name   = "md-1",
1657                 .owner  = THIS_MODULE,
1658                 .load_super         = super_1_load,
1659                 .validate_super     = super_1_validate,
1660                 .sync_super         = super_1_sync,
1661                 .rdev_size_change   = super_1_rdev_size_change,
1662         },
1663 };
1664
1665 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1666 {
1667         mdk_rdev_t *rdev, *rdev2;
1668
1669         rcu_read_lock();
1670         rdev_for_each_rcu(rdev, mddev1)
1671                 rdev_for_each_rcu(rdev2, mddev2)
1672                         if (rdev->bdev->bd_contains ==
1673                             rdev2->bdev->bd_contains) {
1674                                 rcu_read_unlock();
1675                                 return 1;
1676                         }
1677         rcu_read_unlock();
1678         return 0;
1679 }
1680
1681 static LIST_HEAD(pending_raid_disks);
1682
1683 /*
1684  * Try to register data integrity profile for an mddev
1685  *
1686  * This is called when an array is started and after a disk has been kicked
1687  * from the array. It only succeeds if all working and active component devices
1688  * are integrity capable with matching profiles.
1689  */
1690 int md_integrity_register(mddev_t *mddev)
1691 {
1692         mdk_rdev_t *rdev, *reference = NULL;
1693
1694         if (list_empty(&mddev->disks))
1695                 return 0; /* nothing to do */
1696         if (blk_get_integrity(mddev->gendisk))
1697                 return 0; /* already registered */
1698         list_for_each_entry(rdev, &mddev->disks, same_set) {
1699                 /* skip spares and non-functional disks */
1700                 if (test_bit(Faulty, &rdev->flags))
1701                         continue;
1702                 if (rdev->raid_disk < 0)
1703                         continue;
1704                 /*
1705                  * If at least one rdev is not integrity capable, we can not
1706                  * enable data integrity for the md device.
1707                  */
1708                 if (!bdev_get_integrity(rdev->bdev))
1709                         return -EINVAL;
1710                 if (!reference) {
1711                         /* Use the first rdev as the reference */
1712                         reference = rdev;
1713                         continue;
1714                 }
1715                 /* does this rdev's profile match the reference profile? */
1716                 if (blk_integrity_compare(reference->bdev->bd_disk,
1717                                 rdev->bdev->bd_disk) < 0)
1718                         return -EINVAL;
1719         }
1720         /*
1721          * All component devices are integrity capable and have matching
1722          * profiles, register the common profile for the md device.
1723          */
1724         if (blk_integrity_register(mddev->gendisk,
1725                         bdev_get_integrity(reference->bdev)) != 0) {
1726                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1727                         mdname(mddev));
1728                 return -EINVAL;
1729         }
1730         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1731                 mdname(mddev));
1732         return 0;
1733 }
1734 EXPORT_SYMBOL(md_integrity_register);
1735
1736 /* Disable data integrity if non-capable/non-matching disk is being added */
1737 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1738 {
1739         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1740         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1741
1742         if (!bi_mddev) /* nothing to do */
1743                 return;
1744         if (rdev->raid_disk < 0) /* skip spares */
1745                 return;
1746         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1747                                              rdev->bdev->bd_disk) >= 0)
1748                 return;
1749         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1750         blk_integrity_unregister(mddev->gendisk);
1751 }
1752 EXPORT_SYMBOL(md_integrity_add_rdev);
1753
1754 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1755 {
1756         char b[BDEVNAME_SIZE];
1757         struct kobject *ko;
1758         char *s;
1759         int err;
1760
1761         if (rdev->mddev) {
1762                 MD_BUG();
1763                 return -EINVAL;
1764         }
1765
1766         /* prevent duplicates */
1767         if (find_rdev(mddev, rdev->bdev->bd_dev))
1768                 return -EEXIST;
1769
1770         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1771         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1772                         rdev->sectors < mddev->dev_sectors)) {
1773                 if (mddev->pers) {
1774                         /* Cannot change size, so fail
1775                          * If mddev->level <= 0, then we don't care
1776                          * about aligning sizes (e.g. linear)
1777                          */
1778                         if (mddev->level > 0)
1779                                 return -ENOSPC;
1780                 } else
1781                         mddev->dev_sectors = rdev->sectors;
1782         }
1783
1784         /* Verify rdev->desc_nr is unique.
1785          * If it is -1, assign a free number, else
1786          * check number is not in use
1787          */
1788         if (rdev->desc_nr < 0) {
1789                 int choice = 0;
1790                 if (mddev->pers) choice = mddev->raid_disks;
1791                 while (find_rdev_nr(mddev, choice))
1792                         choice++;
1793                 rdev->desc_nr = choice;
1794         } else {
1795                 if (find_rdev_nr(mddev, rdev->desc_nr))
1796                         return -EBUSY;
1797         }
1798         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1799                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1800                        mdname(mddev), mddev->max_disks);
1801                 return -EBUSY;
1802         }
1803         bdevname(rdev->bdev,b);
1804         while ( (s=strchr(b, '/')) != NULL)
1805                 *s = '!';
1806
1807         rdev->mddev = mddev;
1808         printk(KERN_INFO "md: bind<%s>\n", b);
1809
1810         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1811                 goto fail;
1812
1813         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1814         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1815                 kobject_del(&rdev->kobj);
1816                 goto fail;
1817         }
1818         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, NULL, "state");
1819
1820         list_add_rcu(&rdev->same_set, &mddev->disks);
1821         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1822
1823         /* May as well allow recovery to be retried once */
1824         mddev->recovery_disabled = 0;
1825
1826         return 0;
1827
1828  fail:
1829         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1830                b, mdname(mddev));
1831         return err;
1832 }
1833
1834 static void md_delayed_delete(struct work_struct *ws)
1835 {
1836         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1837         kobject_del(&rdev->kobj);
1838         kobject_put(&rdev->kobj);
1839 }
1840
1841 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1842 {
1843         char b[BDEVNAME_SIZE];
1844         if (!rdev->mddev) {
1845                 MD_BUG();
1846                 return;
1847         }
1848         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1849         list_del_rcu(&rdev->same_set);
1850         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1851         rdev->mddev = NULL;
1852         sysfs_remove_link(&rdev->kobj, "block");
1853         sysfs_put(rdev->sysfs_state);
1854         rdev->sysfs_state = NULL;
1855         /* We need to delay this, otherwise we can deadlock when
1856          * writing to 'remove' to "dev/state".  We also need
1857          * to delay it due to rcu usage.
1858          */
1859         synchronize_rcu();
1860         INIT_WORK(&rdev->del_work, md_delayed_delete);
1861         kobject_get(&rdev->kobj);
1862         schedule_work(&rdev->del_work);
1863 }
1864
1865 /*
1866  * prevent the device from being mounted, repartitioned or
1867  * otherwise reused by a RAID array (or any other kernel
1868  * subsystem), by bd_claiming the device.
1869  */
1870 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1871 {
1872         int err = 0;
1873         struct block_device *bdev;
1874         char b[BDEVNAME_SIZE];
1875
1876         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1877         if (IS_ERR(bdev)) {
1878                 printk(KERN_ERR "md: could not open %s.\n",
1879                         __bdevname(dev, b));
1880                 return PTR_ERR(bdev);
1881         }
1882         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1883         if (err) {
1884                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1885                         bdevname(bdev, b));
1886                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1887                 return err;
1888         }
1889         if (!shared)
1890                 set_bit(AllReserved, &rdev->flags);
1891         rdev->bdev = bdev;
1892         return err;
1893 }
1894
1895 static void unlock_rdev(mdk_rdev_t *rdev)
1896 {
1897         struct block_device *bdev = rdev->bdev;
1898         rdev->bdev = NULL;
1899         if (!bdev)
1900                 MD_BUG();
1901         bd_release(bdev);
1902         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1903 }
1904
1905 void md_autodetect_dev(dev_t dev);
1906
1907 static void export_rdev(mdk_rdev_t * rdev)
1908 {
1909         char b[BDEVNAME_SIZE];
1910         printk(KERN_INFO "md: export_rdev(%s)\n",
1911                 bdevname(rdev->bdev,b));
1912         if (rdev->mddev)
1913                 MD_BUG();
1914         free_disk_sb(rdev);
1915 #ifndef MODULE
1916         if (test_bit(AutoDetected, &rdev->flags))
1917                 md_autodetect_dev(rdev->bdev->bd_dev);
1918 #endif
1919         unlock_rdev(rdev);
1920         kobject_put(&rdev->kobj);
1921 }
1922
1923 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1924 {
1925         unbind_rdev_from_array(rdev);
1926         export_rdev(rdev);
1927 }
1928
1929 static void export_array(mddev_t *mddev)
1930 {
1931         mdk_rdev_t *rdev, *tmp;
1932
1933         rdev_for_each(rdev, tmp, mddev) {
1934                 if (!rdev->mddev) {
1935                         MD_BUG();
1936                         continue;
1937                 }
1938                 kick_rdev_from_array(rdev);
1939         }
1940         if (!list_empty(&mddev->disks))
1941                 MD_BUG();
1942         mddev->raid_disks = 0;
1943         mddev->major_version = 0;
1944 }
1945
1946 static void print_desc(mdp_disk_t *desc)
1947 {
1948         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1949                 desc->major,desc->minor,desc->raid_disk,desc->state);
1950 }
1951
1952 static void print_sb_90(mdp_super_t *sb)
1953 {
1954         int i;
1955
1956         printk(KERN_INFO
1957                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1958                 sb->major_version, sb->minor_version, sb->patch_version,
1959                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1960                 sb->ctime);
1961         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1962                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1963                 sb->md_minor, sb->layout, sb->chunk_size);
1964         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1965                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1966                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1967                 sb->failed_disks, sb->spare_disks,
1968                 sb->sb_csum, (unsigned long)sb->events_lo);
1969
1970         printk(KERN_INFO);
1971         for (i = 0; i < MD_SB_DISKS; i++) {
1972                 mdp_disk_t *desc;
1973
1974                 desc = sb->disks + i;
1975                 if (desc->number || desc->major || desc->minor ||
1976                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1977                         printk("     D %2d: ", i);
1978                         print_desc(desc);
1979                 }
1980         }
1981         printk(KERN_INFO "md:     THIS: ");
1982         print_desc(&sb->this_disk);
1983 }
1984
1985 static void print_sb_1(struct mdp_superblock_1 *sb)
1986 {
1987         __u8 *uuid;
1988
1989         uuid = sb->set_uuid;
1990         printk(KERN_INFO
1991                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1992                "md:    Name: \"%s\" CT:%llu\n",
1993                 le32_to_cpu(sb->major_version),
1994                 le32_to_cpu(sb->feature_map),
1995                 uuid,
1996                 sb->set_name,
1997                 (unsigned long long)le64_to_cpu(sb->ctime)
1998                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1999
2000         uuid = sb->device_uuid;
2001         printk(KERN_INFO
2002                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2003                         " RO:%llu\n"
2004                "md:     Dev:%08x UUID: %pU\n"
2005                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2006                "md:         (MaxDev:%u) \n",
2007                 le32_to_cpu(sb->level),
2008                 (unsigned long long)le64_to_cpu(sb->size),
2009                 le32_to_cpu(sb->raid_disks),
2010                 le32_to_cpu(sb->layout),
2011                 le32_to_cpu(sb->chunksize),
2012                 (unsigned long long)le64_to_cpu(sb->data_offset),
2013                 (unsigned long long)le64_to_cpu(sb->data_size),
2014                 (unsigned long long)le64_to_cpu(sb->super_offset),
2015                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2016                 le32_to_cpu(sb->dev_number),
2017                 uuid,
2018                 sb->devflags,
2019                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2020                 (unsigned long long)le64_to_cpu(sb->events),
2021                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2022                 le32_to_cpu(sb->sb_csum),
2023                 le32_to_cpu(sb->max_dev)
2024                 );
2025 }
2026
2027 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2028 {
2029         char b[BDEVNAME_SIZE];
2030         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2031                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2032                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2033                 rdev->desc_nr);
2034         if (rdev->sb_loaded) {
2035                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2036                 switch (major_version) {
2037                 case 0:
2038                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2039                         break;
2040                 case 1:
2041                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2042                         break;
2043                 }
2044         } else
2045                 printk(KERN_INFO "md: no rdev superblock!\n");
2046 }
2047
2048 static void md_print_devices(void)
2049 {
2050         struct list_head *tmp;
2051         mdk_rdev_t *rdev;
2052         mddev_t *mddev;
2053         char b[BDEVNAME_SIZE];
2054
2055         printk("\n");
2056         printk("md:     **********************************\n");
2057         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2058         printk("md:     **********************************\n");
2059         for_each_mddev(mddev, tmp) {
2060
2061                 if (mddev->bitmap)
2062                         bitmap_print_sb(mddev->bitmap);
2063                 else
2064                         printk("%s: ", mdname(mddev));
2065                 list_for_each_entry(rdev, &mddev->disks, same_set)
2066                         printk("<%s>", bdevname(rdev->bdev,b));
2067                 printk("\n");
2068
2069                 list_for_each_entry(rdev, &mddev->disks, same_set)
2070                         print_rdev(rdev, mddev->major_version);
2071         }
2072         printk("md:     **********************************\n");
2073         printk("\n");
2074 }
2075
2076
2077 static void sync_sbs(mddev_t * mddev, int nospares)
2078 {
2079         /* Update each superblock (in-memory image), but
2080          * if we are allowed to, skip spares which already
2081          * have the right event counter, or have one earlier
2082          * (which would mean they aren't being marked as dirty
2083          * with the rest of the array)
2084          */
2085         mdk_rdev_t *rdev;
2086
2087         /* First make sure individual recovery_offsets are correct */
2088         list_for_each_entry(rdev, &mddev->disks, same_set) {
2089                 if (rdev->raid_disk >= 0 &&
2090                     !test_bit(In_sync, &rdev->flags) &&
2091                     mddev->curr_resync_completed > rdev->recovery_offset)
2092                                 rdev->recovery_offset = mddev->curr_resync_completed;
2093
2094         }
2095         list_for_each_entry(rdev, &mddev->disks, same_set) {
2096                 if (rdev->sb_events == mddev->events ||
2097                     (nospares &&
2098                      rdev->raid_disk < 0 &&
2099                      rdev->sb_events+1 == mddev->events)) {
2100                         /* Don't update this superblock */
2101                         rdev->sb_loaded = 2;
2102                 } else {
2103                         super_types[mddev->major_version].
2104                                 sync_super(mddev, rdev);
2105                         rdev->sb_loaded = 1;
2106                 }
2107         }
2108 }
2109
2110 static void md_update_sb(mddev_t * mddev, int force_change)
2111 {
2112         mdk_rdev_t *rdev;
2113         int sync_req;
2114         int nospares = 0;
2115
2116         mddev->utime = get_seconds();
2117         if (mddev->external)
2118                 return;
2119 repeat:
2120         spin_lock_irq(&mddev->write_lock);
2121
2122         set_bit(MD_CHANGE_PENDING, &mddev->flags);
2123         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2124                 force_change = 1;
2125         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2126                 /* just a clean<-> dirty transition, possibly leave spares alone,
2127                  * though if events isn't the right even/odd, we will have to do
2128                  * spares after all
2129                  */
2130                 nospares = 1;
2131         if (force_change)
2132                 nospares = 0;
2133         if (mddev->degraded)
2134                 /* If the array is degraded, then skipping spares is both
2135                  * dangerous and fairly pointless.
2136                  * Dangerous because a device that was removed from the array
2137                  * might have a event_count that still looks up-to-date,
2138                  * so it can be re-added without a resync.
2139                  * Pointless because if there are any spares to skip,
2140                  * then a recovery will happen and soon that array won't
2141                  * be degraded any more and the spare can go back to sleep then.
2142                  */
2143                 nospares = 0;
2144
2145         sync_req = mddev->in_sync;
2146
2147         /* If this is just a dirty<->clean transition, and the array is clean
2148          * and 'events' is odd, we can roll back to the previous clean state */
2149         if (nospares
2150             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2151             && mddev->can_decrease_events
2152             && mddev->events != 1) {
2153                 mddev->events--;
2154                 mddev->can_decrease_events = 0;
2155         } else {
2156                 /* otherwise we have to go forward and ... */
2157                 mddev->events ++;
2158                 mddev->can_decrease_events = nospares;
2159         }
2160
2161         if (!mddev->events) {
2162                 /*
2163                  * oops, this 64-bit counter should never wrap.
2164                  * Either we are in around ~1 trillion A.C., assuming
2165                  * 1 reboot per second, or we have a bug:
2166                  */
2167                 MD_BUG();
2168                 mddev->events --;
2169         }
2170
2171         /*
2172          * do not write anything to disk if using
2173          * nonpersistent superblocks
2174          */
2175         if (!mddev->persistent) {
2176                 if (!mddev->external)
2177                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2178
2179                 spin_unlock_irq(&mddev->write_lock);
2180                 wake_up(&mddev->sb_wait);
2181                 return;
2182         }
2183         sync_sbs(mddev, nospares);
2184         spin_unlock_irq(&mddev->write_lock);
2185
2186         dprintk(KERN_INFO
2187                 "md: updating %s RAID superblock on device (in sync %d)\n",
2188                 mdname(mddev),mddev->in_sync);
2189
2190         bitmap_update_sb(mddev->bitmap);
2191         list_for_each_entry(rdev, &mddev->disks, same_set) {
2192                 char b[BDEVNAME_SIZE];
2193                 dprintk(KERN_INFO "md: ");
2194                 if (rdev->sb_loaded != 1)
2195                         continue; /* no noise on spare devices */
2196                 if (test_bit(Faulty, &rdev->flags))
2197                         dprintk("(skipping faulty ");
2198
2199                 dprintk("%s ", bdevname(rdev->bdev,b));
2200                 if (!test_bit(Faulty, &rdev->flags)) {
2201                         md_super_write(mddev,rdev,
2202                                        rdev->sb_start, rdev->sb_size,
2203                                        rdev->sb_page);
2204                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2205                                 bdevname(rdev->bdev,b),
2206                                 (unsigned long long)rdev->sb_start);
2207                         rdev->sb_events = mddev->events;
2208
2209                 } else
2210                         dprintk(")\n");
2211                 if (mddev->level == LEVEL_MULTIPATH)
2212                         /* only need to write one superblock... */
2213                         break;
2214         }
2215         md_super_wait(mddev);
2216         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2217
2218         spin_lock_irq(&mddev->write_lock);
2219         if (mddev->in_sync != sync_req ||
2220             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2221                 /* have to write it out again */
2222                 spin_unlock_irq(&mddev->write_lock);
2223                 goto repeat;
2224         }
2225         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2226         spin_unlock_irq(&mddev->write_lock);
2227         wake_up(&mddev->sb_wait);
2228         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2229                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2230
2231 }
2232
2233 /* words written to sysfs files may, or may not, be \n terminated.
2234  * We want to accept with case. For this we use cmd_match.
2235  */
2236 static int cmd_match(const char *cmd, const char *str)
2237 {
2238         /* See if cmd, written into a sysfs file, matches
2239          * str.  They must either be the same, or cmd can
2240          * have a trailing newline
2241          */
2242         while (*cmd && *str && *cmd == *str) {
2243                 cmd++;
2244                 str++;
2245         }
2246         if (*cmd == '\n')
2247                 cmd++;
2248         if (*str || *cmd)
2249                 return 0;
2250         return 1;
2251 }
2252
2253 struct rdev_sysfs_entry {
2254         struct attribute attr;
2255         ssize_t (*show)(mdk_rdev_t *, char *);
2256         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2257 };
2258
2259 static ssize_t
2260 state_show(mdk_rdev_t *rdev, char *page)
2261 {
2262         char *sep = "";
2263         size_t len = 0;
2264
2265         if (test_bit(Faulty, &rdev->flags)) {
2266                 len+= sprintf(page+len, "%sfaulty",sep);
2267                 sep = ",";
2268         }
2269         if (test_bit(In_sync, &rdev->flags)) {
2270                 len += sprintf(page+len, "%sin_sync",sep);
2271                 sep = ",";
2272         }
2273         if (test_bit(WriteMostly, &rdev->flags)) {
2274                 len += sprintf(page+len, "%swrite_mostly",sep);
2275                 sep = ",";
2276         }
2277         if (test_bit(Blocked, &rdev->flags)) {
2278                 len += sprintf(page+len, "%sblocked", sep);
2279                 sep = ",";
2280         }
2281         if (!test_bit(Faulty, &rdev->flags) &&
2282             !test_bit(In_sync, &rdev->flags)) {
2283                 len += sprintf(page+len, "%sspare", sep);
2284                 sep = ",";
2285         }
2286         return len+sprintf(page+len, "\n");
2287 }
2288
2289 static ssize_t
2290 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2291 {
2292         /* can write
2293          *  faulty  - simulates and error
2294          *  remove  - disconnects the device
2295          *  writemostly - sets write_mostly
2296          *  -writemostly - clears write_mostly
2297          *  blocked - sets the Blocked flag
2298          *  -blocked - clears the Blocked flag
2299          *  insync - sets Insync providing device isn't active
2300          */
2301         int err = -EINVAL;
2302         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2303                 md_error(rdev->mddev, rdev);
2304                 err = 0;
2305         } else if (cmd_match(buf, "remove")) {
2306                 if (rdev->raid_disk >= 0)
2307                         err = -EBUSY;
2308                 else {
2309                         mddev_t *mddev = rdev->mddev;
2310                         kick_rdev_from_array(rdev);
2311                         if (mddev->pers)
2312                                 md_update_sb(mddev, 1);
2313                         md_new_event(mddev);
2314                         err = 0;
2315                 }
2316         } else if (cmd_match(buf, "writemostly")) {
2317                 set_bit(WriteMostly, &rdev->flags);
2318                 err = 0;
2319         } else if (cmd_match(buf, "-writemostly")) {
2320                 clear_bit(WriteMostly, &rdev->flags);
2321                 err = 0;
2322         } else if (cmd_match(buf, "blocked")) {
2323                 set_bit(Blocked, &rdev->flags);
2324                 err = 0;
2325         } else if (cmd_match(buf, "-blocked")) {
2326                 clear_bit(Blocked, &rdev->flags);
2327                 wake_up(&rdev->blocked_wait);
2328                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2329                 md_wakeup_thread(rdev->mddev->thread);
2330
2331                 err = 0;
2332         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2333                 set_bit(In_sync, &rdev->flags);
2334                 err = 0;
2335         }
2336         if (!err && rdev->sysfs_state)
2337                 sysfs_notify_dirent(rdev->sysfs_state);
2338         return err ? err : len;
2339 }
2340 static struct rdev_sysfs_entry rdev_state =
2341 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2342
2343 static ssize_t
2344 errors_show(mdk_rdev_t *rdev, char *page)
2345 {
2346         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2347 }
2348
2349 static ssize_t
2350 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2351 {
2352         char *e;
2353         unsigned long n = simple_strtoul(buf, &e, 10);
2354         if (*buf && (*e == 0 || *e == '\n')) {
2355                 atomic_set(&rdev->corrected_errors, n);
2356                 return len;
2357         }
2358         return -EINVAL;
2359 }
2360 static struct rdev_sysfs_entry rdev_errors =
2361 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2362
2363 static ssize_t
2364 slot_show(mdk_rdev_t *rdev, char *page)
2365 {
2366         if (rdev->raid_disk < 0)
2367                 return sprintf(page, "none\n");
2368         else
2369                 return sprintf(page, "%d\n", rdev->raid_disk);
2370 }
2371
2372 static ssize_t
2373 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2374 {
2375         char *e;
2376         int err;
2377         char nm[20];
2378         int slot = simple_strtoul(buf, &e, 10);
2379         if (strncmp(buf, "none", 4)==0)
2380                 slot = -1;
2381         else if (e==buf || (*e && *e!= '\n'))
2382                 return -EINVAL;
2383         if (rdev->mddev->pers && slot == -1) {
2384                 /* Setting 'slot' on an active array requires also
2385                  * updating the 'rd%d' link, and communicating
2386                  * with the personality with ->hot_*_disk.
