linux/drivers/md/md.c
<<
>>
Prefs
   1/*
   2   md.c : Multiple Devices driver for Linux
   3          Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5     completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7   Changes:
   8
   9   - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10   - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11   - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12   - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13   - kmod support by: Cyrus Durgin
  14   - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15   - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17   - lots of fixes and improvements to the RAID1/RAID5 and generic
  18     RAID code (such as request based resynchronization):
  19
  20     Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22   - persistent bitmap code
  23     Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25   This program is free software; you can redistribute it and/or modify
  26   it under the terms of the GNU General Public License as published by
  27   the Free Software Foundation; either version 2, or (at your option)
  28   any later version.
  29
  30   You should have received a copy of the GNU General Public License
  31   (for example /usr/src/linux/COPYING); if not, write to the Free
  32   Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33*/
  34
  35#include <linux/kthread.h>
  36#include <linux/blkdev.h>
  37#include <linux/sysctl.h>
  38#include <linux/seq_file.h>
  39#include <linux/fs.h>
  40#include <linux/poll.h>
  41#include <linux/ctype.h>
  42#include <linux/string.h>
  43#include <linux/hdreg.h>
  44#include <linux/proc_fs.h>
  45#include <linux/random.h>
  46#include <linux/module.h>
  47#include <linux/reboot.h>
  48#include <linux/file.h>
  49#include <linux/compat.h>
  50#include <linux/delay.h>
  51#include <linux/raid/md_p.h>
  52#include <linux/raid/md_u.h>
  53#include <linux/slab.h>
  54#include "md.h"
  55#include "bitmap.h"
  56
  57#ifndef MODULE
  58static void autostart_arrays(int part);
  59#endif
  60
  61/* pers_list is a list of registered personalities protected
  62 * by pers_lock.
  63 * pers_lock does extra service to protect accesses to
  64 * mddev->thread when the mutex cannot be held.
  65 */
  66static LIST_HEAD(pers_list);
  67static DEFINE_SPINLOCK(pers_lock);
  68
  69static void md_print_devices(void);
  70
  71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  72static struct workqueue_struct *md_wq;
  73static struct workqueue_struct *md_misc_wq;
  74
  75#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  76
  77/*
  78 * Default number of read corrections we'll attempt on an rdev
  79 * before ejecting it from the array. We divide the read error
  80 * count by 2 for every hour elapsed between read errors.
  81 */
  82#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  83/*
  84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  85 * is 1000 KB/sec, so the extra system load does not show up that much.
  86 * Increase it if you want to have more _guaranteed_ speed. Note that
  87 * the RAID driver will use the maximum available bandwidth if the IO
  88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
  89 * speed limit - in case reconstruction slows down your system despite
  90 * idle IO detection.
  91 *
  92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  93 * or /sys/block/mdX/md/sync_speed_{min,max}
  94 */
  95
  96static int sysctl_speed_limit_min = 1000;
  97static int sysctl_speed_limit_max = 200000;
  98static inline int speed_min(struct mddev *mddev)
  99{
 100        return mddev->sync_speed_min ?
 101                mddev->sync_speed_min : sysctl_speed_limit_min;
 102}
 103
 104static inline int speed_max(struct mddev *mddev)
 105{
 106        return mddev->sync_speed_max ?
 107                mddev->sync_speed_max : sysctl_speed_limit_max;
 108}
 109
 110static struct ctl_table_header *raid_table_header;
 111
 112static ctl_table raid_table[] = {
 113        {
 114                .procname       = "speed_limit_min",
 115                .data           = &sysctl_speed_limit_min,
 116                .maxlen         = sizeof(int),
 117                .mode           = S_IRUGO|S_IWUSR,
 118                .proc_handler   = proc_dointvec,
 119        },
 120        {
 121                .procname       = "speed_limit_max",
 122                .data           = &sysctl_speed_limit_max,
 123                .maxlen         = sizeof(int),
 124                .mode           = S_IRUGO|S_IWUSR,
 125                .proc_handler   = proc_dointvec,
 126        },
 127        { }
 128};
 129
 130static ctl_table raid_dir_table[] = {
 131        {
 132                .procname       = "raid",
 133                .maxlen         = 0,
 134                .mode           = S_IRUGO|S_IXUGO,
 135                .child          = raid_table,
 136        },
 137        { }
 138};
 139
 140static ctl_table raid_root_table[] = {
 141        {
 142                .procname       = "dev",
 143                .maxlen         = 0,
 144                .mode           = 0555,
 145                .child          = raid_dir_table,
 146        },
 147        {  }
 148};
 149
 150static const struct block_device_operations md_fops;
 151
 152static int start_readonly;
 153
 154/* bio_clone_mddev
 155 * like bio_clone, but with a local bio set
 156 */
 157
 158struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 159                            struct mddev *mddev)
 160{
 161        struct bio *b;
 162
 163        if (!mddev || !mddev->bio_set)
 164                return bio_alloc(gfp_mask, nr_iovecs);
 165
 166        b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
 167        if (!b)
 168                return NULL;
 169        return b;
 170}
 171EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 172
 173struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
 174                            struct mddev *mddev)
 175{
 176        if (!mddev || !mddev->bio_set)
 177                return bio_clone(bio, gfp_mask);
 178
 179        return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
 180}
 181EXPORT_SYMBOL_GPL(bio_clone_mddev);
 182
 183void md_trim_bio(struct bio *bio, int offset, int size)
 184{
 185        /* 'bio' is a cloned bio which we need to trim to match
 186         * the given offset and size.
 187         * This requires adjusting bi_sector, bi_size, and bi_io_vec
 188         */
 189        int i;
 190        struct bio_vec *bvec;
 191        int sofar = 0;
 192
 193        size <<= 9;
 194        if (offset == 0 && size == bio->bi_size)
 195                return;
 196
 197        bio->bi_sector += offset;
 198        bio->bi_size = size;
 199        offset <<= 9;
 200        clear_bit(BIO_SEG_VALID, &bio->bi_flags);
 201
 202        while (bio->bi_idx < bio->bi_vcnt &&
 203               bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
 204                /* remove this whole bio_vec */
 205                offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
 206                bio->bi_idx++;
 207        }
 208        if (bio->bi_idx < bio->bi_vcnt) {
 209                bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
 210                bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
 211        }
 212        /* avoid any complications with bi_idx being non-zero*/
 213        if (bio->bi_idx) {
 214                memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
 215                        (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
 216                bio->bi_vcnt -= bio->bi_idx;
 217                bio->bi_idx = 0;
 218        }
 219        /* Make sure vcnt and last bv are not too big */
 220        bio_for_each_segment(bvec, bio, i) {
 221                if (sofar + bvec->bv_len > size)
 222                        bvec->bv_len = size - sofar;
 223                if (bvec->bv_len == 0) {
 224                        bio->bi_vcnt = i;
 225                        break;
 226                }
 227                sofar += bvec->bv_len;
 228        }
 229}
 230EXPORT_SYMBOL_GPL(md_trim_bio);
 231
 232/*
 233 * We have a system wide 'event count' that is incremented
 234 * on any 'interesting' event, and readers of /proc/mdstat
 235 * can use 'poll' or 'select' to find out when the event
 236 * count increases.
 237 *
 238 * Events are:
 239 *  start array, stop array, error, add device, remove device,
 240 *  start build, activate spare
 241 */
 242static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 243static atomic_t md_event_count;
 244void md_new_event(struct mddev *mddev)
 245{
 246        atomic_inc(&md_event_count);
 247        wake_up(&md_event_waiters);
 248}
 249EXPORT_SYMBOL_GPL(md_new_event);
 250
 251/* Alternate version that can be called from interrupts
 252 * when calling sysfs_notify isn't needed.
 253 */
 254static void md_new_event_inintr(struct mddev *mddev)
 255{
 256        atomic_inc(&md_event_count);
 257        wake_up(&md_event_waiters);
 258}
 259
 260/*
 261 * Enables to iterate over all existing md arrays
 262 * all_mddevs_lock protects this list.
 263 */
 264static LIST_HEAD(all_mddevs);
 265static DEFINE_SPINLOCK(all_mddevs_lock);
 266
 267
 268/*
 269 * iterates through all used mddevs in the system.
 270 * We take care to grab the all_mddevs_lock whenever navigating
 271 * the list, and to always hold a refcount when unlocked.
 272 * Any code which breaks out of this loop while own
 273 * a reference to the current mddev and must mddev_put it.
 274 */
 275#define for_each_mddev(_mddev,_tmp)                                     \
 276                                                                        \
 277        for (({ spin_lock(&all_mddevs_lock);                            \
 278                _tmp = all_mddevs.next;                                 \
 279                _mddev = NULL;});                                       \
 280             ({ if (_tmp != &all_mddevs)                                \
 281                        mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
 282                spin_unlock(&all_mddevs_lock);                          \
 283                if (_mddev) mddev_put(_mddev);                          \
 284                _mddev = list_entry(_tmp, struct mddev, all_mddevs);    \
 285                _tmp != &all_mddevs;});                                 \
 286             ({ spin_lock(&all_mddevs_lock);                            \
 287                _tmp = _tmp->next;})                                    \
 288                )
 289
 290
 291/* Rather than calling directly into the personality make_request function,
 292 * IO requests come here first so that we can check if the device is
 293 * being suspended pending a reconfiguration.
 294 * We hold a refcount over the call to ->make_request.  By the time that
 295 * call has finished, the bio has been linked into some internal structure
 296 * and so is visible to ->quiesce(), so we don't need the refcount any more.
 297 */
 298static void md_make_request(struct request_queue *q, struct bio *bio)
 299{
 300        const int rw = bio_data_dir(bio);
 301        struct mddev *mddev = q->queuedata;
 302        int cpu;
 303        unsigned int sectors;
 304
 305        if (mddev == NULL || mddev->pers == NULL
 306            || !mddev->ready) {
 307                bio_io_error(bio);
 308                return;
 309        }
 310        smp_rmb(); /* Ensure implications of  'active' are visible */
 311        rcu_read_lock();
 312        if (mddev->suspended) {
 313                DEFINE_WAIT(__wait);
 314                for (;;) {
 315                        prepare_to_wait(&mddev->sb_wait, &__wait,
 316                                        TASK_UNINTERRUPTIBLE);
 317                        if (!mddev->suspended)
 318                                break;
 319                        rcu_read_unlock();
 320                        schedule();
 321                        rcu_read_lock();
 322                }
 323                finish_wait(&mddev->sb_wait, &__wait);
 324        }
 325        atomic_inc(&mddev->active_io);
 326        rcu_read_unlock();
 327
 328        /*
 329         * save the sectors now since our bio can
 330         * go away inside make_request
 331         */
 332        sectors = bio_sectors(bio);
 333        mddev->pers->make_request(mddev, bio);
 334
 335        cpu = part_stat_lock();
 336        part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 337        part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
 338        part_stat_unlock();
 339
 340        if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 341                wake_up(&mddev->sb_wait);
 342}
 343
 344/* mddev_suspend makes sure no new requests are submitted
 345 * to the device, and that any requests that have been submitted
 346 * are completely handled.
 347 * Once ->stop is called and completes, the module will be completely
 348 * unused.
 349 */
 350void mddev_suspend(struct mddev *mddev)
 351{
 352        BUG_ON(mddev->suspended);
 353        mddev->suspended = 1;
 354        synchronize_rcu();
 355        wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 356        mddev->pers->quiesce(mddev, 1);
 357
 358        del_timer_sync(&mddev->safemode_timer);
 359}
 360EXPORT_SYMBOL_GPL(mddev_suspend);
 361
 362void mddev_resume(struct mddev *mddev)
 363{
 364        mddev->suspended = 0;
 365        wake_up(&mddev->sb_wait);
 366        mddev->pers->quiesce(mddev, 0);
 367
 368        set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 369        md_wakeup_thread(mddev->thread);
 370        md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 371}
 372EXPORT_SYMBOL_GPL(mddev_resume);
 373
 374int mddev_congested(struct mddev *mddev, int bits)
 375{
 376        return mddev->suspended;
 377}
 378EXPORT_SYMBOL(mddev_congested);
 379
 380/*
 381 * Generic flush handling for md
 382 */
 383
 384static void md_end_flush(struct bio *bio, int err)
 385{
 386        struct md_rdev *rdev = bio->bi_private;
 387        struct mddev *mddev = rdev->mddev;
 388
 389        rdev_dec_pending(rdev, mddev);
 390
 391        if (atomic_dec_and_test(&mddev->flush_pending)) {
 392                /* The pre-request flush has finished */
 393                queue_work(md_wq, &mddev->flush_work);
 394        }
 395        bio_put(bio);
 396}
 397
 398static void md_submit_flush_data(struct work_struct *ws);
 399
 400static void submit_flushes(struct work_struct *ws)
 401{
 402        struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 403        struct md_rdev *rdev;
 404
 405        INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 406        atomic_set(&mddev->flush_pending, 1);
 407        rcu_read_lock();
 408        rdev_for_each_rcu(rdev, mddev)
 409                if (rdev->raid_disk >= 0 &&
 410                    !test_bit(Faulty, &rdev->flags)) {
 411                        /* Take two references, one is dropped
 412                         * when request finishes, one after
 413                         * we reclaim rcu_read_lock
 414                         */
 415                        struct bio *bi;
 416                        atomic_inc(&rdev->nr_pending);
 417                        atomic_inc(&rdev->nr_pending);
 418                        rcu_read_unlock();
 419                        bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
 420                        bi->bi_end_io = md_end_flush;
 421                        bi->bi_private = rdev;
 422                        bi->bi_bdev = rdev->bdev;
 423                        atomic_inc(&mddev->flush_pending);
 424                        submit_bio(WRITE_FLUSH, bi);
 425                        rcu_read_lock();
 426                        rdev_dec_pending(rdev, mddev);
 427                }
 428        rcu_read_unlock();
 429        if (atomic_dec_and_test(&mddev->flush_pending))
 430                queue_work(md_wq, &mddev->flush_work);
 431}
 432
 433static void md_submit_flush_data(struct work_struct *ws)
 434{
 435        struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 436        struct bio *bio = mddev->flush_bio;
 437
 438        if (bio->bi_size == 0)
 439                /* an empty barrier - all done */
 440                bio_endio(bio, 0);
 441        else {
 442                bio->bi_rw &= ~REQ_FLUSH;
 443                mddev->pers->make_request(mddev, bio);
 444        }
 445
 446        mddev->flush_bio = NULL;
 447        wake_up(&mddev->sb_wait);
 448}
 449
 450void md_flush_request(struct mddev *mddev, struct bio *bio)
 451{
 452        spin_lock_irq(&mddev->write_lock);
 453        wait_event_lock_irq(mddev->sb_wait,
 454                            !mddev->flush_bio,
 455                            mddev->write_lock, /*nothing*/);
 456        mddev->flush_bio = bio;
 457        spin_unlock_irq(&mddev->write_lock);
 458
 459        INIT_WORK(&mddev->flush_work, submit_flushes);
 460        queue_work(md_wq, &mddev->flush_work);
 461}
 462EXPORT_SYMBOL(md_flush_request);
 463
 464void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
 465{
 466        struct mddev *mddev = cb->data;
 467        md_wakeup_thread(mddev->thread);
 468        kfree(cb);
 469}
 470EXPORT_SYMBOL(md_unplug);
 471
 472static inline struct mddev *mddev_get(struct mddev *mddev)
 473{
 474        atomic_inc(&mddev->active);
 475        return mddev;
 476}
 477
 478static void mddev_delayed_delete(struct work_struct *ws);
 479
 480static void mddev_put(struct mddev *mddev)
 481{
 482        struct bio_set *bs = NULL;
 483
 484        if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 485                return;
 486        if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 487            mddev->ctime == 0 && !mddev->hold_active) {
 488                /* Array is not configured at all, and not held active,
 489                 * so destroy it */
 490                list_del_init(&mddev->all_mddevs);
 491                bs = mddev->bio_set;
 492                mddev->bio_set = NULL;
 493                if (mddev->gendisk) {
 494                        /* We did a probe so need to clean up.  Call
 495                         * queue_work inside the spinlock so that
 496                         * flush_workqueue() after mddev_find will
 497                         * succeed in waiting for the work to be done.
 498                         */
 499                        INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 500                        queue_work(md_misc_wq, &mddev->del_work);
 501                } else
 502                        kfree(mddev);
 503        }
 504        spin_unlock(&all_mddevs_lock);
 505        if (bs)
 506                bioset_free(bs);
 507}
 508
 509void mddev_init(struct mddev *mddev)
 510{
 511        mutex_init(&mddev->open_mutex);
 512        mutex_init(&mddev->reconfig_mutex);
 513        mutex_init(&mddev->bitmap_info.mutex);
 514        INIT_LIST_HEAD(&mddev->disks);
 515        INIT_LIST_HEAD(&mddev->all_mddevs);
 516        init_timer(&mddev->safemode_timer);
 517        atomic_set(&mddev->active, 1);
 518        atomic_set(&mddev->openers, 0);
 519        atomic_set(&mddev->active_io, 0);
 520        spin_lock_init(&mddev->write_lock);
 521        atomic_set(&mddev->flush_pending, 0);
 522        init_waitqueue_head(&mddev->sb_wait);
 523        init_waitqueue_head(&mddev->recovery_wait);
 524        mddev->reshape_position = MaxSector;
 525        mddev->reshape_backwards = 0;
 526        mddev->resync_min = 0;
 527        mddev->resync_max = MaxSector;
 528        mddev->level = LEVEL_NONE;
 529}
 530EXPORT_SYMBOL_GPL(mddev_init);
 531
 532static struct mddev * mddev_find(dev_t unit)
 533{
 534        struct mddev *mddev, *new = NULL;
 535
 536        if (unit && MAJOR(unit) != MD_MAJOR)
 537                unit &= ~((1<<MdpMinorShift)-1);
 538
 539 retry:
 540        spin_lock(&all_mddevs_lock);
 541
 542        if (unit) {
 543                list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 544                        if (mddev->unit == unit) {
 545                                mddev_get(mddev);
 546                                spin_unlock(&all_mddevs_lock);
 547                                kfree(new);
 548                                return mddev;
 549                        }
 550
 551                if (new) {
 552                        list_add(&new->all_mddevs, &all_mddevs);
 553                        spin_unlock(&all_mddevs_lock);
 554                        new->hold_active = UNTIL_IOCTL;
 555                        return new;
 556                }
 557        } else if (new) {
 558                /* find an unused unit number */
 559                static int next_minor = 512;
 560                int start = next_minor;
 561                int is_free = 0;
 562                int dev = 0;
 563                while (!is_free) {
 564                        dev = MKDEV(MD_MAJOR, next_minor);
 565                        next_minor++;
 566                        if (next_minor > MINORMASK)
 567                                next_minor = 0;
 568                        if (next_minor == start) {
 569                                /* Oh dear, all in use. */
 570                                spin_unlock(&all_mddevs_lock);
 571                                kfree(new);
 572                                return NULL;
 573                        }
 574                                
 575                        is_free = 1;
 576                        list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 577                                if (mddev->unit == dev) {
 578                                        is_free = 0;
 579                                        break;
 580                                }
 581                }
 582                new->unit = dev;
 583                new->md_minor = MINOR(dev);
 584                new->hold_active = UNTIL_STOP;
 585                list_add(&new->all_mddevs, &all_mddevs);
 586                spin_unlock(&all_mddevs_lock);
 587                return new;
 588        }
 589        spin_unlock(&all_mddevs_lock);
 590
 591        new = kzalloc(sizeof(*new), GFP_KERNEL);
 592        if (!new)
 593                return NULL;
 594
 595        new->unit = unit;
 596        if (MAJOR(unit) == MD_MAJOR)
 597                new->md_minor = MINOR(unit);
 598        else
 599                new->md_minor = MINOR(unit) >> MdpMinorShift;
 600
 601        mddev_init(new);
 602
 603        goto retry;
 604}
 605
 606static inline int mddev_lock(struct mddev * mddev)
 607{
 608        return mutex_lock_interruptible(&mddev->reconfig_mutex);
 609}
 610
 611static inline int mddev_is_locked(struct mddev *mddev)
 612{
 613        return mutex_is_locked(&mddev->reconfig_mutex);
 614}
 615
 616static inline int mddev_trylock(struct mddev * mddev)
 617{
 618        return mutex_trylock(&mddev->reconfig_mutex);
 619}
 620
 621static struct attribute_group md_redundancy_group;
 622
 623static void mddev_unlock(struct mddev * mddev)
 624{
 625        if (mddev->to_remove) {
 626                /* These cannot be removed under reconfig_mutex as
 627                 * an access to the files will try to take reconfig_mutex
 628                 * while holding the file unremovable, which leads to
 629                 * a deadlock.
 630                 * So hold set sysfs_active while the remove in happeing,
 631                 * and anything else which might set ->to_remove or my
 632                 * otherwise change the sysfs namespace will fail with
 633                 * -EBUSY if sysfs_active is still set.
 634                 * We set sysfs_active under reconfig_mutex and elsewhere
 635                 * test it under the same mutex to ensure its correct value
 636                 * is seen.
