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                     mddev->delta_disks >= 0 &&
2091                     !test_bit(In_sync, &rdev->flags) &&
2092                     mddev->curr_resync_completed > rdev->recovery_offset)
2093                                 rdev->recovery_offset = mddev->curr_resync_completed;
2094
2095         }
2096         list_for_each_entry(rdev, &mddev->disks, same_set) {
2097                 if (rdev->sb_events == mddev->events ||
2098                     (nospares &&
2099                      rdev->raid_disk < 0 &&
2100                      rdev->sb_events+1 == mddev->events)) {
2101                         /* Don't update this superblock */
2102                         rdev->sb_loaded = 2;
2103                 } else {
2104                         super_types[mddev->major_version].
2105                                 sync_super(mddev, rdev);
2106                         rdev->sb_loaded = 1;
2107                 }
2108         }
2109 }
2110
2111 static void md_update_sb(mddev_t * mddev, int force_change)
2112 {
2113         mdk_rdev_t *rdev;
2114         int sync_req;
2115         int nospares = 0;
2116
2117         mddev->utime = get_seconds();
2118         if (mddev->external)
2119                 return;
2120 repeat:
2121         spin_lock_irq(&mddev->write_lock);
2122
2123         set_bit(MD_CHANGE_PENDING, &mddev->flags);
2124         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2125                 force_change = 1;
2126         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2127                 /* just a clean<-> dirty transition, possibly leave spares alone,
2128                  * though if events isn't the right even/odd, we will have to do
2129                  * spares after all
2130                  */
2131                 nospares = 1;
2132         if (force_change)
2133                 nospares = 0;
2134         if (mddev->degraded)
2135                 /* If the array is degraded, then skipping spares is both
2136                  * dangerous and fairly pointless.
2137                  * Dangerous because a device that was removed from the array
2138                  * might have a event_count that still looks up-to-date,
2139                  * so it can be re-added without a resync.
2140                  * Pointless because if there are any spares to skip,
2141                  * then a recovery will happen and soon that array won't
2142                  * be degraded any more and the spare can go back to sleep then.
2143                  */
2144                 nospares = 0;
2145
2146         sync_req = mddev->in_sync;
2147
2148         /* If this is just a dirty<->clean transition, and the array is clean
2149          * and 'events' is odd, we can roll back to the previous clean state */
2150         if (nospares
2151             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2152             && mddev->can_decrease_events
2153             && mddev->events != 1) {
2154                 mddev->events--;
2155                 mddev->can_decrease_events = 0;
2156         } else {
2157                 /* otherwise we have to go forward and ... */
2158                 mddev->events ++;
2159                 mddev->can_decrease_events = nospares;
2160         }
2161
2162         if (!mddev->events) {
2163                 /*
2164                  * oops, this 64-bit counter should never wrap.
2165                  * Either we are in around ~1 trillion A.C., assuming
2166                  * 1 reboot per second, or we have a bug:
2167                  */
2168                 MD_BUG();
2169                 mddev->events --;
2170         }
2171
2172         /*
2173          * do not write anything to disk if using
2174          * nonpersistent superblocks
2175          */
2176         if (!mddev->persistent) {
2177                 if (!mddev->external)
2178                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2179
2180                 spin_unlock_irq(&mddev->write_lock);
2181                 wake_up(&mddev->sb_wait);
2182                 return;
2183         }
2184         sync_sbs(mddev, nospares);
2185         spin_unlock_irq(&mddev->write_lock);
2186
2187         dprintk(KERN_INFO
2188                 "md: updating %s RAID superblock on device (in sync %d)\n",
2189                 mdname(mddev),mddev->in_sync);
2190
2191         bitmap_update_sb(mddev->bitmap);
2192         list_for_each_entry(rdev, &mddev->disks, same_set) {
2193                 char b[BDEVNAME_SIZE];
2194                 dprintk(KERN_INFO "md: ");
2195                 if (rdev->sb_loaded != 1)
2196                         continue; /* no noise on spare devices */
2197                 if (test_bit(Faulty, &rdev->flags))
2198                         dprintk("(skipping faulty ");
2199
2200                 dprintk("%s ", bdevname(rdev->bdev,b));
2201                 if (!test_bit(Faulty, &rdev->flags)) {
2202                         md_super_write(mddev,rdev,
2203                                        rdev->sb_start, rdev->sb_size,
2204                                        rdev->sb_page);
2205                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2206                                 bdevname(rdev->bdev,b),
2207                                 (unsigned long long)rdev->sb_start);
2208                         rdev->sb_events = mddev->events;
2209
2210                 } else
2211                         dprintk(")\n");
2212                 if (mddev->level == LEVEL_MULTIPATH)
2213                         /* only need to write one superblock... */
2214                         break;
2215         }
2216         md_super_wait(mddev);
2217         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2218
2219         spin_lock_irq(&mddev->write_lock);
2220         if (mddev->in_sync != sync_req ||
2221             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2222                 /* have to write it out again */
2223                 spin_unlock_irq(&mddev->write_lock);
2224                 goto repeat;
2225         }
2226         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2227         spin_unlock_irq(&mddev->write_lock);
2228         wake_up(&mddev->sb_wait);
2229         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2230                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2231
2232 }
2233
2234 /* words written to sysfs files may, or may not, be \n terminated.
2235  * We want to accept with case. For this we use cmd_match.
2236  */
2237 static int cmd_match(const char *cmd, const char *str)
2238 {
2239         /* See if cmd, written into a sysfs file, matches
2240          * str.  They must either be the same, or cmd can
2241          * have a trailing newline
2242          */
2243         while (*cmd && *str && *cmd == *str) {
2244                 cmd++;
2245                 str++;
2246         }
2247         if (*cmd == '\n')
2248                 cmd++;
2249         if (*str || *cmd)
2250                 return 0;
2251         return 1;
2252 }
2253
2254 struct rdev_sysfs_entry {
2255         struct attribute attr;
2256         ssize_t (*show)(mdk_rdev_t *, char *);
2257         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2258 };
2259
2260 static ssize_t
2261 state_show(mdk_rdev_t *rdev, char *page)
2262 {
2263         char *sep = "";
2264         size_t len = 0;
2265
2266         if (test_bit(Faulty, &rdev->flags)) {
2267                 len+= sprintf(page+len, "%sfaulty",sep);
2268                 sep = ",";
2269         }
2270         if (test_bit(In_sync, &rdev->flags)) {
2271                 len += sprintf(page+len, "%sin_sync",sep);
2272                 sep = ",";
2273         }
2274         if (test_bit(WriteMostly, &rdev->flags)) {
2275                 len += sprintf(page+len, "%swrite_mostly",sep);
2276                 sep = ",";
2277         }
2278         if (test_bit(Blocked, &rdev->flags)) {
2279                 len += sprintf(page+len, "%sblocked", sep);
2280                 sep = ",";
2281         }
2282         if (!test_bit(Faulty, &rdev->flags) &&
2283             !test_bit(In_sync, &rdev->flags)) {
2284                 len += sprintf(page+len, "%sspare", sep);
2285                 sep = ",";
2286         }
2287         return len+sprintf(page+len, "\n");
2288 }
2289
2290 static ssize_t
2291 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2292 {
2293         /* can write
2294          *  faulty  - simulates and error
2295          *  remove  - disconnects the device
2296          *  writemostly - sets write_mostly
2297          *  -writemostly - clears write_mostly
2298          *  blocked - sets the Blocked flag
2299          *  -blocked - clears the Blocked flag
2300          *  insync - sets Insync providing device isn't active
2301          */
2302         int err = -EINVAL;
2303         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2304                 md_error(rdev->mddev, rdev);
2305                 err = 0;
2306         } else if (cmd_match(buf, "remove")) {
2307                 if (rdev->raid_disk >= 0)
2308                         err = -EBUSY;
2309                 else {
2310                         mddev_t *mddev = rdev->mddev;
2311                         kick_rdev_from_array(rdev);
2312                         if (mddev->pers)
2313                                 md_update_sb(mddev, 1);
2314                         md_new_event(mddev);
2315                         err = 0;
2316                 }
2317         } else if (cmd_match(buf, "writemostly")) {
2318                 set_bit(WriteMostly, &rdev->flags);
2319                 err = 0;
2320         } else if (cmd_match(buf, "-writemostly")) {
2321                 clear_bit(WriteMostly, &rdev->flags);
2322                 err = 0;
2323         } else if (cmd_match(buf, "blocked")) {
2324                 set_bit(Blocked, &rdev->flags);
2325                 err = 0;
2326         } else if (cmd_match(buf, "-blocked")) {
2327                 clear_bit(Blocked, &rdev->flags);
2328                 wake_up(&rdev->blocked_wait);
2329                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2330                 md_wakeup_thread(rdev->mddev->thread);
2331
2332                 err = 0;
2333         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2334                 set_bit(In_sync, &rdev->flags);
2335                 err = 0;
2336         }
2337         if (!err && rdev->sysfs_state)
2338                 sysfs_notify_dirent(rdev->sysfs_state);
2339         return err ? err : len;
2340 }
2341 static struct rdev_sysfs_entry rdev_state =
2342 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2343
2344 static ssize_t
2345 errors_show(mdk_rdev_t *rdev, char *page)
2346 {
2347         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2348 }
2349
2350 static ssize_t
2351 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2352 {
2353         char *e;
2354         unsigned long n = simple_strtoul(buf, &e, 10);
2355         if (*buf && (*e == 0 || *e == '\n')) {
2356                 atomic_set(&rdev->corrected_errors, n);
2357                 return len;
2358         }
2359         return -EINVAL;
2360 }
2361 static struct rdev_sysfs_entry rdev_errors =
2362 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2363
2364 static ssize_t
2365 slot_show(mdk_rdev_t *rdev, char *page)
2366 {
2367         if (rdev->raid_disk < 0)
2368                 return sprintf(page, "none\n");
2369         else
2370                 return sprintf(page, "%d\n", rdev->raid_disk);
2371 }
2372
2373 static ssize_t
2374 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2375 {
2376         char *e;
2377         int err;
2378         char nm[20];
2379         int slot = simple_strtoul(buf, &e, 10);
2380         if (strncmp(buf, "none", 4)==0)
2381                 slot = -1;
2382         else if (e==buf || (*e && *e!= '\n'))
2383                 return -EINVAL;
2384         if (rdev->mddev->pers && slot == -1) {
2385                 /* Setting 'slot' on an active array requires also
2386                  * updating the 'rd%d' link, and communicating
2387                  * with the personality with ->hot_*_disk.
2388                  * For now we only support removing
2389                  * failed/spare devices.  This normally happens automatically,
2390                  * but not when the metadata is externally managed.
2391                  */
2392                 if (rdev->raid_disk == -1)
2393                         return -EEXIST;
2394                 /* personality does all needed checks */
2395                 if (rdev->mddev->pers->hot_add_disk == NULL)
2396                         return -EINVAL;
2397                 err = rdev->mddev->pers->
2398                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2399                 if (err)
2400                         return err;
2401                 sprintf(nm, "rd%d", rdev->raid_disk);
2402                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2403                 rdev->raid_disk = -1;
2404                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2405                 md_wakeup_thread(rdev->mddev->thread);
2406         } else if (rdev->mddev->pers) {
2407                 mdk_rdev_t *rdev2;
2408                 /* Activating a spare .. or possibly reactivating
2409                  * if we ever get bitmaps working here.
2410                  */
2411
2412                 if (rdev->raid_disk != -1)
2413                         return -EBUSY;
2414
2415                 if (rdev->mddev->pers->hot_add_disk == NULL)
2416                         return -EINVAL;
2417
2418                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2419                         if (rdev2->raid_disk == slot)
2420                                 return -EEXIST;
2421
2422                 rdev->raid_disk = slot;
2423                 if (test_bit(In_sync, &rdev->flags))
2424                         rdev->saved_raid_disk = slot;
2425                 else
2426                         rdev->saved_raid_disk = -1;
2427                 err = rdev->mddev->pers->
2428                         hot_add_disk(rdev->mddev, rdev);
2429                 if (err) {
2430                         rdev->raid_disk = -1;
2431                         return err;
2432                 } else
2433                         sysfs_notify_dirent(rdev->sysfs_state);
2434                 sprintf(nm, "rd%d", rdev->raid_disk);
2435                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2436                         printk(KERN_WARNING
2437                                "md: cannot register "
2438                                "%s for %s\n",
2439                                nm, mdname(rdev->mddev));
2440
2441                 /* don't wakeup anyone, leave that to userspace. */
2442         } else {
2443                 if (slot >= rdev->mddev->raid_disks)
2444                         return -ENOSPC;
2445                 rdev->raid_disk = slot;
2446                 /* assume it is working */
2447                 clear_bit(Faulty, &rdev->flags);
2448                 clear_bit(WriteMostly, &rdev->flags);
2449                 set_bit(In_sync, &rdev->flags);
2450                 sysfs_notify_dirent(rdev->sysfs_state);
2451         }
2452         return len;
2453 }
2454
2455
2456 static struct rdev_sysfs_entry rdev_slot =
2457 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2458
2459 static ssize_t
2460 offset_show(mdk_rdev_t *rdev, char *page)
2461 {
2462         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2463 }
2464
2465 static ssize_t
2466 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2467 {
2468         char *e;
2469         unsigned long long offset = simple_strtoull(buf, &e, 10);
2470         if (e==buf || (*e && *e != '\n'))
2471                 return -EINVAL;
2472         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2473                 return -EBUSY;
2474         if (rdev->sectors && rdev->mddev->external)
2475                 /* Must set offset before size, so overlap checks
2476                  * can be sane */
2477                 return -EBUSY;
2478         rdev->data_offset = offset;
2479         return len;
2480 }
2481
2482 static struct rdev_sysfs_entry rdev_offset =
2483 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2484
2485 static ssize_t
2486 rdev_size_show(mdk_rdev_t *rdev, char *page)
2487 {
2488         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2489 }
2490
2491 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2492 {
2493         /* check if two start/length pairs overlap */
2494         if (s1+l1 <= s2)
2495                 return 0;
2496         if (s2+l2 <= s1)
2497                 return 0;
2498         return 1;
2499 }
2500
2501 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2502 {
2503         unsigned long long blocks;
2504         sector_t new;
2505
2506         if (strict_strtoull(buf, 10, &blocks) < 0)
2507                 return -EINVAL;
2508
2509         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2510                 return -EINVAL; /* sector conversion overflow */
2511
2512         new = blocks * 2;
2513         if (new != blocks * 2)
2514                 return -EINVAL; /* unsigned long long to sector_t overflow */
2515
2516         *sectors = new;
2517         return 0;
2518 }
2519
2520 static ssize_t
2521 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2522 {
2523         mddev_t *my_mddev = rdev->mddev;
2524         sector_t oldsectors = rdev->sectors;
2525         sector_t sectors;
2526
2527         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2528                 return -EINVAL;
2529         if (my_mddev->pers && rdev->raid_disk >= 0) {
2530                 if (my_mddev->persistent) {
2531                         sectors = super_types[my_mddev->major_version].
2532                                 rdev_size_change(rdev, sectors);
2533                         if (!sectors)
2534                                 return -EBUSY;
2535                 } else if (!sectors)
2536                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2537                                 rdev->data_offset;
2538         }
2539         if (sectors < my_mddev->dev_sectors)
2540                 return -EINVAL; /* component must fit device */
2541
2542         rdev->sectors = sectors;
2543         if (sectors > oldsectors && my_mddev->external) {
2544                 /* need to check that all other rdevs with the same ->bdev
2545                  * do not overlap.  We need to unlock the mddev to avoid
2546                  * a deadlock.  We have already changed rdev->sectors, and if
2547                  * we have to change it back, we will have the lock again.
2548                  */
2549                 mddev_t *mddev;
2550                 int overlap = 0;
2551                 struct list_head *tmp;
2552
2553                 mddev_unlock(my_mddev);
2554                 for_each_mddev(mddev, tmp) {
2555                         mdk_rdev_t *rdev2;
2556
2557                         mddev_lock(mddev);
2558                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2559                                 if (test_bit(AllReserved, &rdev2->flags) ||
2560                                     (rdev->bdev == rdev2->bdev &&
2561                                      rdev != rdev2 &&
2562                                      overlaps(rdev->data_offset, rdev->sectors,
2563                                               rdev2->data_offset,
2564                                               rdev2->sectors))) {
2565                                         overlap = 1;
2566                                         break;
2567                                 }
2568                         mddev_unlock(mddev);
2569                         if (overlap) {
2570                                 mddev_put(mddev);
2571                                 break;
2572                         }
2573                 }
2574                 mddev_lock(my_mddev);
2575                 if (overlap) {
2576                         /* Someone else could have slipped in a size
2577                          * change here, but doing so is just silly.
2578                          * We put oldsectors back because we *know* it is
2579                          * safe, and trust userspace not to race with
2580                          * itself
2581                          */
2582                         rdev->sectors = oldsectors;
2583                         return -EBUSY;
2584                 }
2585         }
2586         return len;
2587 }
2588
2589 static struct rdev_sysfs_entry rdev_size =
2590 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2591
2592
2593 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2594 {
2595         unsigned long long recovery_start = rdev->recovery_offset;
2596
2597         if (test_bit(In_sync, &rdev->flags) ||
2598             recovery_start == MaxSector)
2599                 return sprintf(page, "none\n");
2600
2601         return sprintf(page, "%llu\n", recovery_start);
2602 }
2603
2604 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2605 {
2606         unsigned long long recovery_start;
2607
2608         if (cmd_match(buf, "none"))
2609                 recovery_start = MaxSector;
2610         else if (strict_strtoull(buf, 10, &recovery_start))
2611                 return -EINVAL;
2612
2613         if (rdev->mddev->pers &&
2614             rdev->raid_disk >= 0)
2615                 return -EBUSY;
2616
2617         rdev->recovery_offset = recovery_start;
2618         if (recovery_start == MaxSector)
2619                 set_bit(In_sync, &rdev->flags);
2620         else
2621                 clear_bit(In_sync, &rdev->flags);
2622         return len;
2623 }
2624
2625 static struct rdev_sysfs_entry rdev_recovery_start =
2626 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2627
2628 static struct attribute *rdev_default_attrs[] = {
2629         &rdev_state.attr,
2630         &rdev_errors.attr,
2631         &rdev_slot.attr,
2632         &rdev_offset.attr,
2633         &rdev_size.attr,
2634         &rdev_recovery_start.attr,
2635         NULL,
2636 };
2637 static ssize_t
2638 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2639 {
2640         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2641         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2642         mddev_t *mddev = rdev->mddev;
2643         ssize_t rv;
2644
2645         if (!entry->show)
2646                 return -EIO;
2647
2648         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2649         if (!rv) {
2650                 if (rdev->mddev == NULL)
2651                         rv = -EBUSY;
2652                 else
2653                         rv = entry->show(rdev, page);
2654                 mddev_unlock(mddev);
2655         }
2656         return rv;
2657 }
2658
2659 static ssize_t
2660 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2661               const char *page, size_t length)
2662 {
2663         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2664         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2665         ssize_t rv;
2666         mddev_t *mddev = rdev->mddev;
2667
2668         if (!entry->store)
2669                 return -EIO;
2670         if (!capable(CAP_SYS_ADMIN))
2671                 return -EACCES;
2672         rv = mddev ? mddev_lock(mddev): -EBUSY;
2673         if (!rv) {
2674                 if (rdev->mddev == NULL)
2675                         rv = -EBUSY;
2676                 else
2677                         rv = entry->store(rdev, page, length);
2678                 mddev_unlock(mddev);
2679         }
2680         return rv;
2681 }
2682
2683 static void rdev_free(struct kobject *ko)
2684 {
2685         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2686         kfree(rdev);
2687 }
2688 static const struct sysfs_ops rdev_sysfs_ops = {
2689         .show           = rdev_attr_show,
2690         .store          = rdev_attr_store,
2691 };
2692 static struct kobj_type rdev_ktype = {
2693         .release        = rdev_free,
2694         .sysfs_ops      = &rdev_sysfs_ops,
2695         .default_attrs  = rdev_default_attrs,
2696 };
2697
2698 /*
2699  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2700  *
2701  * mark the device faulty if:
2702  *
2703  *   - the device is nonexistent (zero size)
2704  *   - the device has no valid superblock
2705  *
2706  * a faulty rdev _never_ has rdev->sb set.
2707  */
2708 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2709 {
2710         char b[BDEVNAME_SIZE];
2711         int err;
2712         mdk_rdev_t *rdev;
2713         sector_t size;
2714
2715         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2716         if (!rdev) {
2717                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2718                 return ERR_PTR(-ENOMEM);
2719         }
2720
2721         if ((err = alloc_disk_sb(rdev)))
2722                 goto abort_free;
2723
2724         err = lock_rdev(rdev, newdev, super_format == -2);
2725         if (err)
2726                 goto abort_free;
2727
2728         kobject_init(&rdev->kobj, &rdev_ktype);
2729
2730         rdev->desc_nr = -1;
2731         rdev->saved_raid_disk = -1;
2732         rdev->raid_disk = -1;
2733         rdev->flags = 0;
2734         rdev->data_offset = 0;
2735         rdev->sb_events = 0;
2736         rdev->last_read_error.tv_sec  = 0;
2737         rdev->last_read_error.tv_nsec = 0;
2738         atomic_set(&rdev->nr_pending, 0);
2739         atomic_set(&rdev->read_errors, 0);
2740         atomic_set(&rdev->corrected_errors, 0);
2741
2742         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2743         if (!size) {
2744                 printk(KERN_WARNING
2745                         "md: %s has zero or unknown size, marking faulty!\n",
2746                         bdevname(rdev->bdev,b));
2747                 err = -EINVAL;
2748                 goto abort_free;
2749         }
2750
2751         if (super_format >= 0) {
2752                 err = super_types[super_format].
