linux/drivers/md/md-bitmap.c
<<
>>
Prefs
   1/*
   2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   3 *
   4 * bitmap_create  - sets up the bitmap structure
   5 * bitmap_destroy - destroys the bitmap structure
   6 *
   7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   8 * - added disk storage for bitmap
   9 * - changes to allow various bitmap chunk sizes
  10 */
  11
  12/*
  13 * Still to do:
  14 *
  15 * flush after percent set rather than just time based. (maybe both).
  16 */
  17
  18#include <linux/blkdev.h>
  19#include <linux/module.h>
  20#include <linux/errno.h>
  21#include <linux/slab.h>
  22#include <linux/init.h>
  23#include <linux/timer.h>
  24#include <linux/sched.h>
  25#include <linux/list.h>
  26#include <linux/file.h>
  27#include <linux/mount.h>
  28#include <linux/buffer_head.h>
  29#include <linux/seq_file.h>
  30#include <trace/events/block.h>
  31#include "md.h"
  32#include "md-bitmap.h"
  33
  34static inline char *bmname(struct bitmap *bitmap)
  35{
  36        return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  37}
  38
  39/*
  40 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
  41 *
  42 * 1) check to see if this page is allocated, if it's not then try to alloc
  43 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
  44 *    page pointer directly as a counter
  45 *
  46 * if we find our page, we increment the page's refcount so that it stays
  47 * allocated while we're using it
  48 */
  49static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
  50                               unsigned long page, int create, int no_hijack)
  51__releases(bitmap->lock)
  52__acquires(bitmap->lock)
  53{
  54        unsigned char *mappage;
  55
  56        if (page >= bitmap->pages) {
  57                /* This can happen if bitmap_start_sync goes beyond
  58                 * End-of-device while looking for a whole page.
  59                 * It is harmless.
  60                 */
  61                return -EINVAL;
  62        }
  63
  64        if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  65                return 0;
  66
  67        if (bitmap->bp[page].map) /* page is already allocated, just return */
  68                return 0;
  69
  70        if (!create)
  71                return -ENOENT;
  72
  73        /* this page has not been allocated yet */
  74
  75        spin_unlock_irq(&bitmap->lock);
  76        /* It is possible that this is being called inside a
  77         * prepare_to_wait/finish_wait loop from raid5c:make_request().
  78         * In general it is not permitted to sleep in that context as it
  79         * can cause the loop to spin freely.
  80         * That doesn't apply here as we can only reach this point
  81         * once with any loop.
  82         * When this function completes, either bp[page].map or
  83         * bp[page].hijacked.  In either case, this function will
  84         * abort before getting to this point again.  So there is
  85         * no risk of a free-spin, and so it is safe to assert
  86         * that sleeping here is allowed.
  87         */
  88        sched_annotate_sleep();
  89        mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  90        spin_lock_irq(&bitmap->lock);
  91
  92        if (mappage == NULL) {
  93                pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  94                /* We don't support hijack for cluster raid */
  95                if (no_hijack)
  96                        return -ENOMEM;
  97                /* failed - set the hijacked flag so that we can use the
  98                 * pointer as a counter */
  99                if (!bitmap->bp[page].map)
 100                        bitmap->bp[page].hijacked = 1;
 101        } else if (bitmap->bp[page].map ||
 102                   bitmap->bp[page].hijacked) {
 103                /* somebody beat us to getting the page */
 104                kfree(mappage);
 105        } else {
 106
 107                /* no page was in place and we have one, so install it */
 108
 109                bitmap->bp[page].map = mappage;
 110                bitmap->missing_pages--;
 111        }
 112        return 0;
 113}
 114
 115/* if page is completely empty, put it back on the free list, or dealloc it */
 116/* if page was hijacked, unmark the flag so it might get alloced next time */
 117/* Note: lock should be held when calling this */
 118static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 119{
 120        char *ptr;
 121
 122        if (bitmap->bp[page].count) /* page is still busy */
 123                return;
 124
 125        /* page is no longer in use, it can be released */
 126
 127        if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 128                bitmap->bp[page].hijacked = 0;
 129                bitmap->bp[page].map = NULL;
 130        } else {
 131                /* normal case, free the page */
 132                ptr = bitmap->bp[page].map;
 133                bitmap->bp[page].map = NULL;
 134                bitmap->missing_pages++;
 135                kfree(ptr);
 136        }
 137}
 138
 139/*
 140 * bitmap file handling - read and write the bitmap file and its superblock
 141 */
 142
 143/*
 144 * basic page I/O operations
 145 */
 146
 147/* IO operations when bitmap is stored near all superblocks */
 148static int read_sb_page(struct mddev *mddev, loff_t offset,
 149                        struct page *page,
 150                        unsigned long index, int size)
 151{
 152        /* choose a good rdev and read the page from there */
 153
 154        struct md_rdev *rdev;
 155        sector_t target;
 156
 157        rdev_for_each(rdev, mddev) {
 158                if (! test_bit(In_sync, &rdev->flags)
 159                    || test_bit(Faulty, &rdev->flags)
 160                    || test_bit(Bitmap_sync, &rdev->flags))
 161                        continue;
 162
 163                target = offset + index * (PAGE_SIZE/512);
 164
 165                if (sync_page_io(rdev, target,
 166                                 roundup(size, bdev_logical_block_size(rdev->bdev)),
 167                                 page, REQ_OP_READ, 0, true)) {
 168                        page->index = index;
 169                        return 0;
 170                }
 171        }
 172        return -EIO;
 173}
 174
 175static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 176{
 177        /* Iterate the disks of an mddev, using rcu to protect access to the
 178         * linked list, and raising the refcount of devices we return to ensure
 179         * they don't disappear while in use.
 180         * As devices are only added or removed when raid_disk is < 0 and
 181         * nr_pending is 0 and In_sync is clear, the entries we return will
 182         * still be in the same position on the list when we re-enter
 183         * list_for_each_entry_continue_rcu.
 184         *
 185         * Note that if entered with 'rdev == NULL' to start at the
 186         * beginning, we temporarily assign 'rdev' to an address which
 187         * isn't really an rdev, but which can be used by
 188         * list_for_each_entry_continue_rcu() to find the first entry.
 189         */
 190        rcu_read_lock();
 191        if (rdev == NULL)
 192                /* start at the beginning */
 193                rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 194        else {
 195                /* release the previous rdev and start from there. */
 196                rdev_dec_pending(rdev, mddev);
 197        }
 198        list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 199                if (rdev->raid_disk >= 0 &&
 200                    !test_bit(Faulty, &rdev->flags)) {
 201                        /* this is a usable devices */
 202                        atomic_inc(&rdev->nr_pending);
 203                        rcu_read_unlock();
 204                        return rdev;
 205                }
 206        }
 207        rcu_read_unlock();
 208        return NULL;
 209}
 210
 211static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 212{
 213        struct md_rdev *rdev;
 214        struct block_device *bdev;
 215        struct mddev *mddev = bitmap->mddev;
 216        struct bitmap_storage *store = &bitmap->storage;
 217
 218restart:
 219        rdev = NULL;
 220        while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 221                int size = PAGE_SIZE;
 222                loff_t offset = mddev->bitmap_info.offset;
 223
 224                bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 225
 226                if (page->index == store->file_pages-1) {
 227                        int last_page_size = store->bytes & (PAGE_SIZE-1);
 228                        if (last_page_size == 0)
 229                                last_page_size = PAGE_SIZE;
 230                        size = roundup(last_page_size,
 231                                       bdev_logical_block_size(bdev));
 232                }
 233                /* Just make sure we aren't corrupting data or
 234                 * metadata
 235                 */
 236                if (mddev->external) {
 237                        /* Bitmap could be anywhere. */
 238                        if (rdev->sb_start + offset + (page->index
 239                                                       * (PAGE_SIZE/512))
 240                            > rdev->data_offset
 241                            &&
 242                            rdev->sb_start + offset
 243                            < (rdev->data_offset + mddev->dev_sectors
 244                             + (PAGE_SIZE/512)))
 245                                goto bad_alignment;
 246                } else if (offset < 0) {
 247                        /* DATA  BITMAP METADATA  */
 248                        if (offset
 249                            + (long)(page->index * (PAGE_SIZE/512))
 250                            + size/512 > 0)
 251                                /* bitmap runs in to metadata */
 252                                goto bad_alignment;
 253                        if (rdev->data_offset + mddev->dev_sectors
 254                            > rdev->sb_start + offset)
 255                                /* data runs in to bitmap */
 256                                goto bad_alignment;
 257                } else if (rdev->sb_start < rdev->data_offset) {
 258                        /* METADATA BITMAP DATA */
 259                        if (rdev->sb_start
 260                            + offset
 261                            + page->index*(PAGE_SIZE/512) + size/512
 262                            > rdev->data_offset)
 263                                /* bitmap runs in to data */
 264                                goto bad_alignment;
 265                } else {
 266                        /* DATA METADATA BITMAP - no problems */
 267                }
 268                md_super_write(mddev, rdev,
 269                               rdev->sb_start + offset
 270                               + page->index * (PAGE_SIZE/512),
 271                               size,
 272                               page);
 273        }
 274
 275        if (wait && md_super_wait(mddev) < 0)
 276                goto restart;
 277        return 0;
 278
 279 bad_alignment:
 280        return -EINVAL;
 281}
 282
 283static void md_bitmap_file_kick(struct bitmap *bitmap);
 284/*
 285 * write out a page to a file
 286 */
 287static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 288{
 289        struct buffer_head *bh;
 290
 291        if (bitmap->storage.file == NULL) {
 292                switch (write_sb_page(bitmap, page, wait)) {
 293                case -EINVAL:
 294                        set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 295                }
 296        } else {
 297
 298                bh = page_buffers(page);
 299
 300                while (bh && bh->b_blocknr) {
 301                        atomic_inc(&bitmap->pending_writes);
 302                        set_buffer_locked(bh);
 303                        set_buffer_mapped(bh);
 304                        submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
 305                        bh = bh->b_this_page;
 306                }
 307
 308                if (wait)
 309                        wait_event(bitmap->write_wait,
 310                                   atomic_read(&bitmap->pending_writes)==0);
 311        }
 312        if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 313                md_bitmap_file_kick(bitmap);
 314}
 315
 316static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 317{
 318        struct bitmap *bitmap = bh->b_private;
 319
 320        if (!uptodate)
 321                set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 322        if (atomic_dec_and_test(&bitmap->pending_writes))
 323                wake_up(&bitmap->write_wait);
 324}
 325
 326/* copied from buffer.c */
 327static void
 328__clear_page_buffers(struct page *page)
 329{
 330        ClearPagePrivate(page);
 331        set_page_private(page, 0);
 332        put_page(page);
 333}
 334static void free_buffers(struct page *page)
 335{
 336        struct buffer_head *bh;
 337
 338        if (!PagePrivate(page))
 339                return;
 340
 341        bh = page_buffers(page);
 342        while (bh) {
 343                struct buffer_head *next = bh->b_this_page;
 344                free_buffer_head(bh);
 345                bh = next;
 346        }
 347        __clear_page_buffers(page);
 348        put_page(page);
 349}
 350
 351/* read a page from a file.
