drivers/md/md-bitmap.c

   1 /*
   2  * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   3  *
   4  * bitmap_create  - sets up the bitmap structure
   5  * bitmap_destroy - destroys the bitmap structure
   6  *
   7  * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   8  * - added disk storage for bitmap
   9  * - changes to allow various bitmap chunk sizes
  10  */
  11
  12 /*
  13  * Still to do:
  14  *
  15  * flush after percent set rather than just time based. (maybe both).
  16  */
  17
  18 #include <linux/blkdev.h>
  19 #include <linux/module.h>
  20 #include <linux/errno.h>
  21 #include <linux/slab.h>
  22 #include <linux/init.h>
  23 #include <linux/timer.h>
  24 #include <linux/sched.h>
  25 #include <linux/list.h>
  26 #include <linux/file.h>
  27 #include <linux/mount.h>
  28 #include <linux/buffer_head.h>
  29 #include <linux/seq_file.h>
  30 #include <trace/events/block.h>
  31 #include "md.h"
  32 #include "md-bitmap.h"
  33
  34 static inline char *bmname(struct bitmap *bitmap)
  35 {
  36         return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  37 }
  38
  39 /*
  40  * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
  41  *
  42  * 1) check to see if this page is allocated, if it's not then try to alloc
  43  * 2) if the alloc fails, set the page's hijacked flag so we'll use the
  44  *    page pointer directly as a counter
  45  *
  46  * if we find our page, we increment the page's refcount so that it stays
  47  * allocated while we're using it
  48  */
  49 static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
  50                                unsigned long page, int create, int no_hijack)
  51 __releases(bitmap->lock)
  52 __acquires(bitmap->lock)
  53 {
  54         unsigned char *mappage;
  55
  56         if (page >= bitmap->pages) {
  57                 /* This can happen if bitmap_start_sync goes beyond
  58                  * End-of-device while looking for a whole page.
  59                  * It is harmless.
  60                  */
  61                 return -EINVAL;
  62         }
  63
  64         if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  65                 return 0;
  66
  67         if (bitmap->bp[page].map) /* page is already allocated, just return */
  68                 return 0;
  69
  70         if (!create)
  71                 return -ENOENT;
  72
  73         /* this page has not been allocated yet */
  74
  75         spin_unlock_irq(&bitmap->lock);
  76         /* It is possible that this is being called inside a
  77          * prepare_to_wait/finish_wait loop from raid5c:make_request().
  78          * In general it is not permitted to sleep in that context as it
  79          * can cause the loop to spin freely.
  80          * That doesn't apply here as we can only reach this point
  81          * once with any loop.
  82          * When this function completes, either bp[page].map or
  83          * bp[page].hijacked.  In either case, this function will
  84          * abort before getting to this point again.  So there is
  85          * no risk of a free-spin, and so it is safe to assert
  86          * that sleeping here is allowed.
  87          */
  88         sched_annotate_sleep();
  89         mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  90         spin_lock_irq(&bitmap->lock);
  91
  92         if (mappage == NULL) {
  93                 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  94                 /* We don't support hijack for cluster raid */
  95                 if (no_hijack)
  96                         return -ENOMEM;
  97                 /* failed - set the hijacked flag so that we can use the
  98                  * pointer as a counter */
  99                 if (!bitmap->bp[page].map)
 100                         bitmap->bp[page].hijacked = 1;
 101         } else if (bitmap->bp[page].map ||
 102                    bitmap->bp[page].hijacked) {
 103                 /* somebody beat us to getting the page */
 104                 kfree(mappage);
 105         } else {
 106
 107                 /* no page was in place and we have one, so install it */
 108
 109                 bitmap->bp[page].map = mappage;
 110                 bitmap->missing_pages--;
 111         }
 112         return 0;
 113 }
 114
 115 /* if page is completely empty, put it back on the free list, or dealloc it */
 116 /* if page was hijacked, unmark the flag so it might get alloced next time */
 117 /* Note: lock should be held when calling this */
 118 static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 119 {
 120         char *ptr;
 121
 122         if (bitmap->bp[page].count) /* page is still busy */
 123                 return;
 124
 125         /* page is no longer in use, it can be released */
 126
 127         if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 128                 bitmap->bp[page].hijacked = 0;
 129                 bitmap->bp[page].map = NULL;
 130         } else {
 131                 /* normal case, free the page */
 132                 ptr = bitmap->bp[page].map;
 133                 bitmap->bp[page].map = NULL;
 134                 bitmap->missing_pages++;
 135                 kfree(ptr);
 136         }
 137 }
 138
 139 /*
 140  * bitmap file handling - read and write the bitmap file and its superblock
 141  */
 142
 143 /*
 144  * basic page I/O operations
 145  */
 146
 147 /* IO operations when bitmap is stored near all superblocks */
 148 static int read_sb_page(struct mddev *mddev, loff_t offset,
 149                         struct page *page,
 150                         unsigned long index, int size)
 151 {
 152         /* choose a good rdev and read the page from there */
 153
 154         struct md_rdev *rdev;
 155         sector_t target;
 156
 157         rdev_for_each(rdev, mddev) {
 158                 if (! test_bit(In_sync, &rdev->flags)
 159                     || test_bit(Faulty, &rdev->flags)
 160                     || test_bit(Bitmap_sync, &rdev->flags))
 161                         continue;
 162
 163                 target = offset + index * (PAGE_SIZE/512);
 164
 165                 if (sync_page_io(rdev, target,
 166                                  roundup(size, bdev_logical_block_size(rdev->bdev)),
 167                                  page, REQ_OP_READ, 0, true)) {
 168                         page->index = index;
 169                         return 0;
 170                 }
 171         }
 172         return -EIO;
 173 }
 174
 175 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 176 {
 177         /* Iterate the disks of an mddev, using rcu to protect access to the
 178          * linked list, and raising the refcount of devices we return to ensure
 179          * they don't disappear while in use.
 180          * As devices are only added or removed when raid_disk is < 0 and
 181          * nr_pending is 0 and In_sync is clear, the entries we return will
 182          * still be in the same position on the list when we re-enter
 183          * list_for_each_entry_continue_rcu.
 184          *
 185          * Note that if entered with 'rdev == NULL' to start at the
 186          * beginning, we temporarily assign 'rdev' to an address which
 187          * isn't really an rdev, but which can be used by
 188          * list_for_each_entry_continue_rcu() to find the first entry.
 189          */
 190         rcu_read_lock();
 191         if (rdev == NULL)
 192                 /* start at the beginning */
 193                 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 194         else {
 195                 /* release the previous rdev and start from there. */
 196                 rdev_dec_pending(rdev, mddev);
 197         }
 198         list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 199                 if (rdev->raid_disk >= 0 &&
 200                     !test_bit(Faulty, &rdev->flags)) {
 201                         /* this is a usable devices */
 202                         atomic_inc(&rdev->nr_pending);
 203                         rcu_read_unlock();
 204                         return rdev;
 205                 }
 206         }
 207         rcu_read_unlock();
 208         return NULL;
 209 }
 210
 211 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 212 {
 213         struct md_rdev *rdev;
 214         struct block_device *bdev;
 215         struct mddev *mddev = bitmap->mddev;
 216         struct bitmap_storage *store = &bitmap->storage;
 217
 218 restart:
 219         rdev = NULL;
 220         while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 221                 int size = PAGE_SIZE;
 222                 loff_t offset = mddev->bitmap_info.offset;
 223
 224                 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 225
 226                 if (page->index == store->file_pages-1) {
 227                         int last_page_size = store->bytes & (PAGE_SIZE-1);
 228                         if (last_page_size == 0)
 229                                 last_page_size = PAGE_SIZE;
 230                         size = roundup(last_page_size,
 231                                        bdev_logical_block_size(bdev));
 232                 }
 233                 /* Just make sure we aren't corrupting data or
 234                  * metadata
 235                  */
 236                 if (mddev->external) {
 237                         /* Bitmap could be anywhere. */
 238                         if (rdev->sb_start + offset + (page->index
 239                                                        * (PAGE_SIZE/512))
 240                             > rdev->data_offset
 241                             &&
 242                             rdev->sb_start + offset
 243                             < (rdev->data_offset + mddev->dev_sectors
 244                              + (PAGE_SIZE/512)))
 245                                 goto bad_alignment;
 246                 } else if (offset < 0) {
 247                         /* DATA  BITMAP METADATA  */
 248                         if (offset
 249                             + (long)(page->index * (PAGE_SIZE/512))
 250                             + size/512 > 0)
 251                                 /* bitmap runs in to metadata */
 252                                 goto bad_alignment;
 253                         if (rdev->data_offset + mddev->dev_sectors
 254                             > rdev->sb_start + offset)
 255                                 /* data runs in to bitmap */
 256                                 goto bad_alignment;
 257                 } else if (rdev->sb_start < rdev->data_offset) {
 258                         /* METADATA BITMAP DATA */
 259                         if (rdev->sb_start
 260                             + offset
 261                             + page->index*(PAGE_SIZE/512) + size/512
 262                             > rdev->data_offset)
 263                                 /* bitmap runs in to data */
 264                                 goto bad_alignment;
 265                 } else {
 266                         /* DATA METADATA BITMAP - no problems */
 267                 }
 268                 md_super_write(mddev, rdev,
 269                                rdev->sb_start + offset
 270                                + page->index * (PAGE_SIZE/512),
 271                                size,
 272                                page);
 273         }
 274
 275         if (wait && md_super_wait(mddev) < 0)
 276                 goto restart;
 277         return 0;
 278
 279  bad_alignment:
 280         return -EINVAL;
 281 }
 282
 283 static void md_bitmap_file_kick(struct bitmap *bitmap);
 284 /*
 285  * write out a page to a file
 286  */
 287 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 288 {
 289         struct buffer_head *bh;
 290
 291         if (bitmap->storage.file == NULL) {
 292                 switch (write_sb_page(bitmap, page, wait)) {
 293                 case -EINVAL:
 294                         set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 295                 }
 296         } else {
 297
 298                 bh = page_buffers(page);
 299
 300                 while (bh && bh->b_blocknr) {
 301                         atomic_inc(&bitmap->pending_writes);
 302                         set_buffer_locked(bh);
 303                         set_buffer_mapped(bh);
 304                         submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
 305                         bh = bh->b_this_page;
 306                 }
 307
 308                 if (wait)
 309                         wait_event(bitmap->write_wait,
 310                                    atomic_read(&bitmap->pending_writes)==0);
 311         }
 312         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 313                 md_bitmap_file_kick(bitmap);
 314 }
 315
 316 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 317 {
 318         struct bitmap *bitmap = bh->b_private;
 319
 320         if (!uptodate)
 321                 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 322         if (atomic_dec_and_test(&bitmap->pending_writes))
 323                 wake_up(&bitmap->write_wait);
 324 }
 325
 326 /* copied from buffer.c */
 327 static void
 328 __clear_page_buffers(struct page *page)
 329 {
 330         ClearPagePrivate(page);
 331         set_page_private(page, 0);
 332         put_page(page);
 333 }
 334 static void free_buffers(struct page *page)
 335 {
 336         struct buffer_head *bh;
 337
 338         if (!PagePrivate(page))
 339                 return;
 340
 341         bh = page_buffers(page);
 342         while (bh) {
 343                 struct buffer_head *next = bh->b_this_page;
 344                 free_buffer_head(bh);
 345                 bh = next;
 346         }
 347         __clear_page_buffers(page);
 348         put_page(page);
 349 }
 350
 351 /* read a page from a file.