2753                         load_super(rdev, NULL, super_minor);
2754                 if (err == -EINVAL) {
2755                         printk(KERN_WARNING
2756                                 "md: %s does not have a valid v%d.%d "
2757                                "superblock, not importing!\n",
2758                                 bdevname(rdev->bdev,b),
2759                                super_format, super_minor);
2760                         goto abort_free;
2761                 }
2762                 if (err < 0) {
2763                         printk(KERN_WARNING
2764                                 "md: could not read %s's sb, not importing!\n",
2765                                 bdevname(rdev->bdev,b));
2766                         goto abort_free;
2767                 }
2768         }
2769
2770         INIT_LIST_HEAD(&rdev->same_set);
2771         init_waitqueue_head(&rdev->blocked_wait);
2772
2773         return rdev;
2774
2775 abort_free:
2776         if (rdev->sb_page) {
2777                 if (rdev->bdev)
2778                         unlock_rdev(rdev);
2779                 free_disk_sb(rdev);
2780         }
2781         kfree(rdev);
2782         return ERR_PTR(err);
2783 }
2784
2785 /*
2786  * Check a full RAID array for plausibility
2787  */
2788
2789
2790 static void analyze_sbs(mddev_t * mddev)
2791 {
2792         int i;
2793         mdk_rdev_t *rdev, *freshest, *tmp;
2794         char b[BDEVNAME_SIZE];
2795
2796         freshest = NULL;
2797         rdev_for_each(rdev, tmp, mddev)
2798                 switch (super_types[mddev->major_version].
2799                         load_super(rdev, freshest, mddev->minor_version)) {
2800                 case 1:
2801                         freshest = rdev;
2802                         break;
2803                 case 0:
2804                         break;
2805                 default:
2806                         printk( KERN_ERR \
2807                                 "md: fatal superblock inconsistency in %s"
2808                                 " -- removing from array\n",
2809                                 bdevname(rdev->bdev,b));
2810                         kick_rdev_from_array(rdev);
2811                 }
2812
2813
2814         super_types[mddev->major_version].
2815                 validate_super(mddev, freshest);
2816
2817         i = 0;
2818         rdev_for_each(rdev, tmp, mddev) {
2819                 if (mddev->max_disks &&
2820                     (rdev->desc_nr >= mddev->max_disks ||
2821                      i > mddev->max_disks)) {
2822                         printk(KERN_WARNING
2823                                "md: %s: %s: only %d devices permitted\n",
2824                                mdname(mddev), bdevname(rdev->bdev, b),
2825                                mddev->max_disks);
2826                         kick_rdev_from_array(rdev);
2827                         continue;
2828                 }
2829                 if (rdev != freshest)
2830                         if (super_types[mddev->major_version].
2831                             validate_super(mddev, rdev)) {
2832                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2833                                         " from array!\n",
2834                                         bdevname(rdev->bdev,b));
2835                                 kick_rdev_from_array(rdev);
2836                                 continue;
2837                         }
2838                 if (mddev->level == LEVEL_MULTIPATH) {
2839                         rdev->desc_nr = i++;
2840                         rdev->raid_disk = rdev->desc_nr;
2841                         set_bit(In_sync, &rdev->flags);
2842                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2843                         rdev->raid_disk = -1;
2844                         clear_bit(In_sync, &rdev->flags);
2845                 }
2846         }
2847 }
2848
2849 /* Read a fixed-point number.
2850  * Numbers in sysfs attributes should be in "standard" units where
2851  * possible, so time should be in seconds.
2852  * However we internally use a a much smaller unit such as
2853  * milliseconds or jiffies.
2854  * This function takes a decimal number with a possible fractional
2855  * component, and produces an integer which is the result of
2856  * multiplying that number by 10^'scale'.
2857  * all without any floating-point arithmetic.
2858  */
2859 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2860 {
2861         unsigned long result = 0;
2862         long decimals = -1;
2863         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2864                 if (*cp == '.')
2865                         decimals = 0;
2866                 else if (decimals < scale) {
2867                         unsigned int value;
2868                         value = *cp - '0';
2869                         result = result * 10 + value;
2870                         if (decimals >= 0)
2871                                 decimals++;
2872                 }
2873                 cp++;
2874         }
2875         if (*cp == '\n')
2876                 cp++;
2877         if (*cp)
2878                 return -EINVAL;
2879         if (decimals < 0)
2880                 decimals = 0;
2881         while (decimals < scale) {
2882                 result *= 10;
2883                 decimals ++;
2884         }
2885         *res = result;
2886         return 0;
2887 }
2888
2889
2890 static void md_safemode_timeout(unsigned long data);
2891
2892 static ssize_t
2893 safe_delay_show(mddev_t *mddev, char *page)
2894 {
2895         int msec = (mddev->safemode_delay*1000)/HZ;
2896         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2897 }
2898 static ssize_t
2899 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2900 {
2901         unsigned long msec;
2902
2903         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2904                 return -EINVAL;
2905         if (msec == 0)
2906                 mddev->safemode_delay = 0;
2907         else {
2908                 unsigned long old_delay = mddev->safemode_delay;
2909                 mddev->safemode_delay = (msec*HZ)/1000;
2910                 if (mddev->safemode_delay == 0)
2911                         mddev->safemode_delay = 1;
2912                 if (mddev->safemode_delay < old_delay)
2913                         md_safemode_timeout((unsigned long)mddev);
2914         }
2915         return len;
2916 }
2917 static struct md_sysfs_entry md_safe_delay =
2918 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2919
2920 static ssize_t
2921 level_show(mddev_t *mddev, char *page)
2922 {
2923         struct mdk_personality *p = mddev->pers;
2924         if (p)
2925                 return sprintf(page, "%s\n", p->name);
2926         else if (mddev->clevel[0])
2927                 return sprintf(page, "%s\n", mddev->clevel);
2928         else if (mddev->level != LEVEL_NONE)
2929                 return sprintf(page, "%d\n", mddev->level);
2930         else
2931                 return 0;
2932 }
2933
2934 static ssize_t
2935 level_store(mddev_t *mddev, const char *buf, size_t len)
2936 {
2937         char clevel[16];
2938         ssize_t rv = len;
2939         struct mdk_personality *pers;
2940         long level;
2941         void *priv;
2942         mdk_rdev_t *rdev;
2943
2944         if (mddev->pers == NULL) {
2945                 if (len == 0)
2946                         return 0;
2947                 if (len >= sizeof(mddev->clevel))
2948                         return -ENOSPC;
2949                 strncpy(mddev->clevel, buf, len);
2950                 if (mddev->clevel[len-1] == '\n')
2951                         len--;
2952                 mddev->clevel[len] = 0;
2953                 mddev->level = LEVEL_NONE;
2954                 return rv;
2955         }
2956
2957         /* request to change the personality.  Need to ensure:
2958          *  - array is not engaged in resync/recovery/reshape
2959          *  - old personality can be suspended
2960          *  - new personality will access other array.
2961          */
2962
2963         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2964                 return -EBUSY;
2965
2966         if (!mddev->pers->quiesce) {
2967                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2968                        mdname(mddev), mddev->pers->name);
2969                 return -EINVAL;
2970         }
2971
2972         /* Now find the new personality */
2973         if (len == 0 || len >= sizeof(clevel))
2974                 return -EINVAL;
2975         strncpy(clevel, buf, len);
2976         if (clevel[len-1] == '\n')
2977                 len--;
2978         clevel[len] = 0;
2979         if (strict_strtol(clevel, 10, &level))
2980                 level = LEVEL_NONE;
2981
2982         if (request_module("md-%s", clevel) != 0)
2983                 request_module("md-level-%s", clevel);
2984         spin_lock(&pers_lock);
2985         pers = find_pers(level, clevel);
2986         if (!pers || !try_module_get(pers->owner)) {
2987                 spin_unlock(&pers_lock);
2988                 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2989                 return -EINVAL;
2990         }
2991         spin_unlock(&pers_lock);
2992
2993         if (pers == mddev->pers) {
2994                 /* Nothing to do! */
2995                 module_put(pers->owner);
2996                 return rv;
2997         }
2998         if (!pers->takeover) {
2999                 module_put(pers->owner);
3000                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3001                        mdname(mddev), clevel);
3002                 return -EINVAL;
3003         }
3004
3005         list_for_each_entry(rdev, &mddev->disks, same_set)
3006                 rdev->new_raid_disk = rdev->raid_disk;
3007
3008         /* ->takeover must set new_* and/or delta_disks
3009          * if it succeeds, and may set them when it fails.
3010          */
3011         priv = pers->takeover(mddev);
3012         if (IS_ERR(priv)) {
3013                 mddev->new_level = mddev->level;
3014                 mddev->new_layout = mddev->layout;
3015                 mddev->new_chunk_sectors = mddev->chunk_sectors;
3016                 mddev->raid_disks -= mddev->delta_disks;
3017                 mddev->delta_disks = 0;
3018                 module_put(pers->owner);
3019                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3020                        mdname(mddev), clevel);
3021                 return PTR_ERR(priv);
3022         }
3023
3024         /* Looks like we have a winner */
3025         mddev_suspend(mddev);
3026         mddev->pers->stop(mddev);
3027
3028         if (mddev->pers->sync_request == NULL &&
3029             pers->sync_request != NULL) {
3030                 /* need to add the md_redundancy_group */
3031                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3032                         printk(KERN_WARNING
3033                                "md: cannot register extra attributes for %s\n",
3034                                mdname(mddev));
3035                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3036         }
3037         if (mddev->pers->sync_request != NULL &&
3038             pers->sync_request == NULL) {
3039                 /* need to remove the md_redundancy_group */
3040                 if (mddev->to_remove == NULL)
3041                         mddev->to_remove = &md_redundancy_group;
3042         }
3043
3044         if (mddev->pers->sync_request == NULL &&
3045             mddev->external) {
3046                 /* We are converting from a no-redundancy array
3047                  * to a redundancy array and metadata is managed
3048                  * externally so we need to be sure that writes
3049                  * won't block due to a need to transition
3050                  *      clean->dirty
3051                  * until external management is started.
3052                  */
3053                 mddev->in_sync = 0;
3054                 mddev->safemode_delay = 0;
3055                 mddev->safemode = 0;
3056         }
3057
3058         list_for_each_entry(rdev, &mddev->disks, same_set) {
3059                 char nm[20];
3060                 if (rdev->raid_disk < 0)
3061                         continue;
3062                 if (rdev->new_raid_disk > mddev->raid_disks)
3063                         rdev->new_raid_disk = -1;
3064                 if (rdev->new_raid_disk == rdev->raid_disk)
3065                         continue;
3066                 sprintf(nm, "rd%d", rdev->raid_disk);
3067                 sysfs_remove_link(&mddev->kobj, nm);
3068         }
3069         list_for_each_entry(rdev, &mddev->disks, same_set) {
3070                 if (rdev->raid_disk < 0)
3071                         continue;
3072                 if (rdev->new_raid_disk == rdev->raid_disk)
3073                         continue;
3074                 rdev->raid_disk = rdev->new_raid_disk;
3075                 if (rdev->raid_disk < 0)
3076                         clear_bit(In_sync, &rdev->flags);
3077                 else {
3078                         char nm[20];
3079                         sprintf(nm, "rd%d", rdev->raid_disk);
3080                         if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3081                                 printk("md: cannot register %s for %s after level change\n",
3082                                        nm, mdname(mddev));
3083                 }
3084         }
3085
3086         module_put(mddev->pers->owner);
3087         mddev->pers = pers;
3088         mddev->private = priv;
3089         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3090         mddev->level = mddev->new_level;
3091         mddev->layout = mddev->new_layout;
3092         mddev->chunk_sectors = mddev->new_chunk_sectors;
3093         mddev->delta_disks = 0;
3094         if (mddev->pers->sync_request == NULL) {
3095                 /* this is now an array without redundancy, so
3096                  * it must always be in_sync
3097                  */
3098                 mddev->in_sync = 1;
3099                 del_timer_sync(&mddev->safemode_timer);
3100         }
3101         pers->run(mddev);
3102         mddev_resume(mddev);
3103         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3104         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3105         md_wakeup_thread(mddev->thread);
3106         sysfs_notify(&mddev->kobj, NULL, "level");
3107         md_new_event(mddev);
3108         return rv;
3109 }
3110
3111 static struct md_sysfs_entry md_level =
3112 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3113
3114
3115 static ssize_t
3116 layout_show(mddev_t *mddev, char *page)
3117 {
3118         /* just a number, not meaningful for all levels */
3119         if (mddev->reshape_position != MaxSector &&
3120             mddev->layout != mddev->new_layout)
3121                 return sprintf(page, "%d (%d)\n",
3122                                mddev->new_layout, mddev->layout);
3123         return sprintf(page, "%d\n", mddev->layout);
3124 }
3125
3126 static ssize_t
3127 layout_store(mddev_t *mddev, const char *buf, size_t len)
3128 {
3129         char *e;
3130         unsigned long n = simple_strtoul(buf, &e, 10);
3131
3132         if (!*buf || (*e && *e != '\n'))
3133                 return -EINVAL;
3134
3135         if (mddev->pers) {
3136                 int err;
3137                 if (mddev->pers->check_reshape == NULL)
3138                         return -EBUSY;
3139                 mddev->new_layout = n;
3140                 err = mddev->pers->check_reshape(mddev);
3141                 if (err) {
3142                         mddev->new_layout = mddev->layout;
3143                         return err;
3144                 }
3145         } else {
3146                 mddev->new_layout = n;
3147                 if (mddev->reshape_position == MaxSector)
3148                         mddev->layout = n;
3149         }
3150         return len;
3151 }
3152 static struct md_sysfs_entry md_layout =
3153 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3154
3155
3156 static ssize_t
3157 raid_disks_show(mddev_t *mddev, char *page)
3158 {
3159         if (mddev->raid_disks == 0)
3160                 return 0;
3161         if (mddev->reshape_position != MaxSector &&
3162             mddev->delta_disks != 0)
3163                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3164                                mddev->raid_disks - mddev->delta_disks);
3165         return sprintf(page, "%d\n", mddev->raid_disks);
3166 }
3167
3168 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3169
3170 static ssize_t
3171 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3172 {
3173         char *e;
3174         int rv = 0;
3175         unsigned long n = simple_strtoul(buf, &e, 10);
3176
3177         if (!*buf || (*e && *e != '\n'))
3178                 return -EINVAL;
3179
3180         if (mddev->pers)
3181                 rv = update_raid_disks(mddev, n);
3182         else if (mddev->reshape_position != MaxSector) {
3183                 int olddisks = mddev->raid_disks - mddev->delta_disks;
3184                 mddev->delta_disks = n - olddisks;
3185                 mddev->raid_disks = n;
3186         } else
3187                 mddev->raid_disks = n;
3188         return rv ? rv : len;
3189 }
3190 static struct md_sysfs_entry md_raid_disks =
3191 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3192
3193 static ssize_t
3194 chunk_size_show(mddev_t *mddev, char *page)
3195 {
3196         if (mddev->reshape_position != MaxSector &&
3197             mddev->chunk_sectors != mddev->new_chunk_sectors)
3198                 return sprintf(page, "%d (%d)\n",
3199                                mddev->new_chunk_sectors << 9,
3200                                mddev->chunk_sectors << 9);
3201         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3202 }
3203
3204 static ssize_t
3205 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3206 {
3207         char *e;
3208         unsigned long n = simple_strtoul(buf, &e, 10);
3209
3210         if (!*buf || (*e && *e != '\n'))
3211                 return -EINVAL;
3212
3213         if (mddev->pers) {
3214                 int err;
3215                 if (mddev->pers->check_reshape == NULL)
3216                         return -EBUSY;
3217                 mddev->new_chunk_sectors = n >> 9;
3218                 err = mddev->pers->check_reshape(mddev);
3219                 if (err) {
3220                         mddev->new_chunk_sectors = mddev->chunk_sectors;
3221                         return err;
3222                 }
3223         } else {
3224                 mddev->new_chunk_sectors = n >> 9;
3225                 if (mddev->reshape_position == MaxSector)
3226                         mddev->chunk_sectors = n >> 9;
3227         }
3228         return len;
3229 }
3230 static struct md_sysfs_entry md_chunk_size =
3231 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3232
3233 static ssize_t
3234 resync_start_show(mddev_t *mddev, char *page)
3235 {
3236         if (mddev->recovery_cp == MaxSector)
3237                 return sprintf(page, "none\n");
3238         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3239 }
3240
3241 static ssize_t
3242 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3243 {
3244         char *e;
3245         unsigned long long n = simple_strtoull(buf, &e, 10);
3246
3247         if (mddev->pers)
3248                 return -EBUSY;
3249         if (cmd_match(buf, "none"))
3250                 n = MaxSector;
3251         else if (!*buf || (*e && *e != '\n'))
3252                 return -EINVAL;
3253
3254         mddev->recovery_cp = n;
3255         return len;
3256 }
3257 static struct md_sysfs_entry md_resync_start =
3258 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3259
3260 /*
3261  * The array state can be:
3262  *
3263  * clear
3264  *     No devices, no size, no level
3265  *     Equivalent to STOP_ARRAY ioctl
3266  * inactive
3267  *     May have some settings, but array is not active
3268  *        all IO results in error
3269  *     When written, doesn't tear down array, but just stops it
3270  * suspended (not supported yet)
3271  *     All IO requests will block. The array can be reconfigured.
3272  *     Writing this, if accepted, will block until array is quiescent
3273  * readonly
3274  *     no resync can happen.  no superblocks get written.
3275  *     write requests fail
3276  * read-auto
3277  *     like readonly, but behaves like 'clean' on a write request.
3278  *
3279  * clean - no pending writes, but otherwise active.
3280  *     When written to inactive array, starts without resync
3281  *     If a write request arrives then
3282  *       if metadata is known, mark 'dirty' and switch to 'active'.
3283  *       if not known, block and switch to write-pending
3284  *     If written to an active array that has pending writes, then fails.
3285  * active
3286  *     fully active: IO and resync can be happening.
3287  *     When written to inactive array, starts with resync
3288  *
3289  * write-pending
3290  *     clean, but writes are blocked waiting for 'active' to be written.
3291  *
3292  * active-idle
3293  *     like active, but no writes have been seen for a while (100msec).