 352 * We both read the page, and attach buffers to the page to record the
 353 * address of each block (using bmap).  These addresses will be used
 354 * to write the block later, completely bypassing the filesystem.
 355 * This usage is similar to how swap files are handled, and allows us
 356 * to write to a file with no concerns of memory allocation failing.
 357 */
 358static int read_page(struct file *file, unsigned long index,
 359                     struct bitmap *bitmap,
 360                     unsigned long count,
 361                     struct page *page)
 362{
 363        int ret = 0;
 364        struct inode *inode = file_inode(file);
 365        struct buffer_head *bh;
 366        sector_t block;
 367        unsigned long blocksize = i_blocksize(inode);
 368
 369        pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 370                 (unsigned long long)index << PAGE_SHIFT);
 371
 372        bh = alloc_page_buffers(page, blocksize, false);
 373        if (!bh) {
 374                ret = -ENOMEM;
 375                goto out;
 376        }
 377        attach_page_buffers(page, bh);
 378        block = index << (PAGE_SHIFT - inode->i_blkbits);
 379        while (bh) {
 380                if (count == 0)
 381                        bh->b_blocknr = 0;
 382                else {
 383                        bh->b_blocknr = bmap(inode, block);
 384                        if (bh->b_blocknr == 0) {
 385                                /* Cannot use this file! */
 386                                ret = -EINVAL;
 387                                goto out;
 388                        }
 389                        bh->b_bdev = inode->i_sb->s_bdev;
 390                        if (count < blocksize)
 391                                count = 0;
 392                        else
 393                                count -= blocksize;
 394
 395                        bh->b_end_io = end_bitmap_write;
 396                        bh->b_private = bitmap;
 397                        atomic_inc(&bitmap->pending_writes);
 398                        set_buffer_locked(bh);
 399                        set_buffer_mapped(bh);
 400                        submit_bh(REQ_OP_READ, 0, bh);
 401                }
 402                block++;
 403                bh = bh->b_this_page;
 404        }
 405        page->index = index;
 406
 407        wait_event(bitmap->write_wait,
 408                   atomic_read(&bitmap->pending_writes)==0);
 409        if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 410                ret = -EIO;
 411out:
 412        if (ret)
 413                pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 414                       (int)PAGE_SIZE,
 415                       (unsigned long long)index << PAGE_SHIFT,
 416                       ret);
 417        return ret;
 418}
 419
 420/*
 421 * bitmap file superblock operations
 422 */
 423
 424/*
 425 * md_bitmap_wait_writes() should be called before writing any bitmap
 426 * blocks, to ensure previous writes, particularly from
 427 * md_bitmap_daemon_work(), have completed.
 428 */
 429static void md_bitmap_wait_writes(struct bitmap *bitmap)
 430{
 431        if (bitmap->storage.file)
 432                wait_event(bitmap->write_wait,
 433                           atomic_read(&bitmap->pending_writes)==0);
 434        else
 435                /* Note that we ignore the return value.  The writes
 436                 * might have failed, but that would just mean that
 437                 * some bits which should be cleared haven't been,
 438                 * which is safe.  The relevant bitmap blocks will
 439                 * probably get written again, but there is no great
 440                 * loss if they aren't.
 441                 */
 442                md_super_wait(bitmap->mddev);
 443}
 444
 445
 446/* update the event counter and sync the superblock to disk */
 447void md_bitmap_update_sb(struct bitmap *bitmap)
 448{
 449        bitmap_super_t *sb;
 450
 451        if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 452                return;
 453        if (bitmap->mddev->bitmap_info.external)
 454                return;
 455        if (!bitmap->storage.sb_page) /* no superblock */
 456                return;
 457        sb = kmap_atomic(bitmap->storage.sb_page);
 458        sb->events = cpu_to_le64(bitmap->mddev->events);
 459        if (bitmap->mddev->events < bitmap->events_cleared)
 460                /* rocking back to read-only */
 461                bitmap->events_cleared = bitmap->mddev->events;
 462        sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 463        /*
 464         * clear BITMAP_WRITE_ERROR bit to protect against the case that
 465         * a bitmap write error occurred but the later writes succeeded.
 466         */
 467        sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 468        /* Just in case these have been changed via sysfs: */
 469        sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 470        sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 471        /* This might have been changed by a reshape */
 472        sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 473        sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 474        sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 475        sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 476                                           bitmap_info.space);
 477        kunmap_atomic(sb);
 478        write_page(bitmap, bitmap->storage.sb_page, 1);
 479}
 480EXPORT_SYMBOL(md_bitmap_update_sb);
 481
 482/* print out the bitmap file superblock */
 483void md_bitmap_print_sb(struct bitmap *bitmap)
 484{
 485        bitmap_super_t *sb;
 486
 487        if (!bitmap || !bitmap->storage.sb_page)
 488                return;
 489        sb = kmap_atomic(bitmap->storage.sb_page);
 490        pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 491        pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 492        pr_debug("       version: %d\n", le32_to_cpu(sb->version));
 493        pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 494                 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 495                 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 496                 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 497                 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 498        pr_debug("        events: %llu\n",
 499                 (unsigned long long) le64_to_cpu(sb->events));
 500        pr_debug("events cleared: %llu\n",
 501                 (unsigned long long) le64_to_cpu(sb->events_cleared));
 502        pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 503        pr_debug("     chunksize: %d B\n", le32_to_cpu(sb->chunksize));
 504        pr_debug("  daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
 505        pr_debug("     sync size: %llu KB\n",
 506                 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 507        pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind));
 508        kunmap_atomic(sb);
 509}
 510
 511/*
 512 * bitmap_new_disk_sb
 513 * @bitmap
 514 *
 515 * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 516 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 517 * This function verifies 'bitmap_info' and populates the on-disk bitmap
 518 * structure, which is to be written to disk.
 519 *
 520 * Returns: 0 on success, -Exxx on error
 521 */
 522static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 523{
 524        bitmap_super_t *sb;
 525        unsigned long chunksize, daemon_sleep, write_behind;
 526
 527        bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 528        if (bitmap->storage.sb_page == NULL)
 529                return -ENOMEM;
 530        bitmap->storage.sb_page->index = 0;
 531
 532        sb = kmap_atomic(bitmap->storage.sb_page);
 533
 534        sb->magic = cpu_to_le32(BITMAP_MAGIC);
 535        sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 536
 537        chunksize = bitmap->mddev->bitmap_info.chunksize;
 538        BUG_ON(!chunksize);
 539        if (!is_power_of_2(chunksize)) {
 540                kunmap_atomic(sb);
 541                pr_warn("bitmap chunksize not a power of 2\n");
 542                return -EINVAL;
 543        }
 544        sb->chunksize = cpu_to_le32(chunksize);
 545
 546        daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 547        if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 548                pr_debug("Choosing daemon_sleep default (5 sec)\n");
 549                daemon_sleep = 5 * HZ;
 550        }
 551        sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 552        bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 553
 554        /*
 555         * FIXME: write_behind for RAID1.  If not specified, what
 556         * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 557         */
 558        write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 559        if (write_behind > COUNTER_MAX)
 560                write_behind = COUNTER_MAX / 2;
 561        sb->write_behind = cpu_to_le32(write_behind);
 562        bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 563
 564        /* keep the array size field of the bitmap superblock up to date */
 565        sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 566
 567        memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 568
 569        set_bit(BITMAP_STALE, &bitmap->flags);
 570        sb->state = cpu_to_le32(bitmap->flags);
 571        bitmap->events_cleared = bitmap->mddev->events;
 572        sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 573        bitmap->mddev->bitmap_info.nodes = 0;
 574
 575        kunmap_atomic(sb);
 576
 577        return 0;
 578}
 579
 580/* read the superblock from the bitmap file and initialize some bitmap fields */
 581static int md_bitmap_read_sb(struct bitmap *bitmap)
 582{
 583        char *reason = NULL;
 584        bitmap_super_t *sb;
 585        unsigned long chunksize, daemon_sleep, write_behind;
 586        unsigned long long events;
 587        int nodes = 0;
 588        unsigned long sectors_reserved = 0;
 589        int err = -EINVAL;
 590        struct page *sb_page;
 591        loff_t offset = bitmap->mddev->bitmap_info.offset;
 592
 593        if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 594                chunksize = 128 * 1024 * 1024;
 595                daemon_sleep = 5 * HZ;
 596                write_behind = 0;
 597                set_bit(BITMAP_STALE, &bitmap->flags);
 598                err = 0;
 599                goto out_no_sb;
 600        }
 601        /* page 0 is the superblock, read it... */
 602        sb_page = alloc_page(GFP_KERNEL);
 603        if (!sb_page)
 604                return -ENOMEM;
 605        bitmap->storage.sb_page = sb_page;
 606
 607re_read:
 608        /* If cluster_slot is set, the cluster is setup */
 609        if (bitmap->cluster_slot >= 0) {
 610                sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 611
 612                bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 613                           (bitmap->mddev->bitmap_info.chunksize >> 9));
 614                /* bits to bytes */
 615                bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 616                /* to 4k blocks */
 617                bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 618                offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
 619                pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 620                        bitmap->cluster_slot, offset);
 621        }
 622
 623        if (bitmap->storage.file) {
 624                loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 625                int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 626
 627                err = read_page(bitmap->storage.file, 0,
 628                                bitmap, bytes, sb_page);
 629        } else {
 630                err = read_sb_page(bitmap->mddev,
 631                                   offset,
 632                                   sb_page,
 633                                   0, sizeof(bitmap_super_t));
 634        }
 635        if (err)
 636                return err;
 637
 638        err = -EINVAL;
 639        sb = kmap_atomic(sb_page);
 640
 641        chunksize = le32_to_cpu(sb->chunksize);
 642        daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 643        write_behind = le32_to_cpu(sb->write_behind);
 644        sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 645        /* Setup nodes/clustername only if bitmap version is
 646         * cluster-compatible
 647         */
 648        if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 649                nodes = le32_to_cpu(sb->nodes);
 650                strlcpy(bitmap->mddev->bitmap_info.cluster_name,
 651                                sb->cluster_name, 64);
 652        }
 653
 654        /* verify that the bitmap-specific fields are valid */
 655        if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 656                reason = "bad magic";
 657        else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 658                 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 659                reason = "unrecognized superblock version";
 660        else if (chunksize < 512)
 661                reason = "bitmap chunksize too small";
 662        else if (!is_power_of_2(chunksize))
 663                reason = "bitmap chunksize not a power of 2";
 664        else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 665                reason = "daemon sleep period out of range";
 666        else if (write_behind > COUNTER_MAX)
 667                reason = "write-behind limit out of range (0 - 16383)";
 668        if (reason) {
 669                pr_warn("%s: invalid bitmap file superblock: %s\n",
 670                        bmname(bitmap), reason);
 671                goto out;
 672        }
 673
 674        /* keep the array size field of the bitmap superblock up to date */
 675        sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 676
 677        if (bitmap->mddev->persistent) {
 678                /*
 679                 * We have a persistent array superblock, so compare the
 680                 * bitmap's UUID and event counter to the mddev's
 681                 */
 682                if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 683                        pr_warn("%s: bitmap superblock UUID mismatch\n",
 684                                bmname(bitmap));
 685                        goto out;
 686                }
 687                events = le64_to_cpu(sb->events);
 688                if (!nodes && (events < bitmap->mddev->events)) {
 689                        pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 690                                bmname(bitmap), events,
 691                                (unsigned long long) bitmap->mddev->events);
 692                        set_bit(BITMAP_STALE, &bitmap->flags);
 693                }
 694        }
 695
 696        /* assign fields using values from superblock */
 697        bitmap->flags |= le32_to_cpu(sb->state);
 698        if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 699                set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 700        bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 701        strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 702        err = 0;
 703
 704out:
 705        kunmap_atomic(sb);
 706        /* Assigning chunksize is required for "re_read" */
 707        bitmap->mddev->bitmap_info.chunksize = chunksize;
 708        if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 709                err = md_setup_cluster(bitmap->mddev, nodes);
 710                if (err) {
 711                        pr_warn("%s: Could not setup cluster service (%d)\n",
 712                                bmname(bitmap), err);
 713                        goto out_no_sb;
 714                }
 715                bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 716                goto re_read;
 717        }
 718
 719
 720out_no_sb:
 721        if (test_bit(BITMAP_STALE, &bitmap->flags))
 722                bitmap->events_cleared = bitmap->mddev->events;
 723        bitmap->mddev->bitmap_info.chunksize = chunksize;
 724        bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 725        bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 726        bitmap->mddev->bitmap_info.nodes = nodes;
 727        if (bitmap->mddev->bitmap_info.space == 0 ||
 728            bitmap->mddev->bitmap_info.space > sectors_reserved)
 729                bitmap->mddev->bitmap_info.space = sectors_reserved;
 730        if (err) {
 731                md_bitmap_print_sb(bitmap);
 732                if (bitmap->cluster_slot < 0)
 733                        md_cluster_stop(bitmap->mddev);
 734        }
 735        return err;
 736}
 737
 738/*
 739 * general bitmap file operations
 740 */
 741
 742/*
 743 * on-disk bitmap:
 744 *
 745 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 746 * file a page at a time. There's a superblock at the start of the file.