 352  * We both read the page, and attach buffers to the page to record the
 353  * address of each block (using bmap).  These addresses will be used
 354  * to write the block later, completely bypassing the filesystem.
 355  * This usage is similar to how swap files are handled, and allows us
 356  * to write to a file with no concerns of memory allocation failing.
 357  */
 358 static int read_page(struct file *file, unsigned long index,
 359                      struct bitmap *bitmap,
 360                      unsigned long count,
 361                      struct page *page)
 362 {
 363         int ret = 0;
 364         struct inode *inode = file_inode(file);
 365         struct buffer_head *bh;
 366         sector_t block;
 367
 368         pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 369                  (unsigned long long)index << PAGE_SHIFT);
 370
 371         bh = alloc_page_buffers(page, 1<<inode->i_blkbits, false);
 372         if (!bh) {
 373                 ret = -ENOMEM;
 374                 goto out;
 375         }
 376         attach_page_buffers(page, bh);
 377         block = index << (PAGE_SHIFT - inode->i_blkbits);
 378         while (bh) {
 379                 if (count == 0)
 380                         bh->b_blocknr = 0;
 381                 else {
 382                         bh->b_blocknr = bmap(inode, block);
 383                         if (bh->b_blocknr == 0) {
 384                                 /* Cannot use this file! */
 385                                 ret = -EINVAL;
 386                                 goto out;
 387                         }
 388                         bh->b_bdev = inode->i_sb->s_bdev;
 389                         if (count < (1<<inode->i_blkbits))
 390                                 count = 0;
 391                         else
 392                                 count -= (1<<inode->i_blkbits);
 393
 394                         bh->b_end_io = end_bitmap_write;
 395                         bh->b_private = bitmap;
 396                         atomic_inc(&bitmap->pending_writes);
 397                         set_buffer_locked(bh);
 398                         set_buffer_mapped(bh);
 399                         submit_bh(REQ_OP_READ, 0, bh);
 400                 }
 401                 block++;
 402                 bh = bh->b_this_page;
 403         }
 404         page->index = index;
 405
 406         wait_event(bitmap->write_wait,
 407                    atomic_read(&bitmap->pending_writes)==0);
 408         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 409                 ret = -EIO;
 410 out:
 411         if (ret)
 412                 pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 413                        (int)PAGE_SIZE,
 414                        (unsigned long long)index << PAGE_SHIFT,
 415                        ret);
 416         return ret;
 417 }
 418
 419 /*
 420  * bitmap file superblock operations
 421  */
 422
 423 /*
 424  * md_bitmap_wait_writes() should be called before writing any bitmap
 425  * blocks, to ensure previous writes, particularly from
 426  * md_bitmap_daemon_work(), have completed.
 427  */
 428 static void md_bitmap_wait_writes(struct bitmap *bitmap)
 429 {
 430         if (bitmap->storage.file)
 431                 wait_event(bitmap->write_wait,
 432                            atomic_read(&bitmap->pending_writes)==0);
 433         else
 434                 /* Note that we ignore the return value.  The writes
 435                  * might have failed, but that would just mean that
 436                  * some bits which should be cleared haven't been,
 437                  * which is safe.  The relevant bitmap blocks will
 438                  * probably get written again, but there is no great
 439                  * loss if they aren't.
 440                  */
 441                 md_super_wait(bitmap->mddev);
 442 }
 443
 444
 445 /* update the event counter and sync the superblock to disk */
 446 void md_bitmap_update_sb(struct bitmap *bitmap)
 447 {
 448         bitmap_super_t *sb;
 449
 450         if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 451                 return;
 452         if (bitmap->mddev->bitmap_info.external)
 453                 return;
 454         if (!bitmap->storage.sb_page) /* no superblock */
 455                 return;
 456         sb = kmap_atomic(bitmap->storage.sb_page);
 457         sb->events = cpu_to_le64(bitmap->mddev->events);
 458         if (bitmap->mddev->events < bitmap->events_cleared)
 459                 /* rocking back to read-only */
 460                 bitmap->events_cleared = bitmap->mddev->events;
 461         sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 462         /*
 463          * clear BITMAP_WRITE_ERROR bit to protect against the case that
 464          * a bitmap write error occurred but the later writes succeeded.
 465          */
 466         sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 467         /* Just in case these have been changed via sysfs: */
 468         sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 469         sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 470         /* This might have been changed by a reshape */
 471         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 472         sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 473         sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 474         sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 475                                            bitmap_info.space);
 476         kunmap_atomic(sb);
 477         write_page(bitmap, bitmap->storage.sb_page, 1);
 478 }
 479 EXPORT_SYMBOL(md_bitmap_update_sb);
 480
 481 /* print out the bitmap file superblock */
 482 void md_bitmap_print_sb(struct bitmap *bitmap)
 483 {
 484         bitmap_super_t *sb;
 485
 486         if (!bitmap || !bitmap->storage.sb_page)
 487                 return;
 488         sb = kmap_atomic(bitmap->storage.sb_page);
 489         pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 490         pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 491         pr_debug("       version: %d\n", le32_to_cpu(sb->version));
 492         pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 493                  le32_to_cpu(*(__u32 *)(sb->uuid+0)),
 494                  le32_to_cpu(*(__u32 *)(sb->uuid+4)),
 495                  le32_to_cpu(*(__u32 *)(sb->uuid+8)),
 496                  le32_to_cpu(*(__u32 *)(sb->uuid+12)));
 497         pr_debug("        events: %llu\n",
 498                  (unsigned long long) le64_to_cpu(sb->events));
 499         pr_debug("events cleared: %llu\n",
 500                  (unsigned long long) le64_to_cpu(sb->events_cleared));
 501         pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 502         pr_debug("     chunksize: %d B\n", le32_to_cpu(sb->chunksize));
 503         pr_debug("  daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
 504         pr_debug("     sync size: %llu KB\n",
 505                  (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 506         pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind));
 507         kunmap_atomic(sb);
 508 }
 509
 510 /*
 511  * bitmap_new_disk_sb
 512  * @bitmap
 513  *
 514  * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 515  * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 516  * This function verifies 'bitmap_info' and populates the on-disk bitmap
 517  * structure, which is to be written to disk.
 518  *
 519  * Returns: 0 on success, -Exxx on error
 520  */
 521 static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 522 {
 523         bitmap_super_t *sb;
 524         unsigned long chunksize, daemon_sleep, write_behind;
 525
 526         bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 527         if (bitmap->storage.sb_page == NULL)
 528                 return -ENOMEM;
 529         bitmap->storage.sb_page->index = 0;
 530
 531         sb = kmap_atomic(bitmap->storage.sb_page);
 532
 533         sb->magic = cpu_to_le32(BITMAP_MAGIC);
 534         sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 535
 536         chunksize = bitmap->mddev->bitmap_info.chunksize;
 537         BUG_ON(!chunksize);
 538         if (!is_power_of_2(chunksize)) {
 539                 kunmap_atomic(sb);
 540                 pr_warn("bitmap chunksize not a power of 2\n");
 541                 return -EINVAL;
 542         }
 543         sb->chunksize = cpu_to_le32(chunksize);
 544
 545         daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 546         if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 547                 pr_debug("Choosing daemon_sleep default (5 sec)\n");
 548                 daemon_sleep = 5 * HZ;
 549         }
 550         sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 551         bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 552
 553         /*
 554          * FIXME: write_behind for RAID1.  If not specified, what
 555          * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 556          */
 557         write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 558         if (write_behind > COUNTER_MAX)
 559                 write_behind = COUNTER_MAX / 2;
 560         sb->write_behind = cpu_to_le32(write_behind);
 561         bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 562
 563         /* keep the array size field of the bitmap superblock up to date */
 564         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 565
 566         memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 567
 568         set_bit(BITMAP_STALE, &bitmap->flags);
 569         sb->state = cpu_to_le32(bitmap->flags);
 570         bitmap->events_cleared = bitmap->mddev->events;
 571         sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 572         bitmap->mddev->bitmap_info.nodes = 0;
 573
 574         kunmap_atomic(sb);
 575
 576         return 0;
 577 }
 578
 579 /* read the superblock from the bitmap file and initialize some bitmap fields */
 580 static int md_bitmap_read_sb(struct bitmap *bitmap)
 581 {
 582         char *reason = NULL;
 583         bitmap_super_t *sb;
 584         unsigned long chunksize, daemon_sleep, write_behind;
 585         unsigned long long events;
 586         int nodes = 0;
 587         unsigned long sectors_reserved = 0;
 588         int err = -EINVAL;
 589         struct page *sb_page;
 590         loff_t offset = bitmap->mddev->bitmap_info.offset;
 591
 592         if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 593                 chunksize = 128 * 1024 * 1024;
 594                 daemon_sleep = 5 * HZ;
 595                 write_behind = 0;
 596                 set_bit(BITMAP_STALE, &bitmap->flags);
 597                 err = 0;
 598                 goto out_no_sb;
 599         }
 600         /* page 0 is the superblock, read it... */
 601         sb_page = alloc_page(GFP_KERNEL);
 602         if (!sb_page)
 603                 return -ENOMEM;
 604         bitmap->storage.sb_page = sb_page;
 605
 606 re_read:
 607         /* If cluster_slot is set, the cluster is setup */
 608         if (bitmap->cluster_slot >= 0) {
 609                 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 610
 611                 sector_div(bm_blocks,
 612                            bitmap->mddev->bitmap_info.chunksize >> 9);
 613                 /* bits to bytes */
 614                 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 615                 /* to 4k blocks */
 616                 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 617                 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
 618                 pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 619                         bitmap->cluster_slot, offset);
 620         }
 621
 622         if (bitmap->storage.file) {
 623                 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 624                 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 625
 626                 err = read_page(bitmap->storage.file, 0,
 627                                 bitmap, bytes, sb_page);
 628         } else {
 629                 err = read_sb_page(bitmap->mddev,
 630                                    offset,
 631                                    sb_page,
 632                                    0, sizeof(bitmap_super_t));
 633         }
 634         if (err)
 635                 return err;
 636
 637         err = -EINVAL;
 638         sb = kmap_atomic(sb_page);
 639
 640         chunksize = le32_to_cpu(sb->chunksize);
 641         daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 642         write_behind = le32_to_cpu(sb->write_behind);
 643         sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 644
 645         /* verify that the bitmap-specific fields are valid */
