drivers/md/md-bitmap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * bitmap.c two-level bitmap (C) Peter T. Breuer ([email protected]) 2003
   4  *
   5  * bitmap_create  - sets up the bitmap structure
   6  * bitmap_destroy - destroys the bitmap structure
   7  *
   8  * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   9  * - added disk storage for bitmap
  10  * - changes to allow various bitmap chunk sizes
  11  */
  12
  13 /*
  14  * Still to do:
  15  *
  16  * flush after percent set rather than just time based. (maybe both).
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/errno.h>
  22 #include <linux/slab.h>
  23 #include <linux/init.h>
  24 #include <linux/timer.h>
  25 #include <linux/sched.h>
  26 #include <linux/list.h>
  27 #include <linux/file.h>
  28 #include <linux/mount.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/seq_file.h>
  31 #include <trace/events/block.h>
  32 #include "md.h"
  33 #include "md-bitmap.h"
  34
  35 static inline char *bmname(struct bitmap *bitmap)
  36 {
  37         return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  38 }
  39
  40 /*
  41  * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
  42  *
  43  * 1) check to see if this page is allocated, if it's not then try to alloc
  44  * 2) if the alloc fails, set the page's hijacked flag so we'll use the
  45  *    page pointer directly as a counter
  46  *
  47  * if we find our page, we increment the page's refcount so that it stays
  48  * allocated while we're using it
  49  */
  50 static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
  51                                unsigned long page, int create, int no_hijack)
  52 __releases(bitmap->lock)
  53 __acquires(bitmap->lock)
  54 {
  55         unsigned char *mappage;
  56
  57         if (page >= bitmap->pages) {
  58                 /* This can happen if bitmap_start_sync goes beyond
  59                  * End-of-device while looking for a whole page.
  60                  * It is harmless.
  61                  */
  62                 return -EINVAL;
  63         }
  64
  65         if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  66                 return 0;
  67
  68         if (bitmap->bp[page].map) /* page is already allocated, just return */
  69                 return 0;
  70
  71         if (!create)
  72                 return -ENOENT;
  73
  74         /* this page has not been allocated yet */
  75
  76         spin_unlock_irq(&bitmap->lock);
  77         /* It is possible that this is being called inside a
  78          * prepare_to_wait/finish_wait loop from raid5c:make_request().
  79          * In general it is not permitted to sleep in that context as it
  80          * can cause the loop to spin freely.
  81          * That doesn't apply here as we can only reach this point
  82          * once with any loop.
  83          * When this function completes, either bp[page].map or
  84          * bp[page].hijacked.  In either case, this function will
  85          * abort before getting to this point again.  So there is
  86          * no risk of a free-spin, and so it is safe to assert
  87          * that sleeping here is allowed.
  88          */
  89         sched_annotate_sleep();
  90         mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  91         spin_lock_irq(&bitmap->lock);
  92
  93         if (mappage == NULL) {
  94                 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  95                 /* We don't support hijack for cluster raid */
  96                 if (no_hijack)
  97                         return -ENOMEM;
  98                 /* failed - set the hijacked flag so that we can use the
  99                  * pointer as a counter */
 100                 if (!bitmap->bp[page].map)
 101                         bitmap->bp[page].hijacked = 1;
 102         } else if (bitmap->bp[page].map ||
 103                    bitmap->bp[page].hijacked) {
 104                 /* somebody beat us to getting the page */
 105                 kfree(mappage);
 106         } else {
 107
 108                 /* no page was in place and we have one, so install it */
 109
 110                 bitmap->bp[page].map = mappage;
 111                 bitmap->missing_pages--;
 112         }
 113         return 0;
 114 }
 115
 116 /* if page is completely empty, put it back on the free list, or dealloc it */
 117 /* if page was hijacked, unmark the flag so it might get alloced next time */
 118 /* Note: lock should be held when calling this */
 119 static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 120 {
 121         char *ptr;
 122
 123         if (bitmap->bp[page].count) /* page is still busy */
 124                 return;
 125
 126         /* page is no longer in use, it can be released */
 127
 128         if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 129                 bitmap->bp[page].hijacked = 0;
 130                 bitmap->bp[page].map = NULL;
 131         } else {
 132                 /* normal case, free the page */
 133                 ptr = bitmap->bp[page].map;
 134                 bitmap->bp[page].map = NULL;
 135                 bitmap->missing_pages++;
 136                 kfree(ptr);
 137         }
 138 }
 139
 140 /*
 141  * bitmap file handling - read and write the bitmap file and its superblock
 142  */
 143
 144 /*
 145  * basic page I/O operations
 146  */
 147
 148 /* IO operations when bitmap is stored near all superblocks */
 149 static int read_sb_page(struct mddev *mddev, loff_t offset,
 150                         struct page *page,
 151                         unsigned long index, int size)
 152 {
 153         /* choose a good rdev and read the page from there */
 154
 155         struct md_rdev *rdev;
 156         sector_t target;
 157
 158         rdev_for_each(rdev, mddev) {
 159                 if (! test_bit(In_sync, &rdev->flags)
 160                     || test_bit(Faulty, &rdev->flags)
 161                     || test_bit(Bitmap_sync, &rdev->flags))
 162                         continue;
 163
 164                 target = offset + index * (PAGE_SIZE/512);
 165
 166                 if (sync_page_io(rdev, target,
 167                                  roundup(size, bdev_logical_block_size(rdev->bdev)),
 168                                  page, REQ_OP_READ, true)) {
 169                         page->index = index;
 170                         return 0;
 171                 }
 172         }
 173         return -EIO;
 174 }
 175
 176 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 177 {
 178         /* Iterate the disks of an mddev, using rcu to protect access to the
 179          * linked list, and raising the refcount of devices we return to ensure
 180          * they don't disappear while in use.
 181          * As devices are only added or removed when raid_disk is < 0 and
 182          * nr_pending is 0 and In_sync is clear, the entries we return will
 183          * still be in the same position on the list when we re-enter
 184          * list_for_each_entry_continue_rcu.
 185          *
 186          * Note that if entered with 'rdev == NULL' to start at the
 187          * beginning, we temporarily assign 'rdev' to an address which
 188          * isn't really an rdev, but which can be used by
 189          * list_for_each_entry_continue_rcu() to find the first entry.
 190          */
 191         rcu_read_lock();
 192         if (rdev == NULL)
 193                 /* start at the beginning */
 194                 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 195         else {
 196                 /* release the previous rdev and start from there. */
 197                 rdev_dec_pending(rdev, mddev);
 198         }
 199         list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 200                 if (rdev->raid_disk >= 0 &&
 201                     !test_bit(Faulty, &rdev->flags)) {
 202                         /* this is a usable devices */
 203                         atomic_inc(&rdev->nr_pending);
 204                         rcu_read_unlock();
 205                         return rdev;
 206                 }
 207         }
 208         rcu_read_unlock();
 209         return NULL;
 210 }
 211
 212 static unsigned int optimal_io_size(struct block_device *bdev,
 213                                     unsigned int last_page_size,
 214                                     unsigned int io_size)
 215 {
 216         if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
 217                 return roundup(last_page_size, bdev_io_opt(bdev));
 218         return io_size;
 219 }
 220
 221 static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
 222                                    loff_t start, loff_t boundary)
 223 {
 224         if (io_size != opt_size &&
 225             start + opt_size / SECTOR_SIZE <= boundary)
 226                 return opt_size;
 227         if (start + io_size / SECTOR_SIZE <= boundary)
 228                 return io_size;
 229
 230         /* Overflows boundary */
 231         return 0;
 232 }
 233
 234 static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 235                            struct page *page)
 236 {
 237         struct block_device *bdev;
 238         struct mddev *mddev = bitmap->mddev;
 239         struct bitmap_storage *store = &bitmap->storage;
 240         loff_t sboff, offset = mddev->bitmap_info.offset;
 241         sector_t ps, doff;
 242         unsigned int size = PAGE_SIZE;
 243         unsigned int opt_size = PAGE_SIZE;
 244
 245         bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 246         if (page->index == store->file_pages - 1) {
 247                 unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 248
 249                 if (last_page_size == 0)
 250                         last_page_size = PAGE_SIZE;
 251                 size = roundup(last_page_size, bdev_logical_block_size(bdev));
 252                 opt_size = optimal_io_size(bdev, last_page_size, size);
 253         }
 254
 255         ps = page->index * PAGE_SIZE / SECTOR_SIZE;
 256         sboff = rdev->sb_start + offset;
 257         doff = rdev->data_offset;
 258
 259         /* Just make sure we aren't corrupting data or metadata */
 260         if (mddev->external) {
 261                 /* Bitmap could be anywhere. */
 262                 if (sboff + ps > doff &&
 263                     sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE))
 264                         return -EINVAL;
 265         } else if (offset < 0) {
 266                 /* DATA  BITMAP METADATA  */
 267                 size = bitmap_io_size(size, opt_size, offset + ps, 0);
 268                 if (size == 0)
 269                         /* bitmap runs in to metadata */
 270                         return -EINVAL;
 271
 272                 if (doff + mddev->dev_sectors > sboff)
 273                         /* data runs in to bitmap */
 274                         return -EINVAL;
 275         } else if (rdev->sb_start < rdev->data_offset) {
 276                 /* METADATA BITMAP DATA */
 277                 size = bitmap_io_size(size, opt_size, sboff + ps, doff);
 278                 if (size == 0)
 279                         /* bitmap runs in to data */
 280                         return -EINVAL;
 281         } else {
 282                 /* DATA METADATA BITMAP - no problems */
 283         }
 284
 285         md_super_write(mddev, rdev, sboff + ps, (int) size, page);
 286         return 0;
 287 }
 288
 289 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 290 {
 291         struct md_rdev *rdev;
 292         struct mddev *mddev = bitmap->mddev;
 293         int ret;
 294
 295         do {
 296                 rdev = NULL;
 297                 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 298                         ret = __write_sb_page(rdev, bitmap, page);
 299                         if (ret)
 300                                 return ret;
 301                 }
 302         } while (wait && md_super_wait(mddev) < 0);
 303
 304         return 0;
 305 }
 306
 307 static void md_bitmap_file_kick(struct bitmap *bitmap);
 308 /*
 309  * write out a page to a file
 310  */
 311 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 312 {
 313         struct buffer_head *bh;
 314
 315         if (bitmap->storage.file == NULL) {
 316                 switch (write_sb_page(bitmap, page, wait)) {
 317                 case -EINVAL:
 318                         set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 319                 }
 320         } else {
 321
 322                 bh = page_buffers(page);
 323
 324                 while (bh && bh->b_blocknr) {
 325                         atomic_inc(&bitmap->pending_writes);
 326                         set_buffer_locked(bh);
 327                         set_buffer_mapped(bh);
 328                         submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
 329                         bh = bh->b_this_page;
 330                 }
 331
 332                 if (wait)
 333                         wait_event(bitmap->write_wait,
 334                                    atomic_read(&bitmap->pending_writes)==0);
 335         }
 336         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 337                 md_bitmap_file_kick(bitmap);
 338 }
 339
 340 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 341 {
 342         struct bitmap *bitmap = bh->b_private;
 343
 344         if (!uptodate)
 345                 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 346         if (atomic_dec_and_test(&bitmap->pending_writes))
 347                 wake_up(&bitmap->write_wait);
 348 }
 349
 350 static void free_buffers(struct page *page)
 351 {
 352         struct buffer_head *bh;
 353
 354         if (!PagePrivate(page))
 355                 return;
 356
 357         bh = page_buffers(page);
 358         while (bh) {
 359                 struct buffer_head *next = bh->b_this_page;
 360                 free_buffer_head(bh);
 361                 bh = next;
 362         }
 363         detach_page_private(page);
 364         put_page(page);
 365 }
 366
 367 /* read a page from a file.