 747 */
 748/* calculate the index of the page that contains this bit */
 749static inline unsigned long file_page_index(struct bitmap_storage *store,
 750                                            unsigned long chunk)
 751{
 752        if (store->sb_page)
 753                chunk += sizeof(bitmap_super_t) << 3;
 754        return chunk >> PAGE_BIT_SHIFT;
 755}
 756
 757/* calculate the (bit) offset of this bit within a page */
 758static inline unsigned long file_page_offset(struct bitmap_storage *store,
 759                                             unsigned long chunk)
 760{
 761        if (store->sb_page)
 762                chunk += sizeof(bitmap_super_t) << 3;
 763        return chunk & (PAGE_BITS - 1);
 764}
 765
 766/*
 767 * return a pointer to the page in the filemap that contains the given bit
 768 *
 769 */
 770static inline struct page *filemap_get_page(struct bitmap_storage *store,
 771                                            unsigned long chunk)
 772{
 773        if (file_page_index(store, chunk) >= store->file_pages)
 774                return NULL;
 775        return store->filemap[file_page_index(store, chunk)];
 776}
 777
 778static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 779                                   unsigned long chunks, int with_super,
 780                                   int slot_number)
 781{
 782        int pnum, offset = 0;
 783        unsigned long num_pages;
 784        unsigned long bytes;
 785
 786        bytes = DIV_ROUND_UP(chunks, 8);
 787        if (with_super)
 788                bytes += sizeof(bitmap_super_t);
 789
 790        num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 791        offset = slot_number * num_pages;
 792
 793        store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 794                                       GFP_KERNEL);
 795        if (!store->filemap)
 796                return -ENOMEM;
 797
 798        if (with_super && !store->sb_page) {
 799                store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 800                if (store->sb_page == NULL)
 801                        return -ENOMEM;
 802        }
 803
 804        pnum = 0;
 805        if (store->sb_page) {
 806                store->filemap[0] = store->sb_page;
 807                pnum = 1;
 808                store->sb_page->index = offset;
 809        }
 810
 811        for ( ; pnum < num_pages; pnum++) {
 812                store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 813                if (!store->filemap[pnum]) {
 814                        store->file_pages = pnum;
 815                        return -ENOMEM;
 816                }
 817                store->filemap[pnum]->index = pnum + offset;
 818        }
 819        store->file_pages = pnum;
 820
 821        /* We need 4 bits per page, rounded up to a multiple
 822         * of sizeof(unsigned long) */
 823        store->filemap_attr = kzalloc(
 824                roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 825                GFP_KERNEL);
 826        if (!store->filemap_attr)
 827                return -ENOMEM;
 828
 829        store->bytes = bytes;
 830
 831        return 0;
 832}
 833
 834static void md_bitmap_file_unmap(struct bitmap_storage *store)
 835{
 836        struct page **map, *sb_page;
 837        int pages;
 838        struct file *file;
 839
 840        file = store->file;
 841        map = store->filemap;
 842        pages = store->file_pages;
 843        sb_page = store->sb_page;
 844
 845        while (pages--)
 846                if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 847                        free_buffers(map[pages]);
 848        kfree(map);
 849        kfree(store->filemap_attr);
 850
 851        if (sb_page)
 852                free_buffers(sb_page);
 853
 854        if (file) {
 855                struct inode *inode = file_inode(file);
 856                invalidate_mapping_pages(inode->i_mapping, 0, -1);
 857                fput(file);
 858        }
 859}
 860
 861/*
 862 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 863 * then it is no longer reliable, so we stop using it and we mark the file
 864 * as failed in the superblock
 865 */
 866static void md_bitmap_file_kick(struct bitmap *bitmap)
 867{
 868        char *path, *ptr = NULL;
 869
 870        if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 871                md_bitmap_update_sb(bitmap);
 872
 873                if (bitmap->storage.file) {
 874                        path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 875                        if (path)
 876                                ptr = file_path(bitmap->storage.file,
 877                                             path, PAGE_SIZE);
 878
 879                        pr_warn("%s: kicking failed bitmap file %s from array!\n",
 880                                bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 881
 882                        kfree(path);
 883                } else
 884                        pr_warn("%s: disabling internal bitmap due to errors\n",
 885                                bmname(bitmap));
 886        }
 887}
 888
 889enum bitmap_page_attr {
 890        BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 891        BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 892                                    * i.e. counter is 1 or 2. */
 893        BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 894};
 895
 896static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 897                                 enum bitmap_page_attr attr)
 898{
 899        set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 900}
 901
 902static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 903                                   enum bitmap_page_attr attr)
 904{
 905        clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 906}
 907
 908static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 909                                 enum bitmap_page_attr attr)
 910{
 911        return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 912}
 913
 914static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 915                                           enum bitmap_page_attr attr)
 916{
 917        return test_and_clear_bit((pnum<<2) + attr,
 918                                  bitmap->storage.filemap_attr);
 919}
 920/*
 921 * bitmap_file_set_bit -- called before performing a write to the md device
 922 * to set (and eventually sync) a particular bit in the bitmap file
 923 *
 924 * we set the bit immediately, then we record the page number so that
 925 * when an unplug occurs, we can flush the dirty pages out to disk
 926 */
 927static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 928{
 929        unsigned long bit;
 930        struct page *page;
 931        void *kaddr;
 932        unsigned long chunk = block >> bitmap->counts.chunkshift;
 933        struct bitmap_storage *store = &bitmap->storage;
 934        unsigned long node_offset = 0;
 935
 936        if (mddev_is_clustered(bitmap->mddev))
 937                node_offset = bitmap->cluster_slot * store->file_pages;
 938
 939        page = filemap_get_page(&bitmap->storage, chunk);
 940        if (!page)
 941                return;
 942        bit = file_page_offset(&bitmap->storage, chunk);
 943
 944        /* set the bit */
 945        kaddr = kmap_atomic(page);
 946        if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 947                set_bit(bit, kaddr);
 948        else
 949                set_bit_le(bit, kaddr);
 950        kunmap_atomic(kaddr);
 951        pr_debug("set file bit %lu page %lu\n", bit, page->index);
 952        /* record page number so it gets flushed to disk when unplug occurs */
 953        set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
 954}
 955
 956static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 957{
 958        unsigned long bit;
 959        struct page *page;
 960        void *paddr;
 961        unsigned long chunk = block >> bitmap->counts.chunkshift;
 962        struct bitmap_storage *store = &bitmap->storage;
 963        unsigned long node_offset = 0;
 964
 965        if (mddev_is_clustered(bitmap->mddev))
 966                node_offset = bitmap->cluster_slot * store->file_pages;
 967
 968        page = filemap_get_page(&bitmap->storage, chunk);
 969        if (!page)
 970                return;
 971        bit = file_page_offset(&bitmap->storage, chunk);
 972        paddr = kmap_atomic(page);
 973        if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 974                clear_bit(bit, paddr);
 975        else
 976                clear_bit_le(bit, paddr);
 977        kunmap_atomic(paddr);
 978        if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
 979                set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
 980                bitmap->allclean = 0;
 981        }
 982}
 983
 984static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
 985{
 986        unsigned long bit;
 987        struct page *page;
 988        void *paddr;
 989        unsigned long chunk = block >> bitmap->counts.chunkshift;
 990        int set = 0;
 991
 992        page = filemap_get_page(&bitmap->storage, chunk);
 993        if (!page)
 994                return -EINVAL;
 995        bit = file_page_offset(&bitmap->storage, chunk);
 996        paddr = kmap_atomic(page);
 997        if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 998                set = test_bit(bit, paddr);
 999        else
1000                set = test_bit_le(bit, paddr);
1001        kunmap_atomic(paddr);
1002        return set;
1003}
1004
1005
1006/* this gets called when the md device is ready to unplug its underlying
1007 * (slave) device queues -- before we let any writes go down, we need to
1008 * sync the dirty pages of the bitmap file to disk */
1009void md_bitmap_unplug(struct bitmap *bitmap)
1010{
1011        unsigned long i;
1012        int dirty, need_write;
1013        int writing = 0;
1014
1015        if (!bitmap || !bitmap->storage.filemap ||
1016            test_bit(BITMAP_STALE, &bitmap->flags))
1017                return;
1018
1019        /* look at each page to see if there are any set bits that need to be
1020         * flushed out to disk */
1021        for (i = 0; i < bitmap->storage.file_pages; i++) {
1022                dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1023                need_write = test_and_clear_page_attr(bitmap, i,
1024                                                      BITMAP_PAGE_NEEDWRITE);
1025                if (dirty || need_write) {
1026                        if (!writing) {
1027                                md_bitmap_wait_writes(bitmap);
1028                                if (bitmap->mddev->queue)
1029                                        blk_add_trace_msg(bitmap->mddev->queue,
1030                                                          "md bitmap_unplug");
1031                        }
1032                        clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1033                        write_page(bitmap, bitmap->storage.filemap[i], 0);
1034                        writing = 1;
1035                }
1036        }
1037        if (writing)
1038                md_bitmap_wait_writes(bitmap);
1039
1040        if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1041                md_bitmap_file_kick(bitmap);
1042}
1043EXPORT_SYMBOL(md_bitmap_unplug);
1044
1045static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1046/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1047 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1048 * memory mapping of the bitmap file
1049 * Special cases:
1050 *   if there's no bitmap file, or if the bitmap file had been
1051 *   previously kicked from the array, we mark all the bits as
1052 *   1's in order to cause a full resync.