 646         if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 647                 reason = "bad magic";
 648         else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 649                  le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 650                 reason = "unrecognized superblock version";
 651         else if (chunksize < 512)
 652                 reason = "bitmap chunksize too small";
 653         else if (!is_power_of_2(chunksize))
 654                 reason = "bitmap chunksize not a power of 2";
 655         else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 656                 reason = "daemon sleep period out of range";
 657         else if (write_behind > COUNTER_MAX)
 658                 reason = "write-behind limit out of range (0 - 16383)";
 659         if (reason) {
 660                 pr_warn("%s: invalid bitmap file superblock: %s\n",
 661                         bmname(bitmap), reason);
 662                 goto out;
 663         }
 664
 665         /*
 666          * Setup nodes/clustername only if bitmap version is
 667          * cluster-compatible
 668          */
 669         if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 670                 nodes = le32_to_cpu(sb->nodes);
 671                 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
 672                                 sb->cluster_name, 64);
 673         }
 674
 675         /* keep the array size field of the bitmap superblock up to date */
 676         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 677
 678         if (bitmap->mddev->persistent) {
 679                 /*
 680                  * We have a persistent array superblock, so compare the
 681                  * bitmap's UUID and event counter to the mddev's
 682                  */
 683                 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 684                         pr_warn("%s: bitmap superblock UUID mismatch\n",
 685                                 bmname(bitmap));
 686                         goto out;
 687                 }
 688                 events = le64_to_cpu(sb->events);
 689                 if (!nodes && (events < bitmap->mddev->events)) {
 690                         pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 691                                 bmname(bitmap), events,
 692                                 (unsigned long long) bitmap->mddev->events);
 693                         set_bit(BITMAP_STALE, &bitmap->flags);
 694                 }
 695         }
 696
 697         /* assign fields using values from superblock */
 698         bitmap->flags |= le32_to_cpu(sb->state);
 699         if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 700                 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 701         bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 702         strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 703         err = 0;
 704
 705 out:
 706         kunmap_atomic(sb);
 707         if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 708                 /* Assigning chunksize is required for "re_read" */
 709                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 710                 err = md_setup_cluster(bitmap->mddev, nodes);
 711                 if (err) {
 712                         pr_warn("%s: Could not setup cluster service (%d)\n",
 713                                 bmname(bitmap), err);
 714                         goto out_no_sb;
 715                 }
 716                 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 717                 goto re_read;
 718         }
 719
 720 out_no_sb:
 721         if (err == 0) {
 722                 if (test_bit(BITMAP_STALE, &bitmap->flags))
 723                         bitmap->events_cleared = bitmap->mddev->events;
 724                 bitmap->mddev->bitmap_info.chunksize = chunksize;
 725                 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 726                 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 727                 bitmap->mddev->bitmap_info.nodes = nodes;
 728                 if (bitmap->mddev->bitmap_info.space == 0 ||
 729                         bitmap->mddev->bitmap_info.space > sectors_reserved)
 730                         bitmap->mddev->bitmap_info.space = sectors_reserved;
 731         } else {
 732                 md_bitmap_print_sb(bitmap);
 733                 if (bitmap->cluster_slot < 0)
 734                         md_cluster_stop(bitmap->mddev);
 735         }
 736         return err;
 737 }
 738
 739 /*
 740  * general bitmap file operations
 741  */
 742
 743 /*
 744  * on-disk bitmap:
 745  *
 746  * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 747  * file a page at a time. There's a superblock at the start of the file.
 748  */
 749 /* calculate the index of the page that contains this bit */
 750 static inline unsigned long file_page_index(struct bitmap_storage *store,
 751                                             unsigned long chunk)
 752 {
 753         if (store->sb_page)
 754                 chunk += sizeof(bitmap_super_t) << 3;
 755         return chunk >> PAGE_BIT_SHIFT;
 756 }
 757
 758 /* calculate the (bit) offset of this bit within a page */
 759 static inline unsigned long file_page_offset(struct bitmap_storage *store,
 760                                              unsigned long chunk)
 761 {
 762         if (store->sb_page)
 763                 chunk += sizeof(bitmap_super_t) << 3;
 764         return chunk & (PAGE_BITS - 1);
 765 }
 766
 767 /*
 768  * return a pointer to the page in the filemap that contains the given bit
 769  *
 770  */
 771 static inline struct page *filemap_get_page(struct bitmap_storage *store,
 772                                             unsigned long chunk)
 773 {
 774         if (file_page_index(store, chunk) >= store->file_pages)
 775                 return NULL;
 776         return store->filemap[file_page_index(store, chunk)];
 777 }
 778
 779 static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 780                                    unsigned long chunks, int with_super,
 781                                    int slot_number)
 782 {
 783         int pnum, offset = 0;
 784         unsigned long num_pages;
 785         unsigned long bytes;
 786
 787         bytes = DIV_ROUND_UP(chunks, 8);
 788         if (with_super)
 789                 bytes += sizeof(bitmap_super_t);
 790
 791         num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 792         offset = slot_number * num_pages;
 793
 794         store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 795                                        GFP_KERNEL);
 796         if (!store->filemap)
 797                 return -ENOMEM;
 798
 799         if (with_super && !store->sb_page) {
 800                 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 801                 if (store->sb_page == NULL)
 802                         return -ENOMEM;
 803         }
 804
 805         pnum = 0;
 806         if (store->sb_page) {
 807                 store->filemap[0] = store->sb_page;
 808                 pnum = 1;
 809                 store->sb_page->index = offset;
 810         }
 811
 812         for ( ; pnum < num_pages; pnum++) {
 813                 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 814                 if (!store->filemap[pnum]) {
 815                         store->file_pages = pnum;
 816                         return -ENOMEM;
 817                 }
 818                 store->filemap[pnum]->index = pnum + offset;
 819         }
 820         store->file_pages = pnum;
 821
 822         /* We need 4 bits per page, rounded up to a multiple
 823          * of sizeof(unsigned long) */
 824         store->filemap_attr = kzalloc(
 825                 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 826                 GFP_KERNEL);
 827         if (!store->filemap_attr)
 828                 return -ENOMEM;
 829
 830         store->bytes = bytes;
 831
 832         return 0;
 833 }
 834
 835 static void md_bitmap_file_unmap(struct bitmap_storage *store)
 836 {
 837         struct page **map, *sb_page;
 838         int pages;
 839         struct file *file;
 840
 841         file = store->file;
 842         map = store->filemap;
 843         pages = store->file_pages;
 844         sb_page = store->sb_page;
 845
 846         while (pages--)
 847                 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 848                         free_buffers(map[pages]);
 849         kfree(map);
 850         kfree(store->filemap_attr);
 851
 852         if (sb_page)
 853                 free_buffers(sb_page);
 854
 855         if (file) {
 856                 struct inode *inode = file_inode(file);
 857                 invalidate_mapping_pages(inode->i_mapping, 0, -1);
 858                 fput(file);
 859         }
 860 }
 861
 862 /*
 863  * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 864  * then it is no longer reliable, so we stop using it and we mark the file
 865  * as failed in the superblock
 866  */
 867 static void md_bitmap_file_kick(struct bitmap *bitmap)
 868 {
 869         char *path, *ptr = NULL;
 870
 871         if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 872                 md_bitmap_update_sb(bitmap);
 873
 874                 if (bitmap->storage.file) {
 875                         path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 876                         if (path)
 877                                 ptr = file_path(bitmap->storage.file,
 878                                              path, PAGE_SIZE);
 879
 880                         pr_warn("%s: kicking failed bitmap file %s from array!\n",
 881                                 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 882
 883                         kfree(path);
 884                 } else
 885                         pr_warn("%s: disabling internal bitmap due to errors\n",
 886                                 bmname(bitmap));
 887         }
 888 }
 889
 890 enum bitmap_page_attr {
 891         BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 892         BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 893                                     * i.e. counter is 1 or 2. */
 894         BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 895 };
 896
 897 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 898                                  enum bitmap_page_attr attr)
 899 {
 900         set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 901 }
 902
 903 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 904                                    enum bitmap_page_attr attr)
 905 {
 906         clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 907 }
 908
 909 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 910                                  enum bitmap_page_attr attr)
 911 {
 912         return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 913 }
 914
 915 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 916                                            enum bitmap_page_attr attr)
 917 {
 918         return test_and_clear_bit((pnum<<2) + attr,
 919                                   bitmap->storage.filemap_attr);
 920 }
 921 /*
 922  * bitmap_file_set_bit -- called before performing a write to the md device
 923  * to set (and eventually sync) a particular bit in the bitmap file
 924  *
 925  * we set the bit immediately, then we record the page number so that
 926  * when an unplug occurs, we can flush the dirty pages out to disk
 927  */
 928 static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 929 {
 930         unsigned long bit;
 931         struct page *page;
 932         void *kaddr;
 933         unsigned long chunk = block >> bitmap->counts.chunkshift;
 934         struct bitmap_storage *store = &bitmap->storage;
 935         unsigned long node_offset = 0;
 936
 937         if (mddev_is_clustered(bitmap->mddev))
 938                 node_offset = bitmap->cluster_slot * store->file_pages;
 939
 940         page = filemap_get_page(&bitmap->storage, chunk);
 941         if (!page)
 942                 return;
 943         bit = file_page_offset(&bitmap->storage, chunk);
 944
 945         /* set the bit */
 946         kaddr = kmap_atomic(page);
 947         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 948                 set_bit(bit, kaddr);
 949         else
 950                 set_bit_le(bit, kaddr);
 951         kunmap_atomic(kaddr);
 952         pr_debug("set file bit %lu page %lu\n", bit, page->index);
 953         /* record page number so it gets flushed to disk when unplug occurs */
 954         set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
 955 }
 956
 957 static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 958 {
 959         unsigned long bit;
 960         struct page *page;