2387                  * For now we only support removing
2388                  * failed/spare devices.  This normally happens automatically,
2389                  * but not when the metadata is externally managed.
2390                  */
2391                 if (rdev->raid_disk == -1)
2392                         return -EEXIST;
2393                 /* personality does all needed checks */
2394                 if (rdev->mddev->pers->hot_add_disk == NULL)
2395                         return -EINVAL;
2396                 err = rdev->mddev->pers->
2397                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2398                 if (err)
2399                         return err;
2400                 sprintf(nm, "rd%d", rdev->raid_disk);
2401                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2402                 rdev->raid_disk = -1;
2403                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2404                 md_wakeup_thread(rdev->mddev->thread);
2405         } else if (rdev->mddev->pers) {
2406                 mdk_rdev_t *rdev2;
2407                 /* Activating a spare .. or possibly reactivating
2408                  * if we ever get bitmaps working here.
2409                  */
2410
2411                 if (rdev->raid_disk != -1)
2412                         return -EBUSY;
2413
2414                 if (rdev->mddev->pers->hot_add_disk == NULL)
2415                         return -EINVAL;
2416
2417                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2418                         if (rdev2->raid_disk == slot)
2419                                 return -EEXIST;
2420
2421                 rdev->raid_disk = slot;
2422                 if (test_bit(In_sync, &rdev->flags))
2423                         rdev->saved_raid_disk = slot;
2424                 else
2425                         rdev->saved_raid_disk = -1;
2426                 err = rdev->mddev->pers->
2427                         hot_add_disk(rdev->mddev, rdev);
2428                 if (err) {
2429                         rdev->raid_disk = -1;
2430                         return err;
2431                 } else
2432                         sysfs_notify_dirent(rdev->sysfs_state);
2433                 sprintf(nm, "rd%d", rdev->raid_disk);
2434                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2435                         printk(KERN_WARNING
2436                                "md: cannot register "
2437                                "%s for %s\n",
2438                                nm, mdname(rdev->mddev));
2439
2440                 /* don't wakeup anyone, leave that to userspace. */
2441         } else {
2442                 if (slot >= rdev->mddev->raid_disks)
2443                         return -ENOSPC;
2444                 rdev->raid_disk = slot;
2445                 /* assume it is working */
2446                 clear_bit(Faulty, &rdev->flags);
2447                 clear_bit(WriteMostly, &rdev->flags);
2448                 set_bit(In_sync, &rdev->flags);
2449                 sysfs_notify_dirent(rdev->sysfs_state);
2450         }
2451         return len;
2452 }
2453
2454
2455 static struct rdev_sysfs_entry rdev_slot =
2456 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2457
2458 static ssize_t
2459 offset_show(mdk_rdev_t *rdev, char *page)
2460 {
2461         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2462 }
2463
2464 static ssize_t
2465 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2466 {
2467         char *e;
2468         unsigned long long offset = simple_strtoull(buf, &e, 10);
2469         if (e==buf || (*e && *e != '\n'))
2470                 return -EINVAL;
2471         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2472                 return -EBUSY;
2473         if (rdev->sectors && rdev->mddev->external)
2474                 /* Must set offset before size, so overlap checks
2475                  * can be sane */
2476                 return -EBUSY;
2477         rdev->data_offset = offset;
2478         return len;
2479 }
2480
2481 static struct rdev_sysfs_entry rdev_offset =
2482 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2483
2484 static ssize_t
2485 rdev_size_show(mdk_rdev_t *rdev, char *page)
2486 {
2487         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2488 }
2489
2490 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2491 {
2492         /* check if two start/length pairs overlap */
2493         if (s1+l1 <= s2)
2494                 return 0;
2495         if (s2+l2 <= s1)
2496                 return 0;
2497         return 1;
2498 }
2499
2500 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2501 {
2502         unsigned long long blocks;
2503         sector_t new;
2504
2505         if (strict_strtoull(buf, 10, &blocks) < 0)
2506                 return -EINVAL;
2507
2508         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2509                 return -EINVAL; /* sector conversion overflow */
2510
2511         new = blocks * 2;
2512         if (new != blocks * 2)
2513                 return -EINVAL; /* unsigned long long to sector_t overflow */
2514
2515         *sectors = new;
2516         return 0;
2517 }
2518
2519 static ssize_t
2520 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2521 {
2522         mddev_t *my_mddev = rdev->mddev;
2523         sector_t oldsectors = rdev->sectors;
2524         sector_t sectors;
2525
2526         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2527                 return -EINVAL;
2528         if (my_mddev->pers && rdev->raid_disk >= 0) {
2529                 if (my_mddev->persistent) {
2530                         sectors = super_types[my_mddev->major_version].
2531                                 rdev_size_change(rdev, sectors);
2532                         if (!sectors)
2533                                 return -EBUSY;
2534                 } else if (!sectors)
2535                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2536                                 rdev->data_offset;
2537         }
2538         if (sectors < my_mddev->dev_sectors)
2539                 return -EINVAL; /* component must fit device */
2540
2541         rdev->sectors = sectors;
2542         if (sectors > oldsectors && my_mddev->external) {
2543                 /* need to check that all other rdevs with the same ->bdev
2544                  * do not overlap.  We need to unlock the mddev to avoid
2545                  * a deadlock.  We have already changed rdev->sectors, and if
2546                  * we have to change it back, we will have the lock again.
2547                  */
2548                 mddev_t *mddev;
2549                 int overlap = 0;
2550                 struct list_head *tmp;
2551
2552                 mddev_unlock(my_mddev);
2553                 for_each_mddev(mddev, tmp) {
2554                         mdk_rdev_t *rdev2;
2555
2556                         mddev_lock(mddev);
2557                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2558                                 if (test_bit(AllReserved, &rdev2->flags) ||
2559                                     (rdev->bdev == rdev2->bdev &&
2560                                      rdev != rdev2 &&
2561                                      overlaps(rdev->data_offset, rdev->sectors,
2562                                               rdev2->data_offset,
2563                                               rdev2->sectors))) {
2564                                         overlap = 1;
2565                                         break;
2566                                 }
2567                         mddev_unlock(mddev);
2568                         if (overlap) {
2569                                 mddev_put(mddev);
2570                                 break;
2571                         }
2572                 }
2573                 mddev_lock(my_mddev);
2574                 if (overlap) {
2575                         /* Someone else could have slipped in a size
2576                          * change here, but doing so is just silly.
2577                          * We put oldsectors back because we *know* it is
2578                          * safe, and trust userspace not to race with
2579                          * itself
2580                          */
2581                         rdev->sectors = oldsectors;
2582                         return -EBUSY;
2583                 }
2584         }
2585         return len;
2586 }
2587
2588 static struct rdev_sysfs_entry rdev_size =
2589 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2590
2591
2592 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2593 {
2594         unsigned long long recovery_start = rdev->recovery_offset;
2595
2596         if (test_bit(In_sync, &rdev->flags) ||
2597             recovery_start == MaxSector)
2598                 return sprintf(page, "none\n");
2599
2600         return sprintf(page, "%llu\n", recovery_start);
2601 }
2602
2603 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2604 {
2605         unsigned long long recovery_start;
2606
2607         if (cmd_match(buf, "none"))
2608                 recovery_start = MaxSector;
2609         else if (strict_strtoull(buf, 10, &recovery_start))
2610                 return -EINVAL;
2611
2612         if (rdev->mddev->pers &&
2613             rdev->raid_disk >= 0)
2614                 return -EBUSY;
2615
2616         rdev->recovery_offset = recovery_start;
2617         if (recovery_start == MaxSector)
2618                 set_bit(In_sync, &rdev->flags);
2619         else
2620                 clear_bit(In_sync, &rdev->flags);
2621         return len;
2622 }
2623
2624 static struct rdev_sysfs_entry rdev_recovery_start =
2625 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2626
2627 static struct attribute *rdev_default_attrs[] = {
2628         &rdev_state.attr,
2629         &rdev_errors.attr,
2630         &rdev_slot.attr,
2631         &rdev_offset.attr,
2632         &rdev_size.attr,
2633         &rdev_recovery_start.attr,
2634         NULL,
2635 };
2636 static ssize_t
2637 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2638 {
2639         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2640         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2641         mddev_t *mddev = rdev->mddev;
2642         ssize_t rv;
2643
2644         if (!entry->show)
2645                 return -EIO;
2646
2647         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2648         if (!rv) {
2649                 if (rdev->mddev == NULL)
2650                         rv = -EBUSY;
2651                 else
2652                         rv = entry->show(rdev, page);
2653                 mddev_unlock(mddev);
2654         }
2655         return rv;
2656 }
2657
2658 static ssize_t
2659 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2660               const char *page, size_t length)
2661 {
2662         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2663         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2664         ssize_t rv;
2665         mddev_t *mddev = rdev->mddev;
2666
2667         if (!entry->store)
2668                 return -EIO;
2669         if (!capable(CAP_SYS_ADMIN))
2670                 return -EACCES;
2671         rv = mddev ? mddev_lock(mddev): -EBUSY;
2672         if (!rv) {
2673                 if (rdev->mddev == NULL)
2674                         rv = -EBUSY;
2675                 else
2676                         rv = entry->store(rdev, page, length);
2677                 mddev_unlock(mddev);
2678         }
2679         return rv;
2680 }
2681
2682 static void rdev_free(struct kobject *ko)
2683 {
2684         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2685         kfree(rdev);
2686 }
2687 static const struct sysfs_ops rdev_sysfs_ops = {
2688         .show           = rdev_attr_show,
2689         .store          = rdev_attr_store,
2690 };
2691 static struct kobj_type rdev_ktype = {
2692         .release        = rdev_free,
2693         .sysfs_ops      = &rdev_sysfs_ops,
2694         .default_attrs  = rdev_default_attrs,
2695 };
2696
2697 /*
2698  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2699  *
2700  * mark the device faulty if:
2701  *
2702  *   - the device is nonexistent (zero size)
2703  *   - the device has no valid superblock
2704  *
2705  * a faulty rdev _never_ has rdev->sb set.
2706  */
2707 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2708 {
2709         char b[BDEVNAME_SIZE];
2710         int err;
2711         mdk_rdev_t *rdev;
2712         sector_t size;
2713
2714         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2715         if (!rdev) {
2716                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2717                 return ERR_PTR(-ENOMEM);
2718         }
2719
2720         if ((err = alloc_disk_sb(rdev)))
2721                 goto abort_free;
2722
2723         err = lock_rdev(rdev, newdev, super_format == -2);
2724         if (err)
2725                 goto abort_free;
2726
2727         kobject_init(&rdev->kobj, &rdev_ktype);
2728
2729         rdev->desc_nr = -1;
2730         rdev->saved_raid_disk = -1;
2731         rdev->raid_disk = -1;
2732         rdev->flags = 0;
2733         rdev->data_offset = 0;
2734         rdev->sb_events = 0;
2735         rdev->last_read_error.tv_sec  = 0;
2736         rdev->last_read_error.tv_nsec = 0;
2737         atomic_set(&rdev->nr_pending, 0);
2738         atomic_set(&rdev->read_errors, 0);
2739         atomic_set(&rdev->corrected_errors, 0);
2740
2741         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2742         if (!size) {
2743                 printk(KERN_WARNING
2744                         "md: %s has zero or unknown size, marking faulty!\n",
2745                         bdevname(rdev->bdev,b));
2746                 err = -EINVAL;
2747                 goto abort_free;
2748         }
2749
2750         if (super_format >= 0) {
2751                 err = super_types[super_format].
2752                         load_super(rdev, NULL, super_minor);
2753                 if (err == -EINVAL) {
2754                         printk(KERN_WARNING
2755                                 "md: %s does not have a valid v%d.%d "
2756                                "superblock, not importing!\n",
2757                                 bdevname(rdev->bdev,b),
2758                                super_format, super_minor);
2759                         goto abort_free;
2760                 }
2761                 if (err < 0) {
2762                         printk(KERN_WARNING
2763                                 "md: could not read %s's sb, not importing!\n",
2764                                 bdevname(rdev->bdev,b));
2765                         goto abort_free;
2766                 }
2767         }
2768
2769         INIT_LIST_HEAD(&rdev->same_set);
2770         init_waitqueue_head(&rdev->blocked_wait);
2771
2772         return rdev;
2773
2774 abort_free:
2775         if (rdev->sb_page) {
2776                 if (rdev->bdev)
2777                         unlock_rdev(rdev);
2778                 free_disk_sb(rdev);
2779         }
2780         kfree(rdev);
2781         return ERR_PTR(err);
2782 }
2783
2784 /*
2785  * Check a full RAID array for plausibility
2786  */
2787
2788
2789 static void analyze_sbs(mddev_t * mddev)
2790 {
2791         int i;
2792         mdk_rdev_t *rdev, *freshest, *tmp;
2793         char b[BDEVNAME_SIZE];
2794
2795         freshest = NULL;
2796         rdev_for_each(rdev, tmp, mddev)
2797                 switch (super_types[mddev->major_version].
2798                         load_super(rdev, freshest, mddev->minor_version)) {
2799                 case 1:
2800                         freshest = rdev;
2801                         break;
2802                 case 0:
2803                         break;
2804                 default:
2805                         printk( KERN_ERR \
2806                                 "md: fatal superblock inconsistency in %s"
2807                                 " -- removing from array\n",
2808                                 bdevname(rdev->bdev,b));
2809                         kick_rdev_from_array(rdev);
2810                 }
2811
2812
2813         super_types[mddev->major_version].
2814                 validate_super(mddev, freshest);
2815
2816         i = 0;
2817         rdev_for_each(rdev, tmp, mddev) {
2818                 if (mddev->max_disks &&
2819                     (rdev->desc_nr >= mddev->max_disks ||
2820                      i > mddev->max_disks)) {
2821                         printk(KERN_WARNING
2822                                "md: %s: %s: only %d devices permitted\n",
2823                                mdname(mddev), bdevname(rdev->bdev, b),
2824                                mddev->max_disks);
2825                         kick_rdev_from_array(rdev);
2826                         continue;
2827                 }
2828                 if (rdev != freshest)
2829                         if (super_types[mddev->major_version].
2830                             validate_super(mddev, rdev)) {
2831                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2832                                         " from array!\n",
2833                                         bdevname(rdev->bdev,b));
2834                                 kick_rdev_from_array(rdev);
2835                                 continue;
2836                         }
2837                 if (mddev->level == LEVEL_MULTIPATH) {
2838                         rdev->desc_nr = i++;
2839                         rdev->raid_disk = rdev->desc_nr;
2840                         set_bit(In_sync, &rdev->flags);
2841                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2842                         rdev->raid_disk = -1;
2843                         clear_bit(In_sync, &rdev->flags);
2844                 }
2845         }
2846 }
2847
2848 /* Read a fixed-point number.
2849  * Numbers in sysfs attributes should be in "standard" units where
2850  * possible, so time should be in seconds.
2851  * However we internally use a a much smaller unit such as
2852  * milliseconds or jiffies.
2853  * This function takes a decimal number with a possible fractional
2854  * component, and produces an integer which is the result of
2855  * multiplying that number by 10^'scale'.
2856  * all without any floating-point arithmetic.
2857  */
2858 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2859 {
2860         unsigned long result = 0;
2861         long decimals = -1;
2862         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2863                 if (*cp == '.')
2864                         decimals = 0;
2865                 else if (decimals < scale) {
2866                         unsigned int value;
2867                         value = *cp - '0';
2868                         result = result * 10 + value;
2869                         if (decimals >= 0)
2870                                 decimals++;
2871                 }
2872                 cp++;
2873         }
2874         if (*cp == '\n')
2875                 cp++;
2876         if (*cp)
2877                 return -EINVAL;
2878         if (decimals < 0)
2879                 decimals = 0;
2880         while (decimals < scale) {
2881                 result *= 10;
2882                 decimals ++;
2883         }
2884         *res = result;
2885         return 0;
2886 }
2887
2888
2889 static void md_safemode_timeout(unsigned long data);
2890
2891 static ssize_t
2892 safe_delay_show(mddev_t *mddev, char *page)
2893 {
2894         int msec = (mddev->safemode_delay*1000)/HZ;
2895         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2896 }
2897 static ssize_t
2898 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2899 {
2900         unsigned long msec;
2901
2902         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2903                 return -EINVAL;
2904         if (msec == 0)
2905                 mddev->safemode_delay = 0;
2906         else {
2907                 unsigned long old_delay = mddev->safemode_delay;
2908                 mddev->safemode_delay = (msec*HZ)/1000;
2909                 if (mddev->safemode_delay == 0)
2910                         mddev->safemode_delay = 1;
2911                 if (mddev->safemode_delay < old_delay)
2912                         md_safemode_timeout((unsigned long)mddev);
2913         }
2914         return len;
2915 }
2916 static struct md_sysfs_entry md_safe_delay =
2917 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2918
2919 static ssize_t
2920 level_show(mddev_t *mddev, char *page)
2921 {
2922         struct mdk_personality *p = mddev->pers;
2923         if (p)
2924                 return sprintf(page, "%s\n", p->name);
2925         else if (mddev->clevel[0])
2926                 return sprintf(page, "%s\n", mddev->clevel);
2927         else if (mddev->level != LEVEL_NONE)
2928                 return sprintf(page, "%d\n", mddev->level);
2929         else
2930                 return 0;
2931 }
2932
2933 static ssize_t
2934 level_store(mddev_t *mddev, const char *buf, size_t len)
2935 {
2936         char clevel[16];
2937         ssize_t rv = len;
2938         struct mdk_personality *pers;
2939         long level;
2940         void *priv;
2941         mdk_rdev_t *rdev;
2942
2943         if (mddev->pers == NULL) {
2944                 if (len == 0)
2945                         return 0;
2946                 if (len >= sizeof(mddev->clevel))
2947                         return -ENOSPC;
2948                 strncpy(mddev->clevel, buf, len);
2949                 if (mddev->clevel[len-1] == '\n')
2950                         len--;
2951                 mddev->clevel[len] = 0;
2952                 mddev->level = LEVEL_NONE;
2953                 return rv;
2954         }
2955
2956         /* request to change the personality.  Need to ensure:
2957          *  - array is not engaged in resync/recovery/reshape
2958          *  - old personality can be suspended
2959          *  - new personality will access other array.
2960          */
2961
2962         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2963                 return -EBUSY;
2964
2965         if (!mddev->pers->quiesce) {
2966                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2967                        mdname(mddev), mddev->pers->name);
2968                 return -EINVAL;
2969         }
2970
2971         /* Now find the new personality */
2972         if (len == 0 || len >= sizeof(clevel))
2973                 return -EINVAL;
2974         strncpy(clevel, buf, len);
2975         if (clevel[len-1] == '\n')
2976                 len--;
2977         clevel[len] = 0;
2978         if (strict_strtol(clevel, 10, &level))
2979                 level = LEVEL_NONE;
2980
2981         if (request_module("md-%s", clevel) != 0)
2982                 request_module("md-level-%s", clevel);
2983         spin_lock(&pers_lock);
2984         pers = find_pers(level, clevel);
2985         if (!pers || !try_module_get(pers->owner)) {
2986                 spin_unlock(&pers_lock);
2987                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2988                 return -EINVAL;
2989         }
2990         spin_unlock(&pers_lock);
2991
2992         if (pers == mddev->pers) {
2993                 /* Nothing to do! */
2994                 module_put(pers->owner);
2995                 return rv;
2996         }
2997         if (!pers->takeover) {
2998                 module_put(pers->owner);
2999                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3000                        mdname(mddev), clevel);
3001                 return -EINVAL;
3002         }
3003
3004         /* ->takeover must set new_* and/or delta_disks
3005          * if it succeeds, and may set them when it fails.
3006          */
3007         priv = pers->takeover(mddev);
3008         if (IS_ERR(priv)) {
3009                 mddev->new_level = mddev->level;
3010                 mddev->new_layout = mddev->layout;
3011                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3012                 mddev->raid_disks -= mddev->delta_disks;
3013                 mddev->delta_disks = 0;
3014                 module_put(pers->owner);
3015                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3016                        mdname(mddev), clevel);
3017                 return PTR_ERR(priv);
3018         }
3019
3020         /* Looks like we have a winner */
3021         mddev_suspend(mddev);
3022         mddev->pers->stop(mddev);
3023
3024         if (mddev->pers->sync_request == NULL &&
3025             pers->sync_request != NULL) {
3026                 /* need to add the md_redundancy_group */
3027                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3028                         printk(KERN_WARNING
3029                                "md: cannot register extra attributes for %s\n",
3030                                mdname(mddev));
3031                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3032         }
3033         if (mddev->pers->sync_request != NULL &&
3034             pers->sync_request == NULL) {
3035                 /* need to remove the md_redundancy_group */
3036                 if (mddev->to_remove == NULL)
3037                         mddev->to_remove = &md_redundancy_group;
3038         }
3039
3040         if (mddev->pers->sync_request == NULL &&
3041             mddev->external) {
3042                 /* We are converting from a no-redundancy array
3043                  * to a redundancy array and metadata is managed
3044                  * externally so we need to be sure that writes
3045                  * won't block due to a need to transition
3046                  *      clean->dirty
3047                  * until external management is started.