3294  *
3295  */
3296 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3297                    write_pending, active_idle, bad_word};
3298 static char *array_states[] = {
3299         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3300         "write-pending", "active-idle", NULL };
3301
3302 static int match_word(const char *word, char **list)
3303 {
3304         int n;
3305         for (n=0; list[n]; n++)
3306                 if (cmd_match(word, list[n]))
3307                         break;
3308         return n;
3309 }
3310
3311 static ssize_t
3312 array_state_show(mddev_t *mddev, char *page)
3313 {
3314         enum array_state st = inactive;
3315
3316         if (mddev->pers)
3317                 switch(mddev->ro) {
3318                 case 1:
3319                         st = readonly;
3320                         break;
3321                 case 2:
3322                         st = read_auto;
3323                         break;
3324                 case 0:
3325                         if (mddev->in_sync)
3326                                 st = clean;
3327                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3328                                 st = write_pending;
3329                         else if (mddev->safemode)
3330                                 st = active_idle;
3331                         else
3332                                 st = active;
3333                 }
3334         else {
3335                 if (list_empty(&mddev->disks) &&
3336                     mddev->raid_disks == 0 &&
3337                     mddev->dev_sectors == 0)
3338                         st = clear;
3339                 else
3340                         st = inactive;
3341         }
3342         return sprintf(page, "%s\n", array_states[st]);
3343 }
3344
3345 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3346 static int md_set_readonly(mddev_t * mddev, int is_open);
3347 static int do_md_run(mddev_t * mddev);
3348 static int restart_array(mddev_t *mddev);
3349
3350 static ssize_t
3351 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3352 {
3353         int err = -EINVAL;
3354         enum array_state st = match_word(buf, array_states);
3355         switch(st) {
3356         case bad_word:
3357                 break;
3358         case clear:
3359                 /* stopping an active array */
3360                 if (atomic_read(&mddev->openers) > 0)
3361                         return -EBUSY;
3362                 err = do_md_stop(mddev, 0, 0);
3363                 break;
3364         case inactive:
3365                 /* stopping an active array */
3366                 if (mddev->pers) {
3367                         if (atomic_read(&mddev->openers) > 0)
3368                                 return -EBUSY;
3369                         err = do_md_stop(mddev, 2, 0);
3370                 } else
3371                         err = 0; /* already inactive */
3372                 break;
3373         case suspended:
3374                 break; /* not supported yet */
3375         case readonly:
3376                 if (mddev->pers)
3377                         err = md_set_readonly(mddev, 0);
3378                 else {
3379                         mddev->ro = 1;
3380                         set_disk_ro(mddev->gendisk, 1);
3381                         err = do_md_run(mddev);
3382                 }
3383                 break;
3384         case read_auto:
3385                 if (mddev->pers) {
3386                         if (mddev->ro == 0)
3387                                 err = md_set_readonly(mddev, 0);
3388                         else if (mddev->ro == 1)
3389                                 err = restart_array(mddev);
3390                         if (err == 0) {
3391                                 mddev->ro = 2;
3392                                 set_disk_ro(mddev->gendisk, 0);
3393                         }
3394                 } else {
3395                         mddev->ro = 2;
3396                         err = do_md_run(mddev);
3397                 }
3398                 break;
3399         case clean:
3400                 if (mddev->pers) {
3401                         restart_array(mddev);
3402                         spin_lock_irq(&mddev->write_lock);
3403                         if (atomic_read(&mddev->writes_pending) == 0) {
3404                                 if (mddev->in_sync == 0) {
3405                                         mddev->in_sync = 1;
3406                                         if (mddev->safemode == 1)
3407                                                 mddev->safemode = 0;
3408                                         if (mddev->persistent)
3409                                                 set_bit(MD_CHANGE_CLEAN,
3410                                                         &mddev->flags);
3411                                 }
3412                                 err = 0;
3413                         } else
3414                                 err = -EBUSY;
3415                         spin_unlock_irq(&mddev->write_lock);
3416                 } else
3417                         err = -EINVAL;
3418                 break;
3419         case active:
3420                 if (mddev->pers) {
3421                         restart_array(mddev);
3422                         if (mddev->external)
3423                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3424                         wake_up(&mddev->sb_wait);
3425                         err = 0;
3426                 } else {
3427                         mddev->ro = 0;
3428                         set_disk_ro(mddev->gendisk, 0);
3429                         err = do_md_run(mddev);
3430                 }
3431                 break;
3432         case write_pending:
3433         case active_idle:
3434                 /* these cannot be set */
3435                 break;
3436         }
3437         if (err)
3438                 return err;
3439         else {
3440                 sysfs_notify_dirent(mddev->sysfs_state);
3441                 return len;
3442         }
3443 }
3444 static struct md_sysfs_entry md_array_state =
3445 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3446
3447 static ssize_t
3448 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3449         return sprintf(page, "%d\n",
3450                        atomic_read(&mddev->max_corr_read_errors));
3451 }
3452
3453 static ssize_t
3454 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3455 {
3456         char *e;
3457         unsigned long n = simple_strtoul(buf, &e, 10);
3458
3459         if (*buf && (*e == 0 || *e == '\n')) {
3460                 atomic_set(&mddev->max_corr_read_errors, n);
3461                 return len;
3462         }
3463         return -EINVAL;
3464 }
3465
3466 static struct md_sysfs_entry max_corr_read_errors =
3467 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3468         max_corrected_read_errors_store);
3469
3470 static ssize_t
3471 null_show(mddev_t *mddev, char *page)
3472 {
3473         return -EINVAL;
3474 }
3475
3476 static ssize_t
3477 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3478 {
3479         /* buf must be %d:%d\n? giving major and minor numbers */
3480         /* The new device is added to the array.
3481          * If the array has a persistent superblock, we read the
3482          * superblock to initialise info and check validity.
3483          * Otherwise, only checking done is that in bind_rdev_to_array,
3484          * which mainly checks size.
3485          */
3486         char *e;
3487         int major = simple_strtoul(buf, &e, 10);
3488         int minor;
3489         dev_t dev;
3490         mdk_rdev_t *rdev;
3491         int err;
3492
3493         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3494                 return -EINVAL;
3495         minor = simple_strtoul(e+1, &e, 10);
3496         if (*e && *e != '\n')
3497                 return -EINVAL;
3498         dev = MKDEV(major, minor);
3499         if (major != MAJOR(dev) ||
3500             minor != MINOR(dev))
3501                 return -EOVERFLOW;
3502
3503
3504         if (mddev->persistent) {
3505                 rdev = md_import_device(dev, mddev->major_version,
3506                                         mddev->minor_version);
3507                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3508                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3509                                                        mdk_rdev_t, same_set);
3510                         err = super_types[mddev->major_version]
3511                                 .load_super(rdev, rdev0, mddev->minor_version);
3512                         if (err < 0)
3513                                 goto out;
3514                 }
3515         } else if (mddev->external)
3516                 rdev = md_import_device(dev, -2, -1);
3517         else
3518                 rdev = md_import_device(dev, -1, -1);
3519
3520         if (IS_ERR(rdev))
3521                 return PTR_ERR(rdev);
3522         err = bind_rdev_to_array(rdev, mddev);
3523  out:
3524         if (err)
3525                 export_rdev(rdev);
3526         return err ? err : len;
3527 }
3528
3529 static struct md_sysfs_entry md_new_device =
3530 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3531
3532 static ssize_t
3533 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3534 {
3535         char *end;
3536         unsigned long chunk, end_chunk;
3537
3538         if (!mddev->bitmap)
3539                 goto out;
3540         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3541         while (*buf) {
3542                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3543                 if (buf == end) break;
3544                 if (*end == '-') { /* range */
3545                         buf = end + 1;
3546                         end_chunk = simple_strtoul(buf, &end, 0);
3547                         if (buf == end) break;
3548                 }
3549                 if (*end && !isspace(*end)) break;
3550                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3551                 buf = skip_spaces(end);
3552         }
3553         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3554 out:
3555         return len;
3556 }
3557
3558 static struct md_sysfs_entry md_bitmap =
3559 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3560
3561 static ssize_t
3562 size_show(mddev_t *mddev, char *page)
3563 {
3564         return sprintf(page, "%llu\n",
3565                 (unsigned long long)mddev->dev_sectors / 2);
3566 }
3567
3568 static int update_size(mddev_t *mddev, sector_t num_sectors);
3569
3570 static ssize_t
3571 size_store(mddev_t *mddev, const char *buf, size_t len)
3572 {
3573         /* If array is inactive, we can reduce the component size, but
3574          * not increase it (except from 0).
3575          * If array is active, we can try an on-line resize
3576          */
3577         sector_t sectors;
3578         int err = strict_blocks_to_sectors(buf, &sectors);
3579
3580         if (err < 0)
3581                 return err;
3582         if (mddev->pers) {
3583                 err = update_size(mddev, sectors);
3584                 md_update_sb(mddev, 1);
3585         } else {
3586                 if (mddev->dev_sectors == 0 ||
3587                     mddev->dev_sectors > sectors)
3588                         mddev->dev_sectors = sectors;
3589                 else
3590                         err = -ENOSPC;
3591         }
3592         return err ? err : len;
3593 }
3594
3595 static struct md_sysfs_entry md_size =
3596 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3597
3598
3599 /* Metdata version.
3600  * This is one of
3601  *   'none' for arrays with no metadata (good luck...)
3602  *   'external' for arrays with externally managed metadata,
3603  * or N.M for internally known formats
3604  */
3605 static ssize_t
3606 metadata_show(mddev_t *mddev, char *page)
3607 {
3608         if (mddev->persistent)
3609                 return sprintf(page, "%d.%d\n",
3610                                mddev->major_version, mddev->minor_version);
3611         else if (mddev->external)
3612                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3613         else
3614                 return sprintf(page, "none\n");
3615 }
3616
3617 static ssize_t
3618 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3619 {
3620         int major, minor;
3621         char *e;
3622         /* Changing the details of 'external' metadata is
3623          * always permitted.  Otherwise there must be
3624          * no devices attached to the array.
3625          */
3626         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3627                 ;
3628         else if (!list_empty(&mddev->disks))
3629                 return -EBUSY;
3630
3631         if (cmd_match(buf, "none")) {
3632                 mddev->persistent = 0;
3633                 mddev->external = 0;
3634                 mddev->major_version = 0;
3635                 mddev->minor_version = 90;
3636                 return len;
3637         }
3638         if (strncmp(buf, "external:", 9) == 0) {
3639                 size_t namelen = len-9;
3640                 if (namelen >= sizeof(mddev->metadata_type))
3641                         namelen = sizeof(mddev->metadata_type)-1;
3642                 strncpy(mddev->metadata_type, buf+9, namelen);
3643                 mddev->metadata_type[namelen] = 0;
3644                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3645                         mddev->metadata_type[--namelen] = 0;
3646                 mddev->persistent = 0;
3647                 mddev->external = 1;
3648                 mddev->major_version = 0;
3649                 mddev->minor_version = 90;
3650                 return len;
3651         }
3652         major = simple_strtoul(buf, &e, 10);
3653         if (e==buf || *e != '.')
3654                 return -EINVAL;
3655         buf = e+1;
3656         minor = simple_strtoul(buf, &e, 10);
3657         if (e==buf || (*e && *e != '\n') )
3658                 return -EINVAL;
3659         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3660                 return -ENOENT;
3661         mddev->major_version = major;
3662         mddev->minor_version = minor;
3663         mddev->persistent = 1;
3664         mddev->external = 0;
3665         return len;
3666 }
3667
3668 static struct md_sysfs_entry md_metadata =
3669 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3670
3671 static ssize_t
3672 action_show(mddev_t *mddev, char *page)
3673 {
3674         char *type = "idle";
3675         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3676                 type = "frozen";
3677         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3678             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3679                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3680                         type = "reshape";
3681                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3682                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3683                                 type = "resync";
3684                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3685                                 type = "check";
3686                         else
3687                                 type = "repair";
3688                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3689                         type = "recover";
3690         }
3691         return sprintf(page, "%s\n", type);
3692 }
3693
3694 static ssize_t
3695 action_store(mddev_t *mddev, const char *page, size_t len)
3696 {
3697         if (!mddev->pers || !mddev->pers->sync_request)
3698                 return -EINVAL;
3699
3700         if (cmd_match(page, "frozen"))
3701                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3702         else
3703                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3704
3705         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3706                 if (mddev->sync_thread) {
3707                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3708                         md_unregister_thread(mddev->sync_thread);
3709                         mddev->sync_thread = NULL;
3710                         mddev->recovery = 0;
3711                 }
3712         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3713                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3714                 return -EBUSY;
3715         else if (cmd_match(page, "resync"))
3716                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3717         else if (cmd_match(page, "recover")) {
3718                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3719                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3720         } else if (cmd_match(page, "reshape")) {
3721                 int err;
3722                 if (mddev->pers->start_reshape == NULL)
3723                         return -EINVAL;
3724                 err = mddev->pers->start_reshape(mddev);
3725                 if (err)
3726                         return err;
3727                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3728         } else {
3729                 if (cmd_match(page, "check"))
3730                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3731                 else if (!cmd_match(page, "repair"))
3732                         return -EINVAL;
3733                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3734                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3735         }
3736         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3737         md_wakeup_thread(mddev->thread);
3738         sysfs_notify_dirent(mddev->sysfs_action);
3739         return len;
3740 }
3741
3742 static ssize_t
3743 mismatch_cnt_show(mddev_t *mddev, char *page)
3744 {
3745         return sprintf(page, "%llu\n",
3746                        (unsigned long long) mddev->resync_mismatches);
3747 }
3748
3749 static struct md_sysfs_entry md_scan_mode =
3750 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3751
3752
3753 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3754
3755 static ssize_t
3756 sync_min_show(mddev_t *mddev, char *page)
3757 {
3758         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3759                        mddev->sync_speed_min ? "local": "system");
3760 }
3761
3762 static ssize_t
3763 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3764 {
3765         int min;
3766         char *e;
3767         if (strncmp(buf, "system", 6)==0) {
3768                 mddev->sync_speed_min = 0;
3769                 return len;
3770         }
3771         min = simple_strtoul(buf, &e, 10);
3772         if (buf == e || (*e && *e != '\n') || min <= 0)
3773                 return -EINVAL;
3774         mddev->sync_speed_min = min;
3775         return len;
3776 }
3777
3778 static struct md_sysfs_entry md_sync_min =
3779 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3780
3781 static ssize_t
3782 sync_max_show(mddev_t *mddev, char *page)
3783 {
3784         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3785                        mddev->sync_speed_max ? "local": "system");
3786 }
3787
3788 static ssize_t
3789 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3790 {
3791         int max;
3792         char *e;
3793         if (strncmp(buf, "system", 6)==0) {
3794                 mddev->sync_speed_max = 0;
3795                 return len;
3796         }
3797         max = simple_strtoul(buf, &e, 10);
3798         if (buf == e || (*e && *e != '\n') || max <= 0)
3799                 return -EINVAL;
3800         mddev->sync_speed_max = max;
3801         return len;
3802 }
3803
3804 static struct md_sysfs_entry md_sync_max =
3805 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3806
3807 static ssize_t
3808 degraded_show(mddev_t *mddev, char *page)
3809 {
3810         return sprintf(page, "%d\n", mddev->degraded);
3811 }
3812 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3813
3814 static ssize_t
3815 sync_force_parallel_show(mddev_t *mddev, char *page)
3816 {
3817         return sprintf(page, "%d\n", mddev->parallel_resync);
3818 }
3819
3820 static ssize_t
3821 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3822 {
3823         long n;
3824
3825         if (strict_strtol(buf, 10, &n))
3826                 return -EINVAL;
3827
3828         if (n != 0 && n != 1)
3829                 return -EINVAL;
3830
3831         mddev->parallel_resync = n;
3832
3833         if (mddev->sync_thread)
3834                 wake_up(&resync_wait);
3835
3836         return len;
3837 }
3838
3839 /* force parallel resync, even with shared block devices */
3840 static struct md_sysfs_entry md_sync_force_parallel =
3841 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3842        sync_force_parallel_show, sync_force_parallel_store);
3843
3844 static ssize_t
3845 sync_speed_show(mddev_t *mddev, char *page)
3846 {
3847         unsigned long resync, dt, db;
3848         if (mddev->curr_resync == 0)
3849                 return sprintf(page, "none\n");
3850         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3851         dt = (jiffies - mddev->resync_mark) / HZ;
3852         if (!dt) dt++;
3853         db = resync - mddev->resync_mark_cnt;
3854         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3855 }
3856
3857 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3858
3859 static ssize_t
3860 sync_completed_show(mddev_t *mddev, char *page)
3861 {
3862         unsigned long max_sectors, resync;
3863
3864         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3865                 return sprintf(page, "none\n");
3866
3867         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3868                 max_sectors = mddev->resync_max_sectors;
3869         else
3870                 max_sectors = mddev->dev_sectors;
3871
3872         resync = mddev->curr_resync_completed;
3873         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3874 }
3875
3876 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3877
3878 static ssize_t
3879 min_sync_show(mddev_t *mddev, char *page)
3880 {
3881         return sprintf(page, "%llu\n",
3882                        (unsigned long long)mddev->resync_min);
3883 }
3884 static ssize_t
3885 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3886 {
3887         unsigned long long min;
3888         if (strict_strtoull(buf, 10, &min))
3889                 return -EINVAL;
3890         if (min > mddev->resync_max)
3891                 return -EINVAL;
3892         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3893                 return -EBUSY;
3894
3895         /* Must be a multiple of chunk_size */
3896         if (mddev->chunk_sectors) {
3897                 sector_t temp = min;
3898                 if (sector_div(temp, mddev->chunk_sectors))
3899                         return -EINVAL;
3900         }
3901         mddev->resync_min = min;
3902
3903         return len;
3904 }
3905
3906 static struct md_sysfs_entry md_min_sync =
3907 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3908
3909 static ssize_t
3910 max_sync_show(mddev_t *mddev, char *page)
3911 {
3912         if (mddev->resync_max == MaxSector)
3913                 return sprintf(page, "max\n");
3914         else
3915                 return sprintf(page, "%llu\n",
3916                                (unsigned long long)mddev->resync_max);
3917 }
3918 static ssize_t
3919 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3920 {
3921         if (strncmp(buf, "max", 3) == 0)
3922                 mddev->resync_max = MaxSector;
3923         else {
3924                 unsigned long long max;
3925                 if (strict_strtoull(buf, 10, &max))
3926                         return -EINVAL;
3927                 if (max < mddev->resync_min)
3928                         return -EINVAL;
3929                 if (max < mddev->resync_max &&
3930                     mddev->ro == 0 &&
3931                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3932                         return -EBUSY;
3933
3934                 /* Must be a multiple of chunk_size */
3935                 if (mddev->chunk_sectors) {
3936                         sector_t temp = max;
3937                         if (sector_div(temp, mddev->chunk_sectors))
3938                                 return -EINVAL;
3939                 }
3940                 mddev->resync_max = max;
3941         }
3942         wake_up(&mddev->recovery_wait);
3943         return len;
3944 }
3945
3946 static struct md_sysfs_entry md_max_sync =
3947 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3948
3949 static ssize_t
3950 suspend_lo_show(mddev_t *mddev, char *page)
3951 {
3952         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3953 }
3954
3955 static ssize_t
3956 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3957 {
3958         char *e;
3959         unsigned long long new = simple_strtoull(buf, &e, 10);
3960
3961         if (mddev->pers == NULL ||
3962             mddev->pers->quiesce == NULL)
3963                 return -EINVAL;
3964         if (buf == e || (*e && *e != '\n'))
3965                 return -EINVAL;
3966         if (new >= mddev->suspend_hi ||
3967             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3968                 mddev->suspend_lo = new;
3969                 mddev->pers->quiesce(mddev, 2);
3970                 return len;
3971         } else
3972                 return -EINVAL;
3973 }
3974 static struct md_sysfs_entry md_suspend_lo =
3975 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3976
3977
3978 static ssize_t
3979 suspend_hi_show(mddev_t *mddev, char *page)
3980 {
3981         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3982 }
3983
3984 static ssize_t
3985 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3986 {
3987         char *e;
3988         unsigned long long new = simple_strtoull(buf, &e, 10);
3989
3990         if (mddev->pers == NULL ||
3991             mddev->pers->quiesce == NULL)
3992                 return -EINVAL;
3993         if (buf == e || (*e && *e != '\n'))
3994                 return -EINVAL;
3995         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3996             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3997                 mddev->suspend_hi = new;
3998                 mddev->pers->quiesce(mddev, 1);
3999                 mddev->pers->quiesce(mddev, 0);
4000                 return len;
4001         } else
4002                 return -EINVAL;
4003 }
4004 static struct md_sysfs_entry md_suspend_hi =
4005 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4006
4007 static ssize_t
4008 reshape_position_show(mddev_t *mddev, char *page)
4009 {
4010         if (mddev->reshape_position != MaxSector)
4011                 return sprintf(page, "%llu\n",
4012                                (unsigned long long)mddev->reshape_position);
4013         strcpy(page, "none\n");
4014         return 5;
4015 }
4016
4017 static ssize_t
4018 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4019 {
4020         char *e;
4021         unsigned long long new = simple_strtoull(buf, &e, 10);
4022         if (mddev->pers)
4023                 return -EBUSY;
4024         if (buf == e || (*e && *e != '\n'))
4025                 return -EINVAL;
4026         mddev->reshape_position = new;
4027         mddev->delta_disks = 0;
4028         mddev->new_level = mddev->level;
4029         mddev->new_layout = mddev->layout;
4030         mddev->new_chunk_sectors = mddev->chunk_sectors;
4031         return len;
4032 }
4033
4034 static struct md_sysfs_entry md_reshape_position =
4035 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4036        reshape_position_store);
4037
4038 static ssize_t
4039 array_size_show(mddev_t *mddev, char *page)
4040 {
4041         if (mddev->external_size)
4042                 return sprintf(page, "%llu\n",
4043                                (unsigned long long)mddev->array_sectors/2);
4044         else
4045                 return sprintf(page, "default\n");
4046 }
4047
4048 static ssize_t
4049 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4050 {
4051         sector_t sectors;
4052
4053         if (strncmp(buf, "default", 7) == 0) {
4054                 if (mddev->pers)
4055                         sectors = mddev->pers->size(mddev, 0, 0);
4056                 else
4057                         sectors = mddev->array_sectors;
4058
4059                 mddev->external_size = 0;
4060         } else {
4061                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4062                         return -EINVAL;
4063                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4064                         return -E2BIG;
4065
4066                 mddev->external_size = 1;
4067         }
4068
4069         mddev->array_sectors = sectors;
4070         set_capacity(mddev->gendisk, mddev->array_sectors);
4071         if (mddev->pers)
4072                 revalidate_disk(mddev->gendisk);
4073
4074         return len;
4075 }
4076
4077 static struct md_sysfs_entry md_array_size =
4078 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4079        array_size_store);
4080
4081 static struct attribute *md_default_attrs[] = {
4082         &md_level.attr,
4083         &md_layout.attr,
4084         &md_raid_disks.attr,
4085         &md_chunk_size.attr,
4086         &md_size.attr,
4087         &md_resync_start.attr,
4088         &md_metadata.attr,
4089         &md_new_device.attr,
4090         &md_safe_delay.attr,
4091         &md_array_state.attr,
4092         &md_reshape_position.attr,
4093         &md_array_size.attr,
4094         &max_corr_read_errors.attr,
4095         NULL,
4096 };
4097
4098 static struct attribute *md_redundancy_attrs[] = {
4099         &md_scan_mode.attr,
4100         &md_mismatches.attr,
4101         &md_sync_min.attr,
4102         &md_sync_max.attr,
4103         &md_sync_speed.attr,
4104         &md_sync_force_parallel.attr,
4105         &md_sync_completed.attr,
4106         &md_min_sync.attr,
4107         &md_max_sync.attr,
4108         &md_suspend_lo.attr,
4109         &md_suspend_hi.attr,
4110         &md_bitmap.attr,
4111         &md_degraded.attr,
4112         NULL,
4113 };
4114 static struct attribute_group md_redundancy_group = {
4115         .name = NULL,
4116         .attrs = md_redundancy_attrs,
4117 };
4118
4119
4120 static ssize_t
4121 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4122 {
4123         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4124         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4125         ssize_t rv;
4126
4127         if (!entry->show)
4128                 return -EIO;
4129         rv = mddev_lock(mddev);
4130         if (!rv) {
4131                 rv = entry->show(mddev, page);
4132                 mddev_unlock(mddev);
4133         }
4134         return rv;
4135 }
4136
4137 static ssize_t
4138 md_attr_store(struct kobject *kobj, struct attribute *attr,
4139               const char *page, size_t length)
4140 {
4141         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4142         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4143         ssize_t rv;
4144
4145         if (!entry->store)
4146                 return -EIO;
4147         if (!capable(CAP_SYS_ADMIN))
4148                 return -EACCES;
4149         rv = mddev_lock(mddev);
4150         if (mddev->hold_active == UNTIL_IOCTL)
4151                 mddev->hold_active = 0;
4152         if (!rv) {
4153                 rv = entry->store(mddev, page, length);
4154                 mddev_unlock(mddev);
4155         }
4156         return rv;
4157 }
4158
4159 static void md_free(struct kobject *ko)
4160 {
4161         mddev_t *mddev = container_of(ko, mddev_t, kobj);
4162
4163         if (mddev->sysfs_state)
4164                 sysfs_put(mddev->sysfs_state);
4165
4166         if (mddev->gendisk) {
4167                 del_gendisk(mddev->gendisk);
4168                 put_disk(mddev->gendisk);
4169         }
4170         if (mddev->queue)
4171                 blk_cleanup_queue(mddev->queue);
4172
4173         kfree(mddev);
4174 }
4175
4176 static const struct sysfs_ops md_sysfs_ops = {
4177         .show   = md_attr_show,
4178         .store  = md_attr_store,
4179 };
4180 static struct kobj_type md_ktype = {
4181         .release        = md_free,
4182         .sysfs_ops      = &md_sysfs_ops,
4183         .default_attrs  = md_default_attrs,
4184 };
4185
4186 int mdp_major = 0;
4187
4188 static void mddev_delayed_delete(struct work_struct *ws)
4189 {
4190         mddev_t *mddev = container_of(ws, mddev_t, del_work);
4191
4192         sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4193         kobject_del(&mddev->kobj);
4194         kobject_put(&mddev->kobj);
4195 }
4196
4197 static int md_alloc(dev_t dev, char *name)
4198 {
4199         static DEFINE_MUTEX(disks_mutex);
4200         mddev_t *mddev = mddev_find(dev);
4201         struct gendisk *disk;
4202         int partitioned;
4203         int shift;
4204         int unit;
4205         int error;
4206
4207         if (!mddev)
4208                 return -ENODEV;
4209
4210         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4211         shift = partitioned ? MdpMinorShift : 0;
4212         unit = MINOR(mddev->unit) >> shift;
4213
4214         /* wait for any previous instance if this device
4215          * to be completed removed (mddev_delayed_delete).