 961         void *paddr;
 962         unsigned long chunk = block >> bitmap->counts.chunkshift;
 963         struct bitmap_storage *store = &bitmap->storage;
 964         unsigned long node_offset = 0;
 965
 966         if (mddev_is_clustered(bitmap->mddev))
 967                 node_offset = bitmap->cluster_slot * store->file_pages;
 968
 969         page = filemap_get_page(&bitmap->storage, chunk);
 970         if (!page)
 971                 return;
 972         bit = file_page_offset(&bitmap->storage, chunk);
 973         paddr = kmap_atomic(page);
 974         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 975                 clear_bit(bit, paddr);
 976         else
 977                 clear_bit_le(bit, paddr);
 978         kunmap_atomic(paddr);
 979         if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
 980                 set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
 981                 bitmap->allclean = 0;
 982         }
 983 }
 984
 985 static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
 986 {
 987         unsigned long bit;
 988         struct page *page;
 989         void *paddr;
 990         unsigned long chunk = block >> bitmap->counts.chunkshift;
 991         int set = 0;
 992
 993         page = filemap_get_page(&bitmap->storage, chunk);
 994         if (!page)
 995                 return -EINVAL;
 996         bit = file_page_offset(&bitmap->storage, chunk);
 997         paddr = kmap_atomic(page);
 998         if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 999                 set = test_bit(bit, paddr);
1000         else
1001                 set = test_bit_le(bit, paddr);
1002         kunmap_atomic(paddr);
1003         return set;
1004 }
1005
1006
1007 /* this gets called when the md device is ready to unplug its underlying
1008  * (slave) device queues -- before we let any writes go down, we need to
1009  * sync the dirty pages of the bitmap file to disk */
1010 void md_bitmap_unplug(struct bitmap *bitmap)
1011 {
1012         unsigned long i;
1013         int dirty, need_write;
1014         int writing = 0;
1015
1016         if (!bitmap || !bitmap->storage.filemap ||
1017             test_bit(BITMAP_STALE, &bitmap->flags))
1018                 return;
1019
1020         /* look at each page to see if there are any set bits that need to be
1021          * flushed out to disk */
1022         for (i = 0; i < bitmap->storage.file_pages; i++) {
1023                 if (!bitmap->storage.filemap)
1024                         return;
1025                 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1026                 need_write = test_and_clear_page_attr(bitmap, i,
1027                                                       BITMAP_PAGE_NEEDWRITE);
1028                 if (dirty || need_write) {
1029                         if (!writing) {
1030                                 md_bitmap_wait_writes(bitmap);
1031                                 if (bitmap->mddev->queue)
1032                                         blk_add_trace_msg(bitmap->mddev->queue,
1033                                                           "md bitmap_unplug");
1034                         }
1035                         clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1036                         write_page(bitmap, bitmap->storage.filemap[i], 0);
1037                         writing = 1;
1038                 }
1039         }
1040         if (writing)
1041                 md_bitmap_wait_writes(bitmap);
1042
1043         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1044                 md_bitmap_file_kick(bitmap);
1045 }
1046 EXPORT_SYMBOL(md_bitmap_unplug);
1047
1048 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1049 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1050  * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1051  * memory mapping of the bitmap file
1052  * Special cases:
1053  *   if there's no bitmap file, or if the bitmap file had been
1054  *   previously kicked from the array, we mark all the bits as
1055  *   1's in order to cause a full resync.
1056  *
1057  * We ignore all bits for sectors that end earlier than 'start'.
1058  * This is used when reading an out-of-date bitmap...
1059  */
1060 static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1061 {
1062         unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1063         struct page *page = NULL;
1064         unsigned long bit_cnt = 0;
1065         struct file *file;
1066         unsigned long offset;
1067         int outofdate;
1068         int ret = -ENOSPC;
1069         void *paddr;
1070         struct bitmap_storage *store = &bitmap->storage;
1071
1072         chunks = bitmap->counts.chunks;
1073         file = store->file;
1074
1075         if (!file && !bitmap->mddev->bitmap_info.offset) {
1076                 /* No permanent bitmap - fill with '1s'. */
1077                 store->filemap = NULL;
1078                 store->file_pages = 0;
1079                 for (i = 0; i < chunks ; i++) {
1080                         /* if the disk bit is set, set the memory bit */
1081                         int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1082                                       >= start);
1083                         md_bitmap_set_memory_bits(bitmap,
1084                                                   (sector_t)i << bitmap->counts.chunkshift,
1085                                                   needed);
1086                 }
1087                 return 0;
1088         }
1089
1090         outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1091         if (outofdate)
1092                 pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1093
1094         if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1095                 pr_warn("%s: bitmap file too short %lu < %lu\n",
1096                         bmname(bitmap),
1097                         (unsigned long) i_size_read(file->f_mapping->host),
1098                         store->bytes);
1099                 goto err;
1100         }
1101
1102         oldindex = ~0L;
1103         offset = 0;
1104         if (!bitmap->mddev->bitmap_info.external)
1105                 offset = sizeof(bitmap_super_t);
1106
1107         if (mddev_is_clustered(bitmap->mddev))
1108                 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1109
1110         for (i = 0; i < chunks; i++) {
1111                 int b;
1112                 index = file_page_index(&bitmap->storage, i);
1113                 bit = file_page_offset(&bitmap->storage, i);
1114                 if (index != oldindex) { /* this is a new page, read it in */
1115                         int count;
1116                         /* unmap the old page, we're done with it */
1117                         if (index == store->file_pages-1)
1118                                 count = store->bytes - index * PAGE_SIZE;
1119                         else
1120                                 count = PAGE_SIZE;
1121                         page = store->filemap[index];
1122                         if (file)
1123                                 ret = read_page(file, index, bitmap,
1124                                                 count, page);
1125                         else
1126                                 ret = read_sb_page(
1127                                         bitmap->mddev,
1128                                         bitmap->mddev->bitmap_info.offset,
1129                                         page,
1130                                         index + node_offset, count);
1131
1132                         if (ret)
1133                                 goto err;
1134
1135                         oldindex = index;
1136
1137                         if (outofdate) {
1138                                 /*
1139                                  * if bitmap is out of date, dirty the
1140                                  * whole page and write it out
1141                                  */
1142                                 paddr = kmap_atomic(page);
1143                                 memset(paddr + offset, 0xff,
1144                                        PAGE_SIZE - offset);
1145                                 kunmap_atomic(paddr);
1146                                 write_page(bitmap, page, 1);
1147
1148                                 ret = -EIO;
1149                                 if (test_bit(BITMAP_WRITE_ERROR,
1150                                              &bitmap->flags))
1151                                         goto err;
1152                         }
1153                 }
1154                 paddr = kmap_atomic(page);
1155                 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1156                         b = test_bit(bit, paddr);
1157                 else
1158                         b = test_bit_le(bit, paddr);
1159                 kunmap_atomic(paddr);
1160                 if (b) {
1161                         /* if the disk bit is set, set the memory bit */
1162                         int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1163                                       >= start);
1164                         md_bitmap_set_memory_bits(bitmap,
1165                                                   (sector_t)i << bitmap->counts.chunkshift,
1166                                                   needed);
1167                         bit_cnt++;
1168                 }
1169                 offset = 0;
1170         }
1171
1172         pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1173                  bmname(bitmap), store->file_pages,
1174                  bit_cnt, chunks);
1175
1176         return 0;
1177
1178  err:
1179         pr_warn("%s: bitmap initialisation failed: %d\n",
1180                 bmname(bitmap), ret);
1181         return ret;
1182 }
1183
1184 void md_bitmap_write_all(struct bitmap *bitmap)
1185 {
1186         /* We don't actually write all bitmap blocks here,
1187          * just flag them as needing to be written
1188          */
1189         int i;
1190
1191         if (!bitmap || !bitmap->storage.filemap)
1192                 return;
1193         if (bitmap->storage.file)
1194                 /* Only one copy, so nothing needed */
1195                 return;
1196
1197         for (i = 0; i < bitmap->storage.file_pages; i++)
1198                 set_page_attr(bitmap, i,
1199                               BITMAP_PAGE_NEEDWRITE);
1200         bitmap->allclean = 0;
1201 }
1202
1203 static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1204                                  sector_t offset, int inc)
1205 {
1206         sector_t chunk = offset >> bitmap->chunkshift;
1207         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1208         bitmap->bp[page].count += inc;
1209         md_bitmap_checkfree(bitmap, page);
1210 }
1211
1212 static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1213 {
1214         sector_t chunk = offset >> bitmap->chunkshift;
1215         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1216         struct bitmap_page *bp = &bitmap->bp[page];
1217
1218         if (!bp->pending)
1219                 bp->pending = 1;
1220 }
1221
1222 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1223                                                sector_t offset, sector_t *blocks,
1224                                                int create);
1225
1226 /*
1227  * bitmap daemon -- periodically wakes up to clean bits and flush pages
1228  *                      out to disk
1229  */
1230
1231 void md_bitmap_daemon_work(struct mddev *mddev)
1232 {
1233         struct bitmap *bitmap;
1234         unsigned long j;
1235         unsigned long nextpage;
1236         sector_t blocks;
1237         struct bitmap_counts *counts;
1238
1239         /* Use a mutex to guard daemon_work against
1240          * bitmap_destroy.
1241          */
1242         mutex_lock(&mddev->bitmap_info.mutex);
1243         bitmap = mddev->bitmap;
1244         if (bitmap == NULL) {
1245                 mutex_unlock(&mddev->bitmap_info.mutex);
1246                 return;
1247         }
1248         if (time_before(jiffies, bitmap->daemon_lastrun
1249                         + mddev->bitmap_info.daemon_sleep))
1250                 goto done;
1251
1252         bitmap->daemon_lastrun = jiffies;
1253         if (bitmap->allclean) {
1254                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1255                 goto done;
1256         }
1257         bitmap->allclean = 1;
1258
1259         if (bitmap->mddev->queue)
1260                 blk_add_trace_msg(bitmap->mddev->queue,
1261                                   "md bitmap_daemon_work");
1262
1263         /* Any file-page which is PENDING now needs to be written.
1264          * So set NEEDWRITE now, then after we make any last-minute changes
1265          * we will write it.