1053 *
1054 * We ignore all bits for sectors that end earlier than 'start'.
1055 * This is used when reading an out-of-date bitmap...
1056 */
1057static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1058{
1059        unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1060        struct page *page = NULL;
1061        unsigned long bit_cnt = 0;
1062        struct file *file;
1063        unsigned long offset;
1064        int outofdate;
1065        int ret = -ENOSPC;
1066        void *paddr;
1067        struct bitmap_storage *store = &bitmap->storage;
1068
1069        chunks = bitmap->counts.chunks;
1070        file = store->file;
1071
1072        if (!file && !bitmap->mddev->bitmap_info.offset) {
1073                /* No permanent bitmap - fill with '1s'. */
1074                store->filemap = NULL;
1075                store->file_pages = 0;
1076                for (i = 0; i < chunks ; i++) {
1077                        /* if the disk bit is set, set the memory bit */
1078                        int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1079                                      >= start);
1080                        md_bitmap_set_memory_bits(bitmap,
1081                                                  (sector_t)i << bitmap->counts.chunkshift,
1082                                                  needed);
1083                }
1084                return 0;
1085        }
1086
1087        outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1088        if (outofdate)
1089                pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1090
1091        if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1092                pr_warn("%s: bitmap file too short %lu < %lu\n",
1093                        bmname(bitmap),
1094                        (unsigned long) i_size_read(file->f_mapping->host),
1095                        store->bytes);
1096                goto err;
1097        }
1098
1099        oldindex = ~0L;
1100        offset = 0;
1101        if (!bitmap->mddev->bitmap_info.external)
1102                offset = sizeof(bitmap_super_t);
1103
1104        if (mddev_is_clustered(bitmap->mddev))
1105                node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1106
1107        for (i = 0; i < chunks; i++) {
1108                int b;
1109                index = file_page_index(&bitmap->storage, i);
1110                bit = file_page_offset(&bitmap->storage, i);
1111                if (index != oldindex) { /* this is a new page, read it in */
1112                        int count;
1113                        /* unmap the old page, we're done with it */
1114                        if (index == store->file_pages-1)
1115                                count = store->bytes - index * PAGE_SIZE;
1116                        else
1117                                count = PAGE_SIZE;
1118                        page = store->filemap[index];
1119                        if (file)
1120                                ret = read_page(file, index, bitmap,
1121                                                count, page);
1122                        else
1123                                ret = read_sb_page(
1124                                        bitmap->mddev,
1125                                        bitmap->mddev->bitmap_info.offset,
1126                                        page,
1127                                        index + node_offset, count);
1128
1129                        if (ret)
1130                                goto err;
1131
1132                        oldindex = index;
1133
1134                        if (outofdate) {
1135                                /*
1136                                 * if bitmap is out of date, dirty the
1137                                 * whole page and write it out
1138                                 */
1139                                paddr = kmap_atomic(page);
1140                                memset(paddr + offset, 0xff,
1141                                       PAGE_SIZE - offset);
1142                                kunmap_atomic(paddr);
1143                                write_page(bitmap, page, 1);
1144
1145                                ret = -EIO;
1146                                if (test_bit(BITMAP_WRITE_ERROR,
1147                                             &bitmap->flags))
1148                                        goto err;
1149                        }
1150                }
1151                paddr = kmap_atomic(page);
1152                if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1153                        b = test_bit(bit, paddr);
1154                else
1155                        b = test_bit_le(bit, paddr);
1156                kunmap_atomic(paddr);
1157                if (b) {
1158                        /* if the disk bit is set, set the memory bit */
1159                        int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1160                                      >= start);
1161                        md_bitmap_set_memory_bits(bitmap,
1162                                                  (sector_t)i << bitmap->counts.chunkshift,
1163                                                  needed);
1164                        bit_cnt++;
1165                }
1166                offset = 0;
1167        }
1168
1169        pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1170                 bmname(bitmap), store->file_pages,
1171                 bit_cnt, chunks);
1172
1173        return 0;
1174
1175 err:
1176        pr_warn("%s: bitmap initialisation failed: %d\n",
1177                bmname(bitmap), ret);
1178        return ret;
1179}
1180
1181void md_bitmap_write_all(struct bitmap *bitmap)
1182{
1183        /* We don't actually write all bitmap blocks here,
1184         * just flag them as needing to be written
1185         */
1186        int i;
1187
1188        if (!bitmap || !bitmap->storage.filemap)
1189                return;
1190        if (bitmap->storage.file)
1191                /* Only one copy, so nothing needed */
1192                return;
1193
1194        for (i = 0; i < bitmap->storage.file_pages; i++)
1195                set_page_attr(bitmap, i,
1196                              BITMAP_PAGE_NEEDWRITE);
1197        bitmap->allclean = 0;
1198}
1199
1200static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1201                                 sector_t offset, int inc)
1202{
1203        sector_t chunk = offset >> bitmap->chunkshift;
1204        unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1205        bitmap->bp[page].count += inc;
1206        md_bitmap_checkfree(bitmap, page);
1207}
1208
1209static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1210{
1211        sector_t chunk = offset >> bitmap->chunkshift;
1212        unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1213        struct bitmap_page *bp = &bitmap->bp[page];
1214
1215        if (!bp->pending)
1216                bp->pending = 1;
1217}
1218
1219static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1220                                               sector_t offset, sector_t *blocks,
1221                                               int create);
1222
1223/*
1224 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1225 *                      out to disk
1226 */
1227
1228void md_bitmap_daemon_work(struct mddev *mddev)
1229{
1230        struct bitmap *bitmap;
1231        unsigned long j;
1232        unsigned long nextpage;
1233        sector_t blocks;
1234        struct bitmap_counts *counts;
1235
1236        /* Use a mutex to guard daemon_work against
1237         * bitmap_destroy.
1238         */
1239        mutex_lock(&mddev->bitmap_info.mutex);
1240        bitmap = mddev->bitmap;
1241        if (bitmap == NULL) {
1242                mutex_unlock(&mddev->bitmap_info.mutex);
1243                return;
1244        }
1245        if (time_before(jiffies, bitmap->daemon_lastrun
1246                        + mddev->bitmap_info.daemon_sleep))
1247                goto done;
1248
1249        bitmap->daemon_lastrun = jiffies;
1250        if (bitmap->allclean) {
1251                mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1252                goto done;
1253        }
1254        bitmap->allclean = 1;
1255
1256        if (bitmap->mddev->queue)
1257                blk_add_trace_msg(bitmap->mddev->queue,
1258                                  "md bitmap_daemon_work");
1259
1260        /* Any file-page which is PENDING now needs to be written.
1261         * So set NEEDWRITE now, then after we make any last-minute changes
1262         * we will write it.
1263         */
1264        for (j = 0; j < bitmap->storage.file_pages; j++)
1265                if (test_and_clear_page_attr(bitmap, j,
1266                                             BITMAP_PAGE_PENDING))
1267                        set_page_attr(bitmap, j,
1268                                      BITMAP_PAGE_NEEDWRITE);
1269
1270        if (bitmap->need_sync &&
1271            mddev->bitmap_info.external == 0) {
1272                /* Arrange for superblock update as well as
1273                 * other changes */
1274                bitmap_super_t *sb;
1275                bitmap->need_sync = 0;
1276                if (bitmap->storage.filemap) {
1277                        sb = kmap_atomic(bitmap->storage.sb_page);
1278                        sb->events_cleared =
1279                                cpu_to_le64(bitmap->events_cleared);
1280                        kunmap_atomic(sb);
1281                        set_page_attr(bitmap, 0,
1282                                      BITMAP_PAGE_NEEDWRITE);
1283                }
1284        }
1285        /* Now look at the bitmap counters and if any are '2' or '1',
1286         * decrement and handle accordingly.
1287         */
1288        counts = &bitmap->counts;
1289        spin_lock_irq(&counts->lock);
1290        nextpage = 0;
1291        for (j = 0; j < counts->chunks; j++) {
1292                bitmap_counter_t *bmc;
1293                sector_t  block = (sector_t)j << counts->chunkshift;
1294
1295                if (j == nextpage) {
1296                        nextpage += PAGE_COUNTER_RATIO;
1297                        if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1298                                j |= PAGE_COUNTER_MASK;
1299                                continue;
1300                        }
1301                        counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1302                }
1303
1304                bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1305                if (!bmc) {
1306                        j |= PAGE_COUNTER_MASK;
1307                        continue;
1308                }
1309                if (*bmc == 1 && !bitmap->need_sync) {
1310                        /* We can clear the bit */
1311                        *bmc = 0;
1312                        md_bitmap_count_page(counts, block, -1);
1313                        md_bitmap_file_clear_bit(bitmap, block);
1314                } else if (*bmc && *bmc <= 2) {
1315                        *bmc = 1;
1316                        md_bitmap_set_pending(counts, block);
1317                        bitmap->allclean = 0;
1318                }
1319        }
1320        spin_unlock_irq(&counts->lock);
1321
1322        md_bitmap_wait_writes(bitmap);
1323        /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1324         * DIRTY pages need to be written by bitmap_unplug so it can wait
1325         * for them.
