Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   4 *
   5 * bitmap_create  - sets up the bitmap structure
   6 * bitmap_destroy - destroys the bitmap structure
   7 *
   8 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   9 * - added disk storage for bitmap
  10 * - changes to allow various bitmap chunk sizes
  11 */
  12
  13/*
  14 * Still to do:
  15 *
  16 * flush after percent set rather than just time based. (maybe both).
  17 */
  18
  19#include <linux/blkdev.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/slab.h>
  23#include <linux/init.h>
  24#include <linux/timer.h>
  25#include <linux/sched.h>
  26#include <linux/list.h>
  27#include <linux/file.h>
  28#include <linux/mount.h>
  29#include <linux/buffer_head.h>
  30#include <linux/seq_file.h>
  31#include <trace/events/block.h>
  32#include "md.h"
  33#include "md-bitmap.h"
  34
  35#define BITMAP_MAJOR_LO 3
  36/* version 4 insists the bitmap is in little-endian order
  37 * with version 3, it is host-endian which is non-portable
  38 * Version 5 is currently set only for clustered devices
  39 */
  40#define BITMAP_MAJOR_HI 4
  41#define BITMAP_MAJOR_CLUSTERED 5
  42#define	BITMAP_MAJOR_HOSTENDIAN 3
  43
  44/*
  45 * in-memory bitmap:
  46 *
  47 * Use 16 bit block counters to track pending writes to each "chunk".
  48 * The 2 high order bits are special-purpose, the first is a flag indicating
  49 * whether a resync is needed.  The second is a flag indicating whether a
  50 * resync is active.
  51 * This means that the counter is actually 14 bits:
  52 *
  53 * +--------+--------+------------------------------------------------+
  54 * | resync | resync |               counter                          |
  55 * | needed | active |                                                |
  56 * |  (0-1) |  (0-1) |              (0-16383)                         |
  57 * +--------+--------+------------------------------------------------+
  58 *
  59 * The "resync needed" bit is set when:
  60 *    a '1' bit is read from storage at startup.
  61 *    a write request fails on some drives
  62 *    a resync is aborted on a chunk with 'resync active' set
  63 * It is cleared (and resync-active set) when a resync starts across all drives
  64 * of the chunk.
  65 *
  66 *
  67 * The "resync active" bit is set when:
  68 *    a resync is started on all drives, and resync_needed is set.
  69 *       resync_needed will be cleared (as long as resync_active wasn't already set).
  70 * It is cleared when a resync completes.
  71 *
  72 * The counter counts pending write requests, plus the on-disk bit.
  73 * When the counter is '1' and the resync bits are clear, the on-disk
  74 * bit can be cleared as well, thus setting the counter to 0.
  75 * When we set a bit, or in the counter (to start a write), if the fields is
  76 * 0, we first set the disk bit and set the counter to 1.
  77 *
  78 * If the counter is 0, the on-disk bit is clear and the stripe is clean
  79 * Anything that dirties the stripe pushes the counter to 2 (at least)
  80 * and sets the on-disk bit (lazily).
  81 * If a periodic sweep find the counter at 2, it is decremented to 1.
  82 * If the sweep find the counter at 1, the on-disk bit is cleared and the
  83 * counter goes to zero.
  84 *
  85 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
  86 * counters as a fallback when "page" memory cannot be allocated:
  87 *
  88 * Normal case (page memory allocated):
  89 *
  90 *     page pointer (32-bit)
  91 *
  92 *     [ ] ------+
  93 *               |
  94 *               +-------> [   ][   ]..[   ] (4096 byte page == 2048 counters)
  95 *                          c1   c2    c2048
  96 *
  97 * Hijacked case (page memory allocation failed):
  98 *
  99 *     hijacked page pointer (32-bit)
 100 *
 101 *     [		  ][		  ] (no page memory allocated)
 102 *      counter #1 (16-bit) counter #2 (16-bit)
 103 *
 104 */
 105
 106#define PAGE_BITS (PAGE_SIZE << 3)
 107#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
 108
 109#define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
 110#define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
 111#define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
 112
 113/* how many counters per page? */
 114#define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
 115/* same, except a shift value for more efficient bitops */
 116#define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
 117/* same, except a mask value for more efficient bitops */
 118#define PAGE_COUNTER_MASK  (PAGE_COUNTER_RATIO - 1)
 119
 120#define BITMAP_BLOCK_SHIFT 9
 121
 122/*
 123 * bitmap structures:
 124 */
 125
 126/* the in-memory bitmap is represented by bitmap_pages */
 127struct bitmap_page {
 128	/*
 129	 * map points to the actual memory page
 130	 */
 131	char *map;
 132	/*
 133	 * in emergencies (when map cannot be alloced), hijack the map
 134	 * pointer and use it as two counters itself
 135	 */
 136	unsigned int hijacked:1;
 137	/*
 138	 * If any counter in this page is '1' or '2' - and so could be
 139	 * cleared then that page is marked as 'pending'
 140	 */
 141	unsigned int pending:1;
 142	/*
 143	 * count of dirty bits on the page
 144	 */
 145	unsigned int  count:30;
 146};
 147
 148/* the main bitmap structure - one per mddev */
 149struct bitmap {
 150
 151	struct bitmap_counts {
 152		spinlock_t lock;
 153		struct bitmap_page *bp;
 154		/* total number of pages in the bitmap */
 155		unsigned long pages;
 156		/* number of pages not yet allocated */
 157		unsigned long missing_pages;
 158		/* chunksize = 2^chunkshift (for bitops) */
 159		unsigned long chunkshift;
 160		/* total number of data chunks for the array */
 161		unsigned long chunks;
 162	} counts;
 163
 164	struct mddev *mddev; /* the md device that the bitmap is for */
 165
 166	__u64	events_cleared;
 167	int need_sync;
 168
 169	struct bitmap_storage {
 170		/* backing disk file */
 171		struct file *file;
 172		/* cached copy of the bitmap file superblock */
 173		struct page *sb_page;
 174		unsigned long sb_index;
 175		/* list of cache pages for the file */
 176		struct page **filemap;
 177		/* attributes associated filemap pages */
 178		unsigned long *filemap_attr;
 179		/* number of pages in the file */
 180		unsigned long file_pages;
 181		/* total bytes in the bitmap */
 182		unsigned long bytes;
 183	} storage;
 184
 185	unsigned long flags;
 186
 187	int allclean;
 188
 189	atomic_t behind_writes;
 190	/* highest actual value at runtime */
 191	unsigned long behind_writes_used;
 192
 193	/*
 194	 * the bitmap daemon - periodically wakes up and sweeps the bitmap
 195	 * file, cleaning up bits and flushing out pages to disk as necessary
 196	 */
 197	unsigned long daemon_lastrun; /* jiffies of last run */
 198	/*
 199	 * when we lasted called end_sync to update bitmap with resync
 200	 * progress.
 201	 */
 202	unsigned long last_end_sync;
 203
 204	/* pending writes to the bitmap file */
 205	atomic_t pending_writes;
 206	wait_queue_head_t write_wait;
 207	wait_queue_head_t overflow_wait;
 208	wait_queue_head_t behind_wait;
 209
 210	struct kernfs_node *sysfs_can_clear;
 211	/* slot offset for clustered env */
 212	int cluster_slot;
 213};
 214
 215static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
 216			   int chunksize, bool init);
 217
 218static inline char *bmname(struct bitmap *bitmap)
 219{
 220	return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
 221}
 222
 223static bool __bitmap_enabled(struct bitmap *bitmap)
 224{
 225	return bitmap->storage.filemap &&
 226	       !test_bit(BITMAP_STALE, &bitmap->flags);
 227}
 228
 229static bool bitmap_enabled(struct mddev *mddev)
 230{
 231	struct bitmap *bitmap = mddev->bitmap;
 232
 233	if (!bitmap)
 234		return false;
 235
 236	return __bitmap_enabled(bitmap);
 237}
 238
 239/*
 240 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
 241 *
 242 * 1) check to see if this page is allocated, if it's not then try to alloc
 243 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
 244 *    page pointer directly as a counter
 245 *
 246 * if we find our page, we increment the page's refcount so that it stays
 247 * allocated while we're using it
 248 */
 249static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
 250			       unsigned long page, int create, int no_hijack)
 251__releases(bitmap->lock)
 252__acquires(bitmap->lock)
 253{
 254	unsigned char *mappage;
 255
 256	WARN_ON_ONCE(page >= bitmap->pages);
 257	if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
 258		return 0;
 259
 260	if (bitmap->bp[page].map) /* page is already allocated, just return */
 261		return 0;
 262
 263	if (!create)
 264		return -ENOENT;
 265
 266	/* this page has not been allocated yet */
 267
 268	spin_unlock_irq(&bitmap->lock);
 269	/* It is possible that this is being called inside a
 270	 * prepare_to_wait/finish_wait loop from raid5c:make_request().
 271	 * In general it is not permitted to sleep in that context as it
 272	 * can cause the loop to spin freely.
 273	 * That doesn't apply here as we can only reach this point
 274	 * once with any loop.
 275	 * When this function completes, either bp[page].map or
 276	 * bp[page].hijacked.  In either case, this function will
 277	 * abort before getting to this point again.  So there is
 278	 * no risk of a free-spin, and so it is safe to assert
 279	 * that sleeping here is allowed.