1266          */
1267         for (j = 0; j < bitmap->storage.file_pages; j++)
1268                 if (test_and_clear_page_attr(bitmap, j,
1269                                              BITMAP_PAGE_PENDING))
1270                         set_page_attr(bitmap, j,
1271                                       BITMAP_PAGE_NEEDWRITE);
1272
1273         if (bitmap->need_sync &&
1274             mddev->bitmap_info.external == 0) {
1275                 /* Arrange for superblock update as well as
1276                  * other changes */
1277                 bitmap_super_t *sb;
1278                 bitmap->need_sync = 0;
1279                 if (bitmap->storage.filemap) {
1280                         sb = kmap_atomic(bitmap->storage.sb_page);
1281                         sb->events_cleared =
1282                                 cpu_to_le64(bitmap->events_cleared);
1283                         kunmap_atomic(sb);
1284                         set_page_attr(bitmap, 0,
1285                                       BITMAP_PAGE_NEEDWRITE);
1286                 }
1287         }
1288         /* Now look at the bitmap counters and if any are '2' or '1',
1289          * decrement and handle accordingly.
1290          */
1291         counts = &bitmap->counts;
1292         spin_lock_irq(&counts->lock);
1293         nextpage = 0;
1294         for (j = 0; j < counts->chunks; j++) {
1295                 bitmap_counter_t *bmc;
1296                 sector_t  block = (sector_t)j << counts->chunkshift;
1297
1298                 if (j == nextpage) {
1299                         nextpage += PAGE_COUNTER_RATIO;
1300                         if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1301                                 j |= PAGE_COUNTER_MASK;
1302                                 continue;
1303                         }
1304                         counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1305                 }
1306
1307                 bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1308                 if (!bmc) {
1309                         j |= PAGE_COUNTER_MASK;
1310                         continue;
1311                 }
1312                 if (*bmc == 1 && !bitmap->need_sync) {
1313                         /* We can clear the bit */
1314                         *bmc = 0;
1315                         md_bitmap_count_page(counts, block, -1);
1316                         md_bitmap_file_clear_bit(bitmap, block);
1317                 } else if (*bmc && *bmc <= 2) {
1318                         *bmc = 1;
1319                         md_bitmap_set_pending(counts, block);
1320                         bitmap->allclean = 0;
1321                 }
1322         }
1323         spin_unlock_irq(&counts->lock);
1324
1325         md_bitmap_wait_writes(bitmap);
1326         /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1327          * DIRTY pages need to be written by bitmap_unplug so it can wait
1328          * for them.
1329          * If we find any DIRTY page we stop there and let bitmap_unplug
1330          * handle all the rest.  This is important in the case where
1331          * the first blocking holds the superblock and it has been updated.
1332          * We mustn't write any other blocks before the superblock.
1333          */
1334         for (j = 0;
1335              j < bitmap->storage.file_pages
1336                      && !test_bit(BITMAP_STALE, &bitmap->flags);
1337              j++) {
1338                 if (test_page_attr(bitmap, j,
1339                                    BITMAP_PAGE_DIRTY))
1340                         /* bitmap_unplug will handle the rest */
1341                         break;
1342                 if (test_and_clear_page_attr(bitmap, j,
1343                                              BITMAP_PAGE_NEEDWRITE)) {
1344                         write_page(bitmap, bitmap->storage.filemap[j], 0);
1345                 }
1346         }
1347
1348  done:
1349         if (bitmap->allclean == 0)
1350                 mddev->thread->timeout =
1351                         mddev->bitmap_info.daemon_sleep;
1352         mutex_unlock(&mddev->bitmap_info.mutex);
1353 }
1354
1355 static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1356                                                sector_t offset, sector_t *blocks,
1357                                                int create)
1358 __releases(bitmap->lock)
1359 __acquires(bitmap->lock)
1360 {
1361         /* If 'create', we might release the lock and reclaim it.
1362          * The lock must have been taken with interrupts enabled.
1363          * If !create, we don't release the lock.
1364          */
1365         sector_t chunk = offset >> bitmap->chunkshift;
1366         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1367         unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1368         sector_t csize;
1369         int err;
1370
1371         err = md_bitmap_checkpage(bitmap, page, create, 0);
1372
1373         if (bitmap->bp[page].hijacked ||
1374             bitmap->bp[page].map == NULL)
1375                 csize = ((sector_t)1) << (bitmap->chunkshift +
1376                                           PAGE_COUNTER_SHIFT);
1377         else
1378                 csize = ((sector_t)1) << bitmap->chunkshift;
1379         *blocks = csize - (offset & (csize - 1));
1380
1381         if (err < 0)
1382                 return NULL;
1383
1384         /* now locked ... */
1385
1386         if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1387                 /* should we use the first or second counter field
1388                  * of the hijacked pointer? */
1389                 int hi = (pageoff > PAGE_COUNTER_MASK);
1390                 return  &((bitmap_counter_t *)
1391                           &bitmap->bp[page].map)[hi];
1392         } else /* page is allocated */
1393                 return (bitmap_counter_t *)
1394                         &(bitmap->bp[page].map[pageoff]);
1395 }
1396
1397 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1398 {
1399         if (!bitmap)
1400                 return 0;
1401
1402         if (behind) {
1403                 int bw;
1404                 atomic_inc(&bitmap->behind_writes);
1405                 bw = atomic_read(&bitmap->behind_writes);
1406                 if (bw > bitmap->behind_writes_used)
1407                         bitmap->behind_writes_used = bw;
1408
1409                 pr_debug("inc write-behind count %d/%lu\n",
1410                          bw, bitmap->mddev->bitmap_info.max_write_behind);
1411         }
1412
1413         while (sectors) {
1414                 sector_t blocks;
1415                 bitmap_counter_t *bmc;
1416
1417                 spin_lock_irq(&bitmap->counts.lock);
1418                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1419                 if (!bmc) {
1420                         spin_unlock_irq(&bitmap->counts.lock);
1421                         return 0;
1422                 }
1423
1424                 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1425                         DEFINE_WAIT(__wait);
1426                         /* note that it is safe to do the prepare_to_wait
1427                          * after the test as long as we do it before dropping
1428                          * the spinlock.
1429                          */
1430                         prepare_to_wait(&bitmap->overflow_wait, &__wait,
1431                                         TASK_UNINTERRUPTIBLE);
1432                         spin_unlock_irq(&bitmap->counts.lock);
1433                         schedule();
1434                         finish_wait(&bitmap->overflow_wait, &__wait);
1435                         continue;
1436                 }
1437
1438                 switch (*bmc) {
1439                 case 0:
1440                         md_bitmap_file_set_bit(bitmap, offset);
1441                         md_bitmap_count_page(&bitmap->counts, offset, 1);
1442                         /* fall through */
1443                 case 1:
1444                         *bmc = 2;
1445                 }
1446
1447                 (*bmc)++;
1448
1449                 spin_unlock_irq(&bitmap->counts.lock);
1450
1451                 offset += blocks;
1452                 if (sectors > blocks)
1453                         sectors -= blocks;
1454                 else
1455                         sectors = 0;
1456         }
1457         return 0;
1458 }
1459 EXPORT_SYMBOL(md_bitmap_startwrite);
1460
1461 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1462                         unsigned long sectors, int success, int behind)
1463 {
1464         if (!bitmap)
1465                 return;
1466         if (behind) {
1467                 if (atomic_dec_and_test(&bitmap->behind_writes))
1468                         wake_up(&bitmap->behind_wait);
1469                 pr_debug("dec write-behind count %d/%lu\n",
1470                          atomic_read(&bitmap->behind_writes),
1471                          bitmap->mddev->bitmap_info.max_write_behind);
1472         }
1473
1474         while (sectors) {
1475                 sector_t blocks;
1476                 unsigned long flags;
1477                 bitmap_counter_t *bmc;
1478
1479                 spin_lock_irqsave(&bitmap->counts.lock, flags);
1480                 bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1481                 if (!bmc) {
1482                         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1483                         return;
1484                 }
1485
1486                 if (success && !bitmap->mddev->degraded &&
1487                     bitmap->events_cleared < bitmap->mddev->events) {
1488                         bitmap->events_cleared = bitmap->mddev->events;
1489                         bitmap->need_sync = 1;
1490                         sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1491                 }
1492
1493                 if (!success && !NEEDED(*bmc))
1494                         *bmc |= NEEDED_MASK;
1495
1496                 if (COUNTER(*bmc) == COUNTER_MAX)
1497                         wake_up(&bitmap->overflow_wait);
1498
1499                 (*bmc)--;
1500                 if (*bmc <= 2) {
1501                         md_bitmap_set_pending(&bitmap->counts, offset);
1502                         bitmap->allclean = 0;
1503                 }
1504                 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1505                 offset += blocks;
1506                 if (sectors > blocks)
1507                         sectors -= blocks;
1508                 else
1509                         sectors = 0;
1510         }
1511 }
1512 EXPORT_SYMBOL(md_bitmap_endwrite);
1513
1514 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1515                                int degraded)
1516 {
1517         bitmap_counter_t *bmc;
1518         int rv;
1519         if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1520                 *blocks = 1024;
1521                 return 1; /* always resync if no bitmap */
1522         }
1523         spin_lock_irq(&bitmap->counts.lock);
1524         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1525         rv = 0;
1526         if (bmc) {
1527                 /* locked */
1528                 if (RESYNC(*bmc))
1529                         rv = 1;
1530                 else if (NEEDED(*bmc)) {
1531                         rv = 1;
1532                         if (!degraded) { /* don't set/clear bits if degraded */
1533                                 *bmc |= RESYNC_MASK;
1534                                 *bmc &= ~NEEDED_MASK;
1535                         }
1536                 }
1537         }
1538         spin_unlock_irq(&bitmap->counts.lock);
1539         return rv;
1540 }
1541
1542 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1543                          int degraded)
1544 {
1545         /* bitmap_start_sync must always report on multiples of whole
1546          * pages, otherwise resync (which is very PAGE_SIZE based) will
1547          * get confused.
1548          * So call __bitmap_start_sync repeatedly (if needed) until
1549          * At least PAGE_SIZE>>9 blocks are covered.
1550          * Return the 'or' of the result.
