drivers/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  * wait if count gets too high, wake when it drops to half.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/errno.h>
  22 #include <linux/slab.h>
  23 #include <linux/init.h>
  24 #include <linux/timer.h>
  25 #include <linux/sched.h>
  26 #include <linux/list.h>
  27 #include <linux/file.h>
  28 #include <linux/mount.h>
  29 #include <linux/buffer_head.h>
  30 #include "md.h"
  31 #include "bitmap.h"
  32
  33 /* debug macros */
  34
  35 #define DEBUG 0
  36
  37 #if DEBUG
  38 /* these are for debugging purposes only! */
  39
  40 /* define one and only one of these */
  41 #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */
  42 #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/
  43 #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */
  44 #define INJECT_FAULTS_4 0 /* undef */
  45 #define INJECT_FAULTS_5 0 /* undef */
  46 #define INJECT_FAULTS_6 0
  47
  48 /* if these are defined, the driver will fail! debug only */
  49 #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */
  50 #define INJECT_FATAL_FAULT_2 0 /* undef */
  51 #define INJECT_FATAL_FAULT_3 0 /* undef */
  52 #endif
  53
  54 //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */
  55 #define DPRINTK(x...) do { } while(0)
  56
  57 #ifndef PRINTK
  58 #  if DEBUG > 0
  59 #    define PRINTK(x...) printk(KERN_DEBUG x)
  60 #  else
  61 #    define PRINTK(x...)
  62 #  endif
  63 #endif
  64
  65 static inline char * bmname(struct bitmap *bitmap)
  66 {
  67         return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  68 }
  69
  70
  71 /*
  72  * just a placeholder - calls kmalloc for bitmap pages
  73  */
  74 static unsigned char *bitmap_alloc_page(struct bitmap *bitmap)
  75 {
  76         unsigned char *page;
  77
  78 #ifdef INJECT_FAULTS_1
  79         page = NULL;
  80 #else
  81         page = kmalloc(PAGE_SIZE, GFP_NOIO);
  82 #endif
  83         if (!page)
  84                 printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap));
  85         else
  86                 PRINTK("%s: bitmap_alloc_page: allocated page at %p\n",
  87                         bmname(bitmap), page);
  88         return page;
  89 }
  90
  91 /*
  92  * for now just a placeholder -- just calls kfree for bitmap pages
  93  */
  94 static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page)
  95 {
  96         PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page);
  97         kfree(page);
  98 }
  99
 100 /*
 101  * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
 102  *
 103  * 1) check to see if this page is allocated, if it's not then try to alloc
 104  * 2) if the alloc fails, set the page's hijacked flag so we'll use the
 105  *    page pointer directly as a counter
 106  *
 107  * if we find our page, we increment the page's refcount so that it stays
 108  * allocated while we're using it
 109  */
 110 static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create)
 111 __releases(bitmap->lock)
 112 __acquires(bitmap->lock)
 113 {
 114         unsigned char *mappage;
 115
 116         if (page >= bitmap->pages) {
 117                 /* This can happen if bitmap_start_sync goes beyond
 118                  * End-of-device while looking for a whole page.
 119                  * It is harmless.
 120                  */
 121                 return -EINVAL;
 122         }
 123
 124
 125         if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
 126                 return 0;
 127
 128         if (bitmap->bp[page].map) /* page is already allocated, just return */
 129                 return 0;
 130
 131         if (!create)
 132                 return -ENOENT;
 133
 134         spin_unlock_irq(&bitmap->lock);
 135
 136         /* this page has not been allocated yet */
 137
 138         if ((mappage = bitmap_alloc_page(bitmap)) == NULL) {
 139                 PRINTK("%s: bitmap map page allocation failed, hijacking\n",
 140                         bmname(bitmap));
 141                 /* failed - set the hijacked flag so that we can use the
 142                  * pointer as a counter */
 143                 spin_lock_irq(&bitmap->lock);
 144                 if (!bitmap->bp[page].map)
 145                         bitmap->bp[page].hijacked = 1;
 146                 goto out;
 147         }
 148
 149         /* got a page */
 150
 151         spin_lock_irq(&bitmap->lock);
 152
 153         /* recheck the page */
 154
 155         if (bitmap->bp[page].map || bitmap->bp[page].hijacked) {
 156                 /* somebody beat us to getting the page */
 157                 bitmap_free_page(bitmap, mappage);
 158                 return 0;
 159         }
 160
 161         /* no page was in place and we have one, so install it */
 162
 163         memset(mappage, 0, PAGE_SIZE);
 164         bitmap->bp[page].map = mappage;
 165         bitmap->missing_pages--;
 166 out:
 167         return 0;
 168 }
 169
 170
 171 /* if page is completely empty, put it back on the free list, or dealloc it */
 172 /* if page was hijacked, unmark the flag so it might get alloced next time */
 173 /* Note: lock should be held when calling this */
 174 static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page)
 175 {
 176         char *ptr;
 177
 178         if (bitmap->bp[page].count) /* page is still busy */
 179                 return;
 180
 181         /* page is no longer in use, it can be released */
 182
 183         if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 184                 bitmap->bp[page].hijacked = 0;
 185                 bitmap->bp[page].map = NULL;
 186                 return;
 187         }
 188
 189         /* normal case, free the page */
 190
 191 #if 0
 192 /* actually ... let's not.  We will probably need the page again exactly when
 193  * memory is tight and we are flusing to disk
 194  */
 195         return;
 196 #else
 197         ptr = bitmap->bp[page].map;
 198         bitmap->bp[page].map = NULL;
 199         bitmap->missing_pages++;
 200         bitmap_free_page(bitmap, ptr);
 201         return;
 202 #endif
 203 }
 204
 205
 206 /*
 207  * bitmap file handling - read and write the bitmap file and its superblock
 208  */
 209
 210 /*
 211  * basic page I/O operations
 212  */
 213
 214 /* IO operations when bitmap is stored near all superblocks */
 215 static struct page *read_sb_page(mddev_t *mddev, long offset,
 216                                  struct page *page,
 217                                  unsigned long index, int size)
 218 {
 219         /* choose a good rdev and read the page from there */
 220
 221         mdk_rdev_t *rdev;
 222         sector_t target;
 223
 224         if (!page)
 225                 page = alloc_page(GFP_KERNEL);
 226         if (!page)
 227                 return ERR_PTR(-ENOMEM);
 228
 229         list_for_each_entry(rdev, &mddev->disks, same_set) {
 230                 if (! test_bit(In_sync, &rdev->flags)
 231                     || test_bit(Faulty, &rdev->flags))
 232                         continue;
 233
 234                 target = rdev->sb_start + offset + index * (PAGE_SIZE/512);
 235
 236                 if (sync_page_io(rdev->bdev, target,
 237                                  roundup(size, bdev_logical_block_size(rdev->bdev)),
 238                                  page, READ)) {
 239                         page->index = index;
 240                         attach_page_buffers(page, NULL); /* so that free_buffer will
 241                                                           * quietly no-op */
 242                         return page;
 243                 }
 244         }
 245         return ERR_PTR(-EIO);
 246
 247 }
 248
 249 static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
 250 {
 251         /* Iterate the disks of an mddev, using rcu to protect access to the
 252          * linked list, and raising the refcount of devices we return to ensure
 253          * they don't disappear while in use.