 637                 */
 638                struct attribute_group *to_remove = mddev->to_remove;
 639                mddev->to_remove = NULL;
 640                mddev->sysfs_active = 1;
 641                mutex_unlock(&mddev->reconfig_mutex);
 642
 643                if (mddev->kobj.sd) {
 644                        if (to_remove != &md_redundancy_group)
 645                                sysfs_remove_group(&mddev->kobj, to_remove);
 646                        if (mddev->pers == NULL ||
 647                            mddev->pers->sync_request == NULL) {
 648                                sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 649                                if (mddev->sysfs_action)
 650                                        sysfs_put(mddev->sysfs_action);
 651                                mddev->sysfs_action = NULL;
 652                        }
 653                }
 654                mddev->sysfs_active = 0;
 655        } else
 656                mutex_unlock(&mddev->reconfig_mutex);
 657
 658        /* As we've dropped the mutex we need a spinlock to
 659         * make sure the thread doesn't disappear
 660         */
 661        spin_lock(&pers_lock);
 662        md_wakeup_thread(mddev->thread);
 663        spin_unlock(&pers_lock);
 664}
 665
 666static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
 667{
 668        struct md_rdev *rdev;
 669
 670        rdev_for_each(rdev, mddev)
 671                if (rdev->desc_nr == nr)
 672                        return rdev;
 673
 674        return NULL;
 675}
 676
 677static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
 678{
 679        struct md_rdev *rdev;
 680
 681        rdev_for_each_rcu(rdev, mddev)
 682                if (rdev->desc_nr == nr)
 683                        return rdev;
 684
 685        return NULL;
 686}
 687
 688static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
 689{
 690        struct md_rdev *rdev;
 691
 692        rdev_for_each(rdev, mddev)
 693                if (rdev->bdev->bd_dev == dev)
 694                        return rdev;
 695
 696        return NULL;
 697}
 698
 699static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
 700{
 701        struct md_rdev *rdev;
 702
 703        rdev_for_each_rcu(rdev, mddev)
 704                if (rdev->bdev->bd_dev == dev)
 705                        return rdev;
 706
 707        return NULL;
 708}
 709
 710static struct md_personality *find_pers(int level, char *clevel)
 711{
 712        struct md_personality *pers;
 713        list_for_each_entry(pers, &pers_list, list) {
 714                if (level != LEVEL_NONE && pers->level == level)
 715                        return pers;
 716                if (strcmp(pers->name, clevel)==0)
 717                        return pers;
 718        }
 719        return NULL;
 720}
 721
 722/* return the offset of the super block in 512byte sectors */
 723static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
 724{
 725        sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 726        return MD_NEW_SIZE_SECTORS(num_sectors);
 727}
 728
 729static int alloc_disk_sb(struct md_rdev * rdev)
 730{
 731        if (rdev->sb_page)
 732                MD_BUG();
 733
 734        rdev->sb_page = alloc_page(GFP_KERNEL);
 735        if (!rdev->sb_page) {
 736                printk(KERN_ALERT "md: out of memory.\n");
 737                return -ENOMEM;
 738        }
 739
 740        return 0;
 741}
 742
 743void md_rdev_clear(struct md_rdev *rdev)
 744{
 745        if (rdev->sb_page) {
 746                put_page(rdev->sb_page);
 747                rdev->sb_loaded = 0;
 748                rdev->sb_page = NULL;
 749                rdev->sb_start = 0;
 750                rdev->sectors = 0;
 751        }
 752        if (rdev->bb_page) {
 753                put_page(rdev->bb_page);
 754                rdev->bb_page = NULL;
 755        }
 756        kfree(rdev->badblocks.page);
 757        rdev->badblocks.page = NULL;
 758}
 759EXPORT_SYMBOL_GPL(md_rdev_clear);
 760
 761static void super_written(struct bio *bio, int error)
 762{
 763        struct md_rdev *rdev = bio->bi_private;
 764        struct mddev *mddev = rdev->mddev;
 765
 766        if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 767                printk("md: super_written gets error=%d, uptodate=%d\n",
 768                       error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 769                WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 770                md_error(mddev, rdev);
 771        }
 772
 773        if (atomic_dec_and_test(&mddev->pending_writes))
 774                wake_up(&mddev->sb_wait);
 775        bio_put(bio);
 776}
 777
 778void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
 779                   sector_t sector, int size, struct page *page)
 780{
 781        /* write first size bytes of page to sector of rdev
 782         * Increment mddev->pending_writes before returning
 783         * and decrement it on completion, waking up sb_wait
 784         * if zero is reached.
 785         * If an error occurred, call md_error
 786         */
 787        struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 788
 789        bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
 790        bio->bi_sector = sector;
 791        bio_add_page(bio, page, size, 0);
 792        bio->bi_private = rdev;
 793        bio->bi_end_io = super_written;
 794
 795        atomic_inc(&mddev->pending_writes);
 796        submit_bio(WRITE_FLUSH_FUA, bio);
 797}
 798
 799void md_super_wait(struct mddev *mddev)
 800{
 801        /* wait for all superblock writes that were scheduled to complete */
 802        DEFINE_WAIT(wq);
 803        for(;;) {
 804                prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 805                if (atomic_read(&mddev->pending_writes)==0)
 806                        break;
 807                schedule();
 808        }
 809        finish_wait(&mddev->sb_wait, &wq);
 810}
 811
 812static void bi_complete(struct bio *bio, int error)
 813{
 814        complete((struct completion*)bio->bi_private);
 815}
 816
 817int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
 818                 struct page *page, int rw, bool metadata_op)
 819{
 820        struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 821        struct completion event;
 822        int ret;
 823
 824        rw |= REQ_SYNC;
 825
 826        bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
 827                rdev->meta_bdev : rdev->bdev;
 828        if (metadata_op)
 829                bio->bi_sector = sector + rdev->sb_start;
 830        else if (rdev->mddev->reshape_position != MaxSector &&
 831                 (rdev->mddev->reshape_backwards ==
 832                  (sector >= rdev->mddev->reshape_position)))
 833                bio->bi_sector = sector + rdev->new_data_offset;
 834        else
 835                bio->bi_sector = sector + rdev->data_offset;
 836        bio_add_page(bio, page, size, 0);
 837        init_completion(&event);
 838        bio->bi_private = &event;
 839        bio->bi_end_io = bi_complete;
 840        submit_bio(rw, bio);
 841        wait_for_completion(&event);
 842
 843        ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 844        bio_put(bio);
 845        return ret;
 846}
 847EXPORT_SYMBOL_GPL(sync_page_io);
 848
 849static int read_disk_sb(struct md_rdev * rdev, int size)
 850{
 851        char b[BDEVNAME_SIZE];
 852        if (!rdev->sb_page) {
 853                MD_BUG();
 854                return -EINVAL;
 855        }
 856        if (rdev->sb_loaded)
 857                return 0;
 858
 859
 860        if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
 861                goto fail;
 862        rdev->sb_loaded = 1;
 863        return 0;
 864
 865fail:
 866        printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 867                bdevname(rdev->bdev,b));
 868        return -EINVAL;
 869}
 870
 871static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 872{
 873        return  sb1->set_uuid0 == sb2->set_uuid0 &&
 874                sb1->set_uuid1 == sb2->set_uuid1 &&
 875                sb1->set_uuid2 == sb2->set_uuid2 &&
 876                sb1->set_uuid3 == sb2->set_uuid3;
 877}
 878
 879static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 880{
 881        int ret;
 882        mdp_super_t *tmp1, *tmp2;
 883
 884        tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 885        tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 886
 887        if (!tmp1 || !tmp2) {
 888                ret = 0;
 889                printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 890                goto abort;
 891        }
 892
 893        *tmp1 = *sb1;
 894        *tmp2 = *sb2;
 895
 896        /*
 897         * nr_disks is not constant
 898         */
 899        tmp1->nr_disks = 0;
 900        tmp2->nr_disks = 0;
 901
 902        ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 903abort:
 904        kfree(tmp1);
 905        kfree(tmp2);
 906        return ret;
 907}
 908
 909
 910static u32 md_csum_fold(u32 csum)
 911{
 912        csum = (csum & 0xffff) + (csum >> 16);
 913        return (csum & 0xffff) + (csum >> 16);
 914}
 915
 916static unsigned int calc_sb_csum(mdp_super_t * sb)
 917{
 918        u64 newcsum = 0;
 919        u32 *sb32 = (u32*)sb;
 920        int i;
 921        unsigned int disk_csum, csum;
 922
 923        disk_csum = sb->sb_csum;
 924        sb->sb_csum = 0;
 925
 926        for (i = 0; i < MD_SB_BYTES/4 ; i++)
 927                newcsum += sb32[i];
 928        csum = (newcsum & 0xffffffff) + (newcsum>>32);
 929
 930
 931#ifdef CONFIG_ALPHA
 932        /* This used to use csum_partial, which was wrong for several
 933         * reasons including that different results are returned on
 934         * different architectures.  It isn't critical that we get exactly
 935         * the same return value as before (we always csum_fold before
 936         * testing, and that removes any differences).  However as we
 937         * know that csum_partial always returned a 16bit value on
 938         * alphas, do a fold to maximise conformity to previous behaviour.
 939         */
 940        sb->sb_csum = md_csum_fold(disk_csum);
 941#else
 942        sb->sb_csum = disk_csum;
 943#endif
 944        return csum;
 945}
 946
 947
 948/*
 949 * Handle superblock details.
 950 * We want to be able to handle multiple superblock formats
 951 * so we have a common interface to them all, and an array of
 952 * different handlers.
 953 * We rely on user-space to write the initial superblock, and support
 954 * reading and updating of superblocks.
 955 * Interface methods are:
 956 *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
 957 *      loads and validates a superblock on dev.
 958 *      if refdev != NULL, compare superblocks on both devices
 959 *    Return:
 960 *      0 - dev has a superblock that is compatible with refdev
 961 *      1 - dev has a superblock that is compatible and newer than refdev
 962 *          so dev should be used as the refdev in future
 963 *     -EINVAL superblock incompatible or invalid
 964 *     -othererror e.g. -EIO
 965 *
 966 *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
 967 *      Verify that dev is acceptable into mddev.
 968 *       The first time, mddev->raid_disks will be 0, and data from
 969 *       dev should be merged in.  Subsequent calls check that dev
 970 *       is new enough.  Return 0 or -EINVAL
 971 *
 972 *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
 973 *     Update the superblock for rdev with data in mddev
 974 *     This does not write to disc.
 975 *
 976 */
 977
 978struct super_type  {
 979        char                *name;
 980        struct module       *owner;
 981        int                 (*load_super)(struct md_rdev *rdev,
 982                                          struct md_rdev *refdev,
 983                                          int minor_version);
 984        int                 (*validate_super)(struct mddev *mddev,
 985                                              struct md_rdev *rdev);
 986        void                (*sync_super)(struct mddev *mddev,
 987                                          struct md_rdev *rdev);
 988        unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
 989                                                sector_t num_sectors);
 990        int                 (*allow_new_offset)(struct md_rdev *rdev,
 991                                                unsigned long long new_offset);
 992};
 993
 994/*
 995 * Check that the given mddev has no bitmap.
 996 *
 997 * This function is called from the run method of all personalities that do not
 998 * support bitmaps. It prints an error message and returns non-zero if mddev
 999 * has a bitmap. Otherwise, it returns 0.
1000 *
1001 */
1002int md_check_no_bitmap(struct mddev *mddev)
1003{
1004        if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1005                return 0;
1006        printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1007                mdname(mddev), mddev->pers->name);
1008        return 1;
1009}
1010EXPORT_SYMBOL(md_check_no_bitmap);
1011
1012/*
1013 * load_super for 0.90.0 
1014 */
1015static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1016{
1017        char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1018        mdp_super_t *sb;
1019        int ret;
1020
1021        /*
1022         * Calculate the position of the superblock (512byte sectors),
1023         * it's at the end of the disk.
1024         *
1025         * It also happens to be a multiple of 4Kb.
1026         */
1027        rdev->sb_start = calc_dev_sboffset(rdev);
1028
1029        ret = read_disk_sb(rdev, MD_SB_BYTES);
1030        if (ret) return ret;
1031
1032        ret = -EINVAL;
1033
1034        bdevname(rdev->bdev, b);
1035        sb = page_address(rdev->sb_page);
1036
1037        if (sb->md_magic != MD_SB_MAGIC) {
1038                printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1039                       b);
1040                goto abort;
1041        }
1042
1043        if (sb->major_version != 0 ||
1044            sb->minor_version < 90 ||
1045            sb->minor_version > 91) {
1046                printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1047                        sb->major_version, sb->minor_version,
1048                        b);
1049                goto abort;
1050        }
1051
1052        if (sb->raid_disks <= 0)
1053                goto abort;
1054
1055        if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1056                printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1057                        b);
1058                goto abort;
1059        }
1060
1061        rdev->preferred_minor = sb->md_minor;
1062        rdev->data_offset = 0;
1063        rdev->new_data_offset = 0;
1064        rdev->sb_size = MD_SB_BYTES;
1065        rdev->badblocks.shift = -1;
1066
1067        if (sb->level == LEVEL_MULTIPATH)
1068                rdev->desc_nr = -1;
1069        else
1070                rdev->desc_nr = sb->this_disk.number;
1071
1072        if (!refdev) {
1073                ret = 1;
1074        } else {
1075                __u64 ev1, ev2;
1076                mdp_super_t *refsb = page_address(refdev->sb_page);
1077                if (!uuid_equal(refsb, sb)) {
1078                        printk(KERN_WARNING "md: %s has different UUID to %s\n",
1079                                b, bdevname(refdev->bdev,b2));
1080                        goto abort;
1081                }
1082                if (!sb_equal(refsb, sb)) {
1083                        printk(KERN_WARNING "md: %s has same UUID"
1084                               " but different superblock to %s\n",
1085                               b, bdevname(refdev->bdev, b2));
1086                        goto abort;
1087                }
1088                ev1 = md_event(sb);
1089                ev2 = md_event(refsb);
1090                if (ev1 > ev2)
1091                        ret = 1;
1092                else 
1093                        ret = 0;
1094        }
1095        rdev->sectors = rdev->sb_start;
1096        /* Limit to 4TB as metadata cannot record more than that.
1097         * (not needed for Linear and RAID0 as metadata doesn't
1098         * record this size)
1099         */
1100        if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1101                rdev->sectors = (2ULL << 32) - 2;
1102
1103        if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1104                /* "this cannot possibly happen" ... */
1105                ret = -EINVAL;
1106
1107 abort:
1108        return ret;
1109}
1110
1111/*
1112 * validate_super for 0.90.0
1113 */
1114static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1115{
1116        mdp_disk_t *desc;
1117        mdp_super_t *sb = page_address(rdev->sb_page);
1118        __u64 ev1 = md_event(sb);
1119
1120        rdev->raid_disk = -1;
1121        clear_bit(Faulty, &rdev->flags);
1122        clear_bit(In_sync, &rdev->flags);
1123        clear_bit(WriteMostly, &rdev->flags);
1124
1125        if (mddev->raid_disks == 0) {
1126                mddev->major_version = 0;
1127                mddev->minor_version = sb->minor_version;
1128                mddev->patch_version = sb->patch_version;
1129                mddev->external = 0;
1130                mddev->chunk_sectors = sb->chunk_size >> 9;
1131                mddev->ctime = sb->ctime;
1132                mddev->utime = sb->utime;
1133                mddev->level = sb->level;
1134                mddev->clevel[0] = 0;
1135                mddev->layout = sb->layout;
1136                mddev->raid_disks = sb->raid_disks;
1137                mddev->dev_sectors = ((sector_t)sb->size) * 2;
1138                mddev->events = ev1;
1139                mddev->bitmap_info.offset = 0;
1140                mddev->bitmap_info.space = 0;
1141                /* bitmap can use 60 K after the 4K superblocks */
1142                mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1143                mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1144                mddev->reshape_backwards = 0;
1145
1146                if (mddev->minor_version >= 91) {
1147                        mddev->reshape_position = sb->reshape_position;
1148                        mddev->delta_disks = sb->delta_disks;
1149                        mddev->new_level = sb->new_level;
1150                        mddev->new_layout = sb->new_layout;
1151                        mddev->new_chunk_sectors = sb->new_chunk >> 9;
1152                        if (mddev->delta_disks < 0)
1153                                mddev->reshape_backwards = 1;
1154                } else {
1155                        mddev->reshape_position = MaxSector;
1156                        mddev->delta_disks = 0;
1157                        mddev->new_level = mddev->level;
1158                        mddev->new_layout = mddev->layout;
1159                        mddev->new_chunk_sectors = mddev->chunk_sectors;
1160                }
1161
1162                if (sb->state & (1<<MD_SB_CLEAN))
1163                        mddev->recovery_cp = MaxSector;
1164                else {
1165                        if (sb->events_hi == sb->cp_events_hi && 
1166                                sb->events_lo == sb->cp_events_lo) {
1167                                mddev->recovery_cp = sb->recovery_cp;
1168                        } else
1169                                mddev->recovery_cp = 0;
1170                }
1171
1172                memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1173                memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1174                memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1175                memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1176
1177                mddev->max_disks = MD_SB_DISKS;
1178
1179                if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1180                    mddev->bitmap_info.file == NULL) {
1181                        mddev->bitmap_info.offset =
1182                                mddev->bitmap_info.default_offset;
1183                        mddev->bitmap_info.space =
1184                                mddev->bitmap_info.space;
1185                }
1186
1187        } else if (mddev->pers == NULL) {
1188                /* Insist on good event counter while assembling, except
1189                 * for spares (which don't need an event count) */
1190                ++ev1;
1191                if (sb->disks[rdev->desc_nr].state & (
1192                            (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1193                        if (ev1 < mddev->events) 
1194                                return -EINVAL;
1195        } else if (mddev->bitmap) {
1196                /* if adding to array with a bitmap, then we can accept an
1197                 * older device ... but not too old.
1198                 */
1199                if (ev1 < mddev->bitmap->events_cleared)
1200                        return 0;
1201        } else {
1202                if (ev1 < mddev->events)
1203                        /* just a hot-add of a new device, leave raid_disk at -1 */
1204                        return 0;
1205        }
1206
1207        if (mddev->level != LEVEL_MULTIPATH) {
1208                desc = sb->disks + rdev->desc_nr;
1209
1210                if (desc->state & (1<<MD_DISK_FAULTY))
1211                        set_bit(Faulty, &rdev->flags);
1212                else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1213                            desc->raid_disk < mddev->raid_disks */) {
1214                        set_bit(In_sync, &rdev->flags);
1215                        rdev->raid_disk = desc->raid_disk;
1216                } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1217                        /* active but not in sync implies recovery up to
1218                         * reshape position.  We don't know exactly where
1219                         * that is, so set to zero for now */
1220                        if (mddev->minor_version >= 91) {
1221                                rdev->recovery_offset = 0;
1222                                rdev->raid_disk = desc->raid_disk;
1223                        }
1224                }
1225                if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1226                        set_bit(WriteMostly, &rdev->flags);
1227        } else /* MULTIPATH are always insync */
1228                set_bit(In_sync, &rdev->flags);
1229        return 0;
1230}
1231
1232/*
1233 * sync_super for 0.90.0
1234 */
1235static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1236{
1237        mdp_super_t *sb;
1238        struct md_rdev *rdev2;
1239        int next_spare = mddev->raid_disks;
1240
1241
1242        /* make rdev->sb match mddev data..
1243         *
1244         * 1/ zero out disks
1245         * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1246         * 3/ any empty disks < next_spare become removed
1247         *
1248         * disks[0] gets initialised to REMOVED because
1249         * we cannot be sure from other fields if it has
1250         * been initialised or not.
1251         */
1252        int i;
1253        int active=0, working=0,failed=0,spare=0,nr_disks=0;
1254
1255        rdev->sb_size = MD_SB_BYTES;
1256
1257        sb = page_address(rdev->sb_page);
1258
1259        memset(sb, 0, sizeof(*sb));
1260
1261        sb->md_magic = MD_SB_MAGIC;
1262        sb->major_version = mddev->major_version;
1263        sb->patch_version = mddev->patch_version;
1264        sb->gvalid_words  = 0; /* ignored */
1265        memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1266        memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1267        memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1268        memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1269
1270        sb->ctime = mddev->ctime;
1271        sb->level = mddev->level;
1272        sb->size = mddev->dev_sectors / 2;
1273        sb->raid_disks = mddev->raid_disks;
1274        sb->md_minor = mddev->md_minor;
1275        sb->not_persistent = 0;
1276        sb->utime = mddev->utime;
1277        sb->state = 0;
1278        sb->events_hi = (mddev->events>>32);
1279        sb->events_lo = (u32)mddev->events;
1280
1281        if (mddev->reshape_position == MaxSector)
1282                sb->minor_version = 90;
1283        else {
1284                sb->minor_version = 91;
1285                sb->reshape_position = mddev->reshape_position;
1286                sb->new_level = mddev->new_level;
1287                sb->delta_disks = mddev->delta_disks;
1288                sb->new_layout = mddev->new_layout;
1289                sb->new_chunk = mddev->new_chunk_sectors << 9;
1290        }
1291        mddev->minor_version = sb->minor_version;
1292        if (mddev->in_sync)
1293        {
1294                sb->recovery_cp = mddev->recovery_cp;
1295                sb->cp_events_hi = (mddev->events>>32);
1296                sb->cp_events_lo = (u32)mddev->events;
1297                if (mddev->recovery_cp == MaxSector)
1298                        sb->state = (1<< MD_SB_CLEAN);
1299        } else
1300                sb->recovery_cp = 0;
1301
1302        sb->layout = mddev->layout;
1303        sb->chunk_size = mddev->chunk_sectors << 9;
1304
1305        if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1306                sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1307
1308        sb->disks[0].state = (1<<MD_DISK_REMOVED);
1309        rdev_for_each(rdev2, mddev) {
1310                mdp_disk_t *d;
1311                int desc_nr;
1312                int is_active = test_bit(In_sync, &rdev2->flags);
1313
1314                if (rdev2->raid_disk >= 0 &&
1315                    sb->minor_version >= 91)
1316                        /* we have nowhere to store the recovery_offset,
1317                         * but if it is not below the reshape_position,
1318                         * we can piggy-back on that.