4216          */
4217         flush_scheduled_work();
4218
4219         mutex_lock(&disks_mutex);
4220         error = -EEXIST;
4221         if (mddev->gendisk)
4222                 goto abort;
4223
4224         if (name) {
4225                 /* Need to ensure that 'name' is not a duplicate.
4226                  */
4227                 mddev_t *mddev2;
4228                 spin_lock(&all_mddevs_lock);
4229
4230                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4231                         if (mddev2->gendisk &&
4232                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
4233                                 spin_unlock(&all_mddevs_lock);
4234                                 goto abort;
4235                         }
4236                 spin_unlock(&all_mddevs_lock);
4237         }
4238
4239         error = -ENOMEM;
4240         mddev->queue = blk_alloc_queue(GFP_KERNEL);
4241         if (!mddev->queue)
4242                 goto abort;
4243         mddev->queue->queuedata = mddev;
4244
4245         /* Can be unlocked because the queue is new: no concurrency */
4246         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4247
4248         blk_queue_make_request(mddev->queue, md_make_request);
4249
4250         disk = alloc_disk(1 << shift);
4251         if (!disk) {
4252                 blk_cleanup_queue(mddev->queue);
4253                 mddev->queue = NULL;
4254                 goto abort;
4255         }
4256         disk->major = MAJOR(mddev->unit);
4257         disk->first_minor = unit << shift;
4258         if (name)
4259                 strcpy(disk->disk_name, name);
4260         else if (partitioned)
4261                 sprintf(disk->disk_name, "md_d%d", unit);
4262         else
4263                 sprintf(disk->disk_name, "md%d", unit);
4264         disk->fops = &md_fops;
4265         disk->private_data = mddev;
4266         disk->queue = mddev->queue;
4267         /* Allow extended partitions.  This makes the
4268          * 'mdp' device redundant, but we can't really
4269          * remove it now.
4270          */
4271         disk->flags |= GENHD_FL_EXT_DEVT;
4272         add_disk(disk);
4273         mddev->gendisk = disk;
4274         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4275                                      &disk_to_dev(disk)->kobj, "%s", "md");
4276         if (error) {
4277                 /* This isn't possible, but as kobject_init_and_add is marked
4278                  * __must_check, we must do something with the result
4279                  */
4280                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4281                        disk->disk_name);
4282                 error = 0;
4283         }
4284         if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4285                 printk(KERN_DEBUG "pointless warning\n");
4286  abort:
4287         mutex_unlock(&disks_mutex);
4288         if (!error) {
4289                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4290                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, NULL, "array_state");
4291         }
4292         mddev_put(mddev);
4293         return error;
4294 }
4295
4296 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4297 {
4298         md_alloc(dev, NULL);
4299         return NULL;
4300 }
4301
4302 static int add_named_array(const char *val, struct kernel_param *kp)
4303 {
4304         /* val must be "md_*" where * is not all digits.
4305          * We allocate an array with a large free minor number, and
4306          * set the name to val.  val must not already be an active name.
4307          */
4308         int len = strlen(val);
4309         char buf[DISK_NAME_LEN];
4310
4311         while (len && val[len-1] == '\n')
4312                 len--;
4313         if (len >= DISK_NAME_LEN)
4314                 return -E2BIG;
4315         strlcpy(buf, val, len+1);
4316         if (strncmp(buf, "md_", 3) != 0)
4317                 return -EINVAL;
4318         return md_alloc(0, buf);
4319 }
4320
4321 static void md_safemode_timeout(unsigned long data)
4322 {
4323         mddev_t *mddev = (mddev_t *) data;
4324
4325         if (!atomic_read(&mddev->writes_pending)) {
4326                 mddev->safemode = 1;
4327                 if (mddev->external)
4328                         sysfs_notify_dirent(mddev->sysfs_state);
4329         }
4330         md_wakeup_thread(mddev->thread);
4331 }
4332
4333 static int start_dirty_degraded;
4334
4335 static int md_run(mddev_t *mddev)
4336 {
4337         int err;
4338         mdk_rdev_t *rdev;
4339         struct mdk_personality *pers;
4340
4341         if (list_empty(&mddev->disks))
4342                 /* cannot run an array with no devices.. */
4343                 return -EINVAL;
4344
4345         if (mddev->pers)
4346                 return -EBUSY;
4347
4348         /* These two calls synchronise us with the
4349          * sysfs_remove_group calls in mddev_unlock,
4350          * so they must have completed.
4351          */
4352         mutex_lock(&mddev->open_mutex);
4353         mutex_unlock(&mddev->open_mutex);
4354
4355         /*
4356          * Analyze all RAID superblock(s)
4357          */
4358         if (!mddev->raid_disks) {
4359                 if (!mddev->persistent)
4360                         return -EINVAL;
4361                 analyze_sbs(mddev);
4362         }
4363
4364         if (mddev->level != LEVEL_NONE)
4365                 request_module("md-level-%d", mddev->level);
4366         else if (mddev->clevel[0])
4367                 request_module("md-%s", mddev->clevel);
4368
4369         /*
4370          * Drop all container device buffers, from now on
4371          * the only valid external interface is through the md
4372          * device.
4373          */
4374         list_for_each_entry(rdev, &mddev->disks, same_set) {
4375                 if (test_bit(Faulty, &rdev->flags))
4376                         continue;
4377                 sync_blockdev(rdev->bdev);
4378                 invalidate_bdev(rdev->bdev);
4379
4380                 /* perform some consistency tests on the device.
4381                  * We don't want the data to overlap the metadata,
4382                  * Internal Bitmap issues have been handled elsewhere.
4383                  */
4384                 if (rdev->data_offset < rdev->sb_start) {
4385                         if (mddev->dev_sectors &&
4386                             rdev->data_offset + mddev->dev_sectors
4387                             > rdev->sb_start) {
4388                                 printk("md: %s: data overlaps metadata\n",
4389                                        mdname(mddev));
4390                                 return -EINVAL;
4391                         }
4392                 } else {
4393                         if (rdev->sb_start + rdev->sb_size/512
4394                             > rdev->data_offset) {
4395                                 printk("md: %s: metadata overlaps data\n",
4396                                        mdname(mddev));
4397                                 return -EINVAL;
4398                         }
4399                 }
4400                 sysfs_notify_dirent(rdev->sysfs_state);
4401         }
4402
4403         spin_lock(&pers_lock);
4404         pers = find_pers(mddev->level, mddev->clevel);
4405         if (!pers || !try_module_get(pers->owner)) {
4406                 spin_unlock(&pers_lock);
4407                 if (mddev->level != LEVEL_NONE)
4408                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4409                                mddev->level);
4410                 else
4411                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4412                                mddev->clevel);
4413                 return -EINVAL;
4414         }
4415         mddev->pers = pers;
4416         spin_unlock(&pers_lock);
4417         if (mddev->level != pers->level) {
4418                 mddev->level = pers->level;
4419                 mddev->new_level = pers->level;
4420         }
4421         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4422
4423         if (mddev->reshape_position != MaxSector &&
4424             pers->start_reshape == NULL) {
4425                 /* This personality cannot handle reshaping... */
4426                 mddev->pers = NULL;
4427                 module_put(pers->owner);
4428                 return -EINVAL;
4429         }
4430
4431         if (pers->sync_request) {
4432                 /* Warn if this is a potentially silly
4433                  * configuration.
4434                  */
4435                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4436                 mdk_rdev_t *rdev2;
4437                 int warned = 0;
4438
4439                 list_for_each_entry(rdev, &mddev->disks, same_set)
4440                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4441                                 if (rdev < rdev2 &&
4442                                     rdev->bdev->bd_contains ==
4443                                     rdev2->bdev->bd_contains) {
4444                                         printk(KERN_WARNING
4445                                                "%s: WARNING: %s appears to be"
4446                                                " on the same physical disk as"
4447                                                " %s.\n",
4448                                                mdname(mddev),
4449                                                bdevname(rdev->bdev,b),
4450                                                bdevname(rdev2->bdev,b2));
4451                                         warned = 1;
4452                                 }
4453                         }
4454
4455                 if (warned)
4456                         printk(KERN_WARNING
4457                                "True protection against single-disk"
4458                                " failure might be compromised.\n");
4459         }
4460
4461         mddev->recovery = 0;
4462         /* may be over-ridden by personality */
4463         mddev->resync_max_sectors = mddev->dev_sectors;
4464
4465         mddev->barriers_work = 1;
4466         mddev->ok_start_degraded = start_dirty_degraded;
4467
4468         if (start_readonly && mddev->ro == 0)
4469                 mddev->ro = 2; /* read-only, but switch on first write */
4470
4471         err = mddev->pers->run(mddev);
4472         if (err)
4473                 printk(KERN_ERR "md: pers->run() failed ...\n");
4474         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4475                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4476                           " but 'external_size' not in effect?\n", __func__);
4477                 printk(KERN_ERR
4478                        "md: invalid array_size %llu > default size %llu\n",
4479                        (unsigned long long)mddev->array_sectors / 2,
4480                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4481                 err = -EINVAL;
4482                 mddev->pers->stop(mddev);
4483         }
4484         if (err == 0 && mddev->pers->sync_request) {
4485                 err = bitmap_create(mddev);
4486                 if (err) {
4487                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4488                                mdname(mddev), err);
4489                         mddev->pers->stop(mddev);
4490                 }
4491         }
4492         if (err) {
4493                 module_put(mddev->pers->owner);
4494                 mddev->pers = NULL;
4495                 bitmap_destroy(mddev);
4496                 return err;
4497         }
4498         if (mddev->pers->sync_request) {
4499                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4500                         printk(KERN_WARNING
4501                                "md: cannot register extra attributes for %s\n",
4502                                mdname(mddev));
4503                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
4504         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4505                 mddev->ro = 0;
4506
4507         atomic_set(&mddev->writes_pending,0);
4508         atomic_set(&mddev->max_corr_read_errors,
4509                    MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4510         mddev->safemode = 0;
4511         mddev->safemode_timer.function = md_safemode_timeout;
4512         mddev->safemode_timer.data = (unsigned long) mddev;
4513         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4514         mddev->in_sync = 1;
4515
4516         list_for_each_entry(rdev, &mddev->disks, same_set)
4517                 if (rdev->raid_disk >= 0) {
4518                         char nm[20];
4519                         sprintf(nm, "rd%d", rdev->raid_disk);
4520                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4521                                 printk("md: cannot register %s for %s\n",
4522                                        nm, mdname(mddev));
4523                 }
4524
4525         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4526
4527         if (mddev->flags)
4528                 md_update_sb(mddev, 0);
4529
4530         md_wakeup_thread(mddev->thread);
4531         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4532
4533         md_new_event(mddev);
4534         sysfs_notify_dirent(mddev->sysfs_state);
4535         if (mddev->sysfs_action)
4536                 sysfs_notify_dirent(mddev->sysfs_action);
4537         sysfs_notify(&mddev->kobj, NULL, "degraded");
4538         return 0;
4539 }
4540
4541 static int do_md_run(mddev_t *mddev)
4542 {
4543         int err;
4544
4545         err = md_run(mddev);
4546         if (err)
4547                 goto out;
4548
4549         set_capacity(mddev->gendisk, mddev->array_sectors);
4550         revalidate_disk(mddev->gendisk);
4551         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4552 out:
4553         return err;
4554 }
4555
4556 static int restart_array(mddev_t *mddev)
4557 {
4558         struct gendisk *disk = mddev->gendisk;
4559
4560         /* Complain if it has no devices */
4561         if (list_empty(&mddev->disks))
4562                 return -ENXIO;
4563         if (!mddev->pers)
4564                 return -EINVAL;
4565         if (!mddev->ro)
4566                 return -EBUSY;
4567         mddev->safemode = 0;
4568         mddev->ro = 0;
4569         set_disk_ro(disk, 0);
4570         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4571                 mdname(mddev));
4572         /* Kick recovery or resync if necessary */
4573         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4574         md_wakeup_thread(mddev->thread);
4575         md_wakeup_thread(mddev->sync_thread);
4576         sysfs_notify_dirent(mddev->sysfs_state);
4577         return 0;
4578 }
4579
4580 /* similar to deny_write_access, but accounts for our holding a reference
4581  * to the file ourselves */
4582 static int deny_bitmap_write_access(struct file * file)
4583 {
4584         struct inode *inode = file->f_mapping->host;
4585
4586         spin_lock(&inode->i_lock);
4587         if (atomic_read(&inode->i_writecount) > 1) {
4588                 spin_unlock(&inode->i_lock);
4589                 return -ETXTBSY;
4590         }
4591         atomic_set(&inode->i_writecount, -1);
4592         spin_unlock(&inode->i_lock);
4593
4594         return 0;
4595 }
4596
4597 void restore_bitmap_write_access(struct file *file)
4598 {
4599         struct inode *inode = file->f_mapping->host;
4600
4601         spin_lock(&inode->i_lock);
4602         atomic_set(&inode->i_writecount, 1);
4603         spin_unlock(&inode->i_lock);
4604 }
4605
4606 static void md_clean(mddev_t *mddev)
4607 {
4608         mddev->array_sectors = 0;
4609         mddev->external_size = 0;
4610         mddev->dev_sectors = 0;
4611         mddev->raid_disks = 0;
4612         mddev->recovery_cp = 0;
4613         mddev->resync_min = 0;
4614         mddev->resync_max = MaxSector;
4615         mddev->reshape_position = MaxSector;
4616         mddev->external = 0;
4617         mddev->persistent = 0;
4618         mddev->level = LEVEL_NONE;
4619         mddev->clevel[0] = 0;
4620         mddev->flags = 0;
4621         mddev->ro = 0;
4622         mddev->metadata_type[0] = 0;
4623         mddev->chunk_sectors = 0;
4624         mddev->ctime = mddev->utime = 0;
4625         mddev->layout = 0;
4626         mddev->max_disks = 0;
4627         mddev->events = 0;
4628         mddev->can_decrease_events = 0;
4629         mddev->delta_disks = 0;
4630         mddev->new_level = LEVEL_NONE;
4631         mddev->new_layout = 0;
4632         mddev->new_chunk_sectors = 0;
4633         mddev->curr_resync = 0;
4634         mddev->resync_mismatches = 0;
4635         mddev->suspend_lo = mddev->suspend_hi = 0;
4636         mddev->sync_speed_min = mddev->sync_speed_max = 0;
4637         mddev->recovery = 0;
4638         mddev->in_sync = 0;
4639         mddev->degraded = 0;
4640         mddev->barriers_work = 0;
4641         mddev->safemode = 0;
4642         mddev->bitmap_info.offset = 0;
4643         mddev->bitmap_info.default_offset = 0;
4644         mddev->bitmap_info.chunksize = 0;
4645         mddev->bitmap_info.daemon_sleep = 0;
4646         mddev->bitmap_info.max_write_behind = 0;
4647 }
4648
4649 static void md_stop_writes(mddev_t *mddev)
4650 {
4651         if (mddev->sync_thread) {
4652                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4653                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4654                 md_unregister_thread(mddev->sync_thread);
4655                 mddev->sync_thread = NULL;
4656         }
4657
4658         del_timer_sync(&mddev->safemode_timer);
4659
4660         bitmap_flush(mddev);
4661         md_super_wait(mddev);
4662
4663         if (!mddev->in_sync || mddev->flags) {
4664                 /* mark array as shutdown cleanly */
4665                 mddev->in_sync = 1;
4666                 md_update_sb(mddev, 1);
4667         }
4668 }
4669
4670 static void md_stop(mddev_t *mddev)
4671 {
4672         md_stop_writes(mddev);
4673
4674         mddev->pers->stop(mddev);
4675         if (mddev->pers->sync_request && mddev->to_remove == NULL)
4676                 mddev->to_remove = &md_redundancy_group;
4677         module_put(mddev->pers->owner);
4678         mddev->pers = NULL;
4679         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4680 }
4681
4682 static int md_set_readonly(mddev_t *mddev, int is_open)
4683 {
4684         int err = 0;
4685         mutex_lock(&mddev->open_mutex);
4686         if (atomic_read(&mddev->openers) > is_open) {
4687                 printk("md: %s still in use.\n",mdname(mddev));
4688                 err = -EBUSY;
4689                 goto out;
4690         }
4691         if (mddev->pers) {
4692                 md_stop_writes(mddev);
4693
4694                 err  = -ENXIO;
4695                 if (mddev->ro==1)
4696                         goto out;
4697                 mddev->ro = 1;
4698                 set_disk_ro(mddev->gendisk, 1);
4699                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4700                 sysfs_notify_dirent(mddev->sysfs_state);
4701                 err = 0;
4702         }
4703 out:
4704         mutex_unlock(&mddev->open_mutex);
4705         return err;
4706 }
4707
4708 /* mode:
4709  *   0 - completely stop and dis-assemble array
4710  *   2 - stop but do not disassemble array
4711  */
4712 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4713 {
4714         int err = 0;
4715         struct gendisk *disk = mddev->gendisk;
4716         mdk_rdev_t *rdev;
4717
4718         mutex_lock(&mddev->open_mutex);
4719         if (atomic_read(&mddev->openers) > is_open) {
4720                 printk("md: %s still in use.\n",mdname(mddev));
4721                 err = -EBUSY;
4722         } else if (mddev->pers) {
4723
4724                 if (mddev->ro)
4725                         set_disk_ro(disk, 0);
4726
4727                 md_stop(mddev);
4728                 mddev->queue->merge_bvec_fn = NULL;
4729                 mddev->queue->unplug_fn = NULL;
4730                 mddev->queue->backing_dev_info.congested_fn = NULL;
4731
4732                 /* tell userspace to handle 'inactive' */
4733                 sysfs_notify_dirent(mddev->sysfs_state);
4734
4735                 list_for_each_entry(rdev, &mddev->disks, same_set)
4736                         if (rdev->raid_disk >= 0) {
4737                                 char nm[20];
4738                                 sprintf(nm, "rd%d", rdev->raid_disk);
4739                                 sysfs_remove_link(&mddev->kobj, nm);
4740                         }
4741
4742                 set_capacity(disk, 0);
4743                 revalidate_disk(disk);
4744
4745                 if (mddev->ro)
4746                         mddev->ro = 0;
4747
4748                 err = 0;
4749         }
4750         mutex_unlock(&mddev->open_mutex);
4751         if (err)
4752                 return err;
4753         /*
4754          * Free resources if final stop
4755          */
4756         if (mode == 0) {
4757
4758                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4759
4760                 bitmap_destroy(mddev);
4761                 if (mddev->bitmap_info.file) {
4762                         restore_bitmap_write_access(mddev->bitmap_info.file);
4763                         fput(mddev->bitmap_info.file);
4764                         mddev->bitmap_info.file = NULL;
4765                 }
4766                 mddev->bitmap_info.offset = 0;
4767
4768                 export_array(mddev);
4769
4770                 md_clean(mddev);
4771                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4772                 if (mddev->hold_active == UNTIL_STOP)
4773                         mddev->hold_active = 0;
4774
4775         }
4776         err = 0;
4777         blk_integrity_unregister(disk);
4778         md_new_event(mddev);
4779         sysfs_notify_dirent(mddev->sysfs_state);
4780         return err;
4781 }
4782
4783 #ifndef MODULE
4784 static void autorun_array(mddev_t *mddev)
4785 {
4786         mdk_rdev_t *rdev;
4787         int err;
4788
4789         if (list_empty(&mddev->disks))
4790                 return;
4791
4792         printk(KERN_INFO "md: running: ");
4793
4794         list_for_each_entry(rdev, &mddev->disks, same_set) {
4795                 char b[BDEVNAME_SIZE];
4796                 printk("<%s>", bdevname(rdev->bdev,b));
4797         }
4798         printk("\n");
4799
4800         err = do_md_run(mddev);
4801         if (err) {
4802                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4803                 do_md_stop(mddev, 0, 0);
4804         }
4805 }
4806
4807 /*
4808  * lets try to run arrays based on all disks that have arrived
4809  * until now. (those are in pending_raid_disks)
4810  *
4811  * the method: pick the first pending disk, collect all disks with
4812  * the same UUID, remove all from the pending list and put them into
4813  * the 'same_array' list. Then order this list based on superblock
4814  * update time (freshest comes first), kick out 'old' disks and
4815  * compare superblocks. If everything's fine then run it.