1326         * If we find any DIRTY page we stop there and let bitmap_unplug
1327         * handle all the rest.  This is important in the case where
1328         * the first blocking holds the superblock and it has been updated.
1329         * We mustn't write any other blocks before the superblock.
1330         */
1331        for (j = 0;
1332             j < bitmap->storage.file_pages
1333                     && !test_bit(BITMAP_STALE, &bitmap->flags);
1334             j++) {
1335                if (test_page_attr(bitmap, j,
1336                                   BITMAP_PAGE_DIRTY))
1337                        /* bitmap_unplug will handle the rest */
1338                        break;
1339                if (bitmap->storage.filemap &&
1340                    test_and_clear_page_attr(bitmap, j,
1341                                             BITMAP_PAGE_NEEDWRITE)) {
1342                        write_page(bitmap, bitmap->storage.filemap[j], 0);
1343                }
1344        }
1345
1346 done:
1347        if (bitmap->allclean == 0)
1348                mddev->thread->timeout =
1349                        mddev->bitmap_info.daemon_sleep;
1350        mutex_unlock(&mddev->bitmap_info.mutex);
1351}
1352
1353static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1354                                               sector_t offset, sector_t *blocks,
1355                                               int create)
1356__releases(bitmap->lock)
1357__acquires(bitmap->lock)
1358{
1359        /* If 'create', we might release the lock and reclaim it.
1360         * The lock must have been taken with interrupts enabled.
1361         * If !create, we don't release the lock.
1362         */
1363        sector_t chunk = offset >> bitmap->chunkshift;
1364        unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1365        unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1366        sector_t csize;
1367        int err;
1368
1369        err = md_bitmap_checkpage(bitmap, page, create, 0);
1370
1371        if (bitmap->bp[page].hijacked ||
1372            bitmap->bp[page].map == NULL)
1373                csize = ((sector_t)1) << (bitmap->chunkshift +
1374                                          PAGE_COUNTER_SHIFT);
1375        else
1376                csize = ((sector_t)1) << bitmap->chunkshift;
1377        *blocks = csize - (offset & (csize - 1));
1378
1379        if (err < 0)
1380                return NULL;
1381
1382        /* now locked ... */
1383
1384        if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1385                /* should we use the first or second counter field
1386                 * of the hijacked pointer? */
1387                int hi = (pageoff > PAGE_COUNTER_MASK);
1388                return  &((bitmap_counter_t *)
1389                          &bitmap->bp[page].map)[hi];
1390        } else /* page is allocated */
1391                return (bitmap_counter_t *)
1392                        &(bitmap->bp[page].map[pageoff]);
1393}
1394
1395int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1396{
1397        if (!bitmap)
1398                return 0;
1399
1400        if (behind) {
1401                int bw;
1402                atomic_inc(&bitmap->behind_writes);
1403                bw = atomic_read(&bitmap->behind_writes);
1404                if (bw > bitmap->behind_writes_used)
1405                        bitmap->behind_writes_used = bw;
1406
1407                pr_debug("inc write-behind count %d/%lu\n",
1408                         bw, bitmap->mddev->bitmap_info.max_write_behind);
1409        }
1410
1411        while (sectors) {
1412                sector_t blocks;
1413                bitmap_counter_t *bmc;
1414
1415                spin_lock_irq(&bitmap->counts.lock);
1416                bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1417                if (!bmc) {
1418                        spin_unlock_irq(&bitmap->counts.lock);
1419                        return 0;
1420                }
1421
1422                if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1423                        DEFINE_WAIT(__wait);
1424                        /* note that it is safe to do the prepare_to_wait
1425                         * after the test as long as we do it before dropping
1426                         * the spinlock.
1427                         */
1428                        prepare_to_wait(&bitmap->overflow_wait, &__wait,
1429                                        TASK_UNINTERRUPTIBLE);
1430                        spin_unlock_irq(&bitmap->counts.lock);
1431                        schedule();
1432                        finish_wait(&bitmap->overflow_wait, &__wait);
1433                        continue;
1434                }
1435
1436                switch (*bmc) {
1437                case 0:
1438                        md_bitmap_file_set_bit(bitmap, offset);
1439                        md_bitmap_count_page(&bitmap->counts, offset, 1);
1440                        /* fall through */
1441                case 1:
1442                        *bmc = 2;
1443                }
1444
1445                (*bmc)++;
1446
1447                spin_unlock_irq(&bitmap->counts.lock);
1448
1449                offset += blocks;
1450                if (sectors > blocks)
1451                        sectors -= blocks;
1452                else
1453                        sectors = 0;
1454        }
1455        return 0;
1456}
1457EXPORT_SYMBOL(md_bitmap_startwrite);
1458
1459void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1460                        unsigned long sectors, int success, int behind)
1461{
1462        if (!bitmap)
1463                return;
1464        if (behind) {
1465                if (atomic_dec_and_test(&bitmap->behind_writes))
1466                        wake_up(&bitmap->behind_wait);
1467                pr_debug("dec write-behind count %d/%lu\n",
1468                         atomic_read(&bitmap->behind_writes),
1469                         bitmap->mddev->bitmap_info.max_write_behind);
1470        }
1471
1472        while (sectors) {
1473                sector_t blocks;
1474                unsigned long flags;
1475                bitmap_counter_t *bmc;
1476
1477                spin_lock_irqsave(&bitmap->counts.lock, flags);
1478                bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1479                if (!bmc) {
1480                        spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1481                        return;
1482                }
1483
1484                if (success && !bitmap->mddev->degraded &&
1485                    bitmap->events_cleared < bitmap->mddev->events) {
1486                        bitmap->events_cleared = bitmap->mddev->events;
1487                        bitmap->need_sync = 1;
1488                        sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1489                }
1490
1491                if (!success && !NEEDED(*bmc))
1492                        *bmc |= NEEDED_MASK;
1493
1494                if (COUNTER(*bmc) == COUNTER_MAX)
1495                        wake_up(&bitmap->overflow_wait);
1496
1497                (*bmc)--;
1498                if (*bmc <= 2) {
1499                        md_bitmap_set_pending(&bitmap->counts, offset);
1500                        bitmap->allclean = 0;
1501                }
1502                spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1503                offset += blocks;
1504                if (sectors > blocks)
1505                        sectors -= blocks;
1506                else
1507                        sectors = 0;
1508        }
1509}
1510EXPORT_SYMBOL(md_bitmap_endwrite);
1511
1512static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1513                               int degraded)
1514{
1515        bitmap_counter_t *bmc;
1516        int rv;
1517        if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1518                *blocks = 1024;
1519                return 1; /* always resync if no bitmap */
1520        }
1521        spin_lock_irq(&bitmap->counts.lock);
1522        bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1523        rv = 0;
1524        if (bmc) {
1525                /* locked */
1526                if (RESYNC(*bmc))
1527                        rv = 1;
1528                else if (NEEDED(*bmc)) {
1529                        rv = 1;
1530                        if (!degraded) { /* don't set/clear bits if degraded */
1531                                *bmc |= RESYNC_MASK;
1532                                *bmc &= ~NEEDED_MASK;
1533                        }
1534                }
1535        }
1536        spin_unlock_irq(&bitmap->counts.lock);
1537        return rv;
1538}
1539
1540int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1541                         int degraded)
1542{
1543        /* bitmap_start_sync must always report on multiples of whole
1544         * pages, otherwise resync (which is very PAGE_SIZE based) will
1545         * get confused.
1546         * So call __bitmap_start_sync repeatedly (if needed) until
1547         * At least PAGE_SIZE>>9 blocks are covered.
1548         * Return the 'or' of the result.
