drivers/md/md-bitmap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 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         WARN_ON_ONCE(page >= bitmap->pages);
  58         if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  59                 return 0;
  60
  61         if (bitmap->bp[page].map) /* page is already allocated, just return */
  62                 return 0;
  63
  64         if (!create)
  65                 return -ENOENT;
  66
  67         /* this page has not been allocated yet */
  68
  69         spin_unlock_irq(&bitmap->lock);
  70         /* It is possible that this is being called inside a
  71          * prepare_to_wait/finish_wait loop from raid5c:make_request().
  72          * In general it is not permitted to sleep in that context as it
  73          * can cause the loop to spin freely.
  74          * That doesn't apply here as we can only reach this point
  75          * once with any loop.
  76          * When this function completes, either bp[page].map or
  77          * bp[page].hijacked.  In either case, this function will
  78          * abort before getting to this point again.  So there is
  79          * no risk of a free-spin, and so it is safe to assert
  80          * that sleeping here is allowed.
  81          */
  82         sched_annotate_sleep();
  83         mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  84         spin_lock_irq(&bitmap->lock);
  85
  86         if (mappage == NULL) {
  87                 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  88                 /* We don't support hijack for cluster raid */
  89                 if (no_hijack)
  90                         return -ENOMEM;
  91                 /* failed - set the hijacked flag so that we can use the
  92                  * pointer as a counter */
  93                 if (!bitmap->bp[page].map)
  94                         bitmap->bp[page].hijacked = 1;
  95         } else if (bitmap->bp[page].map ||
  96                    bitmap->bp[page].hijacked) {
  97                 /* somebody beat us to getting the page */
  98                 kfree(mappage);
  99         } else {
 100
 101                 /* no page was in place and we have one, so install it */
 102
 103                 bitmap->bp[page].map = mappage;
 104                 bitmap->missing_pages--;
 105         }
 106         return 0;
 107 }
 108
 109 /* if page is completely empty, put it back on the free list, or dealloc it */
 110 /* if page was hijacked, unmark the flag so it might get alloced next time */
 111 /* Note: lock should be held when calling this */
 112 static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 113 {
 114         char *ptr;
 115
 116         if (bitmap->bp[page].count) /* page is still busy */
 117                 return;
 118
 119         /* page is no longer in use, it can be released */
 120
 121         if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 122                 bitmap->bp[page].hijacked = 0;
 123                 bitmap->bp[page].map = NULL;
 124         } else {
 125                 /* normal case, free the page */
 126                 ptr = bitmap->bp[page].map;
 127                 bitmap->bp[page].map = NULL;
 128                 bitmap->missing_pages++;
 129                 kfree(ptr);
 130         }
 131 }
 132
 133 /*
 134  * bitmap file handling - read and write the bitmap file and its superblock
 135  */
 136
 137 /*
 138  * basic page I/O operations
 139  */
 140
 141 /* IO operations when bitmap is stored near all superblocks */
 142 static int read_sb_page(struct mddev *mddev, loff_t offset,
 143                         struct page *page,
 144                         unsigned long index, int size)
 145 {
 146         /* choose a good rdev and read the page from there */
 147
 148         struct md_rdev *rdev;
 149         sector_t target;
 150
 151         rdev_for_each(rdev, mddev) {
 152                 if (! test_bit(In_sync, &rdev->flags)
 153                     || test_bit(Faulty, &rdev->flags)
 154                     || test_bit(Bitmap_sync, &rdev->flags))
 155                         continue;
 156
 157                 target = offset + index * (PAGE_SIZE/512);
 158
 159                 if (sync_page_io(rdev, target,
 160                                  roundup(size, bdev_logical_block_size(rdev->bdev)),
 161                                  page, REQ_OP_READ, 0, true)) {
 162                         page->index = index;
 163                         return 0;
 164                 }
 165         }
 166         return -EIO;
 167 }
 168
 169 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 170 {
 171         /* Iterate the disks of an mddev, using rcu to protect access to the
 172          * linked list, and raising the refcount of devices we return to ensure
 173          * they don't disappear while in use.
 174          * As devices are only added or removed when raid_disk is < 0 and
 175          * nr_pending is 0 and In_sync is clear, the entries we return will
 176          * still be in the same position on the list when we re-enter
 177          * list_for_each_entry_continue_rcu.
 178          *
 179          * Note that if entered with 'rdev == NULL' to start at the
 180          * beginning, we temporarily assign 'rdev' to an address which
 181          * isn't really an rdev, but which can be used by
 182          * list_for_each_entry_continue_rcu() to find the first entry.
 183          */
 184         rcu_read_lock();
 185         if (rdev == NULL)
 186                 /* start at the beginning */
 187                 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 188         else {
 189                 /* release the previous rdev and start from there. */
 190                 rdev_dec_pending(rdev, mddev);
 191         }
 192         list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 193                 if (rdev->raid_disk >= 0 &&
 194                     !test_bit(Faulty, &rdev->flags)) {
 195                         /* this is a usable devices */
 196                         atomic_inc(&rdev->nr_pending);
 197                         rcu_read_unlock();
 198                         return rdev;
 199                 }
 200         }
 201         rcu_read_unlock();
 202         return NULL;
 203 }
 204
 205 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 206 {
 207         struct md_rdev *rdev;
 208         struct block_device *bdev;
 209         struct mddev *mddev = bitmap->mddev;
 210         struct bitmap_storage *store = &bitmap->storage;
 211
 212 restart:
 213         rdev = NULL;
 214         while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 215                 int size = PAGE_SIZE;
 216                 loff_t offset = mddev->bitmap_info.offset;
 217
 218                 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 219
 220                 if (page->index == store->file_pages-1) {
 221                         int last_page_size = store->bytes & (PAGE_SIZE-1);
 222                         if (last_page_size == 0)
 223                                 last_page_size = PAGE_SIZE;
 224                         size = roundup(last_page_size,
 225                                        bdev_logical_block_size(bdev));
 226                 }
 227                 /* Just make sure we aren't corrupting data or
 228                  * metadata
 229                  */
 230                 if (mddev->external) {
 231                         /* Bitmap could be anywhere. */
 232                         if (rdev->sb_start + offset + (page->index
 233                                                        * (PAGE_SIZE/512))
 234                             > rdev->data_offset
 235                             &&
 236                             rdev->sb_start + offset
 237                             < (rdev->data_offset + mddev->dev_sectors
 238                              + (PAGE_SIZE/512)))
 239                                 goto bad_alignment;
 240                 } else if (offset < 0) {
 241                         /* DATA  BITMAP METADATA  */
 242                         if (offset
 243                             + (long)(page->index * (PAGE_SIZE/512))
 244                             + size/512 > 0)
 245                                 /* bitmap runs in to metadata */
 246                                 goto bad_alignment;
 247                         if (rdev->data_offset + mddev->dev_sectors
 248                             > rdev->sb_start + offset)
 249                                 /* data runs in to bitmap */
 250                                 goto bad_alignment;
 251                 } else if (rdev->sb_start < rdev->data_offset) {
 252                         /* METADATA BITMAP DATA */
 253                         if (rdev->sb_start
 254                             + offset
 255                             + page->index*(PAGE_SIZE/512) + size/512
 256                             > rdev->data_offset)
 257                                 /* bitmap runs in to data */
 258                                 goto bad_alignment;
 259                 } else {
 260                         /* DATA METADATA BITMAP - no problems */
 261                 }
 262                 md_super_write(mddev, rdev,
 263                                rdev->sb_start + offset
 264                                + page->index * (PAGE_SIZE/512),
 265                                size,
 266                                page);
 267         }
 268
 269         if (wait && md_super_wait(mddev) < 0)
 270                 goto restart;
 271         return 0;
 272
 273  bad_alignment:
 274         return -EINVAL;
 275 }
 276
 277 static void md_bitmap_file_kick(struct bitmap *bitmap);
 278 /*
 279  * write out a page to a file
 280  */
 281 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 282 {
 283         struct buffer_head *bh;
 284
 285         if (bitmap->storage.file == NULL) {
 286                 switch (write_sb_page(bitmap, page, wait)) {
 287                 case -EINVAL:
 288                         set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 289                 }
 290         } else {
 291
 292                 bh = page_buffers(page);
 293
 294                 while (bh && bh->b_blocknr) {
 295                         atomic_inc(&bitmap->pending_writes);
 296                         set_buffer_locked(bh);
 297                         set_buffer_mapped(bh);
 298                         submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
 299                         bh = bh->b_this_page;
 300                 }
 301
 302                 if (wait)
 303                         wait_event(bitmap->write_wait,
 304                                    atomic_read(&bitmap->pending_writes)==0);
 305         }
 306         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 307                 md_bitmap_file_kick(bitmap);
 308 }
 309
 310 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 311 {
 312         struct bitmap *bitmap = bh->b_private;
 313
 314         if (!uptodate)
 315                 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 316         if (atomic_dec_and_test(&bitmap->pending_writes))
 317                 wake_up(&bitmap->write_wait);
 318 }
 319
 320 static void free_buffers(struct page *page)
 321 {
 322         struct buffer_head *bh;
 323
 324         if (!PagePrivate(page))
 325                 return;
 326
 327         bh = page_buffers(page);
 328         while (bh) {
 329                 struct buffer_head *next = bh->b_this_page;
 330                 free_buffer_head(bh);
 331                 bh = next;
 332         }
 333         detach_page_private(page);
 334         put_page(page);
 335 }
 336
 337 /* read a page from a file.
 338  * We both read the page, and attach buffers to the page to record the
 339  * address of each block (using bmap).  These addresses will be used
 340  * to write the block later, completely bypassing the filesystem.
 341  * This usage is similar to how swap files are handled, and allows us
 342  * to write to a file with no concerns of memory allocation failing.