 368  * We both read the page, and attach buffers to the page to record the
 369  * address of each block (using bmap).  These addresses will be used
 370  * to write the block later, completely bypassing the filesystem.
 371  * This usage is similar to how swap files are handled, and allows us
 372  * to write to a file with no concerns of memory allocation failing.
 373  */
 374 static int read_page(struct file *file, unsigned long index,
 375                      struct bitmap *bitmap,
 376                      unsigned long count,
 377                      struct page *page)
 378 {
 379         int ret = 0;
 380         struct inode *inode = file_inode(file);
 381         struct buffer_head *bh;
 382         sector_t block, blk_cur;
 383         unsigned long blocksize = i_blocksize(inode);
 384
 385         pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 386                  (unsigned long long)index << PAGE_SHIFT);
 387
 388         bh = alloc_page_buffers(page, blocksize, false);
 389         if (!bh) {
 390                 ret = -ENOMEM;
 391                 goto out;
 392         }
 393         attach_page_private(page, bh);
 394         blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 395         while (bh) {
 396                 block = blk_cur;
 397
 398                 if (count == 0)
 399                         bh->b_blocknr = 0;
 400                 else {
 401                         ret = bmap(inode, &block);
 402                         if (ret || !block) {
 403                                 ret = -EINVAL;
 404                                 bh->b_blocknr = 0;
 405                                 goto out;
 406                         }
 407
 408                         bh->b_blocknr = block;
 409                         bh->b_bdev = inode->i_sb->s_bdev;
 410                         if (count < blocksize)
 411                                 count = 0;
 412                         else
 413                                 count -= blocksize;
 414
 415                         bh->b_end_io = end_bitmap_write;
 416                         bh->b_private = bitmap;
 417                         atomic_inc(&bitmap->pending_writes);
 418                         set_buffer_locked(bh);
 419                         set_buffer_mapped(bh);
 420                         submit_bh(REQ_OP_READ, bh);
 421                 }
 422                 blk_cur++;
 423                 bh = bh->b_this_page;
 424         }
 425         page->index = index;
 426
 427         wait_event(bitmap->write_wait,
 428                    atomic_read(&bitmap->pending_writes)==0);
 429         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 430                 ret = -EIO;
 431 out:
 432         if (ret)
 433                 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 434                        (int)PAGE_SIZE,
 435                        (unsigned long long)index << PAGE_SHIFT,
 436                        ret);
 437         return ret;
 438 }
 439
 440 /*
 441  * bitmap file superblock operations
 442  */
 443
 444 /*
 445  * md_bitmap_wait_writes() should be called before writing any bitmap
 446  * blocks, to ensure previous writes, particularly from
 447  * md_bitmap_daemon_work(), have completed.
 448  */
 449 static void md_bitmap_wait_writes(struct bitmap *bitmap)
 450 {
 451         if (bitmap->storage.file)
 452                 wait_event(bitmap->write_wait,
 453                            atomic_read(&bitmap->pending_writes)==0);
 454         else
 455                 /* Note that we ignore the return value.  The writes
 456                  * might have failed, but that would just mean that
 457                  * some bits which should be cleared haven't been,
 458                  * which is safe.  The relevant bitmap blocks will
 459                  * probably get written again, but there is no great
 460                  * loss if they aren't.
 461                  */
 462                 md_super_wait(bitmap->mddev);
 463 }
 464
 465
 466 /* update the event counter and sync the superblock to disk */
 467 void md_bitmap_update_sb(struct bitmap *bitmap)
 468 {
 469         bitmap_super_t *sb;
 470
 471         if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 472                 return;
 473         if (bitmap->mddev->bitmap_info.external)
 474                 return;
 475         if (!bitmap->storage.sb_page) /* no superblock */
 476                 return;
 477         sb = kmap_atomic(bitmap->storage.sb_page);
 478         sb->events = cpu_to_le64(bitmap->mddev->events);
 479         if (bitmap->mddev->events < bitmap->events_cleared)
 480                 /* rocking back to read-only */
 481                 bitmap->events_cleared = bitmap->mddev->events;
 482         sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 483         /*
 484          * clear BITMAP_WRITE_ERROR bit to protect against the case that
 485          * a bitmap write error occurred but the later writes succeeded.
 486          */
 487         sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 488         /* Just in case these have been changed via sysfs: */
 489         sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 490         sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 491         /* This might have been changed by a reshape */
 492         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 493         sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 494         sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 495         sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 496                                            bitmap_info.space);
 497         kunmap_atomic(sb);
 498         write_page(bitmap, bitmap->storage.sb_page, 1);
 499 }
 500 EXPORT_SYMBOL(md_bitmap_update_sb);
 501
 502 /* print out the bitmap file superblock */
 503 void md_bitmap_print_sb(struct bitmap *bitmap)
 504 {
 505         bitmap_super_t *sb;
 506
 507         if (!bitmap || !bitmap->storage.sb_page)
 508                 return;
 509         sb = kmap_atomic(bitmap->storage.sb_page);
 510         pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 511         pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 512         pr_debug("       version: %u\n", le32_to_cpu(sb->version));
 513         pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 514                  le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 515                  le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 516                  le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 517                  le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 518         pr_debug("        events: %llu\n",
 519                  (unsigned long long) le64_to_cpu(sb->events));
 520         pr_debug("events cleared: %llu\n",
 521                  (unsigned long long) le64_to_cpu(sb->events_cleared));
 522         pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 523         pr_debug("     chunksize: %u B\n", le32_to_cpu(sb->chunksize));
 524         pr_debug("  daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
 525         pr_debug("     sync size: %llu KB\n",
 526                  (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 527         pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
 528         kunmap_atomic(sb);
 529 }
 530
 531 /*
 532  * bitmap_new_disk_sb
 533  * @bitmap
 534  *
 535  * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 536  * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 537  * This function verifies 'bitmap_info' and populates the on-disk bitmap
 538  * structure, which is to be written to disk.
 539  *
 540  * Returns: 0 on success, -Exxx on error
 541  */
 542 static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 543 {
 544         bitmap_super_t *sb;
 545         unsigned long chunksize, daemon_sleep, write_behind;
 546
 547         bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 548         if (bitmap->storage.sb_page == NULL)
 549                 return -ENOMEM;
 550         bitmap->storage.sb_page->index = 0;
 551
 552         sb = kmap_atomic(bitmap->storage.sb_page);
 553
 554         sb->magic = cpu_to_le32(BITMAP_MAGIC);
 555         sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 556
 557         chunksize = bitmap->mddev->bitmap_info.chunksize;
 558         BUG_ON(!chunksize);
 559         if (!is_power_of_2(chunksize)) {
 560                 kunmap_atomic(sb);
 561                 pr_warn("bitmap chunksize not a power of 2\n");
 562                 return -EINVAL;
 563         }
 564         sb->chunksize = cpu_to_le32(chunksize);
 565
 566         daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 567         if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 568                 pr_debug("Choosing daemon_sleep default (5 sec)\n");
 569                 daemon_sleep = 5 * HZ;
 570         }
 571         sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 572         bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 573
 574         /*
 575          * FIXME: write_behind for RAID1.  If not specified, what
 576          * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 577          */
 578         write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 579         if (write_behind > COUNTER_MAX)
 580                 write_behind = COUNTER_MAX / 2;
 581         sb->write_behind = cpu_to_le32(write_behind);
 582         bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 583
 584         /* keep the array size field of the bitmap superblock up to date */
 585         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 586
 587         memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 588
 589         set_bit(BITMAP_STALE, &bitmap->flags);
 590         sb->state = cpu_to_le32(bitmap->flags);
 591         bitmap->events_cleared = bitmap->mddev->events;
 592         sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 593         bitmap->mddev->bitmap_info.nodes = 0;
 594
 595         kunmap_atomic(sb);
 596
 597         return 0;
 598 }
 599
 600 /* read the superblock from the bitmap file and initialize some bitmap fields */
 601 static int md_bitmap_read_sb(struct bitmap *bitmap)
 602 {
 603         char *reason = NULL;
 604         bitmap_super_t *sb;
 605         unsigned long chunksize, daemon_sleep, write_behind;
 606         unsigned long long events;
 607         int nodes = 0;
 608         unsigned long sectors_reserved = 0;
 609         int err = -EINVAL;
 610         struct page *sb_page;
 611         loff_t offset = bitmap->mddev->bitmap_info.offset;
 612
 613         if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 614                 chunksize = 128 * 1024 * 1024;
 615                 daemon_sleep = 5 * HZ;
 616                 write_behind = 0;
 617                 set_bit(BITMAP_STALE, &bitmap->flags);
 618                 err = 0;
 619                 goto out_no_sb;
 620         }
 621         /* page 0 is the superblock, read it... */
 622         sb_page = alloc_page(GFP_KERNEL);
 623         if (!sb_page)
 624                 return -ENOMEM;
 625         bitmap->storage.sb_page = sb_page;
 626
 627 re_read:
 628         /* If cluster_slot is set, the cluster is setup */
 629         if (bitmap->cluster_slot >= 0) {
 630                 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 631
 632                 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 633                            (bitmap->mddev->bitmap_info.chunksize >> 9));
 634                 /* bits to bytes */
 635                 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 636                 /* to 4k blocks */
 637                 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 638                 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
 639                 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 640                         bitmap->cluster_slot, offset);
 641         }
 642
 643         if (bitmap->storage.file) {
 644                 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 645                 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 646
 647                 err = read_page(bitmap->storage.file, 0,
 648                                 bitmap, bytes, sb_page);
 649         } else {
 650                 err = read_sb_page(bitmap->mddev,
 651                                    offset,
 652                                    sb_page,
 653                                    0, sizeof(bitmap_super_t));
 654         }
 655         if (err)
 656                 return err;
 657
 658         err = -EINVAL;
 659         sb = kmap_atomic(sb_page);
 660
 661         chunksize = le32_to_cpu(sb->chunksize);
 662         daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 663         write_behind = le32_to_cpu(sb->write_behind);
 664         sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 665
 666         /* verify that the bitmap-specific fields are valid */
 667         if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 668                 reason = "bad magic";
 669         else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 670                  le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 671                 reason = "unrecognized superblock version";
 672         else if (chunksize < 512)
 673                 reason = "bitmap chunksize too small";
 674         else if (!is_power_of_2(chunksize))
 675                 reason = "bitmap chunksize not a power of 2";
 676         else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 677                 reason = "daemon sleep period out of range";
 678         else if (write_behind > COUNTER_MAX)
 679                 reason = "write-behind limit out of range (0 - 16383)";
 680         if (reason) {
 681                 pr_warn("%s: invalid bitmap file superblock: %s\n",
 682                         bmname(bitmap), reason);
 683                 goto out;
 684         }
 685
 686         /*
 687          * Setup nodes/clustername only if bitmap version is
 688          * cluster-compatible
 689          */
 690         if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 691                 nodes = le32_to_cpu(sb->nodes);
 692                 strscpy(bitmap->mddev->bitmap_info.cluster_name,
 693                                 sb->cluster_name, 64);
 694         }
 695
 696         /* keep the array size field of the bitmap superblock up to date */
 697         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 698
 699         if (bitmap->mddev->persistent) {
 700                 /*
 701                  * We have a persistent array superblock, so compare the
 702                  * bitmap's UUID and event counter to the mddev's
 703                  */
 704                 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 705                         pr_warn("%s: bitmap superblock UUID mismatch\n",
 706                                 bmname(bitmap));
 707                         goto out;
 708                 }
 709                 events = le64_to_cpu(sb->events);
 710                 if (!nodes && (events < bitmap->mddev->events)) {
 711                         pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 712                                 bmname(bitmap), events,
 713                                 (unsigned long long) bitmap->mddev->events);
 714                         set_bit(BITMAP_STALE, &bitmap->flags);
 715                 }
 716         }
 717
 718         /* assign fields using values from superblock */
 719         bitmap->flags |= le32_to_cpu(sb->state);
 720         if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 721                 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 722         bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 723         err = 0;
 724
 725 out:
 726         kunmap_atomic(sb);
 727         if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 728                 /* Assigning chunksize is required for "re_read" */
 729                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 730                 err = md_setup_cluster(bitmap->mddev, nodes);
 731                 if (err) {
 732                         pr_warn("%s: Could not setup cluster service (%d)\n",
 733                                 bmname(bitmap), err);
 734                         goto out_no_sb;
 735                 }
 736                 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 737                 goto re_read;
 738         }
 739
 740 out_no_sb:
 741         if (err == 0) {
 742                 if (test_bit(BITMAP_STALE, &bitmap->flags))
 743                         bitmap->events_cleared = bitmap->mddev->events;
 744                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 745                 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 746                 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 747                 bitmap->mddev->bitmap_info.nodes = nodes;
 748                 if (bitmap->mddev->bitmap_info.space == 0 ||
 749                         bitmap->mddev->bitmap_info.space > sectors_reserved)
 750                         bitmap->mddev->bitmap_info.space = sectors_reserved;
 751         } else {
 752                 md_bitmap_print_sb(bitmap);
 753                 if (bitmap->cluster_slot < 0)
 754                         md_cluster_stop(bitmap->mddev);
 755         }
 756         return err;
 757 }
 758
 759 /*
 760  * general bitmap file operations
 761  */
 762
 763 /*
 764  * on-disk bitmap:
 765  *
 766  * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 767  * file a page at a time. There's a superblock at the start of the file.