3048                  */
3049                 mddev->in_sync = 0;
3050                 mddev->safemode_delay = 0;
3051                 mddev->safemode = 0;
3052         }
3053
3054         module_put(mddev->pers->owner);
3055         /* Invalidate devices that are now superfluous */
3056         list_for_each_entry(rdev, &mddev->disks, same_set)
3057                 if (rdev->raid_disk >= mddev->raid_disks) {
3058                         rdev->raid_disk = -1;
3059                         clear_bit(In_sync, &rdev->flags);
3060                 }
3061         mddev->pers = pers;
3062         mddev->private = priv;
3063         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3064         mddev->level = mddev->new_level;
3065         mddev->layout = mddev->new_layout;
3066         mddev->chunk_sectors = mddev->new_chunk_sectors;
3067         mddev->delta_disks = 0;
3068         if (mddev->pers->sync_request == NULL) {
3069                 /* this is now an array without redundancy, so
3070                  * it must always be in_sync
3071                  */
3072                 mddev->in_sync = 1;
3073                 del_timer_sync(&mddev->safemode_timer);
3074         }
3075         pers->run(mddev);
3076         mddev_resume(mddev);
3077         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3078         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3079         md_wakeup_thread(mddev->thread);
3080         sysfs_notify(&mddev->kobj, NULL, "level");
3081         md_new_event(mddev);
3082         return rv;
3083 }
3084
3085 static struct md_sysfs_entry md_level =
3086 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3087
3088
3089 static ssize_t
3090 layout_show(mddev_t *mddev, char *page)
3091 {
3092         /* just a number, not meaningful for all levels */
3093         if (mddev->reshape_position != MaxSector &&
3094             mddev->layout != mddev->new_layout)
3095                 return sprintf(page, "%d (%d)\n",
3096                                mddev->new_layout, mddev->layout);
3097         return sprintf(page, "%d\n", mddev->layout);
3098 }
3099
3100 static ssize_t
3101 layout_store(mddev_t *mddev, const char *buf, size_t len)
3102 {
3103         char *e;
3104         unsigned long n = simple_strtoul(buf, &e, 10);
3105
3106         if (!*buf || (*e && *e != '\n'))
3107                 return -EINVAL;
3108
3109         if (mddev->pers) {
3110                 int err;
3111                 if (mddev->pers->check_reshape == NULL)
3112                         return -EBUSY;
3113                 mddev->new_layout = n;
3114                 err = mddev->pers->check_reshape(mddev);
3115                 if (err) {
3116                         mddev->new_layout = mddev->layout;
3117                         return err;
3118                 }
3119         } else {
3120                 mddev->new_layout = n;
3121                 if (mddev->reshape_position == MaxSector)
3122                         mddev->layout = n;
3123         }
3124         return len;
3125 }
3126 static struct md_sysfs_entry md_layout =
3127 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3128
3129
3130 static ssize_t
3131 raid_disks_show(mddev_t *mddev, char *page)
3132 {
3133         if (mddev->raid_disks == 0)
3134                 return 0;
3135         if (mddev->reshape_position != MaxSector &&
3136             mddev->delta_disks != 0)
3137                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3138                                mddev->raid_disks - mddev->delta_disks);
3139         return sprintf(page, "%d\n", mddev->raid_disks);
3140 }
3141
3142 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3143
3144 static ssize_t
3145 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3146 {
3147         char *e;
3148         int rv = 0;
3149         unsigned long n = simple_strtoul(buf, &e, 10);
3150
3151         if (!*buf || (*e && *e != '\n'))
3152                 return -EINVAL;
3153
3154         if (mddev->pers)
3155                 rv = update_raid_disks(mddev, n);
3156         else if (mddev->reshape_position != MaxSector) {
3157                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3158                 mddev->delta_disks = n - olddisks;
3159                 mddev->raid_disks = n;
3160         } else
3161                 mddev->raid_disks = n;
3162         return rv ? rv : len;
3163 }
3164 static struct md_sysfs_entry md_raid_disks =
3165 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3166
3167 static ssize_t
3168 chunk_size_show(mddev_t *mddev, char *page)
3169 {
3170         if (mddev->reshape_position != MaxSector &&
3171             mddev->chunk_sectors != mddev->new_chunk_sectors)
3172                 return sprintf(page, "%d (%d)\n",
3173                                mddev->new_chunk_sectors << 9,
3174                                mddev->chunk_sectors << 9);
3175         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3176 }
3177
3178 static ssize_t
3179 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3180 {
3181         char *e;
3182         unsigned long n = simple_strtoul(buf, &e, 10);
3183
3184         if (!*buf || (*e && *e != '\n'))
3185                 return -EINVAL;
3186
3187         if (mddev->pers) {
3188                 int err;
3189                 if (mddev->pers->check_reshape == NULL)
3190                         return -EBUSY;
3191                 mddev->new_chunk_sectors = n >> 9;
3192                 err = mddev->pers->check_reshape(mddev);
3193                 if (err) {
3194                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3195                         return err;
3196                 }
3197         } else {
3198                 mddev->new_chunk_sectors = n >> 9;
3199                 if (mddev->reshape_position == MaxSector)
3200                         mddev->chunk_sectors = n >> 9;
3201         }
3202         return len;
3203 }
3204 static struct md_sysfs_entry md_chunk_size =
3205 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3206
3207 static ssize_t
3208 resync_start_show(mddev_t *mddev, char *page)
3209 {
3210         if (mddev->recovery_cp == MaxSector)
3211                 return sprintf(page, "none\n");
3212         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3213 }
3214
3215 static ssize_t
3216 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3217 {
3218         char *e;
3219         unsigned long long n = simple_strtoull(buf, &e, 10);
3220
3221         if (mddev->pers)
3222                 return -EBUSY;
3223         if (cmd_match(buf, "none"))
3224                 n = MaxSector;
3225         else if (!*buf || (*e && *e != '\n'))
3226                 return -EINVAL;
3227
3228         mddev->recovery_cp = n;
3229         return len;
3230 }
3231 static struct md_sysfs_entry md_resync_start =
3232 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3233
3234 /*
3235  * The array state can be:
3236  *
3237  * clear
3238  *     No devices, no size, no level
3239  *     Equivalent to STOP_ARRAY ioctl
3240  * inactive
3241  *     May have some settings, but array is not active
3242  *        all IO results in error
3243  *     When written, doesn't tear down array, but just stops it
3244  * suspended (not supported yet)
3245  *     All IO requests will block. The array can be reconfigured.
3246  *     Writing this, if accepted, will block until array is quiescent
3247  * readonly
3248  *     no resync can happen.  no superblocks get written.
3249  *     write requests fail
3250  * read-auto
3251  *     like readonly, but behaves like 'clean' on a write request.
3252  *
3253  * clean - no pending writes, but otherwise active.
3254  *     When written to inactive array, starts without resync
3255  *     If a write request arrives then
3256  *       if metadata is known, mark 'dirty' and switch to 'active'.
3257  *       if not known, block and switch to write-pending
3258  *     If written to an active array that has pending writes, then fails.
3259  * active
3260  *     fully active: IO and resync can be happening.
3261  *     When written to inactive array, starts with resync
3262  *
3263  * write-pending
3264  *     clean, but writes are blocked waiting for 'active' to be written.
3265  *
3266  * active-idle
3267  *     like active, but no writes have been seen for a while (100msec).
3268  *
3269  */
3270 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3271                    write_pending, active_idle, bad_word};
3272 static char *array_states[] = {
3273         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3274         "write-pending", "active-idle", NULL };
3275
3276 static int match_word(const char *word, char **list)
3277 {
3278         int n;
3279         for (n=0; list[n]; n++)
3280                 if (cmd_match(word, list[n]))
3281                         break;
3282         return n;
3283 }
3284
3285 static ssize_t
3286 array_state_show(mddev_t *mddev, char *page)
3287 {
3288         enum array_state st = inactive;
3289
3290         if (mddev->pers)
3291                 switch(mddev->ro) {
3292                 case 1:
3293                         st = readonly;
3294                         break;
3295                 case 2:
3296                         st = read_auto;
3297                         break;
3298                 case 0:
3299                         if (mddev->in_sync)
3300                                 st = clean;
3301                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3302                                 st = write_pending;
3303                         else if (mddev->safemode)
3304                                 st = active_idle;
3305                         else
3306                                 st = active;
3307                 }
3308         else {
3309                 if (list_empty(&mddev->disks) &&
3310                     mddev->raid_disks == 0 &&
3311                     mddev->dev_sectors == 0)
3312                         st = clear;
3313                 else
3314                         st = inactive;
3315         }
3316         return sprintf(page, "%s\n", array_states[st]);
3317 }
3318
3319 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3320 static int md_set_readonly(mddev_t * mddev, int is_open);
3321 static int do_md_run(mddev_t * mddev);
3322 static int restart_array(mddev_t *mddev);
3323
3324 static ssize_t
3325 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3326 {
3327         int err = -EINVAL;
3328         enum array_state st = match_word(buf, array_states);
3329         switch(st) {
3330         case bad_word:
3331                 break;
3332         case clear:
3333                 /* stopping an active array */
3334                 if (atomic_read(&mddev->openers) > 0)
3335                         return -EBUSY;
3336                 err = do_md_stop(mddev, 0, 0);
3337                 break;
3338         case inactive:
3339                 /* stopping an active array */
3340                 if (mddev->pers) {
3341                         if (atomic_read(&mddev->openers) > 0)
3342                                 return -EBUSY;
3343                         err = do_md_stop(mddev, 2, 0);
3344                 } else
3345                         err = 0; /* already inactive */
3346                 break;
3347         case suspended:
3348                 break; /* not supported yet */
3349         case readonly:
3350                 if (mddev->pers)
3351                         err = md_set_readonly(mddev, 0);
3352                 else {
3353                         mddev->ro = 1;
3354                         set_disk_ro(mddev->gendisk, 1);
3355                         err = do_md_run(mddev);
3356                 }
3357                 break;
3358         case read_auto:
3359                 if (mddev->pers) {
3360                         if (mddev->ro == 0)
3361                                 err = md_set_readonly(mddev, 0);
3362                         else if (mddev->ro == 1)
3363                                 err = restart_array(mddev);
3364                         if (err == 0) {
3365                                 mddev->ro = 2;
3366                                 set_disk_ro(mddev->gendisk, 0);
3367                         }
3368                 } else {
3369                         mddev->ro = 2;
3370                         err = do_md_run(mddev);
3371                 }
3372                 break;
3373         case clean:
3374                 if (mddev->pers) {
3375                         restart_array(mddev);
3376                         spin_lock_irq(&mddev->write_lock);
3377                         if (atomic_read(&mddev->writes_pending) == 0) {
3378                                 if (mddev->in_sync == 0) {
3379                                         mddev->in_sync = 1;
3380                                         if (mddev->safemode == 1)
3381                                                 mddev->safemode = 0;
3382                                         if (mddev->persistent)
3383                                                 set_bit(MD_CHANGE_CLEAN,
3384                                                         &mddev->flags);
3385                                 }
3386                                 err = 0;
3387                         } else
3388                                 err = -EBUSY;
3389                         spin_unlock_irq(&mddev->write_lock);
3390                 } else
3391                         err = -EINVAL;
3392                 break;
3393         case active:
3394                 if (mddev->pers) {
3395                         restart_array(mddev);
3396                         if (mddev->external)
3397                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3398                         wake_up(&mddev->sb_wait);
3399                         err = 0;
3400                 } else {
3401                         mddev->ro = 0;
3402                         set_disk_ro(mddev->gendisk, 0);
3403                         err = do_md_run(mddev);
3404                 }
3405                 break;
3406         case write_pending:
3407         case active_idle:
3408                 /* these cannot be set */
3409                 break;
3410         }
3411         if (err)
3412                 return err;
3413         else {
3414                 sysfs_notify_dirent(mddev->sysfs_state);
3415                 return len;
3416         }
3417 }
3418 static struct md_sysfs_entry md_array_state =
3419 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3420
3421 static ssize_t
3422 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3423         return sprintf(page, "%d\n",
3424                        atomic_read(&mddev->max_corr_read_errors));
3425 }
3426
3427 static ssize_t
3428 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3429 {
3430         char *e;
3431         unsigned long n = simple_strtoul(buf, &e, 10);
3432
3433         if (*buf && (*e == 0 || *e == '\n')) {
3434                 atomic_set(&mddev->max_corr_read_errors, n);
3435                 return len;
3436         }
3437         return -EINVAL;
3438 }
3439
3440 static struct md_sysfs_entry max_corr_read_errors =
3441 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3442         max_corrected_read_errors_store);
3443
3444 static ssize_t
3445 null_show(mddev_t *mddev, char *page)
3446 {
3447         return -EINVAL;
3448 }
3449
3450 static ssize_t
3451 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3452 {
3453         /* buf must be %d:%d\n? giving major and minor numbers */
3454         /* The new device is added to the array.
3455          * If the array has a persistent superblock, we read the
3456          * superblock to initialise info and check validity.
3457          * Otherwise, only checking done is that in bind_rdev_to_array,
3458          * which mainly checks size.
3459          */
3460         char *e;
3461         int major = simple_strtoul(buf, &e, 10);
3462         int minor;
3463         dev_t dev;
3464         mdk_rdev_t *rdev;
3465         int err;
3466
3467         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3468                 return -EINVAL;
3469         minor = simple_strtoul(e+1, &e, 10);
3470         if (*e && *e != '\n')
3471                 return -EINVAL;
3472         dev = MKDEV(major, minor);
3473         if (major != MAJOR(dev) ||
3474             minor != MINOR(dev))
3475                 return -EOVERFLOW;
3476
3477
3478         if (mddev->persistent) {
3479                 rdev = md_import_device(dev, mddev->major_version,
3480                                         mddev->minor_version);
3481                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3482                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3483                                                        mdk_rdev_t, same_set);
3484                         err = super_types[mddev->major_version]
3485                                 .load_super(rdev, rdev0, mddev->minor_version);
3486                         if (err < 0)
3487                                 goto out;
3488                 }
3489         } else if (mddev->external)
3490                 rdev = md_import_device(dev, -2, -1);
3491         else
3492                 rdev = md_import_device(dev, -1, -1);
3493
3494         if (IS_ERR(rdev))
3495                 return PTR_ERR(rdev);
3496         err = bind_rdev_to_array(rdev, mddev);
3497  out:
3498         if (err)
3499                 export_rdev(rdev);
3500         return err ? err : len;
3501 }
3502
3503 static struct md_sysfs_entry md_new_device =
3504 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3505
3506 static ssize_t
3507 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3508 {
3509         char *end;
3510         unsigned long chunk, end_chunk;
3511
3512         if (!mddev->bitmap)
3513                 goto out;
3514         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3515         while (*buf) {
3516                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3517                 if (buf == end) break;
3518                 if (*end == '-') { /* range */
3519                         buf = end + 1;
3520                         end_chunk = simple_strtoul(buf, &end, 0);
3521                         if (buf == end) break;
3522                 }
3523                 if (*end && !isspace(*end)) break;
3524                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3525                 buf = skip_spaces(end);
3526         }
3527         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3528 out:
3529         return len;
3530 }
3531
3532 static struct md_sysfs_entry md_bitmap =
3533 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3534
3535 static ssize_t
3536 size_show(mddev_t *mddev, char *page)
3537 {
3538         return sprintf(page, "%llu\n",
3539                 (unsigned long long)mddev->dev_sectors / 2);
3540 }
3541
3542 static int update_size(mddev_t *mddev, sector_t num_sectors);
3543
3544 static ssize_t
3545 size_store(mddev_t *mddev, const char *buf, size_t len)
3546 {
3547         /* If array is inactive, we can reduce the component size, but
3548          * not increase it (except from 0).
3549          * If array is active, we can try an on-line resize
3550          */
3551         sector_t sectors;
3552         int err = strict_blocks_to_sectors(buf, &sectors);
3553
3554         if (err < 0)
3555                 return err;
3556         if (mddev->pers) {
3557                 err = update_size(mddev, sectors);
3558                 md_update_sb(mddev, 1);
3559         } else {
3560                 if (mddev->dev_sectors == 0 ||
3561                     mddev->dev_sectors > sectors)
3562                         mddev->dev_sectors = sectors;
3563                 else
3564                         err = -ENOSPC;
3565         }
3566         return err ? err : len;
3567 }
3568
3569 static struct md_sysfs_entry md_size =
3570 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3571
3572
3573 /* Metdata version.
3574  * This is one of
3575  *   'none' for arrays with no metadata (good luck...)
3576  *   'external' for arrays with externally managed metadata,
3577  * or N.M for internally known formats
3578  */
3579 static ssize_t
3580 metadata_show(mddev_t *mddev, char *page)
3581 {
3582         if (mddev->persistent)
3583                 return sprintf(page, "%d.%d\n",
3584                                mddev->major_version, mddev->minor_version);
3585         else if (mddev->external)
3586                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3587         else
3588                 return sprintf(page, "none\n");
3589 }
3590
3591 static ssize_t
3592 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3593 {
3594         int major, minor;
3595         char *e;
3596         /* Changing the details of 'external' metadata is
3597          * always permitted.  Otherwise there must be
3598          * no devices attached to the array.
3599          */
3600         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3601                 ;
3602         else if (!list_empty(&mddev->disks))
3603                 return -EBUSY;
3604
3605         if (cmd_match(buf, "none")) {
3606                 mddev->persistent = 0;
3607                 mddev->external = 0;
3608                 mddev->major_version = 0;
3609                 mddev->minor_version = 90;
3610                 return len;
3611         }
3612         if (strncmp(buf, "external:", 9) == 0) {
3613                 size_t namelen = len-9;
3614                 if (namelen >= sizeof(mddev->metadata_type))
3615                         namelen = sizeof(mddev->metadata_type)-1;
3616                 strncpy(mddev->metadata_type, buf+9, namelen);
3617                 mddev->metadata_type[namelen] = 0;
3618                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3619                         mddev->metadata_type[--namelen] = 0;
3620                 mddev->persistent = 0;
3621                 mddev->external = 1;
3622                 mddev->major_version = 0;
3623                 mddev->minor_version = 90;
3624                 return len;
3625         }
3626         major = simple_strtoul(buf, &e, 10);
3627         if (e==buf || *e != '.')