4816  *
4817  * If "unit" is allocated, then bump its reference count
4818  */
4819 static void autorun_devices(int part)
4820 {
4821         mdk_rdev_t *rdev0, *rdev, *tmp;
4822         mddev_t *mddev;
4823         char b[BDEVNAME_SIZE];
4824
4825         printk(KERN_INFO "md: autorun ...\n");
4826         while (!list_empty(&pending_raid_disks)) {
4827                 int unit;
4828                 dev_t dev;
4829                 LIST_HEAD(candidates);
4830                 rdev0 = list_entry(pending_raid_disks.next,
4831                                          mdk_rdev_t, same_set);
4832
4833                 printk(KERN_INFO "md: considering %s ...\n",
4834                         bdevname(rdev0->bdev,b));
4835                 INIT_LIST_HEAD(&candidates);
4836                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4837                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4838                                 printk(KERN_INFO "md:  adding %s ...\n",
4839                                         bdevname(rdev->bdev,b));
4840                                 list_move(&rdev->same_set, &candidates);
4841                         }
4842                 /*
4843                  * now we have a set of devices, with all of them having
4844                  * mostly sane superblocks. It's time to allocate the
4845                  * mddev.
4846                  */
4847                 if (part) {
4848                         dev = MKDEV(mdp_major,
4849                                     rdev0->preferred_minor << MdpMinorShift);
4850                         unit = MINOR(dev) >> MdpMinorShift;
4851                 } else {
4852                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4853                         unit = MINOR(dev);
4854                 }
4855                 if (rdev0->preferred_minor != unit) {
4856                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4857                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4858                         break;
4859                 }
4860
4861                 md_probe(dev, NULL, NULL);
4862                 mddev = mddev_find(dev);
4863                 if (!mddev || !mddev->gendisk) {
4864                         if (mddev)
4865                                 mddev_put(mddev);
4866                         printk(KERN_ERR
4867                                 "md: cannot allocate memory for md drive.\n");
4868                         break;
4869                 }
4870                 if (mddev_lock(mddev))
4871                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4872                                mdname(mddev));
4873                 else if (mddev->raid_disks || mddev->major_version
4874                          || !list_empty(&mddev->disks)) {
4875                         printk(KERN_WARNING
4876                                 "md: %s already running, cannot run %s\n",
4877                                 mdname(mddev), bdevname(rdev0->bdev,b));
4878                         mddev_unlock(mddev);
4879                 } else {
4880                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4881                         mddev->persistent = 1;
4882                         rdev_for_each_list(rdev, tmp, &candidates) {
4883                                 list_del_init(&rdev->same_set);
4884                                 if (bind_rdev_to_array(rdev, mddev))
4885                                         export_rdev(rdev);
4886                         }
4887                         autorun_array(mddev);
4888                         mddev_unlock(mddev);
4889                 }
4890                 /* on success, candidates will be empty, on error
4891                  * it won't...
4892                  */
4893                 rdev_for_each_list(rdev, tmp, &candidates) {
4894                         list_del_init(&rdev->same_set);
4895                         export_rdev(rdev);
4896                 }
4897                 mddev_put(mddev);
4898         }
4899         printk(KERN_INFO "md: ... autorun DONE.\n");
4900 }
4901 #endif /* !MODULE */
4902
4903 static int get_version(void __user * arg)
4904 {
4905         mdu_version_t ver;
4906
4907         ver.major = MD_MAJOR_VERSION;
4908         ver.minor = MD_MINOR_VERSION;
4909         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4910
4911         if (copy_to_user(arg, &ver, sizeof(ver)))
4912                 return -EFAULT;
4913
4914         return 0;
4915 }
4916
4917 static int get_array_info(mddev_t * mddev, void __user * arg)
4918 {
4919         mdu_array_info_t info;
4920         int nr,working,insync,failed,spare;
4921         mdk_rdev_t *rdev;
4922
4923         nr=working=insync=failed=spare=0;
4924         list_for_each_entry(rdev, &mddev->disks, same_set) {
4925                 nr++;
4926                 if (test_bit(Faulty, &rdev->flags))
4927                         failed++;
4928                 else {
4929                         working++;
4930                         if (test_bit(In_sync, &rdev->flags))
4931                                 insync++;
4932                         else
4933                                 spare++;
4934                 }
4935         }
4936
4937         info.major_version = mddev->major_version;
4938         info.minor_version = mddev->minor_version;
4939         info.patch_version = MD_PATCHLEVEL_VERSION;
4940         info.ctime         = mddev->ctime;
4941         info.level         = mddev->level;
4942         info.size          = mddev->dev_sectors / 2;
4943         if (info.size != mddev->dev_sectors / 2) /* overflow */
4944                 info.size = -1;
4945         info.nr_disks      = nr;
4946         info.raid_disks    = mddev->raid_disks;
4947         info.md_minor      = mddev->md_minor;
4948         info.not_persistent= !mddev->persistent;
4949
4950         info.utime         = mddev->utime;
4951         info.state         = 0;
4952         if (mddev->in_sync)
4953                 info.state = (1<<MD_SB_CLEAN);
4954         if (mddev->bitmap && mddev->bitmap_info.offset)
4955                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4956         info.active_disks  = insync;
4957         info.working_disks = working;
4958         info.failed_disks  = failed;
4959         info.spare_disks   = spare;
4960
4961         info.layout        = mddev->layout;
4962         info.chunk_size    = mddev->chunk_sectors << 9;
4963
4964         if (copy_to_user(arg, &info, sizeof(info)))
4965                 return -EFAULT;
4966
4967         return 0;
4968 }
4969
4970 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4971 {
4972         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4973         char *ptr, *buf = NULL;
4974         int err = -ENOMEM;
4975
4976         if (md_allow_write(mddev))
4977                 file = kmalloc(sizeof(*file), GFP_NOIO);
4978         else
4979                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4980
4981         if (!file)
4982                 goto out;
4983
4984         /* bitmap disabled, zero the first byte and copy out */
4985         if (!mddev->bitmap || !mddev->bitmap->file) {
4986                 file->pathname[0] = '\0';
4987                 goto copy_out;
4988         }
4989
4990         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4991         if (!buf)
4992                 goto out;
4993
4994         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4995         if (IS_ERR(ptr))
4996                 goto out;
4997
4998         strcpy(file->pathname, ptr);
4999
5000 copy_out:
5001         err = 0;
5002         if (copy_to_user(arg, file, sizeof(*file)))
5003                 err = -EFAULT;
5004 out:
5005         kfree(buf);
5006         kfree(file);
5007         return err;
5008 }
5009
5010 static int get_disk_info(mddev_t * mddev, void __user * arg)
5011 {
5012         mdu_disk_info_t info;
5013         mdk_rdev_t *rdev;
5014
5015         if (copy_from_user(&info, arg, sizeof(info)))
5016                 return -EFAULT;
5017
5018         rdev = find_rdev_nr(mddev, info.number);
5019         if (rdev) {
5020                 info.major = MAJOR(rdev->bdev->bd_dev);
5021                 info.minor = MINOR(rdev->bdev->bd_dev);
5022                 info.raid_disk = rdev->raid_disk;
5023                 info.state = 0;
5024                 if (test_bit(Faulty, &rdev->flags))
5025                         info.state |= (1<<MD_DISK_FAULTY);
5026                 else if (test_bit(In_sync, &rdev->flags)) {
5027                         info.state |= (1<<MD_DISK_ACTIVE);
5028                         info.state |= (1<<MD_DISK_SYNC);
5029                 }
5030                 if (test_bit(WriteMostly, &rdev->flags))
5031                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
5032         } else {
5033                 info.major = info.minor = 0;
5034                 info.raid_disk = -1;
5035                 info.state = (1<<MD_DISK_REMOVED);
5036         }
5037
5038         if (copy_to_user(arg, &info, sizeof(info)))
5039                 return -EFAULT;
5040
5041         return 0;
5042 }
5043
5044 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5045 {
5046         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5047         mdk_rdev_t *rdev;
5048         dev_t dev = MKDEV(info->major,info->minor);
5049
5050         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5051                 return -EOVERFLOW;
5052
5053         if (!mddev->raid_disks) {
5054                 int err;
5055                 /* expecting a device which has a superblock */
5056                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5057                 if (IS_ERR(rdev)) {
5058                         printk(KERN_WARNING
5059                                 "md: md_import_device returned %ld\n",
5060                                 PTR_ERR(rdev));
5061                         return PTR_ERR(rdev);
5062                 }
5063                 if (!list_empty(&mddev->disks)) {
5064                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5065                                                         mdk_rdev_t, same_set);
5066                         err = super_types[mddev->major_version]
5067                                 .load_super(rdev, rdev0, mddev->minor_version);
5068                         if (err < 0) {
5069                                 printk(KERN_WARNING
5070                                         "md: %s has different UUID to %s\n",
5071                                         bdevname(rdev->bdev,b),
5072                                         bdevname(rdev0->bdev,b2));
5073                                 export_rdev(rdev);
5074                                 return -EINVAL;
5075                         }
5076                 }
5077                 err = bind_rdev_to_array(rdev, mddev);
5078                 if (err)
5079                         export_rdev(rdev);
5080                 return err;
5081         }
5082
5083         /*
5084          * add_new_disk can be used once the array is assembled
5085          * to add "hot spares".  They must already have a superblock
5086          * written
5087          */
5088         if (mddev->pers) {
5089                 int err;
5090                 if (!mddev->pers->hot_add_disk) {
5091                         printk(KERN_WARNING
5092                                 "%s: personality does not support diskops!\n",
5093                                mdname(mddev));
5094                         return -EINVAL;
5095                 }
5096                 if (mddev->persistent)
5097                         rdev = md_import_device(dev, mddev->major_version,
5098                                                 mddev->minor_version);
5099                 else
5100                         rdev = md_import_device(dev, -1, -1);
5101                 if (IS_ERR(rdev)) {
5102                         printk(KERN_WARNING
5103                                 "md: md_import_device returned %ld\n",
5104                                 PTR_ERR(rdev));
5105                         return PTR_ERR(rdev);
5106                 }
5107                 /* set save_raid_disk if appropriate */
5108                 if (!mddev->persistent) {
5109                         if (info->state & (1<<MD_DISK_SYNC)  &&
5110                             info->raid_disk < mddev->raid_disks)
5111                                 rdev->raid_disk = info->raid_disk;
5112                         else
5113                                 rdev->raid_disk = -1;
5114                 } else
5115                         super_types[mddev->major_version].
5116                                 validate_super(mddev, rdev);
5117                 rdev->saved_raid_disk = rdev->raid_disk;
5118
5119                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5120                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5121                         set_bit(WriteMostly, &rdev->flags);
5122                 else
5123                         clear_bit(WriteMostly, &rdev->flags);
5124
5125                 rdev->raid_disk = -1;
5126                 err = bind_rdev_to_array(rdev, mddev);
5127                 if (!err && !mddev->pers->hot_remove_disk) {
5128                         /* If there is hot_add_disk but no hot_remove_disk
5129                          * then added disks for geometry changes,
5130                          * and should be added immediately.
5131                          */
5132                         super_types[mddev->major_version].
5133                                 validate_super(mddev, rdev);
5134                         err = mddev->pers->hot_add_disk(mddev, rdev);
5135                         if (err)
5136                                 unbind_rdev_from_array(rdev);
5137                 }
5138                 if (err)
5139                         export_rdev(rdev);
5140                 else
5141                         sysfs_notify_dirent(rdev->sysfs_state);
5142
5143                 md_update_sb(mddev, 1);
5144                 if (mddev->degraded)
5145                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5146                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5147                 md_wakeup_thread(mddev->thread);
5148                 return err;
5149         }
5150
5151         /* otherwise, add_new_disk is only allowed
5152          * for major_version==0 superblocks
5153          */
5154         if (mddev->major_version != 0) {
5155                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5156                        mdname(mddev));
5157                 return -EINVAL;
5158         }
5159
5160         if (!(info->state & (1<<MD_DISK_FAULTY))) {
5161                 int err;
5162                 rdev = md_import_device(dev, -1, 0);
5163                 if (IS_ERR(rdev)) {
5164                         printk(KERN_WARNING
5165                                 "md: error, md_import_device() returned %ld\n",
5166                                 PTR_ERR(rdev));
5167                         return PTR_ERR(rdev);
5168                 }
5169                 rdev->desc_nr = info->number;
5170                 if (info->raid_disk < mddev->raid_disks)
5171                         rdev->raid_disk = info->raid_disk;
5172                 else
5173                         rdev->raid_disk = -1;
5174
5175                 if (rdev->raid_disk < mddev->raid_disks)
5176                         if (info->state & (1<<MD_DISK_SYNC))
5177                                 set_bit(In_sync, &rdev->flags);
5178
5179                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5180                         set_bit(WriteMostly, &rdev->flags);
5181
5182                 if (!mddev->persistent) {
5183                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
5184                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5185                 } else
5186                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5187                 rdev->sectors = rdev->sb_start;
5188
5189                 err = bind_rdev_to_array(rdev, mddev);
5190                 if (err) {
5191                         export_rdev(rdev);
5192                         return err;
5193                 }
5194         }
5195
5196         return 0;
5197 }
5198
5199 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5200 {
5201         char b[BDEVNAME_SIZE];
5202         mdk_rdev_t *rdev;
5203
5204         rdev = find_rdev(mddev, dev);
5205         if (!rdev)
5206                 return -ENXIO;
5207
5208         if (rdev->raid_disk >= 0)
5209                 goto busy;
5210
5211         kick_rdev_from_array(rdev);
5212         md_update_sb(mddev, 1);
5213         md_new_event(mddev);
5214
5215         return 0;
5216 busy:
5217         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5218                 bdevname(rdev->bdev,b), mdname(mddev));
5219         return -EBUSY;
5220 }
5221
5222 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5223 {
5224         char b[BDEVNAME_SIZE];
5225         int err;
5226         mdk_rdev_t *rdev;
5227
5228         if (!mddev->pers)
5229                 return -ENODEV;
5230
5231         if (mddev->major_version != 0) {
5232                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5233                         " version-0 superblocks.\n",
5234                         mdname(mddev));
5235                 return -EINVAL;
5236         }
5237         if (!mddev->pers->hot_add_disk) {
5238                 printk(KERN_WARNING
5239                         "%s: personality does not support diskops!\n",
5240                         mdname(mddev));
5241                 return -EINVAL;
5242         }
5243
5244         rdev = md_import_device(dev, -1, 0);
5245         if (IS_ERR(rdev)) {
5246                 printk(KERN_WARNING
5247                         "md: error, md_import_device() returned %ld\n",
5248                         PTR_ERR(rdev));
5249                 return -EINVAL;
5250         }
5251
5252         if (mddev->persistent)
5253                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5254         else
5255                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5256
5257         rdev->sectors = rdev->sb_start;
5258
5259         if (test_bit(Faulty, &rdev->flags)) {
5260                 printk(KERN_WARNING
5261                         "md: can not hot-add faulty %s disk to %s!\n",
5262                         bdevname(rdev->bdev,b), mdname(mddev));
5263                 err = -EINVAL;
5264                 goto abort_export;
5265         }
5266         clear_bit(In_sync, &rdev->flags);
5267         rdev->desc_nr = -1;
5268         rdev->saved_raid_disk = -1;
5269         err = bind_rdev_to_array(rdev, mddev);
5270         if (err)
5271                 goto abort_export;
5272
5273         /*
5274          * The rest should better be atomic, we can have disk failures
5275          * noticed in interrupt contexts ...
5276          */
5277
5278         rdev->raid_disk = -1;
5279
5280         md_update_sb(mddev, 1);
5281
5282         /*
5283          * Kick recovery, maybe this spare has to be added to the
5284          * array immediately.
5285          */
5286         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5287         md_wakeup_thread(mddev->thread);
5288         md_new_event(mddev);
5289         return 0;
5290
5291 abort_export:
5292         export_rdev(rdev);
5293         return err;
5294 }
5295
5296 static int set_bitmap_file(mddev_t *mddev, int fd)
5297 {
5298         int err;
5299
5300         if (mddev->pers) {
5301                 if (!mddev->pers->quiesce)
5302                         return -EBUSY;
5303                 if (mddev->recovery || mddev->sync_thread)
5304                         return -EBUSY;
5305                 /* we should be able to change the bitmap.. */
5306         }
5307
5308
5309         if (fd >= 0) {
5310                 if (mddev->bitmap)
5311                         return -EEXIST; /* cannot add when bitmap is present */
5312                 mddev->bitmap_info.file = fget(fd);
5313
5314                 if (mddev->bitmap_info.file == NULL) {
5315                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5316                                mdname(mddev));
5317                         return -EBADF;
5318                 }
5319
5320                 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5321                 if (err) {
5322                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5323                                mdname(mddev));
5324                         fput(mddev->bitmap_info.file);
5325                         mddev->bitmap_info.file = NULL;
5326                         return err;
5327                 }
5328                 mddev->bitmap_info.offset = 0; /* file overrides offset */
5329         } else if (mddev->bitmap == NULL)
5330                 return -ENOENT; /* cannot remove what isn't there */
5331         err = 0;
5332         if (mddev->pers) {
5333                 mddev->pers->quiesce(mddev, 1);
5334                 if (fd >= 0)
5335                         err = bitmap_create(mddev);
5336                 if (fd < 0 || err) {
5337                         bitmap_destroy(mddev);
5338                         fd = -1; /* make sure to put the file */
5339                 }
5340                 mddev->pers->quiesce(mddev, 0);
5341         }
5342         if (fd < 0) {
5343                 if (mddev->bitmap_info.file) {
5344                         restore_bitmap_write_access(mddev->bitmap_info.file);
5345                         fput(mddev->bitmap_info.file);
5346                 }
5347                 mddev->bitmap_info.file = NULL;
5348         }
5349
5350         return err;
5351 }
5352
5353 /*
5354  * set_array_info is used two different ways
5355  * The original usage is when creating a new array.
5356  * In this usage, raid_disks is > 0 and it together with
5357  *  level, size, not_persistent,layout,chunksize determine the
5358  *  shape of the array.
5359  *  This will always create an array with a type-0.90.0 superblock.
5360  * The newer usage is when assembling an array.
5361  *  In this case raid_disks will be 0, and the major_version field is
5362  *  use to determine which style super-blocks are to be found on the devices.
5363  *  The minor and patch _version numbers are also kept incase the
5364  *  super_block handler wishes to interpret them.