 280	 */
 281	sched_annotate_sleep();
 282	mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
 283	spin_lock_irq(&bitmap->lock);
 284
 285	if (mappage == NULL) {
 286		pr_debug("md/bitmap: map page allocation failed, hijacking\n");
 287		/* We don't support hijack for cluster raid */
 288		if (no_hijack)
 289			return -ENOMEM;
 290		/* failed - set the hijacked flag so that we can use the
 291		 * pointer as a counter */
 292		if (!bitmap->bp[page].map)
 293			bitmap->bp[page].hijacked = 1;
 294	} else if (bitmap->bp[page].map ||
 295		   bitmap->bp[page].hijacked) {
 296		/* somebody beat us to getting the page */
 297		kfree(mappage);
 298	} else {
 299
 300		/* no page was in place and we have one, so install it */
 301
 302		bitmap->bp[page].map = mappage;
 303		bitmap->missing_pages--;
 304	}
 305	return 0;
 306}
 307
 308/* if page is completely empty, put it back on the free list, or dealloc it */
 309/* if page was hijacked, unmark the flag so it might get alloced next time */
 310/* Note: lock should be held when calling this */
 311static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 312{
 313	char *ptr;
 314
 315	if (bitmap->bp[page].count) /* page is still busy */
 316		return;
 317
 318	/* page is no longer in use, it can be released */
 319
 320	if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 321		bitmap->bp[page].hijacked = 0;
 322		bitmap->bp[page].map = NULL;
 323	} else {
 324		/* normal case, free the page */
 325		ptr = bitmap->bp[page].map;
 326		bitmap->bp[page].map = NULL;
 327		bitmap->missing_pages++;
 328		kfree(ptr);
 329	}
 330}
 331
 332/*
 333 * bitmap file handling - read and write the bitmap file and its superblock
 334 */
 335
 336/*
 337 * basic page I/O operations
 338 */
 339
 340/* IO operations when bitmap is stored near all superblocks */
 341
 342/* choose a good rdev and read the page from there */
 343static int read_sb_page(struct mddev *mddev, loff_t offset,
 344		struct page *page, unsigned long index, int size)
 345{
 346
 347	sector_t sector = mddev->bitmap_info.offset + offset +
 348		index * (PAGE_SIZE / SECTOR_SIZE);
 349	struct md_rdev *rdev;
 350
 351	rdev_for_each(rdev, mddev) {
 352		u32 iosize = roundup(size, bdev_logical_block_size(rdev->bdev));
 353
 354		if (!test_bit(In_sync, &rdev->flags) ||
 355		    test_bit(Faulty, &rdev->flags) ||
 356		    test_bit(Bitmap_sync, &rdev->flags))
 357			continue;
 358
 359		if (sync_page_io(rdev, sector, iosize, page, REQ_OP_READ, true))
 360			return 0;
 361	}
 362	return -EIO;
 363}
 364
 365static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 366{
 367	/* Iterate the disks of an mddev, using rcu to protect access to the
 368	 * linked list, and raising the refcount of devices we return to ensure
 369	 * they don't disappear while in use.
 370	 * As devices are only added or removed when raid_disk is < 0 and
 371	 * nr_pending is 0 and In_sync is clear, the entries we return will
 372	 * still be in the same position on the list when we re-enter
 373	 * list_for_each_entry_continue_rcu.
 374	 *
 375	 * Note that if entered with 'rdev == NULL' to start at the
 376	 * beginning, we temporarily assign 'rdev' to an address which
 377	 * isn't really an rdev, but which can be used by
 378	 * list_for_each_entry_continue_rcu() to find the first entry.
 379	 */
 380	rcu_read_lock();
 381	if (rdev == NULL)
 382		/* start at the beginning */
 383		rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 384	else {
 385		/* release the previous rdev and start from there. */
 386		rdev_dec_pending(rdev, mddev);
 387	}
 388	list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 389		if (rdev->raid_disk >= 0 &&
 390		    !test_bit(Faulty, &rdev->flags)) {
 391			/* this is a usable devices */
 392			atomic_inc(&rdev->nr_pending);
 393			rcu_read_unlock();
 394			return rdev;
 395		}
 396	}
 397	rcu_read_unlock();
 398	return NULL;
 399}
 400
 401static unsigned int optimal_io_size(struct block_device *bdev,
 402				    unsigned int last_page_size,
 403				    unsigned int io_size)
 404{
 405	if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
 406		return roundup(last_page_size, bdev_io_opt(bdev));
 407	return io_size;
 408}
 409
 410static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
 411				   loff_t start, loff_t boundary)
 412{
 413	if (io_size != opt_size &&
 414	    start + opt_size / SECTOR_SIZE <= boundary)
 415		return opt_size;
 416	if (start + io_size / SECTOR_SIZE <= boundary)
 417		return io_size;
 418
 419	/* Overflows boundary */
 420	return 0;
 421}
 422
 423static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 424			   unsigned long pg_index, struct page *page)
 425{
 426	struct block_device *bdev;
 427	struct mddev *mddev = bitmap->mddev;
 428	struct bitmap_storage *store = &bitmap->storage;
 429	unsigned int bitmap_limit = (bitmap->storage.file_pages - pg_index) <<
 430		PAGE_SHIFT;
 431	loff_t sboff, offset = mddev->bitmap_info.offset;
 432	sector_t ps = pg_index * PAGE_SIZE / SECTOR_SIZE;
 433	unsigned int size = PAGE_SIZE;
 434	unsigned int opt_size = PAGE_SIZE;
 435	sector_t doff;
 436
 437	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 438	/* we compare length (page numbers), not page offset. */
 439	if ((pg_index - store->sb_index) == store->file_pages - 1) {
 440		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 441
 442		if (last_page_size == 0)
 443			last_page_size = PAGE_SIZE;
 444		size = roundup(last_page_size, bdev_logical_block_size(bdev));
 445		opt_size = optimal_io_size(bdev, last_page_size, size);
 446	}
 447
 448	sboff = rdev->sb_start + offset;
 449	doff = rdev->data_offset;
 450
 451	/* Just make sure we aren't corrupting data or metadata */
 452	if (mddev->external) {
 453		/* Bitmap could be anywhere. */
 454		if (sboff + ps > doff &&
 455		    sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE))
 456			return -EINVAL;
 457	} else if (offset < 0) {
 458		/* DATA  BITMAP METADATA  */
 459		size = bitmap_io_size(size, opt_size, offset + ps, 0);
 460		if (size == 0)
 461			/* bitmap runs in to metadata */
 462			return -EINVAL;
 463
 464		if (doff + mddev->dev_sectors > sboff)
 465			/* data runs in to bitmap */
 466			return -EINVAL;
 467	} else if (rdev->sb_start < rdev->data_offset) {
 468		/* METADATA BITMAP DATA */
 469		size = bitmap_io_size(size, opt_size, sboff + ps, doff);
 470		if (size == 0)
 471			/* bitmap runs in to data */
 472			return -EINVAL;
 473	}
 474
 475	md_super_write(mddev, rdev, sboff + ps, (int)min(size, bitmap_limit), page);
 476	return 0;
 477}
 478
 479static void write_sb_page(struct bitmap *bitmap, unsigned long pg_index,
 480			  struct page *page, bool wait)
 481{
 482	struct mddev *mddev = bitmap->mddev;
 483
 484	do {
 485		struct md_rdev *rdev = NULL;
 486
 487		while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 488			if (__write_sb_page(rdev, bitmap, pg_index, page) < 0) {
 489				set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 490				return;
 491			}
 492		}
 493	} while (wait && md_super_wait(mddev) < 0);
 494}
 495
 496static void md_bitmap_file_kick(struct bitmap *bitmap);
 497
 498#ifdef CONFIG_MD_BITMAP_FILE
 499static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 500{
 501	struct buffer_head *bh = page_buffers(page);
 502
 503	while (bh && bh->b_blocknr) {
 504		atomic_inc(&bitmap->pending_writes);
 505		set_buffer_locked(bh);
 506		set_buffer_mapped(bh);
 507		submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
 508		bh = bh->b_this_page;
 509	}
 510
 511	if (wait)
 512		wait_event(bitmap->write_wait,
 513			   atomic_read(&bitmap->pending_writes) == 0);
 514}
 515
 516static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 517{
 518	struct bitmap *bitmap = bh->b_private;
 519
 520	if (!uptodate)
 521		set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 522	if (atomic_dec_and_test(&bitmap->pending_writes))
 523		wake_up(&bitmap->write_wait);
 524}
 525
 526static void free_buffers(struct page *page)
 527{
 528	struct buffer_head *bh;
 529
 530	if (!PagePrivate(page))
 531		return;
 532
 533	bh = page_buffers(page);
 534	while (bh) {
 535		struct buffer_head *next = bh->b_this_page;
 536		free_buffer_head(bh);
 537		bh = next;
 538	}
 539	detach_page_private(page);
 540	put_page(page);
 541}
 542
 543/* read a page from a file.
 544 * We both read the page, and attach buffers to the page to record the
 545 * address of each block (using bmap).  These addresses will be used
 546 * to write the block later, completely bypassing the filesystem.
 547 * This usage is similar to how swap files are handled, and allows us
 548 * to write to a file with no concerns of memory allocation failing.
 549 */
 550static int read_file_page(struct file *file, unsigned long index,
 551		struct bitmap *bitmap, unsigned long count, struct page *page)
 552{
 553	int ret = 0;
 554	struct inode *inode = file_inode(file);
 555	struct buffer_head *bh;
 556	sector_t block, blk_cur;
 557	unsigned long blocksize = i_blocksize(inode);
 558
 559	pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 560		 (unsigned long long)index << PAGE_SHIFT);
 561
 562	bh = alloc_page_buffers(page, blocksize);
 563	if (!bh) {
 564		ret = -ENOMEM;
 565		goto out;
 566	}
 567	attach_page_private(page, bh);
 568	blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 569	while (bh) {
 570		block = blk_cur;
 571
 572		if (count == 0)
 573			bh->b_blocknr = 0;
 574		else {
 575			ret = bmap(inode, &block);
 576			if (ret || !block) {
 577				ret = -EINVAL;
 578				bh->b_blocknr = 0;
 579				goto out;
 580			}
 581
 582			bh->b_blocknr = block;
 583			bh->b_bdev = inode->i_sb->s_bdev;
 584			if (count < blocksize)
 585				count = 0;
 586			else
 587				count -= blocksize;
 588
 589			bh->b_end_io = end_bitmap_write;
 590			bh->b_private = bitmap;
 591			atomic_inc(&bitmap->pending_writes);
 592			set_buffer_locked(bh);
 593			set_buffer_mapped(bh);
 594			submit_bh(REQ_OP_READ, bh);
 595		}
 596		blk_cur++;
 597		bh = bh->b_this_page;
 598	}
 599
 600	wait_event(bitmap->write_wait,
 601		   atomic_read(&bitmap->pending_writes)==0);
 602	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 603		ret = -EIO;
 604out:
 605	if (ret)
 606		pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 607		       (int)PAGE_SIZE,
 608		       (unsigned long long)index << PAGE_SHIFT,
 609		       ret);
 610	return ret;
 611}
 612#else /* CONFIG_MD_BITMAP_FILE */
 613static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 614{
 615}
 616static int read_file_page(struct file *file, unsigned long index,
 617		struct bitmap *bitmap, unsigned long count, struct page *page)
 618{
 619	return -EIO;
 620}
 621static void free_buffers(struct page *page)
 622{
 623	put_page(page);
 624}
 625#endif /* CONFIG_MD_BITMAP_FILE */
 626
 627/*
 628 * bitmap file superblock operations
 629 */
 630
 631/*
 632 * write out a page to a file
 633 */
 634static void filemap_write_page(struct bitmap *bitmap, unsigned long pg_index,
 635			       bool wait)
 636{
 637	struct bitmap_storage *store = &bitmap->storage;
 638	struct page *page = store->filemap[pg_index];
 639
 640	if (mddev_is_clustered(bitmap->mddev)) {
 641		/* go to node bitmap area starting point */
 642		pg_index += store->sb_index;
 643	}
 644
 645	if (store->file)
 646		write_file_page(bitmap, page, wait);
 647	else
 648		write_sb_page(bitmap, pg_index, page, wait);
 649}
 650
 651/*
 652 * md_bitmap_wait_writes() should be called before writing any bitmap
 653 * blocks, to ensure previous writes, particularly from
 654 * md_bitmap_daemon_work(), have completed.