3628                 return -EINVAL;
3629         buf = e+1;
3630         minor = simple_strtoul(buf, &e, 10);
3631         if (e==buf || (*e && *e != '\n') )
3632                 return -EINVAL;
3633         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3634                 return -ENOENT;
3635         mddev->major_version = major;
3636         mddev->minor_version = minor;
3637         mddev->persistent = 1;
3638         mddev->external = 0;
3639         return len;
3640 }
3641
3642 static struct md_sysfs_entry md_metadata =
3643 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3644
3645 static ssize_t
3646 action_show(mddev_t *mddev, char *page)
3647 {
3648         char *type = "idle";
3649         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3650                 type = "frozen";
3651         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3652             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3653                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3654                         type = "reshape";
3655                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3656                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3657                                 type = "resync";
3658                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3659                                 type = "check";
3660                         else
3661                                 type = "repair";
3662                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3663                         type = "recover";
3664         }
3665         return sprintf(page, "%s\n", type);
3666 }
3667
3668 static ssize_t
3669 action_store(mddev_t *mddev, const char *page, size_t len)
3670 {
3671         if (!mddev->pers || !mddev->pers->sync_request)
3672                 return -EINVAL;
3673
3674         if (cmd_match(page, "frozen"))
3675                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3676         else
3677                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3678
3679         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3680                 if (mddev->sync_thread) {
3681                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3682                         md_unregister_thread(mddev->sync_thread);
3683                         mddev->sync_thread = NULL;
3684                         mddev->recovery = 0;
3685                 }
3686         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3687                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3688                 return -EBUSY;
3689         else if (cmd_match(page, "resync"))
3690                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3691         else if (cmd_match(page, "recover")) {
3692                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3693                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3694         } else if (cmd_match(page, "reshape")) {
3695                 int err;
3696                 if (mddev->pers->start_reshape == NULL)
3697                         return -EINVAL;
3698                 err = mddev->pers->start_reshape(mddev);
3699                 if (err)
3700                         return err;
3701                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3702         } else {
3703                 if (cmd_match(page, "check"))
3704                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3705                 else if (!cmd_match(page, "repair"))
3706                         return -EINVAL;
3707                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3708                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3709         }
3710         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3711         md_wakeup_thread(mddev->thread);
3712         sysfs_notify_dirent(mddev->sysfs_action);
3713         return len;
3714 }
3715
3716 static ssize_t
3717 mismatch_cnt_show(mddev_t *mddev, char *page)
3718 {
3719         return sprintf(page, "%llu\n",
3720                        (unsigned long long) mddev->resync_mismatches);
3721 }
3722
3723 static struct md_sysfs_entry md_scan_mode =
3724 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3725
3726
3727 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3728
3729 static ssize_t
3730 sync_min_show(mddev_t *mddev, char *page)
3731 {
3732         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3733                        mddev->sync_speed_min ? "local": "system");
3734 }
3735
3736 static ssize_t
3737 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3738 {
3739         int min;
3740         char *e;
3741         if (strncmp(buf, "system", 6)==0) {
3742                 mddev->sync_speed_min = 0;
3743                 return len;
3744         }
3745         min = simple_strtoul(buf, &e, 10);
3746         if (buf == e || (*e && *e != '\n') || min <= 0)
3747                 return -EINVAL;
3748         mddev->sync_speed_min = min;
3749         return len;
3750 }
3751
3752 static struct md_sysfs_entry md_sync_min =
3753 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3754
3755 static ssize_t
3756 sync_max_show(mddev_t *mddev, char *page)
3757 {
3758         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3759                        mddev->sync_speed_max ? "local": "system");
3760 }
3761
3762 static ssize_t
3763 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3764 {
3765         int max;
3766         char *e;
3767         if (strncmp(buf, "system", 6)==0) {
3768                 mddev->sync_speed_max = 0;
3769                 return len;
3770         }
3771         max = simple_strtoul(buf, &e, 10);
3772         if (buf == e || (*e && *e != '\n') || max <= 0)
3773                 return -EINVAL;
3774         mddev->sync_speed_max = max;
3775         return len;
3776 }
3777
3778 static struct md_sysfs_entry md_sync_max =
3779 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3780
3781 static ssize_t
3782 degraded_show(mddev_t *mddev, char *page)
3783 {
3784         return sprintf(page, "%d\n", mddev->degraded);
3785 }
3786 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3787
3788 static ssize_t
3789 sync_force_parallel_show(mddev_t *mddev, char *page)
3790 {
3791         return sprintf(page, "%d\n", mddev->parallel_resync);
3792 }
3793
3794 static ssize_t
3795 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3796 {
3797         long n;
3798
3799         if (strict_strtol(buf, 10, &n))
3800                 return -EINVAL;
3801
3802         if (n != 0 && n != 1)
3803                 return -EINVAL;
3804
3805         mddev->parallel_resync = n;
3806
3807         if (mddev->sync_thread)
3808                 wake_up(&resync_wait);
3809
3810         return len;
3811 }
3812
3813 /* force parallel resync, even with shared block devices */
3814 static struct md_sysfs_entry md_sync_force_parallel =
3815 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3816        sync_force_parallel_show, sync_force_parallel_store);
3817
3818 static ssize_t
3819 sync_speed_show(mddev_t *mddev, char *page)
3820 {
3821         unsigned long resync, dt, db;
3822         if (mddev->curr_resync == 0)
3823                 return sprintf(page, "none\n");
3824         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3825         dt = (jiffies - mddev->resync_mark) / HZ;
3826         if (!dt) dt++;
3827         db = resync - mddev->resync_mark_cnt;
3828         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3829 }
3830
3831 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3832
3833 static ssize_t
3834 sync_completed_show(mddev_t *mddev, char *page)
3835 {
3836         unsigned long max_sectors, resync;
3837
3838         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3839                 return sprintf(page, "none\n");
3840
3841         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3842                 max_sectors = mddev->resync_max_sectors;
3843         else
3844                 max_sectors = mddev->dev_sectors;
3845
3846         resync = mddev->curr_resync_completed;
3847         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3848 }
3849
3850 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3851
3852 static ssize_t
3853 min_sync_show(mddev_t *mddev, char *page)
3854 {
3855         return sprintf(page, "%llu\n",
3856                        (unsigned long long)mddev->resync_min);
3857 }
3858 static ssize_t
3859 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3860 {
3861         unsigned long long min;
3862         if (strict_strtoull(buf, 10, &min))
3863                 return -EINVAL;
3864         if (min > mddev->resync_max)
3865                 return -EINVAL;
3866         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3867                 return -EBUSY;
3868
3869         /* Must be a multiple of chunk_size */
3870         if (mddev->chunk_sectors) {
3871                 sector_t temp = min;
3872                 if (sector_div(temp, mddev->chunk_sectors))
3873                         return -EINVAL;
3874         }
3875         mddev->resync_min = min;
3876
3877         return len;
3878 }
3879
3880 static struct md_sysfs_entry md_min_sync =
3881 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3882
3883 static ssize_t
3884 max_sync_show(mddev_t *mddev, char *page)
3885 {
3886         if (mddev->resync_max == MaxSector)
3887                 return sprintf(page, "max\n");
3888         else
3889                 return sprintf(page, "%llu\n",
3890                                (unsigned long long)mddev->resync_max);
3891 }
3892 static ssize_t
3893 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3894 {
3895         if (strncmp(buf, "max", 3) == 0)
3896                 mddev->resync_max = MaxSector;
3897         else {
3898                 unsigned long long max;
3899                 if (strict_strtoull(buf, 10, &max))
3900                         return -EINVAL;
3901                 if (max < mddev->resync_min)
3902                         return -EINVAL;
3903                 if (max < mddev->resync_max &&
3904                     mddev->ro == 0 &&
3905                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3906                         return -EBUSY;
3907
3908                 /* Must be a multiple of chunk_size */
3909                 if (mddev->chunk_sectors) {
3910                         sector_t temp = max;
3911                         if (sector_div(temp, mddev->chunk_sectors))
3912                                 return -EINVAL;
3913                 }
3914                 mddev->resync_max = max;
3915         }
3916         wake_up(&mddev->recovery_wait);
3917         return len;
3918 }
3919
3920 static struct md_sysfs_entry md_max_sync =
3921 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3922
3923 static ssize_t
3924 suspend_lo_show(mddev_t *mddev, char *page)
3925 {
3926         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3927 }
3928
3929 static ssize_t
3930 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3931 {
3932         char *e;
3933         unsigned long long new = simple_strtoull(buf, &e, 10);
3934
3935         if (mddev->pers == NULL ||
3936             mddev->pers->quiesce == NULL)
3937                 return -EINVAL;
3938         if (buf == e || (*e && *e != '\n'))
3939                 return -EINVAL;
3940         if (new >= mddev->suspend_hi ||
3941             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3942                 mddev->suspend_lo = new;
3943                 mddev->pers->quiesce(mddev, 2);
3944                 return len;
3945         } else
3946                 return -EINVAL;
3947 }
3948 static struct md_sysfs_entry md_suspend_lo =
3949 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3950
3951
3952 static ssize_t
3953 suspend_hi_show(mddev_t *mddev, char *page)
3954 {
3955         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3956 }
3957
3958 static ssize_t
3959 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3960 {
3961         char *e;
3962         unsigned long long new = simple_strtoull(buf, &e, 10);
3963
3964         if (mddev->pers == NULL ||
3965             mddev->pers->quiesce == NULL)
3966                 return -EINVAL;
3967         if (buf == e || (*e && *e != '\n'))
3968                 return -EINVAL;
3969         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3970             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3971                 mddev->suspend_hi = new;
3972                 mddev->pers->quiesce(mddev, 1);
3973                 mddev->pers->quiesce(mddev, 0);
3974                 return len;
3975         } else
3976                 return -EINVAL;
3977 }
3978 static struct md_sysfs_entry md_suspend_hi =
3979 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3980
3981 static ssize_t
3982 reshape_position_show(mddev_t *mddev, char *page)
3983 {
3984         if (mddev->reshape_position != MaxSector)
3985                 return sprintf(page, "%llu\n",
3986                                (unsigned long long)mddev->reshape_position);
3987         strcpy(page, "none\n");
3988         return 5;
3989 }
3990
3991 static ssize_t
3992 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3993 {
3994         char *e;
3995         unsigned long long new = simple_strtoull(buf, &e, 10);
3996         if (mddev->pers)
3997                 return -EBUSY;
3998         if (buf == e || (*e && *e != '\n'))
3999                 return -EINVAL;
4000         mddev->reshape_position = new;
4001         mddev->delta_disks = 0;
4002         mddev->new_level = mddev->level;
4003         mddev->new_layout = mddev->layout;
4004         mddev->new_chunk_sectors = mddev->chunk_sectors;
4005         return len;
4006 }
4007
4008 static struct md_sysfs_entry md_reshape_position =
4009 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4010        reshape_position_store);
4011
4012 static ssize_t
4013 array_size_show(mddev_t *mddev, char *page)
4014 {
4015         if (mddev->external_size)
4016                 return sprintf(page, "%llu\n",
4017                                (unsigned long long)mddev->array_sectors/2);
4018         else
4019                 return sprintf(page, "default\n");
4020 }
4021
4022 static ssize_t
4023 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4024 {
4025         sector_t sectors;
4026
4027         if (strncmp(buf, "default", 7) == 0) {
4028                 if (mddev->pers)
4029                         sectors = mddev->pers->size(mddev, 0, 0);
4030                 else
4031                         sectors = mddev->array_sectors;
4032
4033                 mddev->external_size = 0;
4034         } else {
4035                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4036                         return -EINVAL;
4037                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4038                         return -E2BIG;
4039
4040                 mddev->external_size = 1;
4041         }
4042
4043         mddev->array_sectors = sectors;
4044         set_capacity(mddev->gendisk, mddev->array_sectors);
4045         if (mddev->pers)
4046                 revalidate_disk(mddev->gendisk);
4047
4048         return len;
4049 }
4050
4051 static struct md_sysfs_entry md_array_size =
4052 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4053        array_size_store);
4054
4055 static struct attribute *md_default_attrs[] = {
4056         &md_level.attr,
4057         &md_layout.attr,
4058         &md_raid_disks.attr,
4059         &md_chunk_size.attr,
4060         &md_size.attr,
4061         &md_resync_start.attr,
4062         &md_metadata.attr,
4063         &md_new_device.attr,
4064         &md_safe_delay.attr,
4065         &md_array_state.attr,
4066         &md_reshape_position.attr,
4067         &md_array_size.attr,
4068         &max_corr_read_errors.attr,
4069         NULL,
4070 };
4071
4072 static struct attribute *md_redundancy_attrs[] = {
4073         &md_scan_mode.attr,
4074         &md_mismatches.attr,
4075         &md_sync_min.attr,
4076         &md_sync_max.attr,
4077         &md_sync_speed.attr,
4078         &md_sync_force_parallel.attr,
4079         &md_sync_completed.attr,
4080         &md_min_sync.attr,
4081         &md_max_sync.attr,
4082         &md_suspend_lo.attr,
4083         &md_suspend_hi.attr,
4084         &md_bitmap.attr,
4085         &md_degraded.attr,
4086         NULL,
4087 };
4088 static struct attribute_group md_redundancy_group = {
4089         .name = NULL,
4090         .attrs = md_redundancy_attrs,
4091 };
4092
4093
4094 static ssize_t
4095 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4096 {
4097         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4098         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4099         ssize_t rv;
4100
4101         if (!entry->show)
4102                 return -EIO;
4103         rv = mddev_lock(mddev);
4104         if (!rv) {
4105                 rv = entry->show(mddev, page);
4106                 mddev_unlock(mddev);
4107         }
4108         return rv;
4109 }
4110
4111 static ssize_t
4112 md_attr_store(struct kobject *kobj, struct attribute *attr,
4113               const char *page, size_t length)
4114 {
4115         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4116         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4117         ssize_t rv;
4118
4119         if (!entry->store)
4120                 return -EIO;
4121         if (!capable(CAP_SYS_ADMIN))
4122                 return -EACCES;
4123         rv = mddev_lock(mddev);
4124         if (mddev->hold_active == UNTIL_IOCTL)
4125                 mddev->hold_active = 0;
4126         if (!rv) {
4127                 rv = entry->store(mddev, page, length);
4128                 mddev_unlock(mddev);
4129         }
4130         return rv;
4131 }
4132
4133 static void md_free(struct kobject *ko)
4134 {
4135         mddev_t *mddev = container_of(ko, mddev_t, kobj);
4136
4137         if (mddev->sysfs_state)
4138                 sysfs_put(mddev->sysfs_state);
4139
4140         if (mddev->gendisk) {
4141                 del_gendisk(mddev->gendisk);
4142                 put_disk(mddev->gendisk);
4143         }
4144         if (mddev->queue)
4145                 blk_cleanup_queue(mddev->queue);
4146
4147         kfree(mddev);
4148 }
4149
4150 static const struct sysfs_ops md_sysfs_ops = {
4151         .show   = md_attr_show,
4152         .store  = md_attr_store,
4153 };
4154 static struct kobj_type md_ktype = {
4155         .release        = md_free,
4156         .sysfs_ops      = &md_sysfs_ops,
4157         .default_attrs  = md_default_attrs,
4158 };
4159
4160 int mdp_major = 0;
4161
4162 static void mddev_delayed_delete(struct work_struct *ws)
4163 {
4164         mddev_t *mddev = container_of(ws, mddev_t, del_work);
4165
4166         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4167         kobject_del(&mddev->kobj);
4168         kobject_put(&mddev->kobj);
4169 }
4170
4171 static int md_alloc(dev_t dev, char *name)
4172 {
4173         static DEFINE_MUTEX(disks_mutex);
4174         mddev_t *mddev = mddev_find(dev);
4175         struct gendisk *disk;
4176         int partitioned;
4177         int shift;
4178         int unit;
4179         int error;
4180
4181         if (!mddev)
4182                 return -ENODEV;
4183
4184         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4185         shift = partitioned ? MdpMinorShift : 0;
4186         unit = MINOR(mddev->unit) >> shift;
4187
4188         /* wait for any previous instance if this device
4189          * to be completed removed (mddev_delayed_delete).
4190          */
4191         flush_scheduled_work();
4192
4193         mutex_lock(&disks_mutex);
4194         error = -EEXIST;
4195         if (mddev->gendisk)
4196                 goto abort;
4197
4198         if (name) {
4199                 /* Need to ensure that 'name' is not a duplicate.
4200                  */
4201                 mddev_t *mddev2;
4202                 spin_lock(&all_mddevs_lock);
4203
4204                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4205                         if (mddev2->gendisk &&
4206                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4207                                 spin_unlock(&all_mddevs_lock);
4208                                 goto abort;
4209                         }
4210                 spin_unlock(&all_mddevs_lock);
4211         }
4212
4213         error = -ENOMEM;
4214         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4215         if (!mddev->queue)
4216                 goto abort;
4217         mddev->queue->queuedata = mddev;
4218
4219         /* Can be unlocked because the queue is new: no concurrency */
4220         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4221
4222         blk_queue_make_request(mddev->queue, md_make_request);
4223
4224         disk = alloc_disk(1 << shift);
4225         if (!disk) {
4226                 blk_cleanup_queue(mddev->queue);
4227                 mddev->queue = NULL;
4228                 goto abort;
4229         }
4230         disk->major = MAJOR(mddev->unit);
4231         disk->first_minor = unit << shift;
4232         if (name)
4233                 strcpy(disk->disk_name, name);
4234         else if (partitioned)
4235                 sprintf(disk->disk_name, "md_d%d", unit);
4236         else
4237                 sprintf(disk->disk_name, "md%d", unit);
4238         disk->fops = &md_fops;
4239         disk->private_data = mddev;
4240         disk->queue = mddev->queue;
4241         /* Allow extended partitions.  This makes the
4242          * 'mdp' device redundant, but we can't really
4243          * remove it now.
4244          */
4245         disk->flags |= GENHD_FL_EXT_DEVT;
4246         add_disk(disk);
4247         mddev->gendisk = disk;
4248         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4249                                      &disk_to_dev(disk)->kobj, "%s", "md");
4250         if (error) {
4251                 /* This isn't possible, but as kobject_init_and_add is marked
4252                  * __must_check, we must do something with the result
4253                  */
4254                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4255                        disk->disk_name);
4256                 error = 0;
4257         }
4258         if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4259                 printk(KERN_DEBUG "pointless warning\n");
4260  abort:
4261         mutex_unlock(&disks_mutex);
4262         if (!error) {
4263                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4264                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, NULL, "array_state");
4265         }
4266         mddev_put(mddev);
4267         return error;
4268 }
4269
4270 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4271 {
4272         md_alloc(dev, NULL);
4273         return NULL;
4274 }
4275
4276 static int add_named_array(const char *val, struct kernel_param *kp)
4277 {
4278         /* val must be "md_*" where * is not all digits.
4279          * We allocate an array with a large free minor number, and
4280          * set the name to val.  val must not already be an active name.
4281          */
4282         int len = strlen(val);
4283         char buf[DISK_NAME_LEN];
4284
4285         while (len && val[len-1] == '\n')
4286                 len--;
4287         if (len >= DISK_NAME_LEN)
4288                 return -E2BIG;
4289         strlcpy(buf, val, len+1);
4290         if (strncmp(buf, "md_", 3) != 0)
4291                 return -EINVAL;
4292         return md_alloc(0, buf);
4293 }
4294
4295 static void md_safemode_timeout(unsigned long data)
4296 {
4297         mddev_t *mddev = (mddev_t *) data;
4298
4299         if (!atomic_read(&mddev->writes_pending)) {
4300                 mddev->safemode = 1;
4301                 if (mddev->external)
4302                         sysfs_notify_dirent(mddev->sysfs_state);
4303         }
4304         md_wakeup_thread(mddev->thread);
4305 }
4306
4307 static int start_dirty_degraded;
4308
4309 static int md_run(mddev_t *mddev)
4310 {
4311         int err;
4312         mdk_rdev_t *rdev;
4313         struct mdk_personality *pers;
4314
4315         if (list_empty(&mddev->disks))
4316                 /* cannot run an array with no devices.. */
4317                 return -EINVAL;
4318
4319         if (mddev->pers)
4320                 return -EBUSY;
4321
4322         /* These two calls synchronise us with the
4323          * sysfs_remove_group calls in mddev_unlock,
4324          * so they must have completed.
4325          */
4326         mutex_lock(&mddev->open_mutex);
4327         mutex_unlock(&mddev->open_mutex);
4328
4329         /*
4330          * Analyze all RAID superblock(s)
4331          */
4332         if (!mddev->raid_disks) {
4333                 if (!mddev->persistent)
4334                         return -EINVAL;
4335                 analyze_sbs(mddev);
4336         }
4337
4338         if (mddev->level != LEVEL_NONE)
4339                 request_module("md-level-%d", mddev->level);
4340         else if (mddev->clevel[0])
4341                 request_module("md-%s", mddev->clevel);
4342
4343         /*
4344          * Drop all container device buffers, from now on
4345          * the only valid external interface is through the md
4346          * device.
4347          */
4348         list_for_each_entry(rdev, &mddev->disks, same_set) {
4349                 if (test_bit(Faulty, &rdev->flags))
4350                         continue;
4351                 sync_blockdev(rdev->bdev);
4352                 invalidate_bdev(rdev->bdev);
4353
4354                 /* perform some consistency tests on the device.
4355                  * We don't want the data to overlap the metadata,
4356                  * Internal Bitmap issues have been handled elsewhere.
4357                  */
4358                 if (rdev->data_offset < rdev->sb_start) {
4359                         if (mddev->dev_sectors &&
4360                             rdev->data_offset + mddev->dev_sectors
4361                             > rdev->sb_start) {
4362                                 printk("md: %s: data overlaps metadata\n",
4363                                        mdname(mddev));
4364                                 return -EINVAL;
4365                         }
4366                 } else {
4367                         if (rdev->sb_start + rdev->sb_size/512
4368                             > rdev->data_offset) {
4369                                 printk("md: %s: metadata overlaps data\n",
4370                                        mdname(mddev));
4371                                 return -EINVAL;
4372                         }
4373                 }
4374                 sysfs_notify_dirent(rdev->sysfs_state);
4375         }
4376
4377         spin_lock(&pers_lock);
4378         pers = find_pers(mddev->level, mddev->clevel);
4379         if (!pers || !try_module_get(pers->owner)) {
4380                 spin_unlock(&pers_lock);
4381                 if (mddev->level != LEVEL_NONE)
4382                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4383                                mddev->level);
4384                 else
4385                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4386                                mddev->clevel);
4387                 return -EINVAL;
4388         }
4389         mddev->pers = pers;
4390         spin_unlock(&pers_lock);
4391         if (mddev->level != pers->level) {
4392                 mddev->level = pers->level;
4393                 mddev->new_level = pers->level;
4394         }
4395         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4396
4397         if (mddev->reshape_position != MaxSector &&
4398             pers->start_reshape == NULL) {
4399                 /* This personality cannot handle reshaping... */
4400                 mddev->pers = NULL;
4401                 module_put(pers->owner);
4402                 return -EINVAL;
4403         }
4404
4405         if (pers->sync_request) {
4406                 /* Warn if this is a potentially silly
4407                  * configuration.