1319                         */
1320                        is_active = 1;
1321                if (rdev2->raid_disk < 0 ||
1322                    test_bit(Faulty, &rdev2->flags))
1323                        is_active = 0;
1324                if (is_active)
1325                        desc_nr = rdev2->raid_disk;
1326                else
1327                        desc_nr = next_spare++;
1328                rdev2->desc_nr = desc_nr;
1329                d = &sb->disks[rdev2->desc_nr];
1330                nr_disks++;
1331                d->number = rdev2->desc_nr;
1332                d->major = MAJOR(rdev2->bdev->bd_dev);
1333                d->minor = MINOR(rdev2->bdev->bd_dev);
1334                if (is_active)
1335                        d->raid_disk = rdev2->raid_disk;
1336                else
1337                        d->raid_disk = rdev2->desc_nr; /* compatibility */
1338                if (test_bit(Faulty, &rdev2->flags))
1339                        d->state = (1<<MD_DISK_FAULTY);
1340                else if (is_active) {
1341                        d->state = (1<<MD_DISK_ACTIVE);
1342                        if (test_bit(In_sync, &rdev2->flags))
1343                                d->state |= (1<<MD_DISK_SYNC);
1344                        active++;
1345                        working++;
1346                } else {
1347                        d->state = 0;
1348                        spare++;
1349                        working++;
1350                }
1351                if (test_bit(WriteMostly, &rdev2->flags))
1352                        d->state |= (1<<MD_DISK_WRITEMOSTLY);
1353        }
1354        /* now set the "removed" and "faulty" bits on any missing devices */
1355        for (i=0 ; i < mddev->raid_disks ; i++) {
1356                mdp_disk_t *d = &sb->disks[i];
1357                if (d->state == 0 && d->number == 0) {
1358                        d->number = i;
1359                        d->raid_disk = i;
1360                        d->state = (1<<MD_DISK_REMOVED);
1361                        d->state |= (1<<MD_DISK_FAULTY);
1362                        failed++;
1363                }
1364        }
1365        sb->nr_disks = nr_disks;
1366        sb->active_disks = active;
1367        sb->working_disks = working;
1368        sb->failed_disks = failed;
1369        sb->spare_disks = spare;
1370
1371        sb->this_disk = sb->disks[rdev->desc_nr];
1372        sb->sb_csum = calc_sb_csum(sb);
1373}
1374
1375/*
1376 * rdev_size_change for 0.90.0
1377 */
1378static unsigned long long
1379super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1380{
1381        if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1382                return 0; /* component must fit device */
1383        if (rdev->mddev->bitmap_info.offset)
1384                return 0; /* can't move bitmap */
1385        rdev->sb_start = calc_dev_sboffset(rdev);
1386        if (!num_sectors || num_sectors > rdev->sb_start)
1387                num_sectors = rdev->sb_start;
1388        /* Limit to 4TB as metadata cannot record more than that.
1389         * 4TB == 2^32 KB, or 2*2^32 sectors.
1390         */
1391        if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1392                num_sectors = (2ULL << 32) - 2;
1393        md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1394                       rdev->sb_page);
1395        md_super_wait(rdev->mddev);
1396        return num_sectors;
1397}
1398
1399static int
1400super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1401{
1402        /* non-zero offset changes not possible with v0.90 */
1403        return new_offset == 0;
1404}
1405
1406/*
1407 * version 1 superblock
1408 */
1409
1410static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1411{
1412        __le32 disk_csum;
1413        u32 csum;
1414        unsigned long long newcsum;
1415        int size = 256 + le32_to_cpu(sb->max_dev)*2;
1416        __le32 *isuper = (__le32*)sb;
1417        int i;
1418
1419        disk_csum = sb->sb_csum;
1420        sb->sb_csum = 0;
1421        newcsum = 0;
1422        for (i=0; size>=4; size -= 4 )
1423                newcsum += le32_to_cpu(*isuper++);
1424
1425        if (size == 2)
1426                newcsum += le16_to_cpu(*(__le16*) isuper);
1427
1428        csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1429        sb->sb_csum = disk_csum;
1430        return cpu_to_le32(csum);
1431}
1432
1433static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1434                            int acknowledged);
1435static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1436{
1437        struct mdp_superblock_1 *sb;
1438        int ret;
1439        sector_t sb_start;
1440        sector_t sectors;
1441        char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1442        int bmask;
1443
1444        /*
1445         * Calculate the position of the superblock in 512byte sectors.
1446         * It is always aligned to a 4K boundary and
1447         * depeding on minor_version, it can be:
1448         * 0: At least 8K, but less than 12K, from end of device
1449         * 1: At start of device
1450         * 2: 4K from start of device.
1451         */
1452        switch(minor_version) {
1453        case 0:
1454                sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1455                sb_start -= 8*2;
1456                sb_start &= ~(sector_t)(4*2-1);
1457                break;
1458        case 1:
1459                sb_start = 0;
1460                break;
1461        case 2:
1462                sb_start = 8;
1463                break;
1464        default:
1465                return -EINVAL;
1466        }
1467        rdev->sb_start = sb_start;
1468
1469        /* superblock is rarely larger than 1K, but it can be larger,
1470         * and it is safe to read 4k, so we do that
1471         */
1472        ret = read_disk_sb(rdev, 4096);
1473        if (ret) return ret;
1474
1475
1476        sb = page_address(rdev->sb_page);
1477
1478        if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1479            sb->major_version != cpu_to_le32(1) ||
1480            le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1481            le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1482            (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1483                return -EINVAL;
1484
1485        if (calc_sb_1_csum(sb) != sb->sb_csum) {
1486                printk("md: invalid superblock checksum on %s\n",
1487                        bdevname(rdev->bdev,b));
1488                return -EINVAL;
1489        }
1490        if (le64_to_cpu(sb->data_size) < 10) {
1491                printk("md: data_size too small on %s\n",
1492                       bdevname(rdev->bdev,b));
1493                return -EINVAL;
1494        }
1495        if (sb->pad0 ||
1496            sb->pad3[0] ||
1497            memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1498                /* Some padding is non-zero, might be a new feature */
1499                return -EINVAL;
1500
1501        rdev->preferred_minor = 0xffff;
1502        rdev->data_offset = le64_to_cpu(sb->data_offset);
1503        rdev->new_data_offset = rdev->data_offset;
1504        if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1505            (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1506                rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1507        atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1508
1509        rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1510        bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1511        if (rdev->sb_size & bmask)
1512                rdev->sb_size = (rdev->sb_size | bmask) + 1;
1513
1514        if (minor_version
1515            && rdev->data_offset < sb_start + (rdev->sb_size/512))
1516                return -EINVAL;
1517        if (minor_version
1518            && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1519                return -EINVAL;
1520
1521        if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1522                rdev->desc_nr = -1;
1523        else
1524                rdev->desc_nr = le32_to_cpu(sb->dev_number);
1525
1526        if (!rdev->bb_page) {
1527                rdev->bb_page = alloc_page(GFP_KERNEL);
1528                if (!rdev->bb_page)
1529                        return -ENOMEM;
1530        }
1531        if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1532            rdev->badblocks.count == 0) {
1533                /* need to load the bad block list.
1534                 * Currently we limit it to one page.
1535                 */
1536                s32 offset;
1537                sector_t bb_sector;
1538                u64 *bbp;
1539                int i;
1540                int sectors = le16_to_cpu(sb->bblog_size);
1541                if (sectors > (PAGE_SIZE / 512))
1542                        return -EINVAL;
1543                offset = le32_to_cpu(sb->bblog_offset);
1544                if (offset == 0)
1545                        return -EINVAL;
1546                bb_sector = (long long)offset;
1547                if (!sync_page_io(rdev, bb_sector, sectors << 9,
1548                                  rdev->bb_page, READ, true))
1549                        return -EIO;
1550                bbp = (u64 *)page_address(rdev->bb_page);
1551                rdev->badblocks.shift = sb->bblog_shift;
1552                for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1553                        u64 bb = le64_to_cpu(*bbp);
1554                        int count = bb & (0x3ff);
1555                        u64 sector = bb >> 10;
1556                        sector <<= sb->bblog_shift;
1557                        count <<= sb->bblog_shift;
1558                        if (bb + 1 == 0)
1559                                break;
1560                        if (md_set_badblocks(&rdev->badblocks,
1561                                             sector, count, 1) == 0)
1562                                return -EINVAL;
1563                }
1564        } else if (sb->bblog_offset == 0)
1565                rdev->badblocks.shift = -1;
1566
1567        if (!refdev) {
1568                ret = 1;
1569        } else {
1570                __u64 ev1, ev2;
1571                struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1572
1573                if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1574                    sb->level != refsb->level ||
1575                    sb->layout != refsb->layout ||
1576                    sb->chunksize != refsb->chunksize) {
1577                        printk(KERN_WARNING "md: %s has strangely different"
1578                                " superblock to %s\n",
1579                                bdevname(rdev->bdev,b),
1580                                bdevname(refdev->bdev,b2));
1581                        return -EINVAL;
1582                }
1583                ev1 = le64_to_cpu(sb->events);
1584                ev2 = le64_to_cpu(refsb->events);
1585
1586                if (ev1 > ev2)
1587                        ret = 1;
1588                else
1589                        ret = 0;
1590        }
1591        if (minor_version) {
1592                sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1593                sectors -= rdev->data_offset;
1594        } else
1595                sectors = rdev->sb_start;
1596        if (sectors < le64_to_cpu(sb->data_size))
1597                return -EINVAL;
1598        rdev->sectors = le64_to_cpu(sb->data_size);
1599        return ret;
1600}
1601
1602static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1603{
1604        struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1605        __u64 ev1 = le64_to_cpu(sb->events);
1606
1607        rdev->raid_disk = -1;
1608        clear_bit(Faulty, &rdev->flags);
1609        clear_bit(In_sync, &rdev->flags);
1610        clear_bit(WriteMostly, &rdev->flags);
1611
1612        if (mddev->raid_disks == 0) {
1613                mddev->major_version = 1;
1614                mddev->patch_version = 0;
1615                mddev->external = 0;
1616                mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1617                mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1618                mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1619                mddev->level = le32_to_cpu(sb->level);
1620                mddev->clevel[0] = 0;
1621                mddev->layout = le32_to_cpu(sb->layout);
1622                mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1623                mddev->dev_sectors = le64_to_cpu(sb->size);
1624                mddev->events = ev1;
1625                mddev->bitmap_info.offset = 0;
1626                mddev->bitmap_info.space = 0;
1627                /* Default location for bitmap is 1K after superblock
1628                 * using 3K - total of 4K
1629                 */
1630                mddev->bitmap_info.default_offset = 1024 >> 9;
1631                mddev->bitmap_info.default_space = (4096-1024) >> 9;
1632                mddev->reshape_backwards = 0;
1633
1634                mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1635                memcpy(mddev->uuid, sb->set_uuid, 16);
1636
1637                mddev->max_disks =  (4096-256)/2;
1638
1639                if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1640                    mddev->bitmap_info.file == NULL) {
1641                        mddev->bitmap_info.offset =
1642                                (__s32)le32_to_cpu(sb->bitmap_offset);
1643                        /* Metadata doesn't record how much space is available.
1644                         * For 1.0, we assume we can use up to the superblock
1645                         * if before, else to 4K beyond superblock.
1646                         * For others, assume no change is possible.
1647                         */
1648                        if (mddev->minor_version > 0)
1649                                mddev->bitmap_info.space = 0;
1650                        else if (mddev->bitmap_info.offset > 0)
1651                                mddev->bitmap_info.space =
1652                                        8 - mddev->bitmap_info.offset;
1653                        else
1654                                mddev->bitmap_info.space =
1655                                        -mddev->bitmap_info.offset;
1656                }
1657
1658                if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1659                        mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1660                        mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1661                        mddev->new_level = le32_to_cpu(sb->new_level);
1662                        mddev->new_layout = le32_to_cpu(sb->new_layout);
1663                        mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1664                        if (mddev->delta_disks < 0 ||
1665                            (mddev->delta_disks == 0 &&
1666                             (le32_to_cpu(sb->feature_map)
1667                              & MD_FEATURE_RESHAPE_BACKWARDS)))
1668                                mddev->reshape_backwards = 1;
1669                } else {
1670                        mddev->reshape_position = MaxSector;
1671                        mddev->delta_disks = 0;
1672                        mddev->new_level = mddev->level;
1673                        mddev->new_layout = mddev->layout;
1674                        mddev->new_chunk_sectors = mddev->chunk_sectors;
1675                }
1676
1677        } else if (mddev->pers == NULL) {
1678                /* Insist of good event counter while assembling, except for
1679                 * spares (which don't need an event count) */
1680                ++ev1;
1681                if (rdev->desc_nr >= 0 &&
1682                    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1683                    le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1684                        if (ev1 < mddev->events)
1685                                return -EINVAL;
1686        } else if (mddev->bitmap) {
1687                /* If adding to array with a bitmap, then we can accept an
1688                 * older device, but not too old.
1689                 */
1690                if (ev1 < mddev->bitmap->events_cleared)
1691                        return 0;
1692        } else {
1693                if (ev1 < mddev->events)
1694                        /* just a hot-add of a new device, leave raid_disk at -1 */
1695                        return 0;
1696        }
1697        if (mddev->level != LEVEL_MULTIPATH) {
1698                int role;
1699                if (rdev->desc_nr < 0 ||
1700                    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1701                        role = 0xffff;
1702                        rdev->desc_nr = -1;
1703                } else
1704                        role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1705                switch(role) {
1706                case 0xffff: /* spare */
1707                        break;
1708                case 0xfffe: /* faulty */
1709                        set_bit(Faulty, &rdev->flags);
1710                        break;
1711                default:
1712                        if ((le32_to_cpu(sb->feature_map) &
1713                             MD_FEATURE_RECOVERY_OFFSET))
1714                                rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1715                        else
1716                                set_bit(In_sync, &rdev->flags);
1717                        rdev->raid_disk = role;
1718                        break;
1719                }
1720                if (sb->devflags & WriteMostly1)
1721                        set_bit(WriteMostly, &rdev->flags);
1722                if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1723                        set_bit(Replacement, &rdev->flags);
1724        } else /* MULTIPATH are always insync */
1725                set_bit(In_sync, &rdev->flags);
1726
1727        return 0;
1728}
1729
1730static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1731{
1732        struct mdp_superblock_1 *sb;
1733        struct md_rdev *rdev2;
1734        int max_dev, i;
1735        /* make rdev->sb match mddev and rdev data. */
1736
1737        sb = page_address(rdev->sb_page);
1738
1739        sb->feature_map = 0;
1740        sb->pad0 = 0;
1741        sb->recovery_offset = cpu_to_le64(0);
1742        memset(sb->pad3, 0, sizeof(sb->pad3));
1743
1744        sb->utime = cpu_to_le64((__u64)mddev->utime);
1745        sb->events = cpu_to_le64(mddev->events);
1746        if (mddev->in_sync)
1747                sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1748        else
1749                sb->resync_offset = cpu_to_le64(0);
1750
1751        sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1752
1753        sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1754        sb->size = cpu_to_le64(mddev->dev_sectors);
1755        sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1756        sb->level = cpu_to_le32(mddev->level);
1757        sb->layout = cpu_to_le32(mddev->layout);
1758
1759        if (test_bit(WriteMostly, &rdev->flags))
1760                sb->devflags |= WriteMostly1;
1761        else
1762                sb->devflags &= ~WriteMostly1;
1763        sb->data_offset = cpu_to_le64(rdev->data_offset);
1764        sb->data_size = cpu_to_le64(rdev->sectors);
1765
1766        if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1767                sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1768                sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1769        }
1770
1771        if (rdev->raid_disk >= 0 &&
1772            !test_bit(In_sync, &rdev->flags)) {
1773                sb->feature_map |=
1774                        cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1775                sb->recovery_offset =
1776                        cpu_to_le64(rdev->recovery_offset);
1777        }
1778        if (test_bit(Replacement, &rdev->flags))
1779                sb->feature_map |=
1780                        cpu_to_le32(MD_FEATURE_REPLACEMENT);
1781
1782        if (mddev->reshape_position != MaxSector) {
1783                sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1784                sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1785                sb->new_layout = cpu_to_le32(mddev->new_layout);
1786                sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1787                sb->new_level = cpu_to_le32(mddev->new_level);
1788                sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1789                if (mddev->delta_disks == 0 &&
1790                    mddev->reshape_backwards)
1791                        sb->feature_map
1792                                |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1793                if (rdev->new_data_offset != rdev->data_offset) {
1794                        sb->feature_map
1795                                |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1796                        sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1797                                                             - rdev->data_offset));
1798                }
1799        }
1800
1801        if (rdev->badblocks.count == 0)
1802                /* Nothing to do for bad blocks*/ ;
1803        else if (sb->bblog_offset == 0)
1804                /* Cannot record bad blocks on this device */
1805                md_error(mddev, rdev);
1806        else {
1807                struct badblocks *bb = &rdev->badblocks;
1808                u64 *bbp = (u64 *)page_address(rdev->bb_page);
1809                u64 *p = bb->page;
1810                sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1811                if (bb->changed) {
1812                        unsigned seq;
1813
1814retry:
1815                        seq = read_seqbegin(&bb->lock);
1816
1817                        memset(bbp, 0xff, PAGE_SIZE);
1818
1819                        for (i = 0 ; i < bb->count ; i++) {
1820                                u64 internal_bb = p[i];
1821                                u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1822                                                | BB_LEN(internal_bb));
1823                                bbp[i] = cpu_to_le64(store_bb);
1824                        }
1825                        bb->changed = 0;
1826                        if (read_seqretry(&bb->lock, seq))
1827                                goto retry;
1828
1829                        bb->sector = (rdev->sb_start +
1830                                      (int)le32_to_cpu(sb->bblog_offset));
1831                        bb->size = le16_to_cpu(sb->bblog_size);
1832                }
1833        }
1834
1835        max_dev = 0;
1836        rdev_for_each(rdev2, mddev)
1837                if (rdev2->desc_nr+1 > max_dev)
1838                        max_dev = rdev2->desc_nr+1;
1839
1840        if (max_dev > le32_to_cpu(sb->max_dev)) {
1841                int bmask;
1842                sb->max_dev = cpu_to_le32(max_dev);
1843                rdev->sb_size = max_dev * 2 + 256;
1844                bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1845                if (rdev->sb_size & bmask)
1846                        rdev->sb_size = (rdev->sb_size | bmask) + 1;
1847        } else
1848                max_dev = le32_to_cpu(sb->max_dev);
1849
1850        for (i=0; i<max_dev;i++)
1851                sb->dev_roles[i] = cpu_to_le16(0xfffe);
1852        
1853        rdev_for_each(rdev2, mddev) {
1854                i = rdev2->desc_nr;
1855                if (test_bit(Faulty, &rdev2->flags))
1856                        sb->dev_roles[i] = cpu_to_le16(0xfffe);
1857                else if (test_bit(In_sync, &rdev2->flags))
1858                        sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1859                else if (rdev2->raid_disk >= 0)
1860                        sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1861                else
1862                        sb->dev_roles[i] = cpu_to_le16(0xffff);
1863        }
1864
1865        sb->sb_csum = calc_sb_1_csum(sb);
1866}
1867
1868static unsigned long long
1869super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1870{
1871        struct mdp_superblock_1 *sb;
1872        sector_t max_sectors;
1873        if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1874                return 0; /* component must fit device */
1875        if (rdev->data_offset != rdev->new_data_offset)
1876                return 0; /* too confusing */
1877        if (rdev->sb_start < rdev->data_offset) {
1878                /* minor versions 1 and 2; superblock before data */
1879                max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1880                max_sectors -= rdev->data_offset;
1881                if (!num_sectors || num_sectors > max_sectors)
1882                        num_sectors = max_sectors;
1883        } else if (rdev->mddev->bitmap_info.offset) {
1884                /* minor version 0 with bitmap we can't move */
1885                return 0;
1886        } else {
1887                /* minor version 0; superblock after data */
1888                sector_t sb_start;
1889                sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1890                sb_start &= ~(sector_t)(4*2 - 1);
1891                max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1892                if (!num_sectors || num_sectors > max_sectors)
1893                        num_sectors = max_sectors;
1894                rdev->sb_start = sb_start;
1895        }
1896        sb = page_address(rdev->sb_page);
1897        sb->data_size = cpu_to_le64(num_sectors);
1898        sb->super_offset = rdev->sb_start;
1899        sb->sb_csum = calc_sb_1_csum(sb);
1900        md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1901                       rdev->sb_page);
1902        md_super_wait(rdev->mddev);
1903        return num_sectors;
1904
1905}
1906
1907static int
1908super_1_allow_new_offset(struct md_rdev *rdev,
1909                         unsigned long long new_offset)
1910{
1911        /* All necessary checks on new >= old have been done */
1912        struct bitmap *bitmap;
1913        if (new_offset >= rdev->data_offset)
1914                return 1;
1915
1916        /* with 1.0 metadata, there is no metadata to tread on
1917         * so we can always move back */
1918        if (rdev->mddev->minor_version == 0)
1919                return 1;
1920
1921        /* otherwise we must be sure not to step on
1922         * any metadata, so stay:
1923         * 36K beyond start of superblock
1924         * beyond end of badblocks
1925         * beyond write-intent bitmap
1926         */
1927        if (rdev->sb_start + (32+4)*2 > new_offset)
1928                return 0;
1929        bitmap = rdev->mddev->bitmap;
1930        if (bitmap && !rdev->mddev->bitmap_info.file &&
1931            rdev->sb_start + rdev->mddev->bitmap_info.offset +
1932            bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1933                return 0;
1934        if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1935                return 0;
1936
1937        return 1;
1938}
1939
1940static struct super_type super_types[] = {
1941        [0] = {
1942                .name   = "0.90.0",
1943                .owner  = THIS_MODULE,
1944                .load_super         = super_90_load,
1945                .validate_super     = super_90_validate,
1946                .sync_super         = super_90_sync,
1947                .rdev_size_change   = super_90_rdev_size_change,
1948                .allow_new_offset   = super_90_allow_new_offset,
1949        },
1950        [1] = {
1951                .name   = "md-1",
1952                .owner  = THIS_MODULE,
1953                .load_super         = super_1_load,
1954                .validate_super     = super_1_validate,
1955                .sync_super         = super_1_sync,
1956                .rdev_size_change   = super_1_rdev_size_change,
1957                .allow_new_offset   = super_1_allow_new_offset,
1958        },
1959};
1960
1961static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1962{
1963        if (mddev->sync_super) {
1964                mddev->sync_super(mddev, rdev);
1965                return;
1966        }
1967
1968        BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1969
1970        super_types[mddev->major_version].sync_super(mddev, rdev);
1971}
1972
1973static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1974{
1975        struct md_rdev *rdev, *rdev2;
1976
1977        rcu_read_lock();
1978        rdev_for_each_rcu(rdev, mddev1)
1979                rdev_for_each_rcu(rdev2, mddev2)
1980                        if (rdev->bdev->bd_contains ==
1981                            rdev2->bdev->bd_contains) {
1982                                rcu_read_unlock();
1983                                return 1;
1984                        }
1985        rcu_read_unlock();
1986        return 0;
1987}
1988
1989static LIST_HEAD(pending_raid_disks);
1990
1991/*
1992 * Try to register data integrity profile for an mddev
1993 *
1994 * This is called when an array is started and after a disk has been kicked
1995 * from the array. It only succeeds if all working and active component devices
1996 * are integrity capable with matching profiles.