 655 */
 656static void md_bitmap_wait_writes(struct bitmap *bitmap)
 657{
 658	if (bitmap->storage.file)
 659		wait_event(bitmap->write_wait,
 660			   atomic_read(&bitmap->pending_writes)==0);
 661	else
 662		/* Note that we ignore the return value.  The writes
 663		 * might have failed, but that would just mean that
 664		 * some bits which should be cleared haven't been,
 665		 * which is safe.  The relevant bitmap blocks will
 666		 * probably get written again, but there is no great
 667		 * loss if they aren't.
 668		 */
 669		md_super_wait(bitmap->mddev);
 670}
 671
 672
 673/* update the event counter and sync the superblock to disk */
 674static void bitmap_update_sb(void *data)
 675{
 676	bitmap_super_t *sb;
 677	struct bitmap *bitmap = data;
 678
 679	if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 680		return;
 681	if (bitmap->mddev->bitmap_info.external)
 682		return;
 683	if (!bitmap->storage.sb_page) /* no superblock */
 684		return;
 685	sb = kmap_atomic(bitmap->storage.sb_page);
 686	sb->events = cpu_to_le64(bitmap->mddev->events);
 687	if (bitmap->mddev->events < bitmap->events_cleared)
 688		/* rocking back to read-only */
 689		bitmap->events_cleared = bitmap->mddev->events;
 690	sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 691	/*
 692	 * clear BITMAP_WRITE_ERROR bit to protect against the case that
 693	 * a bitmap write error occurred but the later writes succeeded.
 694	 */
 695	sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 696	/* Just in case these have been changed via sysfs: */
 697	sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 698	sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 699	/* This might have been changed by a reshape */
 700	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 701	sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 702	sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 703	sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 704					   bitmap_info.space);
 705	kunmap_atomic(sb);
 706
 707	if (bitmap->storage.file)
 708		write_file_page(bitmap, bitmap->storage.sb_page, 1);
 709	else
 710		write_sb_page(bitmap, bitmap->storage.sb_index,
 711			      bitmap->storage.sb_page, 1);
 712}
 713
 714static void bitmap_print_sb(struct bitmap *bitmap)
 715{
 716	bitmap_super_t *sb;
 717
 718	if (!bitmap || !bitmap->storage.sb_page)
 719		return;
 720	sb = kmap_atomic(bitmap->storage.sb_page);
 721	pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 722	pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 723	pr_debug("       version: %u\n", le32_to_cpu(sb->version));
 724	pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 725		 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 726		 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 727		 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 728		 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 729	pr_debug("        events: %llu\n",
 730		 (unsigned long long) le64_to_cpu(sb->events));
 731	pr_debug("events cleared: %llu\n",
 732		 (unsigned long long) le64_to_cpu(sb->events_cleared));
 733	pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 734	pr_debug("     chunksize: %u B\n", le32_to_cpu(sb->chunksize));
 735	pr_debug("  daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
 736	pr_debug("     sync size: %llu KB\n",
 737		 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 738	pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
 739	kunmap_atomic(sb);
 740}
 741
 742/*
 743 * bitmap_new_disk_sb
 744 * @bitmap
 745 *
 746 * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 747 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 748 * This function verifies 'bitmap_info' and populates the on-disk bitmap
 749 * structure, which is to be written to disk.
 750 *
 751 * Returns: 0 on success, -Exxx on error
 752 */
 753static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 754{
 755	bitmap_super_t *sb;
 756	unsigned long chunksize, daemon_sleep, write_behind;
 757
 758	bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 759	if (bitmap->storage.sb_page == NULL)
 760		return -ENOMEM;
 761	bitmap->storage.sb_index = 0;
 762
 763	sb = kmap_atomic(bitmap->storage.sb_page);
 764
 765	sb->magic = cpu_to_le32(BITMAP_MAGIC);
 766	sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 767
 768	chunksize = bitmap->mddev->bitmap_info.chunksize;
 769	BUG_ON(!chunksize);
 770	if (!is_power_of_2(chunksize)) {
 771		kunmap_atomic(sb);
 772		pr_warn("bitmap chunksize not a power of 2\n");
 773		return -EINVAL;
 774	}
 775	sb->chunksize = cpu_to_le32(chunksize);
 776
 777	daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 778	if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 779		pr_debug("Choosing daemon_sleep default (5 sec)\n");
 780		daemon_sleep = 5 * HZ;
 781	}
 782	sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 783	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 784
 785	/*
 786	 * FIXME: write_behind for RAID1.  If not specified, what
 787	 * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 788	 */
 789	write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 790	if (write_behind > COUNTER_MAX)
 791		write_behind = COUNTER_MAX / 2;
 792	sb->write_behind = cpu_to_le32(write_behind);
 793	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 794
 795	/* keep the array size field of the bitmap superblock up to date */
 796	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 797
 798	memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 799
 800	set_bit(BITMAP_STALE, &bitmap->flags);
 801	sb->state = cpu_to_le32(bitmap->flags);
 802	bitmap->events_cleared = bitmap->mddev->events;
 803	sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 804	bitmap->mddev->bitmap_info.nodes = 0;
 805
 806	kunmap_atomic(sb);
 807
 808	return 0;
 809}
 810
 811/* read the superblock from the bitmap file and initialize some bitmap fields */
 812static int md_bitmap_read_sb(struct bitmap *bitmap)
 813{
 814	char *reason = NULL;
 815	bitmap_super_t *sb;
 816	unsigned long chunksize, daemon_sleep, write_behind;
 817	unsigned long long events;
 818	int nodes = 0;
 819	unsigned long sectors_reserved = 0;
 820	int err = -EINVAL;
 821	struct page *sb_page;
 822	loff_t offset = 0;
 823
 824	if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 825		chunksize = 128 * 1024 * 1024;
 826		daemon_sleep = 5 * HZ;
 827		write_behind = 0;
 828		set_bit(BITMAP_STALE, &bitmap->flags);
 829		err = 0;
 830		goto out_no_sb;
 831	}
 832	/* page 0 is the superblock, read it... */
 833	sb_page = alloc_page(GFP_KERNEL);
 834	if (!sb_page)
 835		return -ENOMEM;
 836	bitmap->storage.sb_page = sb_page;
 837
 838re_read:
 839	/* If cluster_slot is set, the cluster is setup */
 840	if (bitmap->cluster_slot >= 0) {
 841		sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 842
 843		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 844			   (bitmap->mddev->bitmap_info.chunksize >> 9));
 845		/* bits to bytes */
 846		bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 847		/* to 4k blocks */
 848		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 849		offset = bitmap->cluster_slot * (bm_blocks << 3);
 850		pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 851			bitmap->cluster_slot, offset);
 852	}
 853
 854	if (bitmap->storage.file) {
 855		loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 856		int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 857
 858		err = read_file_page(bitmap->storage.file, 0,
 859				bitmap, bytes, sb_page);
 860	} else {
 861		err = read_sb_page(bitmap->mddev, offset, sb_page, 0,
 862				   sizeof(bitmap_super_t));
 863	}
 864	if (err)
 865		return err;
 866
 867	err = -EINVAL;
 868	sb = kmap_atomic(sb_page);
 869
 870	chunksize = le32_to_cpu(sb->chunksize);
 871	daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 872	write_behind = le32_to_cpu(sb->write_behind);
 873	sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 874
 875	/* verify that the bitmap-specific fields are valid */
 876	if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 877		reason = "bad magic";
 878	else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 879		 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 880		reason = "unrecognized superblock version";
 881	else if (chunksize < 512)
 882		reason = "bitmap chunksize too small";
 883	else if (!is_power_of_2(chunksize))
 884		reason = "bitmap chunksize not a power of 2";
 885	else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 886		reason = "daemon sleep period out of range";
 887	else if (write_behind > COUNTER_MAX)
 888		reason = "write-behind limit out of range (0 - 16383)";
 889	if (reason) {
 890		pr_warn("%s: invalid bitmap file superblock: %s\n",
 891			bmname(bitmap), reason);
 892		goto out;
 893	}
 894
 895	/*
 896	 * Setup nodes/clustername only if bitmap version is
 897	 * cluster-compatible
 898	 */
 899	if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 900		nodes = le32_to_cpu(sb->nodes);
 901		strscpy(bitmap->mddev->bitmap_info.cluster_name,
 902				sb->cluster_name, 64);
 903	}
 904
 905	/* keep the array size field of the bitmap superblock up to date */
 906	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 907
 908	if (bitmap->mddev->persistent) {
 909		/*
 910		 * We have a persistent array superblock, so compare the
 911		 * bitmap's UUID and event counter to the mddev's
 912		 */
 913		if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 914			pr_warn("%s: bitmap superblock UUID mismatch\n",
 915				bmname(bitmap));
 916			goto out;
 917		}
 918		events = le64_to_cpu(sb->events);
 919		if (!nodes && (events < bitmap->mddev->events)) {
 920			pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 921				bmname(bitmap), events,
 922				(unsigned long long) bitmap->mddev->events);
 923			set_bit(BITMAP_STALE, &bitmap->flags);
 924		}
 925	}
 926
 927	/* assign fields using values from superblock */
 928	bitmap->flags |= le32_to_cpu(sb->state);
 929	if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 930		set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 931	bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 932	err = 0;
 933
 934out:
 935	kunmap_atomic(sb);
 936	if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 937		/* Assigning chunksize is required for "re_read" */
 938		bitmap->mddev->bitmap_info.chunksize = chunksize;
 939		err = md_setup_cluster(bitmap->mddev, nodes);
 940		if (err) {
 941			pr_warn("%s: Could not setup cluster service (%d)\n",
 942				bmname(bitmap), err);
 943			goto out_no_sb;
 944		}
 945		bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 946		goto re_read;
 947	}
 948
 949out_no_sb:
 950	if (err == 0) {
 951		if (test_bit(BITMAP_STALE, &bitmap->flags))
 952			bitmap->events_cleared = bitmap->mddev->events;
 953		bitmap->mddev->bitmap_info.chunksize = chunksize;
 954		bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 955		bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 956		bitmap->mddev->bitmap_info.nodes = nodes;
 957		if (bitmap->mddev->bitmap_info.space == 0 ||
 958			bitmap->mddev->bitmap_info.space > sectors_reserved)
 959			bitmap->mddev->bitmap_info.space = sectors_reserved;
 960	} else {
 961		bitmap_print_sb(bitmap);
 962		if (bitmap->cluster_slot < 0)
 963			md_cluster_stop(bitmap->mddev);
 964	}
 965	return err;
 966}
 967
 968/*
 969 * general bitmap file operations
 970 */
 971
 972/*
 973 * on-disk bitmap:
 974 *
 975 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 976 * file a page at a time. There's a superblock at the start of the file.