4408                  */
4409                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4410                 mdk_rdev_t *rdev2;
4411                 int warned = 0;
4412
4413                 list_for_each_entry(rdev, &mddev->disks, same_set)
4414                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4415                                 if (rdev < rdev2 &&
4416                                     rdev->bdev->bd_contains ==
4417                                     rdev2->bdev->bd_contains) {
4418                                         printk(KERN_WARNING
4419                                                "%s: WARNING: %s appears to be"
4420                                                " on the same physical disk as"
4421                                                " %s.\n",
4422                                                mdname(mddev),
4423                                                bdevname(rdev->bdev,b),
4424                                                bdevname(rdev2->bdev,b2));
4425                                         warned = 1;
4426                                 }
4427                         }
4428
4429                 if (warned)
4430                         printk(KERN_WARNING
4431                                "True protection against single-disk"
4432                                " failure might be compromised.\n");
4433         }
4434
4435         mddev->recovery = 0;
4436         /* may be over-ridden by personality */
4437         mddev->resync_max_sectors = mddev->dev_sectors;
4438
4439         mddev->barriers_work = 1;
4440         mddev->ok_start_degraded = start_dirty_degraded;
4441
4442         if (start_readonly && mddev->ro == 0)
4443                 mddev->ro = 2; /* read-only, but switch on first write */
4444
4445         err = mddev->pers->run(mddev);
4446         if (err)
4447                 printk(KERN_ERR "md: pers->run() failed ...\n");
4448         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4449                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4450                           " but 'external_size' not in effect?\n", __func__);
4451                 printk(KERN_ERR
4452                        "md: invalid array_size %llu > default size %llu\n",
4453                        (unsigned long long)mddev->array_sectors / 2,
4454                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4455                 err = -EINVAL;
4456                 mddev->pers->stop(mddev);
4457         }
4458         if (err == 0 && mddev->pers->sync_request) {
4459                 err = bitmap_create(mddev);
4460                 if (err) {
4461                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4462                                mdname(mddev), err);
4463                         mddev->pers->stop(mddev);
4464                 }
4465         }
4466         if (err) {
4467                 module_put(mddev->pers->owner);
4468                 mddev->pers = NULL;
4469                 bitmap_destroy(mddev);
4470                 return err;
4471         }
4472         if (mddev->pers->sync_request) {
4473                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4474                         printk(KERN_WARNING
4475                                "md: cannot register extra attributes for %s\n",
4476                                mdname(mddev));
4477                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
4478         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4479                 mddev->ro = 0;
4480
4481         atomic_set(&mddev->writes_pending,0);
4482         atomic_set(&mddev->max_corr_read_errors,
4483                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4484         mddev->safemode = 0;
4485         mddev->safemode_timer.function = md_safemode_timeout;
4486         mddev->safemode_timer.data = (unsigned long) mddev;
4487         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4488         mddev->in_sync = 1;
4489
4490         list_for_each_entry(rdev, &mddev->disks, same_set)
4491                 if (rdev->raid_disk >= 0) {
4492                         char nm[20];
4493                         sprintf(nm, "rd%d", rdev->raid_disk);
4494                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4495                                 printk("md: cannot register %s for %s\n",
4496                                        nm, mdname(mddev));
4497                 }
4498
4499         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4500
4501         if (mddev->flags)
4502                 md_update_sb(mddev, 0);
4503
4504         md_wakeup_thread(mddev->thread);
4505         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4506
4507         md_new_event(mddev);
4508         sysfs_notify_dirent(mddev->sysfs_state);
4509         if (mddev->sysfs_action)
4510                 sysfs_notify_dirent(mddev->sysfs_action);
4511         sysfs_notify(&mddev->kobj, NULL, "degraded");
4512         return 0;
4513 }
4514
4515 static int do_md_run(mddev_t *mddev)
4516 {
4517         int err;
4518
4519         err = md_run(mddev);
4520         if (err)
4521                 goto out;
4522
4523         set_capacity(mddev->gendisk, mddev->array_sectors);
4524         revalidate_disk(mddev->gendisk);
4525         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4526 out:
4527         return err;
4528 }
4529
4530 static int restart_array(mddev_t *mddev)
4531 {
4532         struct gendisk *disk = mddev->gendisk;
4533
4534         /* Complain if it has no devices */
4535         if (list_empty(&mddev->disks))
4536                 return -ENXIO;
4537         if (!mddev->pers)
4538                 return -EINVAL;
4539         if (!mddev->ro)
4540                 return -EBUSY;
4541         mddev->safemode = 0;
4542         mddev->ro = 0;
4543         set_disk_ro(disk, 0);
4544         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4545                 mdname(mddev));
4546         /* Kick recovery or resync if necessary */
4547         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4548         md_wakeup_thread(mddev->thread);
4549         md_wakeup_thread(mddev->sync_thread);
4550         sysfs_notify_dirent(mddev->sysfs_state);
4551         return 0;
4552 }
4553
4554 /* similar to deny_write_access, but accounts for our holding a reference
4555  * to the file ourselves */
4556 static int deny_bitmap_write_access(struct file * file)
4557 {
4558         struct inode *inode = file->f_mapping->host;
4559
4560         spin_lock(&inode->i_lock);
4561         if (atomic_read(&inode->i_writecount) > 1) {
4562                 spin_unlock(&inode->i_lock);
4563                 return -ETXTBSY;
4564         }
4565         atomic_set(&inode->i_writecount, -1);
4566         spin_unlock(&inode->i_lock);
4567
4568         return 0;
4569 }
4570
4571 void restore_bitmap_write_access(struct file *file)
4572 {
4573         struct inode *inode = file->f_mapping->host;
4574
4575         spin_lock(&inode->i_lock);
4576         atomic_set(&inode->i_writecount, 1);
4577         spin_unlock(&inode->i_lock);
4578 }
4579
4580 static void md_clean(mddev_t *mddev)
4581 {
4582         mddev->array_sectors = 0;
4583         mddev->external_size = 0;
4584         mddev->dev_sectors = 0;
4585         mddev->raid_disks = 0;
4586         mddev->recovery_cp = 0;
4587         mddev->resync_min = 0;
4588         mddev->resync_max = MaxSector;
4589         mddev->reshape_position = MaxSector;
4590         mddev->external = 0;
4591         mddev->persistent = 0;
4592         mddev->level = LEVEL_NONE;
4593         mddev->clevel[0] = 0;
4594         mddev->flags = 0;
4595         mddev->ro = 0;
4596         mddev->metadata_type[0] = 0;
4597         mddev->chunk_sectors = 0;
4598         mddev->ctime = mddev->utime = 0;
4599         mddev->layout = 0;
4600         mddev->max_disks = 0;
4601         mddev->events = 0;
4602         mddev->can_decrease_events = 0;
4603         mddev->delta_disks = 0;
4604         mddev->new_level = LEVEL_NONE;
4605         mddev->new_layout = 0;
4606         mddev->new_chunk_sectors = 0;
4607         mddev->curr_resync = 0;
4608         mddev->resync_mismatches = 0;
4609         mddev->suspend_lo = mddev->suspend_hi = 0;
4610         mddev->sync_speed_min = mddev->sync_speed_max = 0;
4611         mddev->recovery = 0;
4612         mddev->in_sync = 0;
4613         mddev->degraded = 0;
4614         mddev->barriers_work = 0;
4615         mddev->safemode = 0;
4616         mddev->bitmap_info.offset = 0;
4617         mddev->bitmap_info.default_offset = 0;
4618         mddev->bitmap_info.chunksize = 0;
4619         mddev->bitmap_info.daemon_sleep = 0;
4620         mddev->bitmap_info.max_write_behind = 0;
4621 }
4622
4623 static void md_stop_writes(mddev_t *mddev)
4624 {
4625         if (mddev->sync_thread) {
4626                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4627                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4628                 md_unregister_thread(mddev->sync_thread);
4629                 mddev->sync_thread = NULL;
4630         }
4631
4632         del_timer_sync(&mddev->safemode_timer);
4633
4634         bitmap_flush(mddev);
4635         md_super_wait(mddev);
4636
4637         if (!mddev->in_sync || mddev->flags) {
4638                 /* mark array as shutdown cleanly */
4639                 mddev->in_sync = 1;
4640                 md_update_sb(mddev, 1);
4641         }
4642 }
4643
4644 static void md_stop(mddev_t *mddev)
4645 {
4646         md_stop_writes(mddev);
4647
4648         mddev->pers->stop(mddev);
4649         if (mddev->pers->sync_request && mddev->to_remove == NULL)
4650                 mddev->to_remove = &md_redundancy_group;
4651         module_put(mddev->pers->owner);
4652         mddev->pers = NULL;
4653         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4654 }
4655
4656 static int md_set_readonly(mddev_t *mddev, int is_open)
4657 {
4658         int err = 0;
4659         mutex_lock(&mddev->open_mutex);
4660         if (atomic_read(&mddev->openers) > is_open) {
4661                 printk("md: %s still in use.\n",mdname(mddev));
4662                 err = -EBUSY;
4663                 goto out;
4664         }
4665         if (mddev->pers) {
4666                 md_stop_writes(mddev);
4667
4668                 err  = -ENXIO;
4669                 if (mddev->ro==1)
4670                         goto out;
4671                 mddev->ro = 1;
4672                 set_disk_ro(mddev->gendisk, 1);
4673                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4674                 sysfs_notify_dirent(mddev->sysfs_state);
4675                 err = 0;
4676         }
4677 out:
4678         mutex_unlock(&mddev->open_mutex);
4679         return err;
4680 }
4681
4682 /* mode:
4683  *   0 - completely stop and dis-assemble array
4684  *   2 - stop but do not disassemble array
4685  */
4686 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4687 {
4688         int err = 0;
4689         struct gendisk *disk = mddev->gendisk;
4690         mdk_rdev_t *rdev;
4691
4692         mutex_lock(&mddev->open_mutex);
4693         if (atomic_read(&mddev->openers) > is_open) {
4694                 printk("md: %s still in use.\n",mdname(mddev));
4695                 err = -EBUSY;
4696         } else if (mddev->pers) {
4697
4698                 if (mddev->ro)
4699                         set_disk_ro(disk, 0);
4700
4701                 md_stop(mddev);
4702                 mddev->queue->merge_bvec_fn = NULL;
4703                 mddev->queue->unplug_fn = NULL;
4704                 mddev->queue->backing_dev_info.congested_fn = NULL;
4705
4706                 /* tell userspace to handle 'inactive' */
4707                 sysfs_notify_dirent(mddev->sysfs_state);
4708
4709                 list_for_each_entry(rdev, &mddev->disks, same_set)
4710                         if (rdev->raid_disk >= 0) {
4711                                 char nm[20];
4712                                 sprintf(nm, "rd%d", rdev->raid_disk);
4713                                 sysfs_remove_link(&mddev->kobj, nm);
4714                         }
4715
4716                 set_capacity(disk, 0);
4717                 revalidate_disk(disk);
4718
4719                 if (mddev->ro)
4720                         mddev->ro = 0;
4721
4722                 err = 0;
4723         }
4724         mutex_unlock(&mddev->open_mutex);
4725         if (err)
4726                 return err;
4727         /*
4728          * Free resources if final stop
4729          */
4730         if (mode == 0) {
4731
4732                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4733
4734                 bitmap_destroy(mddev);
4735                 if (mddev->bitmap_info.file) {
4736                         restore_bitmap_write_access(mddev->bitmap_info.file);
4737                         fput(mddev->bitmap_info.file);
4738                         mddev->bitmap_info.file = NULL;
4739                 }
4740                 mddev->bitmap_info.offset = 0;
4741
4742                 export_array(mddev);
4743
4744                 md_clean(mddev);
4745                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4746                 if (mddev->hold_active == UNTIL_STOP)
4747                         mddev->hold_active = 0;
4748
4749         }
4750         err = 0;
4751         blk_integrity_unregister(disk);
4752         md_new_event(mddev);
4753         sysfs_notify_dirent(mddev->sysfs_state);
4754         return err;
4755 }
4756
4757 #ifndef MODULE
4758 static void autorun_array(mddev_t *mddev)
4759 {
4760         mdk_rdev_t *rdev;
4761         int err;
4762
4763         if (list_empty(&mddev->disks))
4764                 return;
4765
4766         printk(KERN_INFO "md: running: ");
4767
4768         list_for_each_entry(rdev, &mddev->disks, same_set) {
4769                 char b[BDEVNAME_SIZE];
4770                 printk("<%s>", bdevname(rdev->bdev,b));
4771         }
4772         printk("\n");
4773
4774         err = do_md_run(mddev);
4775         if (err) {
4776                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4777                 do_md_stop(mddev, 0, 0);
4778         }
4779 }
4780
4781 /*
4782  * lets try to run arrays based on all disks that have arrived
4783  * until now. (those are in pending_raid_disks)
4784  *
4785  * the method: pick the first pending disk, collect all disks with
4786  * the same UUID, remove all from the pending list and put them into
4787  * the 'same_array' list. Then order this list based on superblock
4788  * update time (freshest comes first), kick out 'old' disks and
4789  * compare superblocks. If everything's fine then run it.
4790  *
4791  * If "unit" is allocated, then bump its reference count
4792  */
4793 static void autorun_devices(int part)
4794 {
4795         mdk_rdev_t *rdev0, *rdev, *tmp;
4796         mddev_t *mddev;
4797         char b[BDEVNAME_SIZE];
4798
4799         printk(KERN_INFO "md: autorun ...\n");
4800         while (!list_empty(&pending_raid_disks)) {
4801                 int unit;
4802                 dev_t dev;
4803                 LIST_HEAD(candidates);
4804                 rdev0 = list_entry(pending_raid_disks.next,
4805                                          mdk_rdev_t, same_set);
4806
4807                 printk(KERN_INFO "md: considering %s ...\n",
4808                         bdevname(rdev0->bdev,b));
4809                 INIT_LIST_HEAD(&candidates);
4810                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4811                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4812                                 printk(KERN_INFO "md:  adding %s ...\n",
4813                                         bdevname(rdev->bdev,b));
4814                                 list_move(&rdev->same_set, &candidates);
4815                         }
4816                 /*
4817                  * now we have a set of devices, with all of them having
4818                  * mostly sane superblocks. It's time to allocate the
4819                  * mddev.
4820                  */
4821                 if (part) {
4822                         dev = MKDEV(mdp_major,
4823                                     rdev0->preferred_minor << MdpMinorShift);
4824                         unit = MINOR(dev) >> MdpMinorShift;
4825                 } else {
4826                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4827                         unit = MINOR(dev);
4828                 }
4829                 if (rdev0->preferred_minor != unit) {
4830                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4831                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4832                         break;
4833                 }
4834
4835                 md_probe(dev, NULL, NULL);
4836                 mddev = mddev_find(dev);
4837                 if (!mddev || !mddev->gendisk) {
4838                         if (mddev)
4839                                 mddev_put(mddev);
4840                         printk(KERN_ERR
4841                                 "md: cannot allocate memory for md drive.\n");
4842                         break;
4843                 }
4844                 if (mddev_lock(mddev))
4845                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4846                                mdname(mddev));
4847                 else if (mddev->raid_disks || mddev->major_version
4848                          || !list_empty(&mddev->disks)) {
4849                         printk(KERN_WARNING
4850                                 "md: %s already running, cannot run %s\n",
4851                                 mdname(mddev), bdevname(rdev0->bdev,b));
4852                         mddev_unlock(mddev);
4853                 } else {
4854                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4855                         mddev->persistent = 1;
4856                         rdev_for_each_list(rdev, tmp, &candidates) {
4857                                 list_del_init(&rdev->same_set);
4858                                 if (bind_rdev_to_array(rdev, mddev))
4859                                         export_rdev(rdev);
4860                         }
4861                         autorun_array(mddev);
4862                         mddev_unlock(mddev);
4863                 }
4864                 /* on success, candidates will be empty, on error
4865                  * it won't...
4866                  */
4867                 rdev_for_each_list(rdev, tmp, &candidates) {
4868                         list_del_init(&rdev->same_set);
4869                         export_rdev(rdev);
4870                 }
4871                 mddev_put(mddev);
4872         }
4873         printk(KERN_INFO "md: ... autorun DONE.\n");
4874 }
4875 #endif /* !MODULE */
4876
4877 static int get_version(void __user * arg)
4878 {
4879         mdu_version_t ver;
4880
4881         ver.major = MD_MAJOR_VERSION;
4882         ver.minor = MD_MINOR_VERSION;
4883         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4884
4885         if (copy_to_user(arg, &ver, sizeof(ver)))
4886                 return -EFAULT;
4887
4888         return 0;
4889 }
4890
4891 static int get_array_info(mddev_t * mddev, void __user * arg)
4892 {
4893         mdu_array_info_t info;
4894         int nr,working,insync,failed,spare;
4895         mdk_rdev_t *rdev;
4896
4897         nr=working=insync=failed=spare=0;
4898         list_for_each_entry(rdev, &mddev->disks, same_set) {
4899                 nr++;
4900                 if (test_bit(Faulty, &rdev->flags))
4901                         failed++;
4902                 else {
4903                         working++;
4904                         if (test_bit(In_sync, &rdev->flags))
4905                                 insync++;
4906                         else
4907                                 spare++;
4908                 }
4909         }
4910
4911         info.major_version = mddev->major_version;
4912         info.minor_version = mddev->minor_version;
4913         info.patch_version = MD_PATCHLEVEL_VERSION;
4914         info.ctime         = mddev->ctime;
4915         info.level         = mddev->level;
4916         info.size          = mddev->dev_sectors / 2;
4917         if (info.size != mddev->dev_sectors / 2) /* overflow */
4918                 info.size = -1;
4919         info.nr_disks      = nr;
4920         info.raid_disks    = mddev->raid_disks;
4921         info.md_minor      = mddev->md_minor;
4922         info.not_persistent= !mddev->persistent;
4923
4924         info.utime         = mddev->utime;
4925         info.state         = 0;
4926         if (mddev->in_sync)
4927                 info.state = (1<<MD_SB_CLEAN);
4928         if (mddev->bitmap && mddev->bitmap_info.offset)
4929                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4930         info.active_disks  = insync;
4931         info.working_disks = working;
4932         info.failed_disks  = failed;
4933         info.spare_disks   = spare;
4934
4935         info.layout        = mddev->layout;
4936         info.chunk_size    = mddev->chunk_sectors << 9;
4937
4938         if (copy_to_user(arg, &info, sizeof(info)))
4939                 return -EFAULT;
4940
4941         return 0;
4942 }
4943
4944 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4945 {
4946         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4947         char *ptr, *buf = NULL;
4948         int err = -ENOMEM;
4949
4950         if (md_allow_write(mddev))
4951                 file = kmalloc(sizeof(*file), GFP_NOIO);
4952         else
4953                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4954
4955         if (!file)
4956                 goto out;
4957
4958         /* bitmap disabled, zero the first byte and copy out */
4959         if (!mddev->bitmap || !mddev->bitmap->file) {
4960                 file->pathname[0] = '\0';
4961                 goto copy_out;
4962         }
4963
4964         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4965         if (!buf)
4966                 goto out;
4967
4968         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4969         if (IS_ERR(ptr))
4970                 goto out;
4971
4972         strcpy(file->pathname, ptr);
4973
4974 copy_out:
4975         err = 0;
4976         if (copy_to_user(arg, file, sizeof(*file)))
4977                 err = -EFAULT;
4978 out:
4979         kfree(buf);
4980         kfree(file);
4981         return err;
4982 }
4983
4984 static int get_disk_info(mddev_t * mddev, void __user * arg)
4985 {
4986         mdu_disk_info_t info;
4987         mdk_rdev_t *rdev;
4988
4989         if (copy_from_user(&info, arg, sizeof(info)))
4990                 return -EFAULT;
4991
4992         rdev = find_rdev_nr(mddev, info.number);
4993         if (rdev) {
4994                 info.major = MAJOR(rdev->bdev->bd_dev);
4995                 info.minor = MINOR(rdev->bdev->bd_dev);
4996                 info.raid_disk = rdev->raid_disk;
4997                 info.state = 0;
4998                 if (test_bit(Faulty, &rdev->flags))
4999                         info.state |= (1<<MD_DISK_FAULTY);
5000                 else if (test_bit(In_sync, &rdev->flags)) {
5001                         info.state |= (1<<MD_DISK_ACTIVE);
5002                         info.state |= (1<<MD_DISK_SYNC);
5003                 }
5004                 if (test_bit(WriteMostly, &rdev->flags))
5005                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5006         } else {
5007                 info.major = info.minor = 0;
5008                 info.raid_disk = -1;
5009                 info.state = (1<<MD_DISK_REMOVED);
5010         }
5011
5012         if (copy_to_user(arg, &info, sizeof(info)))
5013                 return -EFAULT;
5014
5015         return 0;
5016 }
5017
5018 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5019 {
5020         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5021         mdk_rdev_t *rdev;
5022         dev_t dev = MKDEV(info->major,info->minor);
5023
5024         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5025                 return -EOVERFLOW;
5026
5027         if (!mddev->raid_disks) {
5028                 int err;
5029                 /* expecting a device which has a superblock */
5030                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5031                 if (IS_ERR(rdev)) {
5032                         printk(KERN_WARNING
5033                                 "md: md_import_device returned %ld\n",
5034                                 PTR_ERR(rdev));
5035                         return PTR_ERR(rdev);
5036                 }
5037                 if (!list_empty(&mddev->disks)) {
5038                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5039                                                         mdk_rdev_t, same_set);
5040                         err = super_types[mddev->major_version]
5041                                 .load_super(rdev, rdev0, mddev->minor_version);
5042                         if (err < 0) {
5043                                 printk(KERN_WARNING
5044                                         "md: %s has different UUID to %s\n",
5045                                         bdevname(rdev->bdev,b),
5046                                         bdevname(rdev0->bdev,b2));
5047                                 export_rdev(rdev);
5048                                 return -EINVAL;
5049                         }
5050                 }
5051                 err = bind_rdev_to_array(rdev, mddev);
5052                 if (err)
5053                         export_rdev(rdev);
5054                 return err;
5055         }
5056
5057         /*
5058          * add_new_disk can be used once the array is assembled
5059          * to add "hot spares".  They must already have a superblock
5060          * written
5061          */
5062         if (mddev->pers) {
5063                 int err;
5064                 if (!mddev->pers->hot_add_disk) {
5065                         printk(KERN_WARNING
5066                                 "%s: personality does not support diskops!\n",
5067                                mdname(mddev));
5068                         return -EINVAL;
5069                 }
5070                 if (mddev->persistent)
5071                         rdev = md_import_device(dev, mddev->major_version,
5072                                                 mddev->minor_version);
5073                 else
5074                         rdev = md_import_device(dev, -1, -1);
5075                 if (IS_ERR(rdev)) {
5076                         printk(KERN_WARNING
5077                                 "md: md_import_device returned %ld\n",
5078                                 PTR_ERR(rdev));
5079                         return PTR_ERR(rdev);
5080                 }
5081                 /* set save_raid_disk if appropriate */
5082                 if (!mddev->persistent) {
5083                         if (info->state & (1<<MD_DISK_SYNC)  &&
5084                             info->raid_disk < mddev->raid_disks)
5085                                 rdev->raid_disk = info->raid_disk;
5086                         else
5087                                 rdev->raid_disk = -1;
5088                 } else
5089                         super_types[mddev->major_version].
5090                                 validate_super(mddev, rdev);
5091                 rdev->saved_raid_disk = rdev->raid_disk;
5092
5093                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5094                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5095                         set_bit(WriteMostly, &rdev->flags);
5096                 else
5097                         clear_bit(WriteMostly, &rdev->flags);
5098
5099                 rdev->raid_disk = -1;
5100                 err = bind_rdev_to_array(rdev, mddev);
5101                 if (!err && !mddev->pers->hot_remove_disk) {
5102                         /* If there is hot_add_disk but no hot_remove_disk
5103                          * then added disks for geometry changes,
5104                          * and should be added immediately.
5105                          */
5106                         super_types[mddev->major_version].
5107                                 validate_super(mddev, rdev);
5108                         err = mddev->pers->hot_add_disk(mddev, rdev);
5109                         if (err)
5110                                 unbind_rdev_from_array(rdev);
5111                 }
5112                 if (err)
5113                         export_rdev(rdev);
5114                 else
5115                         sysfs_notify_dirent(rdev->sysfs_state);
5116
5117                 md_update_sb(mddev, 1);
5118                 if (mddev->degraded)
5119                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5120                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5121                 md_wakeup_thread(mddev->thread);
5122                 return err;
5123         }
5124
5125         /* otherwise, add_new_disk is only allowed
5126          * for major_version==0 superblocks
5127          */
5128         if (mddev->major_version != 0) {
5129                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5130                        mdname(mddev));
5131                 return -EINVAL;
5132         }
5133
5134         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5135                 int err;
5136                 rdev = md_import_device(dev, -1, 0);
5137                 if (IS_ERR(rdev)) {
5138                         printk(KERN_WARNING
5139                                 "md: error, md_import_device() returned %ld\n",
5140                                 PTR_ERR(rdev));
5141                         return PTR_ERR(rdev);
5142                 }
5143                 rdev->desc_nr = info->number;
5144                 if (info->raid_disk < mddev->raid_disks)
5145                         rdev->raid_disk = info->raid_disk;
5146                 else
5147                         rdev->raid_disk = -1;
5148
5149                 if (rdev->raid_disk < mddev->raid_disks)
5150                         if (info->state & (1<<MD_DISK_SYNC))
5151                                 set_bit(In_sync, &rdev->flags);
5152
5153                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5154                         set_bit(WriteMostly, &rdev->flags);
5155
5156                 if (!mddev->persistent) {
5157                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5158                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5159                 } else
5160                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5161                 rdev->sectors = rdev->sb_start;
5162
5163                 err = bind_rdev_to_array(rdev, mddev);
5164                 if (err) {
5165                         export_rdev(rdev);
5166                         return err;
5167                 }
5168         }
5169
5170         return 0;
5171 }
5172
5173 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5174 {
5175         char b[BDEVNAME_SIZE];
5176         mdk_rdev_t *rdev;
5177
5178         rdev = find_rdev(mddev, dev);
5179         if (!rdev)
5180                 return -ENXIO;
5181
5182         if (rdev->raid_disk >= 0)
5183                 goto busy;
5184
5185         kick_rdev_from_array(rdev);
5186         md_update_sb(mddev, 1);
5187         md_new_event(mddev);
5188
5189         return 0;
5190 busy:
5191         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5192                 bdevname(rdev->bdev,b), mdname(mddev));
5193         return -EBUSY;
5194 }
5195
5196 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5197 {
5198         char b[BDEVNAME_SIZE];
5199         int err;
5200         mdk_rdev_t *rdev;
5201
5202         if (!mddev->pers)
5203                 return -ENODEV;
5204
5205         if (mddev->major_version != 0) {
5206                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5207                         " version-0 superblocks.\n",
5208                         mdname(mddev));
5209                 return -EINVAL;
5210         }
5211         if (!mddev->pers->hot_add_disk) {
5212                 printk(KERN_WARNING
5213                         "%s: personality does not support diskops!\n",
5214                         mdname(mddev));
5215                 return -EINVAL;
5216         }
5217
5218         rdev = md_import_device(dev, -1, 0);
5219         if (IS_ERR(rdev)) {
5220                 printk(KERN_WARNING
5221                         "md: error, md_import_device() returned %ld\n",
5222                         PTR_ERR(rdev));
5223                 return -EINVAL;
5224         }
5225
5226         if (mddev->persistent)
5227                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5228         else
5229                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5230
5231         rdev->sectors = rdev->sb_start;
5232
5233         if (test_bit(Faulty, &rdev->flags)) {
5234                 printk(KERN_WARNING
5235                         "md: can not hot-add faulty %s disk to %s!\n",
5236                         bdevname(rdev->bdev,b), mdname(mddev));
5237                 err = -EINVAL;
5238                 goto abort_export;
5239         }
5240         clear_bit(In_sync, &rdev->flags);
5241         rdev->desc_nr = -1;
5242         rdev->saved_raid_disk = -1;
5243         err = bind_rdev_to_array(rdev, mddev);
5244         if (err)
5245                 goto abort_export;
5246
5247         /*
5248          * The rest should better be atomic, we can have disk failures
5249          * noticed in interrupt contexts ...