1997 */
1998int md_integrity_register(struct mddev *mddev)
1999{
2000        struct md_rdev *rdev, *reference = NULL;
2001
2002        if (list_empty(&mddev->disks))
2003                return 0; /* nothing to do */
2004        if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2005                return 0; /* shouldn't register, or already is */
2006        rdev_for_each(rdev, mddev) {
2007                /* skip spares and non-functional disks */
2008                if (test_bit(Faulty, &rdev->flags))
2009                        continue;
2010                if (rdev->raid_disk < 0)
2011                        continue;
2012                if (!reference) {
2013                        /* Use the first rdev as the reference */
2014                        reference = rdev;
2015                        continue;
2016                }
2017                /* does this rdev's profile match the reference profile? */
2018                if (blk_integrity_compare(reference->bdev->bd_disk,
2019                                rdev->bdev->bd_disk) < 0)
2020                        return -EINVAL;
2021        }
2022        if (!reference || !bdev_get_integrity(reference->bdev))
2023                return 0;
2024        /*
2025         * All component devices are integrity capable and have matching
2026         * profiles, register the common profile for the md device.
2027         */
2028        if (blk_integrity_register(mddev->gendisk,
2029                        bdev_get_integrity(reference->bdev)) != 0) {
2030                printk(KERN_ERR "md: failed to register integrity for %s\n",
2031                        mdname(mddev));
2032                return -EINVAL;
2033        }
2034        printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2035        if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2036                printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2037                       mdname(mddev));
2038                return -EINVAL;
2039        }
2040        return 0;
2041}
2042EXPORT_SYMBOL(md_integrity_register);
2043
2044/* Disable data integrity if non-capable/non-matching disk is being added */
2045void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2046{
2047        struct blk_integrity *bi_rdev;
2048        struct blk_integrity *bi_mddev;
2049
2050        if (!mddev->gendisk)
2051                return;
2052
2053        bi_rdev = bdev_get_integrity(rdev->bdev);
2054        bi_mddev = blk_get_integrity(mddev->gendisk);
2055
2056        if (!bi_mddev) /* nothing to do */
2057                return;
2058        if (rdev->raid_disk < 0) /* skip spares */
2059                return;
2060        if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2061                                             rdev->bdev->bd_disk) >= 0)
2062                return;
2063        printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2064        blk_integrity_unregister(mddev->gendisk);
2065}
2066EXPORT_SYMBOL(md_integrity_add_rdev);
2067
2068static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2069{
2070        char b[BDEVNAME_SIZE];
2071        struct kobject *ko;
2072        char *s;
2073        int err;
2074
2075        if (rdev->mddev) {
2076                MD_BUG();
2077                return -EINVAL;
2078        }
2079
2080        /* prevent duplicates */
2081        if (find_rdev(mddev, rdev->bdev->bd_dev))
2082                return -EEXIST;
2083
2084        /* make sure rdev->sectors exceeds mddev->dev_sectors */
2085        if (rdev->sectors && (mddev->dev_sectors == 0 ||
2086                        rdev->sectors < mddev->dev_sectors)) {
2087                if (mddev->pers) {
2088                        /* Cannot change size, so fail
2089                         * If mddev->level <= 0, then we don't care
2090                         * about aligning sizes (e.g. linear)
2091                         */
2092                        if (mddev->level > 0)
2093                                return -ENOSPC;
2094                } else
2095                        mddev->dev_sectors = rdev->sectors;
2096        }
2097
2098        /* Verify rdev->desc_nr is unique.
2099         * If it is -1, assign a free number, else
2100         * check number is not in use
2101         */
2102        if (rdev->desc_nr < 0) {
2103                int choice = 0;
2104                if (mddev->pers) choice = mddev->raid_disks;
2105                while (find_rdev_nr(mddev, choice))
2106                        choice++;
2107                rdev->desc_nr = choice;
2108        } else {
2109                if (find_rdev_nr(mddev, rdev->desc_nr))
2110                        return -EBUSY;
2111        }
2112        if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2113                printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2114                       mdname(mddev), mddev->max_disks);
2115                return -EBUSY;
2116        }
2117        bdevname(rdev->bdev,b);
2118        while ( (s=strchr(b, '/')) != NULL)
2119                *s = '!';
2120
2121        rdev->mddev = mddev;
2122        printk(KERN_INFO "md: bind<%s>\n", b);
2123
2124        if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2125                goto fail;
2126
2127        ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2128        if (sysfs_create_link(&rdev->kobj, ko, "block"))
2129                /* failure here is OK */;
2130        rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2131
2132        list_add_rcu(&rdev->same_set, &mddev->disks);
2133        bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2134
2135        /* May as well allow recovery to be retried once */
2136        mddev->recovery_disabled++;
2137
2138        return 0;
2139
2140 fail:
2141        printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2142               b, mdname(mddev));
2143        return err;
2144}
2145
2146static void md_delayed_delete(struct work_struct *ws)
2147{
2148        struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2149        kobject_del(&rdev->kobj);
2150        kobject_put(&rdev->kobj);
2151}
2152
2153static void unbind_rdev_from_array(struct md_rdev * rdev)
2154{
2155        char b[BDEVNAME_SIZE];
2156        if (!rdev->mddev) {
2157                MD_BUG();
2158                return;
2159        }
2160        bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2161        list_del_rcu(&rdev->same_set);
2162        printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2163        rdev->mddev = NULL;
2164        sysfs_remove_link(&rdev->kobj, "block");
2165        sysfs_put(rdev->sysfs_state);
2166        rdev->sysfs_state = NULL;
2167        rdev->badblocks.count = 0;
2168        /* We need to delay this, otherwise we can deadlock when
2169         * writing to 'remove' to "dev/state".  We also need
2170         * to delay it due to rcu usage.
2171         */
2172        synchronize_rcu();
2173        INIT_WORK(&rdev->del_work, md_delayed_delete);
2174        kobject_get(&rdev->kobj);
2175        queue_work(md_misc_wq, &rdev->del_work);
2176}
2177
2178/*
2179 * prevent the device from being mounted, repartitioned or
2180 * otherwise reused by a RAID array (or any other kernel
2181 * subsystem), by bd_claiming the device.
2182 */
2183static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2184{
2185        int err = 0;
2186        struct block_device *bdev;
2187        char b[BDEVNAME_SIZE];
2188
2189        bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2190                                 shared ? (struct md_rdev *)lock_rdev : rdev);
2191        if (IS_ERR(bdev)) {
2192                printk(KERN_ERR "md: could not open %s.\n",
2193                        __bdevname(dev, b));
2194                return PTR_ERR(bdev);
2195        }
2196        rdev->bdev = bdev;
2197        return err;
2198}
2199
2200static void unlock_rdev(struct md_rdev *rdev)
2201{
2202        struct block_device *bdev = rdev->bdev;
2203        rdev->bdev = NULL;
2204        if (!bdev)
2205                MD_BUG();
2206        blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2207}
2208
2209void md_autodetect_dev(dev_t dev);
2210
2211static void export_rdev(struct md_rdev * rdev)
2212{
2213        char b[BDEVNAME_SIZE];
2214        printk(KERN_INFO "md: export_rdev(%s)\n",
2215                bdevname(rdev->bdev,b));
2216        if (rdev->mddev)
2217                MD_BUG();
2218        md_rdev_clear(rdev);
2219#ifndef MODULE
2220        if (test_bit(AutoDetected, &rdev->flags))
2221                md_autodetect_dev(rdev->bdev->bd_dev);
2222#endif
2223        unlock_rdev(rdev);
2224        kobject_put(&rdev->kobj);
2225}
2226
2227static void kick_rdev_from_array(struct md_rdev * rdev)
2228{
2229        unbind_rdev_from_array(rdev);
2230        export_rdev(rdev);
2231}
2232
2233static void export_array(struct mddev *mddev)
2234{
2235        struct md_rdev *rdev, *tmp;
2236
2237        rdev_for_each_safe(rdev, tmp, mddev) {
2238                if (!rdev->mddev) {
2239                        MD_BUG();
2240                        continue;
2241                }
2242                kick_rdev_from_array(rdev);
2243        }
2244        if (!list_empty(&mddev->disks))
2245                MD_BUG();
2246        mddev->raid_disks = 0;
2247        mddev->major_version = 0;
2248}
2249
2250static void print_desc(mdp_disk_t *desc)
2251{
2252        printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2253                desc->major,desc->minor,desc->raid_disk,desc->state);
2254}
2255
2256static void print_sb_90(mdp_super_t *sb)
2257{
2258        int i;
2259
2260        printk(KERN_INFO 
2261                "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2262                sb->major_version, sb->minor_version, sb->patch_version,
2263                sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2264                sb->ctime);
2265        printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2266                sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2267                sb->md_minor, sb->layout, sb->chunk_size);
2268        printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2269                " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2270                sb->utime, sb->state, sb->active_disks, sb->working_disks,
2271                sb->failed_disks, sb->spare_disks,
2272                sb->sb_csum, (unsigned long)sb->events_lo);
2273
2274        printk(KERN_INFO);
2275        for (i = 0; i < MD_SB_DISKS; i++) {
2276                mdp_disk_t *desc;
2277
2278                desc = sb->disks + i;
2279                if (desc->number || desc->major || desc->minor ||
2280                    desc->raid_disk || (desc->state && (desc->state != 4))) {
2281                        printk("     D %2d: ", i);
2282                        print_desc(desc);
2283                }
2284        }
2285        printk(KERN_INFO "md:     THIS: ");
2286        print_desc(&sb->this_disk);
2287}
2288
2289static void print_sb_1(struct mdp_superblock_1 *sb)
2290{
2291        __u8 *uuid;
2292
2293        uuid = sb->set_uuid;
2294        printk(KERN_INFO
2295               "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2296               "md:    Name: \"%s\" CT:%llu\n",
2297                le32_to_cpu(sb->major_version),
2298                le32_to_cpu(sb->feature_map),
2299                uuid,
2300                sb->set_name,
2301                (unsigned long long)le64_to_cpu(sb->ctime)
2302                       & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2303
2304        uuid = sb->device_uuid;
2305        printk(KERN_INFO
2306               "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2307                        " RO:%llu\n"
2308               "md:     Dev:%08x UUID: %pU\n"
2309               "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2310               "md:         (MaxDev:%u) \n",
2311                le32_to_cpu(sb->level),
2312                (unsigned long long)le64_to_cpu(sb->size),
2313                le32_to_cpu(sb->raid_disks),
2314                le32_to_cpu(sb->layout),
2315                le32_to_cpu(sb->chunksize),
2316                (unsigned long long)le64_to_cpu(sb->data_offset),
2317                (unsigned long long)le64_to_cpu(sb->data_size),
2318                (unsigned long long)le64_to_cpu(sb->super_offset),
2319                (unsigned long long)le64_to_cpu(sb->recovery_offset),
2320                le32_to_cpu(sb->dev_number),
2321                uuid,
2322                sb->devflags,
2323                (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2324                (unsigned long long)le64_to_cpu(sb->events),
2325                (unsigned long long)le64_to_cpu(sb->resync_offset),
2326                le32_to_cpu(sb->sb_csum),
2327                le32_to_cpu(sb->max_dev)
2328                );
2329}
2330
2331static void print_rdev(struct md_rdev *rdev, int major_version)
2332{
2333        char b[BDEVNAME_SIZE];
2334        printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2335                bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2336                test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2337                rdev->desc_nr);
2338        if (rdev->sb_loaded) {
2339                printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2340                switch (major_version) {
2341                case 0:
2342                        print_sb_90(page_address(rdev->sb_page));
2343                        break;
2344                case 1:
2345                        print_sb_1(page_address(rdev->sb_page));
2346                        break;
2347                }
2348        } else
2349                printk(KERN_INFO "md: no rdev superblock!\n");
2350}
2351
2352static void md_print_devices(void)
2353{
2354        struct list_head *tmp;
2355        struct md_rdev *rdev;
2356        struct mddev *mddev;
2357        char b[BDEVNAME_SIZE];
2358
2359        printk("\n");
2360        printk("md:     **********************************\n");
2361        printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2362        printk("md:     **********************************\n");
2363        for_each_mddev(mddev, tmp) {
2364
2365                if (mddev->bitmap)
2366                        bitmap_print_sb(mddev->bitmap);
2367                else
2368                        printk("%s: ", mdname(mddev));
2369                rdev_for_each(rdev, mddev)
2370                        printk("<%s>", bdevname(rdev->bdev,b));
2371                printk("\n");
2372
2373                rdev_for_each(rdev, mddev)
2374                        print_rdev(rdev, mddev->major_version);
2375        }
2376        printk("md:     **********************************\n");
2377        printk("\n");
2378}
2379
2380
2381static void sync_sbs(struct mddev * mddev, int nospares)
2382{
2383        /* Update each superblock (in-memory image), but
2384         * if we are allowed to, skip spares which already
2385         * have the right event counter, or have one earlier
2386         * (which would mean they aren't being marked as dirty
2387         * with the rest of the array)
2388         */
2389        struct md_rdev *rdev;
2390        rdev_for_each(rdev, mddev) {
2391                if (rdev->sb_events == mddev->events ||
2392                    (nospares &&
2393                     rdev->raid_disk < 0 &&
2394                     rdev->sb_events+1 == mddev->events)) {
2395                        /* Don't update this superblock */
2396                        rdev->sb_loaded = 2;
2397                } else {
2398                        sync_super(mddev, rdev);
2399                        rdev->sb_loaded = 1;
2400                }
2401        }
2402}
2403
2404static void md_update_sb(struct mddev * mddev, int force_change)
2405{
2406        struct md_rdev *rdev;
2407        int sync_req;
2408        int nospares = 0;
2409        int any_badblocks_changed = 0;
2410
2411repeat:
2412        /* First make sure individual recovery_offsets are correct */
2413        rdev_for_each(rdev, mddev) {
2414                if (rdev->raid_disk >= 0 &&
2415                    mddev->delta_disks >= 0 &&
2416                    !test_bit(In_sync, &rdev->flags) &&
2417                    mddev->curr_resync_completed > rdev->recovery_offset)
2418                                rdev->recovery_offset = mddev->curr_resync_completed;
2419
2420        }       
2421        if (!mddev->persistent) {
2422                clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2423                clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2424                if (!mddev->external) {
2425                        clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2426                        rdev_for_each(rdev, mddev) {
2427                                if (rdev->badblocks.changed) {
2428                                        rdev->badblocks.changed = 0;
2429                                        md_ack_all_badblocks(&rdev->badblocks);
2430                                        md_error(mddev, rdev);
2431                                }
2432                                clear_bit(Blocked, &rdev->flags);
2433                                clear_bit(BlockedBadBlocks, &rdev->flags);
2434                                wake_up(&rdev->blocked_wait);
2435                        }
2436                }
2437                wake_up(&mddev->sb_wait);
2438                return;
2439        }
2440
2441        spin_lock_irq(&mddev->write_lock);
2442
2443        mddev->utime = get_seconds();
2444
2445        if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2446                force_change = 1;
2447        if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2448                /* just a clean<-> dirty transition, possibly leave spares alone,
2449                 * though if events isn't the right even/odd, we will have to do
2450                 * spares after all
2451                 */
2452                nospares = 1;
2453        if (force_change)
2454                nospares = 0;
2455        if (mddev->degraded)
2456                /* If the array is degraded, then skipping spares is both
2457                 * dangerous and fairly pointless.
2458                 * Dangerous because a device that was removed from the array
2459                 * might have a event_count that still looks up-to-date,
2460                 * so it can be re-added without a resync.
2461                 * Pointless because if there are any spares to skip,
2462                 * then a recovery will happen and soon that array won't
2463                 * be degraded any more and the spare can go back to sleep then.
2464                 */
2465                nospares = 0;
2466
2467        sync_req = mddev->in_sync;
2468
2469        /* If this is just a dirty<->clean transition, and the array is clean
2470         * and 'events' is odd, we can roll back to the previous clean state */
2471        if (nospares
2472            && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2473            && mddev->can_decrease_events
2474            && mddev->events != 1) {
2475                mddev->events--;
2476                mddev->can_decrease_events = 0;
2477        } else {
2478                /* otherwise we have to go forward and ... */
2479                mddev->events ++;
2480                mddev->can_decrease_events = nospares;
2481        }
2482
2483        if (!mddev->events) {
2484                /*
2485                 * oops, this 64-bit counter should never wrap.
2486                 * Either we are in around ~1 trillion A.C., assuming
2487                 * 1 reboot per second, or we have a bug:
2488                 */
2489                MD_BUG();
2490                mddev->events --;
2491        }
2492
2493        rdev_for_each(rdev, mddev) {
2494                if (rdev->badblocks.changed)
2495                        any_badblocks_changed++;
2496                if (test_bit(Faulty, &rdev->flags))
2497                        set_bit(FaultRecorded, &rdev->flags);
2498        }
2499
2500        sync_sbs(mddev, nospares);
2501        spin_unlock_irq(&mddev->write_lock);
2502
2503        pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2504                 mdname(mddev), mddev->in_sync);
2505
2506        bitmap_update_sb(mddev->bitmap);
2507        rdev_for_each(rdev, mddev) {
2508                char b[BDEVNAME_SIZE];
2509
2510                if (rdev->sb_loaded != 1)
2511                        continue; /* no noise on spare devices */
2512
2513                if (!test_bit(Faulty, &rdev->flags) &&
2514                    rdev->saved_raid_disk == -1) {
2515                        md_super_write(mddev,rdev,
2516                                       rdev->sb_start, rdev->sb_size,
2517                                       rdev->sb_page);
2518                        pr_debug("md: (write) %s's sb offset: %llu\n",
2519                                 bdevname(rdev->bdev, b),
2520                                 (unsigned long long)rdev->sb_start);
2521                        rdev->sb_events = mddev->events;
2522                        if (rdev->badblocks.size) {
2523                                md_super_write(mddev, rdev,
2524                                               rdev->badblocks.sector,
2525                                               rdev->badblocks.size << 9,
2526                                               rdev->bb_page);
2527                                rdev->badblocks.size = 0;
2528                        }
2529
2530                } else if (test_bit(Faulty, &rdev->flags))
2531                        pr_debug("md: %s (skipping faulty)\n",
2532                                 bdevname(rdev->bdev, b));
2533                else
2534                        pr_debug("(skipping incremental s/r ");
2535
2536                if (mddev->level == LEVEL_MULTIPATH)
2537                        /* only need to write one superblock... */
2538                        break;
2539        }
2540        md_super_wait(mddev);
2541        /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2542
2543        spin_lock_irq(&mddev->write_lock);
2544        if (mddev->in_sync != sync_req ||
2545            test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2546                /* have to write it out again */
2547                spin_unlock_irq(&mddev->write_lock);
2548                goto repeat;
2549        }
2550        clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2551        spin_unlock_irq(&mddev->write_lock);
2552        wake_up(&mddev->sb_wait);
2553        if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2554                sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2555
2556        rdev_for_each(rdev, mddev) {
2557                if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2558                        clear_bit(Blocked, &rdev->flags);
2559
2560                if (any_badblocks_changed)
2561                        md_ack_all_badblocks(&rdev->badblocks);
2562                clear_bit(BlockedBadBlocks, &rdev->flags);
2563                wake_up(&rdev->blocked_wait);
2564        }
2565}
2566
2567/* words written to sysfs files may, or may not, be \n terminated.