 977 */
 978/* calculate the index of the page that contains this bit */
 979static inline unsigned long file_page_index(struct bitmap_storage *store,
 980					    unsigned long chunk)
 981{
 982	if (store->sb_page)
 983		chunk += sizeof(bitmap_super_t) << 3;
 984	return chunk >> PAGE_BIT_SHIFT;
 985}
 986
 987/* calculate the (bit) offset of this bit within a page */
 988static inline unsigned long file_page_offset(struct bitmap_storage *store,
 989					     unsigned long chunk)
 990{
 991	if (store->sb_page)
 992		chunk += sizeof(bitmap_super_t) << 3;
 993	return chunk & (PAGE_BITS - 1);
 994}
 995
 996/*
 997 * return a pointer to the page in the filemap that contains the given bit
 998 *
 999 */
1000static inline struct page *filemap_get_page(struct bitmap_storage *store,
1001					    unsigned long chunk)
1002{
1003	if (file_page_index(store, chunk) >= store->file_pages)
1004		return NULL;
1005	return store->filemap[file_page_index(store, chunk)];
1006}
1007
1008static int md_bitmap_storage_alloc(struct bitmap_storage *store,
1009				   unsigned long chunks, int with_super,
1010				   int slot_number)
1011{
1012	int pnum, offset = 0;
1013	unsigned long num_pages;
1014	unsigned long bytes;
1015
1016	bytes = DIV_ROUND_UP(chunks, 8);
1017	if (with_super)
1018		bytes += sizeof(bitmap_super_t);
1019
1020	num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
1021	offset = slot_number * num_pages;
1022
1023	store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
1024				       GFP_KERNEL);
1025	if (!store->filemap)
1026		return -ENOMEM;
1027
1028	if (with_super && !store->sb_page) {
1029		store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
1030		if (store->sb_page == NULL)
1031			return -ENOMEM;
1032	}
1033
1034	pnum = 0;
1035	if (store->sb_page) {
1036		store->filemap[0] = store->sb_page;
1037		pnum = 1;
1038		store->sb_index = offset;
1039	}
1040
1041	for ( ; pnum < num_pages; pnum++) {
1042		store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
1043		if (!store->filemap[pnum]) {
1044			store->file_pages = pnum;
1045			return -ENOMEM;
1046		}
1047	}
1048	store->file_pages = pnum;
1049
1050	/* We need 4 bits per page, rounded up to a multiple
1051	 * of sizeof(unsigned long) */
1052	store->filemap_attr = kzalloc(
1053		roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
1054		GFP_KERNEL);
1055	if (!store->filemap_attr)
1056		return -ENOMEM;
1057
1058	store->bytes = bytes;
1059
1060	return 0;
1061}
1062
1063static void md_bitmap_file_unmap(struct bitmap_storage *store)
1064{
1065	struct file *file = store->file;
1066	struct page *sb_page = store->sb_page;
1067	struct page **map = store->filemap;
1068	int pages = store->file_pages;
1069
1070	while (pages--)
1071		if (map[pages] != sb_page) /* 0 is sb_page, release it below */
1072			free_buffers(map[pages]);
1073	kfree(map);
1074	kfree(store->filemap_attr);
1075
1076	if (sb_page)
1077		free_buffers(sb_page);
1078
1079	if (file) {
1080		struct inode *inode = file_inode(file);
1081		invalidate_mapping_pages(inode->i_mapping, 0, -1);
1082		fput(file);
1083	}
1084}
1085
1086/*
1087 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
1088 * then it is no longer reliable, so we stop using it and we mark the file
1089 * as failed in the superblock
1090 */
1091static void md_bitmap_file_kick(struct bitmap *bitmap)
1092{
1093	if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
1094		bitmap_update_sb(bitmap);
1095
1096		if (bitmap->storage.file) {
1097			pr_warn("%s: kicking failed bitmap file %pD4 from array!\n",
1098				bmname(bitmap), bitmap->storage.file);
1099
1100		} else
1101			pr_warn("%s: disabling internal bitmap due to errors\n",
1102				bmname(bitmap));
1103	}
1104}
1105
1106enum bitmap_page_attr {
1107	BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
1108	BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
1109				    * i.e. counter is 1 or 2. */
1110	BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
1111};
1112
1113static inline void set_page_attr(struct bitmap *bitmap, int pnum,
1114				 enum bitmap_page_attr attr)
1115{
1116	set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1117}
1118
1119static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
1120				   enum bitmap_page_attr attr)
1121{
1122	clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1123}
1124
1125static inline int test_page_attr(struct bitmap *bitmap, int pnum,
1126				 enum bitmap_page_attr attr)
1127{
1128	return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1129}
1130
1131static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
1132					   enum bitmap_page_attr attr)
1133{
1134	return test_and_clear_bit((pnum<<2) + attr,
1135				  bitmap->storage.filemap_attr);
1136}
1137/*
1138 * bitmap_file_set_bit -- called before performing a write to the md device
1139 * to set (and eventually sync) a particular bit in the bitmap file
1140 *
1141 * we set the bit immediately, then we record the page number so that
1142 * when an unplug occurs, we can flush the dirty pages out to disk
1143 */
1144static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
1145{
1146	unsigned long bit;
1147	struct page *page;
1148	void *kaddr;
1149	unsigned long chunk = block >> bitmap->counts.chunkshift;
1150	struct bitmap_storage *store = &bitmap->storage;
1151	unsigned long index = file_page_index(store, chunk);
1152	unsigned long node_offset = 0;
1153
1154	index += store->sb_index;
1155	if (mddev_is_clustered(bitmap->mddev))
1156		node_offset = bitmap->cluster_slot * store->file_pages;
1157
1158	page = filemap_get_page(&bitmap->storage, chunk);
1159	if (!page)
1160		return;
1161	bit = file_page_offset(&bitmap->storage, chunk);
1162
1163	/* set the bit */
1164	kaddr = kmap_atomic(page);
1165	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1166		set_bit(bit, kaddr);
1167	else
1168		set_bit_le(bit, kaddr);
1169	kunmap_atomic(kaddr);
1170	pr_debug("set file bit %lu page %lu\n", bit, index);
1171	/* record page number so it gets flushed to disk when unplug occurs */
1172	set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_DIRTY);
1173}
1174
1175static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
1176{
1177	unsigned long bit;
1178	struct page *page;
1179	void *paddr;
1180	unsigned long chunk = block >> bitmap->counts.chunkshift;
1181	struct bitmap_storage *store = &bitmap->storage;
1182	unsigned long index = file_page_index(store, chunk);
1183	unsigned long node_offset = 0;
1184
1185	index += store->sb_index;
1186	if (mddev_is_clustered(bitmap->mddev))
1187		node_offset = bitmap->cluster_slot * store->file_pages;
1188
1189	page = filemap_get_page(&bitmap->storage, chunk);
1190	if (!page)
1191		return;
1192	bit = file_page_offset(&bitmap->storage, chunk);
1193	paddr = kmap_atomic(page);
1194	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1195		clear_bit(bit, paddr);
1196	else
1197		clear_bit_le(bit, paddr);
1198	kunmap_atomic(paddr);
1199	if (!test_page_attr(bitmap, index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
1200		set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_PENDING);
1201		bitmap->allclean = 0;
1202	}
1203}
1204
1205static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
1206{
1207	unsigned long bit;
1208	struct page *page;
1209	void *paddr;
1210	unsigned long chunk = block >> bitmap->counts.chunkshift;
1211	int set = 0;
1212
1213	page = filemap_get_page(&bitmap->storage, chunk);
1214	if (!page)
1215		return -EINVAL;
1216	bit = file_page_offset(&bitmap->storage, chunk);
1217	paddr = kmap_atomic(page);
1218	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1219		set = test_bit(bit, paddr);
1220	else
1221		set = test_bit_le(bit, paddr);
1222	kunmap_atomic(paddr);
1223	return set;
1224}
1225
1226/* this gets called when the md device is ready to unplug its underlying
1227 * (slave) device queues -- before we let any writes go down, we need to
1228 * sync the dirty pages of the bitmap file to disk */
1229static void __bitmap_unplug(struct bitmap *bitmap)
1230{
1231	unsigned long i;
1232	int dirty, need_write;
1233	int writing = 0;
1234
1235	if (!__bitmap_enabled(bitmap))
1236		return;
1237
1238	/* look at each page to see if there are any set bits that need to be
1239	 * flushed out to disk */
1240	for (i = 0; i < bitmap->storage.file_pages; i++) {
1241		dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1242		need_write = test_and_clear_page_attr(bitmap, i,
1243						      BITMAP_PAGE_NEEDWRITE);
1244		if (dirty || need_write) {
1245			if (!writing) {
1246				md_bitmap_wait_writes(bitmap);
1247				mddev_add_trace_msg(bitmap->mddev,
1248					"md bitmap_unplug");
1249			}
1250			clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1251			filemap_write_page(bitmap, i, false);
1252			writing = 1;
1253		}
1254	}
1255	if (writing)
1256		md_bitmap_wait_writes(bitmap);
1257
1258	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1259		md_bitmap_file_kick(bitmap);
1260}
1261
1262struct bitmap_unplug_work {
1263	struct work_struct work;
1264	struct bitmap *bitmap;
1265	struct completion *done;
1266};
1267
1268static void md_bitmap_unplug_fn(struct work_struct *work)
1269{
1270	struct bitmap_unplug_work *unplug_work =
1271		container_of(work, struct bitmap_unplug_work, work);
1272
1273	__bitmap_unplug(unplug_work->bitmap);
1274	complete(unplug_work->done);
1275}
1276
1277static void bitmap_unplug_async(struct bitmap *bitmap)
1278{
1279	DECLARE_COMPLETION_ONSTACK(done);
1280	struct bitmap_unplug_work unplug_work;
1281
1282	INIT_WORK_ONSTACK(&unplug_work.work, md_bitmap_unplug_fn);
1283	unplug_work.bitmap = bitmap;
1284	unplug_work.done = &done;
1285
1286	queue_work(md_bitmap_wq, &unplug_work.work);
1287	wait_for_completion(&done);
1288	destroy_work_on_stack(&unplug_work.work);
1289}
1290
1291static void bitmap_unplug(struct mddev *mddev, bool sync)
1292{
1293	struct bitmap *bitmap = mddev->bitmap;
1294
1295	if (!bitmap)
1296		return;
1297
1298	if (sync)
1299		__bitmap_unplug(bitmap);
1300	else
1301		bitmap_unplug_async(bitmap);
1302}
1303
1304static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1305
1306/*
1307 * Initialize the in-memory bitmap from the on-disk bitmap and set up the memory
1308 * mapping of the bitmap file.