 343  */
 344 static int read_page(struct file *file, unsigned long index,
 345                      struct bitmap *bitmap,
 346                      unsigned long count,
 347                      struct page *page)
 348 {
 349         int ret = 0;
 350         struct inode *inode = file_inode(file);
 351         struct buffer_head *bh;
 352         sector_t block, blk_cur;
 353         unsigned long blocksize = i_blocksize(inode);
 354
 355         pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 356                  (unsigned long long)index << PAGE_SHIFT);
 357
 358         bh = alloc_page_buffers(page, blocksize, false);
 359         if (!bh) {
 360                 ret = -ENOMEM;
 361                 goto out;
 362         }
 363         attach_page_private(page, bh);
 364         blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 365         while (bh) {
 366                 block = blk_cur;
 367
 368                 if (count == 0)
 369                         bh->b_blocknr = 0;
 370                 else {
 371                         ret = bmap(inode, &block);
 372                         if (ret || !block) {
 373                                 ret = -EINVAL;
 374                                 bh->b_blocknr = 0;
 375                                 goto out;
 376                         }
 377
 378                         bh->b_blocknr = block;
 379                         bh->b_bdev = inode->i_sb->s_bdev;
 380                         if (count < blocksize)
 381                                 count = 0;
 382                         else
 383                                 count -= blocksize;
 384
 385                         bh->b_end_io = end_bitmap_write;
 386                         bh->b_private = bitmap;
 387                         atomic_inc(&bitmap->pending_writes);
 388                         set_buffer_locked(bh);
 389                         set_buffer_mapped(bh);
 390                         submit_bh(REQ_OP_READ, 0, bh);
 391                 }
 392                 blk_cur++;
 393                 bh = bh->b_this_page;
 394         }
 395         page->index = index;
 396
 397         wait_event(bitmap->write_wait,
 398                    atomic_read(&bitmap->pending_writes)==0);
 399         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 400                 ret = -EIO;
 401 out:
 402         if (ret)
 403                 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 404                        (int)PAGE_SIZE,
 405                        (unsigned long long)index << PAGE_SHIFT,
 406                        ret);
 407         return ret;
 408 }
 409
 410 /*
 411  * bitmap file superblock operations
 412  */
 413
 414 /*
 415  * md_bitmap_wait_writes() should be called before writing any bitmap
 416  * blocks, to ensure previous writes, particularly from
 417  * md_bitmap_daemon_work(), have completed.
 418  */
 419 static void md_bitmap_wait_writes(struct bitmap *bitmap)
 420 {
 421         if (bitmap->storage.file)
 422                 wait_event(bitmap->write_wait,
 423                            atomic_read(&bitmap->pending_writes)==0);
 424         else
 425                 /* Note that we ignore the return value.  The writes
 426                  * might have failed, but that would just mean that
 427                  * some bits which should be cleared haven't been,
 428                  * which is safe.  The relevant bitmap blocks will
 429                  * probably get written again, but there is no great
 430                  * loss if they aren't.
 431                  */
 432                 md_super_wait(bitmap->mddev);
 433 }
 434
 435
 436 /* update the event counter and sync the superblock to disk */
 437 void md_bitmap_update_sb(struct bitmap *bitmap)
 438 {
 439         bitmap_super_t *sb;
 440
 441         if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 442                 return;
 443         if (bitmap->mddev->bitmap_info.external)
 444                 return;
 445         if (!bitmap->storage.sb_page) /* no superblock */
 446                 return;
 447         sb = kmap_atomic(bitmap->storage.sb_page);
 448         sb->events = cpu_to_le64(bitmap->mddev->events);
 449         if (bitmap->mddev->events < bitmap->events_cleared)
 450                 /* rocking back to read-only */
 451                 bitmap->events_cleared = bitmap->mddev->events;
 452         sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 453         /*
 454          * clear BITMAP_WRITE_ERROR bit to protect against the case that
 455          * a bitmap write error occurred but the later writes succeeded.
 456          */
 457         sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 458         /* Just in case these have been changed via sysfs: */
 459         sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 460         sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 461         /* This might have been changed by a reshape */
 462         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 463         sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 464         sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 465         sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 466                                            bitmap_info.space);
 467         kunmap_atomic(sb);
 468         write_page(bitmap, bitmap->storage.sb_page, 1);
 469 }
 470 EXPORT_SYMBOL(md_bitmap_update_sb);
 471
 472 /* print out the bitmap file superblock */
 473 void md_bitmap_print_sb(struct bitmap *bitmap)
 474 {
 475         bitmap_super_t *sb;
 476
 477         if (!bitmap || !bitmap->storage.sb_page)
 478                 return;
 479         sb = kmap_atomic(bitmap->storage.sb_page);
 480         pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 481         pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 482         pr_debug("       version: %u\n", le32_to_cpu(sb->version));
 483         pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 484                  le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 485                  le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 486                  le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 487                  le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 488         pr_debug("        events: %llu\n",
 489                  (unsigned long long) le64_to_cpu(sb->events));
 490         pr_debug("events cleared: %llu\n",
 491                  (unsigned long long) le64_to_cpu(sb->events_cleared));
 492         pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 493         pr_debug("     chunksize: %u B\n", le32_to_cpu(sb->chunksize));
 494         pr_debug("  daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
 495         pr_debug("     sync size: %llu KB\n",
 496                  (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 497         pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
 498         kunmap_atomic(sb);
 499 }
 500
 501 /*
 502  * bitmap_new_disk_sb
 503  * @bitmap
 504  *
 505  * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 506  * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 507  * This function verifies 'bitmap_info' and populates the on-disk bitmap
 508  * structure, which is to be written to disk.
 509  *
 510  * Returns: 0 on success, -Exxx on error
 511  */
 512 static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 513 {
 514         bitmap_super_t *sb;
 515         unsigned long chunksize, daemon_sleep, write_behind;
 516
 517         bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 518         if (bitmap->storage.sb_page == NULL)
 519                 return -ENOMEM;
 520         bitmap->storage.sb_page->index = 0;
 521
 522         sb = kmap_atomic(bitmap->storage.sb_page);
 523
 524         sb->magic = cpu_to_le32(BITMAP_MAGIC);
 525         sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 526
 527         chunksize = bitmap->mddev->bitmap_info.chunksize;
 528         BUG_ON(!chunksize);
 529         if (!is_power_of_2(chunksize)) {
 530                 kunmap_atomic(sb);
 531                 pr_warn("bitmap chunksize not a power of 2\n");
 532                 return -EINVAL;
 533         }
 534         sb->chunksize = cpu_to_le32(chunksize);
 535
 536         daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 537         if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 538                 pr_debug("Choosing daemon_sleep default (5 sec)\n");
 539                 daemon_sleep = 5 * HZ;
 540         }
 541         sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 542         bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 543
 544         /*
 545          * FIXME: write_behind for RAID1.  If not specified, what
 546          * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 547          */
 548         write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 549         if (write_behind > COUNTER_MAX)
 550                 write_behind = COUNTER_MAX / 2;
 551         sb->write_behind = cpu_to_le32(write_behind);
 552         bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 553
 554         /* keep the array size field of the bitmap superblock up to date */
 555         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 556
 557         memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 558
 559         set_bit(BITMAP_STALE, &bitmap->flags);
 560         sb->state = cpu_to_le32(bitmap->flags);
 561         bitmap->events_cleared = bitmap->mddev->events;
 562         sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 563         bitmap->mddev->bitmap_info.nodes = 0;
 564
 565         kunmap_atomic(sb);
 566
 567         return 0;
 568 }
 569
 570 /* read the superblock from the bitmap file and initialize some bitmap fields */
 571 static int md_bitmap_read_sb(struct bitmap *bitmap)
 572 {
 573         char *reason = NULL;
 574         bitmap_super_t *sb;
 575         unsigned long chunksize, daemon_sleep, write_behind;
 576         unsigned long long events;
 577         int nodes = 0;
 578         unsigned long sectors_reserved = 0;
 579         int err = -EINVAL;
 580         struct page *sb_page;
 581         loff_t offset = bitmap->mddev->bitmap_info.offset;
 582
 583         if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 584                 chunksize = 128 * 1024 * 1024;
 585                 daemon_sleep = 5 * HZ;
 586                 write_behind = 0;
 587                 set_bit(BITMAP_STALE, &bitmap->flags);
 588                 err = 0;
 589                 goto out_no_sb;
 590         }
 591         /* page 0 is the superblock, read it... */
 592         sb_page = alloc_page(GFP_KERNEL);
 593         if (!sb_page)
 594                 return -ENOMEM;
 595         bitmap->storage.sb_page = sb_page;
 596
 597 re_read:
 598         /* If cluster_slot is set, the cluster is setup */
 599         if (bitmap->cluster_slot >= 0) {
 600                 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 601
 602                 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 603                            (bitmap->mddev->bitmap_info.chunksize >> 9));
 604                 /* bits to bytes */
 605                 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 606                 /* to 4k blocks */
 607                 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 608                 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
 609                 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 610                         bitmap->cluster_slot, offset);
 611         }
 612
 613         if (bitmap->storage.file) {
 614                 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 615                 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 616
 617                 err = read_page(bitmap->storage.file, 0,
 618                                 bitmap, bytes, sb_page);
 619         } else {
 620                 err = read_sb_page(bitmap->mddev,
 621                                    offset,
 622                                    sb_page,
 623                                    0, sizeof(bitmap_super_t));
 624         }
 625         if (err)
 626                 return err;
 627
 628         err = -EINVAL;
 629         sb = kmap_atomic(sb_page);
 630
 631         chunksize = le32_to_cpu(sb->chunksize);
 632         daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 633         write_behind = le32_to_cpu(sb->write_behind);
 634         sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 635
 636         /* verify that the bitmap-specific fields are valid */
 637         if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 638                 reason = "bad magic";
 639         else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 640                  le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 641                 reason = "unrecognized superblock version";
 642         else if (chunksize < 512)
 643                 reason = "bitmap chunksize too small";
 644         else if (!is_power_of_2(chunksize))
 645                 reason = "bitmap chunksize not a power of 2";
 646         else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 647                 reason = "daemon sleep period out of range";
 648         else if (write_behind > COUNTER_MAX)
 649                 reason = "write-behind limit out of range (0 - 16383)";
 650         if (reason) {
 651                 pr_warn("%s: invalid bitmap file superblock: %s\n",
 652                         bmname(bitmap), reason);
 653                 goto out;
 654         }
 655
 656         /*
 657          * Setup nodes/clustername only if bitmap version is
 658          * cluster-compatible
 659          */
 660         if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 661                 nodes = le32_to_cpu(sb->nodes);
 662                 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
 663                                 sb->cluster_name, 64);
 664         }
 665
 666         /* keep the array size field of the bitmap superblock up to date */
 667         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 668
 669         if (bitmap->mddev->persistent) {
 670                 /*
 671                  * We have a persistent array superblock, so compare the
 672                  * bitmap's UUID and event counter to the mddev's
 673                  */
 674                 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 675                         pr_warn("%s: bitmap superblock UUID mismatch\n",
 676                                 bmname(bitmap));
 677                         goto out;
 678                 }
 679                 events = le64_to_cpu(sb->events);
 680                 if (!nodes && (events < bitmap->mddev->events)) {
 681                         pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 682                                 bmname(bitmap), events,
 683                                 (unsigned long long) bitmap->mddev->events);
 684                         set_bit(BITMAP_STALE, &bitmap->flags);
 685                 }
 686         }
 687
 688         /* assign fields using values from superblock */
 689         bitmap->flags |= le32_to_cpu(sb->state);
 690         if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 691                 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 692         bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 693         strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 694         err = 0;
 695
 696 out:
 697         kunmap_atomic(sb);
 698         if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 699                 /* Assigning chunksize is required for "re_read" */
 700                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 701                 err = md_setup_cluster(bitmap->mddev, nodes);
 702                 if (err) {
 703                         pr_warn("%s: Could not setup cluster service (%d)\n",
 704                                 bmname(bitmap), err);
 705                         goto out_no_sb;
 706                 }
 707                 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 708                 goto re_read;
 709         }
 710
 711 out_no_sb:
 712         if (err == 0) {
 713                 if (test_bit(BITMAP_STALE, &bitmap->flags))
 714                         bitmap->events_cleared = bitmap->mddev->events;
 715                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 716                 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 717                 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 718                 bitmap->mddev->bitmap_info.nodes = nodes;
 719                 if (bitmap->mddev->bitmap_info.space == 0 ||
 720                         bitmap->mddev->bitmap_info.space > sectors_reserved)
 721                         bitmap->mddev->bitmap_info.space = sectors_reserved;
 722         } else {
 723                 md_bitmap_print_sb(bitmap);
 724                 if (bitmap->cluster_slot < 0)
 725                         md_cluster_stop(bitmap->mddev);
 726         }
 727         return err;
 728 }
 729
 730 /*
 731  * general bitmap file operations
 732  */
 733
 734 /*
 735  * on-disk bitmap:
 736  *
 737  * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 738  * file a page at a time. There's a superblock at the start of the file.