2568 * We want to accept with case. For this we use cmd_match.
2569 */
2570static int cmd_match(const char *cmd, const char *str)
2571{
2572        /* See if cmd, written into a sysfs file, matches
2573         * str.  They must either be the same, or cmd can
2574         * have a trailing newline
2575         */
2576        while (*cmd && *str && *cmd == *str) {
2577                cmd++;
2578                str++;
2579        }
2580        if (*cmd == '\n')
2581                cmd++;
2582        if (*str || *cmd)
2583                return 0;
2584        return 1;
2585}
2586
2587struct rdev_sysfs_entry {
2588        struct attribute attr;
2589        ssize_t (*show)(struct md_rdev *, char *);
2590        ssize_t (*store)(struct md_rdev *, const char *, size_t);
2591};
2592
2593static ssize_t
2594state_show(struct md_rdev *rdev, char *page)
2595{
2596        char *sep = "";
2597        size_t len = 0;
2598
2599        if (test_bit(Faulty, &rdev->flags) ||
2600            rdev->badblocks.unacked_exist) {
2601                len+= sprintf(page+len, "%sfaulty",sep);
2602                sep = ",";
2603        }
2604        if (test_bit(In_sync, &rdev->flags)) {
2605                len += sprintf(page+len, "%sin_sync",sep);
2606                sep = ",";
2607        }
2608        if (test_bit(WriteMostly, &rdev->flags)) {
2609                len += sprintf(page+len, "%swrite_mostly",sep);
2610                sep = ",";
2611        }
2612        if (test_bit(Blocked, &rdev->flags) ||
2613            (rdev->badblocks.unacked_exist
2614             && !test_bit(Faulty, &rdev->flags))) {
2615                len += sprintf(page+len, "%sblocked", sep);
2616                sep = ",";
2617        }
2618        if (!test_bit(Faulty, &rdev->flags) &&
2619            !test_bit(In_sync, &rdev->flags)) {
2620                len += sprintf(page+len, "%sspare", sep);
2621                sep = ",";
2622        }
2623        if (test_bit(WriteErrorSeen, &rdev->flags)) {
2624                len += sprintf(page+len, "%swrite_error", sep);
2625                sep = ",";
2626        }
2627        if (test_bit(WantReplacement, &rdev->flags)) {
2628                len += sprintf(page+len, "%swant_replacement", sep);
2629                sep = ",";
2630        }
2631        if (test_bit(Replacement, &rdev->flags)) {
2632                len += sprintf(page+len, "%sreplacement", sep);
2633                sep = ",";
2634        }
2635
2636        return len+sprintf(page+len, "\n");
2637}
2638
2639static ssize_t
2640state_store(struct md_rdev *rdev, const char *buf, size_t len)
2641{
2642        /* can write
2643         *  faulty  - simulates an error
2644         *  remove  - disconnects the device
2645         *  writemostly - sets write_mostly
2646         *  -writemostly - clears write_mostly
2647         *  blocked - sets the Blocked flags
2648         *  -blocked - clears the Blocked and possibly simulates an error
2649         *  insync - sets Insync providing device isn't active
2650         *  write_error - sets WriteErrorSeen
2651         *  -write_error - clears WriteErrorSeen
2652         */
2653        int err = -EINVAL;
2654        if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2655                md_error(rdev->mddev, rdev);
2656                if (test_bit(Faulty, &rdev->flags))
2657                        err = 0;
2658                else
2659                        err = -EBUSY;
2660        } else if (cmd_match(buf, "remove")) {
2661                if (rdev->raid_disk >= 0)
2662                        err = -EBUSY;
2663                else {
2664                        struct mddev *mddev = rdev->mddev;
2665                        kick_rdev_from_array(rdev);
2666                        if (mddev->pers)
2667                                md_update_sb(mddev, 1);
2668                        md_new_event(mddev);
2669                        err = 0;
2670                }
2671        } else if (cmd_match(buf, "writemostly")) {
2672                set_bit(WriteMostly, &rdev->flags);
2673                err = 0;
2674        } else if (cmd_match(buf, "-writemostly")) {
2675                clear_bit(WriteMostly, &rdev->flags);
2676                err = 0;
2677        } else if (cmd_match(buf, "blocked")) {
2678                set_bit(Blocked, &rdev->flags);
2679                err = 0;
2680        } else if (cmd_match(buf, "-blocked")) {
2681                if (!test_bit(Faulty, &rdev->flags) &&
2682                    rdev->badblocks.unacked_exist) {
2683                        /* metadata handler doesn't understand badblocks,
2684                         * so we need to fail the device
2685                         */
2686                        md_error(rdev->mddev, rdev);
2687                }
2688                clear_bit(Blocked, &rdev->flags);
2689                clear_bit(BlockedBadBlocks, &rdev->flags);
2690                wake_up(&rdev->blocked_wait);
2691                set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2692                md_wakeup_thread(rdev->mddev->thread);
2693
2694                err = 0;
2695        } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2696                set_bit(In_sync, &rdev->flags);
2697                err = 0;
2698        } else if (cmd_match(buf, "write_error")) {
2699                set_bit(WriteErrorSeen, &rdev->flags);
2700                err = 0;
2701        } else if (cmd_match(buf, "-write_error")) {
2702                clear_bit(WriteErrorSeen, &rdev->flags);
2703                err = 0;
2704        } else if (cmd_match(buf, "want_replacement")) {
2705                /* Any non-spare device that is not a replacement can
2706                 * become want_replacement at any time, but we then need to
2707                 * check if recovery is needed.
2708                 */
2709                if (rdev->raid_disk >= 0 &&
2710                    !test_bit(Replacement, &rdev->flags))
2711                        set_bit(WantReplacement, &rdev->flags);
2712                set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2713                md_wakeup_thread(rdev->mddev->thread);
2714                err = 0;
2715        } else if (cmd_match(buf, "-want_replacement")) {
2716                /* Clearing 'want_replacement' is always allowed.
2717                 * Once replacements starts it is too late though.
2718                 */
2719                err = 0;
2720                clear_bit(WantReplacement, &rdev->flags);
2721        } else if (cmd_match(buf, "replacement")) {
2722                /* Can only set a device as a replacement when array has not
2723                 * yet been started.  Once running, replacement is automatic
2724                 * from spares, or by assigning 'slot'.
2725                 */
2726                if (rdev->mddev->pers)
2727                        err = -EBUSY;
2728                else {
2729                        set_bit(Replacement, &rdev->flags);
2730                        err = 0;
2731                }
2732        } else if (cmd_match(buf, "-replacement")) {
2733                /* Similarly, can only clear Replacement before start */
2734                if (rdev->mddev->pers)
2735                        err = -EBUSY;
2736                else {
2737                        clear_bit(Replacement, &rdev->flags);
2738                        err = 0;
2739                }
2740        }
2741        if (!err)
2742                sysfs_notify_dirent_safe(rdev->sysfs_state);
2743        return err ? err : len;
2744}
2745static struct rdev_sysfs_entry rdev_state =
2746__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2747
2748static ssize_t
2749errors_show(struct md_rdev *rdev, char *page)
2750{
2751        return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2752}
2753
2754static ssize_t
2755errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2756{
2757        char *e;
2758        unsigned long n = simple_strtoul(buf, &e, 10);
2759        if (*buf && (*e == 0 || *e == '\n')) {
2760                atomic_set(&rdev->corrected_errors, n);
2761                return len;
2762        }
2763        return -EINVAL;
2764}
2765static struct rdev_sysfs_entry rdev_errors =
2766__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2767
2768static ssize_t
2769slot_show(struct md_rdev *rdev, char *page)
2770{
2771        if (rdev->raid_disk < 0)
2772                return sprintf(page, "none\n");
2773        else
2774                return sprintf(page, "%d\n", rdev->raid_disk);
2775}
2776
2777static ssize_t
2778slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2779{
2780        char *e;
2781        int err;
2782        int slot = simple_strtoul(buf, &e, 10);
2783        if (strncmp(buf, "none", 4)==0)
2784                slot = -1;
2785        else if (e==buf || (*e && *e!= '\n'))
2786                return -EINVAL;
2787        if (rdev->mddev->pers && slot == -1) {
2788                /* Setting 'slot' on an active array requires also
2789                 * updating the 'rd%d' link, and communicating
2790                 * with the personality with ->hot_*_disk.
2791                 * For now we only support removing
2792                 * failed/spare devices.  This normally happens automatically,
2793                 * but not when the metadata is externally managed.
2794                 */
2795                if (rdev->raid_disk == -1)
2796                        return -EEXIST;
2797                /* personality does all needed checks */
2798                if (rdev->mddev->pers->hot_remove_disk == NULL)
2799                        return -EINVAL;
2800                err = rdev->mddev->pers->
2801                        hot_remove_disk(rdev->mddev, rdev);
2802                if (err)
2803                        return err;
2804                sysfs_unlink_rdev(rdev->mddev, rdev);
2805                rdev->raid_disk = -1;
2806                set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2807                md_wakeup_thread(rdev->mddev->thread);
2808        } else if (rdev->mddev->pers) {
2809                /* Activating a spare .. or possibly reactivating
2810                 * if we ever get bitmaps working here.
2811                 */
2812
2813                if (rdev->raid_disk != -1)
2814                        return -EBUSY;
2815
2816                if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2817                        return -EBUSY;
2818
2819                if (rdev->mddev->pers->hot_add_disk == NULL)
2820                        return -EINVAL;
2821
2822                if (slot >= rdev->mddev->raid_disks &&
2823                    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2824                        return -ENOSPC;
2825
2826                rdev->raid_disk = slot;
2827                if (test_bit(In_sync, &rdev->flags))
2828                        rdev->saved_raid_disk = slot;
2829                else
2830                        rdev->saved_raid_disk = -1;
2831                clear_bit(In_sync, &rdev->flags);
2832                err = rdev->mddev->pers->
2833                        hot_add_disk(rdev->mddev, rdev);
2834                if (err) {
2835                        rdev->raid_disk = -1;
2836                        return err;
2837                } else
2838                        sysfs_notify_dirent_safe(rdev->sysfs_state);
2839                if (sysfs_link_rdev(rdev->mddev, rdev))
2840                        /* failure here is OK */;
2841                /* don't wakeup anyone, leave that to userspace. */
2842        } else {
2843                if (slot >= rdev->mddev->raid_disks &&
2844                    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2845                        return -ENOSPC;
2846                rdev->raid_disk = slot;
2847                /* assume it is working */
2848                clear_bit(Faulty, &rdev->flags);
2849                clear_bit(WriteMostly, &rdev->flags);
2850                set_bit(In_sync, &rdev->flags);
2851                sysfs_notify_dirent_safe(rdev->sysfs_state);
2852        }
2853        return len;
2854}
2855
2856
2857static struct rdev_sysfs_entry rdev_slot =
2858__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2859
2860static ssize_t
2861offset_show(struct md_rdev *rdev, char *page)
2862{
2863        return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2864}
2865
2866static ssize_t
2867offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2868{
2869        unsigned long long offset;
2870        if (strict_strtoull(buf, 10, &offset) < 0)
2871                return -EINVAL;
2872        if (rdev->mddev->pers && rdev->raid_disk >= 0)
2873                return -EBUSY;
2874        if (rdev->sectors && rdev->mddev->external)
2875                /* Must set offset before size, so overlap checks
2876                 * can be sane */
2877                return -EBUSY;
2878        rdev->data_offset = offset;
2879        rdev->new_data_offset = offset;
2880        return len;
2881}
2882
2883static struct rdev_sysfs_entry rdev_offset =
2884__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2885
2886static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2887{
2888        return sprintf(page, "%llu\n",
2889                       (unsigned long long)rdev->new_data_offset);
2890}
2891
2892static ssize_t new_offset_store(struct md_rdev *rdev,
2893                                const char *buf, size_t len)
2894{
2895        unsigned long long new_offset;
2896        struct mddev *mddev = rdev->mddev;
2897
2898        if (strict_strtoull(buf, 10, &new_offset) < 0)
2899                return -EINVAL;
2900
2901        if (mddev->sync_thread)
2902                return -EBUSY;
2903        if (new_offset == rdev->data_offset)
2904                /* reset is always permitted */
2905                ;
2906        else if (new_offset > rdev->data_offset) {
2907                /* must not push array size beyond rdev_sectors */
2908                if (new_offset - rdev->data_offset
2909                    + mddev->dev_sectors > rdev->sectors)
2910                                return -E2BIG;
2911        }
2912        /* Metadata worries about other space details. */
2913
2914        /* decreasing the offset is inconsistent with a backwards
2915         * reshape.
2916         */
2917        if (new_offset < rdev->data_offset &&
2918            mddev->reshape_backwards)
2919                return -EINVAL;
2920        /* Increasing offset is inconsistent with forwards
2921         * reshape.  reshape_direction should be set to
2922         * 'backwards' first.
2923         */
2924        if (new_offset > rdev->data_offset &&
2925            !mddev->reshape_backwards)
2926                return -EINVAL;
2927
2928        if (mddev->pers && mddev->persistent &&
2929            !super_types[mddev->major_version]
2930            .allow_new_offset(rdev, new_offset))
2931                return -E2BIG;
2932        rdev->new_data_offset = new_offset;
2933        if (new_offset > rdev->data_offset)
2934                mddev->reshape_backwards = 1;
2935        else if (new_offset < rdev->data_offset)
2936                mddev->reshape_backwards = 0;
2937
2938        return len;
2939}
2940static struct rdev_sysfs_entry rdev_new_offset =
2941__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2942
2943static ssize_t
2944rdev_size_show(struct md_rdev *rdev, char *page)
2945{
2946        return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2947}
2948
2949static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2950{
2951        /* check if two start/length pairs overlap */
2952        if (s1+l1 <= s2)
2953                return 0;
2954        if (s2+l2 <= s1)
2955                return 0;
2956        return 1;
2957}
2958
2959static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2960{
2961        unsigned long long blocks;
2962        sector_t new;
2963
2964        if (strict_strtoull(buf, 10, &blocks) < 0)
2965                return -EINVAL;
2966
2967        if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2968                return -EINVAL; /* sector conversion overflow */
2969
2970        new = blocks * 2;
2971        if (new != blocks * 2)
2972                return -EINVAL; /* unsigned long long to sector_t overflow */
2973
2974        *sectors = new;
2975        return 0;
2976}
2977
2978static ssize_t
2979rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2980{
2981        struct mddev *my_mddev = rdev->mddev;
2982        sector_t oldsectors = rdev->sectors;
2983        sector_t sectors;
2984
2985        if (strict_blocks_to_sectors(buf, &sectors) < 0)
2986                return -EINVAL;
2987        if (rdev->data_offset != rdev->new_data_offset)
2988                return -EINVAL; /* too confusing */
2989        if (my_mddev->pers && rdev->raid_disk >= 0) {
2990                if (my_mddev->persistent) {
2991                        sectors = super_types[my_mddev->major_version].
2992                                rdev_size_change(rdev, sectors);
2993                        if (!sectors)
2994                                return -EBUSY;
2995                } else if (!sectors)
2996                        sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2997                                rdev->data_offset;
2998        }
2999        if (sectors < my_mddev->dev_sectors)
3000                return -EINVAL; /* component must fit device */
3001
3002        rdev->sectors = sectors;
3003        if (sectors > oldsectors && my_mddev->external) {
3004                /* need to check that all other rdevs with the same ->bdev
3005                 * do not overlap.  We need to unlock the mddev to avoid
3006                 * a deadlock.  We have already changed rdev->sectors, and if
3007                 * we have to change it back, we will have the lock again.
3008                 */
3009                struct mddev *mddev;
3010                int overlap = 0;
3011                struct list_head *tmp;
3012
3013                mddev_unlock(my_mddev);
3014                for_each_mddev(mddev, tmp) {
3015                        struct md_rdev *rdev2;
3016
3017                        mddev_lock(mddev);
3018                        rdev_for_each(rdev2, mddev)
3019                                if (rdev->bdev == rdev2->bdev &&
3020                                    rdev != rdev2 &&
3021                                    overlaps(rdev->data_offset, rdev->sectors,
3022                                             rdev2->data_offset,
3023                                             rdev2->sectors)) {
3024                                        overlap = 1;
3025                                        break;
3026                                }
3027                        mddev_unlock(mddev);
3028                        if (overlap) {
3029                                mddev_put(mddev);
3030                                break;
3031                        }
3032                }
3033                mddev_lock(my_mddev);
3034                if (overlap) {
3035                        /* Someone else could have slipped in a size
3036                         * change here, but doing so is just silly.
3037                         * We put oldsectors back because we *know* it is
3038                         * safe, and trust userspace not to race with
3039                         * itself
3040                         */
3041                        rdev->sectors = oldsectors;
3042                        return -EBUSY;
3043                }
3044        }
3045        return len;
3046}
3047
3048static struct rdev_sysfs_entry rdev_size =
3049__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3050
3051
3052static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3053{
3054        unsigned long long recovery_start = rdev->recovery_offset;
3055
3056        if (test_bit(In_sync, &rdev->flags) ||
3057            recovery_start == MaxSector)
3058                return sprintf(page, "none\n");
3059
3060        return sprintf(page, "%llu\n", recovery_start);
3061}
3062
3063static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3064{
3065        unsigned long long recovery_start;
3066
3067        if (cmd_match(buf, "none"))
3068                recovery_start = MaxSector;
3069        else if (strict_strtoull(buf, 10, &recovery_start))
3070                return -EINVAL;
3071
3072        if (rdev->mddev->pers &&
3073            rdev->raid_disk >= 0)
3074                return -EBUSY;
3075
3076        rdev->recovery_offset = recovery_start;
3077        if (recovery_start == MaxSector)
3078                set_bit(In_sync, &rdev->flags);
3079        else
3080                clear_bit(In_sync, &rdev->flags);
3081        return len;
3082}
3083
3084static struct rdev_sysfs_entry rdev_recovery_start =
3085__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3086
3087
3088static ssize_t
3089badblocks_show(struct badblocks *bb, char *page, int unack);
3090static ssize_t
3091badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3092
3093static ssize_t bb_show(struct md_rdev *rdev, char *page)
3094{
3095        return badblocks_show(&rdev->badblocks, page, 0);
3096}
3097static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3098{
3099        int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3100        /* Maybe that ack was all we needed */
3101        if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3102                wake_up(&rdev->blocked_wait);
3103        return rv;
3104}
3105static struct rdev_sysfs_entry rdev_bad_blocks =
3106__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3107
3108
3109static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3110{
3111        return badblocks_show(&rdev->badblocks, page, 1);
3112}
3113static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3114{
3115        return badblocks_store(&rdev->badblocks, page, len, 1);
3116}
3117static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3118__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3119
3120static struct attribute *rdev_default_attrs[] = {
3121        &rdev_state.attr,
3122        &rdev_errors.attr,
3123        &rdev_slot.attr,
3124        &rdev_offset.attr,
3125        &rdev_new_offset.attr,
3126        &rdev_size.attr,
3127        &rdev_recovery_start.attr,
3128        &rdev_bad_blocks.attr,
3129        &rdev_unack_bad_blocks.attr,
3130        NULL,
3131};
3132static ssize_t
3133rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3134{
3135        struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3136        struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3137        struct mddev *mddev = rdev->mddev;
3138        ssize_t rv;
3139
3140        if (!entry->show)
3141                return -EIO;
3142
3143        rv = mddev ? mddev_lock(mddev) : -EBUSY;
3144        if (!rv) {
3145                if (rdev->mddev == NULL)
3146                        rv = -EBUSY;
3147                else
3148                        rv = entry->show(rdev, page);
3149                mddev_unlock(mddev);
3150        }
3151        return rv;
3152}
3153
3154static ssize_t
3155rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3156              const char *page, size_t length)
3157{
3158        struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3159        struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3160        ssize_t rv;
3161        struct mddev *mddev = rdev->mddev;
3162
3163        if (!entry->store)
3164                return -EIO;
3165        if (!capable(CAP_SYS_ADMIN))
3166                return -EACCES;
3167        rv = mddev ? mddev_lock(mddev): -EBUSY;
3168        if (!rv) {
3169                if (rdev->mddev == NULL)
3170                        rv = -EBUSY;
3171                else
3172                        rv = entry->store(rdev, page, length);
3173                mddev_unlock(mddev);
3174        }
3175        return rv;
3176}
3177
3178static void rdev_free(struct kobject *ko)
3179{
3180        struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3181        kfree(rdev);
3182}
3183static const struct sysfs_ops rdev_sysfs_ops = {
3184        .show           = rdev_attr_show,
3185        .store          = rdev_attr_store,
3186};
3187static struct kobj_type rdev_ktype = {
3188        .release        = rdev_free,
3189        .sysfs_ops      = &rdev_sysfs_ops,
3190        .default_attrs  = rdev_default_attrs,
3191};
3192
3193int md_rdev_init(struct md_rdev *rdev)
3194{
3195        rdev->desc_nr = -1;
3196        rdev->saved_raid_disk = -1;
3197        rdev->raid_disk = -1;
3198        rdev->flags = 0;
3199        rdev->data_offset = 0;
3200        rdev->new_data_offset = 0;
3201        rdev->sb_events = 0;
3202        rdev->last_read_error.tv_sec  = 0;
3203        rdev->last_read_error.tv_nsec = 0;
3204        rdev->sb_loaded = 0;
3205        rdev->bb_page = NULL;
3206        atomic_set(&rdev->nr_pending, 0);
3207        atomic_set(&rdev->read_errors, 0);
3208        atomic_set(&rdev->corrected_errors, 0);
3209
3210        INIT_LIST_HEAD(&rdev->same_set);
3211        init_waitqueue_head(&rdev->blocked_wait);
3212
3213        /* Add space to store bad block list.