5365  */
5366 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5367 {
5368
5369         if (info->raid_disks == 0) {
5370                 /* just setting version number for superblock loading */
5371                 if (info->major_version < 0 ||
5372                     info->major_version >= ARRAY_SIZE(super_types) ||
5373                     super_types[info->major_version].name == NULL) {
5374                         /* maybe try to auto-load a module? */
5375                         printk(KERN_INFO
5376                                 "md: superblock version %d not known\n",
5377                                 info->major_version);
5378                         return -EINVAL;
5379                 }
5380                 mddev->major_version = info->major_version;
5381                 mddev->minor_version = info->minor_version;
5382                 mddev->patch_version = info->patch_version;
5383                 mddev->persistent = !info->not_persistent;
5384                 /* ensure mddev_put doesn't delete this now that there
5385                  * is some minimal configuration.
5386                  */
5387                 mddev->ctime         = get_seconds();
5388                 return 0;
5389         }
5390         mddev->major_version = MD_MAJOR_VERSION;
5391         mddev->minor_version = MD_MINOR_VERSION;
5392         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5393         mddev->ctime         = get_seconds();
5394
5395         mddev->level         = info->level;
5396         mddev->clevel[0]     = 0;
5397         mddev->dev_sectors   = 2 * (sector_t)info->size;
5398         mddev->raid_disks    = info->raid_disks;
5399         /* don't set md_minor, it is determined by which /dev/md* was
5400          * openned
5401          */
5402         if (info->state & (1<<MD_SB_CLEAN))
5403                 mddev->recovery_cp = MaxSector;
5404         else
5405                 mddev->recovery_cp = 0;
5406         mddev->persistent    = ! info->not_persistent;
5407         mddev->external      = 0;
5408
5409         mddev->layout        = info->layout;
5410         mddev->chunk_sectors = info->chunk_size >> 9;
5411
5412         mddev->max_disks     = MD_SB_DISKS;
5413
5414         if (mddev->persistent)
5415                 mddev->flags         = 0;
5416         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5417
5418         mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5419         mddev->bitmap_info.offset = 0;
5420
5421         mddev->reshape_position = MaxSector;
5422
5423         /*
5424          * Generate a 128 bit UUID
5425          */
5426         get_random_bytes(mddev->uuid, 16);
5427
5428         mddev->new_level = mddev->level;
5429         mddev->new_chunk_sectors = mddev->chunk_sectors;
5430         mddev->new_layout = mddev->layout;
5431         mddev->delta_disks = 0;
5432
5433         return 0;
5434 }
5435
5436 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5437 {
5438         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5439
5440         if (mddev->external_size)
5441                 return;
5442
5443         mddev->array_sectors = array_sectors;
5444 }
5445 EXPORT_SYMBOL(md_set_array_sectors);
5446
5447 static int update_size(mddev_t *mddev, sector_t num_sectors)
5448 {
5449         mdk_rdev_t *rdev;
5450         int rv;
5451         int fit = (num_sectors == 0);
5452
5453         if (mddev->pers->resize == NULL)
5454                 return -EINVAL;
5455         /* The "num_sectors" is the number of sectors of each device that
5456          * is used.  This can only make sense for arrays with redundancy.
5457          * linear and raid0 always use whatever space is available. We can only
5458          * consider changing this number if no resync or reconstruction is
5459          * happening, and if the new size is acceptable. It must fit before the
5460          * sb_start or, if that is <data_offset, it must fit before the size
5461          * of each device.  If num_sectors is zero, we find the largest size
5462          * that fits.
5463
5464          */
5465         if (mddev->sync_thread)
5466                 return -EBUSY;
5467         if (mddev->bitmap)
5468                 /* Sorry, cannot grow a bitmap yet, just remove it,
5469                  * grow, and re-add.
5470                  */
5471                 return -EBUSY;
5472         list_for_each_entry(rdev, &mddev->disks, same_set) {
5473                 sector_t avail = rdev->sectors;
5474
5475                 if (fit && (num_sectors == 0 || num_sectors > avail))
5476                         num_sectors = avail;
5477                 if (avail < num_sectors)
5478                         return -ENOSPC;
5479         }
5480         rv = mddev->pers->resize(mddev, num_sectors);
5481         if (!rv)
5482                 revalidate_disk(mddev->gendisk);
5483         return rv;
5484 }
5485
5486 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5487 {
5488         int rv;
5489         /* change the number of raid disks */
5490         if (mddev->pers->check_reshape == NULL)
5491                 return -EINVAL;
5492         if (raid_disks <= 0 ||
5493             (mddev->max_disks && raid_disks >= mddev->max_disks))
5494                 return -EINVAL;
5495         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5496                 return -EBUSY;
5497         mddev->delta_disks = raid_disks - mddev->raid_disks;
5498
5499         rv = mddev->pers->check_reshape(mddev);
5500         return rv;
5501 }
5502
5503
5504 /*
5505  * update_array_info is used to change the configuration of an
5506  * on-line array.
5507  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5508  * fields in the info are checked against the array.
5509  * Any differences that cannot be handled will cause an error.
5510  * Normally, only one change can be managed at a time.
5511  */
5512 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5513 {
5514         int rv = 0;
5515         int cnt = 0;
5516         int state = 0;
5517
5518         /* calculate expected state,ignoring low bits */
5519         if (mddev->bitmap && mddev->bitmap_info.offset)
5520                 state |= (1 << MD_SB_BITMAP_PRESENT);
5521
5522         if (mddev->major_version != info->major_version ||
5523             mddev->minor_version != info->minor_version ||
5524 /*          mddev->patch_version != info->patch_version || */
5525             mddev->ctime         != info->ctime         ||
5526             mddev->level         != info->level         ||
5527 /*          mddev->layout        != info->layout        || */
5528             !mddev->persistent   != info->not_persistent||
5529             mddev->chunk_sectors != info->chunk_size >> 9 ||
5530             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5531             ((state^info->state) & 0xfffffe00)
5532                 )
5533                 return -EINVAL;
5534         /* Check there is only one change */
5535         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5536                 cnt++;
5537         if (mddev->raid_disks != info->raid_disks)
5538                 cnt++;
5539         if (mddev->layout != info->layout)
5540                 cnt++;
5541         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5542                 cnt++;
5543         if (cnt == 0)
5544                 return 0;
5545         if (cnt > 1)
5546                 return -EINVAL;
5547
5548         if (mddev->layout != info->layout) {
5549                 /* Change layout
5550                  * we don't need to do anything at the md level, the
5551                  * personality will take care of it all.
5552                  */
5553                 if (mddev->pers->check_reshape == NULL)
5554                         return -EINVAL;
5555                 else {
5556                         mddev->new_layout = info->layout;
5557                         rv = mddev->pers->check_reshape(mddev);
5558                         if (rv)
5559                                 mddev->new_layout = mddev->layout;
5560                         return rv;
5561                 }
5562         }
5563         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5564                 rv = update_size(mddev, (sector_t)info->size * 2);
5565
5566         if (mddev->raid_disks    != info->raid_disks)
5567                 rv = update_raid_disks(mddev, info->raid_disks);
5568
5569         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5570                 if (mddev->pers->quiesce == NULL)
5571                         return -EINVAL;
5572                 if (mddev->recovery || mddev->sync_thread)
5573                         return -EBUSY;
5574                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5575                         /* add the bitmap */
5576                         if (mddev->bitmap)
5577                                 return -EEXIST;
5578                         if (mddev->bitmap_info.default_offset == 0)
5579                                 return -EINVAL;
5580                         mddev->bitmap_info.offset =
5581                                 mddev->bitmap_info.default_offset;
5582                         mddev->pers->quiesce(mddev, 1);
5583                         rv = bitmap_create(mddev);
5584                         if (rv)
5585                                 bitmap_destroy(mddev);
5586                         mddev->pers->quiesce(mddev, 0);
5587                 } else {
5588                         /* remove the bitmap */
5589                         if (!mddev->bitmap)
5590                                 return -ENOENT;
5591                         if (mddev->bitmap->file)
5592                                 return -EINVAL;
5593                         mddev->pers->quiesce(mddev, 1);
5594                         bitmap_destroy(mddev);
5595                         mddev->pers->quiesce(mddev, 0);
5596                         mddev->bitmap_info.offset = 0;
5597                 }
5598         }
5599         md_update_sb(mddev, 1);
5600         return rv;
5601 }
5602
5603 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5604 {
5605         mdk_rdev_t *rdev;
5606
5607         if (mddev->pers == NULL)
5608                 return -ENODEV;
5609
5610         rdev = find_rdev(mddev, dev);
5611         if (!rdev)
5612                 return -ENODEV;
5613
5614         md_error(mddev, rdev);
5615         return 0;
5616 }
5617
5618 /*
5619  * We have a problem here : there is no easy way to give a CHS
5620  * virtual geometry. We currently pretend that we have a 2 heads
5621  * 4 sectors (with a BIG number of cylinders...). This drives
5622  * dosfs just mad... ;-)
5623  */
5624 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5625 {
5626         mddev_t *mddev = bdev->bd_disk->private_data;
5627
5628         geo->heads = 2;
5629         geo->sectors = 4;
5630         geo->cylinders = mddev->array_sectors / 8;
5631         return 0;
5632 }
5633
5634 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5635                         unsigned int cmd, unsigned long arg)
5636 {
5637         int err = 0;
5638         void __user *argp = (void __user *)arg;
5639         mddev_t *mddev = NULL;
5640         int ro;
5641
5642         if (!capable(CAP_SYS_ADMIN))
5643                 return -EACCES;
5644
5645         /*
5646          * Commands dealing with the RAID driver but not any
5647          * particular array:
5648          */
5649         switch (cmd)
5650         {
5651                 case RAID_VERSION:
5652                         err = get_version(argp);
5653                         goto done;
5654
5655                 case PRINT_RAID_DEBUG:
5656                         err = 0;
5657                         md_print_devices();
5658                         goto done;
5659
5660 #ifndef MODULE
5661                 case RAID_AUTORUN:
5662                         err = 0;
5663                         autostart_arrays(arg);
5664                         goto done;
5665 #endif
5666                 default:;
5667         }
5668
5669         /*
5670          * Commands creating/starting a new array:
5671          */
5672
5673         mddev = bdev->bd_disk->private_data;
5674
5675         if (!mddev) {
5676                 BUG();
5677                 goto abort;
5678         }
5679
5680         err = mddev_lock(mddev);
5681         if (err) {
5682                 printk(KERN_INFO
5683                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5684                         err, cmd);
5685                 goto abort;
5686         }
5687
5688         switch (cmd)
5689         {
5690                 case SET_ARRAY_INFO:
5691                         {
5692                                 mdu_array_info_t info;
5693                                 if (!arg)
5694                                         memset(&info, 0, sizeof(info));
5695                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5696                                         err = -EFAULT;
5697                                         goto abort_unlock;
5698                                 }
5699                                 if (mddev->pers) {
5700                                         err = update_array_info(mddev, &info);
5701                                         if (err) {
5702                                                 printk(KERN_WARNING "md: couldn't update"
5703                                                        " array info. %d\n", err);
5704                                                 goto abort_unlock;
5705                                         }
5706                                         goto done_unlock;
5707                                 }
5708                                 if (!list_empty(&mddev->disks)) {
5709                                         printk(KERN_WARNING
5710                                                "md: array %s already has disks!\n",
5711                                                mdname(mddev));
5712                                         err = -EBUSY;
5713                                         goto abort_unlock;
5714                                 }
5715                                 if (mddev->raid_disks) {
5716                                         printk(KERN_WARNING
5717                                                "md: array %s already initialised!\n",
5718                                                mdname(mddev));
5719                                         err = -EBUSY;
5720                                         goto abort_unlock;
5721                                 }
5722                                 err = set_array_info(mddev, &info);
5723                                 if (err) {
5724                                         printk(KERN_WARNING "md: couldn't set"
5725                                                " array info. %d\n", err);
5726                                         goto abort_unlock;
5727                                 }
5728                         }
5729                         goto done_unlock;
5730
5731                 default:;
5732         }
5733
5734         /*
5735          * Commands querying/configuring an existing array:
5736          */
5737         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5738          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5739         if ((!mddev->raid_disks && !mddev->external)
5740             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5741             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5742             && cmd != GET_BITMAP_FILE) {
5743                 err = -ENODEV;
5744                 goto abort_unlock;
5745         }
5746
5747         /*
5748          * Commands even a read-only array can execute:
5749          */
5750         switch (cmd)
5751         {
5752                 case GET_ARRAY_INFO:
5753                         err = get_array_info(mddev, argp);
5754                         goto done_unlock;
5755
5756                 case GET_BITMAP_FILE:
5757                         err = get_bitmap_file(mddev, argp);
5758                         goto done_unlock;
5759
5760                 case GET_DISK_INFO:
5761                         err = get_disk_info(mddev, argp);
5762                         goto done_unlock;
5763
5764                 case RESTART_ARRAY_RW:
5765                         err = restart_array(mddev);
5766                         goto done_unlock;
5767
5768                 case STOP_ARRAY:
5769                         err = do_md_stop(mddev, 0, 1);
5770                         goto done_unlock;
5771
5772                 case STOP_ARRAY_RO:
5773                         err = md_set_readonly(mddev, 1);
5774                         goto done_unlock;
5775
5776                 case BLKROSET:
5777                         if (get_user(ro, (int __user *)(arg))) {
5778                                 err = -EFAULT;
5779                                 goto done_unlock;
5780                         }
5781                         err = -EINVAL;
5782
5783                         /* if the bdev is going readonly the value of mddev->ro
5784                          * does not matter, no writes are coming
5785                          */
5786                         if (ro)
5787                                 goto done_unlock;
5788
5789                         /* are we are already prepared for writes? */
5790                         if (mddev->ro != 1)
5791                                 goto done_unlock;
5792
5793                         /* transitioning to readauto need only happen for
5794                          * arrays that call md_write_start
5795                          */
5796                         if (mddev->pers) {
5797                                 err = restart_array(mddev);
5798                                 if (err == 0) {
5799                                         mddev->ro = 2;
5800                                         set_disk_ro(mddev->gendisk, 0);
5801                                 }
5802                         }
5803                         goto done_unlock;
5804         }
5805
5806         /*
5807          * The remaining ioctls are changing the state of the
5808          * superblock, so we do not allow them on read-only arrays.
5809          * However non-MD ioctls (e.g. get-size) will still come through
5810          * here and hit the 'default' below, so only disallow
5811          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5812          */
5813         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5814                 if (mddev->ro == 2) {
5815                         mddev->ro = 0;
5816                         sysfs_notify_dirent(mddev->sysfs_state);
5817                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5818                         md_wakeup_thread(mddev->thread);
5819                 } else {
5820                         err = -EROFS;
5821                         goto abort_unlock;
5822                 }
5823         }
5824
5825         switch (cmd)
5826         {
5827                 case ADD_NEW_DISK:
5828                 {
5829                         mdu_disk_info_t info;
5830                         if (copy_from_user(&info, argp, sizeof(info)))
5831                                 err = -EFAULT;
5832                         else
5833                                 err = add_new_disk(mddev, &info);
5834                         goto done_unlock;
5835                 }
5836
5837                 case HOT_REMOVE_DISK:
5838                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5839                         goto done_unlock;
5840
5841                 case HOT_ADD_DISK:
5842                         err = hot_add_disk(mddev, new_decode_dev(arg));
5843                         goto done_unlock;
5844
5845                 case SET_DISK_FAULTY:
5846                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5847                         goto done_unlock;
5848
5849                 case RUN_ARRAY:
5850                         err = do_md_run(mddev);
5851                         goto done_unlock;
5852
5853                 case SET_BITMAP_FILE:
5854                         err = set_bitmap_file(mddev, (int)arg);
5855                         goto done_unlock;
5856
5857                 default:
5858                         err = -EINVAL;
5859                         goto abort_unlock;
5860         }
5861
5862 done_unlock:
5863 abort_unlock:
5864         if (mddev->hold_active == UNTIL_IOCTL &&
5865             err != -EINVAL)
5866                 mddev->hold_active = 0;
5867         mddev_unlock(mddev);
5868
5869         return err;
5870 done:
5871         if (err)
5872                 MD_BUG();
5873 abort:
5874         return err;
5875 }
5876 #ifdef CONFIG_COMPAT
5877 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5878                     unsigned int cmd, unsigned long arg)
5879 {
5880         switch (cmd) {
5881         case HOT_REMOVE_DISK:
5882         case HOT_ADD_DISK:
5883         case SET_DISK_FAULTY:
5884         case SET_BITMAP_FILE:
5885                 /* These take in integer arg, do not convert */
5886                 break;
5887         default:
5888                 arg = (unsigned long)compat_ptr(arg);
5889                 break;
5890         }
5891
5892         return md_ioctl(bdev, mode, cmd, arg);
5893 }
5894 #endif /* CONFIG_COMPAT */
5895
5896 static int md_open(struct block_device *bdev, fmode_t mode)
5897 {
5898         /*
5899          * Succeed if we can lock the mddev, which confirms that
5900          * it isn't being stopped right now.
5901          */
5902         mddev_t *mddev = mddev_find(bdev->bd_dev);
5903         int err;
5904
5905         if (mddev->gendisk != bdev->bd_disk) {
5906                 /* we are racing with mddev_put which is discarding this
5907                  * bd_disk.
5908                  */
5909                 mddev_put(mddev);
5910                 /* Wait until bdev->bd_disk is definitely gone */
5911                 flush_scheduled_work();
5912                 /* Then retry the open from the top */
5913                 return -ERESTARTSYS;
5914         }
5915         BUG_ON(mddev != bdev->bd_disk->private_data);
5916
5917         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5918                 goto out;
5919
5920         err = 0;
5921         atomic_inc(&mddev->openers);
5922         mutex_unlock(&mddev->open_mutex);
5923
5924         check_disk_size_change(mddev->gendisk, bdev);
5925  out:
5926         return err;
5927 }
5928
5929 static int md_release(struct gendisk *disk, fmode_t mode)
5930 {
5931         mddev_t *mddev = disk->private_data;
5932
5933         BUG_ON(!mddev);
5934         atomic_dec(&mddev->openers);
5935         mddev_put(mddev);
5936
5937         return 0;
5938 }
5939 static const struct block_device_operations md_fops =
5940 {
5941         .owner          = THIS_MODULE,
5942         .open           = md_open,
5943         .release        = md_release,
5944         .ioctl          = md_ioctl,
5945 #ifdef CONFIG_COMPAT
5946         .compat_ioctl   = md_compat_ioctl,
5947 #endif
5948         .getgeo         = md_getgeo,
5949 };
5950
5951 static int md_thread(void * arg)
5952 {
5953         mdk_thread_t *thread = arg;
5954
5955         /*
5956          * md_thread is a 'system-thread', it's priority should be very
5957          * high. We avoid resource deadlocks individually in each
5958          * raid personality. (RAID5 does preallocation) We also use RR and
5959          * the very same RT priority as kswapd, thus we will never get
5960          * into a priority inversion deadlock.
5961          *
5962          * we definitely have to have equal or higher priority than
5963          * bdflush, otherwise bdflush will deadlock if there are too
5964          * many dirty RAID5 blocks.
5965          */
5966
5967         allow_signal(SIGKILL);
5968         while (!kthread_should_stop()) {
5969
5970                 /* We need to wait INTERRUPTIBLE so that
5971                  * we don't add to the load-average.
5972                  * That means we need to be sure no signals are
5973                  * pending
5974                  */
5975                 if (signal_pending(current))
5976                         flush_signals(current);
5977
5978                 wait_event_interruptible_timeout
5979                         (thread->wqueue,
5980                          test_bit(THREAD_WAKEUP, &thread->flags)
5981                          || kthread_should_stop(),
5982                          thread->timeout);
5983
5984                 clear_bit(THREAD_WAKEUP, &thread->flags);
5985
5986                 thread->run(thread->mddev);
5987         }
5988
5989         return 0;
5990 }
5991
5992 void md_wakeup_thread(mdk_thread_t *thread)
5993 {
5994         if (thread) {
5995                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5996                 set_bit(THREAD_WAKEUP, &thread->flags);
5997                 wake_up(&thread->wqueue);
5998         }
5999 }
6000
6001 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6002                                  const char *name)
6003 {
6004         mdk_thread_t *thread;
6005
6006         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6007         if (!thread)
6008                 return NULL;
6009
6010         init_waitqueue_head(&thread->wqueue);
6011
6012         thread->run = run;
6013         thread->mddev = mddev;
6014         thread->timeout = MAX_SCHEDULE_TIMEOUT;
6015         thread->tsk = kthread_run(md_thread, thread,
6016                                   "%s_%s",
6017                                   mdname(thread->mddev),
6018                                   name ?: mddev->pers->name);
6019         if (IS_ERR(thread->tsk)) {
6020                 kfree(thread);
6021                 return NULL;
6022         }
6023         return thread;
6024 }
6025
6026 void md_unregister_thread(mdk_thread_t *thread)
6027 {
6028         if (!thread)
6029                 return;
6030         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6031
6032         kthread_stop(thread->tsk);
6033         kfree(thread);
6034 }
6035
6036 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6037 {
6038         if (!mddev) {
6039                 MD_BUG();
6040                 return;
6041         }
6042
6043         if (!rdev || test_bit(Faulty, &rdev->flags))
6044                 return;
6045
6046         if (mddev->external)
6047                 set_bit(Blocked, &rdev->flags);
6048 /*
6049         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6050                 mdname(mddev),
6051                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6052                 __builtin_return_address(0),__builtin_return_address(1),
6053                 __builtin_return_address(2),__builtin_return_address(3));
6054 */
6055         if (!mddev->pers)
6056                 return;
6057         if (!mddev->pers->error_handler)
6058                 return;
6059         mddev->pers->error_handler(mddev,rdev);
6060         if (mddev->degraded)
6061                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6062         sysfs_notify_dirent(rdev->sysfs_state);
6063         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6064         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6065         md_wakeup_thread(mddev->thread);
6066         md_new_event_inintr(mddev);
6067 }
6068
6069 /* seq_file implementation /proc/mdstat */
6070
6071 static void status_unused(struct seq_file *seq)
6072 {
6073         int i = 0;
6074         mdk_rdev_t *rdev;
6075
6076         seq_printf(seq, "unused devices: ");
6077
6078         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6079                 char b[BDEVNAME_SIZE];
6080                 i++;
6081                 seq_printf(seq, "%s ",
6082                               bdevname(rdev->bdev,b));
6083         }
6084         if (!i)
6085                 seq_printf(seq, "<none>");
6086
6087         seq_printf(seq, "\n");
6088 }
6089
6090
6091 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6092 {
6093         sector_t max_sectors, resync, res;
6094         unsigned long dt, db;
6095         sector_t rt;
6096         int scale;
6097         unsigned int per_milli;
6098
6099         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6100
6101         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6102                 max_sectors = mddev->resync_max_sectors;
6103         else
6104                 max_sectors = mddev->dev_sectors;
6105
6106         /*
6107          * Should not happen.