 739  */
 740 /* calculate the index of the page that contains this bit */
 741 static inline unsigned long file_page_index(struct bitmap_storage *store,
 742                                             unsigned long chunk)
 743 {
 744         if (store->sb_page)
 745                 chunk += sizeof(bitmap_super_t) << 3;
 746         return chunk >> PAGE_BIT_SHIFT;
 747 }
 748
 749 /* calculate the (bit) offset of this bit within a page */
 750 static inline unsigned long file_page_offset(struct bitmap_storage *store,
 751                                              unsigned long chunk)
 752 {
 753         if (store->sb_page)
 754                 chunk += sizeof(bitmap_super_t) << 3;
 755         return chunk & (PAGE_BITS - 1);
 756 }
 757
 758 /*
 759  * return a pointer to the page in the filemap that contains the given bit
 760  *
 761  */
 762 static inline struct page *filemap_get_page(struct bitmap_storage *store,
 763                                             unsigned long chunk)
 764 {
 765         if (file_page_index(store, chunk) >= store->file_pages)
 766                 return NULL;
 767         return store->filemap[file_page_index(store, chunk)];
 768 }
 769
 770 static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 771                                    unsigned long chunks, int with_super,
 772                                    int slot_number)
 773 {
 774         int pnum, offset = 0;
 775         unsigned long num_pages;
 776         unsigned long bytes;
 777
 778         bytes = DIV_ROUND_UP(chunks, 8);
 779         if (with_super)
 780                 bytes += sizeof(bitmap_super_t);
 781
 782         num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 783         offset = slot_number * num_pages;
 784
 785         store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 786                                        GFP_KERNEL);
 787         if (!store->filemap)
 788                 return -ENOMEM;
 789
 790         if (with_super && !store->sb_page) {
 791                 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 792                 if (store->sb_page == NULL)
 793                         return -ENOMEM;
 794         }
 795
 796         pnum = 0;
 797         if (store->sb_page) {
 798                 store->filemap[0] = store->sb_page;
 799                 pnum = 1;
 800                 store->sb_page->index = offset;
 801         }
 802
 803         for ( ; pnum < num_pages; pnum++) {
 804                 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 805                 if (!store->filemap[pnum]) {
 806                         store->file_pages = pnum;
 807                         return -ENOMEM;
 808                 }
 809                 store->filemap[pnum]->index = pnum + offset;
 810         }
 811         store->file_pages = pnum;
 812
 813         /* We need 4 bits per page, rounded up to a multiple
 814          * of sizeof(unsigned long) */
 815         store->filemap_attr = kzalloc(
 816                 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 817                 GFP_KERNEL);
 818         if (!store->filemap_attr)
 819                 return -ENOMEM;
 820
 821         store->bytes = bytes;
 822
 823         return 0;
 824 }
 825
 826 static void md_bitmap_file_unmap(struct bitmap_storage *store)
 827 {
 828         struct page **map, *sb_page;
 829         int pages;
 830         struct file *file;
 831
 832         file = store->file;
 833         map = store->filemap;
 834         pages = store->file_pages;
 835         sb_page = store->sb_page;
 836
 837         while (pages--)
 838                 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 839                         free_buffers(map[pages]);
 840         kfree(map);
 841         kfree(store->filemap_attr);
 842
 843         if (sb_page)
 844                 free_buffers(sb_page);
 845
 846         if (file) {
 847                 struct inode *inode = file_inode(file);
 848                 invalidate_mapping_pages(inode->i_mapping, 0, -1);
 849                 fput(file);
 850         }
 851 }
 852
 853 /*
 854  * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 855  * then it is no longer reliable, so we stop using it and we mark the file
 856  * as failed in the superblock
 857  */
 858 static void md_bitmap_file_kick(struct bitmap *bitmap)
 859 {
 860         char *path, *ptr = NULL;
 861
 862         if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 863                 md_bitmap_update_sb(bitmap);
 864
 865                 if (bitmap->storage.file) {
 866                         path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 867                         if (path)
 868                                 ptr = file_path(bitmap->storage.file,
 869                                              path, PAGE_SIZE);
 870
 871                         pr_warn("%s: kicking failed bitmap file %s from array!\n",
 872                                 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 873
 874                         kfree(path);
 875                 } else
 876                         pr_warn("%s: disabling internal bitmap due to errors\n",
 877                                 bmname(bitmap));
 878         }
 879 }
 880
 881 enum bitmap_page_attr {
 882         BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 883         BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 884                                     * i.e. counter is 1 or 2. */
 885         BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 886 };
 887
 888 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 889                                  enum bitmap_page_attr attr)
 890 {
 891         set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 892 }
 893
 894 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 895                                    enum bitmap_page_attr attr)
 896 {
 897         clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 898 }
 899
 900 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 901                                  enum bitmap_page_attr attr)
 902 {
 903         return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 904 }
 905
 906 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 907                                            enum bitmap_page_attr attr)
 908 {
 909         return test_and_clear_bit((pnum<<2) + attr,
 910                                   bitmap->storage.filemap_attr);
 911 }
 912 /*
 913  * bitmap_file_set_bit -- called before performing a write to the md device
 914  * to set (and eventually sync) a particular bit in the bitmap file
 915  *
 916  * we set the bit immediately, then we record the page number so that
 917  * when an unplug occurs, we can flush the dirty pages out to disk
 918  */
 919 static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 920 {
 921         unsigned long bit;
 922         struct page *page;
 923         void *kaddr;
 924         unsigned long chunk = block >> bitmap->counts.chunkshift;
 925         struct bitmap_storage *store = &bitmap->storage;
 926         unsigned long node_offset = 0;
 927
 928         if (mddev_is_clustered(bitmap->mddev))
 929                 node_offset = bitmap->cluster_slot * store->file_pages;
 930
 931         page = filemap_get_page(&bitmap->storage, chunk);
 932         if (!page)
 933                 return;
 934         bit = file_page_offset(&bitmap->storage, chunk);
 935
 936         /* set the bit */
 937         kaddr = kmap_atomic(page);
 938         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 939                 set_bit(bit, kaddr);
 940         else
 941                 set_bit_le(bit, kaddr);
 942         kunmap_atomic(kaddr);
 943         pr_debug("set file bit %lu page %lu\n", bit, page->index);
 944         /* record page number so it gets flushed to disk when unplug occurs */
 945         set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
 946 }
 947
 948 static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 949 {
 950         unsigned long bit;
 951         struct page *page;
 952         void *paddr;
 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         paddr = kmap_atomic(page);
 965         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 966                 clear_bit(bit, paddr);
 967         else
 968                 clear_bit_le(bit, paddr);
 969         kunmap_atomic(paddr);
 970         if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
 971                 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
 972                 bitmap->allclean = 0;
 973         }
 974 }
 975
 976 static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
 977 {
 978         unsigned long bit;
 979         struct page *page;
 980         void *paddr;
 981         unsigned long chunk = block >> bitmap->counts.chunkshift;
 982         int set = 0;
 983
 984         page = filemap_get_page(&bitmap->storage, chunk);
 985         if (!page)
 986                 return -EINVAL;
 987         bit = file_page_offset(&bitmap->storage, chunk);
 988         paddr = kmap_atomic(page);
 989         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 990                 set = test_bit(bit, paddr);
 991         else
 992                 set = test_bit_le(bit, paddr);
 993         kunmap_atomic(paddr);
 994         return set;
 995 }
 996
 997
 998 /* this gets called when the md device is ready to unplug its underlying
 999  * (slave) device queues -- before we let any writes go down, we need to
1000  * sync the dirty pages of the bitmap file to disk */
1001 void md_bitmap_unplug(struct bitmap *bitmap)
1002 {
1003         unsigned long i;
1004         int dirty, need_write;
1005         int writing = 0;
1006
1007         if (!bitmap || !bitmap->storage.filemap ||
1008             test_bit(BITMAP_STALE, &bitmap->flags))
1009                 return;
1010
1011         /* look at each page to see if there are any set bits that need to be
1012          * flushed out to disk */
1013         for (i = 0; i < bitmap->storage.file_pages; i++) {
1014                 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1015                 need_write = test_and_clear_page_attr(bitmap, i,
1016                                                       BITMAP_PAGE_NEEDWRITE);
1017                 if (dirty || need_write) {
1018                         if (!writing) {
1019                                 md_bitmap_wait_writes(bitmap);
1020                                 if (bitmap->mddev->queue)
1021                                         blk_add_trace_msg(bitmap->mddev->queue,
1022                                                           "md bitmap_unplug");
1023                         }
1024                         clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1025                         write_page(bitmap, bitmap->storage.filemap[i], 0);
1026                         writing = 1;
1027                 }
1028         }
1029         if (writing)
1030                 md_bitmap_wait_writes(bitmap);
1031
1032         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1033                 md_bitmap_file_kick(bitmap);
1034 }
1035 EXPORT_SYMBOL(md_bitmap_unplug);
1036
1037 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1038 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1039  * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1040  * memory mapping of the bitmap file
1041  * Special cases:
1042  *   if there's no bitmap file, or if the bitmap file had been
1043  *   previously kicked from the array, we mark all the bits as
1044  *   1's in order to cause a full resync.
1045  *
1046  * We ignore all bits for sectors that end earlier than 'start'.
1047  * This is used when reading an out-of-date bitmap...