1551          */
1552         int rv = 0;
1553         sector_t blocks1;
1554
1555         *blocks = 0;
1556         while (*blocks < (PAGE_SIZE>>9)) {
1557                 rv |= __bitmap_start_sync(bitmap, offset,
1558                                           &blocks1, degraded);
1559                 offset += blocks1;
1560                 *blocks += blocks1;
1561         }
1562         return rv;
1563 }
1564 EXPORT_SYMBOL(md_bitmap_start_sync);
1565
1566 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1567 {
1568         bitmap_counter_t *bmc;
1569         unsigned long flags;
1570
1571         if (bitmap == NULL) {
1572                 *blocks = 1024;
1573                 return;
1574         }
1575         spin_lock_irqsave(&bitmap->counts.lock, flags);
1576         bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1577         if (bmc == NULL)
1578                 goto unlock;
1579         /* locked */
1580         if (RESYNC(*bmc)) {
1581                 *bmc &= ~RESYNC_MASK;
1582
1583                 if (!NEEDED(*bmc) && aborted)
1584                         *bmc |= NEEDED_MASK;
1585                 else {
1586                         if (*bmc <= 2) {
1587                                 md_bitmap_set_pending(&bitmap->counts, offset);
1588                                 bitmap->allclean = 0;
1589                         }
1590                 }
1591         }
1592  unlock:
1593         spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1594 }
1595 EXPORT_SYMBOL(md_bitmap_end_sync);
1596
1597 void md_bitmap_close_sync(struct bitmap *bitmap)
1598 {
1599         /* Sync has finished, and any bitmap chunks that weren't synced
1600          * properly have been aborted.  It remains to us to clear the
1601          * RESYNC bit wherever it is still on
1602          */
1603         sector_t sector = 0;
1604         sector_t blocks;
1605         if (!bitmap)
1606                 return;
1607         while (sector < bitmap->mddev->resync_max_sectors) {
1608                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1609                 sector += blocks;
1610         }
1611 }
1612 EXPORT_SYMBOL(md_bitmap_close_sync);
1613
1614 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1615 {
1616         sector_t s = 0;
1617         sector_t blocks;
1618
1619         if (!bitmap)
1620                 return;
1621         if (sector == 0) {
1622                 bitmap->last_end_sync = jiffies;
1623                 return;
1624         }
1625         if (!force && time_before(jiffies, (bitmap->last_end_sync
1626                                   + bitmap->mddev->bitmap_info.daemon_sleep)))
1627                 return;
1628         wait_event(bitmap->mddev->recovery_wait,
1629                    atomic_read(&bitmap->mddev->recovery_active) == 0);
1630
1631         bitmap->mddev->curr_resync_completed = sector;
1632         set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1633         sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1634         s = 0;
1635         while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1636                 md_bitmap_end_sync(bitmap, s, &blocks, 0);
1637                 s += blocks;
1638         }
1639         bitmap->last_end_sync = jiffies;
1640         sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1641 }
1642 EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1643
1644 void md_bitmap_sync_with_cluster(struct mddev *mddev,
1645                               sector_t old_lo, sector_t old_hi,
1646                               sector_t new_lo, sector_t new_hi)
1647 {
1648         struct bitmap *bitmap = mddev->bitmap;
1649         sector_t sector, blocks = 0;
1650
1651         for (sector = old_lo; sector < new_lo; ) {
1652                 md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1653                 sector += blocks;
1654         }
1655         WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1656
1657         for (sector = old_hi; sector < new_hi; ) {
1658                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1659                 sector += blocks;
1660         }
1661         WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1662 }
1663 EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1664
1665 static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1666 {
1667         /* For each chunk covered by any of these sectors, set the
1668          * counter to 2 and possibly set resync_needed.  They should all
1669          * be 0 at this point
1670          */
1671
1672         sector_t secs;
1673         bitmap_counter_t *bmc;
1674         spin_lock_irq(&bitmap->counts.lock);
1675         bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1676         if (!bmc) {
1677                 spin_unlock_irq(&bitmap->counts.lock);
1678                 return;
1679         }
1680         if (!*bmc) {
1681                 *bmc = 2;
1682                 md_bitmap_count_page(&bitmap->counts, offset, 1);
1683                 md_bitmap_set_pending(&bitmap->counts, offset);
1684                 bitmap->allclean = 0;
1685         }
1686         if (needed)
1687                 *bmc |= NEEDED_MASK;
1688         spin_unlock_irq(&bitmap->counts.lock);
1689 }
1690
1691 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1692 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1693 {
1694         unsigned long chunk;
1695
1696         for (chunk = s; chunk <= e; chunk++) {
1697                 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1698                 md_bitmap_set_memory_bits(bitmap, sec, 1);
1699                 md_bitmap_file_set_bit(bitmap, sec);
1700                 if (sec < bitmap->mddev->recovery_cp)
1701                         /* We are asserting that the array is dirty,
1702                          * so move the recovery_cp address back so
1703                          * that it is obvious that it is dirty
1704                          */
1705                         bitmap->mddev->recovery_cp = sec;
1706         }
1707 }
1708
1709 /*
1710  * flush out any pending updates
1711  */
1712 void md_bitmap_flush(struct mddev *mddev)
1713 {
1714         struct bitmap *bitmap = mddev->bitmap;
1715         long sleep;
1716
1717         if (!bitmap) /* there was no bitmap */
1718                 return;
1719
1720         /* run the daemon_work three time to ensure everything is flushed
1721          * that can be
1722          */
1723         sleep = mddev->bitmap_info.daemon_sleep * 2;
1724         bitmap->daemon_lastrun -= sleep;
1725         md_bitmap_daemon_work(mddev);
1726         bitmap->daemon_lastrun -= sleep;
1727         md_bitmap_daemon_work(mddev);
1728         bitmap->daemon_lastrun -= sleep;
1729         md_bitmap_daemon_work(mddev);
1730         if (mddev->bitmap_info.external)
1731                 md_super_wait(mddev);
1732         md_bitmap_update_sb(bitmap);
1733 }
1734
1735 /*
1736  * free memory that was allocated
1737  */
1738 void md_bitmap_free(struct bitmap *bitmap)
1739 {
1740         unsigned long k, pages;
1741         struct bitmap_page *bp;
1742
1743         if (!bitmap) /* there was no bitmap */
1744                 return;
1745
1746         if (bitmap->sysfs_can_clear)
1747                 sysfs_put(bitmap->sysfs_can_clear);
1748
1749         if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1750                 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1751                 md_cluster_stop(bitmap->mddev);
1752
1753         /* Shouldn't be needed - but just in case.... */
1754         wait_event(bitmap->write_wait,
1755                    atomic_read(&bitmap->pending_writes) == 0);
1756
1757         /* release the bitmap file  */
1758         md_bitmap_file_unmap(&bitmap->storage);
1759
1760         bp = bitmap->counts.bp;
1761         pages = bitmap->counts.pages;
1762
1763         /* free all allocated memory */
1764
1765         if (bp) /* deallocate the page memory */
1766                 for (k = 0; k < pages; k++)
1767                         if (bp[k].map && !bp[k].hijacked)
1768                                 kfree(bp[k].map);
1769         kfree(bp);
1770         kfree(bitmap);
1771 }
1772 EXPORT_SYMBOL(md_bitmap_free);
1773
1774 void md_bitmap_wait_behind_writes(struct mddev *mddev)
1775 {
1776         struct bitmap *bitmap = mddev->bitmap;
1777
1778         /* wait for behind writes to complete */
1779         if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1780                 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1781                          mdname(mddev));
1782                 /* need to kick something here to make sure I/O goes? */
1783                 wait_event(bitmap->behind_wait,
1784                            atomic_read(&bitmap->behind_writes) == 0);
1785         }
1786 }
1787
1788 void md_bitmap_destroy(struct mddev *mddev)
1789 {
1790         struct bitmap *bitmap = mddev->bitmap;
1791
1792         if (!bitmap) /* there was no bitmap */
1793                 return;
1794
1795         md_bitmap_wait_behind_writes(mddev);
1796
1797         mutex_lock(&mddev->bitmap_info.mutex);
1798         spin_lock(&mddev->lock);
1799         mddev->bitmap = NULL; /* disconnect from the md device */
1800         spin_unlock(&mddev->lock);
1801         mutex_unlock(&mddev->bitmap_info.mutex);
1802         if (mddev->thread)
1803                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1804
1805         md_bitmap_free(bitmap);
1806 }
1807
1808 /*
1809  * initialize the bitmap structure
1810  * if this returns an error, bitmap_destroy must be called to do clean up
1811  * once mddev->bitmap is set
1812  */
1813 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1814 {
1815         struct bitmap *bitmap;
1816         sector_t blocks = mddev->resync_max_sectors;
1817         struct file *file = mddev->bitmap_info.file;
1818         int err;
1819         struct kernfs_node *bm = NULL;
1820
1821         BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1822
1823         BUG_ON(file && mddev->bitmap_info.offset);
1824
1825         if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1826                 pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1827                           mdname(mddev));
1828                 return ERR_PTR(-EBUSY);
1829         }
1830
1831         bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1832         if (!bitmap)
1833                 return ERR_PTR(-ENOMEM);
1834
1835         spin_lock_init(&bitmap->counts.lock);
1836         atomic_set(&bitmap->pending_writes, 0);
1837         init_waitqueue_head(&bitmap->write_wait);
1838         init_waitqueue_head(&bitmap->overflow_wait);
1839         init_waitqueue_head(&bitmap->behind_wait);
1840
1841         bitmap->mddev = mddev;
1842         bitmap->cluster_slot = slot;
1843
1844         if (mddev->kobj.sd)
1845                 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1846         if (bm) {
1847                 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1848                 sysfs_put(bm);
1849         } else
1850                 bitmap->sysfs_can_clear = NULL;
1851
1852         bitmap->storage.file = file;
1853         if (file) {
1854                 get_file(file);
1855                 /* As future accesses to this file will use bmap,
1856                  * and bypass the page cache, we must sync the file
1857                  * first.
1858                  */
1859                 vfs_fsync(file, 1);
1860         }
1861         /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1862         if (!mddev->bitmap_info.external) {
1863                 /*
1864                  * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1865                  * instructing us to create a new on-disk bitmap instance.
1866                  */
1867                 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1868                         err = md_bitmap_new_disk_sb(bitmap);
1869                 else
1870                         err = md_bitmap_read_sb(bitmap);
1871         } else {
1872                 err = 0;
1873                 if (mddev->bitmap_info.chunksize == 0 ||
1874                     mddev->bitmap_info.daemon_sleep == 0)
1875                         /* chunksize and time_base need to be
1876                          * set first. */
1877                         err = -EINVAL;
1878         }
1879         if (err)
1880                 goto error;
1881
1882         bitmap->daemon_lastrun = jiffies;
1883         err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1884         if (err)
1885                 goto error;
1886
1887         pr_debug("created bitmap (%lu pages) for device %s\n",
1888                  bitmap->counts.pages, bmname(bitmap));
1889
1890         err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1891         if (err)
1892                 goto error;
1893
1894         return bitmap;
1895  error:
1896         md_bitmap_free(bitmap);
1897         return ERR_PTR(err);
1898 }
1899
1900 int md_bitmap_load(struct mddev *mddev)
1901 {
1902         int err = 0;
1903         sector_t start = 0;
1904         sector_t sector = 0;
1905         struct bitmap *bitmap = mddev->bitmap;
1906
1907         if (!bitmap)
1908                 goto out;
1909
1910         if (mddev_is_clustered(mddev))
1911                 md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1912
1913         /* Clear out old bitmap info first:  Either there is none, or we
1914          * are resuming after someone else has possibly changed things,
1915          * so we should forget old cached info.