 254          * As devices are only added or removed when raid_disk is < 0 and
 255          * nr_pending is 0 and In_sync is clear, the entries we return will
 256          * still be in the same position on the list when we re-enter
 257          * list_for_each_continue_rcu.
 258          */
 259         struct list_head *pos;
 260         rcu_read_lock();
 261         if (rdev == NULL)
 262                 /* start at the beginning */
 263                 pos = &mddev->disks;
 264         else {
 265                 /* release the previous rdev and start from there. */
 266                 rdev_dec_pending(rdev, mddev);
 267                 pos = &rdev->same_set;
 268         }
 269         list_for_each_continue_rcu(pos, &mddev->disks) {
 270                 rdev = list_entry(pos, mdk_rdev_t, same_set);
 271                 if (rdev->raid_disk >= 0 &&
 272                     !test_bit(Faulty, &rdev->flags)) {
 273                         /* this is a usable devices */
 274                         atomic_inc(&rdev->nr_pending);
 275                         rcu_read_unlock();
 276                         return rdev;
 277                 }
 278         }
 279         rcu_read_unlock();
 280         return NULL;
 281 }
 282
 283 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 284 {
 285         mdk_rdev_t *rdev = NULL;
 286         mddev_t *mddev = bitmap->mddev;
 287
 288         while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 289                         int size = PAGE_SIZE;
 290                         if (page->index == bitmap->file_pages-1)
 291                                 size = roundup(bitmap->last_page_size,
 292                                                bdev_logical_block_size(rdev->bdev));
 293                         /* Just make sure we aren't corrupting data or
 294                          * metadata
 295                          */
 296                         if (bitmap->offset < 0) {
 297                                 /* DATA  BITMAP METADATA  */
 298                                 if (bitmap->offset
 299                                     + (long)(page->index * (PAGE_SIZE/512))
 300                                     + size/512 > 0)
 301                                         /* bitmap runs in to metadata */
 302                                         goto bad_alignment;
 303                                 if (rdev->data_offset + mddev->dev_sectors
 304                                     > rdev->sb_start + bitmap->offset)
 305                                         /* data runs in to bitmap */
 306                                         goto bad_alignment;
 307                         } else if (rdev->sb_start < rdev->data_offset) {
 308                                 /* METADATA BITMAP DATA */
 309                                 if (rdev->sb_start
 310                                     + bitmap->offset
 311                                     + page->index*(PAGE_SIZE/512) + size/512
 312                                     > rdev->data_offset)
 313                                         /* bitmap runs in to data */
 314                                         goto bad_alignment;
 315                         } else {
 316                                 /* DATA METADATA BITMAP - no problems */
 317                         }
 318                         md_super_write(mddev, rdev,
 319                                        rdev->sb_start + bitmap->offset
 320                                        + page->index * (PAGE_SIZE/512),
 321                                        size,
 322                                        page);
 323         }
 324
 325         if (wait)
 326                 md_super_wait(mddev);
 327         return 0;
 328
 329  bad_alignment:
 330         return -EINVAL;
 331 }
 332
 333 static void bitmap_file_kick(struct bitmap *bitmap);
 334 /*
 335  * write out a page to a file
 336  */
 337 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 338 {
 339         struct buffer_head *bh;
 340
 341         if (bitmap->file == NULL) {
 342                 switch (write_sb_page(bitmap, page, wait)) {
 343                 case -EINVAL:
 344                         bitmap->flags |= BITMAP_WRITE_ERROR;
 345                 }
 346         } else {
 347
 348                 bh = page_buffers(page);
 349
 350                 while (bh && bh->b_blocknr) {
 351                         atomic_inc(&bitmap->pending_writes);
 352                         set_buffer_locked(bh);
 353                         set_buffer_mapped(bh);
 354                         submit_bh(WRITE, bh);
 355                         bh = bh->b_this_page;
 356                 }
 357
 358                 if (wait) {
 359                         wait_event(bitmap->write_wait,
 360                                    atomic_read(&bitmap->pending_writes)==0);
 361                 }
 362         }
 363         if (bitmap->flags & BITMAP_WRITE_ERROR)
 364                 bitmap_file_kick(bitmap);
 365 }
 366
 367 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 368 {
 369         struct bitmap *bitmap = bh->b_private;
 370         unsigned long flags;
 371
 372         if (!uptodate) {
 373                 spin_lock_irqsave(&bitmap->lock, flags);
 374                 bitmap->flags |= BITMAP_WRITE_ERROR;
 375                 spin_unlock_irqrestore(&bitmap->lock, flags);
 376         }
 377         if (atomic_dec_and_test(&bitmap->pending_writes))
 378                 wake_up(&bitmap->write_wait);
 379 }
 380
 381 /* copied from buffer.c */
 382 static void
 383 __clear_page_buffers(struct page *page)
 384 {
 385         ClearPagePrivate(page);
 386         set_page_private(page, 0);
 387         page_cache_release(page);
 388 }
 389 static void free_buffers(struct page *page)
 390 {
 391         struct buffer_head *bh = page_buffers(page);
 392
 393         while (bh) {
 394                 struct buffer_head *next = bh->b_this_page;
 395                 free_buffer_head(bh);
 396                 bh = next;
 397         }
 398         __clear_page_buffers(page);
 399         put_page(page);
 400 }
 401
 402 /* read a page from a file.
 403  * We both read the page, and attach buffers to the page to record the
 404  * address of each block (using bmap).  These addresses will be used
 405  * to write the block later, completely bypassing the filesystem.
 406  * This usage is similar to how swap files are handled, and allows us
 407  * to write to a file with no concerns of memory allocation failing.