1048  */
1049 static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1050 {
1051         unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1052         struct page *page = NULL;
1053         unsigned long bit_cnt = 0;
1054         struct file *file;
1055         unsigned long offset;
1056         int outofdate;
1057         int ret = -ENOSPC;
1058         void *paddr;
1059         struct bitmap_storage *store = &bitmap->storage;
1060
1061         chunks = bitmap->counts.chunks;
1062         file = store->file;
1063
1064         if (!file && !bitmap->mddev->bitmap_info.offset) {
1065                 /* No permanent bitmap - fill with '1s'. */
1066                 store->filemap = NULL;
1067                 store->file_pages = 0;
1068                 for (i = 0; i < chunks ; i++) {
1069                         /* if the disk bit is set, set the memory bit */
1070                         int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1071                                       >= start);
1072                         md_bitmap_set_memory_bits(bitmap,
1073                                                   (sector_t)i << bitmap->counts.chunkshift,
1074                                                   needed);
1075                 }
1076                 return 0;
1077         }
1078
1079         outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1080         if (outofdate)
1081                 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1082
1083         if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1084                 pr_warn("%s: bitmap file too short %lu < %lu\n",
1085                         bmname(bitmap),
1086                         (unsigned long) i_size_read(file->f_mapping->host),
1087                         store->bytes);
1088                 goto err;
1089         }
1090
1091         oldindex = ~0L;
1092         offset = 0;
1093         if (!bitmap->mddev->bitmap_info.external)
1094                 offset = sizeof(bitmap_super_t);
1095
1096         if (mddev_is_clustered(bitmap->mddev))
1097                 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1098
1099         for (i = 0; i < chunks; i++) {
1100                 int b;
1101                 index = file_page_index(&bitmap->storage, i);
1102                 bit = file_page_offset(&bitmap->storage, i);
1103                 if (index != oldindex) { /* this is a new page, read it in */
1104                         int count;
1105                         /* unmap the old page, we're done with it */
1106                         if (index == store->file_pages-1)
1107                                 count = store->bytes - index * PAGE_SIZE;
1108                         else
1109                                 count = PAGE_SIZE;
1110                         page = store->filemap[index];
1111                         if (file)
1112                                 ret = read_page(file, index, bitmap,
1113                                                 count, page);
1114                         else
1115                                 ret = read_sb_page(
1116                                         bitmap->mddev,
1117                                         bitmap->mddev->bitmap_info.offset,
1118                                         page,
1119                                         index + node_offset, count);
1120
1121                         if (ret)
1122                                 goto err;
1123
1124                         oldindex = index;
1125
1126                         if (outofdate) {
1127                                 /*
1128                                  * if bitmap is out of date, dirty the
1129                                  * whole page and write it out
1130                                  */
1131                                 paddr = kmap_atomic(page);
1132                                 memset(paddr + offset, 0xff,
1133                                        PAGE_SIZE - offset);
1134                                 kunmap_atomic(paddr);
1135                                 write_page(bitmap, page, 1);
1136
1137                                 ret = -EIO;
1138                                 if (test_bit(BITMAP_WRITE_ERROR,
1139                                              &bitmap->flags))
1140                                         goto err;
1141                         }
1142                 }
1143                 paddr = kmap_atomic(page);
1144                 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1145                         b = test_bit(bit, paddr);
1146                 else
1147                         b = test_bit_le(bit, paddr);
1148                 kunmap_atomic(paddr);
1149                 if (b) {
1150                         /* if the disk bit is set, set the memory bit */
1151                         int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1152                                       >= start);
1153                         md_bitmap_set_memory_bits(bitmap,
1154                                                   (sector_t)i << bitmap->counts.chunkshift,
1155                                                   needed);
1156                         bit_cnt++;
1157                 }
1158                 offset = 0;
1159         }
1160
1161         pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1162                  bmname(bitmap), store->file_pages,
1163                  bit_cnt, chunks);
1164
1165         return 0;
1166
1167  err:
1168         pr_warn("%s: bitmap initialisation failed: %d\n",
1169                 bmname(bitmap), ret);
1170         return ret;
1171 }
1172
1173 void md_bitmap_write_all(struct bitmap *bitmap)
1174 {
1175         /* We don't actually write all bitmap blocks here,
1176          * just flag them as needing to be written
1177          */
1178         int i;
1179
1180         if (!bitmap || !bitmap->storage.filemap)
1181                 return;
1182         if (bitmap->storage.file)
1183                 /* Only one copy, so nothing needed */
1184                 return;
1185
1186         for (i = 0; i < bitmap->storage.file_pages; i++)
1187                 set_page_attr(bitmap, i,
1188                               BITMAP_PAGE_NEEDWRITE);
1189         bitmap->allclean = 0;
1190 }
1191
1192 static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1193                                  sector_t offset, int inc)
1194 {
1195         sector_t chunk = offset >> bitmap->chunkshift;
1196         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1197         bitmap->bp[page].count += inc;
1198         md_bitmap_checkfree(bitmap, page);
1199 }
1200
1201 static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1202 {
1203         sector_t chunk = offset >> bitmap->chunkshift;
1204         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1205         struct bitmap_page *bp = &bitmap->bp[page];
1206
1207         if (!bp->pending)
1208                 bp->pending = 1;
1209 }
1210
1211 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1212                                                sector_t offset, sector_t *blocks,
1213                                                int create);
1214
1215 /*
1216  * bitmap daemon -- periodically wakes up to clean bits and flush pages
1217  *                      out to disk
1218  */
1219
1220 void md_bitmap_daemon_work(struct mddev *mddev)
1221 {
1222         struct bitmap *bitmap;
1223         unsigned long j;
1224         unsigned long nextpage;
1225         sector_t blocks;
1226         struct bitmap_counts *counts;
1227
1228         /* Use a mutex to guard daemon_work against
1229          * bitmap_destroy.
1230          */
1231         mutex_lock(&mddev->bitmap_info.mutex);
1232         bitmap = mddev->bitmap;
1233         if (bitmap == NULL) {
1234                 mutex_unlock(&mddev->bitmap_info.mutex);
1235                 return;
1236         }
1237         if (time_before(jiffies, bitmap->daemon_lastrun
1238                         + mddev->bitmap_info.daemon_sleep))
1239                 goto done;
1240
1241         bitmap->daemon_lastrun = jiffies;
1242         if (bitmap->allclean) {
1243                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1244                 goto done;
1245         }
1246         bitmap->allclean = 1;
1247
1248         if (bitmap->mddev->queue)
1249                 blk_add_trace_msg(bitmap->mddev->queue,
1250                                   "md bitmap_daemon_work");
1251
1252         /* Any file-page which is PENDING now needs to be written.
1253          * So set NEEDWRITE now, then after we make any last-minute changes
1254          * we will write it.
1255          */
1256         for (j = 0; j < bitmap->storage.file_pages; j++)
1257                 if (test_and_clear_page_attr(bitmap, j,
1258                                              BITMAP_PAGE_PENDING))
1259                         set_page_attr(bitmap, j,
1260                                       BITMAP_PAGE_NEEDWRITE);
1261
1262         if (bitmap->need_sync &&
1263             mddev->bitmap_info.external == 0) {
1264                 /* Arrange for superblock update as well as
1265                  * other changes */
1266                 bitmap_super_t *sb;
1267                 bitmap->need_sync = 0;
1268                 if (bitmap->storage.filemap) {
1269                         sb = kmap_atomic(bitmap->storage.sb_page);
1270                         sb->events_cleared =
1271                                 cpu_to_le64(bitmap->events_cleared);
1272                         kunmap_atomic(sb);
1273                         set_page_attr(bitmap, 0,
1274                                       BITMAP_PAGE_NEEDWRITE);
1275                 }
1276         }
1277         /* Now look at the bitmap counters and if any are '2' or '1',
1278          * decrement and handle accordingly.
1279          */
1280         counts = &bitmap->counts;
1281         spin_lock_irq(&counts->lock);
1282         nextpage = 0;
1283         for (j = 0; j < counts->chunks; j++) {
1284                 bitmap_counter_t *bmc;
1285                 sector_t  block = (sector_t)j << counts->chunkshift;
1286
1287                 if (j == nextpage) {
1288                         nextpage += PAGE_COUNTER_RATIO;
1289                         if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1290                                 j |= PAGE_COUNTER_MASK;
1291                                 continue;
1292                         }
1293                         counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1294                 }
1295
1296                 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1297                 if (!bmc) {
1298                         j |= PAGE_COUNTER_MASK;
1299                         continue;
1300                 }
1301                 if (*bmc == 1 && !bitmap->need_sync) {
1302                         /* We can clear the bit */
1303                         *bmc = 0;
1304                         md_bitmap_count_page(counts, block, -1);
1305                         md_bitmap_file_clear_bit(bitmap, block);
1306                 } else if (*bmc && *bmc <= 2) {
1307                         *bmc = 1;
1308                         md_bitmap_set_pending(counts, block);
1309                         bitmap->allclean = 0;
1310                 }
1311         }
1312         spin_unlock_irq(&counts->lock);
1313
1314         md_bitmap_wait_writes(bitmap);
1315         /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1316          * DIRTY pages need to be written by bitmap_unplug so it can wait
1317          * for them.
1318          * If we find any DIRTY page we stop there and let bitmap_unplug
1319          * handle all the rest.  This is important in the case where
1320          * the first blocking holds the superblock and it has been updated.
1321          * We mustn't write any other blocks before the superblock.
1322          */
1323         for (j = 0;
1324              j < bitmap->storage.file_pages
1325                      && !test_bit(BITMAP_STALE, &bitmap->flags);
1326              j++) {
1327                 if (test_page_attr(bitmap, j,
1328                                    BITMAP_PAGE_DIRTY))
1329                         /* bitmap_unplug will handle the rest */
1330                         break;
1331                 if (bitmap->storage.filemap &&
1332                     test_and_clear_page_attr(bitmap, j,
1333                                              BITMAP_PAGE_NEEDWRITE)) {
1334                         write_page(bitmap, bitmap->storage.filemap[j], 0);
1335                 }
1336         }
1337
1338  done:
1339         if (bitmap->allclean == 0)
1340                 mddev->thread->timeout =
1341                         mddev->bitmap_info.daemon_sleep;
1342         mutex_unlock(&mddev->bitmap_info.mutex);
1343 }
1344
1345 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1346                                                sector_t offset, sector_t *blocks,
1347                                                int create)
1348 __releases(bitmap->lock)
1349 __acquires(bitmap->lock)
1350 {
1351         /* If 'create', we might release the lock and reclaim it.
1352          * The lock must have been taken with interrupts enabled.
1353          * If !create, we don't release the lock.
