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