1309 *
1310 * Special case: If there's no bitmap file, or if the bitmap file had been
1311 * previously kicked from the array, we mark all the bits as 1's in order to
1312 * cause a full resync.
1313 *
1314 * We ignore all bits for sectors that end earlier than 'start'.
1315 * This is used when reading an out-of-date bitmap.
1316 */
1317static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1318{
1319	bool outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1320	struct mddev *mddev = bitmap->mddev;
1321	unsigned long chunks = bitmap->counts.chunks;
1322	struct bitmap_storage *store = &bitmap->storage;
1323	struct file *file = store->file;
1324	unsigned long node_offset = 0;
1325	unsigned long bit_cnt = 0;
1326	unsigned long i;
1327	int ret;
1328
1329	if (!file && !mddev->bitmap_info.offset) {
1330		/* No permanent bitmap - fill with '1s'. */
1331		store->filemap = NULL;
1332		store->file_pages = 0;
1333		for (i = 0; i < chunks ; i++) {
1334			/* if the disk bit is set, set the memory bit */
1335			int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1336				      >= start);
1337			md_bitmap_set_memory_bits(bitmap,
1338						  (sector_t)i << bitmap->counts.chunkshift,
1339						  needed);
1340		}
1341		return 0;
1342	}
1343
1344	if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1345		pr_warn("%s: bitmap file too short %lu < %lu\n",
1346			bmname(bitmap),
1347			(unsigned long) i_size_read(file->f_mapping->host),
1348			store->bytes);
1349		ret = -ENOSPC;
1350		goto err;
1351	}
1352
1353	if (mddev_is_clustered(mddev))
1354		node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1355
1356	for (i = 0; i < store->file_pages; i++) {
1357		struct page *page = store->filemap[i];
1358		int count;
1359
1360		/* unmap the old page, we're done with it */
1361		if (i == store->file_pages - 1)
1362			count = store->bytes - i * PAGE_SIZE;
1363		else
1364			count = PAGE_SIZE;
1365
1366		if (file)
1367			ret = read_file_page(file, i, bitmap, count, page);
1368		else
1369			ret = read_sb_page(mddev, 0, page, i + node_offset,
1370					   count);
1371		if (ret)
1372			goto err;
1373	}
1374
1375	if (outofdate) {
1376		pr_warn("%s: bitmap file is out of date, doing full recovery\n",
1377			bmname(bitmap));
1378
1379		for (i = 0; i < store->file_pages; i++) {
1380			struct page *page = store->filemap[i];
1381			unsigned long offset = 0;
1382			void *paddr;
1383
1384			if (i == 0 && !mddev->bitmap_info.external)
1385				offset = sizeof(bitmap_super_t);
1386
1387			/*
1388			 * If the bitmap is out of date, dirty the whole page
1389			 * and write it out
1390			 */
1391			paddr = kmap_atomic(page);
1392			memset(paddr + offset, 0xff, PAGE_SIZE - offset);
1393			kunmap_atomic(paddr);
1394
1395			filemap_write_page(bitmap, i, true);
1396			if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) {
1397				ret = -EIO;
1398				goto err;
1399			}
1400		}
1401	}
1402
1403	for (i = 0; i < chunks; i++) {
1404		struct page *page = filemap_get_page(&bitmap->storage, i);
1405		unsigned long bit = file_page_offset(&bitmap->storage, i);
1406		void *paddr;
1407		bool was_set;
1408
1409		paddr = kmap_atomic(page);
1410		if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1411			was_set = test_bit(bit, paddr);
1412		else
1413			was_set = test_bit_le(bit, paddr);
1414		kunmap_atomic(paddr);
1415
1416		if (was_set) {
1417			/* if the disk bit is set, set the memory bit */
1418			int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1419				      >= start);
1420			md_bitmap_set_memory_bits(bitmap,
1421						  (sector_t)i << bitmap->counts.chunkshift,
1422						  needed);
1423			bit_cnt++;
1424		}
1425	}
1426
1427	pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1428		 bmname(bitmap), store->file_pages,
1429		 bit_cnt, chunks);
1430
1431	return 0;
1432
1433 err:
1434	pr_warn("%s: bitmap initialisation failed: %d\n",
1435		bmname(bitmap), ret);
1436	return ret;
1437}
1438
1439/* just flag bitmap pages as needing to be written. */
1440static void bitmap_write_all(struct mddev *mddev)
1441{
1442	int i;
1443	struct bitmap *bitmap = mddev->bitmap;
1444
1445	if (!bitmap || !bitmap->storage.filemap)
1446		return;
1447
1448	/* Only one copy, so nothing needed */
1449	if (bitmap->storage.file)
1450		return;
1451
1452	for (i = 0; i < bitmap->storage.file_pages; i++)
1453		set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
1454	bitmap->allclean = 0;
1455}
1456
1457static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1458				 sector_t offset, int inc)
1459{
1460	sector_t chunk = offset >> bitmap->chunkshift;
1461	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1462	bitmap->bp[page].count += inc;
1463	md_bitmap_checkfree(bitmap, page);
1464}
1465
1466static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1467{
1468	sector_t chunk = offset >> bitmap->chunkshift;
1469	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1470	struct bitmap_page *bp = &bitmap->bp[page];
1471
1472	if (!bp->pending)
1473		bp->pending = 1;
1474}
1475
1476static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1477					       sector_t offset, sector_t *blocks,
1478					       int create);
1479
1480static void mddev_set_timeout(struct mddev *mddev, unsigned long timeout,
1481			      bool force)
1482{
1483	struct md_thread *thread;
1484
1485	rcu_read_lock();
1486	thread = rcu_dereference(mddev->thread);
1487
1488	if (!thread)
1489		goto out;
1490
1491	if (force || thread->timeout < MAX_SCHEDULE_TIMEOUT)
1492		thread->timeout = timeout;
1493
1494out:
1495	rcu_read_unlock();
1496}
1497
1498/*
1499 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1500 *			out to disk
1501 */
1502static void bitmap_daemon_work(struct mddev *mddev)
1503{
1504	struct bitmap *bitmap;
1505	unsigned long j;
1506	unsigned long nextpage;
1507	sector_t blocks;
1508	struct bitmap_counts *counts;
1509
1510	/* Use a mutex to guard daemon_work against
1511	 * bitmap_destroy.
1512	 */
1513	mutex_lock(&mddev->bitmap_info.mutex);
1514	bitmap = mddev->bitmap;
1515	if (bitmap == NULL) {
1516		mutex_unlock(&mddev->bitmap_info.mutex);
1517		return;
1518	}
1519	if (time_before(jiffies, bitmap->daemon_lastrun
1520			+ mddev->bitmap_info.daemon_sleep))
1521		goto done;
1522
1523	bitmap->daemon_lastrun = jiffies;
1524	if (bitmap->allclean) {
1525		mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
1526		goto done;
1527	}
1528	bitmap->allclean = 1;
1529
1530	mddev_add_trace_msg(bitmap->mddev, "md bitmap_daemon_work");
1531
1532	/* Any file-page which is PENDING now needs to be written.
1533	 * So set NEEDWRITE now, then after we make any last-minute changes
1534	 * we will write it.