5250          */
5251
5252         rdev->raid_disk = -1;
5253
5254         md_update_sb(mddev, 1);
5255
5256         /*
5257          * Kick recovery, maybe this spare has to be added to the
5258          * array immediately.
5259          */
5260         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5261         md_wakeup_thread(mddev->thread);
5262         md_new_event(mddev);
5263         return 0;
5264
5265 abort_export:
5266         export_rdev(rdev);
5267         return err;
5268 }
5269
5270 static int set_bitmap_file(mddev_t *mddev, int fd)
5271 {
5272         int err;
5273
5274         if (mddev->pers) {
5275                 if (!mddev->pers->quiesce)
5276                         return -EBUSY;
5277                 if (mddev->recovery || mddev->sync_thread)
5278                         return -EBUSY;
5279                 /* we should be able to change the bitmap.. */
5280         }
5281
5282
5283         if (fd >= 0) {
5284                 if (mddev->bitmap)
5285                         return -EEXIST; /* cannot add when bitmap is present */
5286                 mddev->bitmap_info.file = fget(fd);
5287
5288                 if (mddev->bitmap_info.file == NULL) {
5289                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5290                                mdname(mddev));
5291                         return -EBADF;
5292                 }
5293
5294                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5295                 if (err) {
5296                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5297                                mdname(mddev));
5298                         fput(mddev->bitmap_info.file);
5299                         mddev->bitmap_info.file = NULL;
5300                         return err;
5301                 }
5302                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5303         } else if (mddev->bitmap == NULL)
5304                 return -ENOENT; /* cannot remove what isn't there */
5305         err = 0;
5306         if (mddev->pers) {
5307                 mddev->pers->quiesce(mddev, 1);
5308                 if (fd >= 0)
5309                         err = bitmap_create(mddev);
5310                 if (fd < 0 || err) {
5311                         bitmap_destroy(mddev);
5312                         fd = -1; /* make sure to put the file */
5313                 }
5314                 mddev->pers->quiesce(mddev, 0);
5315         }
5316         if (fd < 0) {
5317                 if (mddev->bitmap_info.file) {
5318                         restore_bitmap_write_access(mddev->bitmap_info.file);
5319                         fput(mddev->bitmap_info.file);
5320                 }
5321                 mddev->bitmap_info.file = NULL;
5322         }
5323
5324         return err;
5325 }
5326
5327 /*
5328  * set_array_info is used two different ways
5329  * The original usage is when creating a new array.
5330  * In this usage, raid_disks is > 0 and it together with
5331  *  level, size, not_persistent,layout,chunksize determine the
5332  *  shape of the array.
5333  *  This will always create an array with a type-0.90.0 superblock.
5334  * The newer usage is when assembling an array.
5335  *  In this case raid_disks will be 0, and the major_version field is
5336  *  use to determine which style super-blocks are to be found on the devices.
5337  *  The minor and patch _version numbers are also kept incase the
5338  *  super_block handler wishes to interpret them.
5339  */
5340 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5341 {
5342
5343         if (info->raid_disks == 0) {
5344                 /* just setting version number for superblock loading */
5345                 if (info->major_version < 0 ||
5346                     info->major_version >= ARRAY_SIZE(super_types) ||
5347                     super_types[info->major_version].name == NULL) {
5348                         /* maybe try to auto-load a module? */
5349                         printk(KERN_INFO
5350                                 "md: superblock version %d not known\n",
5351                                 info->major_version);
5352                         return -EINVAL;
5353                 }
5354                 mddev->major_version = info->major_version;
5355                 mddev->minor_version = info->minor_version;
5356                 mddev->patch_version = info->patch_version;
5357                 mddev->persistent = !info->not_persistent;
5358                 /* ensure mddev_put doesn't delete this now that there
5359                  * is some minimal configuration.
5360                  */
5361                 mddev->ctime         = get_seconds();
5362                 return 0;
5363         }
5364         mddev->major_version = MD_MAJOR_VERSION;
5365         mddev->minor_version = MD_MINOR_VERSION;
5366         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5367         mddev->ctime         = get_seconds();
5368
5369         mddev->level         = info->level;
5370         mddev->clevel[0]     = 0;
5371         mddev->dev_sectors   = 2 * (sector_t)info->size;
5372         mddev->raid_disks    = info->raid_disks;
5373         /* don't set md_minor, it is determined by which /dev/md* was
5374          * openned
5375          */
5376         if (info->state & (1<<MD_SB_CLEAN))
5377                 mddev->recovery_cp = MaxSector;
5378         else
5379                 mddev->recovery_cp = 0;
5380         mddev->persistent    = ! info->not_persistent;
5381         mddev->external      = 0;
5382
5383         mddev->layout        = info->layout;
5384         mddev->chunk_sectors = info->chunk_size >> 9;
5385
5386         mddev->max_disks     = MD_SB_DISKS;
5387
5388         if (mddev->persistent)
5389                 mddev->flags         = 0;
5390         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5391
5392         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5393         mddev->bitmap_info.offset = 0;
5394
5395         mddev->reshape_position = MaxSector;
5396
5397         /*
5398          * Generate a 128 bit UUID
5399          */
5400         get_random_bytes(mddev->uuid, 16);
5401
5402         mddev->new_level = mddev->level;
5403         mddev->new_chunk_sectors = mddev->chunk_sectors;
5404         mddev->new_layout = mddev->layout;
5405         mddev->delta_disks = 0;
5406
5407         return 0;
5408 }
5409
5410 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5411 {
5412         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5413
5414         if (mddev->external_size)
5415                 return;
5416
5417         mddev->array_sectors = array_sectors;
5418 }
5419 EXPORT_SYMBOL(md_set_array_sectors);
5420
5421 static int update_size(mddev_t *mddev, sector_t num_sectors)
5422 {
5423         mdk_rdev_t *rdev;
5424         int rv;
5425         int fit = (num_sectors == 0);
5426
5427         if (mddev->pers->resize == NULL)
5428                 return -EINVAL;
5429         /* The "num_sectors" is the number of sectors of each device that
5430          * is used.  This can only make sense for arrays with redundancy.
5431          * linear and raid0 always use whatever space is available. We can only
5432          * consider changing this number if no resync or reconstruction is
5433          * happening, and if the new size is acceptable. It must fit before the
5434          * sb_start or, if that is <data_offset, it must fit before the size
5435          * of each device.  If num_sectors is zero, we find the largest size
5436          * that fits.
5437
5438          */
5439         if (mddev->sync_thread)
5440                 return -EBUSY;
5441         if (mddev->bitmap)
5442                 /* Sorry, cannot grow a bitmap yet, just remove it,
5443                  * grow, and re-add.
5444                  */
5445                 return -EBUSY;
5446         list_for_each_entry(rdev, &mddev->disks, same_set) {
5447                 sector_t avail = rdev->sectors;
5448
5449                 if (fit && (num_sectors == 0 || num_sectors > avail))
5450                         num_sectors = avail;
5451                 if (avail < num_sectors)
5452                         return -ENOSPC;
5453         }
5454         rv = mddev->pers->resize(mddev, num_sectors);
5455         if (!rv)
5456                 revalidate_disk(mddev->gendisk);
5457         return rv;
5458 }
5459
5460 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5461 {
5462         int rv;
5463         /* change the number of raid disks */
5464         if (mddev->pers->check_reshape == NULL)
5465                 return -EINVAL;
5466         if (raid_disks <= 0 ||
5467             (mddev->max_disks && raid_disks >= mddev->max_disks))
5468                 return -EINVAL;
5469         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5470                 return -EBUSY;
5471         mddev->delta_disks = raid_disks - mddev->raid_disks;
5472
5473         rv = mddev->pers->check_reshape(mddev);
5474         return rv;
5475 }
5476
5477
5478 /*
5479  * update_array_info is used to change the configuration of an
5480  * on-line array.
5481  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5482  * fields in the info are checked against the array.
5483  * Any differences that cannot be handled will cause an error.
5484  * Normally, only one change can be managed at a time.
5485  */
5486 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5487 {
5488         int rv = 0;
5489         int cnt = 0;
5490         int state = 0;
5491
5492         /* calculate expected state,ignoring low bits */
5493         if (mddev->bitmap && mddev->bitmap_info.offset)
5494                 state |= (1 << MD_SB_BITMAP_PRESENT);
5495
5496         if (mddev->major_version != info->major_version ||
5497             mddev->minor_version != info->minor_version ||
5498 /*          mddev->patch_version != info->patch_version || */
5499             mddev->ctime         != info->ctime         ||
5500             mddev->level         != info->level         ||
5501 /*          mddev->layout        != info->layout        || */
5502             !mddev->persistent   != info->not_persistent||
5503             mddev->chunk_sectors != info->chunk_size >> 9 ||
5504             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5505             ((state^info->state) & 0xfffffe00)
5506                 )
5507                 return -EINVAL;
5508         /* Check there is only one change */
5509         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5510                 cnt++;
5511         if (mddev->raid_disks != info->raid_disks)
5512                 cnt++;
5513         if (mddev->layout != info->layout)
5514                 cnt++;
5515         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5516                 cnt++;
5517         if (cnt == 0)
5518                 return 0;
5519         if (cnt > 1)
5520                 return -EINVAL;
5521
5522         if (mddev->layout != info->layout) {
5523                 /* Change layout
5524                  * we don't need to do anything at the md level, the
5525                  * personality will take care of it all.
5526                  */
5527                 if (mddev->pers->check_reshape == NULL)
5528                         return -EINVAL;
5529                 else {
5530                         mddev->new_layout = info->layout;
5531                         rv = mddev->pers->check_reshape(mddev);
5532                         if (rv)
5533                                 mddev->new_layout = mddev->layout;
5534                         return rv;
5535                 }
5536         }
5537         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5538                 rv = update_size(mddev, (sector_t)info->size * 2);
5539
5540         if (mddev->raid_disks    != info->raid_disks)
5541                 rv = update_raid_disks(mddev, info->raid_disks);
5542
5543         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5544                 if (mddev->pers->quiesce == NULL)
5545                         return -EINVAL;
5546                 if (mddev->recovery || mddev->sync_thread)
5547                         return -EBUSY;
5548                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5549                         /* add the bitmap */
5550                         if (mddev->bitmap)
5551                                 return -EEXIST;
5552                         if (mddev->bitmap_info.default_offset == 0)
5553                                 return -EINVAL;
5554                         mddev->bitmap_info.offset =
5555                                 mddev->bitmap_info.default_offset;
5556                         mddev->pers->quiesce(mddev, 1);
5557                         rv = bitmap_create(mddev);
5558                         if (rv)
5559                                 bitmap_destroy(mddev);
5560                         mddev->pers->quiesce(mddev, 0);
5561                 } else {
5562                         /* remove the bitmap */
5563                         if (!mddev->bitmap)
5564                                 return -ENOENT;
5565                         if (mddev->bitmap->file)
5566                                 return -EINVAL;
5567                         mddev->pers->quiesce(mddev, 1);
5568                         bitmap_destroy(mddev);
5569                         mddev->pers->quiesce(mddev, 0);
5570                         mddev->bitmap_info.offset = 0;
5571                 }
5572         }
5573         md_update_sb(mddev, 1);
5574         return rv;
5575 }
5576
5577 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5578 {
5579         mdk_rdev_t *rdev;
5580
5581         if (mddev->pers == NULL)
5582                 return -ENODEV;
5583
5584         rdev = find_rdev(mddev, dev);
5585         if (!rdev)
5586                 return -ENODEV;
5587
5588         md_error(mddev, rdev);
5589         return 0;
5590 }
5591
5592 /*
5593  * We have a problem here : there is no easy way to give a CHS
5594  * virtual geometry. We currently pretend that we have a 2 heads
5595  * 4 sectors (with a BIG number of cylinders...). This drives
5596  * dosfs just mad... ;-)
5597  */
5598 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5599 {
5600         mddev_t *mddev = bdev->bd_disk->private_data;
5601
5602         geo->heads = 2;
5603         geo->sectors = 4;
5604         geo->cylinders = mddev->array_sectors / 8;
5605         return 0;
5606 }
5607
5608 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5609                         unsigned int cmd, unsigned long arg)
5610 {
5611         int err = 0;
5612         void __user *argp = (void __user *)arg;
5613         mddev_t *mddev = NULL;
5614         int ro;
5615
5616         if (!capable(CAP_SYS_ADMIN))
5617                 return -EACCES;
5618
5619         /*
5620          * Commands dealing with the RAID driver but not any
5621          * particular array:
5622          */
5623         switch (cmd)
5624         {
5625                 case RAID_VERSION:
5626                         err = get_version(argp);
5627                         goto done;
5628
5629                 case PRINT_RAID_DEBUG:
5630                         err = 0;
5631                         md_print_devices();
5632                         goto done;
5633
5634 #ifndef MODULE
5635                 case RAID_AUTORUN:
5636                         err = 0;
5637                         autostart_arrays(arg);
5638                         goto done;
5639 #endif
5640                 default:;
5641         }
5642
5643         /*
5644          * Commands creating/starting a new array:
5645          */
5646
5647         mddev = bdev->bd_disk->private_data;
5648
5649         if (!mddev) {
5650                 BUG();
5651                 goto abort;
5652         }
5653
5654         err = mddev_lock(mddev);
5655         if (err) {
5656                 printk(KERN_INFO
5657                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5658                         err, cmd);
5659                 goto abort;
5660         }
5661
5662         switch (cmd)
5663         {
5664                 case SET_ARRAY_INFO:
5665                         {
5666                                 mdu_array_info_t info;
5667                                 if (!arg)
5668                                         memset(&info, 0, sizeof(info));
5669                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5670                                         err = -EFAULT;
5671                                         goto abort_unlock;
5672                                 }
5673                                 if (mddev->pers) {
5674                                         err = update_array_info(mddev, &info);
5675                                         if (err) {
5676                                                 printk(KERN_WARNING "md: couldn't update"
5677                                                        " array info. %d\n", err);
5678                                                 goto abort_unlock;
5679                                         }
5680                                         goto done_unlock;
5681                                 }
5682                                 if (!list_empty(&mddev->disks)) {
5683                                         printk(KERN_WARNING
5684                                                "md: array %s already has disks!\n",
5685                                                mdname(mddev));
5686                                         err = -EBUSY;
5687                                         goto abort_unlock;
5688                                 }
5689                                 if (mddev->raid_disks) {
5690                                         printk(KERN_WARNING
5691                                                "md: array %s already initialised!\n",
5692                                                mdname(mddev));
5693                                         err = -EBUSY;
5694                                         goto abort_unlock;
5695                                 }
5696                                 err = set_array_info(mddev, &info);
5697                                 if (err) {
5698                                         printk(KERN_WARNING "md: couldn't set"
5699                                                " array info. %d\n", err);
5700                                         goto abort_unlock;
5701                                 }
5702                         }
5703                         goto done_unlock;
5704
5705                 default:;
5706         }
5707
5708         /*
5709          * Commands querying/configuring an existing array:
5710          */
5711         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5712          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5713         if ((!mddev->raid_disks && !mddev->external)
5714             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5715             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5716             && cmd != GET_BITMAP_FILE) {
5717                 err = -ENODEV;
5718                 goto abort_unlock;
5719         }
5720
5721         /*
5722          * Commands even a read-only array can execute:
5723          */
5724         switch (cmd)
5725         {
5726                 case GET_ARRAY_INFO:
5727                         err = get_array_info(mddev, argp);
5728                         goto done_unlock;
5729
5730                 case GET_BITMAP_FILE:
5731                         err = get_bitmap_file(mddev, argp);
5732                         goto done_unlock;
5733
5734                 case GET_DISK_INFO:
5735                         err = get_disk_info(mddev, argp);
5736                         goto done_unlock;
5737
5738                 case RESTART_ARRAY_RW:
5739                         err = restart_array(mddev);
5740                         goto done_unlock;
5741
5742                 case STOP_ARRAY:
5743                         err = do_md_stop(mddev, 0, 1);
5744                         goto done_unlock;
5745
5746                 case STOP_ARRAY_RO:
5747                         err = md_set_readonly(mddev, 1);
5748                         goto done_unlock;
5749
5750                 case BLKROSET:
5751                         if (get_user(ro, (int __user *)(arg))) {
5752                                 err = -EFAULT;
5753                                 goto done_unlock;
5754                         }
5755                         err = -EINVAL;
5756
5757                         /* if the bdev is going readonly the value of mddev->ro
5758                          * does not matter, no writes are coming
5759                          */
5760                         if (ro)
5761                                 goto done_unlock;
5762
5763                         /* are we are already prepared for writes? */
5764                         if (mddev->ro != 1)
5765                                 goto done_unlock;
5766
5767                         /* transitioning to readauto need only happen for
5768                          * arrays that call md_write_start
5769                          */
5770                         if (mddev->pers) {
5771                                 err = restart_array(mddev);
5772                                 if (err == 0) {
5773                                         mddev->ro = 2;
5774                                         set_disk_ro(mddev->gendisk, 0);
5775                                 }
5776                         }
5777                         goto done_unlock;
5778         }
5779
5780         /*
5781          * The remaining ioctls are changing the state of the
5782          * superblock, so we do not allow them on read-only arrays.
5783          * However non-MD ioctls (e.g. get-size) will still come through
5784          * here and hit the 'default' below, so only disallow
5785          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5786          */
5787         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5788                 if (mddev->ro == 2) {
5789                         mddev->ro = 0;
5790                         sysfs_notify_dirent(mddev->sysfs_state);
5791                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5792                         md_wakeup_thread(mddev->thread);
5793                 } else {
5794                         err = -EROFS;
5795                         goto abort_unlock;
5796                 }
5797         }
5798
5799         switch (cmd)
5800         {
5801                 case ADD_NEW_DISK:
5802                 {
5803                         mdu_disk_info_t info;
5804                         if (copy_from_user(&info, argp, sizeof(info)))
5805                                 err = -EFAULT;
5806                         else
5807                                 err = add_new_disk(mddev, &info);
5808                         goto done_unlock;
5809                 }
5810
5811                 case HOT_REMOVE_DISK:
5812                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5813                         goto done_unlock;
5814
5815                 case HOT_ADD_DISK:
5816                         err = hot_add_disk(mddev, new_decode_dev(arg));
5817                         goto done_unlock;
5818
5819                 case SET_DISK_FAULTY:
5820                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5821                         goto done_unlock;
5822
5823                 case RUN_ARRAY:
5824                         err = do_md_run(mddev);
5825                         goto done_unlock;
5826
5827                 case SET_BITMAP_FILE:
5828                         err = set_bitmap_file(mddev, (int)arg);
5829                         goto done_unlock;
5830
5831                 default:
5832                         err = -EINVAL;
5833                         goto abort_unlock;
5834         }
5835
5836 done_unlock:
5837 abort_unlock:
5838         if (mddev->hold_active == UNTIL_IOCTL &&
5839             err != -EINVAL)
5840                 mddev->hold_active = 0;
5841         mddev_unlock(mddev);
5842
5843         return err;
5844 done:
5845         if (err)
5846                 MD_BUG();
5847 abort:
5848         return err;
5849 }
5850 #ifdef CONFIG_COMPAT
5851 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5852                     unsigned int cmd, unsigned long arg)
5853 {
5854         switch (cmd) {
5855         case HOT_REMOVE_DISK:
5856         case HOT_ADD_DISK:
5857         case SET_DISK_FAULTY:
5858         case SET_BITMAP_FILE:
5859                 /* These take in integer arg, do not convert */
5860                 break;
5861         default:
5862                 arg = (unsigned long)compat_ptr(arg);
5863                 break;
5864         }
5865
5866         return md_ioctl(bdev, mode, cmd, arg);
5867 }
5868 #endif /* CONFIG_COMPAT */
5869
5870 static int md_open(struct block_device *bdev, fmode_t mode)
5871 {
5872         /*
5873          * Succeed if we can lock the mddev, which confirms that
5874          * it isn't being stopped right now.
5875          */
5876         mddev_t *mddev = mddev_find(bdev->bd_dev);
5877         int err;
5878
5879         if (mddev->gendisk != bdev->bd_disk) {
5880                 /* we are racing with mddev_put which is discarding this
5881                  * bd_disk.
5882                  */
5883                 mddev_put(mddev);
5884                 /* Wait until bdev->bd_disk is definitely gone */
5885                 flush_scheduled_work();
5886                 /* Then retry the open from the top */
5887                 return -ERESTARTSYS;
5888         }
5889         BUG_ON(mddev != bdev->bd_disk->private_data);
5890
5891         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5892                 goto out;
5893
5894         err = 0;
5895         atomic_inc(&mddev->openers);
5896         mutex_unlock(&mddev->open_mutex);
5897
5898  out:
5899         return err;
5900 }
5901
5902 static int md_release(struct gendisk *disk, fmode_t mode)
5903 {
5904         mddev_t *mddev = disk->private_data;
5905
5906         BUG_ON(!mddev);
5907         atomic_dec(&mddev->openers);
5908         mddev_put(mddev);
5909
5910         return 0;
5911 }
5912 static const struct block_device_operations md_fops =
5913 {
5914         .owner          = THIS_MODULE,
5915         .open           = md_open,
5916         .release        = md_release,
5917         .ioctl          = md_ioctl,
5918 #ifdef CONFIG_COMPAT
5919         .compat_ioctl   = md_compat_ioctl,
5920 #endif
5921         .getgeo         = md_getgeo,
5922 };
5923
5924 static int md_thread(void * arg)
5925 {
5926         mdk_thread_t *thread = arg;
5927
5928         /*
5929          * md_thread is a 'system-thread', it's priority should be very
5930          * high. We avoid resource deadlocks individually in each
5931          * raid personality. (RAID5 does preallocation) We also use RR and
5932          * the very same RT priority as kswapd, thus we will never get
5933          * into a priority inversion deadlock.