3214         * This reserves the space even on arrays where it cannot
3215         * be used - I wonder if that matters
3216         */
3217        rdev->badblocks.count = 0;
3218        rdev->badblocks.shift = 0;
3219        rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3220        seqlock_init(&rdev->badblocks.lock);
3221        if (rdev->badblocks.page == NULL)
3222                return -ENOMEM;
3223
3224        return 0;
3225}
3226EXPORT_SYMBOL_GPL(md_rdev_init);
3227/*
3228 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3229 *
3230 * mark the device faulty if:
3231 *
3232 *   - the device is nonexistent (zero size)
3233 *   - the device has no valid superblock
3234 *
3235 * a faulty rdev _never_ has rdev->sb set.
3236 */
3237static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3238{
3239        char b[BDEVNAME_SIZE];
3240        int err;
3241        struct md_rdev *rdev;
3242        sector_t size;
3243
3244        rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3245        if (!rdev) {
3246                printk(KERN_ERR "md: could not alloc mem for new device!\n");
3247                return ERR_PTR(-ENOMEM);
3248        }
3249
3250        err = md_rdev_init(rdev);
3251        if (err)
3252                goto abort_free;
3253        err = alloc_disk_sb(rdev);
3254        if (err)
3255                goto abort_free;
3256
3257        err = lock_rdev(rdev, newdev, super_format == -2);
3258        if (err)
3259                goto abort_free;
3260
3261        kobject_init(&rdev->kobj, &rdev_ktype);
3262
3263        size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3264        if (!size) {
3265                printk(KERN_WARNING 
3266                        "md: %s has zero or unknown size, marking faulty!\n",
3267                        bdevname(rdev->bdev,b));
3268                err = -EINVAL;
3269                goto abort_free;
3270        }
3271
3272        if (super_format >= 0) {
3273                err = super_types[super_format].
3274                        load_super(rdev, NULL, super_minor);
3275                if (err == -EINVAL) {
3276                        printk(KERN_WARNING
3277                                "md: %s does not have a valid v%d.%d "
3278                               "superblock, not importing!\n",
3279                                bdevname(rdev->bdev,b),
3280                               super_format, super_minor);
3281                        goto abort_free;
3282                }
3283                if (err < 0) {
3284                        printk(KERN_WARNING 
3285                                "md: could not read %s's sb, not importing!\n",
3286                                bdevname(rdev->bdev,b));
3287                        goto abort_free;
3288                }
3289        }
3290        if (super_format == -1)
3291                /* hot-add for 0.90, or non-persistent: so no badblocks */
3292                rdev->badblocks.shift = -1;
3293
3294        return rdev;
3295
3296abort_free:
3297        if (rdev->bdev)
3298                unlock_rdev(rdev);
3299        md_rdev_clear(rdev);
3300        kfree(rdev);
3301        return ERR_PTR(err);
3302}
3303
3304/*
3305 * Check a full RAID array for plausibility
3306 */
3307
3308
3309static void analyze_sbs(struct mddev * mddev)
3310{
3311        int i;
3312        struct md_rdev *rdev, *freshest, *tmp;
3313        char b[BDEVNAME_SIZE];
3314
3315        freshest = NULL;
3316        rdev_for_each_safe(rdev, tmp, mddev)
3317                switch (super_types[mddev->major_version].
3318                        load_super(rdev, freshest, mddev->minor_version)) {
3319                case 1:
3320                        freshest = rdev;
3321                        break;
3322                case 0:
3323                        break;
3324                default:
3325                        printk( KERN_ERR \
3326                                "md: fatal superblock inconsistency in %s"
3327                                " -- removing from array\n", 
3328                                bdevname(rdev->bdev,b));
3329                        kick_rdev_from_array(rdev);
3330                }
3331
3332
3333        super_types[mddev->major_version].
3334                validate_super(mddev, freshest);
3335
3336        i = 0;
3337        rdev_for_each_safe(rdev, tmp, mddev) {
3338                if (mddev->max_disks &&
3339                    (rdev->desc_nr >= mddev->max_disks ||
3340                     i > mddev->max_disks)) {
3341                        printk(KERN_WARNING
3342                               "md: %s: %s: only %d devices permitted\n",
3343                               mdname(mddev), bdevname(rdev->bdev, b),
3344                               mddev->max_disks);
3345                        kick_rdev_from_array(rdev);
3346                        continue;
3347                }
3348                if (rdev != freshest)
3349                        if (super_types[mddev->major_version].
3350                            validate_super(mddev, rdev)) {
3351                                printk(KERN_WARNING "md: kicking non-fresh %s"
3352                                        " from array!\n",
3353                                        bdevname(rdev->bdev,b));
3354                                kick_rdev_from_array(rdev);
3355                                continue;
3356                        }
3357                if (mddev->level == LEVEL_MULTIPATH) {
3358                        rdev->desc_nr = i++;
3359                        rdev->raid_disk = rdev->desc_nr;
3360                        set_bit(In_sync, &rdev->flags);
3361                } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3362                        rdev->raid_disk = -1;
3363                        clear_bit(In_sync, &rdev->flags);
3364                }
3365        }
3366}
3367
3368/* Read a fixed-point number.
3369 * Numbers in sysfs attributes should be in "standard" units where
3370 * possible, so time should be in seconds.
3371 * However we internally use a a much smaller unit such as 
3372 * milliseconds or jiffies.
3373 * This function takes a decimal number with a possible fractional
3374 * component, and produces an integer which is the result of
3375 * multiplying that number by 10^'scale'.
3376 * all without any floating-point arithmetic.
3377 */
3378int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3379{
3380        unsigned long result = 0;
3381        long decimals = -1;
3382        while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3383                if (*cp == '.')
3384                        decimals = 0;
3385                else if (decimals < scale) {
3386                        unsigned int value;
3387                        value = *cp - '0';
3388                        result = result * 10 + value;
3389                        if (decimals >= 0)
3390                                decimals++;
3391                }
3392                cp++;
3393        }
3394        if (*cp == '\n')
3395                cp++;
3396        if (*cp)
3397                return -EINVAL;
3398        if (decimals < 0)
3399                decimals = 0;
3400        while (decimals < scale) {
3401                result *= 10;
3402                decimals ++;
3403        }
3404        *res = result;
3405        return 0;
3406}
3407
3408
3409static void md_safemode_timeout(unsigned long data);
3410
3411static ssize_t
3412safe_delay_show(struct mddev *mddev, char *page)
3413{
3414        int msec = (mddev->safemode_delay*1000)/HZ;
3415        return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3416}
3417static ssize_t
3418safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3419{
3420        unsigned long msec;
3421
3422        if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3423                return -EINVAL;
3424        if (msec == 0)
3425                mddev->safemode_delay = 0;
3426        else {
3427                unsigned long old_delay = mddev->safemode_delay;
3428                mddev->safemode_delay = (msec*HZ)/1000;
3429                if (mddev->safemode_delay == 0)
3430                        mddev->safemode_delay = 1;
3431                if (mddev->safemode_delay < old_delay)
3432                        md_safemode_timeout((unsigned long)mddev);
3433        }
3434        return len;
3435}
3436static struct md_sysfs_entry md_safe_delay =
3437__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3438
3439static ssize_t
3440level_show(struct mddev *mddev, char *page)
3441{
3442        struct md_personality *p = mddev->pers;
3443        if (p)
3444                return sprintf(page, "%s\n", p->name);
3445        else if (mddev->clevel[0])
3446                return sprintf(page, "%s\n", mddev->clevel);
3447        else if (mddev->level != LEVEL_NONE)
3448                return sprintf(page, "%d\n", mddev->level);
3449        else
3450                return 0;
3451}
3452
3453static ssize_t
3454level_store(struct mddev *mddev, const char *buf, size_t len)
3455{
3456        char clevel[16];
3457        ssize_t rv = len;
3458        struct md_personality *pers;
3459        long level;
3460        void *priv;
3461        struct md_rdev *rdev;
3462
3463        if (mddev->pers == NULL) {
3464                if (len == 0)
3465                        return 0;
3466                if (len >= sizeof(mddev->clevel))
3467                        return -ENOSPC;
3468                strncpy(mddev->clevel, buf, len);
3469                if (mddev->clevel[len-1] == '\n')
3470                        len--;
3471                mddev->clevel[len] = 0;
3472                mddev->level = LEVEL_NONE;
3473                return rv;
3474        }
3475
3476        /* request to change the personality.  Need to ensure:
3477         *  - array is not engaged in resync/recovery/reshape
3478         *  - old personality can be suspended
3479         *  - new personality will access other array.
3480         */
3481
3482        if (mddev->sync_thread ||
3483            mddev->reshape_position != MaxSector ||
3484            mddev->sysfs_active)
3485                return -EBUSY;
3486
3487        if (!mddev->pers->quiesce) {
3488                printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3489                       mdname(mddev), mddev->pers->name);
3490                return -EINVAL;
3491        }
3492
3493        /* Now find the new personality */
3494        if (len == 0 || len >= sizeof(clevel))
3495                return -EINVAL;
3496        strncpy(clevel, buf, len);
3497        if (clevel[len-1] == '\n')
3498                len--;
3499        clevel[len] = 0;
3500        if (strict_strtol(clevel, 10, &level))
3501                level = LEVEL_NONE;
3502
3503        if (request_module("md-%s", clevel) != 0)
3504                request_module("md-level-%s", clevel);
3505        spin_lock(&pers_lock);
3506        pers = find_pers(level, clevel);
3507        if (!pers || !try_module_get(pers->owner)) {
3508                spin_unlock(&pers_lock);
3509                printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3510                return -EINVAL;
3511        }
3512        spin_unlock(&pers_lock);
3513
3514        if (pers == mddev->pers) {
3515                /* Nothing to do! */
3516                module_put(pers->owner);
3517                return rv;
3518        }
3519        if (!pers->takeover) {
3520                module_put(pers->owner);
3521                printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3522                       mdname(mddev), clevel);
3523                return -EINVAL;
3524        }
3525
3526        rdev_for_each(rdev, mddev)
3527                rdev->new_raid_disk = rdev->raid_disk;
3528
3529        /* ->takeover must set new_* and/or delta_disks
3530         * if it succeeds, and may set them when it fails.
3531         */
3532        priv = pers->takeover(mddev);
3533        if (IS_ERR(priv)) {
3534                mddev->new_level = mddev->level;
3535                mddev->new_layout = mddev->layout;
3536                mddev->new_chunk_sectors = mddev->chunk_sectors;
3537                mddev->raid_disks -= mddev->delta_disks;
3538                mddev->delta_disks = 0;
3539                mddev->reshape_backwards = 0;
3540                module_put(pers->owner);
3541                printk(KERN_WARNING "md: %s: %s would not accept array\n",
3542                       mdname(mddev), clevel);
3543                return PTR_ERR(priv);
3544        }
3545
3546        /* Looks like we have a winner */
3547        mddev_suspend(mddev);
3548        mddev->pers->stop(mddev);
3549        
3550        if (mddev->pers->sync_request == NULL &&
3551            pers->sync_request != NULL) {
3552                /* need to add the md_redundancy_group */
3553                if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3554                        printk(KERN_WARNING
3555                               "md: cannot register extra attributes for %s\n",
3556                               mdname(mddev));
3557                mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3558        }               
3559        if (mddev->pers->sync_request != NULL &&
3560            pers->sync_request == NULL) {
3561                /* need to remove the md_redundancy_group */
3562                if (mddev->to_remove == NULL)
3563                        mddev->to_remove = &md_redundancy_group;
3564        }
3565
3566        if (mddev->pers->sync_request == NULL &&
3567            mddev->external) {
3568                /* We are converting from a no-redundancy array
3569                 * to a redundancy array and metadata is managed
3570                 * externally so we need to be sure that writes
3571                 * won't block due to a need to transition
3572                 *      clean->dirty
3573                 * until external management is started.
3574                 */
3575                mddev->in_sync = 0;
3576                mddev->safemode_delay = 0;
3577                mddev->safemode = 0;
3578        }
3579
3580        rdev_for_each(rdev, mddev) {
3581                if (rdev->raid_disk < 0)
3582                        continue;
3583                if (rdev->new_raid_disk >= mddev->raid_disks)
3584                        rdev->new_raid_disk = -1;
3585                if (rdev->new_raid_disk == rdev->raid_disk)
3586                        continue;
3587                sysfs_unlink_rdev(mddev, rdev);
3588        }
3589        rdev_for_each(rdev, mddev) {
3590                if (rdev->raid_disk < 0)
3591                        continue;
3592                if (rdev->new_raid_disk == rdev->raid_disk)
3593                        continue;
3594                rdev->raid_disk = rdev->new_raid_disk;
3595                if (rdev->raid_disk < 0)
3596                        clear_bit(In_sync, &rdev->flags);
3597                else {
3598                        if (sysfs_link_rdev(mddev, rdev))
3599                                printk(KERN_WARNING "md: cannot register rd%d"
3600                                       " for %s after level change\n",
3601                                       rdev->raid_disk, mdname(mddev));
3602                }
3603        }
3604
3605        module_put(mddev->pers->owner);
3606        mddev->pers = pers;
3607        mddev->private = priv;
3608        strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3609        mddev->level = mddev->new_level;
3610        mddev->layout = mddev->new_layout;
3611        mddev->chunk_sectors = mddev->new_chunk_sectors;
3612        mddev->delta_disks = 0;
3613        mddev->reshape_backwards = 0;
3614        mddev->degraded = 0;
3615        if (mddev->pers->sync_request == NULL) {
3616                /* this is now an array without redundancy, so
3617                 * it must always be in_sync
3618                 */
3619                mddev->in_sync = 1;
3620                del_timer_sync(&mddev->safemode_timer);
3621        }
3622        pers->run(mddev);
3623        set_bit(MD_CHANGE_DEVS, &mddev->flags);
3624        mddev_resume(mddev);
3625        sysfs_notify(&mddev->kobj, NULL, "level");
3626        md_new_event(mddev);
3627        return rv;
3628}
3629
3630static struct md_sysfs_entry md_level =
3631__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3632
3633
3634static ssize_t
3635layout_show(struct mddev *mddev, char *page)
3636{
3637        /* just a number, not meaningful for all levels */
3638        if (mddev->reshape_position != MaxSector &&
3639            mddev->layout != mddev->new_layout)
3640                return sprintf(page, "%d (%d)\n",
3641                               mddev->new_layout, mddev->layout);
3642        return sprintf(page, "%d\n", mddev->layout);
3643}
3644
3645static ssize_t
3646layout_store(struct mddev *mddev, const char *buf, size_t len)
3647{
3648        char *e;
3649        unsigned long n = simple_strtoul(buf, &e, 10);
3650
3651        if (!*buf || (*e && *e != '\n'))
3652                return -EINVAL;
3653
3654        if (mddev->pers) {
3655                int err;
3656                if (mddev->pers->check_reshape == NULL)
3657                        return -EBUSY;
3658                mddev->new_layout = n;
3659                err = mddev->pers->check_reshape(mddev);
3660                if (err) {
3661                        mddev->new_layout = mddev->layout;
3662                        return err;
3663                }
3664        } else {
3665                mddev->new_layout = n;
3666                if (mddev->reshape_position == MaxSector)
3667                        mddev->layout = n;
3668        }
3669        return len;
3670}
3671static struct md_sysfs_entry md_layout =
3672__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3673
3674
3675static ssize_t
3676raid_disks_show(struct mddev *mddev, char *page)
3677{
3678        if (mddev->raid_disks == 0)
3679                return 0;
3680        if (mddev->reshape_position != MaxSector &&
3681            mddev->delta_disks != 0)
3682                return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3683                               mddev->raid_disks - mddev->delta_disks);
3684        return sprintf(page, "%d\n", mddev->raid_disks);
3685}
3686
3687static int update_raid_disks(struct mddev *mddev, int raid_disks);
3688
3689static ssize_t
3690raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3691{
3692        char *e;
3693        int rv = 0;
3694        unsigned long n = simple_strtoul(buf, &e, 10);
3695
3696        if (!*buf || (*e && *e != '\n'))
3697                return -EINVAL;
3698
3699        if (mddev->pers)
3700                rv = update_raid_disks(mddev, n);
3701        else if (mddev->reshape_position != MaxSector) {
3702                struct md_rdev *rdev;
3703                int olddisks = mddev->raid_disks - mddev->delta_disks;
3704
3705                rdev_for_each(rdev, mddev) {
3706                        if (olddisks < n &&
3707                            rdev->data_offset < rdev->new_data_offset)
3708                                return -EINVAL;
3709                        if (olddisks > n &&
3710                            rdev->data_offset > rdev->new_data_offset)
3711                                return -EINVAL;
3712                }
3713                mddev->delta_disks = n - olddisks;
3714                mddev->raid_disks = n;
3715                mddev->reshape_backwards = (mddev->delta_disks < 0);
3716        } else
3717                mddev->raid_disks = n;
3718        return rv ? rv : len;
3719}
3720static struct md_sysfs_entry md_raid_disks =
3721__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3722
3723static ssize_t
3724chunk_size_show(struct mddev *mddev, char *page)
3725{
3726        if (mddev->reshape_position != MaxSector &&
3727            mddev->chunk_sectors != mddev->new_chunk_sectors)
3728                return sprintf(page, "%d (%d)\n",
3729                               mddev->new_chunk_sectors << 9,
3730                               mddev->chunk_sectors << 9);
3731        return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3732}
3733
3734static ssize_t
3735chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3736{
3737        char *e;
3738        unsigned long n = simple_strtoul(buf, &e, 10);
3739
3740        if (!*buf || (*e && *e != '\n'))
3741                return -EINVAL;
3742
3743        if (mddev->pers) {
3744                int err;
3745                if (mddev->pers->check_reshape == NULL)
3746                        return -EBUSY;
3747                mddev->new_chunk_sectors = n >> 9;
3748                err = mddev->pers->check_reshape(mddev);
3749                if (err) {
3750                        mddev->new_chunk_sectors = mddev->chunk_sectors;
3751                        return err;
3752                }
3753        } else {
3754                mddev->new_chunk_sectors = n >> 9;
3755                if (mddev->reshape_position == MaxSector)
3756                        mddev->chunk_sectors = n >> 9;
3757        }
3758        return len;
3759}
3760static struct md_sysfs_entry md_chunk_size =
3761__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3762
3763static ssize_t
3764resync_start_show(struct mddev *mddev, char *page)
3765{
3766        if (mddev->recovery_cp == MaxSector)
3767                return sprintf(page, "none\n");
3768        return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3769}
3770
3771static ssize_t
3772resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3773{
3774        char *e;
3775        unsigned long long n = simple_strtoull(buf, &e, 10);
3776
3777        if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3778                return -EBUSY;
3779        if (cmd_match(buf, "none"))
3780                n = MaxSector;
3781        else if (!*buf || (*e && *e != '\n'))
3782                return -EINVAL;
3783
3784        mddev->recovery_cp = n;
3785        if (mddev->pers)
3786                set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3787        return len;
3788}
3789static struct md_sysfs_entry md_resync_start =
3790__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3791
3792/*
3793 * The array state can be:
3794 *
3795 * clear
3796 *     No devices, no size, no level
3797 *     Equivalent to STOP_ARRAY ioctl
3798 * inactive
3799 *     May have some settings, but array is not active
3800 *        all IO results in error
3801 *     When written, doesn't tear down array, but just stops it
3802 * suspended (not supported yet)
3803 *     All IO requests will block. The array can be reconfigured.
3804 *     Writing this, if accepted, will block until array is quiescent
3805 * readonly
3806 *     no resync can happen.  no superblocks get written.
3807 *     write requests fail
3808 * read-auto
3809 *     like readonly, but behaves like 'clean' on a write request.
3810 *
3811 * clean - no pending writes, but otherwise active.