1535	 */
1536	for (j = 0; j < bitmap->storage.file_pages; j++)
1537		if (test_and_clear_page_attr(bitmap, j,
1538					     BITMAP_PAGE_PENDING))
1539			set_page_attr(bitmap, j,
1540				      BITMAP_PAGE_NEEDWRITE);
1541
1542	if (bitmap->need_sync &&
1543	    mddev->bitmap_info.external == 0) {
1544		/* Arrange for superblock update as well as
1545		 * other changes */
1546		bitmap_super_t *sb;
1547		bitmap->need_sync = 0;
1548		if (bitmap->storage.filemap) {
1549			sb = kmap_atomic(bitmap->storage.sb_page);
1550			sb->events_cleared =
1551				cpu_to_le64(bitmap->events_cleared);
1552			kunmap_atomic(sb);
1553			set_page_attr(bitmap, 0,
1554				      BITMAP_PAGE_NEEDWRITE);
1555		}
1556	}
1557	/* Now look at the bitmap counters and if any are '2' or '1',
1558	 * decrement and handle accordingly.
1559	 */
1560	counts = &bitmap->counts;
1561	spin_lock_irq(&counts->lock);
1562	nextpage = 0;
1563	for (j = 0; j < counts->chunks; j++) {
1564		bitmap_counter_t *bmc;
1565		sector_t  block = (sector_t)j << counts->chunkshift;
1566
1567		if (j == nextpage) {
1568			nextpage += PAGE_COUNTER_RATIO;
1569			if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1570				j |= PAGE_COUNTER_MASK;
1571				continue;
1572			}
1573			counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1574		}
1575
1576		bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1577		if (!bmc) {
1578			j |= PAGE_COUNTER_MASK;
1579			continue;
1580		}
1581		if (*bmc == 1 && !bitmap->need_sync) {
1582			/* We can clear the bit */
1583			*bmc = 0;
1584			md_bitmap_count_page(counts, block, -1);
1585			md_bitmap_file_clear_bit(bitmap, block);
1586		} else if (*bmc && *bmc <= 2) {
1587			*bmc = 1;
1588			md_bitmap_set_pending(counts, block);
1589			bitmap->allclean = 0;
1590		}
1591	}
1592	spin_unlock_irq(&counts->lock);
1593
1594	md_bitmap_wait_writes(bitmap);
1595	/* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1596	 * DIRTY pages need to be written by bitmap_unplug so it can wait
1597	 * for them.
1598	 * If we find any DIRTY page we stop there and let bitmap_unplug
1599	 * handle all the rest.  This is important in the case where
1600	 * the first blocking holds the superblock and it has been updated.
1601	 * We mustn't write any other blocks before the superblock.
1602	 */
1603	for (j = 0;
1604	     j < bitmap->storage.file_pages
1605		     && !test_bit(BITMAP_STALE, &bitmap->flags);
1606	     j++) {
1607		if (test_page_attr(bitmap, j,
1608				   BITMAP_PAGE_DIRTY))
1609			/* bitmap_unplug will handle the rest */
1610			break;
1611		if (bitmap->storage.filemap &&
1612		    test_and_clear_page_attr(bitmap, j,
1613					     BITMAP_PAGE_NEEDWRITE))
1614			filemap_write_page(bitmap, j, false);
1615	}
1616
1617 done:
1618	if (bitmap->allclean == 0)
1619		mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
1620	mutex_unlock(&mddev->bitmap_info.mutex);
1621}
1622
1623static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1624					       sector_t offset, sector_t *blocks,
1625					       int create)
1626__releases(bitmap->lock)
1627__acquires(bitmap->lock)
1628{
1629	/* If 'create', we might release the lock and reclaim it.
1630	 * The lock must have been taken with interrupts enabled.
1631	 * If !create, we don't release the lock.
1632	 */
1633	sector_t chunk = offset >> bitmap->chunkshift;
1634	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1635	unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1636	sector_t csize = ((sector_t)1) << bitmap->chunkshift;
1637	int err;
1638
1639	if (page >= bitmap->pages) {
1640		/*
1641		 * This can happen if bitmap_start_sync goes beyond
1642		 * End-of-device while looking for a whole page or
1643		 * user set a huge number to sysfs bitmap_set_bits.
1644		 */
1645		*blocks = csize - (offset & (csize - 1));
1646		return NULL;
1647	}
1648	err = md_bitmap_checkpage(bitmap, page, create, 0);
1649
1650	if (bitmap->bp[page].hijacked ||
1651	    bitmap->bp[page].map == NULL)
1652		csize = ((sector_t)1) << (bitmap->chunkshift +
1653					  PAGE_COUNTER_SHIFT);
1654
1655	*blocks = csize - (offset & (csize - 1));
1656
1657	if (err < 0)
1658		return NULL;
1659
1660	/* now locked ... */
1661
1662	if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1663		/* should we use the first or second counter field
1664		 * of the hijacked pointer? */
1665		int hi = (pageoff > PAGE_COUNTER_MASK);
1666		return  &((bitmap_counter_t *)
1667			  &bitmap->bp[page].map)[hi];
1668	} else /* page is allocated */
1669		return (bitmap_counter_t *)
1670			&(bitmap->bp[page].map[pageoff]);
1671}
1672
1673static int bitmap_startwrite(struct mddev *mddev, sector_t offset,
1674			     unsigned long sectors)
1675{
1676	struct bitmap *bitmap = mddev->bitmap;
1677
1678	if (!bitmap)
1679		return 0;
1680
1681	while (sectors) {
1682		sector_t blocks;
1683		bitmap_counter_t *bmc;
1684
1685		spin_lock_irq(&bitmap->counts.lock);
1686		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1687		if (!bmc) {
1688			spin_unlock_irq(&bitmap->counts.lock);
1689			return 0;
1690		}
1691
1692		if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1693			DEFINE_WAIT(__wait);
1694			/* note that it is safe to do the prepare_to_wait
1695			 * after the test as long as we do it before dropping
1696			 * the spinlock.
1697			 */
1698			prepare_to_wait(&bitmap->overflow_wait, &__wait,
1699					TASK_UNINTERRUPTIBLE);
1700			spin_unlock_irq(&bitmap->counts.lock);
1701			schedule();
1702			finish_wait(&bitmap->overflow_wait, &__wait);
1703			continue;
1704		}
1705
1706		switch (*bmc) {
1707		case 0:
1708			md_bitmap_file_set_bit(bitmap, offset);
1709			md_bitmap_count_page(&bitmap->counts, offset, 1);
1710			fallthrough;
1711		case 1:
1712			*bmc = 2;
1713		}
1714
1715		(*bmc)++;
1716
1717		spin_unlock_irq(&bitmap->counts.lock);
1718
1719		offset += blocks;
1720		if (sectors > blocks)
1721			sectors -= blocks;
1722		else
1723			sectors = 0;
1724	}
1725	return 0;
1726}
1727
1728static void bitmap_endwrite(struct mddev *mddev, sector_t offset,
1729			    unsigned long sectors)
1730{
1731	struct bitmap *bitmap = mddev->bitmap;
1732
1733	if (!bitmap)
1734		return;
1735
1736	while (sectors) {
1737		sector_t blocks;
1738		unsigned long flags;
1739		bitmap_counter_t *bmc;
1740
1741		spin_lock_irqsave(&bitmap->counts.lock, flags);
1742		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1743		if (!bmc) {
1744			spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1745			return;
1746		}
1747
1748		if (!bitmap->mddev->degraded) {
1749			if (bitmap->events_cleared < bitmap->mddev->events) {
1750				bitmap->events_cleared = bitmap->mddev->events;
1751				bitmap->need_sync = 1;
1752				sysfs_notify_dirent_safe(
1753						bitmap->sysfs_can_clear);
1754			}
1755		} else if (!NEEDED(*bmc)) {
1756			*bmc |= NEEDED_MASK;
1757		}
1758
1759		if (COUNTER(*bmc) == COUNTER_MAX)
1760			wake_up(&bitmap->overflow_wait);
1761
1762		(*bmc)--;
1763		if (*bmc <= 2) {
1764			md_bitmap_set_pending(&bitmap->counts, offset);
1765			bitmap->allclean = 0;
1766		}
1767		spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1768		offset += blocks;
1769		if (sectors > blocks)
1770			sectors -= blocks;
1771		else
1772			sectors = 0;
1773	}
1774}
1775
1776static bool __bitmap_start_sync(struct bitmap *bitmap, sector_t offset,
1777				sector_t *blocks, bool degraded)
1778{
1779	bitmap_counter_t *bmc;
1780	bool rv;
1781
1782	if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1783		*blocks = 1024;
1784		return true; /* always resync if no bitmap */
1785	}
1786	spin_lock_irq(&bitmap->counts.lock);
1787
1788	rv = false;
1789	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1790	if (bmc) {
1791		/* locked */
1792		if (RESYNC(*bmc)) {
1793			rv = true;
1794		} else if (NEEDED(*bmc)) {
1795			rv = true;
1796			if (!degraded) { /* don't set/clear bits if degraded */
1797				*bmc |= RESYNC_MASK;
1798				*bmc &= ~NEEDED_MASK;
1799			}
1800		}
1801	}
1802	spin_unlock_irq(&bitmap->counts.lock);
1803
1804	return rv;
1805}
1806
1807static bool bitmap_start_sync(struct mddev *mddev, sector_t offset,
1808			      sector_t *blocks, bool degraded)
1809{
1810	/* bitmap_start_sync must always report on multiples of whole
1811	 * pages, otherwise resync (which is very PAGE_SIZE based) will
1812	 * get confused.
1813	 * So call __bitmap_start_sync repeatedly (if needed) until
1814	 * At least PAGE_SIZE>>9 blocks are covered.
1815	 * Return the 'or' of the result.