5934          *
5935          * we definitely have to have equal or higher priority than
5936          * bdflush, otherwise bdflush will deadlock if there are too
5937          * many dirty RAID5 blocks.
5938          */
5939
5940         allow_signal(SIGKILL);
5941         while (!kthread_should_stop()) {
5942
5943                 /* We need to wait INTERRUPTIBLE so that
5944                  * we don't add to the load-average.
5945                  * That means we need to be sure no signals are
5946                  * pending
5947                  */
5948                 if (signal_pending(current))
5949                         flush_signals(current);
5950
5951                 wait_event_interruptible_timeout
5952                         (thread->wqueue,
5953                          test_bit(THREAD_WAKEUP, &thread->flags)
5954                          || kthread_should_stop(),
5955                          thread->timeout);
5956
5957                 clear_bit(THREAD_WAKEUP, &thread->flags);
5958
5959                 thread->run(thread->mddev);
5960         }
5961
5962         return 0;
5963 }
5964
5965 void md_wakeup_thread(mdk_thread_t *thread)
5966 {
5967         if (thread) {
5968                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5969                 set_bit(THREAD_WAKEUP, &thread->flags);
5970                 wake_up(&thread->wqueue);
5971         }
5972 }
5973
5974 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5975                                  const char *name)
5976 {
5977         mdk_thread_t *thread;
5978
5979         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5980         if (!thread)
5981                 return NULL;
5982
5983         init_waitqueue_head(&thread->wqueue);
5984
5985         thread->run = run;
5986         thread->mddev = mddev;
5987         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5988         thread->tsk = kthread_run(md_thread, thread,
5989                                   "%s_%s",
5990                                   mdname(thread->mddev),
5991                                   name ?: mddev->pers->name);
5992         if (IS_ERR(thread->tsk)) {
5993                 kfree(thread);
5994                 return NULL;
5995         }
5996         return thread;
5997 }
5998
5999 void md_unregister_thread(mdk_thread_t *thread)
6000 {
6001         if (!thread)
6002                 return;
6003         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6004
6005         kthread_stop(thread->tsk);
6006         kfree(thread);
6007 }
6008
6009 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6010 {
6011         if (!mddev) {
6012                 MD_BUG();
6013                 return;
6014         }
6015
6016         if (!rdev || test_bit(Faulty, &rdev->flags))
6017                 return;
6018
6019         if (mddev->external)
6020                 set_bit(Blocked, &rdev->flags);
6021 /*
6022         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6023                 mdname(mddev),
6024                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6025                 __builtin_return_address(0),__builtin_return_address(1),
6026                 __builtin_return_address(2),__builtin_return_address(3));
6027 */
6028         if (!mddev->pers)
6029                 return;
6030         if (!mddev->pers->error_handler)
6031                 return;
6032         mddev->pers->error_handler(mddev,rdev);
6033         if (mddev->degraded)
6034                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6035         sysfs_notify_dirent(rdev->sysfs_state);
6036         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6037         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6038         md_wakeup_thread(mddev->thread);
6039         md_new_event_inintr(mddev);
6040 }
6041
6042 /* seq_file implementation /proc/mdstat */
6043
6044 static void status_unused(struct seq_file *seq)
6045 {
6046         int i = 0;
6047         mdk_rdev_t *rdev;
6048
6049         seq_printf(seq, "unused devices: ");
6050
6051         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6052                 char b[BDEVNAME_SIZE];
6053                 i++;
6054                 seq_printf(seq, "%s ",
6055                               bdevname(rdev->bdev,b));
6056         }
6057         if (!i)
6058                 seq_printf(seq, "<none>");
6059
6060         seq_printf(seq, "\n");
6061 }
6062
6063
6064 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6065 {
6066         sector_t max_sectors, resync, res;
6067         unsigned long dt, db;
6068         sector_t rt;
6069         int scale;
6070         unsigned int per_milli;
6071
6072         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6073
6074         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6075                 max_sectors = mddev->resync_max_sectors;
6076         else
6077                 max_sectors = mddev->dev_sectors;
6078
6079         /*
6080          * Should not happen.
6081          */
6082         if (!max_sectors) {
6083                 MD_BUG();
6084                 return;
6085         }
6086         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6087          * in a sector_t, and (max_sectors>>scale) will fit in a
6088          * u32, as those are the requirements for sector_div.
6089          * Thus 'scale' must be at least 10
6090          */
6091         scale = 10;
6092         if (sizeof(sector_t) > sizeof(unsigned long)) {
6093                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6094                         scale++;
6095         }
6096         res = (resync>>scale)*1000;
6097         sector_div(res, (u32)((max_sectors>>scale)+1));
6098
6099         per_milli = res;
6100         {
6101                 int i, x = per_milli/50, y = 20-x;
6102                 seq_printf(seq, "[");
6103                 for (i = 0; i < x; i++)
6104                         seq_printf(seq, "=");
6105                 seq_printf(seq, ">");
6106                 for (i = 0; i < y; i++)
6107                         seq_printf(seq, ".");
6108                 seq_printf(seq, "] ");
6109         }
6110         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6111                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6112                     "reshape" :
6113                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6114                      "check" :
6115                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6116                       "resync" : "recovery"))),
6117                    per_milli/10, per_milli % 10,
6118                    (unsigned long long) resync/2,
6119                    (unsigned long long) max_sectors/2);
6120
6121         /*
6122          * dt: time from mark until now
6123          * db: blocks written from mark until now
6124          * rt: remaining time
6125          *
6126          * rt is a sector_t, so could be 32bit or 64bit.
6127          * So we divide before multiply in case it is 32bit and close
6128          * to the limit.
6129          * We scale the divisor (db) by 32 to avoid loosing precision
6130          * near the end of resync when the number of remaining sectors
6131          * is close to 'db'.
6132          * We then divide rt by 32 after multiplying by db to compensate.
6133          * The '+1' avoids division by zero if db is very small.
6134          */
6135         dt = ((jiffies - mddev->resync_mark) / HZ);
6136         if (!dt) dt++;
6137         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6138                 - mddev->resync_mark_cnt;
6139
6140         rt = max_sectors - resync;    /* number of remaining sectors */
6141         sector_div(rt, db/32+1);
6142         rt *= dt;
6143         rt >>= 5;
6144
6145         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6146                    ((unsigned long)rt % 60)/6);
6147
6148         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6149 }
6150
6151 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6152 {
6153         struct list_head *tmp;
6154         loff_t l = *pos;
6155         mddev_t *mddev;
6156
6157         if (l >= 0x10000)
6158                 return NULL;
6159         if (!l--)
6160                 /* header */
6161                 return (void*)1;
6162
6163         spin_lock(&all_mddevs_lock);
6164         list_for_each(tmp,&all_mddevs)
6165                 if (!l--) {
6166                         mddev = list_entry(tmp, mddev_t, all_mddevs);
6167                         mddev_get(mddev);
6168                         spin_unlock(&all_mddevs_lock);
6169                         return mddev;
6170                 }
6171         spin_unlock(&all_mddevs_lock);
6172         if (!l--)
6173                 return (void*)2;/* tail */
6174         return NULL;
6175 }
6176
6177 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6178 {
6179         struct list_head *tmp;
6180         mddev_t *next_mddev, *mddev = v;
6181
6182         ++*pos;
6183         if (v == (void*)2)
6184                 return NULL;
6185
6186         spin_lock(&all_mddevs_lock);
6187         if (v == (void*)1)
6188                 tmp = all_mddevs.next;
6189         else
6190                 tmp = mddev->all_mddevs.next;
6191         if (tmp != &all_mddevs)
6192                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6193         else {
6194                 next_mddev = (void*)2;
6195                 *pos = 0x10000;
6196         }
6197         spin_unlock(&all_mddevs_lock);
6198
6199         if (v != (void*)1)
6200                 mddev_put(mddev);
6201         return next_mddev;
6202
6203 }
6204
6205 static void md_seq_stop(struct seq_file *seq, void *v)
6206 {
6207         mddev_t *mddev = v;
6208
6209         if (mddev && v != (void*)1 && v != (void*)2)
6210                 mddev_put(mddev);
6211 }
6212
6213 struct mdstat_info {
6214         int event;
6215 };
6216
6217 static int md_seq_show(struct seq_file *seq, void *v)
6218 {
6219         mddev_t *mddev = v;
6220         sector_t sectors;
6221         mdk_rdev_t *rdev;
6222         struct mdstat_info *mi = seq->private;
6223         struct bitmap *bitmap;
6224
6225         if (v == (void*)1) {
6226                 struct mdk_personality *pers;
6227                 seq_printf(seq, "Personalities : ");
6228                 spin_lock(&pers_lock);
6229                 list_for_each_entry(pers, &pers_list, list)
6230                         seq_printf(seq, "[%s] ", pers->name);
6231
6232                 spin_unlock(&pers_lock);
6233                 seq_printf(seq, "\n");
6234                 mi->event = atomic_read(&md_event_count);
6235                 return 0;
6236         }
6237         if (v == (void*)2) {
6238                 status_unused(seq);
6239                 return 0;
6240         }
6241
6242         if (mddev_lock(mddev) < 0)
6243                 return -EINTR;
6244
6245         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6246                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6247                                                 mddev->pers ? "" : "in");
6248                 if (mddev->pers) {
6249                         if (mddev->ro==1)
6250                                 seq_printf(seq, " (read-only)");
6251                         if (mddev->ro==2)
6252                                 seq_printf(seq, " (auto-read-only)");
6253                         seq_printf(seq, " %s", mddev->pers->name);
6254                 }
6255
6256                 sectors = 0;
6257                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6258                         char b[BDEVNAME_SIZE];
6259                         seq_printf(seq, " %s[%d]",
6260                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6261                         if (test_bit(WriteMostly, &rdev->flags))
6262                                 seq_printf(seq, "(W)");
6263                         if (test_bit(Faulty, &rdev->flags)) {
6264                                 seq_printf(seq, "(F)");
6265                                 continue;
6266                         } else if (rdev->raid_disk < 0)
6267                                 seq_printf(seq, "(S)"); /* spare */
6268                         sectors += rdev->sectors;
6269                 }
6270
6271                 if (!list_empty(&mddev->disks)) {
6272                         if (mddev->pers)
6273                                 seq_printf(seq, "\n      %llu blocks",
6274                                            (unsigned long long)
6275                                            mddev->array_sectors / 2);
6276                         else
6277                                 seq_printf(seq, "\n      %llu blocks",
6278                                            (unsigned long long)sectors / 2);
6279                 }
6280                 if (mddev->persistent) {
6281                         if (mddev->major_version != 0 ||
6282                             mddev->minor_version != 90) {
6283                                 seq_printf(seq," super %d.%d",
6284                                            mddev->major_version,
6285                                            mddev->minor_version);
6286                         }
6287                 } else if (mddev->external)
6288                         seq_printf(seq, " super external:%s",
6289                                    mddev->metadata_type);
6290                 else
6291                         seq_printf(seq, " super non-persistent");
6292
6293                 if (mddev->pers) {
6294                         mddev->pers->status(seq, mddev);
6295                         seq_printf(seq, "\n      ");
6296                         if (mddev->pers->sync_request) {
6297                                 if (mddev->curr_resync > 2) {
6298                                         status_resync(seq, mddev);
6299                                         seq_printf(seq, "\n      ");
6300                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6301                                         seq_printf(seq, "\tresync=DELAYED\n      ");
6302                                 else if (mddev->recovery_cp < MaxSector)
6303                                         seq_printf(seq, "\tresync=PENDING\n      ");
6304                         }
6305                 } else
6306                         seq_printf(seq, "\n       ");
6307
6308                 if ((bitmap = mddev->bitmap)) {
6309                         unsigned long chunk_kb;
6310                         unsigned long flags;
6311                         spin_lock_irqsave(&bitmap->lock, flags);
6312                         chunk_kb = mddev->bitmap_info.chunksize >> 10;
6313                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6314                                 "%lu%s chunk",
6315                                 bitmap->pages - bitmap->missing_pages,
6316                                 bitmap->pages,
6317                                 (bitmap->pages - bitmap->missing_pages)
6318                                         << (PAGE_SHIFT - 10),
6319                                 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6320                                 chunk_kb ? "KB" : "B");
6321                         if (bitmap->file) {
6322                                 seq_printf(seq, ", file: ");
6323                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
6324                         }
6325
6326                         seq_printf(seq, "\n");
6327                         spin_unlock_irqrestore(&bitmap->lock, flags);
6328                 }
6329
6330                 seq_printf(seq, "\n");
6331         }
6332         mddev_unlock(mddev);
6333
6334         return 0;
6335 }
6336
6337 static const struct seq_operations md_seq_ops = {
6338         .start  = md_seq_start,
6339         .next   = md_seq_next,
6340         .stop   = md_seq_stop,
6341         .show   = md_seq_show,
6342 };
6343
6344 static int md_seq_open(struct inode *inode, struct file *file)
6345 {
6346         int error;
6347         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6348         if (mi == NULL)
6349                 return -ENOMEM;
6350
6351         error = seq_open(file, &md_seq_ops);
6352         if (error)
6353                 kfree(mi);
6354         else {
6355                 struct seq_file *p = file->private_data;
6356                 p->private = mi;
6357                 mi->event = atomic_read(&md_event_count);
6358         }
6359         return error;
6360 }
6361
6362 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6363 {
6364         struct seq_file *m = filp->private_data;
6365         struct mdstat_info *mi = m->private;
6366         int mask;
6367
6368         poll_wait(filp, &md_event_waiters, wait);
6369
6370         /* always allow read */
6371         mask = POLLIN | POLLRDNORM;
6372
6373         if (mi->event != atomic_read(&md_event_count))
6374                 mask |= POLLERR | POLLPRI;
6375         return mask;
6376 }
6377
6378 static const struct file_operations md_seq_fops = {
6379         .owner          = THIS_MODULE,
6380         .open           = md_seq_open,
6381         .read           = seq_read,
6382         .llseek         = seq_lseek,
6383         .release        = seq_release_private,
6384         .poll           = mdstat_poll,
6385 };
6386
6387 int register_md_personality(struct mdk_personality *p)
6388 {
6389         spin_lock(&pers_lock);
6390         list_add_tail(&p->list, &pers_list);
6391         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6392         spin_unlock(&pers_lock);
6393         return 0;
6394 }
6395
6396 int unregister_md_personality(struct mdk_personality *p)
6397 {
6398         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6399         spin_lock(&pers_lock);
6400         list_del_init(&p->list);
6401         spin_unlock(&pers_lock);
6402         return 0;
6403 }
6404
6405 static int is_mddev_idle(mddev_t *mddev, int init)
6406 {
6407         mdk_rdev_t * rdev;
6408         int idle;
6409         int curr_events;
6410
6411         idle = 1;
6412         rcu_read_lock();
6413         rdev_for_each_rcu(rdev, mddev) {
6414                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6415                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6416                               (int)part_stat_read(&disk->part0, sectors[1]) -
6417                               atomic_read(&disk->sync_io);
6418                 /* sync IO will cause sync_io to increase before the disk_stats
6419                  * as sync_io is counted when a request starts, and
6420                  * disk_stats is counted when it completes.
6421                  * So resync activity will cause curr_events to be smaller than
6422                  * when there was no such activity.
6423                  * non-sync IO will cause disk_stat to increase without
6424                  * increasing sync_io so curr_events will (eventually)
6425                  * be larger than it was before.  Once it becomes
6426                  * substantially larger, the test below will cause
6427                  * the array to appear non-idle, and resync will slow
6428                  * down.
6429                  * If there is a lot of outstanding resync activity when
6430                  * we set last_event to curr_events, then all that activity
6431                  * completing might cause the array to appear non-idle
6432                  * and resync will be slowed down even though there might
6433                  * not have been non-resync activity.  This will only
6434                  * happen once though.  'last_events' will soon reflect
6435                  * the state where there is little or no outstanding
6436                  * resync requests, and further resync activity will
6437                  * always make curr_events less than last_events.
6438                  *
6439                  */
6440                 if (init || curr_events - rdev->last_events > 64) {
6441                         rdev->last_events = curr_events;
6442                         idle = 0;
6443                 }
6444         }
6445         rcu_read_unlock();
6446         return idle;
6447 }
6448
6449 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6450 {
6451         /* another "blocks" (512byte) blocks have been synced */
6452         atomic_sub(blocks, &mddev->recovery_active);
6453         wake_up(&mddev->recovery_wait);
6454         if (!ok) {
6455                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6456                 md_wakeup_thread(mddev->thread);
6457                 // stop recovery, signal do_sync ....
6458         }
6459 }
6460
6461
6462 /* md_write_start(mddev, bi)
6463  * If we need to update some array metadata (e.g. 'active' flag
6464  * in superblock) before writing, schedule a superblock update
6465  * and wait for it to complete.
6466  */
6467 void md_write_start(mddev_t *mddev, struct bio *bi)
6468 {
6469         int did_change = 0;
6470         if (bio_data_dir(bi) != WRITE)
6471                 return;
6472
6473         BUG_ON(mddev->ro == 1);
6474         if (mddev->ro == 2) {
6475                 /* need to switch to read/write */
6476                 mddev->ro = 0;
6477                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6478                 md_wakeup_thread(mddev->thread);
6479                 md_wakeup_thread(mddev->sync_thread);
6480                 did_change = 1;
6481         }
6482         atomic_inc(&mddev->writes_pending);
6483         if (mddev->safemode == 1)
6484                 mddev->safemode = 0;
6485         if (mddev->in_sync) {
6486                 spin_lock_irq(&mddev->write_lock);
6487                 if (mddev->in_sync) {
6488                         mddev->in_sync = 0;
6489                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6490                         md_wakeup_thread(mddev->thread);
6491                         did_change = 1;
6492                 }
6493                 spin_unlock_irq(&mddev->write_lock);
6494         }
6495         if (did_change)
6496                 sysfs_notify_dirent(mddev->sysfs_state);
6497         wait_event(mddev->sb_wait,
6498                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6499                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6500 }
6501
6502 void md_write_end(mddev_t *mddev)
6503 {
6504         if (atomic_dec_and_test(&mddev->writes_pending)) {
6505                 if (mddev->safemode == 2)
6506                         md_wakeup_thread(mddev->thread);
6507                 else if (mddev->safemode_delay)
6508                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6509         }
6510 }
6511
6512 /* md_allow_write(mddev)
6513  * Calling this ensures that the array is marked 'active' so that writes
6514  * may proceed without blocking.  It is important to call this before
6515  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6516  * Must be called with mddev_lock held.
6517  *
6518  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6519  * is dropped, so return -EAGAIN after notifying userspace.
6520  */
6521 int md_allow_write(mddev_t *mddev)
6522 {
6523         if (!mddev->pers)
6524                 return 0;
6525         if (mddev->ro)
6526                 return 0;
6527         if (!mddev->pers->sync_request)
6528                 return 0;
6529
6530         spin_lock_irq(&mddev->write_lock);
6531         if (mddev->in_sync) {
6532                 mddev->in_sync = 0;
6533                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6534                 if (mddev->safemode_delay &&
6535                     mddev->safemode == 0)
6536                         mddev->safemode = 1;
6537                 spin_unlock_irq(&mddev->write_lock);
6538                 md_update_sb(mddev, 0);
6539                 sysfs_notify_dirent(mddev->sysfs_state);
6540         } else
6541                 spin_unlock_irq(&mddev->write_lock);
6542
6543         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6544                 return -EAGAIN;
6545         else
6546                 return 0;
6547 }
6548 EXPORT_SYMBOL_GPL(md_allow_write);
6549
6550 #define SYNC_MARKS      10
6551 #define SYNC_MARK_STEP  (3*HZ)
6552 void md_do_sync(mddev_t *mddev)
6553 {
6554         mddev_t *mddev2;
6555         unsigned int currspeed = 0,
6556                  window;
6557         sector_t max_sectors,j, io_sectors;
6558         unsigned long mark[SYNC_MARKS];
6559         sector_t mark_cnt[SYNC_MARKS];
6560         int last_mark,m;
6561         struct list_head *tmp;
6562         sector_t last_check;
6563         int skipped = 0;
6564         mdk_rdev_t *rdev;
6565         char *desc;
6566
6567         /* just incase thread restarts... */
6568         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6569                 return;
6570         if (mddev->ro) /* never try to sync a read-only array */
6571                 return;
6572
6573         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6574                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6575                         desc = "data-check";
6576                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6577                         desc = "requested-resync";
6578                 else
6579                         desc = "resync";
6580         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6581                 desc = "reshape";
6582         else
6583                 desc = "recovery";
6584
6585         /* we overload curr_resync somewhat here.
6586          * 0 == not engaged in resync at all
6587          * 2 == checking that there is no conflict with another sync
6588          * 1 == like 2, but have yielded to allow conflicting resync to
6589          *              commense
6590          * other == active in resync - this many blocks
6591          *
6592          * Before starting a resync we must have set curr_resync to
6593          * 2, and then checked that every "conflicting" array has curr_resync
6594          * less than ours.  When we find one that is the same or higher
6595          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6596          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6597          * This will mean we have to start checking from the beginning again.