 408  */
 409 static struct page *read_page(struct file *file, unsigned long index,
 410                               struct bitmap *bitmap,
 411                               unsigned long count)
 412 {
 413         struct page *page = NULL;
 414         struct inode *inode = file->f_path.dentry->d_inode;
 415         struct buffer_head *bh;
 416         sector_t block;
 417
 418         PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE,
 419                         (unsigned long long)index << PAGE_SHIFT);
 420
 421         page = alloc_page(GFP_KERNEL);
 422         if (!page)
 423                 page = ERR_PTR(-ENOMEM);
 424         if (IS_ERR(page))
 425                 goto out;
 426
 427         bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
 428         if (!bh) {
 429                 put_page(page);
 430                 page = ERR_PTR(-ENOMEM);
 431                 goto out;
 432         }
 433         attach_page_buffers(page, bh);
 434         block = index << (PAGE_SHIFT - inode->i_blkbits);
 435         while (bh) {
 436                 if (count == 0)
 437                         bh->b_blocknr = 0;
 438                 else {
 439                         bh->b_blocknr = bmap(inode, block);
 440                         if (bh->b_blocknr == 0) {
 441                                 /* Cannot use this file! */
 442                                 free_buffers(page);
 443                                 page = ERR_PTR(-EINVAL);
 444                                 goto out;
 445                         }
 446                         bh->b_bdev = inode->i_sb->s_bdev;
 447                         if (count < (1<<inode->i_blkbits))
 448                                 count = 0;
 449                         else
 450                                 count -= (1<<inode->i_blkbits);
 451
 452                         bh->b_end_io = end_bitmap_write;
 453                         bh->b_private = bitmap;
 454                         atomic_inc(&bitmap->pending_writes);
 455                         set_buffer_locked(bh);
 456                         set_buffer_mapped(bh);
 457                         submit_bh(READ, bh);
 458                 }
 459                 block++;
 460                 bh = bh->b_this_page;
 461         }
 462         page->index = index;
 463
 464         wait_event(bitmap->write_wait,
 465                    atomic_read(&bitmap->pending_writes)==0);
 466         if (bitmap->flags & BITMAP_WRITE_ERROR) {
 467                 free_buffers(page);
 468                 page = ERR_PTR(-EIO);
 469         }
 470 out:
 471         if (IS_ERR(page))
 472                 printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n",
 473                         (int)PAGE_SIZE,
 474                         (unsigned long long)index << PAGE_SHIFT,
 475                         PTR_ERR(page));
 476         return page;
 477 }
 478
 479 /*
 480  * bitmap file superblock operations
 481  */
 482
 483 /* update the event counter and sync the superblock to disk */
 484 void bitmap_update_sb(struct bitmap *bitmap)
 485 {
 486         bitmap_super_t *sb;
 487         unsigned long flags;
 488
 489         if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 490                 return;
 491         spin_lock_irqsave(&bitmap->lock, flags);
 492         if (!bitmap->sb_page) { /* no superblock */
 493                 spin_unlock_irqrestore(&bitmap->lock, flags);
 494                 return;
 495         }
 496         spin_unlock_irqrestore(&bitmap->lock, flags);
 497         sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
 498         sb->events = cpu_to_le64(bitmap->mddev->events);
 499         if (bitmap->mddev->events < bitmap->events_cleared) {
 500                 /* rocking back to read-only */
 501                 bitmap->events_cleared = bitmap->mddev->events;
 502                 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 503         }
 504         kunmap_atomic(sb, KM_USER0);
 505         write_page(bitmap, bitmap->sb_page, 1);
 506 }
 507
 508 /* print out the bitmap file superblock */
 509 void bitmap_print_sb(struct bitmap *bitmap)
 510 {
 511         bitmap_super_t *sb;
 512
 513         if (!bitmap || !bitmap->sb_page)
 514                 return;
 515         sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
 516         printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
 517         printk(KERN_DEBUG "         magic: %08x\n", le32_to_cpu(sb->magic));
 518         printk(KERN_DEBUG "       version: %d\n", le32_to_cpu(sb->version));
 519         printk(KERN_DEBUG "          uuid: %08x.%08x.%08x.%08x\n",
 520                                         *(__u32 *)(sb->uuid+0),
 521                                         *(__u32 *)(sb->uuid+4),
 522                                         *(__u32 *)(sb->uuid+8),
 523                                         *(__u32 *)(sb->uuid+12));
 524         printk(KERN_DEBUG "        events: %llu\n",
 525                         (unsigned long long) le64_to_cpu(sb->events));
 526         printk(KERN_DEBUG "events cleared: %llu\n",
 527                         (unsigned long long) le64_to_cpu(sb->events_cleared));
 528         printk(KERN_DEBUG "         state: %08x\n", le32_to_cpu(sb->state));
 529         printk(KERN_DEBUG "     chunksize: %d B\n", le32_to_cpu(sb->chunksize));
 530         printk(KERN_DEBUG "  daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
 531         printk(KERN_DEBUG "     sync size: %llu KB\n",
 532                         (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 533         printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
 534         kunmap_atomic(sb, KM_USER0);
 535 }
 536
 537 /* read the superblock from the bitmap file and initialize some bitmap fields */
 538 static int bitmap_read_sb(struct bitmap *bitmap)
 539 {
 540         char *reason = NULL;
 541         bitmap_super_t *sb;
 542         unsigned long chunksize, daemon_sleep, write_behind;
 543         unsigned long long events;
 544         int err = -EINVAL;
 545
 546         /* page 0 is the superblock, read it... */
 547         if (bitmap->file) {
 548                 loff_t isize = i_size_read(bitmap->file->f_mapping->host);
 549                 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 550
 551                 bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes);
 552         } else {
 553                 bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset,
 554                                                NULL,
 555                                                0, sizeof(bitmap_super_t));
 556         }
 557         if (IS_ERR(bitmap->sb_page)) {
 558                 err = PTR_ERR(bitmap->sb_page);
 559                 bitmap->sb_page = NULL;
 560                 return err;
 561         }
 562
 563         sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
 564
 565         chunksize = le32_to_cpu(sb->chunksize);
 566         daemon_sleep = le32_to_cpu(sb->daemon_sleep);
 567         write_behind = le32_to_cpu(sb->write_behind);
 568
 569         /* verify that the bitmap-specific fields are valid */
 570         if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 571                 reason = "bad magic";
 572         else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 573                  le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
 574                 reason = "unrecognized superblock version";
 575         else if (chunksize < 512)
 576                 reason = "bitmap chunksize too small";
 577         else if ((1 << ffz(~chunksize)) != chunksize)
 578                 reason = "bitmap chunksize not a power of 2";
 579         else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ)
 580                 reason = "daemon sleep period out of range";
 581         else if (write_behind > COUNTER_MAX)
 582                 reason = "write-behind limit out of range (0 - 16383)";
 583         if (reason) {
 584                 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
 585                         bmname(bitmap), reason);
 586                 goto out;
 587         }
 588
 589         /* keep the array size field of the bitmap superblock up to date */
 590         sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 591
 592         if (!bitmap->mddev->persistent)
 593                 goto success;
 594
 595         /*
 596          * if we have a persistent array superblock, compare the
 597          * bitmap's UUID and event counter to the mddev's
 598          */
 599         if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 600                 printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n",
 601                         bmname(bitmap));
 602                 goto out;
 603         }
 604         events = le64_to_cpu(sb->events);
 605         if (events < bitmap->mddev->events) {
 606                 printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) "
 607                         "-- forcing full recovery\n", bmname(bitmap), events,
 608                         (unsigned long long) bitmap->mddev->events);
 609                 sb->state |= cpu_to_le32(BITMAP_STALE);
 610         }
 611 success:
 612         /* assign fields using values from superblock */
 613         bitmap->chunksize = chunksize;
 614         bitmap->daemon_sleep = daemon_sleep;
 615         bitmap->daemon_lastrun = jiffies;
 616         bitmap->max_write_behind = write_behind;
 617         bitmap->flags |= le32_to_cpu(sb->state);
 618         if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 619                 bitmap->flags |= BITMAP_HOSTENDIAN;
 620         bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 621         if (sb->state & cpu_to_le32(BITMAP_STALE))
 622                 bitmap->events_cleared = bitmap->mddev->events;
 623         err = 0;
 624 out:
 625         kunmap_atomic(sb, KM_USER0);
 626         if (err)
 627                 bitmap_print_sb(bitmap);