6108          */
6109         if (!max_sectors) {
6110                 MD_BUG();
6111                 return;
6112         }
6113         /* Pick 'scale' such that (resync>>scale)*1000 will fit
6114          * in a sector_t, and (max_sectors>>scale) will fit in a
6115          * u32, as those are the requirements for sector_div.
6116          * Thus 'scale' must be at least 10
6117          */
6118         scale = 10;
6119         if (sizeof(sector_t) > sizeof(unsigned long)) {
6120                 while ( max_sectors/2 > (1ULL<<(scale+32)))
6121                         scale++;
6122         }
6123         res = (resync>>scale)*1000;
6124         sector_div(res, (u32)((max_sectors>>scale)+1));
6125
6126         per_milli = res;
6127         {
6128                 int i, x = per_milli/50, y = 20-x;
6129                 seq_printf(seq, "[");
6130                 for (i = 0; i < x; i++)
6131                         seq_printf(seq, "=");
6132                 seq_printf(seq, ">");
6133                 for (i = 0; i < y; i++)
6134                         seq_printf(seq, ".");
6135                 seq_printf(seq, "] ");
6136         }
6137         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6138                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6139                     "reshape" :
6140                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6141                      "check" :
6142                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6143                       "resync" : "recovery"))),
6144                    per_milli/10, per_milli % 10,
6145                    (unsigned long long) resync/2,
6146                    (unsigned long long) max_sectors/2);
6147
6148         /*
6149          * dt: time from mark until now
6150          * db: blocks written from mark until now
6151          * rt: remaining time
6152          *
6153          * rt is a sector_t, so could be 32bit or 64bit.
6154          * So we divide before multiply in case it is 32bit and close
6155          * to the limit.
6156          * We scale the divisor (db) by 32 to avoid loosing precision
6157          * near the end of resync when the number of remaining sectors
6158          * is close to 'db'.
6159          * We then divide rt by 32 after multiplying by db to compensate.
6160          * The '+1' avoids division by zero if db is very small.
6161          */
6162         dt = ((jiffies - mddev->resync_mark) / HZ);
6163         if (!dt) dt++;
6164         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6165                 - mddev->resync_mark_cnt;
6166
6167         rt = max_sectors - resync;    /* number of remaining sectors */
6168         sector_div(rt, db/32+1);
6169         rt *= dt;
6170         rt >>= 5;
6171
6172         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6173                    ((unsigned long)rt % 60)/6);
6174
6175         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6176 }
6177
6178 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6179 {
6180         struct list_head *tmp;
6181         loff_t l = *pos;
6182         mddev_t *mddev;
6183
6184         if (l >= 0x10000)
6185                 return NULL;
6186         if (!l--)
6187                 /* header */
6188                 return (void*)1;
6189
6190         spin_lock(&all_mddevs_lock);
6191         list_for_each(tmp,&all_mddevs)
6192                 if (!l--) {
6193                         mddev = list_entry(tmp, mddev_t, all_mddevs);
6194                         mddev_get(mddev);
6195                         spin_unlock(&all_mddevs_lock);
6196                         return mddev;
6197                 }
6198         spin_unlock(&all_mddevs_lock);
6199         if (!l--)
6200                 return (void*)2;/* tail */
6201         return NULL;
6202 }
6203
6204 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6205 {
6206         struct list_head *tmp;
6207         mddev_t *next_mddev, *mddev = v;
6208
6209         ++*pos;
6210         if (v == (void*)2)
6211                 return NULL;
6212
6213         spin_lock(&all_mddevs_lock);
6214         if (v == (void*)1)
6215                 tmp = all_mddevs.next;
6216         else
6217                 tmp = mddev->all_mddevs.next;
6218         if (tmp != &all_mddevs)
6219                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6220         else {
6221                 next_mddev = (void*)2;
6222                 *pos = 0x10000;
6223         }
6224         spin_unlock(&all_mddevs_lock);
6225
6226         if (v != (void*)1)
6227                 mddev_put(mddev);
6228         return next_mddev;
6229
6230 }
6231
6232 static void md_seq_stop(struct seq_file *seq, void *v)
6233 {
6234         mddev_t *mddev = v;
6235
6236         if (mddev && v != (void*)1 && v != (void*)2)
6237                 mddev_put(mddev);
6238 }
6239
6240 struct mdstat_info {
6241         int event;
6242 };
6243
6244 static int md_seq_show(struct seq_file *seq, void *v)
6245 {
6246         mddev_t *mddev = v;
6247         sector_t sectors;
6248         mdk_rdev_t *rdev;
6249         struct mdstat_info *mi = seq->private;
6250         struct bitmap *bitmap;
6251
6252         if (v == (void*)1) {
6253                 struct mdk_personality *pers;
6254                 seq_printf(seq, "Personalities : ");
6255                 spin_lock(&pers_lock);
6256                 list_for_each_entry(pers, &pers_list, list)
6257                         seq_printf(seq, "[%s] ", pers->name);
6258
6259                 spin_unlock(&pers_lock);
6260                 seq_printf(seq, "\n");
6261                 mi->event = atomic_read(&md_event_count);
6262                 return 0;
6263         }
6264         if (v == (void*)2) {
6265                 status_unused(seq);
6266                 return 0;
6267         }
6268
6269         if (mddev_lock(mddev) < 0)
6270                 return -EINTR;
6271
6272         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6273                 seq_printf(seq, "%s : %sactive", mdname(mddev),
6274                                                 mddev->pers ? "" : "in");
6275                 if (mddev->pers) {
6276                         if (mddev->ro==1)
6277                                 seq_printf(seq, " (read-only)");
6278                         if (mddev->ro==2)
6279                                 seq_printf(seq, " (auto-read-only)");
6280                         seq_printf(seq, " %s", mddev->pers->name);
6281                 }
6282
6283                 sectors = 0;
6284                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6285                         char b[BDEVNAME_SIZE];
6286                         seq_printf(seq, " %s[%d]",
6287                                 bdevname(rdev->bdev,b), rdev->desc_nr);
6288                         if (test_bit(WriteMostly, &rdev->flags))
6289                                 seq_printf(seq, "(W)");
6290                         if (test_bit(Faulty, &rdev->flags)) {
6291                                 seq_printf(seq, "(F)");
6292                                 continue;
6293                         } else if (rdev->raid_disk < 0)
6294                                 seq_printf(seq, "(S)"); /* spare */
6295                         sectors += rdev->sectors;
6296                 }
6297
6298                 if (!list_empty(&mddev->disks)) {
6299                         if (mddev->pers)
6300                                 seq_printf(seq, "\n      %llu blocks",
6301                                            (unsigned long long)
6302                                            mddev->array_sectors / 2);
6303                         else
6304                                 seq_printf(seq, "\n      %llu blocks",
6305                                            (unsigned long long)sectors / 2);
6306                 }
6307                 if (mddev->persistent) {
6308                         if (mddev->major_version != 0 ||
6309                             mddev->minor_version != 90) {
6310                                 seq_printf(seq," super %d.%d",
6311                                            mddev->major_version,
6312                                            mddev->minor_version);
6313                         }
6314                 } else if (mddev->external)
6315                         seq_printf(seq, " super external:%s",
6316                                    mddev->metadata_type);
6317                 else
6318                         seq_printf(seq, " super non-persistent");
6319
6320                 if (mddev->pers) {
6321                         mddev->pers->status(seq, mddev);
6322                         seq_printf(seq, "\n      ");
6323                         if (mddev->pers->sync_request) {
6324                                 if (mddev->curr_resync > 2) {
6325                                         status_resync(seq, mddev);
6326                                         seq_printf(seq, "\n      ");
6327                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6328                                         seq_printf(seq, "\tresync=DELAYED\n      ");
6329                                 else if (mddev->recovery_cp < MaxSector)
6330                                         seq_printf(seq, "\tresync=PENDING\n      ");
6331                         }
6332                 } else
6333                         seq_printf(seq, "\n       ");
6334
6335                 if ((bitmap = mddev->bitmap)) {
6336                         unsigned long chunk_kb;
6337                         unsigned long flags;
6338                         spin_lock_irqsave(&bitmap->lock, flags);
6339                         chunk_kb = mddev->bitmap_info.chunksize >> 10;
6340                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6341                                 "%lu%s chunk",
6342                                 bitmap->pages - bitmap->missing_pages,
6343                                 bitmap->pages,
6344                                 (bitmap->pages - bitmap->missing_pages)
6345                                         << (PAGE_SHIFT - 10),
6346                                 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6347                                 chunk_kb ? "KB" : "B");
6348                         if (bitmap->file) {
6349                                 seq_printf(seq, ", file: ");
6350                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
6351                         }
6352
6353                         seq_printf(seq, "\n");
6354                         spin_unlock_irqrestore(&bitmap->lock, flags);
6355                 }
6356
6357                 seq_printf(seq, "\n");
6358         }
6359         mddev_unlock(mddev);
6360
6361         return 0;
6362 }
6363
6364 static const struct seq_operations md_seq_ops = {
6365         .start  = md_seq_start,
6366         .next   = md_seq_next,
6367         .stop   = md_seq_stop,
6368         .show   = md_seq_show,
6369 };
6370
6371 static int md_seq_open(struct inode *inode, struct file *file)
6372 {
6373         int error;
6374         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6375         if (mi == NULL)
6376                 return -ENOMEM;
6377
6378         error = seq_open(file, &md_seq_ops);
6379         if (error)
6380                 kfree(mi);
6381         else {
6382                 struct seq_file *p = file->private_data;
6383                 p->private = mi;
6384                 mi->event = atomic_read(&md_event_count);
6385         }
6386         return error;
6387 }
6388
6389 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6390 {
6391         struct seq_file *m = filp->private_data;
6392         struct mdstat_info *mi = m->private;
6393         int mask;
6394
6395         poll_wait(filp, &md_event_waiters, wait);
6396
6397         /* always allow read */
6398         mask = POLLIN | POLLRDNORM;
6399
6400         if (mi->event != atomic_read(&md_event_count))
6401                 mask |= POLLERR | POLLPRI;
6402         return mask;
6403 }
6404
6405 static const struct file_operations md_seq_fops = {
6406         .owner          = THIS_MODULE,
6407         .open           = md_seq_open,
6408         .read           = seq_read,
6409         .llseek         = seq_lseek,
6410         .release        = seq_release_private,
6411         .poll           = mdstat_poll,
6412 };
6413
6414 int register_md_personality(struct mdk_personality *p)
6415 {
6416         spin_lock(&pers_lock);
6417         list_add_tail(&p->list, &pers_list);
6418         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6419         spin_unlock(&pers_lock);
6420         return 0;
6421 }
6422
6423 int unregister_md_personality(struct mdk_personality *p)
6424 {
6425         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6426         spin_lock(&pers_lock);
6427         list_del_init(&p->list);
6428         spin_unlock(&pers_lock);
6429         return 0;
6430 }
6431
6432 static int is_mddev_idle(mddev_t *mddev, int init)
6433 {
6434         mdk_rdev_t * rdev;
6435         int idle;
6436         int curr_events;
6437
6438         idle = 1;
6439         rcu_read_lock();
6440         rdev_for_each_rcu(rdev, mddev) {
6441                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6442                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6443                               (int)part_stat_read(&disk->part0, sectors[1]) -
6444                               atomic_read(&disk->sync_io);
6445                 /* sync IO will cause sync_io to increase before the disk_stats
6446                  * as sync_io is counted when a request starts, and
6447                  * disk_stats is counted when it completes.
6448                  * So resync activity will cause curr_events to be smaller than
6449                  * when there was no such activity.
6450                  * non-sync IO will cause disk_stat to increase without
6451                  * increasing sync_io so curr_events will (eventually)
6452                  * be larger than it was before.  Once it becomes
6453                  * substantially larger, the test below will cause
6454                  * the array to appear non-idle, and resync will slow
6455                  * down.
6456                  * If there is a lot of outstanding resync activity when
6457                  * we set last_event to curr_events, then all that activity
6458                  * completing might cause the array to appear non-idle
6459                  * and resync will be slowed down even though there might
6460                  * not have been non-resync activity.  This will only
6461                  * happen once though.  'last_events' will soon reflect
6462                  * the state where there is little or no outstanding
6463                  * resync requests, and further resync activity will
6464                  * always make curr_events less than last_events.
6465                  *
6466                  */
6467                 if (init || curr_events - rdev->last_events > 64) {
6468                         rdev->last_events = curr_events;
6469                         idle = 0;
6470                 }
6471         }
6472         rcu_read_unlock();
6473         return idle;
6474 }
6475
6476 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6477 {
6478         /* another "blocks" (512byte) blocks have been synced */
6479         atomic_sub(blocks, &mddev->recovery_active);
6480         wake_up(&mddev->recovery_wait);
6481         if (!ok) {
6482                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6483                 md_wakeup_thread(mddev->thread);
6484                 // stop recovery, signal do_sync ....
6485         }
6486 }
6487
6488
6489 /* md_write_start(mddev, bi)
6490  * If we need to update some array metadata (e.g. 'active' flag
6491  * in superblock) before writing, schedule a superblock update
6492  * and wait for it to complete.
6493  */
6494 void md_write_start(mddev_t *mddev, struct bio *bi)
6495 {
6496         int did_change = 0;
6497         if (bio_data_dir(bi) != WRITE)
6498                 return;
6499
6500         BUG_ON(mddev->ro == 1);
6501         if (mddev->ro == 2) {
6502                 /* need to switch to read/write */
6503                 mddev->ro = 0;
6504                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6505                 md_wakeup_thread(mddev->thread);
6506                 md_wakeup_thread(mddev->sync_thread);
6507                 did_change = 1;
6508         }
6509         atomic_inc(&mddev->writes_pending);
6510         if (mddev->safemode == 1)
6511                 mddev->safemode = 0;
6512         if (mddev->in_sync) {
6513                 spin_lock_irq(&mddev->write_lock);
6514                 if (mddev->in_sync) {
6515                         mddev->in_sync = 0;
6516                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6517                         md_wakeup_thread(mddev->thread);
6518                         did_change = 1;
6519                 }
6520                 spin_unlock_irq(&mddev->write_lock);
6521         }
6522         if (did_change)
6523                 sysfs_notify_dirent(mddev->sysfs_state);
6524         wait_event(mddev->sb_wait,
6525                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6526                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6527 }
6528
6529 void md_write_end(mddev_t *mddev)
6530 {
6531         if (atomic_dec_and_test(&mddev->writes_pending)) {
6532                 if (mddev->safemode == 2)
6533                         md_wakeup_thread(mddev->thread);
6534                 else if (mddev->safemode_delay)
6535                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6536         }
6537 }
6538
6539 /* md_allow_write(mddev)
6540  * Calling this ensures that the array is marked 'active' so that writes
6541  * may proceed without blocking.  It is important to call this before
6542  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6543  * Must be called with mddev_lock held.
6544  *
6545  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6546  * is dropped, so return -EAGAIN after notifying userspace.
6547  */
6548 int md_allow_write(mddev_t *mddev)
6549 {
6550         if (!mddev->pers)
6551                 return 0;
6552         if (mddev->ro)
6553                 return 0;
6554         if (!mddev->pers->sync_request)
6555                 return 0;
6556
6557         spin_lock_irq(&mddev->write_lock);
6558         if (mddev->in_sync) {
6559                 mddev->in_sync = 0;
6560                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6561                 if (mddev->safemode_delay &&
6562                     mddev->safemode == 0)
6563                         mddev->safemode = 1;
6564                 spin_unlock_irq(&mddev->write_lock);
6565                 md_update_sb(mddev, 0);
6566                 sysfs_notify_dirent(mddev->sysfs_state);
6567         } else
6568                 spin_unlock_irq(&mddev->write_lock);
6569
6570         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6571                 return -EAGAIN;
6572         else
6573                 return 0;
6574 }
6575 EXPORT_SYMBOL_GPL(md_allow_write);
6576
6577 #define SYNC_MARKS      10
6578 #define SYNC_MARK_STEP  (3*HZ)
6579 void md_do_sync(mddev_t *mddev)
6580 {
6581         mddev_t *mddev2;
6582         unsigned int currspeed = 0,
6583                  window;
6584         sector_t max_sectors,j, io_sectors;
6585         unsigned long mark[SYNC_MARKS];
6586         sector_t mark_cnt[SYNC_MARKS];
6587         int last_mark,m;
6588         struct list_head *tmp;
6589         sector_t last_check;
6590         int skipped = 0;
6591         mdk_rdev_t *rdev;
6592         char *desc;
6593
6594         /* just incase thread restarts... */
6595         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6596                 return;
6597         if (mddev->ro) /* never try to sync a read-only array */
6598                 return;
6599
6600         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6601                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6602                         desc = "data-check";
6603                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6604                         desc = "requested-resync";
6605                 else
6606                         desc = "resync";
6607         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6608                 desc = "reshape";
6609         else
6610                 desc = "recovery";
6611
6612         /* we overload curr_resync somewhat here.
6613          * 0 == not engaged in resync at all
6614          * 2 == checking that there is no conflict with another sync
6615          * 1 == like 2, but have yielded to allow conflicting resync to
6616          *              commense
6617          * other == active in resync - this many blocks
6618          *
6619          * Before starting a resync we must have set curr_resync to
6620          * 2, and then checked that every "conflicting" array has curr_resync
6621          * less than ours.  When we find one that is the same or higher
6622          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6623          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6624          * This will mean we have to start checking from the beginning again.
6625          *
6626          */
6627
6628         do {
6629                 mddev->curr_resync = 2;
6630
6631         try_again:
6632                 if (kthread_should_stop())
6633                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6634
6635                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6636                         goto skip;
6637                 for_each_mddev(mddev2, tmp) {
6638                         if (mddev2 == mddev)
6639                                 continue;
6640                         if (!mddev->parallel_resync
6641                         &&  mddev2->curr_resync
6642                         &&  match_mddev_units(mddev, mddev2)) {
6643                                 DEFINE_WAIT(wq);
6644                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6645                                         /* arbitrarily yield */
6646                                         mddev->curr_resync = 1;
6647                                         wake_up(&resync_wait);
6648                                 }
6649                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6650                                         /* no need to wait here, we can wait the next
6651                                          * time 'round when curr_resync == 2
6652                                          */
6653                                         continue;
6654                                 /* We need to wait 'interruptible' so as not to
6655                                  * contribute to the load average, and not to
6656                                  * be caught by 'softlockup'
6657                                  */
6658                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6659                                 if (!kthread_should_stop() &&
6660                                     mddev2->curr_resync >= mddev->curr_resync) {
6661                                         printk(KERN_INFO "md: delaying %s of %s"
6662                                                " until %s has finished (they"
6663                                                " share one or more physical units)\n",
6664                                                desc, mdname(mddev), mdname(mddev2));
6665                                         mddev_put(mddev2);
6666                                         if (signal_pending(current))
6667                                                 flush_signals(current);
6668                                         schedule();
6669                                         finish_wait(&resync_wait, &wq);
6670                                         goto try_again;
6671                                 }
6672                                 finish_wait(&resync_wait, &wq);
6673                         }
6674                 }
6675         } while (mddev->curr_resync < 2);
6676
6677         j = 0;
6678         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6679                 /* resync follows the size requested by the personality,
6680                  * which defaults to physical size, but can be virtual size
6681                  */
6682                 max_sectors = mddev->resync_max_sectors;
6683                 mddev->resync_mismatches = 0;
6684                 /* we don't use the checkpoint if there's a bitmap */
6685                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6686                         j = mddev->resync_min;
6687                 else if (!mddev->bitmap)
6688                         j = mddev->recovery_cp;
6689
6690         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6691                 max_sectors = mddev->dev_sectors;
6692         else {
6693                 /* recovery follows the physical size of devices */
6694                 max_sectors = mddev->dev_sectors;
6695                 j = MaxSector;
6696                 rcu_read_lock();
6697                 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6698                         if (rdev->raid_disk >= 0 &&
6699                             !test_bit(Faulty, &rdev->flags) &&
6700                             !test_bit(In_sync, &rdev->flags) &&
6701                             rdev->recovery_offset < j)
6702                                 j = rdev->recovery_offset;
6703                 rcu_read_unlock();
6704         }
6705
6706         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6707         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6708                 " %d KB/sec/disk.\n", speed_min(mddev));
6709         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6710                "(but not more than %d KB/sec) for %s.\n",
6711                speed_max(mddev), desc);
6712
6713         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6714
6715         io_sectors = 0;
6716         for (m = 0; m < SYNC_MARKS; m++) {
6717                 mark[m] = jiffies;
6718                 mark_cnt[m] = io_sectors;
6719         }
6720         last_mark = 0;
6721         mddev->resync_mark = mark[last_mark];
6722         mddev->resync_mark_cnt = mark_cnt[last_mark];
6723
6724         /*
6725          * Tune reconstruction:
6726          */
6727         window = 32*(PAGE_SIZE/512);
6728         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6729                 window/2,(unsigned long long) max_sectors/2);
6730
6731         atomic_set(&mddev->recovery_active, 0);
6732         last_check = 0;
6733
6734         if (j>2) {
6735                 printk(KERN_INFO
6736                        "md: resuming %s of %s from checkpoint.\n",
6737                        desc, mdname(mddev));
6738                 mddev->curr_resync = j;
6739         }
6740         mddev->curr_resync_completed = mddev->curr_resync;
6741
6742         while (j < max_sectors) {
6743                 sector_t sectors;
6744
6745                 skipped = 0;
6746
6747                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6748                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6749                       (mddev->curr_resync - mddev->curr_resync_completed)
6750                       > (max_sectors >> 4)) ||
6751                      (j - mddev->curr_resync_completed)*2
6752                      >= mddev->resync_max - mddev->curr_resync_completed
6753                             )) {
6754                         /* time to update curr_resync_completed */
6755                         blk_unplug(mddev->queue);
6756                         wait_event(mddev->recovery_wait,
6757                                    atomic_read(&mddev->recovery_active) == 0);
6758                         mddev->curr_resync_completed =
6759                                 mddev->curr_resync;
6760                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6761                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6762                 }
6763
6764                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6765                         /* As this condition is controlled by user-space,
6766                          * we can block indefinitely, so use '_interruptible'
6767                          * to avoid triggering warnings.
