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