 628         return err;
 629 }
 630
 631 enum bitmap_mask_op {
 632         MASK_SET,
 633         MASK_UNSET
 634 };
 635
 636 /* record the state of the bitmap in the superblock.  Return the old value */
 637 static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits,
 638                              enum bitmap_mask_op op)
 639 {
 640         bitmap_super_t *sb;
 641         unsigned long flags;
 642         int old;
 643
 644         spin_lock_irqsave(&bitmap->lock, flags);
 645         if (!bitmap->sb_page) { /* can't set the state */
 646                 spin_unlock_irqrestore(&bitmap->lock, flags);
 647                 return 0;
 648         }
 649         spin_unlock_irqrestore(&bitmap->lock, flags);
 650         sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0);
 651         old = le32_to_cpu(sb->state) & bits;
 652         switch (op) {
 653                 case MASK_SET: sb->state |= cpu_to_le32(bits);
 654                                 break;
 655                 case MASK_UNSET: sb->state &= cpu_to_le32(~bits);
 656                                 break;
 657                 default: BUG();
 658         }
 659         kunmap_atomic(sb, KM_USER0);
 660         return old;
 661 }
 662
 663 /*
 664  * general bitmap file operations
 665  */
 666
 667 /* calculate the index of the page that contains this bit */
 668 static inline unsigned long file_page_index(unsigned long chunk)
 669 {
 670         return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT;
 671 }
 672
 673 /* calculate the (bit) offset of this bit within a page */
 674 static inline unsigned long file_page_offset(unsigned long chunk)
 675 {
 676         return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1);
 677 }
 678
 679 /*
 680  * return a pointer to the page in the filemap that contains the given bit
 681  *
 682  * this lookup is complicated by the fact that the bitmap sb might be exactly
 683  * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page
 684  * 0 or page 1
 685  */
 686 static inline struct page *filemap_get_page(struct bitmap *bitmap,
 687                                         unsigned long chunk)
 688 {
 689         if (file_page_index(chunk) >= bitmap->file_pages) return NULL;
 690         return bitmap->filemap[file_page_index(chunk) - file_page_index(0)];
 691 }
 692
 693
 694 static void bitmap_file_unmap(struct bitmap *bitmap)
 695 {
 696         struct page **map, *sb_page;
 697         unsigned long *attr;
 698         int pages;
 699         unsigned long flags;
 700
 701         spin_lock_irqsave(&bitmap->lock, flags);
 702         map = bitmap->filemap;
 703         bitmap->filemap = NULL;
 704         attr = bitmap->filemap_attr;
 705         bitmap->filemap_attr = NULL;
 706         pages = bitmap->file_pages;
 707         bitmap->file_pages = 0;
 708         sb_page = bitmap->sb_page;
 709         bitmap->sb_page = NULL;
 710         spin_unlock_irqrestore(&bitmap->lock, flags);
 711
 712         while (pages--)
 713                 if (map[pages]->index != 0) /* 0 is sb_page, release it below */
 714                         free_buffers(map[pages]);
 715         kfree(map);
 716         kfree(attr);
 717
 718         if (sb_page)
 719                 free_buffers(sb_page);
 720 }
 721
 722 static void bitmap_file_put(struct bitmap *bitmap)
 723 {
 724         struct file *file;
 725         unsigned long flags;
 726
 727         spin_lock_irqsave(&bitmap->lock, flags);
 728         file = bitmap->file;
 729         bitmap->file = NULL;
 730         spin_unlock_irqrestore(&bitmap->lock, flags);
 731
 732         if (file)
 733                 wait_event(bitmap->write_wait,
 734                            atomic_read(&bitmap->pending_writes)==0);
 735         bitmap_file_unmap(bitmap);
 736
 737         if (file) {
 738                 struct inode *inode = file->f_path.dentry->d_inode;
 739                 invalidate_mapping_pages(inode->i_mapping, 0, -1);
 740                 fput(file);
 741         }
 742 }
 743
 744
 745 /*
 746  * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 747  * then it is no longer reliable, so we stop using it and we mark the file
 748  * as failed in the superblock
 749  */
 750 static void bitmap_file_kick(struct bitmap *bitmap)
 751 {
 752         char *path, *ptr = NULL;
 753
 754         if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) {
 755                 bitmap_update_sb(bitmap);
 756
 757                 if (bitmap->file) {
 758                         path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 759                         if (path)
 760                                 ptr = d_path(&bitmap->file->f_path, path,
 761                                              PAGE_SIZE);
 762
 763
 764                         printk(KERN_ALERT
 765                               "%s: kicking failed bitmap file %s from array!\n",
 766                               bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 767
 768                         kfree(path);
 769                 } else
 770                         printk(KERN_ALERT
 771                                "%s: disabling internal bitmap due to errors\n",
 772                                bmname(bitmap));
 773         }
 774
 775         bitmap_file_put(bitmap);
 776
 777         return;
 778 }
 779
 780 enum bitmap_page_attr {
 781         BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced
 782         BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared
 783         BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced
 784 };
 785
 786 static inline void set_page_attr(struct bitmap *bitmap, struct page *page,
 787                                 enum bitmap_page_attr attr)
 788 {
 789         __set_bit((page->index<<2) + attr, bitmap->filemap_attr);
 790 }
 791
 792 static inline void clear_page_attr(struct bitmap *bitmap, struct page *page,
 793                                 enum bitmap_page_attr attr)
 794 {
 795         __clear_bit((page->index<<2) + attr, bitmap->filemap_attr);
 796 }
 797
 798 static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page,
 799                                            enum bitmap_page_attr attr)
 800 {
 801         return test_bit((page->index<<2) + attr, bitmap->filemap_attr);
 802 }
 803
 804 /*
 805  * bitmap_file_set_bit -- called before performing a write to the md device
 806  * to set (and eventually sync) a particular bit in the bitmap file
 807  *
 808  * we set the bit immediately, then we record the page number so that
 809  * when an unplug occurs, we can flush the dirty pages out to disk
 810  */
 811 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 812 {
 813         unsigned long bit;
 814         struct page *page;
 815         void *kaddr;
 816         unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap);
 817
 818         if (!bitmap->filemap) {
 819                 return;
 820         }
 821
 822         page = filemap_get_page(bitmap, chunk);
 823         if (!page) return;
 824         bit = file_page_offset(chunk);
 825
 826         /* set the bit */
 827         kaddr = kmap_atomic(page, KM_USER0);
 828         if (bitmap->flags & BITMAP_HOSTENDIAN)
 829                 set_bit(bit, kaddr);
 830         else
 831                 ext2_set_bit(bit, kaddr);
 832         kunmap_atomic(kaddr, KM_USER0);
 833         PRINTK("set file bit %lu page %lu\n", bit, page->index);
 834
 835         /* record page number so it gets flushed to disk when unplug occurs */
 836         set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
 837
 838 }
 839
 840 /* this gets called when the md device is ready to unplug its underlying
 841  * (slave) device queues -- before we let any writes go down, we need to
 842  * sync the dirty pages of the bitmap file to disk */
 843 void bitmap_unplug(struct bitmap *bitmap)
 844 {
 845         unsigned long i, flags;
 846         int dirty, need_write;
 847         struct page *page;
 848         int wait = 0;
 849
 850         if (!bitmap)
 851                 return;
 852
 853         /* look at each page to see if there are any set bits that need to be
 854          * flushed out to disk */
 855         for (i = 0; i < bitmap->file_pages; i++) {
 856                 spin_lock_irqsave(&bitmap->lock, flags);
 857                 if (!bitmap->filemap) {
 858                         spin_unlock_irqrestore(&bitmap->lock, flags);
 859                         return;
 860                 }
 861                 page = bitmap->filemap[i];
 862                 dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
 863                 need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
 864                 clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY);
 865                 clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
 866                 if (dirty)
 867                         wait = 1;
 868                 spin_unlock_irqrestore(&bitmap->lock, flags);
 869
 870                 if (dirty | need_write)
 871                         write_page(bitmap, page, 0);
 872         }
 873         if (wait) { /* if any writes were performed, we need to wait on them */
 874                 if (bitmap->file)
 875                         wait_event(bitmap->write_wait,
 876                                    atomic_read(&bitmap->pending_writes)==0);
 877                 else
 878                         md_super_wait(bitmap->mddev);
 879         }
 880         if (bitmap->flags & BITMAP_WRITE_ERROR)
 881                 bitmap_file_kick(bitmap);
 882 }
 883
 884 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
 885 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
 886  * the in-memory bitmap from the on-disk bitmap -- also, sets up the
 887  * memory mapping of the bitmap file
 888  * Special cases:
 889  *   if there's no bitmap file, or if the bitmap file had been
 890  *   previously kicked from the array, we mark all the bits as
 891  *   1's in order to cause a full resync.