1816	 */
1817	bool rv = false;
1818	sector_t blocks1;
1819
1820	*blocks = 0;
1821	while (*blocks < (PAGE_SIZE>>9)) {
1822		rv |= __bitmap_start_sync(mddev->bitmap, offset,
1823					  &blocks1, degraded);
1824		offset += blocks1;
1825		*blocks += blocks1;
1826	}
1827
1828	return rv;
1829}
1830
1831static void __bitmap_end_sync(struct bitmap *bitmap, sector_t offset,
1832			      sector_t *blocks, bool aborted)
1833{
1834	bitmap_counter_t *bmc;
1835	unsigned long flags;
1836
1837	if (bitmap == NULL) {
1838		*blocks = 1024;
1839		return;
1840	}
1841	spin_lock_irqsave(&bitmap->counts.lock, flags);
1842	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1843	if (bmc == NULL)
1844		goto unlock;
1845	/* locked */
1846	if (RESYNC(*bmc)) {
1847		*bmc &= ~RESYNC_MASK;
1848
1849		if (!NEEDED(*bmc) && aborted)
1850			*bmc |= NEEDED_MASK;
1851		else {
1852			if (*bmc <= 2) {
1853				md_bitmap_set_pending(&bitmap->counts, offset);
1854				bitmap->allclean = 0;
1855			}
1856		}
1857	}
1858 unlock:
1859	spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1860}
1861
1862static void bitmap_end_sync(struct mddev *mddev, sector_t offset,
1863			    sector_t *blocks)
1864{
1865	__bitmap_end_sync(mddev->bitmap, offset, blocks, true);
1866}
1867
1868static void bitmap_close_sync(struct mddev *mddev)
1869{
1870	/* Sync has finished, and any bitmap chunks that weren't synced
1871	 * properly have been aborted.  It remains to us to clear the
1872	 * RESYNC bit wherever it is still on
1873	 */
1874	sector_t sector = 0;
1875	sector_t blocks;
1876	struct bitmap *bitmap = mddev->bitmap;
1877
1878	if (!bitmap)
1879		return;
1880
1881	while (sector < bitmap->mddev->resync_max_sectors) {
1882		__bitmap_end_sync(bitmap, sector, &blocks, false);
1883		sector += blocks;
1884	}
1885}
1886
1887static void bitmap_cond_end_sync(struct mddev *mddev, sector_t sector,
1888				 bool force)
1889{
1890	sector_t s = 0;
1891	sector_t blocks;
1892	struct bitmap *bitmap = mddev->bitmap;
1893
1894	if (!bitmap)
1895		return;
1896	if (sector == 0) {
1897		bitmap->last_end_sync = jiffies;
1898		return;
1899	}
1900	if (!force && time_before(jiffies, (bitmap->last_end_sync
1901				  + bitmap->mddev->bitmap_info.daemon_sleep)))
1902		return;
1903	wait_event(bitmap->mddev->recovery_wait,
1904		   atomic_read(&bitmap->mddev->recovery_active) == 0);
1905
1906	bitmap->mddev->curr_resync_completed = sector;
1907	set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1908	sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1909	s = 0;
1910	while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1911		__bitmap_end_sync(bitmap, s, &blocks, false);
1912		s += blocks;
1913	}
1914	bitmap->last_end_sync = jiffies;
1915	sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1916}
1917
1918static void bitmap_sync_with_cluster(struct mddev *mddev,
1919				     sector_t old_lo, sector_t old_hi,
1920				     sector_t new_lo, sector_t new_hi)
1921{
1922	struct bitmap *bitmap = mddev->bitmap;
1923	sector_t sector, blocks = 0;
1924
1925	for (sector = old_lo; sector < new_lo; ) {
1926		__bitmap_end_sync(bitmap, sector, &blocks, false);
1927		sector += blocks;
1928	}
1929	WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1930
1931	for (sector = old_hi; sector < new_hi; ) {
1932		bitmap_start_sync(mddev, sector, &blocks, false);
1933		sector += blocks;
1934	}
1935	WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1936}
1937
1938static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1939{
1940	/* For each chunk covered by any of these sectors, set the
1941	 * counter to 2 and possibly set resync_needed.  They should all
1942	 * be 0 at this point
1943	 */
1944
1945	sector_t secs;
1946	bitmap_counter_t *bmc;
1947	spin_lock_irq(&bitmap->counts.lock);
1948	bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1949	if (!bmc) {
1950		spin_unlock_irq(&bitmap->counts.lock);
1951		return;
1952	}
1953	if (!*bmc) {
1954		*bmc = 2;
1955		md_bitmap_count_page(&bitmap->counts, offset, 1);
1956		md_bitmap_set_pending(&bitmap->counts, offset);
1957		bitmap->allclean = 0;
1958	}
1959	if (needed)
1960		*bmc |= NEEDED_MASK;
1961	spin_unlock_irq(&bitmap->counts.lock);
1962}
1963
1964/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1965static void bitmap_dirty_bits(struct mddev *mddev, unsigned long s,
1966			      unsigned long e)
1967{
1968	unsigned long chunk;
1969	struct bitmap *bitmap = mddev->bitmap;
1970
1971	if (!bitmap)
1972		return;
1973
1974	for (chunk = s; chunk <= e; chunk++) {
1975		sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1976
1977		md_bitmap_set_memory_bits(bitmap, sec, 1);
1978		md_bitmap_file_set_bit(bitmap, sec);
1979		if (sec < bitmap->mddev->recovery_cp)
1980			/* We are asserting that the array is dirty,
1981			 * so move the recovery_cp address back so
1982			 * that it is obvious that it is dirty
1983			 */
1984			bitmap->mddev->recovery_cp = sec;
1985	}
1986}
1987
1988static void bitmap_flush(struct mddev *mddev)
1989{
1990	struct bitmap *bitmap = mddev->bitmap;
1991	long sleep;
1992
1993	if (!bitmap) /* there was no bitmap */
1994		return;
1995
1996	/* run the daemon_work three time to ensure everything is flushed
1997	 * that can be
1998	 */
1999	sleep = mddev->bitmap_info.daemon_sleep * 2;
2000	bitmap->daemon_lastrun -= sleep;
2001	bitmap_daemon_work(mddev);
2002	bitmap->daemon_lastrun -= sleep;
2003	bitmap_daemon_work(mddev);
2004	bitmap->daemon_lastrun -= sleep;
2005	bitmap_daemon_work(mddev);
2006	if (mddev->bitmap_info.external)
2007		md_super_wait(mddev);
2008	bitmap_update_sb(bitmap);
2009}
2010
2011static void md_bitmap_free(void *data)
2012{
2013	unsigned long k, pages;
2014	struct bitmap_page *bp;
2015	struct bitmap *bitmap = data;
2016
2017	if (!bitmap) /* there was no bitmap */
2018		return;
2019
2020	if (bitmap->sysfs_can_clear)
2021		sysfs_put(bitmap->sysfs_can_clear);
2022
2023	if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
2024		bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
2025		md_cluster_stop(bitmap->mddev);
2026
2027	/* Shouldn't be needed - but just in case.... */
2028	wait_event(bitmap->write_wait,
2029		   atomic_read(&bitmap->pending_writes) == 0);
2030
2031	/* release the bitmap file  */
2032	md_bitmap_file_unmap(&bitmap->storage);
2033
2034	bp = bitmap->counts.bp;
2035	pages = bitmap->counts.pages;
2036
2037	/* free all allocated memory */
2038
2039	if (bp) /* deallocate the page memory */
2040		for (k = 0; k < pages; k++)
2041			if (bp[k].map && !bp[k].hijacked)
2042				kfree(bp[k].map);
2043	kfree(bp);
2044	kfree(bitmap);
2045}
2046
2047static void bitmap_start_behind_write(struct mddev *mddev)
2048{
2049	struct bitmap *bitmap = mddev->bitmap;
2050	int bw;
2051
2052	if (!bitmap)
2053		return;
2054
2055	atomic_inc(&bitmap->behind_writes);
2056	bw = atomic_read(&bitmap->behind_writes);
2057	if (bw > bitmap->behind_writes_used)
2058		bitmap->behind_writes_used = bw;
2059
2060	pr_debug("inc write-behind count %d/%lu\n",
2061		 bw, bitmap->mddev->bitmap_info.max_write_behind);
2062}
2063
2064static void bitmap_end_behind_write(struct mddev *mddev)
2065{
2066	struct bitmap *bitmap = mddev->bitmap;
2067
2068	if (!bitmap)
2069		return;
2070
2071	if (atomic_dec_and_test(&bitmap->behind_writes))
2072		wake_up(&bitmap->behind_wait);
2073	pr_debug("dec write-behind count %d/%lu\n",
2074		 atomic_read(&bitmap->behind_writes),
2075		 bitmap->mddev->bitmap_info.max_write_behind);
2076}
2077
2078static void bitmap_wait_behind_writes(struct mddev *mddev)
2079{
2080	struct bitmap *bitmap = mddev->bitmap;
2081
2082	/* wait for behind writes to complete */
2083	if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2084		pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
2085			 mdname(mddev));
2086		/* need to kick something here to make sure I/O goes? */
2087		wait_event(bitmap->behind_wait,
2088			   atomic_read(&bitmap->behind_writes) == 0);
2089	}
2090}
2091
2092static void bitmap_destroy(struct mddev *mddev)
2093{
2094	struct bitmap *bitmap = mddev->bitmap;
2095
2096	if (!bitmap) /* there was no bitmap */
2097		return;
2098
2099	bitmap_wait_behind_writes(mddev);
2100	if (!mddev->serialize_policy)
2101		mddev_destroy_serial_pool(mddev, NULL);
2102
2103	mutex_lock(&mddev->bitmap_info.mutex);
2104	spin_lock(&mddev->lock);
2105	mddev->bitmap = NULL; /* disconnect from the md device */
2106	spin_unlock(&mddev->lock);
2107	mutex_unlock(&mddev->bitmap_info.mutex);
2108	mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
2109
2110	md_bitmap_free(bitmap);
2111}
2112
2113/*
2114 * initialize the bitmap structure
2115 * if this returns an error, bitmap_destroy must be called to do clean up
2116 * once mddev->bitmap is set
2117 */
2118static struct bitmap *__bitmap_create(struct mddev *mddev, int slot)
2119{
2120	struct bitmap *bitmap;
2121	sector_t blocks = mddev->resync_max_sectors;
2122	struct file *file = mddev->bitmap_info.file;
2123	int err;
2124	struct kernfs_node *bm = NULL;
2125
2126	BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
2127
2128	BUG_ON(file && mddev->bitmap_info.offset);
2129
2130	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
2131		pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
2132			  mdname(mddev));
2133		return ERR_PTR(-EBUSY);
2134	}
2135
2136	bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
2137	if (!bitmap)
2138		return ERR_PTR(-ENOMEM);
2139
2140	spin_lock_init(&bitmap->counts.lock);
2141	atomic_set(&bitmap->pending_writes, 0);
2142	init_waitqueue_head(&bitmap->write_wait);
2143	init_waitqueue_head(&bitmap->overflow_wait);
2144	init_waitqueue_head(&bitmap->behind_wait);
2145
2146	bitmap->mddev = mddev;
2147	bitmap->cluster_slot = slot;
2148
2149	if (mddev->kobj.sd)
2150		bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
2151	if (bm) {
2152		bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
2153		sysfs_put(bm);
2154	} else
2155		bitmap->sysfs_can_clear = NULL;
2156
2157	bitmap->storage.file = file;
2158	if (file) {
2159		get_file(file);
2160		/* As future accesses to this file will use bmap,
2161		 * and bypass the page cache, we must sync the file
2162		 * first.