3812 *     When written to inactive array, starts without resync
3813 *     If a write request arrives then
3814 *       if metadata is known, mark 'dirty' and switch to 'active'.
3815 *       if not known, block and switch to write-pending
3816 *     If written to an active array that has pending writes, then fails.
3817 * active
3818 *     fully active: IO and resync can be happening.
3819 *     When written to inactive array, starts with resync
3820 *
3821 * write-pending
3822 *     clean, but writes are blocked waiting for 'active' to be written.
3823 *
3824 * active-idle
3825 *     like active, but no writes have been seen for a while (100msec).
3826 *
3827 */
3828enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3829                   write_pending, active_idle, bad_word};
3830static char *array_states[] = {
3831        "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3832        "write-pending", "active-idle", NULL };
3833
3834static int match_word(const char *word, char **list)
3835{
3836        int n;
3837        for (n=0; list[n]; n++)
3838                if (cmd_match(word, list[n]))
3839                        break;
3840        return n;
3841}
3842
3843static ssize_t
3844array_state_show(struct mddev *mddev, char *page)
3845{
3846        enum array_state st = inactive;
3847
3848        if (mddev->pers)
3849                switch(mddev->ro) {
3850                case 1:
3851                        st = readonly;
3852                        break;
3853                case 2:
3854                        st = read_auto;
3855                        break;
3856                case 0:
3857                        if (mddev->in_sync)
3858                                st = clean;
3859                        else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3860                                st = write_pending;
3861                        else if (mddev->safemode)
3862                                st = active_idle;
3863                        else
3864                                st = active;
3865                }
3866        else {
3867                if (list_empty(&mddev->disks) &&
3868                    mddev->raid_disks == 0 &&
3869                    mddev->dev_sectors == 0)
3870                        st = clear;
3871                else
3872                        st = inactive;
3873        }
3874        return sprintf(page, "%s\n", array_states[st]);
3875}
3876
3877static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3878static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3879static int do_md_run(struct mddev * mddev);
3880static int restart_array(struct mddev *mddev);
3881
3882static ssize_t
3883array_state_store(struct mddev *mddev, const char *buf, size_t len)
3884{
3885        int err = -EINVAL;
3886        enum array_state st = match_word(buf, array_states);
3887        switch(st) {
3888        case bad_word:
3889                break;
3890        case clear:
3891                /* stopping an active array */
3892                err = do_md_stop(mddev, 0, NULL);
3893                break;
3894        case inactive:
3895                /* stopping an active array */
3896                if (mddev->pers)
3897                        err = do_md_stop(mddev, 2, NULL);
3898                else
3899                        err = 0; /* already inactive */
3900                break;
3901        case suspended:
3902                break; /* not supported yet */
3903        case readonly:
3904                if (mddev->pers)
3905                        err = md_set_readonly(mddev, NULL);
3906                else {
3907                        mddev->ro = 1;
3908                        set_disk_ro(mddev->gendisk, 1);
3909                        err = do_md_run(mddev);
3910                }
3911                break;
3912        case read_auto:
3913                if (mddev->pers) {
3914                        if (mddev->ro == 0)
3915                                err = md_set_readonly(mddev, NULL);
3916                        else if (mddev->ro == 1)
3917                                err = restart_array(mddev);
3918                        if (err == 0) {
3919                                mddev->ro = 2;
3920                                set_disk_ro(mddev->gendisk, 0);
3921                        }
3922                } else {
3923                        mddev->ro = 2;
3924                        err = do_md_run(mddev);
3925                }
3926                break;
3927        case clean:
3928                if (mddev->pers) {
3929                        restart_array(mddev);
3930                        spin_lock_irq(&mddev->write_lock);
3931                        if (atomic_read(&mddev->writes_pending) == 0) {
3932                                if (mddev->in_sync == 0) {
3933                                        mddev->in_sync = 1;
3934                                        if (mddev->safemode == 1)
3935                                                mddev->safemode = 0;
3936                                        set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3937                                }
3938                                err = 0;
3939                        } else
3940                                err = -EBUSY;
3941                        spin_unlock_irq(&mddev->write_lock);
3942                } else
3943                        err = -EINVAL;
3944                break;
3945        case active:
3946                if (mddev->pers) {
3947                        restart_array(mddev);
3948                        clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3949                        wake_up(&mddev->sb_wait);
3950                        err = 0;
3951                } else {
3952                        mddev->ro = 0;
3953                        set_disk_ro(mddev->gendisk, 0);
3954                        err = do_md_run(mddev);
3955                }
3956                break;
3957        case write_pending:
3958        case active_idle:
3959                /* these cannot be set */
3960                break;
3961        }
3962        if (err)
3963                return err;
3964        else {
3965                if (mddev->hold_active == UNTIL_IOCTL)
3966                        mddev->hold_active = 0;
3967                sysfs_notify_dirent_safe(mddev->sysfs_state);
3968                return len;
3969        }
3970}
3971static struct md_sysfs_entry md_array_state =
3972__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3973
3974static ssize_t
3975max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3976        return sprintf(page, "%d\n",
3977                       atomic_read(&mddev->max_corr_read_errors));
3978}
3979
3980static ssize_t
3981max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3982{
3983        char *e;
3984        unsigned long n = simple_strtoul(buf, &e, 10);
3985
3986        if (*buf && (*e == 0 || *e == '\n')) {
3987                atomic_set(&mddev->max_corr_read_errors, n);
3988                return len;
3989        }
3990        return -EINVAL;
3991}
3992
3993static struct md_sysfs_entry max_corr_read_errors =
3994__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3995        max_corrected_read_errors_store);
3996
3997static ssize_t
3998null_show(struct mddev *mddev, char *page)
3999{
4000        return -EINVAL;
4001}
4002
4003static ssize_t
4004new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4005{
4006        /* buf must be %d:%d\n? giving major and minor numbers */
4007        /* The new device is added to the array.
4008         * If the array has a persistent superblock, we read the
4009         * superblock to initialise info and check validity.
4010         * Otherwise, only checking done is that in bind_rdev_to_array,
4011         * which mainly checks size.
4012         */
4013        char *e;
4014        int major = simple_strtoul(buf, &e, 10);
4015        int minor;
4016        dev_t dev;
4017        struct md_rdev *rdev;
4018        int err;
4019
4020        if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4021                return -EINVAL;
4022        minor = simple_strtoul(e+1, &e, 10);
4023        if (*e && *e != '\n')
4024                return -EINVAL;
4025        dev = MKDEV(major, minor);
4026        if (major != MAJOR(dev) ||
4027            minor != MINOR(dev))
4028                return -EOVERFLOW;
4029
4030
4031        if (mddev->persistent) {
4032                rdev = md_import_device(dev, mddev->major_version,
4033                                        mddev->minor_version);
4034                if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4035                        struct md_rdev *rdev0
4036                                = list_entry(mddev->disks.next,
4037                                             struct md_rdev, same_set);
4038                        err = super_types[mddev->major_version]
4039                                .load_super(rdev, rdev0, mddev->minor_version);
4040                        if (err < 0)
4041                                goto out;
4042                }
4043        } else if (mddev->external)
4044                rdev = md_import_device(dev, -2, -1);
4045        else
4046                rdev = md_import_device(dev, -1, -1);
4047
4048        if (IS_ERR(rdev))
4049                return PTR_ERR(rdev);
4050        err = bind_rdev_to_array(rdev, mddev);
4051 out:
4052        if (err)
4053                export_rdev(rdev);
4054        return err ? err : len;
4055}
4056
4057static struct md_sysfs_entry md_new_device =
4058__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4059
4060static ssize_t
4061bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4062{
4063        char *end;
4064        unsigned long chunk, end_chunk;
4065
4066        if (!mddev->bitmap)
4067                goto out;
4068        /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4069        while (*buf) {
4070                chunk = end_chunk = simple_strtoul(buf, &end, 0);
4071                if (buf == end) break;
4072                if (*end == '-') { /* range */
4073                        buf = end + 1;
4074                        end_chunk = simple_strtoul(buf, &end, 0);
4075                        if (buf == end) break;
4076                }
4077                if (*end && !isspace(*end)) break;
4078                bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4079                buf = skip_spaces(end);
4080        }
4081        bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4082out:
4083        return len;
4084}
4085
4086static struct md_sysfs_entry md_bitmap =
4087__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4088
4089static ssize_t
4090size_show(struct mddev *mddev, char *page)
4091{
4092        return sprintf(page, "%llu\n",
4093                (unsigned long long)mddev->dev_sectors / 2);
4094}
4095
4096static int update_size(struct mddev *mddev, sector_t num_sectors);
4097
4098static ssize_t
4099size_store(struct mddev *mddev, const char *buf, size_t len)
4100{
4101        /* If array is inactive, we can reduce the component size, but
4102         * not increase it (except from 0).
4103         * If array is active, we can try an on-line resize
4104         */
4105        sector_t sectors;
4106        int err = strict_blocks_to_sectors(buf, &sectors);
4107
4108        if (err < 0)
4109                return err;
4110        if (mddev->pers) {
4111                err = update_size(mddev, sectors);
4112                md_update_sb(mddev, 1);
4113        } else {
4114                if (mddev->dev_sectors == 0 ||
4115                    mddev->dev_sectors > sectors)
4116                        mddev->dev_sectors = sectors;
4117                else
4118                        err = -ENOSPC;
4119        }
4120        return err ? err : len;
4121}
4122
4123static struct md_sysfs_entry md_size =
4124__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4125
4126
4127/* Metdata version.
4128 * This is one of
4129 *   'none' for arrays with no metadata (good luck...)
4130 *   'external' for arrays with externally managed metadata,
4131 * or N.M for internally known formats
4132 */
4133static ssize_t
4134metadata_show(struct mddev *mddev, char *page)
4135{
4136        if (mddev->persistent)
4137                return sprintf(page, "%d.%d\n",
4138                               mddev->major_version, mddev->minor_version);
4139        else if (mddev->external)
4140                return sprintf(page, "external:%s\n", mddev->metadata_type);
4141        else
4142                return sprintf(page, "none\n");
4143}
4144
4145static ssize_t
4146metadata_store(struct mddev *mddev, const char *buf, size_t len)
4147{
4148        int major, minor;
4149        char *e;
4150        /* Changing the details of 'external' metadata is
4151         * always permitted.  Otherwise there must be
4152         * no devices attached to the array.
4153         */
4154        if (mddev->external && strncmp(buf, "external:", 9) == 0)
4155                ;
4156        else if (!list_empty(&mddev->disks))
4157                return -EBUSY;
4158
4159        if (cmd_match(buf, "none")) {
4160                mddev->persistent = 0;
4161                mddev->external = 0;
4162                mddev->major_version = 0;
4163                mddev->minor_version = 90;
4164                return len;
4165        }
4166        if (strncmp(buf, "external:", 9) == 0) {
4167                size_t namelen = len-9;
4168                if (namelen >= sizeof(mddev->metadata_type))
4169                        namelen = sizeof(mddev->metadata_type)-1;
4170                strncpy(mddev->metadata_type, buf+9, namelen);
4171                mddev->metadata_type[namelen] = 0;
4172                if (namelen && mddev->metadata_type[namelen-1] == '\n')
4173                        mddev->metadata_type[--namelen] = 0;
4174                mddev->persistent = 0;
4175                mddev->external = 1;
4176                mddev->major_version = 0;
4177                mddev->minor_version = 90;
4178                return len;
4179        }
4180        major = simple_strtoul(buf, &e, 10);
4181        if (e==buf || *e != '.')
4182                return -EINVAL;
4183        buf = e+1;
4184        minor = simple_strtoul(buf, &e, 10);
4185        if (e==buf || (*e && *e != '\n') )
4186                return -EINVAL;
4187        if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4188                return -ENOENT;
4189        mddev->major_version = major;
4190        mddev->minor_version = minor;
4191        mddev->persistent = 1;
4192        mddev->external = 0;
4193        return len;
4194}
4195
4196static struct md_sysfs_entry md_metadata =
4197__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4198
4199static ssize_t
4200action_show(struct mddev *mddev, char *page)
4201{
4202        char *type = "idle";
4203        if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4204                type = "frozen";
4205        else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4206            (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4207                if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4208                        type = "reshape";
4209                else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4210                        if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4211                                type = "resync";
4212                        else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4213                                type = "check";
4214                        else
4215                                type = "repair";
4216                } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4217                        type = "recover";
4218        }
4219        return sprintf(page, "%s\n", type);
4220}
4221
4222static void reap_sync_thread(struct mddev *mddev);
4223
4224static ssize_t
4225action_store(struct mddev *mddev, const char *page, size_t len)
4226{
4227        if (!mddev->pers || !mddev->pers->sync_request)
4228                return -EINVAL;
4229
4230        if (cmd_match(page, "frozen"))
4231                set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4232        else
4233                clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4234
4235        if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4236                if (mddev->sync_thread) {
4237                        set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4238                        reap_sync_thread(mddev);
4239                }
4240        } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4241                   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4242                return -EBUSY;
4243        else if (cmd_match(page, "resync"))
4244                set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4245        else if (cmd_match(page, "recover")) {
4246                set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4247                set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4248        } else if (cmd_match(page, "reshape")) {
4249                int err;
4250                if (mddev->pers->start_reshape == NULL)
4251                        return -EINVAL;
4252                err = mddev->pers->start_reshape(mddev);
4253                if (err)
4254                        return err;
4255                sysfs_notify(&mddev->kobj, NULL, "degraded");
4256        } else {
4257                if (cmd_match(page, "check"))
4258                        set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4259                else if (!cmd_match(page, "repair"))
4260                        return -EINVAL;
4261                set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4262                set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4263        }
4264        if (mddev->ro == 2) {
4265                /* A write to sync_action is enough to justify
4266                 * canceling read-auto mode
4267                 */
4268                mddev->ro = 0;
4269                md_wakeup_thread(mddev->sync_thread);
4270        }
4271        set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4272        md_wakeup_thread(mddev->thread);
4273        sysfs_notify_dirent_safe(mddev->sysfs_action);
4274        return len;
4275}
4276
4277static ssize_t
4278mismatch_cnt_show(struct mddev *mddev, char *page)
4279{
4280        return sprintf(page, "%llu\n",
4281                       (unsigned long long)
4282                       atomic64_read(&mddev->resync_mismatches));
4283}
4284
4285static struct md_sysfs_entry md_scan_mode =
4286__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4287
4288
4289static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4290
4291static ssize_t
4292sync_min_show(struct mddev *mddev, char *page)
4293{
4294        return sprintf(page, "%d (%s)\n", speed_min(mddev),
4295                       mddev->sync_speed_min ? "local": "system");
4296}
4297
4298static ssize_t
4299sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4300{
4301        int min;
4302        char *e;
4303        if (strncmp(buf, "system", 6)==0) {
4304                mddev->sync_speed_min = 0;
4305                return len;
4306        }
4307        min = simple_strtoul(buf, &e, 10);
4308        if (buf == e || (*e && *e != '\n') || min <= 0)
4309                return -EINVAL;
4310        mddev->sync_speed_min = min;
4311        return len;
4312}
4313
4314static struct md_sysfs_entry md_sync_min =
4315__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4316
4317static ssize_t
4318sync_max_show(struct mddev *mddev, char *page)
4319{
4320        return sprintf(page, "%d (%s)\n", speed_max(mddev),
4321                       mddev->sync_speed_max ? "local": "system");
4322}
4323
4324static ssize_t
4325sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4326{
4327        int max;
4328        char *e;
4329        if (strncmp(buf, "system", 6)==0) {
4330                mddev->sync_speed_max = 0;
4331                return len;
4332        }
4333        max = simple_strtoul(buf, &e, 10);
4334        if (buf == e || (*e && *e != '\n') || max <= 0)
4335                return -EINVAL;
4336        mddev->sync_speed_max = max;
4337        return len;
4338}
4339
4340static struct md_sysfs_entry md_sync_max =
4341__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4342
4343static ssize_t
4344degraded_show(struct mddev *mddev, char *page)
4345{
4346        return sprintf(page, "%d\n", mddev->degraded);
4347}
4348static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4349
4350static ssize_t
4351sync_force_parallel_show(struct mddev *mddev, char *page)
4352{
4353        return sprintf(page, "%d\n", mddev->parallel_resync);
4354}
4355
4356static ssize_t
4357sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4358{
4359        long n;
4360
4361        if (strict_strtol(buf, 10, &n))
4362                return -EINVAL;
4363
4364        if (n != 0 && n != 1)
4365                return -EINVAL;
4366
4367        mddev->parallel_resync = n;
4368
4369        if (mddev->sync_thread)
4370                wake_up(&resync_wait);
4371
4372        return len;
4373}
4374
4375/* force parallel resync, even with shared block devices */
4376static struct md_sysfs_entry md_sync_force_parallel =
4377__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4378       sync_force_parallel_show, sync_force_parallel_store);
4379
4380static ssize_t
4381sync_speed_show(struct mddev *mddev, char *page)
4382{
4383        unsigned long resync, dt, db;
4384        if (mddev->curr_resync == 0)
4385                return sprintf(page, "none\n");
4386        resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4387        dt = (jiffies - mddev->resync_mark) / HZ;
4388        if (!dt) dt++;
4389        db = resync - mddev->resync_mark_cnt;
4390        return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4391}
4392
4393static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4394
4395static ssize_t
4396sync_completed_show(struct mddev *mddev, char *page)
4397{
4398        unsigned long long max_sectors, resync;
4399
4400        if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4401                return sprintf(page, "none\n");
4402
4403        if (mddev->curr_resync == 1 ||
4404            mddev->curr_resync == 2)
4405                return sprintf(page, "delayed\n");
4406
4407        if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4408            test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4409                max_sectors = mddev->resync_max_sectors;
4410        else
4411                max_sectors = mddev->dev_sectors;
4412
4413        resync = mddev->curr_resync_completed;
4414        return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4415}
4416
4417static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4418
4419static ssize_t
4420min_sync_show(struct mddev *mddev, char *page)
4421{
4422        return sprintf(page, "%llu\n",
4423                       (unsigned long long)mddev->resync_min);
4424}
4425static ssize_t
4426min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4427{
4428        unsigned long long min;
4429        if (strict_strtoull(buf, 10, &min))
4430                return -EINVAL;
4431        if (min > mddev->resync_max)
4432                return -EINVAL;
4433        if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4434                return -EBUSY;
4435
4436        /* Must be a multiple of chunk_size */
4437        if (mddev->chunk_sectors) {
4438                sector_t temp = min;
4439                if (sector_div(temp, mddev->chunk_sectors))
4440                        return -EINVAL;
4441        }
4442        mddev->resync_min = min;
4443
4444        return len;
4445}
4446
4447static struct md_sysfs_entry md_min_sync =
4448__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4449
4450static ssize_t
4451max_sync_show(struct mddev *mddev, char *page)
4452{
4453        if (mddev->resync_max == MaxSector)
4454                return sprintf(page, "max\n");
4455        else
4456                return sprintf(page, "%llu\n",
4457                               (unsigned long long)mddev->resync_max);
4458}
4459static ssize_t
4460max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4461{
4462        if (strncmp(buf, "max", 3) == 0)
4463                mddev->resync_max = MaxSector;
4464        else {
4465                unsigned long long max;
4466                if (strict_strtoull(buf, 10, &max))
4467                        return -EINVAL;
4468                if (max < mddev->resync_min)
4469                        return -EINVAL;
4470                if (max < mddev->resync_max &&
4471                    mddev->ro == 0 &&
4472                    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4473                        return -EBUSY;
4474
4475                /* Must be a multiple of chunk_size */
4476                if (mddev->chunk_sectors) {
4477                        sector_t temp = max;
4478                        if (sector_div(temp, mddev->chunk_sectors))
4479                                return -EINVAL;
4480                }
4481                mddev->resync_max = max;
4482        }
4483        wake_up(&mddev->recovery_wait);
4484        return len;
4485}
4486
4487static struct md_sysfs_entry md_max_sync =
4488__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4489
4490static ssize_t
4491suspend_lo_show(struct mddev *mddev, char *page)
4492{
4493        return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4494}
4495
4496static ssize_t
4497suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4498{