 768  */
 769 /* calculate the index of the page that contains this bit */
 770 static inline unsigned long file_page_index(struct bitmap_storage *store,
 771                                             unsigned long chunk)
 772 {
 773         if (store->sb_page)
 774                 chunk += sizeof(bitmap_super_t) << 3;
 775         return chunk >> PAGE_BIT_SHIFT;
 776 }
 777
 778 /* calculate the (bit) offset of this bit within a page */
 779 static inline unsigned long file_page_offset(struct bitmap_storage *store,
 780                                              unsigned long chunk)
 781 {
 782         if (store->sb_page)
 783                 chunk += sizeof(bitmap_super_t) << 3;
 784         return chunk & (PAGE_BITS - 1);
 785 }
 786
 787 /*
 788  * return a pointer to the page in the filemap that contains the given bit
 789  *
 790  */
 791 static inline struct page *filemap_get_page(struct bitmap_storage *store,
 792                                             unsigned long chunk)
 793 {
 794         if (file_page_index(store, chunk) >= store->file_pages)
 795                 return NULL;
 796         return store->filemap[file_page_index(store, chunk)];
 797 }
 798
 799 static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 800                                    unsigned long chunks, int with_super,
 801                                    int slot_number)
 802 {
 803         int pnum, offset = 0;
 804         unsigned long num_pages;
 805         unsigned long bytes;
 806
 807         bytes = DIV_ROUND_UP(chunks, 8);
 808         if (with_super)
 809                 bytes += sizeof(bitmap_super_t);
 810
 811         num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 812         offset = slot_number * num_pages;
 813
 814         store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 815                                        GFP_KERNEL);
 816         if (!store->filemap)
 817                 return -ENOMEM;
 818
 819         if (with_super && !store->sb_page) {
 820                 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 821                 if (store->sb_page == NULL)
 822                         return -ENOMEM;
 823         }
 824
 825         pnum = 0;
 826         if (store->sb_page) {
 827                 store->filemap[0] = store->sb_page;
 828                 pnum = 1;
 829                 store->sb_page->index = offset;
 830         }
 831
 832         for ( ; pnum < num_pages; pnum++) {
 833                 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 834                 if (!store->filemap[pnum]) {
 835                         store->file_pages = pnum;
 836                         return -ENOMEM;
 837                 }
 838                 store->filemap[pnum]->index = pnum + offset;
 839         }
 840         store->file_pages = pnum;
 841
 842         /* We need 4 bits per page, rounded up to a multiple
 843          * of sizeof(unsigned long) */
 844         store->filemap_attr = kzalloc(
 845                 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 846                 GFP_KERNEL);
 847         if (!store->filemap_attr)
 848                 return -ENOMEM;
 849
 850         store->bytes = bytes;
 851
 852         return 0;
 853 }
 854
 855 static void md_bitmap_file_unmap(struct bitmap_storage *store)
 856 {
 857         struct page **map, *sb_page;
 858         int pages;
 859         struct file *file;
 860
 861         file = store->file;
 862         map = store->filemap;
 863         pages = store->file_pages;
 864         sb_page = store->sb_page;
 865
 866         while (pages--)
 867                 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 868                         free_buffers(map[pages]);
 869         kfree(map);
 870         kfree(store->filemap_attr);
 871
 872         if (sb_page)
 873                 free_buffers(sb_page);
 874
 875         if (file) {
 876                 struct inode *inode = file_inode(file);
 877                 invalidate_mapping_pages(inode->i_mapping, 0, -1);
 878                 fput(file);
 879         }
 880 }
 881
 882 /*
 883  * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 884  * then it is no longer reliable, so we stop using it and we mark the file
 885  * as failed in the superblock
 886  */
 887 static void md_bitmap_file_kick(struct bitmap *bitmap)
 888 {
 889         char *path, *ptr = NULL;
 890
 891         if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 892                 md_bitmap_update_sb(bitmap);
 893
 894                 if (bitmap->storage.file) {
 895                         path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 896                         if (path)
 897                                 ptr = file_path(bitmap->storage.file,
 898                                              path, PAGE_SIZE);
 899
 900                         pr_warn("%s: kicking failed bitmap file %s from array!\n",
 901                                 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 902
 903                         kfree(path);
 904                 } else
 905                         pr_warn("%s: disabling internal bitmap due to errors\n",
 906                                 bmname(bitmap));
 907         }
 908 }
 909
 910 enum bitmap_page_attr {
 911         BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 912         BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 913                                     * i.e. counter is 1 or 2. */
 914         BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 915 };
 916
 917 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 918                                  enum bitmap_page_attr attr)
 919 {
 920         set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 921 }
 922
 923 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 924                                    enum bitmap_page_attr attr)
 925 {
 926         clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 927 }
 928
 929 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 930                                  enum bitmap_page_attr attr)
 931 {
 932         return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 933 }
 934
 935 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 936                                            enum bitmap_page_attr attr)
 937 {
 938         return test_and_clear_bit((pnum<<2) + attr,
 939                                   bitmap->storage.filemap_attr);
 940 }
 941 /*
 942  * bitmap_file_set_bit -- called before performing a write to the md device
 943  * to set (and eventually sync) a particular bit in the bitmap file
 944  *
 945  * we set the bit immediately, then we record the page number so that
 946  * when an unplug occurs, we can flush the dirty pages out to disk
 947  */
 948 static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 949 {
 950         unsigned long bit;
 951         struct page *page;
 952         void *kaddr;
 953         unsigned long chunk = block >> bitmap->counts.chunkshift;
 954         struct bitmap_storage *store = &bitmap->storage;
 955         unsigned long node_offset = 0;
 956
 957         if (mddev_is_clustered(bitmap->mddev))
 958                 node_offset = bitmap->cluster_slot * store->file_pages;
 959
 960         page = filemap_get_page(&bitmap->storage, chunk);
 961         if (!page)
 962                 return;
 963         bit = file_page_offset(&bitmap->storage, chunk);
 964
 965         /* set the bit */
 966         kaddr = kmap_atomic(page);
 967         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 968                 set_bit(bit, kaddr);
 969         else
 970                 set_bit_le(bit, kaddr);
 971         kunmap_atomic(kaddr);
 972         pr_debug("set file bit %lu page %lu\n", bit, page->index);
 973         /* record page number so it gets flushed to disk when unplug occurs */
 974         set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
 975 }
 976
 977 static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 978 {
 979         unsigned long bit;
 980         struct page *page;
 981         void *paddr;
 982         unsigned long chunk = block >> bitmap->counts.chunkshift;
 983         struct bitmap_storage *store = &bitmap->storage;
 984         unsigned long node_offset = 0;
 985
 986         if (mddev_is_clustered(bitmap->mddev))
 987                 node_offset = bitmap->cluster_slot * store->file_pages;
 988
 989         page = filemap_get_page(&bitmap->storage, chunk);
 990         if (!page)
 991                 return;
 992         bit = file_page_offset(&bitmap->storage, chunk);
 993         paddr = kmap_atomic(page);
 994         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 995                 clear_bit(bit, paddr);
 996         else
 997                 clear_bit_le(bit, paddr);
 998         kunmap_atomic(paddr);
 999         if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
1000                 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
1001                 bitmap->allclean = 0;
1002         }
1003 }
1004
1005 static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
1006 {
1007         unsigned long bit;
1008         struct page *page;
1009         void *paddr;
1010         unsigned long chunk = block >> bitmap->counts.chunkshift;
1011         int set = 0;
1012
1013         page = filemap_get_page(&bitmap->storage, chunk);
1014         if (!page)
1015                 return -EINVAL;
1016         bit = file_page_offset(&bitmap->storage, chunk);
1017         paddr = kmap_atomic(page);
1018         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1019                 set = test_bit(bit, paddr);
1020         else
1021                 set = test_bit_le(bit, paddr);
1022         kunmap_atomic(paddr);
1023         return set;
1024 }
1025
1026
1027 /* this gets called when the md device is ready to unplug its underlying
1028  * (slave) device queues -- before we let any writes go down, we need to
1029  * sync the dirty pages of the bitmap file to disk */
1030 void md_bitmap_unplug(struct bitmap *bitmap)
1031 {
1032         unsigned long i;
1033         int dirty, need_write;
1034         int writing = 0;
1035
1036         if (!bitmap || !bitmap->storage.filemap ||
1037             test_bit(BITMAP_STALE, &bitmap->flags))
1038                 return;
1039
1040         /* look at each page to see if there are any set bits that need to be
1041          * flushed out to disk */
1042         for (i = 0; i < bitmap->storage.file_pages; i++) {
1043                 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1044                 need_write = test_and_clear_page_attr(bitmap, i,
1045                                                       BITMAP_PAGE_NEEDWRITE);
1046                 if (dirty || need_write) {
1047                         if (!writing) {
1048                                 md_bitmap_wait_writes(bitmap);
1049                                 if (bitmap->mddev->queue)
1050                                         blk_add_trace_msg(bitmap->mddev->queue,
1051                                                           "md bitmap_unplug");
1052                         }
1053                         clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1054                         write_page(bitmap, bitmap->storage.filemap[i], 0);
1055                         writing = 1;
1056                 }
1057         }
1058         if (writing)
1059                 md_bitmap_wait_writes(bitmap);
1060
1061         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1062                 md_bitmap_file_kick(bitmap);
1063 }
1064 EXPORT_SYMBOL(md_bitmap_unplug);
1065
1066 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1067 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1068  * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1069  * memory mapping of the bitmap file
1070  * Special cases:
1071  *   if there's no bitmap file, or if the bitmap file had been
1072  *   previously kicked from the array, we mark all the bits as
1073  *   1's in order to cause a full resync.
