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 686                         pr_warn("%s: bitmap file is out of date (%llu 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