1354          */
1355         sector_t chunk = offset >> bitmap->chunkshift;
1356         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1357         unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1358         sector_t csize;
1359         int err;
1360
1361         if (page >= bitmap->pages) {
1362                 /*
1363                  * This can happen if bitmap_start_sync goes beyond
1364                  * End-of-device while looking for a whole page or
1365                  * user set a huge number to sysfs bitmap_set_bits.
1366                  */
1367                 return NULL;
1368         }
1369         err = md_bitmap_checkpage(bitmap, page, create, 0);
1370
1371         if (bitmap->bp[page].hijacked ||
1372             bitmap->bp[page].map == NULL)
1373                 csize = ((sector_t)1) << (bitmap->chunkshift +
1374                                           PAGE_COUNTER_SHIFT);
1375         else
1376                 csize = ((sector_t)1) << bitmap->chunkshift;
1377         *blocks = csize - (offset & (csize - 1));
1378
1379         if (err < 0)
1380                 return NULL;
1381
1382         /* now locked ... */
1383
1384         if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1385                 /* should we use the first or second counter field
1386                  * of the hijacked pointer? */
1387                 int hi = (pageoff > PAGE_COUNTER_MASK);
1388                 return  &((bitmap_counter_t *)
1389                           &bitmap->bp[page].map)[hi];
1390         } else /* page is allocated */
1391                 return (bitmap_counter_t *)
1392                         &(bitmap->bp[page].map[pageoff]);
1393 }
1394
1395 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1396 {
1397         if (!bitmap)
1398                 return 0;
1399
1400         if (behind) {
1401                 int bw;
1402                 atomic_inc(&bitmap->behind_writes);
1403                 bw = atomic_read(&bitmap->behind_writes);
1404                 if (bw > bitmap->behind_writes_used)
1405                         bitmap->behind_writes_used = bw;
1406
1407                 pr_debug("inc write-behind count %d/%lu\n",
1408                          bw, bitmap->mddev->bitmap_info.max_write_behind);
1409         }
1410
1411         while (sectors) {
1412                 sector_t blocks;
1413                 bitmap_counter_t *bmc;
1414
1415                 spin_lock_irq(&bitmap->counts.lock);
1416                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1417                 if (!bmc) {
1418                         spin_unlock_irq(&bitmap->counts.lock);
1419                         return 0;
1420                 }
1421
1422                 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1423                         DEFINE_WAIT(__wait);
1424                         /* note that it is safe to do the prepare_to_wait
1425                          * after the test as long as we do it before dropping
1426                          * the spinlock.
1427                          */
1428                         prepare_to_wait(&bitmap->overflow_wait, &__wait,
1429                                         TASK_UNINTERRUPTIBLE);
1430                         spin_unlock_irq(&bitmap->counts.lock);
1431                         schedule();
1432                         finish_wait(&bitmap->overflow_wait, &__wait);
1433                         continue;
1434                 }
1435
1436                 switch (*bmc) {
1437                 case 0:
1438                         md_bitmap_file_set_bit(bitmap, offset);
1439                         md_bitmap_count_page(&bitmap->counts, offset, 1);
1440                         fallthrough;
1441                 case 1:
1442                         *bmc = 2;
1443                 }
1444
1445                 (*bmc)++;
1446
1447                 spin_unlock_irq(&bitmap->counts.lock);
1448
1449                 offset += blocks;
1450                 if (sectors > blocks)
1451                         sectors -= blocks;
1452                 else
1453                         sectors = 0;
1454         }
1455         return 0;
1456 }
1457 EXPORT_SYMBOL(md_bitmap_startwrite);
1458
1459 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1460                         unsigned long sectors, int success, int behind)
1461 {
1462         if (!bitmap)
1463                 return;
1464         if (behind) {
1465                 if (atomic_dec_and_test(&bitmap->behind_writes))
1466                         wake_up(&bitmap->behind_wait);
1467                 pr_debug("dec write-behind count %d/%lu\n",
1468                          atomic_read(&bitmap->behind_writes),
1469                          bitmap->mddev->bitmap_info.max_write_behind);
1470         }
1471
1472         while (sectors) {
1473                 sector_t blocks;
1474                 unsigned long flags;
1475                 bitmap_counter_t *bmc;
1476
1477                 spin_lock_irqsave(&bitmap->counts.lock, flags);
1478                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1479                 if (!bmc) {
1480                         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1481                         return;
1482                 }
1483
1484                 if (success && !bitmap->mddev->degraded &&
1485                     bitmap->events_cleared < bitmap->mddev->events) {
1486                         bitmap->events_cleared = bitmap->mddev->events;
1487                         bitmap->need_sync = 1;
1488                         sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1489                 }
1490
1491                 if (!success && !NEEDED(*bmc))
1492                         *bmc |= NEEDED_MASK;
1493
1494                 if (COUNTER(*bmc) == COUNTER_MAX)
1495                         wake_up(&bitmap->overflow_wait);
1496
1497                 (*bmc)--;
1498                 if (*bmc <= 2) {
1499                         md_bitmap_set_pending(&bitmap->counts, offset);
1500                         bitmap->allclean = 0;
1501                 }
1502                 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1503                 offset += blocks;
1504                 if (sectors > blocks)
1505                         sectors -= blocks;
1506                 else
1507                         sectors = 0;
1508         }
1509 }
1510 EXPORT_SYMBOL(md_bitmap_endwrite);
1511
1512 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1513                                int degraded)
1514 {
1515         bitmap_counter_t *bmc;
1516         int rv;
1517         if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1518                 *blocks = 1024;
1519                 return 1; /* always resync if no bitmap */
1520         }
1521         spin_lock_irq(&bitmap->counts.lock);
1522         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1523         rv = 0;
1524         if (bmc) {
1525                 /* locked */
1526                 if (RESYNC(*bmc))
1527                         rv = 1;
1528                 else if (NEEDED(*bmc)) {
1529                         rv = 1;
1530                         if (!degraded) { /* don't set/clear bits if degraded */
1531                                 *bmc |= RESYNC_MASK;
1532                                 *bmc &= ~NEEDED_MASK;
1533                         }
1534                 }
1535         }
1536         spin_unlock_irq(&bitmap->counts.lock);
1537         return rv;
1538 }
1539
1540 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1541                          int degraded)
1542 {
1543         /* bitmap_start_sync must always report on multiples of whole
1544          * pages, otherwise resync (which is very PAGE_SIZE based) will
1545          * get confused.
1546          * So call __bitmap_start_sync repeatedly (if needed) until
1547          * At least PAGE_SIZE>>9 blocks are covered.
1548          * Return the 'or' of the result.
1549          */
1550         int rv = 0;
1551         sector_t blocks1;
1552
1553         *blocks = 0;
1554         while (*blocks < (PAGE_SIZE>>9)) {
1555                 rv |= __bitmap_start_sync(bitmap, offset,
1556                                           &blocks1, degraded);
1557                 offset += blocks1;
1558                 *blocks += blocks1;
1559         }
1560         return rv;
1561 }
1562 EXPORT_SYMBOL(md_bitmap_start_sync);
1563
1564 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1565 {
1566         bitmap_counter_t *bmc;
1567         unsigned long flags;
1568
1569         if (bitmap == NULL) {
1570                 *blocks = 1024;
1571                 return;
1572         }
1573         spin_lock_irqsave(&bitmap->counts.lock, flags);
1574         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1575         if (bmc == NULL)
1576                 goto unlock;
1577         /* locked */
1578         if (RESYNC(*bmc)) {
1579                 *bmc &= ~RESYNC_MASK;
1580
1581                 if (!NEEDED(*bmc) && aborted)
1582                         *bmc |= NEEDED_MASK;
1583                 else {
1584                         if (*bmc <= 2) {
1585                                 md_bitmap_set_pending(&bitmap->counts, offset);
1586                                 bitmap->allclean = 0;
1587                         }
1588                 }
1589         }
1590  unlock:
1591         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1592 }
1593 EXPORT_SYMBOL(md_bitmap_end_sync);
1594
1595 void md_bitmap_close_sync(struct bitmap *bitmap)
1596 {
1597         /* Sync has finished, and any bitmap chunks that weren't synced
1598          * properly have been aborted.  It remains to us to clear the
1599          * RESYNC bit wherever it is still on
1600          */
1601         sector_t sector = 0;
1602         sector_t blocks;
1603         if (!bitmap)
1604                 return;
1605         while (sector < bitmap->mddev->resync_max_sectors) {
1606                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1607                 sector += blocks;
1608         }
1609 }
1610 EXPORT_SYMBOL(md_bitmap_close_sync);
1611
1612 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1613 {
1614         sector_t s = 0;
1615         sector_t blocks;
1616
1617         if (!bitmap)
1618                 return;
1619         if (sector == 0) {
1620                 bitmap->last_end_sync = jiffies;
1621                 return;
1622         }
1623         if (!force && time_before(jiffies, (bitmap->last_end_sync
1624                                   + bitmap->mddev->bitmap_info.daemon_sleep)))
1625                 return;
1626         wait_event(bitmap->mddev->recovery_wait,
1627                    atomic_read(&bitmap->mddev->recovery_active) == 0);
1628
1629         bitmap->mddev->curr_resync_completed = sector;
1630         set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1631         sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1632         s = 0;
1633         while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1634                 md_bitmap_end_sync(bitmap, s, &blocks, 0);
1635                 s += blocks;
1636         }
1637         bitmap->last_end_sync = jiffies;
1638         sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1639 }
1640 EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1641
1642 void md_bitmap_sync_with_cluster(struct mddev *mddev,
1643                               sector_t old_lo, sector_t old_hi,
1644                               sector_t new_lo, sector_t new_hi)
1645 {
1646         struct bitmap *bitmap = mddev->bitmap;
1647         sector_t sector, blocks = 0;
1648
1649         for (sector = old_lo; sector < new_lo; ) {
1650                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1651                 sector += blocks;
1652         }
1653         WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1654
1655         for (sector = old_hi; sector < new_hi; ) {
1656                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1657                 sector += blocks;
1658         }
1659         WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1660 }
1661 EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1662
1663 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1664 {
1665         /* For each chunk covered by any of these sectors, set the
1666          * counter to 2 and possibly set resync_needed.  They should all
1667          * be 0 at this point
1668          */
1669
1670         sector_t secs;
1671         bitmap_counter_t *bmc;
1672         spin_lock_irq(&bitmap->counts.lock);
1673         bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1674         if (!bmc) {
1675                 spin_unlock_irq(&bitmap->counts.lock);
1676                 return;
1677         }
1678         if (!*bmc) {
1679                 *bmc = 2;
1680                 md_bitmap_count_page(&bitmap->counts, offset, 1);
1681                 md_bitmap_set_pending(&bitmap->counts, offset);
1682                 bitmap->allclean = 0;
1683         }
1684         if (needed)
1685                 *bmc |= NEEDED_MASK;
1686         spin_unlock_irq(&bitmap->counts.lock);
1687 }
1688
1689 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1690 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1691 {
1692         unsigned long chunk;
1693
1694         for (chunk = s; chunk <= e; chunk++) {
1695                 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1696                 md_bitmap_set_memory_bits(bitmap, sec, 1);