1074  *
1075  * We ignore all bits for sectors that end earlier than 'start'.
1076  * This is used when reading an out-of-date bitmap...
1077  */
1078 static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1079 {
1080         unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1081         struct page *page = NULL;
1082         unsigned long bit_cnt = 0;
1083         struct file *file;
1084         unsigned long offset;
1085         int outofdate;
1086         int ret = -ENOSPC;
1087         void *paddr;
1088         struct bitmap_storage *store = &bitmap->storage;
1089
1090         chunks = bitmap->counts.chunks;
1091         file = store->file;
1092
1093         if (!file && !bitmap->mddev->bitmap_info.offset) {
1094                 /* No permanent bitmap - fill with '1s'. */
1095                 store->filemap = NULL;
1096                 store->file_pages = 0;
1097                 for (i = 0; i < chunks ; i++) {
1098                         /* if the disk bit is set, set the memory bit */
1099                         int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1100                                       >= start);
1101                         md_bitmap_set_memory_bits(bitmap,
1102                                                   (sector_t)i << bitmap->counts.chunkshift,
1103                                                   needed);
1104                 }
1105                 return 0;
1106         }
1107
1108         outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1109         if (outofdate)
1110                 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1111
1112         if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1113                 pr_warn("%s: bitmap file too short %lu < %lu\n",
1114                         bmname(bitmap),
1115                         (unsigned long) i_size_read(file->f_mapping->host),
1116                         store->bytes);
1117                 goto err;
1118         }
1119
1120         oldindex = ~0L;
1121         offset = 0;
1122         if (!bitmap->mddev->bitmap_info.external)
1123                 offset = sizeof(bitmap_super_t);
1124
1125         if (mddev_is_clustered(bitmap->mddev))
1126                 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1127
1128         for (i = 0; i < chunks; i++) {
1129                 int b;
1130                 index = file_page_index(&bitmap->storage, i);
1131                 bit = file_page_offset(&bitmap->storage, i);
1132                 if (index != oldindex) { /* this is a new page, read it in */
1133                         int count;
1134                         /* unmap the old page, we're done with it */
1135                         if (index == store->file_pages-1)
1136                                 count = store->bytes - index * PAGE_SIZE;
1137                         else
1138                                 count = PAGE_SIZE;
1139                         page = store->filemap[index];
1140                         if (file)
1141                                 ret = read_page(file, index, bitmap,
1142                                                 count, page);
1143                         else
1144                                 ret = read_sb_page(
1145                                         bitmap->mddev,
1146                                         bitmap->mddev->bitmap_info.offset,
1147                                         page,
1148                                         index + node_offset, count);
1149
1150                         if (ret)
1151                                 goto err;
1152
1153                         oldindex = index;
1154
1155                         if (outofdate) {
1156                                 /*
1157                                  * if bitmap is out of date, dirty the
1158                                  * whole page and write it out
1159                                  */
1160                                 paddr = kmap_atomic(page);
1161                                 memset(paddr + offset, 0xff,
1162                                        PAGE_SIZE - offset);
1163                                 kunmap_atomic(paddr);
1164                                 write_page(bitmap, page, 1);
1165
1166                                 ret = -EIO;
1167                                 if (test_bit(BITMAP_WRITE_ERROR,
1168                                              &bitmap->flags))
1169                                         goto err;
1170                         }
1171                 }
1172                 paddr = kmap_atomic(page);
1173                 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1174                         b = test_bit(bit, paddr);
1175                 else
1176                         b = test_bit_le(bit, paddr);
1177                 kunmap_atomic(paddr);
1178                 if (b) {
1179                         /* if the disk bit is set, set the memory bit */
1180                         int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1181                                       >= start);
1182                         md_bitmap_set_memory_bits(bitmap,
1183                                                   (sector_t)i << bitmap->counts.chunkshift,
1184                                                   needed);
1185                         bit_cnt++;
1186                 }
1187                 offset = 0;
1188         }
1189
1190         pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1191                  bmname(bitmap), store->file_pages,
1192                  bit_cnt, chunks);
1193
1194         return 0;
1195
1196  err:
1197         pr_warn("%s: bitmap initialisation failed: %d\n",
1198                 bmname(bitmap), ret);
1199         return ret;
1200 }
1201
1202 void md_bitmap_write_all(struct bitmap *bitmap)
1203 {
1204         /* We don't actually write all bitmap blocks here,
1205          * just flag them as needing to be written
1206          */
1207         int i;
1208
1209         if (!bitmap || !bitmap->storage.filemap)
1210                 return;
1211         if (bitmap->storage.file)
1212                 /* Only one copy, so nothing needed */
1213                 return;
1214
1215         for (i = 0; i < bitmap->storage.file_pages; i++)
1216                 set_page_attr(bitmap, i,
1217                               BITMAP_PAGE_NEEDWRITE);
1218         bitmap->allclean = 0;
1219 }
1220
1221 static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1222                                  sector_t offset, int inc)
1223 {
1224         sector_t chunk = offset >> bitmap->chunkshift;
1225         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1226         bitmap->bp[page].count += inc;
1227         md_bitmap_checkfree(bitmap, page);
1228 }
1229
1230 static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1231 {
1232         sector_t chunk = offset >> bitmap->chunkshift;
1233         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1234         struct bitmap_page *bp = &bitmap->bp[page];
1235
1236         if (!bp->pending)
1237                 bp->pending = 1;
1238 }
1239
1240 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1241                                                sector_t offset, sector_t *blocks,
1242                                                int create);
1243
1244 /*
1245  * bitmap daemon -- periodically wakes up to clean bits and flush pages
1246  *                      out to disk
1247  */
1248
1249 void md_bitmap_daemon_work(struct mddev *mddev)
1250 {
1251         struct bitmap *bitmap;
1252         unsigned long j;
1253         unsigned long nextpage;
1254         sector_t blocks;
1255         struct bitmap_counts *counts;
1256
1257         /* Use a mutex to guard daemon_work against
1258          * bitmap_destroy.
1259          */
1260         mutex_lock(&mddev->bitmap_info.mutex);
1261         bitmap = mddev->bitmap;
1262         if (bitmap == NULL) {
1263                 mutex_unlock(&mddev->bitmap_info.mutex);
1264                 return;
1265         }
1266         if (time_before(jiffies, bitmap->daemon_lastrun
1267                         + mddev->bitmap_info.daemon_sleep))
1268                 goto done;
1269
1270         bitmap->daemon_lastrun = jiffies;
1271         if (bitmap->allclean) {
1272                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1273                 goto done;
1274         }
1275         bitmap->allclean = 1;
1276
1277         if (bitmap->mddev->queue)
1278                 blk_add_trace_msg(bitmap->mddev->queue,
1279                                   "md bitmap_daemon_work");
1280
1281         /* Any file-page which is PENDING now needs to be written.
1282          * So set NEEDWRITE now, then after we make any last-minute changes
1283          * we will write it.
1284          */
1285         for (j = 0; j < bitmap->storage.file_pages; j++)
1286                 if (test_and_clear_page_attr(bitmap, j,
1287                                              BITMAP_PAGE_PENDING))
1288                         set_page_attr(bitmap, j,
1289                                       BITMAP_PAGE_NEEDWRITE);
1290
1291         if (bitmap->need_sync &&
1292             mddev->bitmap_info.external == 0) {
1293                 /* Arrange for superblock update as well as
1294                  * other changes */
1295                 bitmap_super_t *sb;
1296                 bitmap->need_sync = 0;
1297                 if (bitmap->storage.filemap) {
1298                         sb = kmap_atomic(bitmap->storage.sb_page);
1299                         sb->events_cleared =
1300                                 cpu_to_le64(bitmap->events_cleared);
1301                         kunmap_atomic(sb);
1302                         set_page_attr(bitmap, 0,
1303                                       BITMAP_PAGE_NEEDWRITE);
1304                 }
1305         }
1306         /* Now look at the bitmap counters and if any are '2' or '1',
1307          * decrement and handle accordingly.
1308          */
1309         counts = &bitmap->counts;
1310         spin_lock_irq(&counts->lock);
1311         nextpage = 0;
1312         for (j = 0; j < counts->chunks; j++) {
1313                 bitmap_counter_t *bmc;
1314                 sector_t  block = (sector_t)j << counts->chunkshift;
1315
1316                 if (j == nextpage) {
1317                         nextpage += PAGE_COUNTER_RATIO;
1318                         if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1319                                 j |= PAGE_COUNTER_MASK;
1320                                 continue;
1321                         }
1322                         counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1323                 }
1324
1325                 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1326                 if (!bmc) {
1327                         j |= PAGE_COUNTER_MASK;
1328                         continue;
1329                 }
1330                 if (*bmc == 1 && !bitmap->need_sync) {
1331                         /* We can clear the bit */
1332                         *bmc = 0;
1333                         md_bitmap_count_page(counts, block, -1);
1334                         md_bitmap_file_clear_bit(bitmap, block);
1335                 } else if (*bmc && *bmc <= 2) {
1336                         *bmc = 1;
1337                         md_bitmap_set_pending(counts, block);
1338                         bitmap->allclean = 0;
1339                 }
1340         }
1341         spin_unlock_irq(&counts->lock);
1342
1343         md_bitmap_wait_writes(bitmap);
1344         /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1345          * DIRTY pages need to be written by bitmap_unplug so it can wait
1346          * for them.