1549         */
1550        int rv = 0;
1551        sector_t blocks1;
1552
1553        *blocks = 0;
1554        while (*blocks < (PAGE_SIZE>>9)) {
1555                rv |= __bitmap_start_sync(bitmap, offset,
1556                                          &blocks1, degraded);
1557                offset += blocks1;
1558                *blocks += blocks1;
1559        }
1560        return rv;
1561}
1562EXPORT_SYMBOL(md_bitmap_start_sync);
1563
1564void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1565{
1566        bitmap_counter_t *bmc;
1567        unsigned long flags;
1568
1569        if (bitmap == NULL) {
1570                *blocks = 1024;
1571                return;
1572        }
1573        spin_lock_irqsave(&bitmap->counts.lock, flags);
1574        bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1575        if (bmc == NULL)
1576                goto unlock;
1577        /* locked */
1578        if (RESYNC(*bmc)) {
1579                *bmc &= ~RESYNC_MASK;
1580
1581                if (!NEEDED(*bmc) && aborted)
1582                        *bmc |= NEEDED_MASK;
1583                else {
1584                        if (*bmc <= 2) {
1585                                md_bitmap_set_pending(&bitmap->counts, offset);
1586                                bitmap->allclean = 0;
1587                        }
1588                }
1589        }
1590 unlock:
1591        spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1592}
1593EXPORT_SYMBOL(md_bitmap_end_sync);
1594
1595void md_bitmap_close_sync(struct bitmap *bitmap)
1596{
1597        /* Sync has finished, and any bitmap chunks that weren't synced
1598         * properly have been aborted.  It remains to us to clear the
1599         * RESYNC bit wherever it is still on
1600         */
1601        sector_t sector = 0;
1602        sector_t blocks;
1603        if (!bitmap)
1604                return;
1605        while (sector < bitmap->mddev->resync_max_sectors) {
1606                md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1607                sector += blocks;
1608        }
1609}
1610EXPORT_SYMBOL(md_bitmap_close_sync);
1611
1612void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1613{
1614        sector_t s = 0;
1615        sector_t blocks;
1616
1617        if (!bitmap)
1618                return;
1619        if (sector == 0) {
1620                bitmap->last_end_sync = jiffies;
1621                return;
1622        }
1623        if (!force && time_before(jiffies, (bitmap->last_end_sync
1624                                  + bitmap->mddev->bitmap_info.daemon_sleep)))
1625                return;
1626        wait_event(bitmap->mddev->recovery_wait,
1627                   atomic_read(&bitmap->mddev->recovery_active) == 0);
1628
1629        bitmap->mddev->curr_resync_completed = sector;
1630        set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1631        sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1632        s = 0;
1633        while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1634                md_bitmap_end_sync(bitmap, s, &blocks, 0);
1635                s += blocks;
1636        }
1637        bitmap->last_end_sync = jiffies;
1638        sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1639}
1640EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1641
1642void md_bitmap_sync_with_cluster(struct mddev *mddev,
1643                              sector_t old_lo, sector_t old_hi,
1644                              sector_t new_lo, sector_t new_hi)
1645{
1646        struct bitmap *bitmap = mddev->bitmap;
1647        sector_t sector, blocks = 0;
1648
1649        for (sector = old_lo; sector < new_lo; ) {
1650                md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1651                sector += blocks;
1652        }
1653        WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1654
1655        for (sector = old_hi; sector < new_hi; ) {
1656                md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1657                sector += blocks;
1658        }
1659        WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1660}
1661EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1662
1663static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1664{
1665        /* For each chunk covered by any of these sectors, set the
1666         * counter to 2 and possibly set resync_needed.  They should all
1667         * be 0 at this point
1668         */
1669
1670        sector_t secs;
1671        bitmap_counter_t *bmc;
1672        spin_lock_irq(&bitmap->counts.lock);
1673        bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1674        if (!bmc) {
1675                spin_unlock_irq(&bitmap->counts.lock);
1676                return;
1677        }
1678        if (!*bmc) {
1679                *bmc = 2;
1680                md_bitmap_count_page(&bitmap->counts, offset, 1);
1681                md_bitmap_set_pending(&bitmap->counts, offset);
1682                bitmap->allclean = 0;
1683        }
1684        if (needed)
1685                *bmc |= NEEDED_MASK;
1686        spin_unlock_irq(&bitmap->counts.lock);
1687}
1688
1689/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1690void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1691{
1692        unsigned long chunk;
1693
1694        for (chunk = s; chunk <= e; chunk++) {
1695                sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1696                md_bitmap_set_memory_bits(bitmap, sec, 1);
1697                md_bitmap_file_set_bit(bitmap, sec);
1698                if (sec < bitmap->mddev->recovery_cp)
1699                        /* We are asserting that the array is dirty,
1700                         * so move the recovery_cp address back so
1701                         * that it is obvious that it is dirty
1702                         */
1703                        bitmap->mddev->recovery_cp = sec;
1704        }
1705}
1706
1707/*
1708 * flush out any pending updates
1709 */
1710void md_bitmap_flush(struct mddev *mddev)
1711{
1712        struct bitmap *bitmap = mddev->bitmap;
1713        long sleep;
1714
1715        if (!bitmap) /* there was no bitmap */
1716                return;
1717
1718        /* run the daemon_work three time to ensure everything is flushed
1719         * that can be
1720         */
1721        sleep = mddev->bitmap_info.daemon_sleep * 2;
1722        bitmap->daemon_lastrun -= sleep;
1723        md_bitmap_daemon_work(mddev);
1724        bitmap->daemon_lastrun -= sleep;
1725        md_bitmap_daemon_work(mddev);
1726        bitmap->daemon_lastrun -= sleep;
1727        md_bitmap_daemon_work(mddev);
1728        md_bitmap_update_sb(bitmap);
1729}
1730
1731/*
1732 * free memory that was allocated
1733 */
1734void md_bitmap_free(struct bitmap *bitmap)
1735{
1736        unsigned long k, pages;
1737        struct bitmap_page *bp;
1738
1739        if (!bitmap) /* there was no bitmap */
1740                return;
1741
1742        if (bitmap->sysfs_can_clear)
1743                sysfs_put(bitmap->sysfs_can_clear);
1744
1745        if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1746                bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1747                md_cluster_stop(bitmap->mddev);
1748
1749        /* Shouldn't be needed - but just in case.... */
1750        wait_event(bitmap->write_wait,
1751                   atomic_read(&bitmap->pending_writes) == 0);
1752
1753        /* release the bitmap file  */
1754        md_bitmap_file_unmap(&bitmap->storage);
1755
1756        bp = bitmap->counts.bp;
1757        pages = bitmap->counts.pages;
1758
1759        /* free all allocated memory */
1760
1761        if (bp) /* deallocate the page memory */
1762                for (k = 0; k < pages; k++)
1763                        if (bp[k].map && !bp[k].hijacked)
1764                                kfree(bp[k].map);
1765        kfree(bp);
1766        kfree(bitmap);
1767}
1768EXPORT_SYMBOL(md_bitmap_free);
1769
1770void md_bitmap_wait_behind_writes(struct mddev *mddev)
1771{
1772        struct bitmap *bitmap = mddev->bitmap;
1773
1774        /* wait for behind writes to complete */
1775        if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1776                pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1777                         mdname(mddev));
1778                /* need to kick something here to make sure I/O goes? */
1779                wait_event(bitmap->behind_wait,
1780                           atomic_read(&bitmap->behind_writes) == 0);
1781        }
1782}
1783
1784void md_bitmap_destroy(struct mddev *mddev)
1785{
1786        struct bitmap *bitmap = mddev->bitmap;
1787
1788        if (!bitmap) /* there was no bitmap */
1789                return;
1790
1791        md_bitmap_wait_behind_writes(mddev);
1792        if (!mddev->serialize_policy)
1793                mddev_destroy_serial_pool(mddev, NULL, true);
1794
1795        mutex_lock(&mddev->bitmap_info.mutex);
1796        spin_lock(&mddev->lock);
1797        mddev->bitmap = NULL; /* disconnect from the md device */
1798        spin_unlock(&mddev->lock);
1799        mutex_unlock(&mddev->bitmap_info.mutex);
1800        if (mddev->thread)
1801                mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1802
1803        md_bitmap_free(bitmap);
1804}
1805
1806/*
1807 * initialize the bitmap structure
1808 * if this returns an error, bitmap_destroy must be called to do clean up
1809 * once mddev->bitmap is set
1810 */
1811struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1812{
1813        struct bitmap *bitmap;
1814        sector_t blocks = mddev->resync_max_sectors;
1815        struct file *file = mddev->bitmap_info.file;
1816        int err;
1817        struct kernfs_node *bm = NULL;
1818
1819        BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1820
1821        BUG_ON(file && mddev->bitmap_info.offset);
1822
1823        if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1824                pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1825                          mdname(mddev));
1826                return ERR_PTR(-EBUSY);
1827        }
1828
1829        bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1830        if (!bitmap)
1831                return ERR_PTR(-ENOMEM);
1832
1833        spin_lock_init(&bitmap->counts.lock);
1834        atomic_set(&bitmap->pending_writes, 0);
1835        init_waitqueue_head(&bitmap->write_wait);
1836        init_waitqueue_head(&bitmap->overflow_wait);
1837        init_waitqueue_head(&bitmap->behind_wait);
1838
1839        bitmap->mddev = mddev;
1840        bitmap->cluster_slot = slot;
1841
1842        if (mddev->kobj.sd)
1843                bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1844        if (bm) {
1845                bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1846                sysfs_put(bm);
1847        } else
1848                bitmap->sysfs_can_clear = NULL;
1849
1850        bitmap->storage.file = file;
1851        if (file) {
1852                get_file(file);
1853                /* As future accesses to this file will use bmap,
1854                 * and bypass the page cache, we must sync the file
1855                 * first.
1856                 */
1857                vfs_fsync(file, 1);
1858        }
1859        /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1860        if (!mddev->bitmap_info.external) {
1861                /*
1862                 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1863                 * instructing us to create a new on-disk bitmap instance.
1864                 */
1865                if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1866                        err = md_bitmap_new_disk_sb(bitmap);
1867                else
1868                        err = md_bitmap_read_sb(bitmap);
1869        } else {
1870                err = 0;
1871                if (mddev->bitmap_info.chunksize == 0 ||
1872                    mddev->bitmap_info.daemon_sleep == 0)
1873                        /* chunksize and time_base need to be
1874                         * set first. */
1875                        err = -EINVAL;
1876        }
1877        if (err)
1878                goto error;
1879
1880        bitmap->daemon_lastrun = jiffies;
1881        err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1882        if (err)
1883                goto error;
1884
1885        pr_debug("created bitmap (%lu pages) for device %s\n",
1886                 bitmap->counts.pages, bmname(bitmap));
1887
1888        err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1889        if (err)
1890                goto error;
1891
1892        return bitmap;
1893 error:
1894        md_bitmap_free(bitmap);
1895        return ERR_PTR(err);
1896}
1897
1898int md_bitmap_load(struct mddev *mddev)
1899{
1900        int err = 0;
1901        sector_t start = 0;
1902        sector_t sector = 0;
1903        struct bitmap *bitmap = mddev->bitmap;
1904        struct md_rdev *rdev;
1905
1906        if (!bitmap)
1907                goto out;
1908
1909        rdev_for_each(rdev, mddev)
1910                mddev_create_serial_pool(mddev, rdev, true);
1911
1912        if (mddev_is_clustered(mddev))
1913                md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1914
1915        /* Clear out old bitmap info first:  Either there is none, or we
1916         * are resuming after someone else has possibly changed things,
1917         * so we should forget old cached info.
1918         * All chunks should be clean, but some might need_sync.