 892  *
 893  * We ignore all bits for sectors that end earlier than 'start'.
 894  * This is used when reading an out-of-date bitmap...
 895  */
 896 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
 897 {
 898         unsigned long i, chunks, index, oldindex, bit;
 899         struct page *page = NULL, *oldpage = NULL;
 900         unsigned long num_pages, bit_cnt = 0;
 901         struct file *file;
 902         unsigned long bytes, offset;
 903         int outofdate;
 904         int ret = -ENOSPC;
 905         void *paddr;
 906
 907         chunks = bitmap->chunks;
 908         file = bitmap->file;
 909
 910         BUG_ON(!file && !bitmap->offset);
 911
 912 #ifdef INJECT_FAULTS_3
 913         outofdate = 1;
 914 #else
 915         outofdate = bitmap->flags & BITMAP_STALE;
 916 #endif
 917         if (outofdate)
 918                 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
 919                         "recovery\n", bmname(bitmap));
 920
 921         bytes = (chunks + 7) / 8;
 922
 923         num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
 924
 925         if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
 926                 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
 927                         bmname(bitmap),
 928                         (unsigned long) i_size_read(file->f_mapping->host),
 929                         bytes + sizeof(bitmap_super_t));
 930                 goto err;
 931         }
 932
 933         ret = -ENOMEM;
 934
 935         bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
 936         if (!bitmap->filemap)
 937                 goto err;
 938
 939         /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
 940         bitmap->filemap_attr = kzalloc(
 941                 roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 942                 GFP_KERNEL);
 943         if (!bitmap->filemap_attr)
 944                 goto err;
 945
 946         oldindex = ~0L;
 947
 948         for (i = 0; i < chunks; i++) {
 949                 int b;
 950                 index = file_page_index(i);
 951                 bit = file_page_offset(i);
 952                 if (index != oldindex) { /* this is a new page, read it in */
 953                         int count;
 954                         /* unmap the old page, we're done with it */
 955                         if (index == num_pages-1)
 956                                 count = bytes + sizeof(bitmap_super_t)
 957                                         - index * PAGE_SIZE;
 958                         else
 959                                 count = PAGE_SIZE;
 960                         if (index == 0) {
 961                                 /*
 962                                  * if we're here then the superblock page
 963                                  * contains some bits (PAGE_SIZE != sizeof sb)
 964                                  * we've already read it in, so just use it
 965                                  */
 966                                 page = bitmap->sb_page;
 967                                 offset = sizeof(bitmap_super_t);
 968                                 if (!file)
 969                                         read_sb_page(bitmap->mddev,
 970                                                      bitmap->offset,
 971                                                      page,
 972                                                      index, count);
 973                         } else if (file) {
 974                                 page = read_page(file, index, bitmap, count);
 975                                 offset = 0;
 976                         } else {
 977                                 page = read_sb_page(bitmap->mddev, bitmap->offset,
 978                                                     NULL,
 979                                                     index, count);
 980                                 offset = 0;
 981                         }
 982                         if (IS_ERR(page)) { /* read error */
 983                                 ret = PTR_ERR(page);
 984                                 goto err;
 985                         }
 986
 987                         oldindex = index;
 988                         oldpage = page;
 989
 990                         bitmap->filemap[bitmap->file_pages++] = page;
 991                         bitmap->last_page_size = count;
 992
 993                         if (outofdate) {
 994                                 /*
 995                                  * if bitmap is out of date, dirty the
 996                                  * whole page and write it out
 997                                  */
 998                                 paddr = kmap_atomic(page, KM_USER0);
 999                                 memset(paddr + offset, 0xff,
1000                                        PAGE_SIZE - offset);
1001                                 kunmap_atomic(paddr, KM_USER0);
1002                                 write_page(bitmap, page, 1);
1003
1004                                 ret = -EIO;
1005                                 if (bitmap->flags & BITMAP_WRITE_ERROR)
1006                                         goto err;
1007                         }
1008                 }
1009                 paddr = kmap_atomic(page, KM_USER0);
1010                 if (bitmap->flags & BITMAP_HOSTENDIAN)
1011                         b = test_bit(bit, paddr);
1012                 else
1013                         b = ext2_test_bit(bit, paddr);
1014                 kunmap_atomic(paddr, KM_USER0);
1015                 if (b) {
1016                         /* if the disk bit is set, set the memory bit */
1017                         int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap))
1018                                       >= start);
1019                         bitmap_set_memory_bits(bitmap,
1020                                                (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap),
1021                                                needed);
1022                         bit_cnt++;
1023                         set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1024                 }
1025         }
1026
1027         /* everything went OK */
1028         ret = 0;
1029         bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
1030
1031         if (bit_cnt) { /* Kick recovery if any bits were set */
1032                 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1033                 md_wakeup_thread(bitmap->mddev->thread);
1034         }
1035
1036         printk(KERN_INFO "%s: bitmap initialized from disk: "
1037                 "read %lu/%lu pages, set %lu bits\n",
1038                 bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt);
1039
1040         return 0;
1041
1042  err:
1043         printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1044                bmname(bitmap), ret);
1045         return ret;
1046 }
1047
1048 void bitmap_write_all(struct bitmap *bitmap)
1049 {
1050         /* We don't actually write all bitmap blocks here,
1051          * just flag them as needing to be written
1052          */
1053         int i;
1054
1055         for (i=0; i < bitmap->file_pages; i++)
1056                 set_page_attr(bitmap, bitmap->filemap[i],
1057                               BITMAP_PAGE_NEEDWRITE);
1058 }
1059
1060
1061 static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc)
1062 {
1063         sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1064         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1065         bitmap->bp[page].count += inc;
1066 /*
1067         if (page == 0) printk("count page 0, offset %llu: %d gives %d\n",
1068                               (unsigned long long)offset, inc, bitmap->bp[page].count);
1069 */
1070         bitmap_checkfree(bitmap, page);
1071 }
1072 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1073                                             sector_t offset, int *blocks,
1074                                             int create);
1075
1076 /*
1077  * bitmap daemon -- periodically wakes up to clean bits and flush pages
1078  *                      out to disk
1079  */
1080
1081 void bitmap_daemon_work(struct bitmap *bitmap)
1082 {
1083         unsigned long j;
1084         unsigned long flags;
1085         struct page *page = NULL, *lastpage = NULL;
1086         int blocks;
1087         void *paddr;
1088
1089         if (bitmap == NULL)
1090                 return;
1091         if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ))
1092                 goto done;
1093
1094         bitmap->daemon_lastrun = jiffies;
1095         if (bitmap->allclean) {
1096                 bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1097                 return;
1098         }
1099         bitmap->allclean = 1;
1100