1347          * If we find any DIRTY page we stop there and let bitmap_unplug
1348          * handle all the rest.  This is important in the case where
1349          * the first blocking holds the superblock and it has been updated.
1350          * We mustn't write any other blocks before the superblock.
1351          */
1352         for (j = 0;
1353              j < bitmap->storage.file_pages
1354                      && !test_bit(BITMAP_STALE, &bitmap->flags);
1355              j++) {
1356                 if (test_page_attr(bitmap, j,
1357                                    BITMAP_PAGE_DIRTY))
1358                         /* bitmap_unplug will handle the rest */
1359                         break;
1360                 if (bitmap->storage.filemap &&
1361                     test_and_clear_page_attr(bitmap, j,
1362                                              BITMAP_PAGE_NEEDWRITE)) {
1363                         write_page(bitmap, bitmap->storage.filemap[j], 0);
1364                 }
1365         }
1366
1367  done:
1368         if (bitmap->allclean == 0)
1369                 mddev->thread->timeout =
1370                         mddev->bitmap_info.daemon_sleep;
1371         mutex_unlock(&mddev->bitmap_info.mutex);
1372 }
1373
1374 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1375                                                sector_t offset, sector_t *blocks,
1376                                                int create)
1377 __releases(bitmap->lock)
1378 __acquires(bitmap->lock)
1379 {
1380         /* If 'create', we might release the lock and reclaim it.
1381          * The lock must have been taken with interrupts enabled.
1382          * If !create, we don't release the lock.
1383          */
1384         sector_t chunk = offset >> bitmap->chunkshift;
1385         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1386         unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1387         sector_t csize;
1388         int err;
1389
1390         err = md_bitmap_checkpage(bitmap, page, create, 0);
1391
1392         if (bitmap->bp[page].hijacked ||
1393             bitmap->bp[page].map == NULL)
1394                 csize = ((sector_t)1) << (bitmap->chunkshift +
1395                                           PAGE_COUNTER_SHIFT);
1396         else
1397                 csize = ((sector_t)1) << bitmap->chunkshift;
1398         *blocks = csize - (offset & (csize - 1));
1399
1400         if (err < 0)
1401                 return NULL;
1402
1403         /* now locked ... */
1404
1405         if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1406                 /* should we use the first or second counter field
1407                  * of the hijacked pointer? */
1408                 int hi = (pageoff > PAGE_COUNTER_MASK);
1409                 return  &((bitmap_counter_t *)
1410                           &bitmap->bp[page].map)[hi];
1411         } else /* page is allocated */
1412                 return (bitmap_counter_t *)
1413                         &(bitmap->bp[page].map[pageoff]);
1414 }
1415
1416 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1417 {
1418         if (!bitmap)
1419                 return 0;
1420
1421         if (behind) {
1422                 int bw;
1423                 atomic_inc(&bitmap->behind_writes);
1424                 bw = atomic_read(&bitmap->behind_writes);
1425                 if (bw > bitmap->behind_writes_used)
1426                         bitmap->behind_writes_used = bw;
1427
1428                 pr_debug("inc write-behind count %d/%lu\n",
1429                          bw, bitmap->mddev->bitmap_info.max_write_behind);
1430         }
1431
1432         while (sectors) {
1433                 sector_t blocks;
1434                 bitmap_counter_t *bmc;
1435
1436                 spin_lock_irq(&bitmap->counts.lock);
1437                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1438                 if (!bmc) {
1439                         spin_unlock_irq(&bitmap->counts.lock);
1440                         return 0;
1441                 }
1442
1443                 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1444                         DEFINE_WAIT(__wait);
1445                         /* note that it is safe to do the prepare_to_wait
1446                          * after the test as long as we do it before dropping
1447                          * the spinlock.
1448                          */
1449                         prepare_to_wait(&bitmap->overflow_wait, &__wait,
1450                                         TASK_UNINTERRUPTIBLE);
1451                         spin_unlock_irq(&bitmap->counts.lock);
1452                         schedule();
1453                         finish_wait(&bitmap->overflow_wait, &__wait);
1454                         continue;
1455                 }
1456
1457                 switch (*bmc) {
1458                 case 0:
1459                         md_bitmap_file_set_bit(bitmap, offset);
1460                         md_bitmap_count_page(&bitmap->counts, offset, 1);
1461                         fallthrough;
1462                 case 1:
1463                         *bmc = 2;
1464                 }
1465
1466                 (*bmc)++;
1467
1468                 spin_unlock_irq(&bitmap->counts.lock);
1469
1470                 offset += blocks;
1471                 if (sectors > blocks)
1472                         sectors -= blocks;
1473                 else
1474                         sectors = 0;
1475         }
1476         return 0;
1477 }
1478 EXPORT_SYMBOL(md_bitmap_startwrite);
1479
1480 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1481                         unsigned long sectors, int success, int behind)
1482 {
1483         if (!bitmap)
1484                 return;
1485         if (behind) {
1486                 if (atomic_dec_and_test(&bitmap->behind_writes))
1487                         wake_up(&bitmap->behind_wait);
1488                 pr_debug("dec write-behind count %d/%lu\n",
1489                          atomic_read(&bitmap->behind_writes),
1490                          bitmap->mddev->bitmap_info.max_write_behind);
1491         }
1492
1493         while (sectors) {
1494                 sector_t blocks;
1495                 unsigned long flags;
1496                 bitmap_counter_t *bmc;
1497
1498                 spin_lock_irqsave(&bitmap->counts.lock, flags);
1499                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1500                 if (!bmc) {
1501                         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1502                         return;
1503                 }
1504
1505                 if (success && !bitmap->mddev->degraded &&
1506                     bitmap->events_cleared < bitmap->mddev->events) {
1507                         bitmap->events_cleared = bitmap->mddev->events;
1508                         bitmap->need_sync = 1;
1509                         sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1510                 }
1511
1512                 if (!success && !NEEDED(*bmc))
1513                         *bmc |= NEEDED_MASK;
1514
1515                 if (COUNTER(*bmc) == COUNTER_MAX)
1516                         wake_up(&bitmap->overflow_wait);
1517
1518                 (*bmc)--;
1519                 if (*bmc <= 2) {
1520                         md_bitmap_set_pending(&bitmap->counts, offset);
1521                         bitmap->allclean = 0;
1522                 }
1523                 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1524                 offset += blocks;
1525                 if (sectors > blocks)
1526                         sectors -= blocks;
1527                 else
1528                         sectors = 0;
1529         }
1530 }
1531 EXPORT_SYMBOL(md_bitmap_endwrite);
1532
1533 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1534                                int degraded)
1535 {
1536         bitmap_counter_t *bmc;
1537         int rv;
1538         if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1539                 *blocks = 1024;
1540                 return 1; /* always resync if no bitmap */
1541         }
1542         spin_lock_irq(&bitmap->counts.lock);
1543         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1544         rv = 0;
1545         if (bmc) {
1546                 /* locked */
1547                 if (RESYNC(*bmc))
1548                         rv = 1;
1549                 else if (NEEDED(*bmc)) {
1550                         rv = 1;
1551                         if (!degraded) { /* don't set/clear bits if degraded */
1552                                 *bmc |= RESYNC_MASK;
1553                                 *bmc &= ~NEEDED_MASK;
1554                         }
1555                 }
1556         }
1557         spin_unlock_irq(&bitmap->counts.lock);
1558         return rv;
1559 }
1560
1561 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1562                          int degraded)
1563 {
1564         /* bitmap_start_sync must always report on multiples of whole
1565          * pages, otherwise resync (which is very PAGE_SIZE based) will
1566          * get confused.
1567          * So call __bitmap_start_sync repeatedly (if needed) until
1568          * At least PAGE_SIZE>>9 blocks are covered.
1569          * Return the 'or' of the result.