1919         */
1920        while (sector < mddev->resync_max_sectors) {
1921                sector_t blocks;
1922                md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1923                sector += blocks;
1924        }
1925        md_bitmap_close_sync(bitmap);
1926
1927        if (mddev->degraded == 0
1928            || bitmap->events_cleared == mddev->events)
1929                /* no need to keep dirty bits to optimise a
1930                 * re-add of a missing device */
1931                start = mddev->recovery_cp;
1932
1933        mutex_lock(&mddev->bitmap_info.mutex);
1934        err = md_bitmap_init_from_disk(bitmap, start);
1935        mutex_unlock(&mddev->bitmap_info.mutex);
1936
1937        if (err)
1938                goto out;
1939        clear_bit(BITMAP_STALE, &bitmap->flags);
1940
1941        /* Kick recovery in case any bits were set */
1942        set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1943
1944        mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1945        md_wakeup_thread(mddev->thread);
1946
1947        md_bitmap_update_sb(bitmap);
1948
1949        if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1950                err = -EIO;
1951out:
1952        return err;
1953}
1954EXPORT_SYMBOL_GPL(md_bitmap_load);
1955
1956/* caller need to free returned bitmap with md_bitmap_free() */
1957struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1958{
1959        int rv = 0;
1960        struct bitmap *bitmap;
1961
1962        bitmap = md_bitmap_create(mddev, slot);
1963        if (IS_ERR(bitmap)) {
1964                rv = PTR_ERR(bitmap);
1965                return ERR_PTR(rv);
1966        }
1967
1968        rv = md_bitmap_init_from_disk(bitmap, 0);
1969        if (rv) {
1970                md_bitmap_free(bitmap);
1971                return ERR_PTR(rv);
1972        }
1973
1974        return bitmap;
1975}
1976EXPORT_SYMBOL(get_bitmap_from_slot);
1977
1978/* Loads the bitmap associated with slot and copies the resync information
1979 * to our bitmap
1980 */
1981int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
1982                sector_t *low, sector_t *high, bool clear_bits)
1983{
1984        int rv = 0, i, j;
1985        sector_t block, lo = 0, hi = 0;
1986        struct bitmap_counts *counts;
1987        struct bitmap *bitmap;
1988
1989        bitmap = get_bitmap_from_slot(mddev, slot);
1990        if (IS_ERR(bitmap)) {
1991                pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
1992                return -1;
1993        }
1994
1995        counts = &bitmap->counts;
1996        for (j = 0; j < counts->chunks; j++) {
1997                block = (sector_t)j << counts->chunkshift;
1998                if (md_bitmap_file_test_bit(bitmap, block)) {
1999                        if (!lo)
2000                                lo = block;
2001                        hi = block;
2002                        md_bitmap_file_clear_bit(bitmap, block);
2003                        md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2004                        md_bitmap_file_set_bit(mddev->bitmap, block);
2005                }
2006        }
2007
2008        if (clear_bits) {
2009                md_bitmap_update_sb(bitmap);
2010                /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2011                 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2012                for (i = 0; i < bitmap->storage.file_pages; i++)
2013                        if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2014                                set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2015                md_bitmap_unplug(bitmap);
2016        }
2017        md_bitmap_unplug(mddev->bitmap);
2018        *low = lo;
2019        *high = hi;
2020        md_bitmap_free(bitmap);
2021
2022        return rv;
2023}
2024EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2025
2026
2027void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2028{
2029        unsigned long chunk_kb;
2030        struct bitmap_counts *counts;
2031
2032        if (!bitmap)
2033                return;
2034
2035        counts = &bitmap->counts;
2036
2037        chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2038        seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2039                   "%lu%s chunk",
2040                   counts->pages - counts->missing_pages,
2041                   counts->pages,
2042                   (counts->pages - counts->missing_pages)
2043                   << (PAGE_SHIFT - 10),
2044                   chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2045                   chunk_kb ? "KB" : "B");
2046        if (bitmap->storage.file) {
2047                seq_printf(seq, ", file: ");
2048                seq_file_path(seq, bitmap->storage.file, " \t\n");
2049        }
2050
2051        seq_printf(seq, "\n");
2052}
2053
2054int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2055                  int chunksize, int init)
2056{
2057        /* If chunk_size is 0, choose an appropriate chunk size.
2058         * Then possibly allocate new storage space.
2059         * Then quiesce, copy bits, replace bitmap, and re-start
2060         *
2061         * This function is called both to set up the initial bitmap
2062         * and to resize the bitmap while the array is active.
2063         * If this happens as a result of the array being resized,
2064         * chunksize will be zero, and we need to choose a suitable
2065         * chunksize, otherwise we use what we are given.
2066         */
2067        struct bitmap_storage store;
2068        struct bitmap_counts old_counts;
2069        unsigned long chunks;
2070        sector_t block;
2071        sector_t old_blocks, new_blocks;
2072        int chunkshift;
2073        int ret = 0;
2074        long pages;
2075        struct bitmap_page *new_bp;
2076
2077        if (bitmap->storage.file && !init) {
2078                pr_info("md: cannot resize file-based bitmap\n");
2079                return -EINVAL;
2080        }
2081
2082        if (chunksize == 0) {
2083                /* If there is enough space, leave the chunk size unchanged,
2084                 * else increase by factor of two until there is enough space.
2085                 */
2086                long bytes;
2087                long space = bitmap->mddev->bitmap_info.space;
2088
2089                if (space == 0) {
2090                        /* We don't know how much space there is, so limit
2091                         * to current size - in sectors.
2092                         */
2093                        bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2094                        if (!bitmap->mddev->bitmap_info.external)
2095                                bytes += sizeof(bitmap_super_t);
2096                        space = DIV_ROUND_UP(bytes, 512);
2097                        bitmap->mddev->bitmap_info.space = space;
2098                }
2099                chunkshift = bitmap->counts.chunkshift;
2100                chunkshift--;
2101                do {
2102                        /* 'chunkshift' is shift from block size to chunk size */
2103                        chunkshift++;
2104                        chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2105                        bytes = DIV_ROUND_UP(chunks, 8);
2106                        if (!bitmap->mddev->bitmap_info.external)
2107                                bytes += sizeof(bitmap_super_t);
2108                } while (bytes > (space << 9));
2109        } else
2110                chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2111
2112        chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2113        memset(&store, 0, sizeof(store));
2114        if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2115                ret = md_bitmap_storage_alloc(&store, chunks,
2116                                              !bitmap->mddev->bitmap_info.external,
2117                                              mddev_is_clustered(bitmap->mddev)
2118                                              ? bitmap->cluster_slot : 0);
2119        if (ret) {
2120                md_bitmap_file_unmap(&store);
2121                goto err;
2122        }
2123
2124        pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2125
2126        new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2127        ret = -ENOMEM;
2128        if (!new_bp) {
2129                md_bitmap_file_unmap(&store);
2130                goto err;
2131        }
2132
2133        if (!init)
2134                bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2135
2136        store.file = bitmap->storage.file;
2137        bitmap->storage.file = NULL;
2138
2139        if (store.sb_page && bitmap->storage.sb_page)
2140                memcpy(page_address(store.sb_page),
2141                       page_address(bitmap->storage.sb_page),
2142                       sizeof(bitmap_super_t));
2143        spin_lock_irq(&bitmap->counts.lock);
2144        md_bitmap_file_unmap(&bitmap->storage);
2145        bitmap->storage = store;
2146
2147        old_counts = bitmap->counts;
2148        bitmap->counts.bp = new_bp;
2149        bitmap->counts.pages = pages;
2150        bitmap->counts.missing_pages = pages;
2151        bitmap->counts.chunkshift = chunkshift;
2152        bitmap->counts.chunks = chunks;
2153        bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2154                                                     BITMAP_BLOCK_SHIFT);
2155
2156        blocks = min(old_counts.chunks << old_counts.chunkshift,
2157                     chunks << chunkshift);
2158
2159        /* For cluster raid, need to pre-allocate bitmap */
2160        if (mddev_is_clustered(bitmap->mddev)) {
2161                unsigned long page;
2162                for (page = 0; page < pages; page++) {
2163                        ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2164                        if (ret) {
2165                                unsigned long k;
2166
2167                                /* deallocate the page memory */
2168                                for (k = 0; k < page; k++) {
2169                                        kfree(new_bp[k].map);
2170                                }
2171                                kfree(new_bp);
2172
2173                                /* restore some fields from old_counts */
2174                                bitmap->counts.bp = old_counts.bp;
2175                                bitmap->counts.pages = old_counts.pages;
2176                                bitmap->counts.missing_pages = old_counts.pages;
2177                                bitmap->counts.chunkshift = old_counts.chunkshift;
2178                                bitmap->counts.chunks = old_counts.chunks;
2179                                bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2180                                                                             BITMAP_BLOCK_SHIFT);
2181                                blocks = old_counts.chunks << old_counts.chunkshift;
2182                                pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2183                                break;
2184                        } else
2185                                bitmap->counts.bp[page].count += 1;
2186                }
2187        }
2188
2189        for (block = 0; block < blocks; ) {
2190                bitmap_counter_t *bmc_old, *bmc_new;
2191                int set;
2192
2193                bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2194                set = bmc_old && NEEDED(*bmc_old);
2195
2196                if (set) {
2197                        bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2198                        if (*bmc_new == 0) {
2199                                /* need to set on-disk bits too. */
2200                                sector_t end = block + new_blocks;
2201                                sector_t start = block >> chunkshift;
2202                                start <<= chunkshift;
2203                                while (start < end) {
2204                                        md_bitmap_file_set_bit(bitmap, block);
2205                                        start += 1 << chunkshift;
2206                                }
2207                                *bmc_new = 2;
2208                                md_bitmap_count_page(&bitmap->counts, block, 1);
2209                                md_bitmap_set_pending(&bitmap->counts, block);
2210                        }
2211                        *bmc_new |= NEEDED_MASK;
2212                        if (new_blocks < old_blocks)
2213                                old_blocks = new_blocks;
2214                }
2215                block += old_blocks;
2216        }
2217
2218        if (bitmap->counts.bp != old_counts.bp) {
2219                unsigned long k;
2220                for (k = 0; k < old_counts.pages; k++)
2221                        if (!old_counts.bp[k].hijacked)
2222                                kfree(old_counts.bp[k].map);
2223                kfree(old_counts.bp);
2224        }
2225
2226        if (!init) {
2227                int i;
2228                while (block < (chunks << chunkshift)) {
2229                        bitmap_counter_t *bmc;
2230                        bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2231                        if (bmc) {
2232                                /* new space.  It needs to be resynced, so
2233                                 * we set NEEDED_MASK.