1101         spin_lock_irqsave(&bitmap->lock, flags);
1102         for (j = 0; j < bitmap->chunks; j++) {
1103                 bitmap_counter_t *bmc;
1104                 if (!bitmap->filemap)
1105                         /* error or shutdown */
1106                         break;
1107
1108                 page = filemap_get_page(bitmap, j);
1109
1110                 if (page != lastpage) {
1111                         /* skip this page unless it's marked as needing cleaning */
1112                         if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) {
1113                                 int need_write = test_page_attr(bitmap, page,
1114                                                                 BITMAP_PAGE_NEEDWRITE);
1115                                 if (need_write)
1116                                         clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE);
1117
1118                                 spin_unlock_irqrestore(&bitmap->lock, flags);
1119                                 if (need_write) {
1120                                         write_page(bitmap, page, 0);
1121                                         bitmap->allclean = 0;
1122                                 }
1123                                 spin_lock_irqsave(&bitmap->lock, flags);
1124                                 j |= (PAGE_BITS - 1);
1125                                 continue;
1126                         }
1127
1128                         /* grab the new page, sync and release the old */
1129                         if (lastpage != NULL) {
1130                                 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1131                                         clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1132                                         spin_unlock_irqrestore(&bitmap->lock, flags);
1133                                         write_page(bitmap, lastpage, 0);
1134                                 } else {
1135                                         set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1136                                         spin_unlock_irqrestore(&bitmap->lock, flags);
1137                                 }
1138                         } else
1139                                 spin_unlock_irqrestore(&bitmap->lock, flags);
1140                         lastpage = page;
1141
1142                         /* We are possibly going to clear some bits, so make
1143                          * sure that events_cleared is up-to-date.
1144                          */
1145                         if (bitmap->need_sync) {
1146                                 bitmap_super_t *sb;
1147                                 bitmap->need_sync = 0;
1148                                 sb = kmap_atomic(bitmap->sb_page, KM_USER0);
1149                                 sb->events_cleared =
1150                                         cpu_to_le64(bitmap->events_cleared);
1151                                 kunmap_atomic(sb, KM_USER0);
1152                                 write_page(bitmap, bitmap->sb_page, 1);
1153                         }
1154                         spin_lock_irqsave(&bitmap->lock, flags);
1155                         clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1156                 }
1157                 bmc = bitmap_get_counter(bitmap,
1158                                          (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1159                                          &blocks, 0);
1160                 if (bmc) {
1161 /*
1162   if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc);
1163 */
1164                         if (*bmc)
1165                                 bitmap->allclean = 0;
1166
1167                         if (*bmc == 2) {
1168                                 *bmc=1; /* maybe clear the bit next time */
1169                                 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1170                         } else if (*bmc == 1) {
1171                                 /* we can clear the bit */
1172                                 *bmc = 0;
1173                                 bitmap_count_page(bitmap,
1174                                                   (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap),
1175                                                   -1);
1176
1177                                 /* clear the bit */
1178                                 paddr = kmap_atomic(page, KM_USER0);
1179                                 if (bitmap->flags & BITMAP_HOSTENDIAN)
1180                                         clear_bit(file_page_offset(j), paddr);
1181                                 else
1182                                         ext2_clear_bit(file_page_offset(j), paddr);
1183                                 kunmap_atomic(paddr, KM_USER0);
1184                         }
1185                 } else
1186                         j |= PAGE_COUNTER_MASK;
1187         }
1188         spin_unlock_irqrestore(&bitmap->lock, flags);
1189
1190         /* now sync the final page */
1191         if (lastpage != NULL) {
1192                 spin_lock_irqsave(&bitmap->lock, flags);
1193                 if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) {
1194                         clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1195                         spin_unlock_irqrestore(&bitmap->lock, flags);
1196                         write_page(bitmap, lastpage, 0);
1197                 } else {
1198                         set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE);
1199                         spin_unlock_irqrestore(&bitmap->lock, flags);
1200                 }
1201         }
1202
1203  done:
1204         if (bitmap->allclean == 0)
1205                 bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1206 }
1207
1208 static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap,
1209                                             sector_t offset, int *blocks,
1210                                             int create)
1211 __releases(bitmap->lock)
1212 __acquires(bitmap->lock)
1213 {
1214         /* If 'create', we might release the lock and reclaim it.
1215          * The lock must have been taken with interrupts enabled.
1216          * If !create, we don't release the lock.
1217          */
1218         sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap);
1219         unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1220         unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1221         sector_t csize;
1222
1223         if (bitmap_checkpage(bitmap, page, create) < 0) {
1224                 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1225                 *blocks = csize - (offset & (csize- 1));
1226                 return NULL;
1227         }
1228         /* now locked ... */
1229
1230         if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1231                 /* should we use the first or second counter field
1232                  * of the hijacked pointer? */
1233                 int hi = (pageoff > PAGE_COUNTER_MASK);
1234                 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) +
1235                                           PAGE_COUNTER_SHIFT - 1);
1236                 *blocks = csize - (offset & (csize- 1));
1237                 return  &((bitmap_counter_t *)
1238                           &bitmap->bp[page].map)[hi];
1239         } else { /* page is allocated */
1240                 csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap));
1241                 *blocks = csize - (offset & (csize- 1));
1242                 return (bitmap_counter_t *)
1243                         &(bitmap->bp[page].map[pageoff]);
1244         }
1245 }
1246
1247 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1248 {
1249         if (!bitmap) return 0;
1250
1251         if (behind) {
1252                 atomic_inc(&bitmap->behind_writes);
1253                 PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n",
1254                   atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1255         }
1256
1257         while (sectors) {
1258                 int blocks;
1259                 bitmap_counter_t *bmc;
1260
1261                 spin_lock_irq(&bitmap->lock);
1262                 bmc = bitmap_get_counter(bitmap, offset, &blocks, 1);
1263                 if (!bmc) {
1264                         spin_unlock_irq(&bitmap->lock);
1265                         return 0;
1266                 }
1267
1268                 if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) {
1269                         DEFINE_WAIT(__wait);
1270                         /* note that it is safe to do the prepare_to_wait
1271                          * after the test as long as we do it before dropping
1272                          * the spinlock.