1570          */
1571         int rv = 0;
1572         sector_t blocks1;
1573
1574         *blocks = 0;
1575         while (*blocks < (PAGE_SIZE>>9)) {
1576                 rv |= __bitmap_start_sync(bitmap, offset,
1577                                           &blocks1, degraded);
1578                 offset += blocks1;
1579                 *blocks += blocks1;
1580         }
1581         return rv;
1582 }
1583 EXPORT_SYMBOL(md_bitmap_start_sync);
1584
1585 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1586 {
1587         bitmap_counter_t *bmc;
1588         unsigned long flags;
1589
1590         if (bitmap == NULL) {
1591                 *blocks = 1024;
1592                 return;
1593         }
1594         spin_lock_irqsave(&bitmap->counts.lock, flags);
1595         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1596         if (bmc == NULL)
1597                 goto unlock;
1598         /* locked */
1599         if (RESYNC(*bmc)) {
1600                 *bmc &= ~RESYNC_MASK;
1601
1602                 if (!NEEDED(*bmc) && aborted)
1603                         *bmc |= NEEDED_MASK;
1604                 else {
1605                         if (*bmc <= 2) {
1606                                 md_bitmap_set_pending(&bitmap->counts, offset);
1607                                 bitmap->allclean = 0;
1608                         }
1609                 }
1610         }
1611  unlock:
1612         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1613 }
1614 EXPORT_SYMBOL(md_bitmap_end_sync);
1615
1616 void md_bitmap_close_sync(struct bitmap *bitmap)
1617 {
1618         /* Sync has finished, and any bitmap chunks that weren't synced
1619          * properly have been aborted.  It remains to us to clear the
1620          * RESYNC bit wherever it is still on
1621          */
1622         sector_t sector = 0;
1623         sector_t blocks;
1624         if (!bitmap)
1625                 return;
1626         while (sector < bitmap->mddev->resync_max_sectors) {
1627                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1628                 sector += blocks;
1629         }
1630 }
1631 EXPORT_SYMBOL(md_bitmap_close_sync);
1632
1633 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1634 {
1635         sector_t s = 0;
1636         sector_t blocks;
1637
1638         if (!bitmap)
1639                 return;
1640         if (sector == 0) {
1641                 bitmap->last_end_sync = jiffies;
1642                 return;
1643         }
1644         if (!force && time_before(jiffies, (bitmap->last_end_sync
1645                                   + bitmap->mddev->bitmap_info.daemon_sleep)))
1646                 return;
1647         wait_event(bitmap->mddev->recovery_wait,
1648                    atomic_read(&bitmap->mddev->recovery_active) == 0);
1649
1650         bitmap->mddev->curr_resync_completed = sector;
1651         set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1652         sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1653         s = 0;
1654         while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1655                 md_bitmap_end_sync(bitmap, s, &blocks, 0);
1656                 s += blocks;
1657         }
1658         bitmap->last_end_sync = jiffies;
1659         sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1660 }
1661 EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1662
1663 void md_bitmap_sync_with_cluster(struct mddev *mddev,
1664                               sector_t old_lo, sector_t old_hi,
1665                               sector_t new_lo, sector_t new_hi)
1666 {
1667         struct bitmap *bitmap = mddev->bitmap;
1668         sector_t sector, blocks = 0;
1669
1670         for (sector = old_lo; sector < new_lo; ) {
1671                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1672                 sector += blocks;
1673         }
1674         WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1675
1676         for (sector = old_hi; sector < new_hi; ) {
1677                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1678                 sector += blocks;
1679         }
1680         WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1681 }
1682 EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1683
1684 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1685 {
1686         /* For each chunk covered by any of these sectors, set the
1687          * counter to 2 and possibly set resync_needed.  They should all
1688          * be 0 at this point
1689          */
1690
1691         sector_t secs;
1692         bitmap_counter_t *bmc;
1693         spin_lock_irq(&bitmap->counts.lock);
1694         bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1695         if (!bmc) {
1696                 spin_unlock_irq(&bitmap->counts.lock);
1697                 return;
1698         }
1699         if (!*bmc) {
1700                 *bmc = 2;
1701                 md_bitmap_count_page(&bitmap->counts, offset, 1);
1702                 md_bitmap_set_pending(&bitmap->counts, offset);
1703                 bitmap->allclean = 0;
1704         }
1705         if (needed)
1706                 *bmc |= NEEDED_MASK;
1707         spin_unlock_irq(&bitmap->counts.lock);
1708 }
1709
1710 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1711 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1712 {
1713         unsigned long chunk;
1714
1715         for (chunk = s; chunk <= e; chunk++) {
1716                 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1717                 md_bitmap_set_memory_bits(bitmap, sec, 1);
1718                 md_bitmap_file_set_bit(bitmap, sec);
1719                 if (sec < bitmap->mddev->recovery_cp)
1720                         /* We are asserting that the array is dirty,
1721                          * so move the recovery_cp address back so
1722                          * that it is obvious that it is dirty
1723                          */
1724                         bitmap->mddev->recovery_cp = sec;
1725         }
1726 }
1727
1728 /*
1729  * flush out any pending updates
1730  */
1731 void md_bitmap_flush(struct mddev *mddev)
1732 {
1733         struct bitmap *bitmap = mddev->bitmap;
1734         long sleep;
1735
1736         if (!bitmap) /* there was no bitmap */
1737                 return;
1738
1739         /* run the daemon_work three time to ensure everything is flushed
1740          * that can be
1741          */
1742         sleep = mddev->bitmap_info.daemon_sleep * 2;
1743         bitmap->daemon_lastrun -= sleep;
1744         md_bitmap_daemon_work(mddev);
1745         bitmap->daemon_lastrun -= sleep;
1746         md_bitmap_daemon_work(mddev);
1747         bitmap->daemon_lastrun -= sleep;
1748         md_bitmap_daemon_work(mddev);
1749         if (mddev->bitmap_info.external)
1750                 md_super_wait(mddev);
1751         md_bitmap_update_sb(bitmap);
1752 }
1753
1754 /*
1755  * free memory that was allocated
1756  */
1757 void md_bitmap_free(struct bitmap *bitmap)
1758 {
1759         unsigned long k, pages;
1760         struct bitmap_page *bp;
1761
1762         if (!bitmap) /* there was no bitmap */
1763                 return;
1764
1765         if (bitmap->sysfs_can_clear)
1766                 sysfs_put(bitmap->sysfs_can_clear);
1767
1768         if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1769                 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1770                 md_cluster_stop(bitmap->mddev);
1771
1772         /* Shouldn't be needed - but just in case.... */
1773         wait_event(bitmap->write_wait,
1774                    atomic_read(&bitmap->pending_writes) == 0);
1775
1776         /* release the bitmap file  */
1777         md_bitmap_file_unmap(&bitmap->storage);
1778
1779         bp = bitmap->counts.bp;
1780         pages = bitmap->counts.pages;
1781
1782         /* free all allocated memory */
1783
1784         if (bp) /* deallocate the page memory */
1785                 for (k = 0; k < pages; k++)
1786                         if (bp[k].map && !bp[k].hijacked)
1787                                 kfree(bp[k].map);
1788         kfree(bp);
1789         kfree(bitmap);
1790 }
1791 EXPORT_SYMBOL(md_bitmap_free);
1792
1793 void md_bitmap_wait_behind_writes(struct mddev *mddev)
1794 {
1795         struct bitmap *bitmap = mddev->bitmap;
1796
1797         /* wait for behind writes to complete */
1798         if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1799                 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1800                          mdname(mddev));
1801                 /* need to kick something here to make sure I/O goes? */
1802                 wait_event(bitmap->behind_wait,
1803                            atomic_read(&bitmap->behind_writes) == 0);
1804         }
1805 }
1806
1807 void md_bitmap_destroy(struct mddev *mddev)
1808 {
1809         struct bitmap *bitmap = mddev->bitmap;
1810
1811         if (!bitmap) /* there was no bitmap */
1812                 return;
1813
1814         md_bitmap_wait_behind_writes(mddev);
1815         if (!mddev->serialize_policy)
1816                 mddev_destroy_serial_pool(mddev, NULL, true);
1817
1818         mutex_lock(&mddev->bitmap_info.mutex);
1819         spin_lock(&mddev->lock);
1820         mddev->bitmap = NULL; /* disconnect from the md device */
1821         spin_unlock(&mddev->lock);
1822         mutex_unlock(&mddev->bitmap_info.mutex);
1823         if (mddev->thread)
1824                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1825
1826         md_bitmap_free(bitmap);
1827 }
1828
1829 /*
1830  * initialize the bitmap structure
1831  * if this returns an error, bitmap_destroy must be called to do clean up
1832  * once mddev->bitmap is set
1833  */
1834 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1835 {
1836         struct bitmap *bitmap;
1837         sector_t blocks = mddev->resync_max_sectors;
1838         struct file *file = mddev->bitmap_info.file;
1839         int err;
1840         struct kernfs_node *bm = NULL;
1841
1842         BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1843
1844         BUG_ON(file && mddev->bitmap_info.offset);
1845
1846         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1847                 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1848                           mdname(mddev));
1849                 return ERR_PTR(-EBUSY);
1850         }
1851
1852         bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1853         if (!bitmap)
1854                 return ERR_PTR(-ENOMEM);
1855
1856         spin_lock_init(&bitmap->counts.lock);
1857         atomic_set(&bitmap->pending_writes, 0);
1858         init_waitqueue_head(&bitmap->write_wait);
1859         init_waitqueue_head(&bitmap->overflow_wait);
1860         init_waitqueue_head(&bitmap->behind_wait);
1861
1862         bitmap->mddev = mddev;
1863         bitmap->cluster_slot = slot;
1864
1865         if (mddev->kobj.sd)
1866                 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1867         if (bm) {
1868                 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1869                 sysfs_put(bm);
1870         } else
1871                 bitmap->sysfs_can_clear = NULL;
1872
1873         bitmap->storage.file = file;
1874         if (file) {
1875                 get_file(file);
1876                 /* As future accesses to this file will use bmap,
1877                  * and bypass the page cache, we must sync the file
1878                  * first.
1879                  */
1880                 vfs_fsync(file, 1);
1881         }
1882         /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1883         if (!mddev->bitmap_info.external) {
1884                 /*
1885                  * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1886                  * instructing us to create a new on-disk bitmap instance.
1887                  */
1888                 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1889                         err = md_bitmap_new_disk_sb(bitmap);
1890                 else
1891                         err = md_bitmap_read_sb(bitmap);
1892         } else {
1893                 err = 0;
1894                 if (mddev->bitmap_info.chunksize == 0 ||
1895                     mddev->bitmap_info.daemon_sleep == 0)
1896                         /* chunksize and time_base need to be
1897                          * set first. */
1898                         err = -EINVAL;
1899         }
1900         if (err)
1901                 goto error;
1902
1903         bitmap->daemon_lastrun = jiffies;
1904         err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1905         if (err)
1906                 goto error;
1907
1908         pr_debug("created bitmap (%lu pages) for device %s\n",
1909                  bitmap->counts.pages, bmname(bitmap));
1910
1911         err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1912         if (err)
1913                 goto error;
1914
1915         return bitmap;
1916  error:
1917         md_bitmap_free(bitmap);
1918         return ERR_PTR(err);
1919 }
1920
1921 int md_bitmap_load(struct mddev *mddev)
1922 {
1923         int err = 0;
1924         sector_t start = 0;
1925         sector_t sector = 0;
1926         struct bitmap *bitmap = mddev->bitmap;
1927         struct md_rdev *rdev;
1928
1929         if (!bitmap)
1930                 goto out;
1931
1932         rdev_for_each(rdev, mddev)
1933                 mddev_create_serial_pool(mddev, rdev, true);
1934
1935         if (mddev_is_clustered(mddev))
1936                 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1937
1938         /* Clear out old bitmap info first:  Either there is none, or we
1939          * are resuming after someone else has possibly changed things,
1940          * so we should forget old cached info.
1941          * All chunks should be clean, but some might need_sync.