1697                 md_bitmap_file_set_bit(bitmap, sec);
1698                 if (sec < bitmap->mddev->recovery_cp)
1699                         /* We are asserting that the array is dirty,
1700                          * so move the recovery_cp address back so
1701                          * that it is obvious that it is dirty
1702                          */
1703                         bitmap->mddev->recovery_cp = sec;
1704         }
1705 }
1706
1707 /*
1708  * flush out any pending updates
1709  */
1710 void md_bitmap_flush(struct mddev *mddev)
1711 {
1712         struct bitmap *bitmap = mddev->bitmap;
1713         long sleep;
1714
1715         if (!bitmap) /* there was no bitmap */
1716                 return;
1717
1718         /* run the daemon_work three time to ensure everything is flushed
1719          * that can be
1720          */
1721         sleep = mddev->bitmap_info.daemon_sleep * 2;
1722         bitmap->daemon_lastrun -= sleep;
1723         md_bitmap_daemon_work(mddev);
1724         bitmap->daemon_lastrun -= sleep;
1725         md_bitmap_daemon_work(mddev);
1726         bitmap->daemon_lastrun -= sleep;
1727         md_bitmap_daemon_work(mddev);
1728         if (mddev->bitmap_info.external)
1729                 md_super_wait(mddev);
1730         md_bitmap_update_sb(bitmap);
1731 }
1732
1733 /*
1734  * free memory that was allocated
1735  */
1736 void md_bitmap_free(struct bitmap *bitmap)
1737 {
1738         unsigned long k, pages;
1739         struct bitmap_page *bp;
1740
1741         if (!bitmap) /* there was no bitmap */
1742                 return;
1743
1744         if (bitmap->sysfs_can_clear)
1745                 sysfs_put(bitmap->sysfs_can_clear);
1746
1747         if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1748                 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1749                 md_cluster_stop(bitmap->mddev);
1750
1751         /* Shouldn't be needed - but just in case.... */
1752         wait_event(bitmap->write_wait,
1753                    atomic_read(&bitmap->pending_writes) == 0);
1754
1755         /* release the bitmap file  */
1756         md_bitmap_file_unmap(&bitmap->storage);
1757
1758         bp = bitmap->counts.bp;
1759         pages = bitmap->counts.pages;
1760
1761         /* free all allocated memory */
1762
1763         if (bp) /* deallocate the page memory */
1764                 for (k = 0; k < pages; k++)
1765                         if (bp[k].map && !bp[k].hijacked)
1766                                 kfree(bp[k].map);
1767         kfree(bp);
1768         kfree(bitmap);
1769 }
1770 EXPORT_SYMBOL(md_bitmap_free);
1771
1772 void md_bitmap_wait_behind_writes(struct mddev *mddev)
1773 {
1774         struct bitmap *bitmap = mddev->bitmap;
1775
1776         /* wait for behind writes to complete */
1777         if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1778                 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1779                          mdname(mddev));
1780                 /* need to kick something here to make sure I/O goes? */
1781                 wait_event(bitmap->behind_wait,
1782                            atomic_read(&bitmap->behind_writes) == 0);
1783         }
1784 }
1785
1786 void md_bitmap_destroy(struct mddev *mddev)
1787 {
1788         struct bitmap *bitmap = mddev->bitmap;
1789
1790         if (!bitmap) /* there was no bitmap */
1791                 return;
1792
1793         md_bitmap_wait_behind_writes(mddev);
1794         if (!mddev->serialize_policy)
1795                 mddev_destroy_serial_pool(mddev, NULL, true);
1796
1797         mutex_lock(&mddev->bitmap_info.mutex);
1798         spin_lock(&mddev->lock);
1799         mddev->bitmap = NULL; /* disconnect from the md device */
1800         spin_unlock(&mddev->lock);
1801         mutex_unlock(&mddev->bitmap_info.mutex);
1802         if (mddev->thread)
1803                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1804
1805         md_bitmap_free(bitmap);
1806 }
1807
1808 /*
1809  * initialize the bitmap structure
1810  * if this returns an error, bitmap_destroy must be called to do clean up
1811  * once mddev->bitmap is set
1812  */
1813 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1814 {
1815         struct bitmap *bitmap;
1816         sector_t blocks = mddev->resync_max_sectors;
1817         struct file *file = mddev->bitmap_info.file;
1818         int err;
1819         struct kernfs_node *bm = NULL;
1820
1821         BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1822
1823         BUG_ON(file && mddev->bitmap_info.offset);
1824
1825         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1826                 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1827                           mdname(mddev));
1828                 return ERR_PTR(-EBUSY);
1829         }
1830
1831         bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1832         if (!bitmap)
1833                 return ERR_PTR(-ENOMEM);
1834
1835         spin_lock_init(&bitmap->counts.lock);
1836         atomic_set(&bitmap->pending_writes, 0);
1837         init_waitqueue_head(&bitmap->write_wait);
1838         init_waitqueue_head(&bitmap->overflow_wait);
1839         init_waitqueue_head(&bitmap->behind_wait);
1840
1841         bitmap->mddev = mddev;
1842         bitmap->cluster_slot = slot;
1843
1844         if (mddev->kobj.sd)
1845                 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1846         if (bm) {
1847                 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1848                 sysfs_put(bm);
1849         } else
1850                 bitmap->sysfs_can_clear = NULL;
1851
1852         bitmap->storage.file = file;
1853         if (file) {
1854                 get_file(file);
1855                 /* As future accesses to this file will use bmap,
1856                  * and bypass the page cache, we must sync the file
1857                  * first.
1858                  */
1859                 vfs_fsync(file, 1);
1860         }
1861         /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1862         if (!mddev->bitmap_info.external) {
1863                 /*
1864                  * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1865                  * instructing us to create a new on-disk bitmap instance.
1866                  */
1867                 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1868                         err = md_bitmap_new_disk_sb(bitmap);
1869                 else
1870                         err = md_bitmap_read_sb(bitmap);
1871         } else {
1872                 err = 0;
1873                 if (mddev->bitmap_info.chunksize == 0 ||
1874                     mddev->bitmap_info.daemon_sleep == 0)
1875                         /* chunksize and time_base need to be
1876                          * set first. */
1877                         err = -EINVAL;
1878         }
1879         if (err)
1880                 goto error;
1881
1882         bitmap->daemon_lastrun = jiffies;
1883         err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1884         if (err)
1885                 goto error;
1886
1887         pr_debug("created bitmap (%lu pages) for device %s\n",
1888                  bitmap->counts.pages, bmname(bitmap));
1889
1890         err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1891         if (err)
1892                 goto error;
1893
1894         return bitmap;
1895  error:
1896         md_bitmap_free(bitmap);
1897         return ERR_PTR(err);
1898 }
1899
1900 int md_bitmap_load(struct mddev *mddev)
1901 {
1902         int err = 0;
1903         sector_t start = 0;
1904         sector_t sector = 0;
1905         struct bitmap *bitmap = mddev->bitmap;
1906         struct md_rdev *rdev;
1907
1908         if (!bitmap)
1909                 goto out;
1910
1911         rdev_for_each(rdev, mddev)
1912                 mddev_create_serial_pool(mddev, rdev, true);
1913
1914         if (mddev_is_clustered(mddev))
1915                 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1916
1917         /* Clear out old bitmap info first:  Either there is none, or we
1918          * are resuming after someone else has possibly changed things,
1919          * so we should forget old cached info.
1920          * All chunks should be clean, but some might need_sync.
1921          */
1922         while (sector < mddev->resync_max_sectors) {
1923                 sector_t blocks;
1924                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1925                 sector += blocks;
1926         }
1927         md_bitmap_close_sync(bitmap);
1928
1929         if (mddev->degraded == 0
1930             || bitmap->events_cleared == mddev->events)
1931                 /* no need to keep dirty bits to optimise a
1932                  * re-add of a missing device */
1933                 start = mddev->recovery_cp;
1934
1935         mutex_lock(&mddev->bitmap_info.mutex);
1936         err = md_bitmap_init_from_disk(bitmap, start);
1937         mutex_unlock(&mddev->bitmap_info.mutex);
1938
1939         if (err)
1940                 goto out;
1941         clear_bit(BITMAP_STALE, &bitmap->flags);
1942
1943         /* Kick recovery in case any bits were set */
1944         set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1945
1946         mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1947         md_wakeup_thread(mddev->thread);
1948
1949         md_bitmap_update_sb(bitmap);
1950
1951         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1952                 err = -EIO;
1953 out:
1954         return err;
1955 }
1956 EXPORT_SYMBOL_GPL(md_bitmap_load);
1957
1958 /* caller need to free returned bitmap with md_bitmap_free() */
1959 struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1960 {
1961         int rv = 0;
1962         struct bitmap *bitmap;
1963
1964         bitmap = md_bitmap_create(mddev, slot);
1965         if (IS_ERR(bitmap)) {
1966                 rv = PTR_ERR(bitmap);
1967                 return ERR_PTR(rv);
1968         }
1969
1970         rv = md_bitmap_init_from_disk(bitmap, 0);
1971         if (rv) {
1972                 md_bitmap_free(bitmap);
1973                 return ERR_PTR(rv);
1974         }
1975
1976         return bitmap;
1977 }
1978 EXPORT_SYMBOL(get_bitmap_from_slot);
1979
1980 /* Loads the bitmap associated with slot and copies the resync information
1981  * to our bitmap
1982  */
1983 int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
1984                 sector_t *low, sector_t *high, bool clear_bits)
1985 {
1986         int rv = 0, i, j;
1987         sector_t block, lo = 0, hi = 0;
1988         struct bitmap_counts *counts;
1989         struct bitmap *bitmap;
1990
1991         bitmap = get_bitmap_from_slot(mddev, slot);
1992         if (IS_ERR(bitmap)) {
1993                 pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
1994                 return -1;
1995         }
1996
1997         counts = &bitmap->counts;
1998         for (j = 0; j < counts->chunks; j++) {
1999                 block = (sector_t)j << counts->chunkshift;
2000                 if (md_bitmap_file_test_bit(bitmap, block)) {
2001                         if (!lo)
2002                                 lo = block;
2003                         hi = block;
2004                         md_bitmap_file_clear_bit(bitmap, block);
2005                         md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2006                         md_bitmap_file_set_bit(mddev->bitmap, block);
2007                 }
2008         }
2009
2010         if (clear_bits) {
2011                 md_bitmap_update_sb(bitmap);
2012                 /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2013                  * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2014                 for (i = 0; i < bitmap->storage.file_pages; i++)
2015                         if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2016                                 set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2017                 md_bitmap_unplug(bitmap);
2018         }
2019         md_bitmap_unplug(mddev->bitmap);
2020         *low = lo;
2021         *high = hi;
2022         md_bitmap_free(bitmap);
2023
2024         return rv;
2025 }
2026 EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2027
2028
2029 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2030 {
2031         unsigned long chunk_kb;
2032         struct bitmap_counts *counts;
2033
2034         if (!bitmap)
2035                 return;
2036
2037         counts = &bitmap->counts;
2038
2039         chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2040         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2041                    "%lu%s chunk",
2042                    counts->pages - counts->missing_pages,
2043                    counts->pages,
2044                    (counts->pages - counts->missing_pages)
2045                    << (PAGE_SHIFT - 10),
2046                    chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2047                    chunk_kb ? "KB" : "B");
2048         if (bitmap->storage.file) {
2049                 seq_printf(seq, ", file: ");
2050                 seq_file_path(seq, bitmap->storage.file, " \t\n");
2051         }
2052
2053         seq_printf(seq, "\n");
2054 }
2055
2056 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2057                   int chunksize, int init)
2058 {
2059         /* If chunk_size is 0, choose an appropriate chunk size.