1273                          */
1274                         prepare_to_wait(&bitmap->overflow_wait, &__wait,
1275                                         TASK_UNINTERRUPTIBLE);
1276                         spin_unlock_irq(&bitmap->lock);
1277                         blk_unplug(bitmap->mddev->queue);
1278                         schedule();
1279                         finish_wait(&bitmap->overflow_wait, &__wait);
1280                         continue;
1281                 }
1282
1283                 switch(*bmc) {
1284                 case 0:
1285                         bitmap_file_set_bit(bitmap, offset);
1286                         bitmap_count_page(bitmap,offset, 1);
1287                         blk_plug_device_unlocked(bitmap->mddev->queue);
1288                         /* fall through */
1289                 case 1:
1290                         *bmc = 2;
1291                 }
1292
1293                 (*bmc)++;
1294
1295                 spin_unlock_irq(&bitmap->lock);
1296
1297                 offset += blocks;
1298                 if (sectors > blocks)
1299                         sectors -= blocks;
1300                 else sectors = 0;
1301         }
1302         bitmap->allclean = 0;
1303         return 0;
1304 }
1305
1306 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1307                      int success, int behind)
1308 {
1309         if (!bitmap) return;
1310         if (behind) {
1311                 atomic_dec(&bitmap->behind_writes);
1312                 PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n",
1313                   atomic_read(&bitmap->behind_writes), bitmap->max_write_behind);
1314         }
1315         if (bitmap->mddev->degraded)
1316                 /* Never clear bits or update events_cleared when degraded */
1317                 success = 0;
1318
1319         while (sectors) {
1320                 int blocks;
1321                 unsigned long flags;
1322                 bitmap_counter_t *bmc;
1323
1324                 spin_lock_irqsave(&bitmap->lock, flags);
1325                 bmc = bitmap_get_counter(bitmap, offset, &blocks, 0);
1326                 if (!bmc) {
1327                         spin_unlock_irqrestore(&bitmap->lock, flags);
1328                         return;
1329                 }
1330
1331                 if (success &&
1332                     bitmap->events_cleared < bitmap->mddev->events) {
1333                         bitmap->events_cleared = bitmap->mddev->events;
1334                         bitmap->need_sync = 1;
1335                 }
1336
1337                 if (!success && ! (*bmc & NEEDED_MASK))
1338                         *bmc |= NEEDED_MASK;
1339
1340                 if ((*bmc & COUNTER_MAX) == COUNTER_MAX)
1341                         wake_up(&bitmap->overflow_wait);
1342
1343                 (*bmc)--;
1344                 if (*bmc <= 2) {
1345                         set_page_attr(bitmap,
1346                                       filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1347                                       BITMAP_PAGE_CLEAN);
1348                 }
1349                 spin_unlock_irqrestore(&bitmap->lock, flags);
1350                 offset += blocks;
1351                 if (sectors > blocks)
1352                         sectors -= blocks;
1353                 else sectors = 0;
1354         }
1355 }
1356
1357 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1358                                int degraded)
1359 {
1360         bitmap_counter_t *bmc;
1361         int rv;
1362         if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1363                 *blocks = 1024;
1364                 return 1; /* always resync if no bitmap */
1365         }
1366         spin_lock_irq(&bitmap->lock);
1367         bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1368         rv = 0;
1369         if (bmc) {
1370                 /* locked */
1371                 if (RESYNC(*bmc))
1372                         rv = 1;
1373                 else if (NEEDED(*bmc)) {
1374                         rv = 1;
1375                         if (!degraded) { /* don't set/clear bits if degraded */
1376                                 *bmc |= RESYNC_MASK;
1377                                 *bmc &= ~NEEDED_MASK;
1378                         }
1379                 }
1380         }
1381         spin_unlock_irq(&bitmap->lock);
1382         bitmap->allclean = 0;
1383         return rv;
1384 }
1385
1386 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks,
1387                       int degraded)
1388 {
1389         /* bitmap_start_sync must always report on multiples of whole
1390          * pages, otherwise resync (which is very PAGE_SIZE based) will
1391          * get confused.
1392          * So call __bitmap_start_sync repeatedly (if needed) until
1393          * At least PAGE_SIZE>>9 blocks are covered.
1394          * Return the 'or' of the result.
1395          */
1396         int rv = 0;
1397         int blocks1;
1398
1399         *blocks = 0;
1400         while (*blocks < (PAGE_SIZE>>9)) {
1401                 rv |= __bitmap_start_sync(bitmap, offset,
1402                                           &blocks1, degraded);
1403                 offset += blocks1;
1404                 *blocks += blocks1;
1405         }
1406         return rv;
1407 }
1408
1409 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted)
1410 {
1411         bitmap_counter_t *bmc;
1412         unsigned long flags;
1413 /*
1414         if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted);
1415 */      if (bitmap == NULL) {
1416                 *blocks = 1024;
1417                 return;
1418         }
1419         spin_lock_irqsave(&bitmap->lock, flags);
1420         bmc = bitmap_get_counter(bitmap, offset, blocks, 0);
1421         if (bmc == NULL)
1422                 goto unlock;
1423         /* locked */
1424 /*
1425         if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks);
1426 */
1427         if (RESYNC(*bmc)) {
1428                 *bmc &= ~RESYNC_MASK;
1429
1430                 if (!NEEDED(*bmc) && aborted)
1431                         *bmc |= NEEDED_MASK;
1432                 else {
1433                         if (*bmc <= 2) {
1434                                 set_page_attr(bitmap,
1435                                               filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)),
1436                                               BITMAP_PAGE_CLEAN);
1437                         }
1438                 }
1439         }
1440  unlock:
1441         spin_unlock_irqrestore(&bitmap->lock, flags);
1442         bitmap->allclean = 0;
1443 }
1444
1445 void bitmap_close_sync(struct bitmap *bitmap)
1446 {
1447         /* Sync has finished, and any bitmap chunks that weren't synced
1448          * properly have been aborted.  It remains to us to clear the
1449          * RESYNC bit wherever it is still on
1450          */
1451         sector_t sector = 0;
1452         int blocks;
1453         if (!bitmap)
1454                 return;
1455         while (sector < bitmap->mddev->resync_max_sectors) {
1456                 bitmap_end_sync(bitmap, sector, &blocks, 0);
1457                 sector += blocks;
1458         }
1459 }
1460
1461 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1462 {
1463         sector_t s = 0;
1464         int blocks;
1465
1466         if (!bitmap)
1467                 return;
1468         if (sector == 0) {
1469                 bitmap->last_end_sync = jiffies;
1470                 return;
1471         }
1472         if (time_before(jiffies, (bitmap->last_end_sync
1473                                   + bitmap->daemon_sleep * HZ)))
1474                 return;
1475         wait_event(bitmap->mddev->recovery_wait,