1942          */
1943         while (sector < mddev->resync_max_sectors) {
1944                 sector_t blocks;
1945                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1946                 sector += blocks;
1947         }
1948         md_bitmap_close_sync(bitmap);
1949
1950         if (mddev->degraded == 0
1951             || bitmap->events_cleared == mddev->events)
1952                 /* no need to keep dirty bits to optimise a
1953                  * re-add of a missing device */
1954                 start = mddev->recovery_cp;
1955
1956         mutex_lock(&mddev->bitmap_info.mutex);
1957         err = md_bitmap_init_from_disk(bitmap, start);
1958         mutex_unlock(&mddev->bitmap_info.mutex);
1959
1960         if (err)
1961                 goto out;
1962         clear_bit(BITMAP_STALE, &bitmap->flags);
1963
1964         /* Kick recovery in case any bits were set */
1965         set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1966
1967         mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1968         md_wakeup_thread(mddev->thread);
1969
1970         md_bitmap_update_sb(bitmap);
1971
1972         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1973                 err = -EIO;
1974 out:
1975         return err;
1976 }
1977 EXPORT_SYMBOL_GPL(md_bitmap_load);
1978
1979 /* caller need to free returned bitmap with md_bitmap_free() */
1980 struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1981 {
1982         int rv = 0;
1983         struct bitmap *bitmap;
1984
1985         bitmap = md_bitmap_create(mddev, slot);
1986         if (IS_ERR(bitmap)) {
1987                 rv = PTR_ERR(bitmap);
1988                 return ERR_PTR(rv);
1989         }
1990
1991         rv = md_bitmap_init_from_disk(bitmap, 0);
1992         if (rv) {
1993                 md_bitmap_free(bitmap);
1994                 return ERR_PTR(rv);
1995         }
1996
1997         return bitmap;
1998 }
1999 EXPORT_SYMBOL(get_bitmap_from_slot);
2000
2001 /* Loads the bitmap associated with slot and copies the resync information
2002  * to our bitmap
2003  */
2004 int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
2005                 sector_t *low, sector_t *high, bool clear_bits)
2006 {
2007         int rv = 0, i, j;
2008         sector_t block, lo = 0, hi = 0;
2009         struct bitmap_counts *counts;
2010         struct bitmap *bitmap;
2011
2012         bitmap = get_bitmap_from_slot(mddev, slot);
2013         if (IS_ERR(bitmap)) {
2014                 pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
2015                 return -1;
2016         }
2017
2018         counts = &bitmap->counts;
2019         for (j = 0; j < counts->chunks; j++) {
2020                 block = (sector_t)j << counts->chunkshift;
2021                 if (md_bitmap_file_test_bit(bitmap, block)) {
2022                         if (!lo)
2023                                 lo = block;
2024                         hi = block;
2025                         md_bitmap_file_clear_bit(bitmap, block);
2026                         md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2027                         md_bitmap_file_set_bit(mddev->bitmap, block);
2028                 }
2029         }
2030
2031         if (clear_bits) {
2032                 md_bitmap_update_sb(bitmap);
2033                 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2034                  * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2035                 for (i = 0; i < bitmap->storage.file_pages; i++)
2036                         if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2037                                 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2038                 md_bitmap_unplug(bitmap);
2039         }
2040         md_bitmap_unplug(mddev->bitmap);
2041         *low = lo;
2042         *high = hi;
2043         md_bitmap_free(bitmap);
2044
2045         return rv;
2046 }
2047 EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2048
2049
2050 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2051 {
2052         unsigned long chunk_kb;
2053         struct bitmap_counts *counts;
2054
2055         if (!bitmap)
2056                 return;
2057
2058         counts = &bitmap->counts;
2059
2060         chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2061         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2062                    "%lu%s chunk",
2063                    counts->pages - counts->missing_pages,
2064                    counts->pages,
2065                    (counts->pages - counts->missing_pages)
2066                    << (PAGE_SHIFT - 10),
2067                    chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2068                    chunk_kb ? "KB" : "B");
2069         if (bitmap->storage.file) {
2070                 seq_printf(seq, ", file: ");
2071                 seq_file_path(seq, bitmap->storage.file, " \t\n");
2072         }
2073
2074         seq_printf(seq, "\n");
2075 }
2076
2077 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2078                   int chunksize, int init)
2079 {
2080         /* If chunk_size is 0, choose an appropriate chunk size.
2081          * Then possibly allocate new storage space.
2082          * Then quiesce, copy bits, replace bitmap, and re-start
2083          *
2084          * This function is called both to set up the initial bitmap
2085          * and to resize the bitmap while the array is active.
2086          * If this happens as a result of the array being resized,
2087          * chunksize will be zero, and we need to choose a suitable
2088          * chunksize, otherwise we use what we are given.
2089          */
2090         struct bitmap_storage store;
2091         struct bitmap_counts old_counts;
2092         unsigned long chunks;
2093         sector_t block;
2094         sector_t old_blocks, new_blocks;
2095         int chunkshift;
2096         int ret = 0;
2097         long pages;
2098         struct bitmap_page *new_bp;
2099
2100         if (bitmap->storage.file && !init) {
2101                 pr_info("md: cannot resize file-based bitmap\n");
2102                 return -EINVAL;
2103         }
2104
2105         if (chunksize == 0) {
2106                 /* If there is enough space, leave the chunk size unchanged,
2107                  * else increase by factor of two until there is enough space.
2108                  */
2109                 long bytes;
2110                 long space = bitmap->mddev->bitmap_info.space;
2111
2112                 if (space == 0) {
2113                         /* We don't know how much space there is, so limit
2114                          * to current size - in sectors.
2115                          */
2116                         bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2117                         if (!bitmap->mddev->bitmap_info.external)
2118                                 bytes += sizeof(bitmap_super_t);
2119                         space = DIV_ROUND_UP(bytes, 512);
2120                         bitmap->mddev->bitmap_info.space = space;
2121                 }
2122                 chunkshift = bitmap->counts.chunkshift;
2123                 chunkshift--;
2124                 do {
2125                         /* 'chunkshift' is shift from block size to chunk size */
2126                         chunkshift++;
2127                         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2128                         bytes = DIV_ROUND_UP(chunks, 8);
2129                         if (!bitmap->mddev->bitmap_info.external)
2130                                 bytes += sizeof(bitmap_super_t);
2131                 } while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2132                         (BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2133         } else
2134                 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2135
2136         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2137         memset(&store, 0, sizeof(store));
2138         if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2139                 ret = md_bitmap_storage_alloc(&store, chunks,
2140                                               !bitmap->mddev->bitmap_info.external,
2141                                               mddev_is_clustered(bitmap->mddev)
2142                                               ? bitmap->cluster_slot : 0);
2143         if (ret) {
2144                 md_bitmap_file_unmap(&store);
2145                 goto err;
2146         }
2147
2148         pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2149
2150         new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2151         ret = -ENOMEM;
2152         if (!new_bp) {
2153                 md_bitmap_file_unmap(&store);
2154                 goto err;
2155         }
2156
2157         if (!init)
2158                 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2159
2160         store.file = bitmap->storage.file;
2161         bitmap->storage.file = NULL;
2162
2163         if (store.sb_page && bitmap->storage.sb_page)
2164                 memcpy(page_address(store.sb_page),
2165                        page_address(bitmap->storage.sb_page),
2166                        sizeof(bitmap_super_t));
2167         spin_lock_irq(&bitmap->counts.lock);
2168         md_bitmap_file_unmap(&bitmap->storage);
2169         bitmap->storage = store;
2170
2171         old_counts = bitmap->counts;
2172         bitmap->counts.bp = new_bp;
2173         bitmap->counts.pages = pages;
2174         bitmap->counts.missing_pages = pages;
2175         bitmap->counts.chunkshift = chunkshift;
2176         bitmap->counts.chunks = chunks;
2177         bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2178                                                      BITMAP_BLOCK_SHIFT);
2179
2180         blocks = min(old_counts.chunks << old_counts.chunkshift,
2181                      chunks << chunkshift);
2182
2183         /* For cluster raid, need to pre-allocate bitmap */
2184         if (mddev_is_clustered(bitmap->mddev)) {
2185                 unsigned long page;
2186                 for (page = 0; page < pages; page++) {
2187                         ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2188                         if (ret) {
2189                                 unsigned long k;
2190
2191                                 /* deallocate the page memory */
2192                                 for (k = 0; k < page; k++) {
2193                                         kfree(new_bp[k].map);
2194                                 }
2195                                 kfree(new_bp);
2196
2197                                 /* restore some fields from old_counts */
2198                                 bitmap->counts.bp = old_counts.bp;
2199                                 bitmap->counts.pages = old_counts.pages;
2200                                 bitmap->counts.missing_pages = old_counts.pages;
2201                                 bitmap->counts.chunkshift = old_counts.chunkshift;
2202                                 bitmap->counts.chunks = old_counts.chunks;
2203                                 bitmap->mddev->bitmap_info.chunksize =
2204                                         1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2205                                 blocks = old_counts.chunks << old_counts.chunkshift;
2206                                 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2207                                 break;
2208                         } else
2209                                 bitmap->counts.bp[page].count += 1;
2210                 }
2211         }
2212
2213         for (block = 0; block < blocks; ) {
2214                 bitmap_counter_t *bmc_old, *bmc_new;
2215                 int set;
2216
2217                 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2218                 set = bmc_old && NEEDED(*bmc_old);
2219
2220                 if (set) {
2221                         bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2222                         if (bmc_new) {
2223                                 if (*bmc_new == 0) {
2224                                         /* need to set on-disk bits too. */
2225                                         sector_t end = block + new_blocks;
2226                                         sector_t start = block >> chunkshift;
2227
2228                                         start <<= chunkshift;
2229                                         while (start < end) {
2230                                                 md_bitmap_file_set_bit(bitmap, block);
2231                                                 start += 1 << chunkshift;
2232                                         }
2233                                         *bmc_new = 2;
2234                                         md_bitmap_count_page(&bitmap->counts, block, 1);
2235                                         md_bitmap_set_pending(&bitmap->counts, block);
2236                                 }
2237                                 *bmc_new |= NEEDED_MASK;
2238                         }
2239                         if (new_blocks < old_blocks)
2240                                 old_blocks = new_blocks;
2241                 }
2242                 block += old_blocks;
2243         }
2244
2245         if (bitmap->counts.bp != old_counts.bp) {
2246                 unsigned long k;
2247                 for (k = 0; k < old_counts.pages; k++)
2248                         if (!old_counts.bp[k].hijacked)
2249                                 kfree(old_counts.bp[k].map);
2250                 kfree(old_counts.bp);
2251         }
2252
2253         if (!init) {
2254                 int i;
2255                 while (block < (chunks << chunkshift)) {
2256                         bitmap_counter_t *bmc;
2257                         bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2258                         if (bmc) {
2259                                 /* new space.  It needs to be resynced, so
2260                                  * we set NEEDED_MASK.