6598          *
6599          */
6600
6601         do {
6602                 mddev->curr_resync = 2;
6603
6604         try_again:
6605                 if (kthread_should_stop())
6606                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6607
6608                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6609                         goto skip;
6610                 for_each_mddev(mddev2, tmp) {
6611                         if (mddev2 == mddev)
6612                                 continue;
6613                         if (!mddev->parallel_resync
6614                         &&  mddev2->curr_resync
6615                         &&  match_mddev_units(mddev, mddev2)) {
6616                                 DEFINE_WAIT(wq);
6617                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6618                                         /* arbitrarily yield */
6619                                         mddev->curr_resync = 1;
6620                                         wake_up(&resync_wait);
6621                                 }
6622                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6623                                         /* no need to wait here, we can wait the next
6624                                          * time 'round when curr_resync == 2
6625                                          */
6626                                         continue;
6627                                 /* We need to wait 'interruptible' so as not to
6628                                  * contribute to the load average, and not to
6629                                  * be caught by 'softlockup'
6630                                  */
6631                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6632                                 if (!kthread_should_stop() &&
6633                                     mddev2->curr_resync >= mddev->curr_resync) {
6634                                         printk(KERN_INFO "md: delaying %s of %s"
6635                                                " until %s has finished (they"
6636                                                " share one or more physical units)\n",
6637                                                desc, mdname(mddev), mdname(mddev2));
6638                                         mddev_put(mddev2);
6639                                         if (signal_pending(current))
6640                                                 flush_signals(current);
6641                                         schedule();
6642                                         finish_wait(&resync_wait, &wq);
6643                                         goto try_again;
6644                                 }
6645                                 finish_wait(&resync_wait, &wq);
6646                         }
6647                 }
6648         } while (mddev->curr_resync < 2);
6649
6650         j = 0;
6651         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6652                 /* resync follows the size requested by the personality,
6653                  * which defaults to physical size, but can be virtual size
6654                  */
6655                 max_sectors = mddev->resync_max_sectors;
6656                 mddev->resync_mismatches = 0;
6657                 /* we don't use the checkpoint if there's a bitmap */
6658                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6659                         j = mddev->resync_min;
6660                 else if (!mddev->bitmap)
6661                         j = mddev->recovery_cp;
6662
6663         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6664                 max_sectors = mddev->dev_sectors;
6665         else {
6666                 /* recovery follows the physical size of devices */
6667                 max_sectors = mddev->dev_sectors;
6668                 j = MaxSector;
6669                 rcu_read_lock();
6670                 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6671                         if (rdev->raid_disk >= 0 &&
6672                             !test_bit(Faulty, &rdev->flags) &&
6673                             !test_bit(In_sync, &rdev->flags) &&
6674                             rdev->recovery_offset < j)
6675                                 j = rdev->recovery_offset;
6676                 rcu_read_unlock();
6677         }
6678
6679         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6680         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6681                 " %d KB/sec/disk.\n", speed_min(mddev));
6682         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6683                "(but not more than %d KB/sec) for %s.\n",
6684                speed_max(mddev), desc);
6685
6686         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6687
6688         io_sectors = 0;
6689         for (m = 0; m < SYNC_MARKS; m++) {
6690                 mark[m] = jiffies;
6691                 mark_cnt[m] = io_sectors;
6692         }
6693         last_mark = 0;
6694         mddev->resync_mark = mark[last_mark];
6695         mddev->resync_mark_cnt = mark_cnt[last_mark];
6696
6697         /*
6698          * Tune reconstruction:
6699          */
6700         window = 32*(PAGE_SIZE/512);
6701         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6702                 window/2,(unsigned long long) max_sectors/2);
6703
6704         atomic_set(&mddev->recovery_active, 0);
6705         last_check = 0;
6706
6707         if (j>2) {
6708                 printk(KERN_INFO
6709                        "md: resuming %s of %s from checkpoint.\n",
6710                        desc, mdname(mddev));
6711                 mddev->curr_resync = j;
6712         }
6713         mddev->curr_resync_completed = mddev->curr_resync;
6714
6715         while (j < max_sectors) {
6716                 sector_t sectors;
6717
6718                 skipped = 0;
6719
6720                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6721                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6722                       (mddev->curr_resync - mddev->curr_resync_completed)
6723                       > (max_sectors >> 4)) ||
6724                      (j - mddev->curr_resync_completed)*2
6725                      >= mddev->resync_max - mddev->curr_resync_completed
6726                             )) {
6727                         /* time to update curr_resync_completed */
6728                         blk_unplug(mddev->queue);
6729                         wait_event(mddev->recovery_wait,
6730                                    atomic_read(&mddev->recovery_active) == 0);
6731                         mddev->curr_resync_completed =
6732                                 mddev->curr_resync;
6733                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6734                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6735                 }
6736
6737                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6738                         /* As this condition is controlled by user-space,
6739                          * we can block indefinitely, so use '_interruptible'
6740                          * to avoid triggering warnings.
6741                          */
6742                         flush_signals(current); /* just in case */
6743                         wait_event_interruptible(mddev->recovery_wait,
6744                                                  mddev->resync_max > j
6745                                                  || kthread_should_stop());
6746                 }
6747
6748                 if (kthread_should_stop())
6749                         goto interrupted;
6750
6751                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6752                                                   currspeed < speed_min(mddev));
6753                 if (sectors == 0) {
6754                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6755                         goto out;
6756                 }
6757
6758                 if (!skipped) { /* actual IO requested */
6759                         io_sectors += sectors;
6760                         atomic_add(sectors, &mddev->recovery_active);
6761                 }
6762
6763                 j += sectors;
6764                 if (j>1) mddev->curr_resync = j;
6765                 mddev->curr_mark_cnt = io_sectors;
6766                 if (last_check == 0)
6767                         /* this is the earliers that rebuilt will be
6768                          * visible in /proc/mdstat
6769                          */
6770                         md_new_event(mddev);
6771
6772                 if (last_check + window > io_sectors || j == max_sectors)
6773                         continue;
6774
6775                 last_check = io_sectors;
6776
6777                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6778                         break;
6779
6780         repeat:
6781                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6782                         /* step marks */
6783                         int next = (last_mark+1) % SYNC_MARKS;
6784
6785                         mddev->resync_mark = mark[next];
6786                         mddev->resync_mark_cnt = mark_cnt[next];
6787                         mark[next] = jiffies;
6788                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6789                         last_mark = next;
6790                 }
6791
6792
6793                 if (kthread_should_stop())
6794                         goto interrupted;
6795
6796
6797                 /*
6798                  * this loop exits only if either when we are slower than
6799                  * the 'hard' speed limit, or the system was IO-idle for
6800                  * a jiffy.
6801                  * the system might be non-idle CPU-wise, but we only care
6802                  * about not overloading the IO subsystem. (things like an
6803                  * e2fsck being done on the RAID array should execute fast)
6804                  */
6805                 blk_unplug(mddev->queue);
6806                 cond_resched();
6807
6808                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6809                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6810
6811                 if (currspeed > speed_min(mddev)) {
6812                         if ((currspeed > speed_max(mddev)) ||
6813                                         !is_mddev_idle(mddev, 0)) {
6814                                 msleep(500);
6815                                 goto repeat;
6816                         }
6817                 }
6818         }
6819         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6820         /*
6821          * this also signals 'finished resyncing' to md_stop
6822          */
6823  out:
6824         blk_unplug(mddev->queue);
6825
6826         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6827
6828         /* tell personality that we are finished */
6829         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6830
6831         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6832             mddev->curr_resync > 2) {
6833                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6834                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6835                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6836                                         printk(KERN_INFO
6837                                                "md: checkpointing %s of %s.\n",
6838                                                desc, mdname(mddev));
6839                                         mddev->recovery_cp = mddev->curr_resync;
6840                                 }
6841                         } else
6842                                 mddev->recovery_cp = MaxSector;
6843                 } else {
6844                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6845                                 mddev->curr_resync = MaxSector;
6846                         rcu_read_lock();
6847                         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6848                                 if (rdev->raid_disk >= 0 &&
6849                                     !test_bit(Faulty, &rdev->flags) &&
6850                                     !test_bit(In_sync, &rdev->flags) &&
6851                                     rdev->recovery_offset < mddev->curr_resync)
6852                                         rdev->recovery_offset = mddev->curr_resync;
6853                         rcu_read_unlock();
6854                 }
6855         }
6856         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6857
6858  skip:
6859         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6860                 /* We completed so min/max setting can be forgotten if used. */
6861                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6862                         mddev->resync_min = 0;
6863                 mddev->resync_max = MaxSector;
6864         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6865                 mddev->resync_min = mddev->curr_resync_completed;
6866         mddev->curr_resync = 0;
6867         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6868                 mddev->curr_resync_completed = 0;
6869         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6870         wake_up(&resync_wait);
6871         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6872         md_wakeup_thread(mddev->thread);
6873         return;
6874
6875  interrupted:
6876         /*
6877          * got a signal, exit.
6878          */
6879         printk(KERN_INFO
6880                "md: md_do_sync() got signal ... exiting\n");
6881         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6882         goto out;
6883
6884 }
6885 EXPORT_SYMBOL_GPL(md_do_sync);
6886
6887
6888 static int remove_and_add_spares(mddev_t *mddev)
6889 {
6890         mdk_rdev_t *rdev;
6891         int spares = 0;
6892
6893         mddev->curr_resync_completed = 0;
6894
6895         list_for_each_entry(rdev, &mddev->disks, same_set)
6896                 if (rdev->raid_disk >= 0 &&
6897                     !test_bit(Blocked, &rdev->flags) &&
6898                     (test_bit(Faulty, &rdev->flags) ||
6899                      ! test_bit(In_sync, &rdev->flags)) &&
6900                     atomic_read(&rdev->nr_pending)==0) {
6901                         if (mddev->pers->hot_remove_disk(
6902                                     mddev, rdev->raid_disk)==0) {
6903                                 char nm[20];
6904                                 sprintf(nm,"rd%d", rdev->raid_disk);
6905                                 sysfs_remove_link(&mddev->kobj, nm);
6906                                 rdev->raid_disk = -1;
6907                         }
6908                 }
6909
6910         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6911                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6912                         if (rdev->raid_disk >= 0 &&
6913                             !test_bit(In_sync, &rdev->flags) &&
6914                             !test_bit(Blocked, &rdev->flags))
6915                                 spares++;
6916                         if (rdev->raid_disk < 0
6917                             && !test_bit(Faulty, &rdev->flags)) {
6918                                 rdev->recovery_offset = 0;
6919                                 if (mddev->pers->
6920                                     hot_add_disk(mddev, rdev) == 0) {
6921                                         char nm[20];
6922                                         sprintf(nm, "rd%d", rdev->raid_disk);
6923                                         if (sysfs_create_link(&mddev->kobj,
6924                                                               &rdev->kobj, nm))
6925                                                 printk(KERN_WARNING
6926                                                        "md: cannot register "
6927                                                        "%s for %s\n",
6928                                                        nm, mdname(mddev));
6929                                         spares++;
6930                                         md_new_event(mddev);
6931                                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6932                                 } else
6933                                         break;
6934                         }
6935                 }
6936         }
6937         return spares;
6938 }
6939 /*
6940  * This routine is regularly called by all per-raid-array threads to
6941  * deal with generic issues like resync and super-block update.
6942  * Raid personalities that don't have a thread (linear/raid0) do not
6943  * need this as they never do any recovery or update the superblock.
6944  *
6945  * It does not do any resync itself, but rather "forks" off other threads
6946  * to do that as needed.
6947  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6948  * "->recovery" and create a thread at ->sync_thread.
6949  * When the thread finishes it sets MD_RECOVERY_DONE
6950  * and wakeups up this thread which will reap the thread and finish up.
6951  * This thread also removes any faulty devices (with nr_pending == 0).
6952  *
6953  * The overall approach is:
6954  *  1/ if the superblock needs updating, update it.
6955  *  2/ If a recovery thread is running, don't do anything else.
6956  *  3/ If recovery has finished, clean up, possibly marking spares active.
6957  *  4/ If there are any faulty devices, remove them.
6958  *  5/ If array is degraded, try to add spares devices
6959  *  6/ If array has spares or is not in-sync, start a resync thread.
6960  */
6961 void md_check_recovery(mddev_t *mddev)
6962 {
6963         mdk_rdev_t *rdev;
6964
6965
6966         if (mddev->bitmap)
6967                 bitmap_daemon_work(mddev);
6968
6969         if (mddev->ro)
6970                 return;
6971
6972         if (signal_pending(current)) {
6973                 if (mddev->pers->sync_request && !mddev->external) {
6974                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6975                                mdname(mddev));
6976                         mddev->safemode = 2;
6977                 }
6978                 flush_signals(current);
6979         }
6980
6981         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6982                 return;
6983         if ( ! (
6984                 (mddev->flags && !mddev->external) ||
6985                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6986                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6987                 (mddev->external == 0 && mddev->safemode == 1) ||
6988                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6989                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6990                 ))
6991                 return;
6992
6993         if (mddev_trylock(mddev)) {
6994                 int spares = 0;
6995
6996                 if (mddev->ro) {
6997                         /* Only thing we do on a ro array is remove
6998                          * failed devices.
6999                          */
7000                         remove_and_add_spares(mddev);
7001                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7002                         goto unlock;
7003                 }
7004
7005                 if (!mddev->external) {
7006                         int did_change = 0;
7007                         spin_lock_irq(&mddev->write_lock);
7008                         if (mddev->safemode &&
7009                             !atomic_read(&mddev->writes_pending) &&
7010                             !mddev->in_sync &&
7011                             mddev->recovery_cp == MaxSector) {
7012                                 mddev->in_sync = 1;
7013                                 did_change = 1;
7014                                 if (mddev->persistent)
7015                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7016                         }
7017                         if (mddev->safemode == 1)
7018                                 mddev->safemode = 0;
7019                         spin_unlock_irq(&mddev->write_lock);
7020                         if (did_change)
7021                                 sysfs_notify_dirent(mddev->sysfs_state);
7022                 }
7023
7024                 if (mddev->flags)
7025                         md_update_sb(mddev, 0);
7026
7027                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7028                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7029                         /* resync/recovery still happening */
7030                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7031                         goto unlock;
7032                 }
7033                 if (mddev->sync_thread) {
7034                         /* resync has finished, collect result */
7035                         md_unregister_thread(mddev->sync_thread);
7036                         mddev->sync_thread = NULL;
7037                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7038                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7039                                 /* success...*/
7040                                 /* activate any spares */
7041                                 if (mddev->pers->spare_active(mddev))
7042                                         sysfs_notify(&mddev->kobj, NULL,
7043                                                      "degraded");
7044                         }
7045                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7046                             mddev->pers->finish_reshape)
7047                                 mddev->pers->finish_reshape(mddev);
7048                         md_update_sb(mddev, 1);
7049
7050                         /* if array is no-longer degraded, then any saved_raid_disk
7051                          * information must be scrapped
7052                          */
7053                         if (!mddev->degraded)
7054                                 list_for_each_entry(rdev, &mddev->disks, same_set)
7055                                         rdev->saved_raid_disk = -1;
7056
7057                         mddev->recovery = 0;
7058                         /* flag recovery needed just to double check */
7059                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7060                         sysfs_notify_dirent(mddev->sysfs_action);
7061                         md_new_event(mddev);
7062                         goto unlock;
7063                 }
7064                 /* Set RUNNING before clearing NEEDED to avoid
7065                  * any transients in the value of "sync_action".
7066                  */
7067                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7068                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7069                 /* Clear some bits that don't mean anything, but
7070                  * might be left set
7071                  */
7072                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7073                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7074
7075                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7076                         goto unlock;
7077                 /* no recovery is running.
7078                  * remove any failed drives, then
7079                  * add spares if possible.
7080                  * Spare are also removed and re-added, to allow
7081                  * the personality to fail the re-add.
7082                  */
7083
7084                 if (mddev->reshape_position != MaxSector) {
7085                         if (mddev->pers->check_reshape == NULL ||
7086                             mddev->pers->check_reshape(mddev) != 0)
7087                                 /* Cannot proceed */
7088                                 goto unlock;
7089                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7090                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7091                 } else if ((spares = remove_and_add_spares(mddev))) {
7092                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7093                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7094                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7095                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7096                 } else if (mddev->recovery_cp < MaxSector) {
7097                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7098                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7099                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7100                         /* nothing to be done ... */
7101                         goto unlock;
7102
7103                 if (mddev->pers->sync_request) {
7104                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7105                                 /* We are adding a device or devices to an array
7106                                  * which has the bitmap stored on all devices.
7107                                  * So make sure all bitmap pages get written
7108                                  */
7109                                 bitmap_write_all(mddev->bitmap);
7110                         }
7111                         mddev->sync_thread = md_register_thread(md_do_sync,
7112                                                                 mddev,
7113                                                                 "resync");
7114                         if (!mddev->sync_thread) {
7115                                 printk(KERN_ERR "%s: could not start resync"
7116                                         " thread...\n",
7117                                         mdname(mddev));
7118                                 /* leave the spares where they are, it shouldn't hurt */
7119                                 mddev->recovery = 0;
7120                         } else
7121                                 md_wakeup_thread(mddev->sync_thread);
7122                         sysfs_notify_dirent(mddev->sysfs_action);
7123                         md_new_event(mddev);
7124                 }
7125         unlock:
7126                 if (!mddev->sync_thread) {
7127                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7128                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7129                                                &mddev->recovery))
7130                                 if (mddev->sysfs_action)
7131                                         sysfs_notify_dirent(mddev->sysfs_action);
7132                 }
7133                 mddev_unlock(mddev);
7134         }
7135 }
7136
7137 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7138 {
7139         sysfs_notify_dirent(rdev->sysfs_state);
7140         wait_event_timeout(rdev->blocked_wait,
7141                            !test_bit(Blocked, &rdev->flags),
7142                            msecs_to_jiffies(5000));
7143         rdev_dec_pending(rdev, mddev);
7144 }
7145 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7146
7147 static int md_notify_reboot(struct notifier_block *this,
7148                             unsigned long code, void *x)
7149 {
7150         struct list_head *tmp;
7151         mddev_t *mddev;
7152
7153         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7154
7155                 printk(KERN_INFO "md: stopping all md devices.\n");
7156
7157                 for_each_mddev(mddev, tmp)
7158                         if (mddev_trylock(mddev)) {
7159                                 /* Force a switch to readonly even array
7160                                  * appears to still be in use.  Hence
7161                                  * the '100'.
7162                                  */
7163                                 md_set_readonly(mddev, 100);
7164                                 mddev_unlock(mddev);
7165                         }
7166                 /*
7167                  * certain more exotic SCSI devices are known to be
7168                  * volatile wrt too early system reboots. While the
7169                  * right place to handle this issue is the given
7170                  * driver, we do want to have a safe RAID driver ...
7171                  */
7172                 mdelay(1000*1);
7173         }
7174         return NOTIFY_DONE;
7175 }
7176
7177 static struct notifier_block md_notifier = {
7178         .notifier_call  = md_notify_reboot,
7179         .next           = NULL,
7180         .priority       = INT_MAX, /* before any real devices */
7181 };
7182
7183 static void md_geninit(void)
7184 {
7185         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7186
7187         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7188 }
7189
7190 static int __init md_init(void)
7191 {
7192         if (register_blkdev(MD_MAJOR, "md"))
7193                 return -1;
7194         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7195                 unregister_blkdev(MD_MAJOR, "md");
7196                 return -1;
7197         }
7198         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7199                             md_probe, NULL, NULL);
7200         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7201                             md_probe, NULL, NULL);
7202
7203         register_reboot_notifier(&md_notifier);
7204         raid_table_header = register_sysctl_table(raid_root_table);
7205
7206         md_geninit();
7207         return 0;
7208 }
7209
7210
7211 #ifndef MODULE
7212
7213 /*
7214  * Searches all registered partitions for autorun RAID arrays
7215  * at boot time.
7216  */
7217
7218 static LIST_HEAD(all_detected_devices);
7219 struct detected_devices_node {
7220         struct list_head list;
7221         dev_t dev;
7222 };
7223
7224 void md_autodetect_dev(dev_t dev)
7225 {
7226         struct detected_devices_node *node_detected_dev;
7227
7228         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7229         if (node_detected_dev) {
7230                 node_detected_dev->dev = dev;
7231                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7232         } else {
7233                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7234                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7235         }
7236 }
7237
7238
7239 static void autostart_arrays(int part)
7240 {
7241         mdk_rdev_t *rdev;
7242         struct detected_devices_node *node_detected_dev;
7243         dev_t dev;
7244         int i_scanned, i_passed;
7245
7246         i_scanned = 0;
7247         i_passed = 0;
7248
7249         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7250
7251         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7252                 i_scanned++;
7253                 node_detected_dev = list_entry(all_detected_devices.next,
7254                                         struct detected_devices_node, list);
7255                 list_del(&node_detected_dev->list);
7256                 dev = node_detected_dev->dev;
7257                 kfree(node_detected_dev);
7258                 rdev = md_import_device(dev,0, 90);
7259                 if (IS_ERR(rdev))
7260                         continue;
7261
7262                 if (test_bit(Faulty, &rdev->flags)) {
7263                         MD_BUG();
7264                         continue;
7265                 }
7266                 set_bit(AutoDetected, &rdev->flags);
7267                 list_add(&rdev->same_set, &pending_raid_disks);
7268                 i_passed++;
7269         }
7270
7271         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7272                                                 i_scanned, i_passed);
7273
7274         autorun_devices(part);
7275 }
7276
7277 #endif /* !MODULE */
7278
7279 static __exit void md_exit(void)
7280 {
7281         mddev_t *mddev;
7282         struct list_head *tmp;
7283
7284         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7285         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7286
7287         unregister_blkdev(MD_MAJOR,"md");
7288         unregister_blkdev(mdp_major, "mdp");
7289         unregister_reboot_notifier(&md_notifier);
7290         unregister_sysctl_table(raid_table_header);
7291         remove_proc_entry("mdstat", NULL);
7292         for_each_mddev(mddev, tmp) {
7293                 export_array(mddev);
7294                 mddev->hold_active = 0;
7295         }
7296 }
7297
7298 subsys_initcall(md_init);
7299 module_exit(md_exit)
7300
7301 static int get_ro(char *buffer, struct kernel_param *kp)
7302 {
7303         return sprintf(buffer, "%d", start_readonly);
7304 }
7305 static int set_ro(const char *val, struct kernel_param *kp)
7306 {
7307         char *e;
7308         int num = simple_strtoul(val, &e, 10);
7309         if (*val && (*e == '\0' || *e == '\n')) {
7310                 start_readonly = num;
7311                 return 0;
7312         }
7313         return -EINVAL;
7314 }
7315
7316 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7317 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7318
7319 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7320
7321 EXPORT_SYMBOL(register_md_personality);
7322 EXPORT_SYMBOL(unregister_md_personality);
7323 EXPORT_SYMBOL(md_error);
7324 EXPORT_SYMBOL(md_done_sync);
7325 EXPORT_SYMBOL(md_write_start);
7326 EXPORT_SYMBOL(md_write_end);
7327 EXPORT_SYMBOL(md_register_thread);
7328 EXPORT_SYMBOL(md_unregister_thread);
7329 EXPORT_SYMBOL(md_wakeup_thread);
7330 EXPORT_SYMBOL(md_check_recovery);
7331 MODULE_LICENSE("GPL");
7332 MODULE_DESCRIPTION("MD RAID framework");
7333 MODULE_ALIAS("md");
7334 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);