2060          * Then possibly allocate new storage space.
2061          * Then quiesce, copy bits, replace bitmap, and re-start
2062          *
2063          * This function is called both to set up the initial bitmap
2064          * and to resize the bitmap while the array is active.
2065          * If this happens as a result of the array being resized,
2066          * chunksize will be zero, and we need to choose a suitable
2067          * chunksize, otherwise we use what we are given.
2068          */
2069         struct bitmap_storage store;
2070         struct bitmap_counts old_counts;
2071         unsigned long chunks;
2072         sector_t block;
2073         sector_t old_blocks, new_blocks;
2074         int chunkshift;
2075         int ret = 0;
2076         long pages;
2077         struct bitmap_page *new_bp;
2078
2079         if (bitmap->storage.file && !init) {
2080                 pr_info("md: cannot resize file-based bitmap\n");
2081                 return -EINVAL;
2082         }
2083
2084         if (chunksize == 0) {
2085                 /* If there is enough space, leave the chunk size unchanged,
2086                  * else increase by factor of two until there is enough space.
2087                  */
2088                 long bytes;
2089                 long space = bitmap->mddev->bitmap_info.space;
2090
2091                 if (space == 0) {
2092                         /* We don't know how much space there is, so limit
2093                          * to current size - in sectors.
2094                          */
2095                         bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2096                         if (!bitmap->mddev->bitmap_info.external)
2097                                 bytes += sizeof(bitmap_super_t);
2098                         space = DIV_ROUND_UP(bytes, 512);
2099                         bitmap->mddev->bitmap_info.space = space;
2100                 }
2101                 chunkshift = bitmap->counts.chunkshift;
2102                 chunkshift--;
2103                 do {
2104                         /* 'chunkshift' is shift from block size to chunk size */
2105                         chunkshift++;
2106                         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2107                         bytes = DIV_ROUND_UP(chunks, 8);
2108                         if (!bitmap->mddev->bitmap_info.external)
2109                                 bytes += sizeof(bitmap_super_t);
2110                 } while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2111                         (BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2112         } else
2113                 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2114
2115         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2116         memset(&store, 0, sizeof(store));
2117         if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2118                 ret = md_bitmap_storage_alloc(&store, chunks,
2119                                               !bitmap->mddev->bitmap_info.external,
2120                                               mddev_is_clustered(bitmap->mddev)
2121                                               ? bitmap->cluster_slot : 0);
2122         if (ret) {
2123                 md_bitmap_file_unmap(&store);
2124                 goto err;
2125         }
2126
2127         pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2128
2129         new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2130         ret = -ENOMEM;
2131         if (!new_bp) {
2132                 md_bitmap_file_unmap(&store);
2133                 goto err;
2134         }
2135
2136         if (!init)
2137                 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2138
2139         store.file = bitmap->storage.file;
2140         bitmap->storage.file = NULL;
2141
2142         if (store.sb_page && bitmap->storage.sb_page)
2143                 memcpy(page_address(store.sb_page),
2144                        page_address(bitmap->storage.sb_page),
2145                        sizeof(bitmap_super_t));
2146         spin_lock_irq(&bitmap->counts.lock);
2147         md_bitmap_file_unmap(&bitmap->storage);
2148         bitmap->storage = store;
2149
2150         old_counts = bitmap->counts;
2151         bitmap->counts.bp = new_bp;
2152         bitmap->counts.pages = pages;
2153         bitmap->counts.missing_pages = pages;
2154         bitmap->counts.chunkshift = chunkshift;
2155         bitmap->counts.chunks = chunks;
2156         bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2157                                                      BITMAP_BLOCK_SHIFT);
2158
2159         blocks = min(old_counts.chunks << old_counts.chunkshift,
2160                      chunks << chunkshift);
2161
2162         /* For cluster raid, need to pre-allocate bitmap */
2163         if (mddev_is_clustered(bitmap->mddev)) {
2164                 unsigned long page;
2165                 for (page = 0; page < pages; page++) {
2166                         ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2167                         if (ret) {
2168                                 unsigned long k;
2169
2170                                 /* deallocate the page memory */
2171                                 for (k = 0; k < page; k++) {
2172                                         kfree(new_bp[k].map);
2173                                 }
2174                                 kfree(new_bp);
2175
2176                                 /* restore some fields from old_counts */
2177                                 bitmap->counts.bp = old_counts.bp;
2178                                 bitmap->counts.pages = old_counts.pages;
2179                                 bitmap->counts.missing_pages = old_counts.pages;
2180                                 bitmap->counts.chunkshift = old_counts.chunkshift;
2181                                 bitmap->counts.chunks = old_counts.chunks;
2182                                 bitmap->mddev->bitmap_info.chunksize =
2183                                         1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2184                                 blocks = old_counts.chunks << old_counts.chunkshift;
2185                                 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2186                                 break;
2187                         } else
2188                                 bitmap->counts.bp[page].count += 1;
2189                 }
2190         }
2191
2192         for (block = 0; block < blocks; ) {
2193                 bitmap_counter_t *bmc_old, *bmc_new;
2194                 int set;
2195
2196                 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2197                 set = bmc_old && NEEDED(*bmc_old);
2198
2199                 if (set) {
2200                         bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2201                         if (bmc_new) {
2202                                 if (*bmc_new == 0) {
2203                                         /* need to set on-disk bits too. */
2204                                         sector_t end = block + new_blocks;
2205                                         sector_t start = block >> chunkshift;
2206
2207                                         start <<= chunkshift;
2208                                         while (start < end) {
2209                                                 md_bitmap_file_set_bit(bitmap, block);
2210                                                 start += 1 << chunkshift;
2211                                         }
2212                                         *bmc_new = 2;
2213                                         md_bitmap_count_page(&bitmap->counts, block, 1);
2214                                         md_bitmap_set_pending(&bitmap->counts, block);
2215                                 }
2216                                 *bmc_new |= NEEDED_MASK;
2217                         }
2218                         if (new_blocks < old_blocks)
2219                                 old_blocks = new_blocks;
2220                 }
2221                 block += old_blocks;
2222         }
2223
2224         if (bitmap->counts.bp != old_counts.bp) {
2225                 unsigned long k;
2226                 for (k = 0; k < old_counts.pages; k++)
2227                         if (!old_counts.bp[k].hijacked)
2228                                 kfree(old_counts.bp[k].map);
2229                 kfree(old_counts.bp);
2230         }
2231
2232         if (!init) {
2233                 int i;
2234                 while (block < (chunks << chunkshift)) {
2235                         bitmap_counter_t *bmc;
2236                         bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2237                         if (bmc) {
2238                                 /* new space.  It needs to be resynced, so
2239                                  * we set NEEDED_MASK.
2240                                  */
2241                                 if (*bmc == 0) {
2242                                         *bmc = NEEDED_MASK | 2;
2243                                         md_bitmap_count_page(&bitmap->counts, block, 1);
2244                                         md_bitmap_set_pending(&bitmap->counts, block);
2245                                 }
2246                         }
2247                         block += new_blocks;
2248                 }
2249                 for (i = 0; i < bitmap->storage.file_pages; i++)
2250                         set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2251         }
2252         spin_unlock_irq(&bitmap->counts.lock);
2253
2254         if (!init) {
2255                 md_bitmap_unplug(bitmap);
2256                 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2257         }
2258         ret = 0;
2259 err:
2260         return ret;
2261 }
2262 EXPORT_SYMBOL_GPL(md_bitmap_resize);
2263
2264 static ssize_t
2265 location_show(struct mddev *mddev, char *page)
2266 {
2267         ssize_t len;
2268         if (mddev->bitmap_info.file)
2269                 len = sprintf(page, "file");
2270         else if (mddev->bitmap_info.offset)
2271                 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2272         else
2273                 len = sprintf(page, "none");
2274         len += sprintf(page+len, "\n");
2275         return len;
2276 }
2277
2278 static ssize_t
2279 location_store(struct mddev *mddev, const char *buf, size_t len)
2280 {
2281         int rv;
2282
2283         rv = mddev_lock(mddev);
2284         if (rv)
2285                 return rv;
2286         if (mddev->pers) {
2287                 if (!mddev->pers->quiesce) {
2288                         rv = -EBUSY;
2289                         goto out;
2290                 }
2291                 if (mddev->recovery || mddev->sync_thread) {
2292                         rv = -EBUSY;
2293                         goto out;
2294                 }
2295         }
2296
2297         if (mddev->bitmap || mddev->bitmap_info.file ||
2298             mddev->bitmap_info.offset) {
2299                 /* bitmap already configured.  Only option is to clear it */
2300                 if (strncmp(buf, "none", 4) != 0) {
2301                         rv = -EBUSY;
2302                         goto out;
2303                 }
2304                 if (mddev->pers) {
2305                         mddev_suspend(mddev);
2306                         md_bitmap_destroy(mddev);
2307                         mddev_resume(mddev);
2308                 }
2309                 mddev->bitmap_info.offset = 0;
2310                 if (mddev->bitmap_info.file) {
2311                         struct file *f = mddev->bitmap_info.file;
2312                         mddev->bitmap_info.file = NULL;
2313                         fput(f);
2314                 }
2315         } else {
2316                 /* No bitmap, OK to set a location */
2317                 long long offset;
2318                 if (strncmp(buf, "none", 4) == 0)
2319                         /* nothing to be done */;
2320                 else if (strncmp(buf, "file:", 5) == 0) {
2321                         /* Not supported yet */
2322                         rv = -EINVAL;
2323                         goto out;
2324                 } else {
2325                         if (buf[0] == '+')
2326                                 rv = kstrtoll(buf+1, 10, &offset);
2327                         else
2328                                 rv = kstrtoll(buf, 10, &offset);
2329                         if (rv)
2330                                 goto out;
2331                         if (offset == 0) {
2332                                 rv = -EINVAL;
2333                                 goto out;
2334                         }
2335                         if (mddev->bitmap_info.external == 0 &&
2336                             mddev->major_version == 0 &&
2337                             offset != mddev->bitmap_info.default_offset) {
2338                                 rv = -EINVAL;
2339                                 goto out;
2340                         }
2341                         mddev->bitmap_info.offset = offset;
2342                         if (mddev->pers) {
2343                                 struct bitmap *bitmap;
2344                                 bitmap = md_bitmap_create(mddev, -1);
2345                                 mddev_suspend(mddev);
2346                                 if (IS_ERR(bitmap))
2347                                         rv = PTR_ERR(bitmap);
2348                                 else {
2349                                         mddev->bitmap = bitmap;
2350                                         rv = md_bitmap_load(mddev);
2351                                         if (rv)
2352                                                 mddev->bitmap_info.offset = 0;
2353                                 }
2354                                 if (rv) {
2355                                         md_bitmap_destroy(mddev);
2356                                         mddev_resume(mddev);
2357                                         goto out;
2358                                 }
2359                                 mddev_resume(mddev);
2360                         }
2361                 }
2362         }
2363         if (!mddev->external) {
2364                 /* Ensure new bitmap info is stored in
2365                  * metadata promptly.