1476                    atomic_read(&bitmap->mddev->recovery_active) == 0);
1477
1478         bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync;
1479         set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1480         sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1);
1481         s = 0;
1482         while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1483                 bitmap_end_sync(bitmap, s, &blocks, 0);
1484                 s += blocks;
1485         }
1486         bitmap->last_end_sync = jiffies;
1487         sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1488 }
1489
1490 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1491 {
1492         /* For each chunk covered by any of these sectors, set the
1493          * counter to 1 and set resync_needed.  They should all
1494          * be 0 at this point
1495          */
1496
1497         int secs;
1498         bitmap_counter_t *bmc;
1499         spin_lock_irq(&bitmap->lock);
1500         bmc = bitmap_get_counter(bitmap, offset, &secs, 1);
1501         if (!bmc) {
1502                 spin_unlock_irq(&bitmap->lock);
1503                 return;
1504         }
1505         if (! *bmc) {
1506                 struct page *page;
1507                 *bmc = 1 | (needed?NEEDED_MASK:0);
1508                 bitmap_count_page(bitmap, offset, 1);
1509                 page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap));
1510                 set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
1511         }
1512         spin_unlock_irq(&bitmap->lock);
1513         bitmap->allclean = 0;
1514 }
1515
1516 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1517 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1518 {
1519         unsigned long chunk;
1520
1521         for (chunk = s; chunk <= e; chunk++) {
1522                 sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap);
1523                 bitmap_set_memory_bits(bitmap, sec, 1);
1524                 bitmap_file_set_bit(bitmap, sec);
1525         }
1526 }
1527
1528 /*
1529  * flush out any pending updates
1530  */
1531 void bitmap_flush(mddev_t *mddev)
1532 {
1533         struct bitmap *bitmap = mddev->bitmap;
1534         int sleep;
1535
1536         if (!bitmap) /* there was no bitmap */
1537                 return;
1538
1539         /* run the daemon_work three time to ensure everything is flushed
1540          * that can be
1541          */
1542         sleep = bitmap->daemon_sleep;
1543         bitmap->daemon_sleep = 0;
1544         bitmap_daemon_work(bitmap);
1545         bitmap_daemon_work(bitmap);
1546         bitmap_daemon_work(bitmap);
1547         bitmap->daemon_sleep = sleep;
1548         bitmap_update_sb(bitmap);
1549 }
1550
1551 /*
1552  * free memory that was allocated
1553  */
1554 static void bitmap_free(struct bitmap *bitmap)
1555 {
1556         unsigned long k, pages;
1557         struct bitmap_page *bp;
1558
1559         if (!bitmap) /* there was no bitmap */
1560                 return;
1561
1562         /* release the bitmap file and kill the daemon */
1563         bitmap_file_put(bitmap);
1564
1565         bp = bitmap->bp;
1566         pages = bitmap->pages;
1567
1568         /* free all allocated memory */
1569
1570         if (bp) /* deallocate the page memory */
1571                 for (k = 0; k < pages; k++)
1572                         if (bp[k].map && !bp[k].hijacked)
1573                                 kfree(bp[k].map);
1574         kfree(bp);
1575         kfree(bitmap);
1576 }
1577 void bitmap_destroy(mddev_t *mddev)
1578 {
1579         struct bitmap *bitmap = mddev->bitmap;
1580
1581         if (!bitmap) /* there was no bitmap */
1582                 return;
1583
1584         mddev->bitmap = NULL; /* disconnect from the md device */
1585         if (mddev->thread)
1586                 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1587
1588         bitmap_free(bitmap);
1589 }
1590
1591 /*
1592  * initialize the bitmap structure
1593  * if this returns an error, bitmap_destroy must be called to do clean up
1594  */
1595 int bitmap_create(mddev_t *mddev)
1596 {
1597         struct bitmap *bitmap;
1598         sector_t blocks = mddev->resync_max_sectors;
1599         unsigned long chunks;
1600         unsigned long pages;
1601         struct file *file = mddev->bitmap_file;
1602         int err;
1603         sector_t start;
1604
1605         BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1606
1607         if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */
1608                 return 0;
1609
1610         BUG_ON(file && mddev->bitmap_offset);
1611
1612         bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1613         if (!bitmap)
1614                 return -ENOMEM;
1615
1616         spin_lock_init(&bitmap->lock);
1617         atomic_set(&bitmap->pending_writes, 0);
1618         init_waitqueue_head(&bitmap->write_wait);
1619         init_waitqueue_head(&bitmap->overflow_wait);
1620
1621         bitmap->mddev = mddev;
1622
1623         bitmap->file = file;
1624         bitmap->offset = mddev->bitmap_offset;
1625         if (file) {
1626                 get_file(file);
1627                 do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX,
1628                                       SYNC_FILE_RANGE_WAIT_BEFORE |
1629                                       SYNC_FILE_RANGE_WRITE |
1630                                       SYNC_FILE_RANGE_WAIT_AFTER);
1631         }
1632         /* read superblock from bitmap file (this sets bitmap->chunksize) */
1633         err = bitmap_read_sb(bitmap);
1634         if (err)
1635                 goto error;
1636
1637         bitmap->chunkshift = ffz(~bitmap->chunksize);
1638
1639         /* now that chunksize and chunkshift are set, we can use these macros */
1640         chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >>
1641                         CHUNK_BLOCK_SHIFT(bitmap);
1642         pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO;
1643
1644         BUG_ON(!pages);
1645
1646         bitmap->chunks = chunks;
1647         bitmap->pages = pages;
1648         bitmap->missing_pages = pages;
1649         bitmap->counter_bits = COUNTER_BITS;
1650
1651         bitmap->syncchunk = ~0UL;
1652
1653 #ifdef INJECT_FATAL_FAULT_1
1654         bitmap->bp = NULL;
1655 #else
1656         bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL);
1657 #endif
1658         err = -ENOMEM;
1659         if (!bitmap->bp)
1660                 goto error;
1661
1662         /* now that we have some pages available, initialize the in-memory
1663          * bitmap from the on-disk bitmap */
1664         start = 0;
1665         if (mddev->degraded == 0
1666             || bitmap->events_cleared == mddev->events)
1667                 /* no need to keep dirty bits to optimise a re-add of a missing device */
1668                 start = mddev->recovery_cp;
1669         err = bitmap_init_from_disk(bitmap, start);
1670
1671         if (err)
1672                 goto error;
1673
1674         printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1675                 pages, bmname(bitmap));
1676
1677         mddev->bitmap = bitmap;
1678
1679         mddev->thread->timeout = bitmap->daemon_sleep * HZ;
1680
1681         bitmap_update_sb(bitmap);
1682
1683         return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0;
1684
1685  error:
1686         bitmap_free(bitmap);
1687         return err;
1688 }
1689
1690 /* the bitmap API -- for raid personalities */
1691 EXPORT_SYMBOL(bitmap_startwrite);
1692 EXPORT_SYMBOL(bitmap_endwrite);
1693 EXPORT_SYMBOL(bitmap_start_sync);
1694 EXPORT_SYMBOL(bitmap_end_sync);
1695 EXPORT_SYMBOL(bitmap_unplug);
1696 EXPORT_SYMBOL(bitmap_close_sync);
1697 EXPORT_SYMBOL(bitmap_cond_end_sync);