2234                                 */
2235                                if (*bmc == 0) {
2236                                        *bmc = NEEDED_MASK | 2;
2237                                        md_bitmap_count_page(&bitmap->counts, block, 1);
2238                                        md_bitmap_set_pending(&bitmap->counts, block);
2239                                }
2240                        }
2241                        block += new_blocks;
2242                }
2243                for (i = 0; i < bitmap->storage.file_pages; i++)
2244                        set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2245        }
2246        spin_unlock_irq(&bitmap->counts.lock);
2247
2248        if (!init) {
2249                md_bitmap_unplug(bitmap);
2250                bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2251        }
2252        ret = 0;
2253err:
2254        return ret;
2255}
2256EXPORT_SYMBOL_GPL(md_bitmap_resize);
2257
2258static ssize_t
2259location_show(struct mddev *mddev, char *page)
2260{
2261        ssize_t len;
2262        if (mddev->bitmap_info.file)
2263                len = sprintf(page, "file");
2264        else if (mddev->bitmap_info.offset)
2265                len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2266        else
2267                len = sprintf(page, "none");
2268        len += sprintf(page+len, "\n");
2269        return len;
2270}
2271
2272static ssize_t
2273location_store(struct mddev *mddev, const char *buf, size_t len)
2274{
2275        int rv;
2276
2277        rv = mddev_lock(mddev);
2278        if (rv)
2279                return rv;
2280        if (mddev->pers) {
2281                if (!mddev->pers->quiesce) {
2282                        rv = -EBUSY;
2283                        goto out;
2284                }
2285                if (mddev->recovery || mddev->sync_thread) {
2286                        rv = -EBUSY;
2287                        goto out;
2288                }
2289        }
2290
2291        if (mddev->bitmap || mddev->bitmap_info.file ||
2292            mddev->bitmap_info.offset) {
2293                /* bitmap already configured.  Only option is to clear it */
2294                if (strncmp(buf, "none", 4) != 0) {
2295                        rv = -EBUSY;
2296                        goto out;
2297                }
2298                if (mddev->pers) {
2299                        mddev_suspend(mddev);
2300                        md_bitmap_destroy(mddev);
2301                        mddev_resume(mddev);
2302                }
2303                mddev->bitmap_info.offset = 0;
2304                if (mddev->bitmap_info.file) {
2305                        struct file *f = mddev->bitmap_info.file;
2306                        mddev->bitmap_info.file = NULL;
2307                        fput(f);
2308                }
2309        } else {
2310                /* No bitmap, OK to set a location */
2311                long long offset;
2312                if (strncmp(buf, "none", 4) == 0)
2313                        /* nothing to be done */;
2314                else if (strncmp(buf, "file:", 5) == 0) {
2315                        /* Not supported yet */
2316                        rv = -EINVAL;
2317                        goto out;
2318                } else {
2319                        if (buf[0] == '+')
2320                                rv = kstrtoll(buf+1, 10, &offset);
2321                        else
2322                                rv = kstrtoll(buf, 10, &offset);
2323                        if (rv)
2324                                goto out;
2325                        if (offset == 0) {
2326                                rv = -EINVAL;
2327                                goto out;
2328                        }
2329                        if (mddev->bitmap_info.external == 0 &&
2330                            mddev->major_version == 0 &&
2331                            offset != mddev->bitmap_info.default_offset) {
2332                                rv = -EINVAL;
2333                                goto out;
2334                        }
2335                        mddev->bitmap_info.offset = offset;
2336                        if (mddev->pers) {
2337                                struct bitmap *bitmap;
2338                                bitmap = md_bitmap_create(mddev, -1);
2339                                mddev_suspend(mddev);
2340                                if (IS_ERR(bitmap))
2341                                        rv = PTR_ERR(bitmap);
2342                                else {
2343                                        mddev->bitmap = bitmap;
2344                                        rv = md_bitmap_load(mddev);
2345                                        if (rv)
2346                                                mddev->bitmap_info.offset = 0;
2347                                }
2348                                if (rv) {
2349                                        md_bitmap_destroy(mddev);
2350                                        mddev_resume(mddev);
2351                                        goto out;
2352                                }
2353                                mddev_resume(mddev);
2354                        }
2355                }
2356        }
2357        if (!mddev->external) {
2358                /* Ensure new bitmap info is stored in
2359                 * metadata promptly.
2360                 */
2361                set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2362                md_wakeup_thread(mddev->thread);
2363        }
2364        rv = 0;
2365out:
2366        mddev_unlock(mddev);
2367        if (rv)
2368                return rv;
2369        return len;
2370}
2371
2372static struct md_sysfs_entry bitmap_location =
2373__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2374
2375/* 'bitmap/space' is the space available at 'location' for the
2376 * bitmap.  This allows the kernel to know when it is safe to
2377 * resize the bitmap to match a resized array.
2378 */
2379static ssize_t
2380space_show(struct mddev *mddev, char *page)
2381{
2382        return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2383}
2384
2385static ssize_t
2386space_store(struct mddev *mddev, const char *buf, size_t len)
2387{
2388        unsigned long sectors;
2389        int rv;
2390
2391        rv = kstrtoul(buf, 10, &sectors);
2392        if (rv)
2393                return rv;
2394
2395        if (sectors == 0)
2396                return -EINVAL;
2397
2398        if (mddev->bitmap &&
2399            sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2400                return -EFBIG; /* Bitmap is too big for this small space */
2401
2402        /* could make sure it isn't too big, but that isn't really
2403         * needed - user-space should be careful.
2404         */
2405        mddev->bitmap_info.space = sectors;
2406        return len;
2407}
2408
2409static struct md_sysfs_entry bitmap_space =
2410__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2411
2412static ssize_t
2413timeout_show(struct mddev *mddev, char *page)
2414{
2415        ssize_t len;
2416        unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2417        unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2418
2419        len = sprintf(page, "%lu", secs);
2420        if (jifs)
2421                len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2422        len += sprintf(page+len, "\n");
2423        return len;
2424}
2425
2426static ssize_t
2427timeout_store(struct mddev *mddev, const char *buf, size_t len)
2428{
2429        /* timeout can be set at any time */
2430        unsigned long timeout;
2431        int rv = strict_strtoul_scaled(buf, &timeout, 4);
2432        if (rv)
2433                return rv;
2434
2435        /* just to make sure we don't overflow... */
2436        if (timeout >= LONG_MAX / HZ)
2437                return -EINVAL;
2438
2439        timeout = timeout * HZ / 10000;
2440
2441        if (timeout >= MAX_SCHEDULE_TIMEOUT)
2442                timeout = MAX_SCHEDULE_TIMEOUT-1;
2443        if (timeout < 1)
2444                timeout = 1;
2445        mddev->bitmap_info.daemon_sleep = timeout;
2446        if (mddev->thread) {
2447                /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2448                 * the bitmap is all clean and we don't need to
2449                 * adjust the timeout right now
2450                 */
2451                if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2452                        mddev->thread->timeout = timeout;
2453                        md_wakeup_thread(mddev->thread);
2454                }
2455        }
2456        return len;
2457}
2458
2459static struct md_sysfs_entry bitmap_timeout =
2460__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2461
2462static ssize_t
2463backlog_show(struct mddev *mddev, char *page)
2464{
2465        return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2466}
2467
2468static ssize_t
2469backlog_store(struct mddev *mddev, const char *buf, size_t len)
2470{
2471        unsigned long backlog;
2472        unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2473        int rv = kstrtoul(buf, 10, &backlog);
2474        if (rv)
2475                return rv;
2476        if (backlog > COUNTER_MAX)
2477                return -EINVAL;
2478        mddev->bitmap_info.max_write_behind = backlog;
2479        if (!backlog && mddev->serial_info_pool) {
2480                /* serial_info_pool is not needed if backlog is zero */
2481                if (!mddev->serialize_policy)
2482                        mddev_destroy_serial_pool(mddev, NULL, false);
2483        } else if (backlog && !mddev->serial_info_pool) {
2484                /* serial_info_pool is needed since backlog is not zero */
2485                struct md_rdev *rdev;
2486
2487                rdev_for_each(rdev, mddev)
2488                        mddev_create_serial_pool(mddev, rdev, false);
2489        }
2490        if (old_mwb != backlog)
2491                md_bitmap_update_sb(mddev->bitmap);
2492        return len;
2493}
2494
2495static struct md_sysfs_entry bitmap_backlog =
2496__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2497
2498static ssize_t
2499chunksize_show(struct mddev *mddev, char *page)
2500{
2501        return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2502}
2503
2504static ssize_t
2505chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2506{
2507        /* Can only be changed when no bitmap is active */
2508        int rv;
2509        unsigned long csize;
2510        if (mddev->bitmap)
2511                return -EBUSY;
2512        rv = kstrtoul(buf, 10, &csize);
2513        if (rv)
2514                return rv;
2515        if (csize < 512 ||
2516            !is_power_of_2(csize))
2517                return -EINVAL;
2518        mddev->bitmap_info.chunksize = csize;
2519        return len;
2520}
2521
2522static struct md_sysfs_entry bitmap_chunksize =
2523__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2524
2525static ssize_t metadata_show(struct mddev *mddev, char *page)
2526{
2527        if (mddev_is_clustered(mddev))
2528                return sprintf(page, "clustered\n");
2529        return sprintf(page, "%s\n", (mddev->bitmap_info.external
2530                                      ? "external" : "internal"));
2531}
2532
2533static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2534{
2535        if (mddev->bitmap ||
2536            mddev->bitmap_info.file ||
2537            mddev->bitmap_info.offset)
2538                return -EBUSY;
2539        if (strncmp(buf, "external", 8) == 0)
2540                mddev->bitmap_info.external = 1;
2541        else if ((strncmp(buf, "internal", 8) == 0) ||
2542                        (strncmp(buf, "clustered", 9) == 0))
2543                mddev->bitmap_info.external = 0;
2544        else
2545                return -EINVAL;
2546        return len;
2547}
2548
2549static struct md_sysfs_entry bitmap_metadata =
2550__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2551
2552static ssize_t can_clear_show(struct mddev *mddev, char *page)
2553{
2554        int len;
2555        spin_lock(&mddev->lock);
2556        if (mddev->bitmap)
2557                len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2558                                             "false" : "true"));
2559        else
2560                len = sprintf(page, "\n");
2561        spin_unlock(&mddev->lock);
2562        return len;
2563}
2564
2565static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2566{
2567        if (mddev->bitmap == NULL)
2568                return -ENOENT;
2569        if (strncmp(buf, "false", 5) == 0)
2570                mddev->bitmap->need_sync = 1;
2571        else if (strncmp(buf, "true", 4) == 0) {
2572                if (mddev->degraded)
2573                        return -EBUSY;
2574                mddev->bitmap->need_sync = 0;
2575        } else
2576                return -EINVAL;
2577        return len;
2578}
2579
2580static struct md_sysfs_entry bitmap_can_clear =
2581__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2582
2583static ssize_t
2584behind_writes_used_show(struct mddev *mddev, char *page)
2585{
2586        ssize_t ret;
2587        spin_lock(&mddev->lock);
2588        if (mddev->bitmap == NULL)
2589                ret = sprintf(page, "0\n");
2590        else
2591                ret = sprintf(page, "%lu\n",
2592                              mddev->bitmap->behind_writes_used);
2593        spin_unlock(&mddev->lock);
2594        return ret;
2595}
2596
2597static ssize_t
2598behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2599{
2600        if (mddev->bitmap)
2601                mddev->bitmap->behind_writes_used = 0;
2602        return len;
2603}
2604
2605static struct md_sysfs_entry max_backlog_used =
2606__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2607       behind_writes_used_show, behind_writes_used_reset);
2608
2609static struct attribute *md_bitmap_attrs[] = {
2610        &bitmap_location.attr,
2611        &bitmap_space.attr,
2612        &bitmap_timeout.attr,
2613        &bitmap_backlog.attr,
2614        &bitmap_chunksize.attr,
2615        &bitmap_metadata.attr,
2616        &bitmap_can_clear.attr,
2617        &max_backlog_used.attr,
2618        NULL
2619};
2620struct attribute_group md_bitmap_group = {
2621        .name = "bitmap",
2622        .attrs = md_bitmap_attrs,
2623};
2624