2366                  */
2367                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2368                 md_wakeup_thread(mddev->thread);
2369         }
2370         rv = 0;
2371 out:
2372         mddev_unlock(mddev);
2373         if (rv)
2374                 return rv;
2375         return len;
2376 }
2377
2378 static struct md_sysfs_entry bitmap_location =
2379 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2380
2381 /* 'bitmap/space' is the space available at 'location' for the
2382  * bitmap.  This allows the kernel to know when it is safe to
2383  * resize the bitmap to match a resized array.
2384  */
2385 static ssize_t
2386 space_show(struct mddev *mddev, char *page)
2387 {
2388         return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2389 }
2390
2391 static ssize_t
2392 space_store(struct mddev *mddev, const char *buf, size_t len)
2393 {
2394         unsigned long sectors;
2395         int rv;
2396
2397         rv = kstrtoul(buf, 10, &sectors);
2398         if (rv)
2399                 return rv;
2400
2401         if (sectors == 0)
2402                 return -EINVAL;
2403
2404         if (mddev->bitmap &&
2405             sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2406                 return -EFBIG; /* Bitmap is too big for this small space */
2407
2408         /* could make sure it isn't too big, but that isn't really
2409          * needed - user-space should be careful.
2410          */
2411         mddev->bitmap_info.space = sectors;
2412         return len;
2413 }
2414
2415 static struct md_sysfs_entry bitmap_space =
2416 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2417
2418 static ssize_t
2419 timeout_show(struct mddev *mddev, char *page)
2420 {
2421         ssize_t len;
2422         unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2423         unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2424
2425         len = sprintf(page, "%lu", secs);
2426         if (jifs)
2427                 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2428         len += sprintf(page+len, "\n");
2429         return len;
2430 }
2431
2432 static ssize_t
2433 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2434 {
2435         /* timeout can be set at any time */
2436         unsigned long timeout;
2437         int rv = strict_strtoul_scaled(buf, &timeout, 4);
2438         if (rv)
2439                 return rv;
2440
2441         /* just to make sure we don't overflow... */
2442         if (timeout >= LONG_MAX / HZ)
2443                 return -EINVAL;
2444
2445         timeout = timeout * HZ / 10000;
2446
2447         if (timeout >= MAX_SCHEDULE_TIMEOUT)
2448                 timeout = MAX_SCHEDULE_TIMEOUT-1;
2449         if (timeout < 1)
2450                 timeout = 1;
2451         mddev->bitmap_info.daemon_sleep = timeout;
2452         if (mddev->thread) {
2453                 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2454                  * the bitmap is all clean and we don't need to
2455                  * adjust the timeout right now
2456                  */
2457                 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2458                         mddev->thread->timeout = timeout;
2459                         md_wakeup_thread(mddev->thread);
2460                 }
2461         }
2462         return len;
2463 }
2464
2465 static struct md_sysfs_entry bitmap_timeout =
2466 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2467
2468 static ssize_t
2469 backlog_show(struct mddev *mddev, char *page)
2470 {
2471         return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2472 }
2473
2474 static ssize_t
2475 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2476 {
2477         unsigned long backlog;
2478         unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2479         struct md_rdev *rdev;
2480         bool has_write_mostly = false;
2481         int rv = kstrtoul(buf, 10, &backlog);
2482         if (rv)
2483                 return rv;
2484         if (backlog > COUNTER_MAX)
2485                 return -EINVAL;
2486
2487         rv = mddev_lock(mddev);
2488         if (rv)
2489                 return rv;
2490
2491         /*
2492          * Without write mostly device, it doesn't make sense to set
2493          * backlog for max_write_behind.
2494          */
2495         rdev_for_each(rdev, mddev) {
2496                 if (test_bit(WriteMostly, &rdev->flags)) {
2497                         has_write_mostly = true;
2498                         break;
2499                 }
2500         }
2501         if (!has_write_mostly) {
2502                 pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2503                                     mdname(mddev));
2504                 mddev_unlock(mddev);
2505                 return -EINVAL;
2506         }
2507
2508         mddev->bitmap_info.max_write_behind = backlog;
2509         if (!backlog && mddev->serial_info_pool) {
2510                 /* serial_info_pool is not needed if backlog is zero */
2511                 if (!mddev->serialize_policy)
2512                         mddev_destroy_serial_pool(mddev, NULL, false);
2513         } else if (backlog && !mddev->serial_info_pool) {
2514                 /* serial_info_pool is needed since backlog is not zero */
2515                 rdev_for_each(rdev, mddev)
2516                         mddev_create_serial_pool(mddev, rdev, false);
2517         }
2518         if (old_mwb != backlog)
2519                 md_bitmap_update_sb(mddev->bitmap);
2520
2521         mddev_unlock(mddev);
2522         return len;
2523 }
2524
2525 static struct md_sysfs_entry bitmap_backlog =
2526 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2527
2528 static ssize_t
2529 chunksize_show(struct mddev *mddev, char *page)
2530 {
2531         return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2532 }
2533
2534 static ssize_t
2535 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2536 {
2537         /* Can only be changed when no bitmap is active */
2538         int rv;
2539         unsigned long csize;
2540         if (mddev->bitmap)
2541                 return -EBUSY;
2542         rv = kstrtoul(buf, 10, &csize);
2543         if (rv)
2544                 return rv;
2545         if (csize < 512 ||
2546             !is_power_of_2(csize))
2547                 return -EINVAL;
2548         if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2549                 sizeof(((bitmap_super_t *)0)->chunksize))))
2550                 return -EOVERFLOW;
2551         mddev->bitmap_info.chunksize = csize;
2552         return len;
2553 }
2554
2555 static struct md_sysfs_entry bitmap_chunksize =
2556 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2557
2558 static ssize_t metadata_show(struct mddev *mddev, char *page)
2559 {
2560         if (mddev_is_clustered(mddev))
2561                 return sprintf(page, "clustered\n");
2562         return sprintf(page, "%s\n", (mddev->bitmap_info.external
2563                                       ? "external" : "internal"));
2564 }
2565
2566 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2567 {
2568         if (mddev->bitmap ||
2569             mddev->bitmap_info.file ||
2570             mddev->bitmap_info.offset)
2571                 return -EBUSY;
2572         if (strncmp(buf, "external", 8) == 0)
2573                 mddev->bitmap_info.external = 1;
2574         else if ((strncmp(buf, "internal", 8) == 0) ||
2575                         (strncmp(buf, "clustered", 9) == 0))
2576                 mddev->bitmap_info.external = 0;
2577         else
2578                 return -EINVAL;
2579         return len;
2580 }
2581
2582 static struct md_sysfs_entry bitmap_metadata =
2583 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2584
2585 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2586 {
2587         int len;
2588         spin_lock(&mddev->lock);
2589         if (mddev->bitmap)
2590                 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2591                                              "false" : "true"));
2592         else
2593                 len = sprintf(page, "\n");
2594         spin_unlock(&mddev->lock);
2595         return len;
2596 }
2597
2598 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2599 {
2600         if (mddev->bitmap == NULL)
2601                 return -ENOENT;
2602         if (strncmp(buf, "false", 5) == 0)
2603                 mddev->bitmap->need_sync = 1;
2604         else if (strncmp(buf, "true", 4) == 0) {
2605                 if (mddev->degraded)
2606                         return -EBUSY;
2607                 mddev->bitmap->need_sync = 0;
2608         } else
2609                 return -EINVAL;
2610         return len;
2611 }
2612
2613 static struct md_sysfs_entry bitmap_can_clear =
2614 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2615
2616 static ssize_t
2617 behind_writes_used_show(struct mddev *mddev, char *page)
2618 {
2619         ssize_t ret;
2620         spin_lock(&mddev->lock);
2621         if (mddev->bitmap == NULL)
2622                 ret = sprintf(page, "0\n");
2623         else
2624                 ret = sprintf(page, "%lu\n",
2625                               mddev->bitmap->behind_writes_used);
2626         spin_unlock(&mddev->lock);
2627         return ret;
2628 }
2629
2630 static ssize_t
2631 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2632 {
2633         if (mddev->bitmap)
2634                 mddev->bitmap->behind_writes_used = 0;
2635         return len;
2636 }
2637
2638 static struct md_sysfs_entry max_backlog_used =
2639 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2640        behind_writes_used_show, behind_writes_used_reset);
2641
2642 static struct attribute *md_bitmap_attrs[] = {
2643         &bitmap_location.attr,
2644         &bitmap_space.attr,
2645         &bitmap_timeout.attr,
2646         &bitmap_backlog.attr,
2647         &bitmap_chunksize.attr,
2648         &bitmap_metadata.attr,
2649         &bitmap_can_clear.attr,
2650         &max_backlog_used.attr,
2651         NULL
2652 };
2653 struct attribute_group md_bitmap_group = {
2654         .name = "bitmap",
2655         .attrs = md_bitmap_attrs,
2656 };