1916          * All chunks should be clean, but some might need_sync.
1917          */
1918         while (sector < mddev->resync_max_sectors) {
1919                 sector_t blocks;
1920                 md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1921                 sector += blocks;
1922         }
1923         md_bitmap_close_sync(bitmap);
1924
1925         if (mddev->degraded == 0
1926             || bitmap->events_cleared == mddev->events)
1927                 /* no need to keep dirty bits to optimise a
1928                  * re-add of a missing device */
1929                 start = mddev->recovery_cp;
1930
1931         mutex_lock(&mddev->bitmap_info.mutex);
1932         err = md_bitmap_init_from_disk(bitmap, start);
1933         mutex_unlock(&mddev->bitmap_info.mutex);
1934
1935         if (err)
1936                 goto out;
1937         clear_bit(BITMAP_STALE, &bitmap->flags);
1938
1939         /* Kick recovery in case any bits were set */
1940         set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1941
1942         mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1943         md_wakeup_thread(mddev->thread);
1944
1945         md_bitmap_update_sb(bitmap);
1946
1947         if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1948                 err = -EIO;
1949 out:
1950         return err;
1951 }
1952 EXPORT_SYMBOL_GPL(md_bitmap_load);
1953
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
2018         return rv;
2019 }
2020 EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2021
2022
2023 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2024 {
2025         unsigned long chunk_kb;
2026         struct bitmap_counts *counts;
2027
2028         if (!bitmap)
2029                 return;
2030
2031         counts = &bitmap->counts;
2032
2033         chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2034         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2035                    "%lu%s chunk",
2036                    counts->pages - counts->missing_pages,
2037                    counts->pages,
2038                    (counts->pages - counts->missing_pages)
2039                    << (PAGE_SHIFT - 10),
2040                    chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2041                    chunk_kb ? "KB" : "B");
2042         if (bitmap->storage.file) {
2043                 seq_printf(seq, ", file: ");
2044                 seq_file_path(seq, bitmap->storage.file, " \t\n");
2045         }
2046
2047         seq_printf(seq, "\n");
2048 }
2049
2050 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2051                   int chunksize, int init)
2052 {
2053         /* If chunk_size is 0, choose an appropriate chunk size.
2054          * Then possibly allocate new storage space.
2055          * Then quiesce, copy bits, replace bitmap, and re-start
2056          *
2057          * This function is called both to set up the initial bitmap
2058          * and to resize the bitmap while the array is active.
2059          * If this happens as a result of the array being resized,
2060          * chunksize will be zero, and we need to choose a suitable
2061          * chunksize, otherwise we use what we are given.
2062          */
2063         struct bitmap_storage store;
2064         struct bitmap_counts old_counts;
2065         unsigned long chunks;
2066         sector_t block;
2067         sector_t old_blocks, new_blocks;
2068         int chunkshift;
2069         int ret = 0;
2070         long pages;
2071         struct bitmap_page *new_bp;
2072
2073         if (bitmap->storage.file && !init) {
2074                 pr_info("md: cannot resize file-based bitmap\n");
2075                 return -EINVAL;
2076         }
2077
2078         if (chunksize == 0) {
2079                 /* If there is enough space, leave the chunk size unchanged,
2080                  * else increase by factor of two until there is enough space.
2081                  */
2082                 long bytes;
2083                 long space = bitmap->mddev->bitmap_info.space;
2084
2085                 if (space == 0) {
2086                         /* We don't know how much space there is, so limit
2087                          * to current size - in sectors.
2088                          */
2089                         bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2090                         if (!bitmap->mddev->bitmap_info.external)
2091                                 bytes += sizeof(bitmap_super_t);
2092                         space = DIV_ROUND_UP(bytes, 512);
2093                         bitmap->mddev->bitmap_info.space = space;
2094                 }
2095                 chunkshift = bitmap->counts.chunkshift;
2096                 chunkshift--;
2097                 do {
2098                         /* 'chunkshift' is shift from block size to chunk size */
2099                         chunkshift++;
2100                         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2101                         bytes = DIV_ROUND_UP(chunks, 8);
2102                         if (!bitmap->mddev->bitmap_info.external)
2103                                 bytes += sizeof(bitmap_super_t);
2104                 } while (bytes > (space << 9));
2105         } else
2106                 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2107
2108         chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2109         memset(&store, 0, sizeof(store));
2110         if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2111                 ret = md_bitmap_storage_alloc(&store, chunks,
2112                                               !bitmap->mddev->bitmap_info.external,
2113                                               mddev_is_clustered(bitmap->mddev)
2114                                               ? bitmap->cluster_slot : 0);
2115         if (ret) {
2116                 md_bitmap_file_unmap(&store);
2117                 goto err;
2118         }
2119
2120         pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2121
2122         new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2123         ret = -ENOMEM;
2124         if (!new_bp) {
2125                 md_bitmap_file_unmap(&store);
2126                 goto err;
2127         }
2128
2129         if (!init)
2130                 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2131
2132         store.file = bitmap->storage.file;
2133         bitmap->storage.file = NULL;
2134
2135         if (store.sb_page && bitmap->storage.sb_page)
2136                 memcpy(page_address(store.sb_page),
2137                        page_address(bitmap->storage.sb_page),
2138                        sizeof(bitmap_super_t));
2139         spin_lock_irq(&bitmap->counts.lock);
2140         md_bitmap_file_unmap(&bitmap->storage);
2141         bitmap->storage = store;
2142
2143         old_counts = bitmap->counts;
2144         bitmap->counts.bp = new_bp;
2145         bitmap->counts.pages = pages;
2146         bitmap->counts.missing_pages = pages;
2147         bitmap->counts.chunkshift = chunkshift;
2148         bitmap->counts.chunks = chunks;
2149         bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2150                                                      BITMAP_BLOCK_SHIFT);
2151
2152         blocks = min(old_counts.chunks << old_counts.chunkshift,
2153                      chunks << chunkshift);
2154
2155         /* For cluster raid, need to pre-allocate bitmap */
2156         if (mddev_is_clustered(bitmap->mddev)) {
2157                 unsigned long page;
2158                 for (page = 0; page < pages; page++) {
2159                         ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2160                         if (ret) {
2161                                 unsigned long k;
2162
2163                                 /* deallocate the page memory */
2164                                 for (k = 0; k < page; k++) {
2165                                         kfree(new_bp[k].map);
2166                                 }
2167                                 kfree(new_bp);
2168
2169                                 /* restore some fields from old_counts */
2170                                 bitmap->counts.bp = old_counts.bp;
2171                                 bitmap->counts.pages = old_counts.pages;
2172                                 bitmap->counts.missing_pages = old_counts.pages;
2173                                 bitmap->counts.chunkshift = old_counts.chunkshift;
2174                                 bitmap->counts.chunks = old_counts.chunks;
2175                                 bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2176                                                                              BITMAP_BLOCK_SHIFT);
2177                                 blocks = old_counts.chunks << old_counts.chunkshift;
2178                                 pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2179                                 break;
2180                         } else
2181                                 bitmap->counts.bp[page].count += 1;
2182                 }
2183         }
2184
2185         for (block = 0; block < blocks; ) {
2186                 bitmap_counter_t *bmc_old, *bmc_new;
2187                 int set;
2188
2189                 bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2190                 set = bmc_old && NEEDED(*bmc_old);
2191
2192                 if (set) {
2193                         bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2194                         if (*bmc_new == 0) {
2195                                 /* need to set on-disk bits too. */
2196                                 sector_t end = block + new_blocks;
2197                                 sector_t start = block >> chunkshift;
2198                                 start <<= chunkshift;
2199                                 while (start < end) {
2200                                         md_bitmap_file_set_bit(bitmap, block);
2201                                         start += 1 << chunkshift;
2202                                 }
2203                                 *bmc_new = 2;
2204                                 md_bitmap_count_page(&bitmap->counts, block, 1);
2205                                 md_bitmap_set_pending(&bitmap->counts, block);
2206                         }
2207                         *bmc_new |= NEEDED_MASK;
2208                         if (new_blocks < old_blocks)
2209                                 old_blocks = new_blocks;
2210                 }
2211                 block += old_blocks;
2212         }
2213
2214         if (bitmap->counts.bp != old_counts.bp) {
2215                 unsigned long k;
2216                 for (k = 0; k < old_counts.pages; k++)
2217                         if (!old_counts.bp[k].hijacked)
2218                                 kfree(old_counts.bp[k].map);
2219                 kfree(old_counts.bp);
2220         }
2221
2222         if (!init) {
2223                 int i;
2224                 while (block < (chunks << chunkshift)) {
2225                         bitmap_counter_t *bmc;
2226                         bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2227                         if (bmc) {
2228                                 /* new space.  It needs to be resynced, so
2229                                  * we set NEEDED_MASK.