2261                                  */
2262                                 if (*bmc == 0) {
2263                                         *bmc = NEEDED_MASK | 2;
2264                                         md_bitmap_count_page(&bitmap->counts, block, 1);
2265                                         md_bitmap_set_pending(&bitmap->counts, block);
2266                                 }
2267                         }
2268                         block += new_blocks;
2269                 }
2270                 for (i = 0; i < bitmap->storage.file_pages; i++)
2271                         set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2272         }
2273         spin_unlock_irq(&bitmap->counts.lock);
2274
2275         if (!init) {
2276                 md_bitmap_unplug(bitmap);
2277                 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2278         }
2279         ret = 0;
2280 err:
2281         return ret;
2282 }
2283 EXPORT_SYMBOL_GPL(md_bitmap_resize);
2284
2285 static ssize_t
2286 location_show(struct mddev *mddev, char *page)
2287 {
2288         ssize_t len;
2289         if (mddev->bitmap_info.file)
2290                 len = sprintf(page, "file");
2291         else if (mddev->bitmap_info.offset)
2292                 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2293         else
2294                 len = sprintf(page, "none");
2295         len += sprintf(page+len, "\n");
2296         return len;
2297 }
2298
2299 static ssize_t
2300 location_store(struct mddev *mddev, const char *buf, size_t len)
2301 {
2302         int rv;
2303
2304         rv = mddev_lock(mddev);
2305         if (rv)
2306                 return rv;
2307         if (mddev->pers) {
2308                 if (!mddev->pers->quiesce) {
2309                         rv = -EBUSY;
2310                         goto out;
2311                 }
2312                 if (mddev->recovery || mddev->sync_thread) {
2313                         rv = -EBUSY;
2314                         goto out;
2315                 }
2316         }
2317
2318         if (mddev->bitmap || mddev->bitmap_info.file ||
2319             mddev->bitmap_info.offset) {
2320                 /* bitmap already configured.  Only option is to clear it */
2321                 if (strncmp(buf, "none", 4) != 0) {
2322                         rv = -EBUSY;
2323                         goto out;
2324                 }
2325                 if (mddev->pers) {
2326                         mddev_suspend(mddev);
2327                         md_bitmap_destroy(mddev);
2328                         mddev_resume(mddev);
2329                 }
2330                 mddev->bitmap_info.offset = 0;
2331                 if (mddev->bitmap_info.file) {
2332                         struct file *f = mddev->bitmap_info.file;
2333                         mddev->bitmap_info.file = NULL;
2334                         fput(f);
2335                 }
2336         } else {
2337                 /* No bitmap, OK to set a location */
2338                 long long offset;
2339                 if (strncmp(buf, "none", 4) == 0)
2340                         /* nothing to be done */;
2341                 else if (strncmp(buf, "file:", 5) == 0) {
2342                         /* Not supported yet */
2343                         rv = -EINVAL;
2344                         goto out;
2345                 } else {
2346                         if (buf[0] == '+')
2347                                 rv = kstrtoll(buf+1, 10, &offset);
2348                         else
2349                                 rv = kstrtoll(buf, 10, &offset);
2350                         if (rv)
2351                                 goto out;
2352                         if (offset == 0) {
2353                                 rv = -EINVAL;
2354                                 goto out;
2355                         }
2356                         if (mddev->bitmap_info.external == 0 &&
2357                             mddev->major_version == 0 &&
2358                             offset != mddev->bitmap_info.default_offset) {
2359                                 rv = -EINVAL;
2360                                 goto out;
2361                         }
2362                         mddev->bitmap_info.offset = offset;
2363                         if (mddev->pers) {
2364                                 struct bitmap *bitmap;
2365                                 bitmap = md_bitmap_create(mddev, -1);
2366                                 mddev_suspend(mddev);
2367                                 if (IS_ERR(bitmap))
2368                                         rv = PTR_ERR(bitmap);
2369                                 else {
2370                                         mddev->bitmap = bitmap;
2371                                         rv = md_bitmap_load(mddev);
2372                                         if (rv)
2373                                                 mddev->bitmap_info.offset = 0;
2374                                 }
2375                                 if (rv) {
2376                                         md_bitmap_destroy(mddev);
2377                                         mddev_resume(mddev);
2378                                         goto out;
2379                                 }
2380                                 mddev_resume(mddev);
2381                         }
2382                 }
2383         }
2384         if (!mddev->external) {
2385                 /* Ensure new bitmap info is stored in
2386                  * metadata promptly.
2387                  */
2388                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2389                 md_wakeup_thread(mddev->thread);
2390         }
2391         rv = 0;
2392 out:
2393         mddev_unlock(mddev);
2394         if (rv)
2395                 return rv;
2396         return len;
2397 }
2398
2399 static struct md_sysfs_entry bitmap_location =
2400 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2401
2402 /* 'bitmap/space' is the space available at 'location' for the
2403  * bitmap.  This allows the kernel to know when it is safe to
2404  * resize the bitmap to match a resized array.
2405  */
2406 static ssize_t
2407 space_show(struct mddev *mddev, char *page)
2408 {
2409         return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2410 }
2411
2412 static ssize_t
2413 space_store(struct mddev *mddev, const char *buf, size_t len)
2414 {
2415         unsigned long sectors;
2416         int rv;
2417
2418         rv = kstrtoul(buf, 10, &sectors);
2419         if (rv)
2420                 return rv;
2421
2422         if (sectors == 0)
2423                 return -EINVAL;
2424
2425         if (mddev->bitmap &&
2426             sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2427                 return -EFBIG; /* Bitmap is too big for this small space */
2428
2429         /* could make sure it isn't too big, but that isn't really
2430          * needed - user-space should be careful.
2431          */
2432         mddev->bitmap_info.space = sectors;
2433         return len;
2434 }
2435
2436 static struct md_sysfs_entry bitmap_space =
2437 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2438
2439 static ssize_t
2440 timeout_show(struct mddev *mddev, char *page)
2441 {
2442         ssize_t len;
2443         unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2444         unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2445
2446         len = sprintf(page, "%lu", secs);
2447         if (jifs)
2448                 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2449         len += sprintf(page+len, "\n");
2450         return len;
2451 }
2452
2453 static ssize_t
2454 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2455 {
2456         /* timeout can be set at any time */
2457         unsigned long timeout;
2458         int rv = strict_strtoul_scaled(buf, &timeout, 4);
2459         if (rv)
2460                 return rv;
2461
2462         /* just to make sure we don't overflow... */
2463         if (timeout >= LONG_MAX / HZ)
2464                 return -EINVAL;
2465
2466         timeout = timeout * HZ / 10000;
2467
2468         if (timeout >= MAX_SCHEDULE_TIMEOUT)
2469                 timeout = MAX_SCHEDULE_TIMEOUT-1;
2470         if (timeout < 1)
2471                 timeout = 1;
2472         mddev->bitmap_info.daemon_sleep = timeout;
2473         if (mddev->thread) {
2474                 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2475                  * the bitmap is all clean and we don't need to
2476                  * adjust the timeout right now
2477                  */
2478                 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2479                         mddev->thread->timeout = timeout;
2480                         md_wakeup_thread(mddev->thread);
2481                 }
2482         }
2483         return len;
2484 }
2485
2486 static struct md_sysfs_entry bitmap_timeout =
2487 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2488
2489 static ssize_t
2490 backlog_show(struct mddev *mddev, char *page)
2491 {
2492         return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2493 }
2494
2495 static ssize_t
2496 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2497 {
2498         unsigned long backlog;
2499         unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2500         struct md_rdev *rdev;
2501         bool has_write_mostly = false;
2502         int rv = kstrtoul(buf, 10, &backlog);
2503         if (rv)
2504                 return rv;
2505         if (backlog > COUNTER_MAX)
2506                 return -EINVAL;
2507
2508         /*
2509          * Without write mostly device, it doesn't make sense to set
2510          * backlog for max_write_behind.
2511          */
2512         rdev_for_each(rdev, mddev) {
2513                 if (test_bit(WriteMostly, &rdev->flags)) {
2514                         has_write_mostly = true;
2515                         break;
2516                 }
2517         }
2518         if (!has_write_mostly) {
2519                 pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2520                                     mdname(mddev));
2521                 return -EINVAL;
2522         }
2523
2524         mddev->bitmap_info.max_write_behind = backlog;
2525         if (!backlog && mddev->serial_info_pool) {
2526                 /* serial_info_pool is not needed if backlog is zero */
2527                 if (!mddev->serialize_policy)
2528                         mddev_destroy_serial_pool(mddev, NULL, false);
2529         } else if (backlog && !mddev->serial_info_pool) {
2530                 /* serial_info_pool is needed since backlog is not zero */
2531                 struct md_rdev *rdev;
2532
2533                 rdev_for_each(rdev, mddev)
2534                         mddev_create_serial_pool(mddev, rdev, false);
2535         }
2536         if (old_mwb != backlog)
2537                 md_bitmap_update_sb(mddev->bitmap);
2538         return len;
2539 }
2540
2541 static struct md_sysfs_entry bitmap_backlog =
2542 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2543
2544 static ssize_t
2545 chunksize_show(struct mddev *mddev, char *page)
2546 {
2547         return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2548 }
2549
2550 static ssize_t
2551 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2552 {
2553         /* Can only be changed when no bitmap is active */
2554         int rv;
2555         unsigned long csize;
2556         if (mddev->bitmap)
2557                 return -EBUSY;
2558         rv = kstrtoul(buf, 10, &csize);
2559         if (rv)
2560                 return rv;
2561         if (csize < 512 ||
2562             !is_power_of_2(csize))
2563                 return -EINVAL;
2564         if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2565                 sizeof(((bitmap_super_t *)0)->chunksize))))
2566                 return -EOVERFLOW;
2567         mddev->bitmap_info.chunksize = csize;
2568         return len;
2569 }
2570
2571 static struct md_sysfs_entry bitmap_chunksize =
2572 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2573
2574 static ssize_t metadata_show(struct mddev *mddev, char *page)
2575 {
2576         if (mddev_is_clustered(mddev))
2577                 return sprintf(page, "clustered\n");
2578         return sprintf(page, "%s\n", (mddev->bitmap_info.external
2579                                       ? "external" : "internal"));
2580 }
2581
2582 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2583 {
2584         if (mddev->bitmap ||
2585             mddev->bitmap_info.file ||
2586             mddev->bitmap_info.offset)
2587                 return -EBUSY;
2588         if (strncmp(buf, "external", 8) == 0)
2589                 mddev->bitmap_info.external = 1;
2590         else if ((strncmp(buf, "internal", 8) == 0) ||
2591                         (strncmp(buf, "clustered", 9) == 0))
2592                 mddev->bitmap_info.external = 0;
2593         else
2594                 return -EINVAL;
2595         return len;
2596 }
2597
2598 static struct md_sysfs_entry bitmap_metadata =
2599 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2600
2601 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2602 {
2603         int len;
2604         spin_lock(&mddev->lock);
2605         if (mddev->bitmap)
2606                 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2607                                              "false" : "true"));
2608         else
2609                 len = sprintf(page, "\n");
2610         spin_unlock(&mddev->lock);
2611         return len;
2612 }
2613
2614 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2615 {
2616         if (mddev->bitmap == NULL)
2617                 return -ENOENT;
2618         if (strncmp(buf, "false", 5) == 0)
2619                 mddev->bitmap->need_sync = 1;
2620         else if (strncmp(buf, "true", 4) == 0) {
2621                 if (mddev->degraded)
2622                         return -EBUSY;
2623                 mddev->bitmap->need_sync = 0;
2624         } else
2625                 return -EINVAL;
2626         return len;
2627 }
2628
2629 static struct md_sysfs_entry bitmap_can_clear =
2630 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2631
2632 static ssize_t
2633 behind_writes_used_show(struct mddev *mddev, char *page)
2634 {
2635         ssize_t ret;
2636         spin_lock(&mddev->lock);
2637         if (mddev->bitmap == NULL)
2638                 ret = sprintf(page, "0\n");
2639         else
2640                 ret = sprintf(page, "%lu\n",
2641                               mddev->bitmap->behind_writes_used);
2642         spin_unlock(&mddev->lock);
2643         return ret;
2644 }
2645
2646 static ssize_t
2647 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2648 {
2649         if (mddev->bitmap)
2650                 mddev->bitmap->behind_writes_used = 0;
2651         return len;
2652 }
2653
2654 static struct md_sysfs_entry max_backlog_used =
2655 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2656        behind_writes_used_show, behind_writes_used_reset);
2657
2658 static struct attribute *md_bitmap_attrs[] = {
2659         &bitmap_location.attr,
2660         &bitmap_space.attr,
2661         &bitmap_timeout.attr,
2662         &bitmap_backlog.attr,
2663         &bitmap_chunksize.attr,
2664         &bitmap_metadata.attr,
2665         &bitmap_can_clear.attr,
2666         &max_backlog_used.attr,
2667         NULL
2668 };
2669 const struct attribute_group md_bitmap_group = {
2670         .name = "bitmap",
2671         .attrs = md_bitmap_attrs,
2672 };