4499        char *e;
4500        unsigned long long new = simple_strtoull(buf, &e, 10);
4501        unsigned long long old = mddev->suspend_lo;
4502
4503        if (mddev->pers == NULL || 
4504            mddev->pers->quiesce == NULL)
4505                return -EINVAL;
4506        if (buf == e || (*e && *e != '\n'))
4507                return -EINVAL;
4508
4509        mddev->suspend_lo = new;
4510        if (new >= old)
4511                /* Shrinking suspended region */
4512                mddev->pers->quiesce(mddev, 2);
4513        else {
4514                /* Expanding suspended region - need to wait */
4515                mddev->pers->quiesce(mddev, 1);
4516                mddev->pers->quiesce(mddev, 0);
4517        }
4518        return len;
4519}
4520static struct md_sysfs_entry md_suspend_lo =
4521__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4522
4523
4524static ssize_t
4525suspend_hi_show(struct mddev *mddev, char *page)
4526{
4527        return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4528}
4529
4530static ssize_t
4531suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4532{
4533        char *e;
4534        unsigned long long new = simple_strtoull(buf, &e, 10);
4535        unsigned long long old = mddev->suspend_hi;
4536
4537        if (mddev->pers == NULL ||
4538            mddev->pers->quiesce == NULL)
4539                return -EINVAL;
4540        if (buf == e || (*e && *e != '\n'))
4541                return -EINVAL;
4542
4543        mddev->suspend_hi = new;
4544        if (new <= old)
4545                /* Shrinking suspended region */
4546                mddev->pers->quiesce(mddev, 2);
4547        else {
4548                /* Expanding suspended region - need to wait */
4549                mddev->pers->quiesce(mddev, 1);
4550                mddev->pers->quiesce(mddev, 0);
4551        }
4552        return len;
4553}
4554static struct md_sysfs_entry md_suspend_hi =
4555__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4556
4557static ssize_t
4558reshape_position_show(struct mddev *mddev, char *page)
4559{
4560        if (mddev->reshape_position != MaxSector)
4561                return sprintf(page, "%llu\n",
4562                               (unsigned long long)mddev->reshape_position);
4563        strcpy(page, "none\n");
4564        return 5;
4565}
4566
4567static ssize_t
4568reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4569{
4570        struct md_rdev *rdev;
4571        char *e;
4572        unsigned long long new = simple_strtoull(buf, &e, 10);
4573        if (mddev->pers)
4574                return -EBUSY;
4575        if (buf == e || (*e && *e != '\n'))
4576                return -EINVAL;
4577        mddev->reshape_position = new;
4578        mddev->delta_disks = 0;
4579        mddev->reshape_backwards = 0;
4580        mddev->new_level = mddev->level;
4581        mddev->new_layout = mddev->layout;
4582        mddev->new_chunk_sectors = mddev->chunk_sectors;
4583        rdev_for_each(rdev, mddev)
4584                rdev->new_data_offset = rdev->data_offset;
4585        return len;
4586}
4587
4588static struct md_sysfs_entry md_reshape_position =
4589__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4590       reshape_position_store);
4591
4592static ssize_t
4593reshape_direction_show(struct mddev *mddev, char *page)
4594{
4595        return sprintf(page, "%s\n",
4596                       mddev->reshape_backwards ? "backwards" : "forwards");
4597}
4598
4599static ssize_t
4600reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4601{
4602        int backwards = 0;
4603        if (cmd_match(buf, "forwards"))
4604                backwards = 0;
4605        else if (cmd_match(buf, "backwards"))
4606                backwards = 1;
4607        else
4608                return -EINVAL;
4609        if (mddev->reshape_backwards == backwards)
4610                return len;
4611
4612        /* check if we are allowed to change */
4613        if (mddev->delta_disks)
4614                return -EBUSY;
4615
4616        if (mddev->persistent &&
4617            mddev->major_version == 0)
4618                return -EINVAL;
4619
4620        mddev->reshape_backwards = backwards;
4621        return len;
4622}
4623
4624static struct md_sysfs_entry md_reshape_direction =
4625__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4626       reshape_direction_store);
4627
4628static ssize_t
4629array_size_show(struct mddev *mddev, char *page)
4630{
4631        if (mddev->external_size)
4632                return sprintf(page, "%llu\n",
4633                               (unsigned long long)mddev->array_sectors/2);
4634        else
4635                return sprintf(page, "default\n");
4636}
4637
4638static ssize_t
4639array_size_store(struct mddev *mddev, const char *buf, size_t len)
4640{
4641        sector_t sectors;
4642
4643        if (strncmp(buf, "default", 7) == 0) {
4644                if (mddev->pers)
4645                        sectors = mddev->pers->size(mddev, 0, 0);
4646                else
4647                        sectors = mddev->array_sectors;
4648
4649                mddev->external_size = 0;
4650        } else {
4651                if (strict_blocks_to_sectors(buf, &sectors) < 0)
4652                        return -EINVAL;
4653                if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4654                        return -E2BIG;
4655
4656                mddev->external_size = 1;
4657        }
4658
4659        mddev->array_sectors = sectors;
4660        if (mddev->pers) {
4661                set_capacity(mddev->gendisk, mddev->array_sectors);
4662                revalidate_disk(mddev->gendisk);
4663        }
4664        return len;
4665}
4666
4667static struct md_sysfs_entry md_array_size =
4668__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4669       array_size_store);
4670
4671static struct attribute *md_default_attrs[] = {
4672        &md_level.attr,
4673        &md_layout.attr,
4674        &md_raid_disks.attr,
4675        &md_chunk_size.attr,
4676        &md_size.attr,
4677        &md_resync_start.attr,
4678        &md_metadata.attr,
4679        &md_new_device.attr,
4680        &md_safe_delay.attr,
4681        &md_array_state.attr,
4682        &md_reshape_position.attr,
4683        &md_reshape_direction.attr,
4684        &md_array_size.attr,
4685        &max_corr_read_errors.attr,
4686        NULL,
4687};
4688
4689static struct attribute *md_redundancy_attrs[] = {
4690        &md_scan_mode.attr,
4691        &md_mismatches.attr,
4692        &md_sync_min.attr,
4693        &md_sync_max.attr,
4694        &md_sync_speed.attr,
4695        &md_sync_force_parallel.attr,
4696        &md_sync_completed.attr,
4697        &md_min_sync.attr,
4698        &md_max_sync.attr,
4699        &md_suspend_lo.attr,
4700        &md_suspend_hi.attr,
4701        &md_bitmap.attr,
4702        &md_degraded.attr,
4703        NULL,
4704};
4705static struct attribute_group md_redundancy_group = {
4706        .name = NULL,
4707        .attrs = md_redundancy_attrs,
4708};
4709
4710
4711static ssize_t
4712md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4713{
4714        struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4715        struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4716        ssize_t rv;
4717
4718        if (!entry->show)
4719                return -EIO;
4720        spin_lock(&all_mddevs_lock);
4721        if (list_empty(&mddev->all_mddevs)) {
4722                spin_unlock(&all_mddevs_lock);
4723                return -EBUSY;
4724        }
4725        mddev_get(mddev);
4726        spin_unlock(&all_mddevs_lock);
4727
4728        rv = mddev_lock(mddev);
4729        if (!rv) {
4730                rv = entry->show(mddev, page);
4731                mddev_unlock(mddev);
4732        }
4733        mddev_put(mddev);
4734        return rv;
4735}
4736
4737static ssize_t
4738md_attr_store(struct kobject *kobj, struct attribute *attr,
4739              const char *page, size_t length)
4740{
4741        struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4742        struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4743        ssize_t rv;
4744
4745        if (!entry->store)
4746                return -EIO;
4747        if (!capable(CAP_SYS_ADMIN))
4748                return -EACCES;
4749        spin_lock(&all_mddevs_lock);
4750        if (list_empty(&mddev->all_mddevs)) {
4751                spin_unlock(&all_mddevs_lock);
4752                return -EBUSY;
4753        }
4754        mddev_get(mddev);
4755        spin_unlock(&all_mddevs_lock);
4756        rv = mddev_lock(mddev);
4757        if (!rv) {
4758                rv = entry->store(mddev, page, length);
4759                mddev_unlock(mddev);
4760        }
4761        mddev_put(mddev);
4762        return rv;
4763}
4764
4765static void md_free(struct kobject *ko)
4766{
4767        struct mddev *mddev = container_of(ko, struct mddev, kobj);
4768
4769        if (mddev->sysfs_state)
4770                sysfs_put(mddev->sysfs_state);
4771
4772        if (mddev->gendisk) {
4773                del_gendisk(mddev->gendisk);
4774                put_disk(mddev->gendisk);
4775        }
4776        if (mddev->queue)
4777                blk_cleanup_queue(mddev->queue);
4778
4779        kfree(mddev);
4780}
4781
4782static const struct sysfs_ops md_sysfs_ops = {
4783        .show   = md_attr_show,
4784        .store  = md_attr_store,
4785};
4786static struct kobj_type md_ktype = {
4787        .release        = md_free,
4788        .sysfs_ops      = &md_sysfs_ops,
4789        .default_attrs  = md_default_attrs,
4790};
4791
4792int mdp_major = 0;
4793
4794static void mddev_delayed_delete(struct work_struct *ws)
4795{
4796        struct mddev *mddev = container_of(ws, struct mddev, del_work);
4797
4798        sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4799        kobject_del(&mddev->kobj);
4800        kobject_put(&mddev->kobj);
4801}
4802
4803static int md_alloc(dev_t dev, char *name)
4804{
4805        static DEFINE_MUTEX(disks_mutex);
4806        struct mddev *mddev = mddev_find(dev);
4807        struct gendisk *disk;
4808        int partitioned;
4809        int shift;
4810        int unit;
4811        int error;
4812
4813        if (!mddev)
4814                return -ENODEV;
4815
4816        partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4817        shift = partitioned ? MdpMinorShift : 0;
4818        unit = MINOR(mddev->unit) >> shift;
4819
4820        /* wait for any previous instance of this device to be
4821         * completely removed (mddev_delayed_delete).
4822         */
4823        flush_workqueue(md_misc_wq);
4824
4825        mutex_lock(&disks_mutex);
4826        error = -EEXIST;
4827        if (mddev->gendisk)
4828                goto abort;
4829
4830        if (name) {
4831                /* Need to ensure that 'name' is not a duplicate.
4832                 */
4833                struct mddev *mddev2;
4834                spin_lock(&all_mddevs_lock);
4835
4836                list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4837                        if (mddev2->gendisk &&
4838                            strcmp(mddev2->gendisk->disk_name, name) == 0) {
4839                                spin_unlock(&all_mddevs_lock);
4840                                goto abort;
4841                        }
4842                spin_unlock(&all_mddevs_lock);
4843        }
4844
4845        error = -ENOMEM;
4846        mddev->queue = blk_alloc_queue(GFP_KERNEL);
4847        if (!mddev->queue)
4848                goto abort;
4849        mddev->queue->queuedata = mddev;
4850
4851        blk_queue_make_request(mddev->queue, md_make_request);
4852        blk_set_stacking_limits(&mddev->queue->limits);
4853
4854        disk = alloc_disk(1 << shift);
4855        if (!disk) {
4856                blk_cleanup_queue(mddev->queue);
4857                mddev->queue = NULL;
4858                goto abort;
4859        }
4860        disk->major = MAJOR(mddev->unit);
4861        disk->first_minor = unit << shift;
4862        if (name)
4863                strcpy(disk->disk_name, name);
4864        else if (partitioned)
4865                sprintf(disk->disk_name, "md_d%d", unit);
4866        else
4867                sprintf(disk->disk_name, "md%d", unit);
4868        disk->fops = &md_fops;
4869        disk->private_data = mddev;
4870        disk->queue = mddev->queue;
4871        blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4872        /* Allow extended partitions.  This makes the
4873         * 'mdp' device redundant, but we can't really
4874         * remove it now.
4875         */
4876        disk->flags |= GENHD_FL_EXT_DEVT;
4877        mddev->gendisk = disk;
4878        /* As soon as we call add_disk(), another thread could get
4879         * through to md_open, so make sure it doesn't get too far
4880         */
4881        mutex_lock(&mddev->open_mutex);
4882        add_disk(disk);
4883
4884        error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4885                                     &disk_to_dev(disk)->kobj, "%s", "md");
4886        if (error) {
4887                /* This isn't possible, but as kobject_init_and_add is marked
4888                 * __must_check, we must do something with the result
4889                 */
4890                printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4891                       disk->disk_name);
4892                error = 0;
4893        }
4894        if (mddev->kobj.sd &&
4895            sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4896                printk(KERN_DEBUG "pointless warning\n");
4897        mutex_unlock(&mddev->open_mutex);
4898 abort:
4899        mutex_unlock(&disks_mutex);
4900        if (!error && mddev->kobj.sd) {
4901                kobject_uevent(&mddev->kobj, KOBJ_ADD);
4902                mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4903        }
4904        mddev_put(mddev);
4905        return error;
4906}
4907
4908static struct kobject *md_probe(dev_t dev, int *part, void *data)
4909{
4910        md_alloc(dev, NULL);
4911        return NULL;
4912}
4913
4914static int add_named_array(const char *val, struct kernel_param *kp)
4915{
4916        /* val must be "md_*" where * is not all digits.
4917         * We allocate an array with a large free minor number, and
4918         * set the name to val.  val must not already be an active name.
4919         */
4920        int len = strlen(val);
4921        char buf[DISK_NAME_LEN];
4922
4923        while (len && val[len-1] == '\n')
4924                len--;
4925        if (len >= DISK_NAME_LEN)
4926                return -E2BIG;
4927        strlcpy(buf, val, len+1);
4928        if (strncmp(buf, "md_", 3) != 0)
4929                return -EINVAL;
4930        return md_alloc(0, buf);
4931}
4932
4933static void md_safemode_timeout(unsigned long data)
4934{
4935        struct mddev *mddev = (struct mddev *) data;
4936
4937        if (!atomic_read(&mddev->writes_pending)) {
4938                mddev->safemode = 1;
4939                if (mddev->external)
4940                        sysfs_notify_dirent_safe(mddev->sysfs_state);
4941        }
4942        md_wakeup_thread(mddev->thread);
4943}
4944
4945static int start_dirty_degraded;
4946
4947int md_run(struct mddev *mddev)
4948{
4949        int err;
4950        struct md_rdev *rdev;
4951        struct md_personality *pers;
4952
4953        if (list_empty(&mddev->disks))
4954                /* cannot run an array with no devices.. */
4955                return -EINVAL;
4956
4957        if (mddev->pers)
4958                return -EBUSY;
4959        /* Cannot run until previous stop completes properly */
4960        if (mddev->sysfs_active)
4961                return -EBUSY;
4962
4963        /*
4964         * Analyze all RAID superblock(s)
4965         */
4966        if (!mddev->raid_disks) {
4967                if (!mddev->persistent)
4968                        return -EINVAL;
4969                analyze_sbs(mddev);
4970        }
4971
4972        if (mddev->level != LEVEL_NONE)
4973                request_module("md-level-%d", mddev->level);
4974        else if (mddev->clevel[0])
4975                request_module("md-%s", mddev->clevel);
4976
4977        /*
4978         * Drop all container device buffers, from now on
4979         * the only valid external interface is through the md
4980         * device.
4981         */
4982        rdev_for_each(rdev, mddev) {
4983                if (test_bit(Faulty, &rdev->flags))
4984                        continue;
4985                sync_blockdev(rdev->bdev);
4986                invalidate_bdev(rdev->bdev);
4987
4988                /* perform some consistency tests on the device.
4989                 * We don't want the data to overlap the metadata,
4990                 * Internal Bitmap issues have been handled elsewhere.
4991                 */
4992                if (rdev->meta_bdev) {
4993                        /* Nothing to check */;
4994                } else if (rdev->data_offset < rdev->sb_start) {
4995                        if (mddev->dev_sectors &&
4996                            rdev->data_offset + mddev->dev_sectors
4997                            > rdev->sb_start) {
4998                                printk("md: %s: data overlaps metadata\n",
4999                                       mdname(mddev));
5000                                return -EINVAL;
5001                        }
5002                } else {
5003                        if (rdev->sb_start + rdev->sb_size/512
5004                            > rdev->data_offset) {
5005                                printk("md: %s: metadata overlaps data\n",
5006                                       mdname(mddev));
5007                                return -EINVAL;
5008                        }
5009                }
5010                sysfs_notify_dirent_safe(rdev->sysfs_state);
5011        }
5012
5013        if (mddev->bio_set == NULL)
5014                mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5015
5016        spin_lock(&pers_lock);
5017        pers = find_pers(mddev->level, mddev->clevel);
5018        if (!pers || !try_module_get(pers->owner)) {
5019                spin_unlock(&pers_lock);
5020                if (mddev->level != LEVEL_NONE)
5021                        printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5022                               mddev->level);
5023                else
5024                        printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5025                               mddev->clevel);
5026                return -EINVAL;
5027        }
5028        mddev->pers = pers;
5029        spin_unlock(&pers_lock);
5030        if (mddev->level != pers->level) {
5031                mddev->level = pers->level;
5032                mddev->new_level = pers->level;
5033        }
5034        strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5035
5036        if (mddev->reshape_position != MaxSector &&
5037            pers->start_reshape == NULL) {
5038                /* This personality cannot handle reshaping... */
5039                mddev->pers = NULL;
5040                module_put(pers->owner);
5041                return -EINVAL;
5042        }
5043
5044        if (pers->sync_request) {
5045                /* Warn if this is a potentially silly
5046                 * configuration.
5047                 */
5048                char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5049                struct md_rdev *rdev2;
5050                int warned = 0;
5051
5052                rdev_for_each(rdev, mddev)
5053                        rdev_for_each(rdev2, mddev) {
5054                                if (rdev < rdev2 &&
5055                                    rdev->bdev->bd_contains ==
5056                                    rdev2->bdev->bd_contains) {
5057                                        printk(KERN_WARNING
5058                                               "%s: WARNING: %s appears to be"
5059                                               " on the same physical disk as"
5060                                               " %s.\n",
5061                                               mdname(mddev),
5062                                               bdevname(rdev->bdev,b),
5063                                               bdevname(rdev2->bdev,b2));
5064                                        warned = 1;
5065                                }
5066                        }
5067
5068                if (warned)
5069                        printk(KERN_WARNING
5070                               "True protection against single-disk"
5071                               " failure might be compromised.\n");
5072        }
5073
5074        mddev->recovery = 0;
5075        /* may be over-ridden by personality */
5076        mddev->resync_max_sectors = mddev->dev_sectors;
5077
5078        mddev->ok_start_degraded = start_dirty_degraded;
5079
5080        if (start_readonly && mddev->ro == 0)
5081                mddev->ro = 2; /* read-only, but switch on first write */
5082
5083        err = mddev->pers->run(mddev);
5084        if (err)
5085                printk(KERN_ERR "md: pers->run() failed ...\n");
5086        else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5087                WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5088                          " but 'external_size' not in effect?\n", __func__);
5089                printk(KERN_ERR
5090                       "md: invalid array_size %llu > default size %llu\n",
5091                       (unsigned long long)mddev->array_sectors / 2,
5092                       (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5093                err = -EINVAL;
5094                mddev->pers->stop(mddev);
5095        }
5096        if (err == 0 && mddev->pers->sync_request &&
5097            (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5098                err = bitmap_create(mddev);
5099                if (err) {
5100                        printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5101                               mdname(mddev), err);
5102                        mddev->pers->stop(mddev);
5103                }
5104        }
5105        if (err) {
5106                module_put(mddev->pers->owner);
5107                mddev->pers = NULL;
5108                bitmap_destroy(mddev);
5109                return err;
5110        }
5111        if (mddev->pers->sync_request) {
5112                if (mddev->kobj.sd &&
5113                    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5114                        printk(KERN_WARNING
5115                               "md: cannot register extra attributes for %s\n",
5116                               mdname(mddev));
5117                mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5118        } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5119                mddev->ro = 0;
5120
5121        atomic_set(&mddev->writes_pending,0);
5122        atomic_set(&mddev->max_corr_read_errors,
5123                   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5124        mddev->safemode = 0;
5125        mddev->safemode_timer.function = md_safemode_timeout;
5126        mddev->safemode_timer.data = (unsigned long) mddev;
5127        mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5128        mddev->in_sync = 1;
5129        smp_wmb();
5130        mddev->ready = 1;
5131        rdev_for_each(rdev, mddev)
5132                if (rdev->raid_disk >= 0)
5133                        if (sysfs_link_rdev(mddev, rdev))
5134                                /* failure here is OK */;
5135        
5136        set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5137        
5138        if (mddev->flags)
5139                md_update_sb(mddev, 0);
5140
5141        md_new_event(mddev);
5142        sysfs_notify_dirent_safe(mddev->sysfs_state);
5143        sysfs_notify_dirent_safe(mddev->sysfs_action);
5144        sysfs_notify(&mddev->kobj, NULL, "degraded");
5145        return 0;
5146}
5147EXPORT_SYMBOL_GPL(md_run);
5148
5149static int do_md_run(struct mddev *mddev)
5150{
5151        int err;
5152
5153        err = md_run(mddev);
5154        if (err)
5155                goto out;
5156        err = bitmap_load(mddev);
5157        if (err) {
5158                bitmap_destroy(mddev);
5159                goto out;
5160        }
5161
5162        md_wakeup_thread(mddev->thread);
5163        md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5164
5165        set_capacity(mddev->gendisk, mddev->array_sectors);
5166        revalidate_disk(mddev->gendisk);
5167        mddev->changed = 1;
5168        kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5169out:
5170        return err;
5171}
5172
5173static int restart_array(struct mddev *mddev)
5174{
5175        struct gendisk *disk = mddev->gendisk;
5176
5177        /* Complain if it has no devices */
5178        if (list_empty(&mddev->disks))
5179                return -ENXIO;
5180        if (!mddev->pers)
5181                return -EINVAL;
5182        if (!mddev->ro)
5183                return -EBUSY;
5184        mddev->safemode = 0;
5185        mddev->ro = 0;
5186        set_disk_ro(disk, 0);
5187        printk(KERN_INFO "md: %s switched to read-write mode.\n",
5188                mdname(mddev));
5189        /* Kick recovery or resync if necessary */
5190        set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5191        md_wakeup_thread(mddev->thread);
5192        md_wakeup_thread(mddev->sync_thread);
5193        sysfs_notify_dirent_safe(mddev->sysfs_state);
5194        return 0;
5195}
5196
5197/* similar to deny_write_access, but accounts for our holding a reference