2230                                  */
2231                                 if (*bmc == 0) {
2232                                         *bmc = NEEDED_MASK | 2;
2233                                         md_bitmap_count_page(&bitmap->counts, block, 1);
2234                                         md_bitmap_set_pending(&bitmap->counts, block);
2235                                 }
2236                         }
2237                         block += new_blocks;
2238                 }
2239                 for (i = 0; i < bitmap->storage.file_pages; i++)
2240                         set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2241         }
2242         spin_unlock_irq(&bitmap->counts.lock);
2243
2244         if (!init) {
2245                 md_bitmap_unplug(bitmap);
2246                 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2247         }
2248         ret = 0;
2249 err:
2250         return ret;
2251 }
2252 EXPORT_SYMBOL_GPL(md_bitmap_resize);
2253
2254 static ssize_t
2255 location_show(struct mddev *mddev, char *page)
2256 {
2257         ssize_t len;
2258         if (mddev->bitmap_info.file)
2259                 len = sprintf(page, "file");
2260         else if (mddev->bitmap_info.offset)
2261                 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2262         else
2263                 len = sprintf(page, "none");
2264         len += sprintf(page+len, "\n");
2265         return len;
2266 }
2267
2268 static ssize_t
2269 location_store(struct mddev *mddev, const char *buf, size_t len)
2270 {
2271         int rv;
2272
2273         rv = mddev_lock(mddev);
2274         if (rv)
2275                 return rv;
2276         if (mddev->pers) {
2277                 if (!mddev->pers->quiesce) {
2278                         rv = -EBUSY;
2279                         goto out;
2280                 }
2281                 if (mddev->recovery || mddev->sync_thread) {
2282                         rv = -EBUSY;
2283                         goto out;
2284                 }
2285         }
2286
2287         if (mddev->bitmap || mddev->bitmap_info.file ||
2288             mddev->bitmap_info.offset) {
2289                 /* bitmap already configured.  Only option is to clear it */
2290                 if (strncmp(buf, "none", 4) != 0) {
2291                         rv = -EBUSY;
2292                         goto out;
2293                 }
2294                 if (mddev->pers) {
2295                         mddev->pers->quiesce(mddev, 1);
2296                         md_bitmap_destroy(mddev);
2297                         mddev->pers->quiesce(mddev, 0);
2298                 }
2299                 mddev->bitmap_info.offset = 0;
2300                 if (mddev->bitmap_info.file) {
2301                         struct file *f = mddev->bitmap_info.file;
2302                         mddev->bitmap_info.file = NULL;
2303                         fput(f);
2304                 }
2305         } else {
2306                 /* No bitmap, OK to set a location */
2307                 long long offset;
2308                 if (strncmp(buf, "none", 4) == 0)
2309                         /* nothing to be done */;
2310                 else if (strncmp(buf, "file:", 5) == 0) {
2311                         /* Not supported yet */
2312                         rv = -EINVAL;
2313                         goto out;
2314                 } else {
2315                         if (buf[0] == '+')
2316                                 rv = kstrtoll(buf+1, 10, &offset);
2317                         else
2318                                 rv = kstrtoll(buf, 10, &offset);
2319                         if (rv)
2320                                 goto out;
2321                         if (offset == 0) {
2322                                 rv = -EINVAL;
2323                                 goto out;
2324                         }
2325                         if (mddev->bitmap_info.external == 0 &&
2326                             mddev->major_version == 0 &&
2327                             offset != mddev->bitmap_info.default_offset) {
2328                                 rv = -EINVAL;
2329                                 goto out;
2330                         }
2331                         mddev->bitmap_info.offset = offset;
2332                         if (mddev->pers) {
2333                                 struct bitmap *bitmap;
2334                                 mddev->pers->quiesce(mddev, 1);
2335                                 bitmap = md_bitmap_create(mddev, -1);
2336                                 if (IS_ERR(bitmap))
2337                                         rv = PTR_ERR(bitmap);
2338                                 else {
2339                                         mddev->bitmap = bitmap;
2340                                         rv = md_bitmap_load(mddev);
2341                                         if (rv)
2342                                                 mddev->bitmap_info.offset = 0;
2343                                 }
2344                                 mddev->pers->quiesce(mddev, 0);
2345                                 if (rv) {
2346                                         md_bitmap_destroy(mddev);
2347                                         goto out;
2348                                 }
2349                         }
2350                 }
2351         }
2352         if (!mddev->external) {
2353                 /* Ensure new bitmap info is stored in
2354                  * metadata promptly.
2355                  */
2356                 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2357                 md_wakeup_thread(mddev->thread);
2358         }
2359         rv = 0;
2360 out:
2361         mddev_unlock(mddev);
2362         if (rv)
2363                 return rv;
2364         return len;
2365 }
2366
2367 static struct md_sysfs_entry bitmap_location =
2368 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2369
2370 /* 'bitmap/space' is the space available at 'location' for the
2371  * bitmap.  This allows the kernel to know when it is safe to
2372  * resize the bitmap to match a resized array.
2373  */
2374 static ssize_t
2375 space_show(struct mddev *mddev, char *page)
2376 {
2377         return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2378 }
2379
2380 static ssize_t
2381 space_store(struct mddev *mddev, const char *buf, size_t len)
2382 {
2383         unsigned long sectors;
2384         int rv;
2385
2386         rv = kstrtoul(buf, 10, &sectors);
2387         if (rv)
2388                 return rv;
2389
2390         if (sectors == 0)
2391                 return -EINVAL;
2392
2393         if (mddev->bitmap &&
2394             sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2395                 return -EFBIG; /* Bitmap is too big for this small space */
2396
2397         /* could make sure it isn't too big, but that isn't really
2398          * needed - user-space should be careful.
2399          */
2400         mddev->bitmap_info.space = sectors;
2401         return len;
2402 }
2403
2404 static struct md_sysfs_entry bitmap_space =
2405 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2406
2407 static ssize_t
2408 timeout_show(struct mddev *mddev, char *page)
2409 {
2410         ssize_t len;
2411         unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2412         unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2413
2414         len = sprintf(page, "%lu", secs);
2415         if (jifs)
2416                 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2417         len += sprintf(page+len, "\n");
2418         return len;
2419 }
2420
2421 static ssize_t
2422 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2423 {
2424         /* timeout can be set at any time */
2425         unsigned long timeout;
2426         int rv = strict_strtoul_scaled(buf, &timeout, 4);
2427         if (rv)
2428                 return rv;
2429
2430         /* just to make sure we don't overflow... */
2431         if (timeout >= LONG_MAX / HZ)
2432                 return -EINVAL;
2433
2434         timeout = timeout * HZ / 10000;
2435
2436         if (timeout >= MAX_SCHEDULE_TIMEOUT)
2437                 timeout = MAX_SCHEDULE_TIMEOUT-1;
2438         if (timeout < 1)
2439                 timeout = 1;
2440         mddev->bitmap_info.daemon_sleep = timeout;
2441         if (mddev->thread) {
2442                 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2443                  * the bitmap is all clean and we don't need to
2444                  * adjust the timeout right now
2445                  */
2446                 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2447                         mddev->thread->timeout = timeout;
2448                         md_wakeup_thread(mddev->thread);
2449                 }
2450         }
2451         return len;
2452 }
2453
2454 static struct md_sysfs_entry bitmap_timeout =
2455 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2456
2457 static ssize_t
2458 backlog_show(struct mddev *mddev, char *page)
2459 {
2460         return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2461 }
2462
2463 static ssize_t
2464 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2465 {
2466         unsigned long backlog;
2467         int rv = kstrtoul(buf, 10, &backlog);
2468         if (rv)
2469                 return rv;
2470         if (backlog > COUNTER_MAX)
2471                 return -EINVAL;
2472         mddev->bitmap_info.max_write_behind = backlog;
2473         return len;
2474 }
2475
2476 static struct md_sysfs_entry bitmap_backlog =
2477 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2478
2479 static ssize_t
2480 chunksize_show(struct mddev *mddev, char *page)
2481 {
2482         return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2483 }
2484
2485 static ssize_t
2486 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2487 {
2488         /* Can only be changed when no bitmap is active */
2489         int rv;
2490         unsigned long csize;
2491         if (mddev->bitmap)
2492                 return -EBUSY;
2493         rv = kstrtoul(buf, 10, &csize);
2494         if (rv)
2495                 return rv;
2496         if (csize < 512 ||
2497             !is_power_of_2(csize))
2498                 return -EINVAL;
2499         mddev->bitmap_info.chunksize = csize;
2500         return len;
2501 }
2502
2503 static struct md_sysfs_entry bitmap_chunksize =
2504 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2505
2506 static ssize_t metadata_show(struct mddev *mddev, char *page)
2507 {
2508         if (mddev_is_clustered(mddev))
2509                 return sprintf(page, "clustered\n");
2510         return sprintf(page, "%s\n", (mddev->bitmap_info.external
2511                                       ? "external" : "internal"));
2512 }
2513
2514 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2515 {
2516         if (mddev->bitmap ||
2517             mddev->bitmap_info.file ||
2518             mddev->bitmap_info.offset)
2519                 return -EBUSY;
2520         if (strncmp(buf, "external", 8) == 0)
2521                 mddev->bitmap_info.external = 1;
2522         else if ((strncmp(buf, "internal", 8) == 0) ||
2523                         (strncmp(buf, "clustered", 9) == 0))
2524                 mddev->bitmap_info.external = 0;
2525         else
2526                 return -EINVAL;
2527         return len;
2528 }
2529
2530 static struct md_sysfs_entry bitmap_metadata =
2531 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2532
2533 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2534 {
2535         int len;
2536         spin_lock(&mddev->lock);
2537         if (mddev->bitmap)
2538                 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2539                                              "false" : "true"));
2540         else
2541                 len = sprintf(page, "\n");
2542         spin_unlock(&mddev->lock);
2543         return len;
2544 }
2545
2546 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2547 {
2548         if (mddev->bitmap == NULL)
2549                 return -ENOENT;
2550         if (strncmp(buf, "false", 5) == 0)
2551                 mddev->bitmap->need_sync = 1;
2552         else if (strncmp(buf, "true", 4) == 0) {
2553                 if (mddev->degraded)
2554                         return -EBUSY;
2555                 mddev->bitmap->need_sync = 0;
2556         } else
2557                 return -EINVAL;
2558         return len;
2559 }
2560
2561 static struct md_sysfs_entry bitmap_can_clear =
2562 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2563
2564 static ssize_t
2565 behind_writes_used_show(struct mddev *mddev, char *page)
2566 {
2567         ssize_t ret;
2568         spin_lock(&mddev->lock);
2569         if (mddev->bitmap == NULL)
2570                 ret = sprintf(page, "0\n");
2571         else
2572                 ret = sprintf(page, "%lu\n",
2573                               mddev->bitmap->behind_writes_used);
2574         spin_unlock(&mddev->lock);
2575         return ret;
2576 }
2577
2578 static ssize_t
2579 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2580 {
2581         if (mddev->bitmap)
2582                 mddev->bitmap->behind_writes_used = 0;
2583         return len;
2584 }
2585
2586 static struct md_sysfs_entry max_backlog_used =
2587 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2588        behind_writes_used_show, behind_writes_used_reset);
2589
2590 static struct attribute *md_bitmap_attrs[] = {
2591         &bitmap_location.attr,
2592         &bitmap_space.attr,
2593         &bitmap_timeout.attr,
2594         &bitmap_backlog.attr,
2595         &bitmap_chunksize.attr,
2596         &bitmap_metadata.attr,
2597         &bitmap_can_clear.attr,
2598         &max_backlog_used.attr,
2599         NULL
2600 };
2601 struct attribute_group md_bitmap_group = {
2602         .name = "bitmap",
2603         .attrs = md_bitmap_attrs,
2604 };
2605