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