6768                          */
6769                         flush_signals(current); /* just in case */
6770                         wait_event_interruptible(mddev->recovery_wait,
6771                                                  mddev->resync_max > j
6772                                                  || kthread_should_stop());
6773                 }
6774
6775                 if (kthread_should_stop())
6776                         goto interrupted;
6777
6778                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6779                                                   currspeed < speed_min(mddev));
6780                 if (sectors == 0) {
6781                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6782                         goto out;
6783                 }
6784
6785                 if (!skipped) { /* actual IO requested */
6786                         io_sectors += sectors;
6787                         atomic_add(sectors, &mddev->recovery_active);
6788                 }
6789
6790                 j += sectors;
6791                 if (j>1) mddev->curr_resync = j;
6792                 mddev->curr_mark_cnt = io_sectors;
6793                 if (last_check == 0)
6794                         /* this is the earliers that rebuilt will be
6795                          * visible in /proc/mdstat
6796                          */
6797                         md_new_event(mddev);
6798
6799                 if (last_check + window > io_sectors || j == max_sectors)
6800                         continue;
6801
6802                 last_check = io_sectors;
6803
6804                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6805                         break;
6806
6807         repeat:
6808                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6809                         /* step marks */
6810                         int next = (last_mark+1) % SYNC_MARKS;
6811
6812                         mddev->resync_mark = mark[next];
6813                         mddev->resync_mark_cnt = mark_cnt[next];
6814                         mark[next] = jiffies;
6815                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6816                         last_mark = next;
6817                 }
6818
6819
6820                 if (kthread_should_stop())
6821                         goto interrupted;
6822
6823
6824                 /*
6825                  * this loop exits only if either when we are slower than
6826                  * the 'hard' speed limit, or the system was IO-idle for
6827                  * a jiffy.
6828                  * the system might be non-idle CPU-wise, but we only care
6829                  * about not overloading the IO subsystem. (things like an
6830                  * e2fsck being done on the RAID array should execute fast)
6831                  */
6832                 blk_unplug(mddev->queue);
6833                 cond_resched();
6834
6835                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6836                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6837
6838                 if (currspeed > speed_min(mddev)) {
6839                         if ((currspeed > speed_max(mddev)) ||
6840                                         !is_mddev_idle(mddev, 0)) {
6841                                 msleep(500);
6842                                 goto repeat;
6843                         }
6844                 }
6845         }
6846         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6847         /*
6848          * this also signals 'finished resyncing' to md_stop
6849          */
6850  out:
6851         blk_unplug(mddev->queue);
6852
6853         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6854
6855         /* tell personality that we are finished */
6856         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6857
6858         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6859             mddev->curr_resync > 2) {
6860                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6861                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6862                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6863                                         printk(KERN_INFO
6864                                                "md: checkpointing %s of %s.\n",
6865                                                desc, mdname(mddev));
6866                                         mddev->recovery_cp = mddev->curr_resync;
6867                                 }
6868                         } else
6869                                 mddev->recovery_cp = MaxSector;
6870                 } else {
6871                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6872                                 mddev->curr_resync = MaxSector;
6873                         rcu_read_lock();
6874                         list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6875                                 if (rdev->raid_disk >= 0 &&
6876                                     mddev->delta_disks >= 0 &&
6877                                     !test_bit(Faulty, &rdev->flags) &&
6878                                     !test_bit(In_sync, &rdev->flags) &&
6879                                     rdev->recovery_offset < mddev->curr_resync)
6880                                         rdev->recovery_offset = mddev->curr_resync;
6881                         rcu_read_unlock();
6882                 }
6883         }
6884         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6885
6886  skip:
6887         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6888                 /* We completed so min/max setting can be forgotten if used. */
6889                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6890                         mddev->resync_min = 0;
6891                 mddev->resync_max = MaxSector;
6892         } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6893                 mddev->resync_min = mddev->curr_resync_completed;
6894         mddev->curr_resync = 0;
6895         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6896                 mddev->curr_resync_completed = 0;
6897         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6898         wake_up(&resync_wait);
6899         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6900         md_wakeup_thread(mddev->thread);
6901         return;
6902
6903  interrupted:
6904         /*
6905          * got a signal, exit.
6906          */
6907         printk(KERN_INFO
6908                "md: md_do_sync() got signal ... exiting\n");
6909         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6910         goto out;
6911
6912 }
6913 EXPORT_SYMBOL_GPL(md_do_sync);
6914
6915
6916 static int remove_and_add_spares(mddev_t *mddev)
6917 {
6918         mdk_rdev_t *rdev;
6919         int spares = 0;
6920
6921         mddev->curr_resync_completed = 0;
6922
6923         list_for_each_entry(rdev, &mddev->disks, same_set)
6924                 if (rdev->raid_disk >= 0 &&
6925                     !test_bit(Blocked, &rdev->flags) &&
6926                     (test_bit(Faulty, &rdev->flags) ||
6927                      ! test_bit(In_sync, &rdev->flags)) &&
6928                     atomic_read(&rdev->nr_pending)==0) {
6929                         if (mddev->pers->hot_remove_disk(
6930                                     mddev, rdev->raid_disk)==0) {
6931                                 char nm[20];
6932                                 sprintf(nm,"rd%d", rdev->raid_disk);
6933                                 sysfs_remove_link(&mddev->kobj, nm);
6934                                 rdev->raid_disk = -1;
6935                         }
6936                 }
6937
6938         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6939                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6940                         if (rdev->raid_disk >= 0 &&
6941                             !test_bit(In_sync, &rdev->flags) &&
6942                             !test_bit(Blocked, &rdev->flags))
6943                                 spares++;
6944                         if (rdev->raid_disk < 0
6945                             && !test_bit(Faulty, &rdev->flags)) {
6946                                 rdev->recovery_offset = 0;
6947                                 if (mddev->pers->
6948                                     hot_add_disk(mddev, rdev) == 0) {
6949                                         char nm[20];
6950                                         sprintf(nm, "rd%d", rdev->raid_disk);
6951                                         if (sysfs_create_link(&mddev->kobj,
6952                                                               &rdev->kobj, nm))
6953                                                 printk(KERN_WARNING
6954                                                        "md: cannot register "
6955                                                        "%s for %s\n",
6956                                                        nm, mdname(mddev));
6957                                         spares++;
6958                                         md_new_event(mddev);
6959                                         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6960                                 } else
6961                                         break;
6962                         }
6963                 }
6964         }
6965         return spares;
6966 }
6967 /*
6968  * This routine is regularly called by all per-raid-array threads to
6969  * deal with generic issues like resync and super-block update.
6970  * Raid personalities that don't have a thread (linear/raid0) do not
6971  * need this as they never do any recovery or update the superblock.
6972  *
6973  * It does not do any resync itself, but rather "forks" off other threads
6974  * to do that as needed.
6975  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6976  * "->recovery" and create a thread at ->sync_thread.
6977  * When the thread finishes it sets MD_RECOVERY_DONE
6978  * and wakeups up this thread which will reap the thread and finish up.
6979  * This thread also removes any faulty devices (with nr_pending == 0).
6980  *
6981  * The overall approach is:
6982  *  1/ if the superblock needs updating, update it.
6983  *  2/ If a recovery thread is running, don't do anything else.
6984  *  3/ If recovery has finished, clean up, possibly marking spares active.
6985  *  4/ If there are any faulty devices, remove them.
6986  *  5/ If array is degraded, try to add spares devices
6987  *  6/ If array has spares or is not in-sync, start a resync thread.
6988  */
6989 void md_check_recovery(mddev_t *mddev)
6990 {
6991         mdk_rdev_t *rdev;
6992
6993
6994         if (mddev->bitmap)
6995                 bitmap_daemon_work(mddev);
6996
6997         if (mddev->ro)
6998                 return;
6999
7000         if (signal_pending(current)) {
7001                 if (mddev->pers->sync_request && !mddev->external) {
7002                         printk(KERN_INFO "md: %s in immediate safe mode\n",
7003                                mdname(mddev));
7004                         mddev->safemode = 2;
7005                 }
7006                 flush_signals(current);
7007         }
7008
7009         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7010                 return;
7011         if ( ! (
7012                 (mddev->flags && !mddev->external) ||
7013                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7014                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7015                 (mddev->external == 0 && mddev->safemode == 1) ||
7016                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7017                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7018                 ))
7019                 return;
7020
7021         if (mddev_trylock(mddev)) {
7022                 int spares = 0;
7023
7024                 if (mddev->ro) {
7025                         /* Only thing we do on a ro array is remove
7026                          * failed devices.
7027                          */
7028                         remove_and_add_spares(mddev);
7029                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7030                         goto unlock;
7031                 }
7032
7033                 if (!mddev->external) {
7034                         int did_change = 0;
7035                         spin_lock_irq(&mddev->write_lock);
7036                         if (mddev->safemode &&
7037                             !atomic_read(&mddev->writes_pending) &&
7038                             !mddev->in_sync &&
7039                             mddev->recovery_cp == MaxSector) {
7040                                 mddev->in_sync = 1;
7041                                 did_change = 1;
7042                                 if (mddev->persistent)
7043                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7044                         }
7045                         if (mddev->safemode == 1)
7046                                 mddev->safemode = 0;
7047                         spin_unlock_irq(&mddev->write_lock);
7048                         if (did_change)
7049                                 sysfs_notify_dirent(mddev->sysfs_state);
7050                 }
7051
7052                 if (mddev->flags)
7053                         md_update_sb(mddev, 0);
7054
7055                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7056                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7057                         /* resync/recovery still happening */
7058                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7059                         goto unlock;
7060                 }
7061                 if (mddev->sync_thread) {
7062                         /* resync has finished, collect result */
7063                         md_unregister_thread(mddev->sync_thread);
7064                         mddev->sync_thread = NULL;
7065                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7066                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7067                                 /* success...*/
7068                                 /* activate any spares */
7069                                 if (mddev->pers->spare_active(mddev))
7070                                         sysfs_notify(&mddev->kobj, NULL,
7071                                                      "degraded");
7072                         }
7073                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7074                             mddev->pers->finish_reshape)
7075                                 mddev->pers->finish_reshape(mddev);
7076                         md_update_sb(mddev, 1);
7077
7078                         /* if array is no-longer degraded, then any saved_raid_disk
7079                          * information must be scrapped
7080                          */
7081                         if (!mddev->degraded)
7082                                 list_for_each_entry(rdev, &mddev->disks, same_set)
7083                                         rdev->saved_raid_disk = -1;
7084
7085                         mddev->recovery = 0;
7086                         /* flag recovery needed just to double check */
7087                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7088                         sysfs_notify_dirent(mddev->sysfs_action);
7089                         md_new_event(mddev);
7090                         goto unlock;
7091                 }
7092                 /* Set RUNNING before clearing NEEDED to avoid
7093                  * any transients in the value of "sync_action".
7094                  */
7095                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7096                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7097                 /* Clear some bits that don't mean anything, but
7098                  * might be left set
7099                  */
7100                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7101                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7102
7103                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7104                         goto unlock;
7105                 /* no recovery is running.
7106                  * remove any failed drives, then
7107                  * add spares if possible.
7108                  * Spare are also removed and re-added, to allow
7109                  * the personality to fail the re-add.
7110                  */
7111
7112                 if (mddev->reshape_position != MaxSector) {
7113                         if (mddev->pers->check_reshape == NULL ||
7114                             mddev->pers->check_reshape(mddev) != 0)
7115                                 /* Cannot proceed */
7116                                 goto unlock;
7117                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7118                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7119                 } else if ((spares = remove_and_add_spares(mddev))) {
7120                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7121                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7122                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7123                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7124                 } else if (mddev->recovery_cp < MaxSector) {
7125                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7126                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7127                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7128                         /* nothing to be done ... */
7129                         goto unlock;
7130
7131                 if (mddev->pers->sync_request) {
7132                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7133                                 /* We are adding a device or devices to an array
7134                                  * which has the bitmap stored on all devices.
7135                                  * So make sure all bitmap pages get written
7136                                  */
7137                                 bitmap_write_all(mddev->bitmap);
7138                         }
7139                         mddev->sync_thread = md_register_thread(md_do_sync,
7140                                                                 mddev,
7141                                                                 "resync");
7142                         if (!mddev->sync_thread) {
7143                                 printk(KERN_ERR "%s: could not start resync"
7144                                         " thread...\n",
7145                                         mdname(mddev));
7146                                 /* leave the spares where they are, it shouldn't hurt */
7147                                 mddev->recovery = 0;
7148                         } else
7149                                 md_wakeup_thread(mddev->sync_thread);
7150                         sysfs_notify_dirent(mddev->sysfs_action);
7151                         md_new_event(mddev);
7152                 }
7153         unlock:
7154                 if (!mddev->sync_thread) {
7155                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7156                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7157                                                &mddev->recovery))
7158                                 if (mddev->sysfs_action)
7159                                         sysfs_notify_dirent(mddev->sysfs_action);
7160                 }
7161                 mddev_unlock(mddev);
7162         }
7163 }
7164
7165 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7166 {
7167         sysfs_notify_dirent(rdev->sysfs_state);
7168         wait_event_timeout(rdev->blocked_wait,
7169                            !test_bit(Blocked, &rdev->flags),
7170                            msecs_to_jiffies(5000));
7171         rdev_dec_pending(rdev, mddev);
7172 }
7173 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7174
7175 static int md_notify_reboot(struct notifier_block *this,
7176                             unsigned long code, void *x)
7177 {
7178         struct list_head *tmp;
7179         mddev_t *mddev;
7180
7181         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7182
7183                 printk(KERN_INFO "md: stopping all md devices.\n");
7184
7185                 for_each_mddev(mddev, tmp)
7186                         if (mddev_trylock(mddev)) {
7187                                 /* Force a switch to readonly even array
7188                                  * appears to still be in use.  Hence
7189                                  * the '100'.
7190                                  */
7191                                 md_set_readonly(mddev, 100);
7192                                 mddev_unlock(mddev);
7193                         }
7194                 /*
7195                  * certain more exotic SCSI devices are known to be
7196                  * volatile wrt too early system reboots. While the
7197                  * right place to handle this issue is the given
7198                  * driver, we do want to have a safe RAID driver ...
7199                  */
7200                 mdelay(1000*1);
7201         }
7202         return NOTIFY_DONE;
7203 }
7204
7205 static struct notifier_block md_notifier = {
7206         .notifier_call  = md_notify_reboot,
7207         .next           = NULL,
7208         .priority       = INT_MAX, /* before any real devices */
7209 };
7210
7211 static void md_geninit(void)
7212 {
7213         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7214
7215         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7216 }
7217
7218 static int __init md_init(void)
7219 {
7220         if (register_blkdev(MD_MAJOR, "md"))
7221                 return -1;
7222         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7223                 unregister_blkdev(MD_MAJOR, "md");
7224                 return -1;
7225         }
7226         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7227                             md_probe, NULL, NULL);
7228         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7229                             md_probe, NULL, NULL);
7230
7231         register_reboot_notifier(&md_notifier);
7232         raid_table_header = register_sysctl_table(raid_root_table);
7233
7234         md_geninit();
7235         return 0;
7236 }
7237
7238
7239 #ifndef MODULE
7240
7241 /*
7242  * Searches all registered partitions for autorun RAID arrays
7243  * at boot time.
7244  */
7245
7246 static LIST_HEAD(all_detected_devices);
7247 struct detected_devices_node {
7248         struct list_head list;
7249         dev_t dev;
7250 };
7251
7252 void md_autodetect_dev(dev_t dev)
7253 {
7254         struct detected_devices_node *node_detected_dev;
7255
7256         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7257         if (node_detected_dev) {
7258                 node_detected_dev->dev = dev;
7259                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7260         } else {
7261                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7262                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7263         }
7264 }
7265
7266
7267 static void autostart_arrays(int part)
7268 {
7269         mdk_rdev_t *rdev;
7270         struct detected_devices_node *node_detected_dev;
7271         dev_t dev;
7272         int i_scanned, i_passed;
7273
7274         i_scanned = 0;
7275         i_passed = 0;
7276
7277         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7278
7279         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7280                 i_scanned++;
7281                 node_detected_dev = list_entry(all_detected_devices.next,
7282                                         struct detected_devices_node, list);
7283                 list_del(&node_detected_dev->list);
7284                 dev = node_detected_dev->dev;
7285                 kfree(node_detected_dev);
7286                 rdev = md_import_device(dev,0, 90);
7287                 if (IS_ERR(rdev))
7288                         continue;
7289
7290                 if (test_bit(Faulty, &rdev->flags)) {
7291                         MD_BUG();
7292                         continue;
7293                 }
7294                 set_bit(AutoDetected, &rdev->flags);
7295                 list_add(&rdev->same_set, &pending_raid_disks);
7296                 i_passed++;
7297         }
7298
7299         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7300                                                 i_scanned, i_passed);
7301
7302         autorun_devices(part);
7303 }
7304
7305 #endif /* !MODULE */
7306
7307 static __exit void md_exit(void)
7308 {
7309         mddev_t *mddev;
7310         struct list_head *tmp;
7311
7312         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7313         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7314
7315         unregister_blkdev(MD_MAJOR,"md");
7316         unregister_blkdev(mdp_major, "mdp");
7317         unregister_reboot_notifier(&md_notifier);
7318         unregister_sysctl_table(raid_table_header);
7319         remove_proc_entry("mdstat", NULL);
7320         for_each_mddev(mddev, tmp) {
7321                 export_array(mddev);
7322                 mddev->hold_active = 0;
7323         }
7324 }
7325
7326 subsys_initcall(md_init);
7327 module_exit(md_exit)
7328
7329 static int get_ro(char *buffer, struct kernel_param *kp)
7330 {
7331         return sprintf(buffer, "%d", start_readonly);
7332 }
7333 static int set_ro(const char *val, struct kernel_param *kp)
7334 {
7335         char *e;
7336         int num = simple_strtoul(val, &e, 10);
7337         if (*val && (*e == '\0' || *e == '\n')) {
7338                 start_readonly = num;
7339                 return 0;
7340         }
7341         return -EINVAL;
7342 }
7343
7344 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7345 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7346
7347 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7348
7349 EXPORT_SYMBOL(register_md_personality);
7350 EXPORT_SYMBOL(unregister_md_personality);
7351 EXPORT_SYMBOL(md_error);
7352 EXPORT_SYMBOL(md_done_sync);
7353 EXPORT_SYMBOL(md_write_start);
7354 EXPORT_SYMBOL(md_write_end);
7355 EXPORT_SYMBOL(md_register_thread);
7356 EXPORT_SYMBOL(md_unregister_thread);
7357 EXPORT_SYMBOL(md_wakeup_thread);
7358 EXPORT_SYMBOL(md_check_recovery);
7359 MODULE_LICENSE("GPL");
7360 MODULE_DESCRIPTION("MD RAID framework");
7361 MODULE_ALIAS("md");
7362 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);