2163		 */
2164		vfs_fsync(file, 1);
2165	}
2166	/* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
2167	if (!mddev->bitmap_info.external) {
2168		/*
2169		 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
2170		 * instructing us to create a new on-disk bitmap instance.
2171		 */
2172		if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
2173			err = md_bitmap_new_disk_sb(bitmap);
2174		else
2175			err = md_bitmap_read_sb(bitmap);
2176	} else {
2177		err = 0;
2178		if (mddev->bitmap_info.chunksize == 0 ||
2179		    mddev->bitmap_info.daemon_sleep == 0)
2180			/* chunksize and time_base need to be
2181			 * set first. */
2182			err = -EINVAL;
2183	}
2184	if (err)
2185		goto error;
2186
2187	bitmap->daemon_lastrun = jiffies;
2188	err = __bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize,
2189			      true);
2190	if (err)
2191		goto error;
2192
2193	pr_debug("created bitmap (%lu pages) for device %s\n",
2194		 bitmap->counts.pages, bmname(bitmap));
2195
2196	err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
2197	if (err)
2198		goto error;
2199
2200	return bitmap;
2201 error:
2202	md_bitmap_free(bitmap);
2203	return ERR_PTR(err);
2204}
2205
2206static int bitmap_create(struct mddev *mddev, int slot)
2207{
2208	struct bitmap *bitmap = __bitmap_create(mddev, slot);
2209
2210	if (IS_ERR(bitmap))
2211		return PTR_ERR(bitmap);
2212
2213	mddev->bitmap = bitmap;
2214	return 0;
2215}
2216
2217static int bitmap_load(struct mddev *mddev)
2218{
2219	int err = 0;
2220	sector_t start = 0;
2221	sector_t sector = 0;
2222	struct bitmap *bitmap = mddev->bitmap;
2223	struct md_rdev *rdev;
2224
2225	if (!bitmap)
2226		goto out;
2227
2228	rdev_for_each(rdev, mddev)
2229		mddev_create_serial_pool(mddev, rdev);
2230
2231	if (mddev_is_clustered(mddev))
2232		md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
2233
2234	/* Clear out old bitmap info first:  Either there is none, or we
2235	 * are resuming after someone else has possibly changed things,
2236	 * so we should forget old cached info.
2237	 * All chunks should be clean, but some might need_sync.
2238	 */
2239	while (sector < mddev->resync_max_sectors) {
2240		sector_t blocks;
2241		bitmap_start_sync(mddev, sector, &blocks, false);
2242		sector += blocks;
2243	}
2244	bitmap_close_sync(mddev);
2245
2246	if (mddev->degraded == 0
2247	    || bitmap->events_cleared == mddev->events)
2248		/* no need to keep dirty bits to optimise a
2249		 * re-add of a missing device */
2250		start = mddev->recovery_cp;
2251
2252	mutex_lock(&mddev->bitmap_info.mutex);
2253	err = md_bitmap_init_from_disk(bitmap, start);
2254	mutex_unlock(&mddev->bitmap_info.mutex);
2255
2256	if (err)
2257		goto out;
2258	clear_bit(BITMAP_STALE, &bitmap->flags);
2259
2260	/* Kick recovery in case any bits were set */
2261	set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
2262
2263	mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
2264	md_wakeup_thread(mddev->thread);
2265
2266	bitmap_update_sb(bitmap);
2267
2268	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
2269		err = -EIO;
2270out:
2271	return err;
2272}
2273
2274/* caller need to free returned bitmap with md_bitmap_free() */
2275static void *bitmap_get_from_slot(struct mddev *mddev, int slot)
2276{
2277	int rv = 0;
2278	struct bitmap *bitmap;
2279
2280	bitmap = __bitmap_create(mddev, slot);
2281	if (IS_ERR(bitmap)) {
2282		rv = PTR_ERR(bitmap);
2283		return ERR_PTR(rv);
2284	}
2285
2286	rv = md_bitmap_init_from_disk(bitmap, 0);
2287	if (rv) {
2288		md_bitmap_free(bitmap);
2289		return ERR_PTR(rv);
2290	}
2291
2292	return bitmap;
2293}
2294
2295/* Loads the bitmap associated with slot and copies the resync information
2296 * to our bitmap
2297 */
2298static int bitmap_copy_from_slot(struct mddev *mddev, int slot, sector_t *low,
2299				 sector_t *high, bool clear_bits)
2300{
2301	int rv = 0, i, j;
2302	sector_t block, lo = 0, hi = 0;
2303	struct bitmap_counts *counts;
2304	struct bitmap *bitmap;
2305
2306	bitmap = bitmap_get_from_slot(mddev, slot);
2307	if (IS_ERR(bitmap)) {
2308		pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
2309		return -1;
2310	}
2311
2312	counts = &bitmap->counts;
2313	for (j = 0; j < counts->chunks; j++) {
2314		block = (sector_t)j << counts->chunkshift;
2315		if (md_bitmap_file_test_bit(bitmap, block)) {
2316			if (!lo)
2317				lo = block;
2318			hi = block;
2319			md_bitmap_file_clear_bit(bitmap, block);
2320			md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2321			md_bitmap_file_set_bit(mddev->bitmap, block);
2322		}
2323	}
2324
2325	if (clear_bits) {
2326		bitmap_update_sb(bitmap);
2327		/* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2328		 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2329		for (i = 0; i < bitmap->storage.file_pages; i++)
2330			if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2331				set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2332		__bitmap_unplug(bitmap);
2333	}
2334	__bitmap_unplug(mddev->bitmap);
2335	*low = lo;
2336	*high = hi;
2337	md_bitmap_free(bitmap);
2338
2339	return rv;
2340}
2341
2342static void bitmap_set_pages(void *data, unsigned long pages)
2343{
2344	struct bitmap *bitmap = data;
2345
2346	bitmap->counts.pages = pages;
2347}
2348
2349static int bitmap_get_stats(void *data, struct md_bitmap_stats *stats)
2350{
2351	struct bitmap_storage *storage;
2352	struct bitmap_counts *counts;
2353	struct bitmap *bitmap = data;
2354	bitmap_super_t *sb;
2355
2356	if (!bitmap)
2357		return -ENOENT;
2358	if (bitmap->mddev->bitmap_info.external)
2359		return -ENOENT;
2360	if (!bitmap->storage.sb_page) /* no superblock */
2361		return -EINVAL;
2362	sb = kmap_local_page(bitmap->storage.sb_page);
2363	stats->sync_size = le64_to_cpu(sb->sync_size);
2364	kunmap_local(sb);
2365
2366	counts = &bitmap->counts;
2367	stats->missing_pages = counts->missing_pages;
2368	stats->pages = counts->pages;
2369
2370	storage = &bitmap->storage;
2371	stats->file_pages = storage->file_pages;
2372	stats->file = storage->file;
2373
2374	stats->behind_writes = atomic_read(&bitmap->behind_writes);
2375	stats->behind_wait = wq_has_sleeper(&bitmap->behind_wait);
2376	stats->events_cleared = bitmap->events_cleared;
2377	return 0;
2378}
2379
2380static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2381			   int chunksize, bool init)
2382{
2383	/* If chunk_size is 0, choose an appropriate chunk size.
2384	 * Then possibly allocate new storage space.
2385	 * Then quiesce, copy bits, replace bitmap, and re-start
2386	 *
2387	 * This function is called both to set up the initial bitmap
2388	 * and to resize the bitmap while the array is active.
2389	 * If this happens as a result of the array being resized,
2390	 * chunksize will be zero, and we need to choose a suitable
2391	 * chunksize, otherwise we use what we are given.
2392	 */
2393	struct bitmap_storage store;
2394	struct bitmap_counts old_counts;
2395	unsigned long chunks;
2396	sector_t block;
2397	sector_t old_blocks, new_blocks;
2398	int chunkshift;
2399	int ret = 0;
2400	long pages;
2401	struct bitmap_page *new_bp;
2402
2403	if (bitmap->storage.file && !init) {
2404		pr_info("md: cannot resize file-based bitmap\n");
2405		return -EINVAL;
2406	}
2407
2408	if (chunksize == 0) {
2409		/* If there is enough space, leave the chunk size unchanged,
2410		 * else increase by factor of two until there is enough space.
2411		 */
2412		long bytes;
2413		long space = bitmap->mddev->bitmap_info.space;
2414
2415		if (space == 0) {
2416			/* We don't know how much space there is, so limit
2417			 * to current size - in sectors.
2418			 */
2419			bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2420			if (!bitmap->mddev->bitmap_info.external)
2421				bytes += sizeof(bitmap_super_t);
2422			space = DIV_ROUND_UP(bytes, 512);
2423			bitmap->mddev->bitmap_info.space = space;
2424		}
2425		chunkshift = bitmap->counts.chunkshift;
2426		chunkshift--;
2427		do {
2428			/* 'chunkshift' is shift from block size to chunk size */
2429			chunkshift++;
2430			chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2431			bytes = DIV_ROUND_UP(chunks, 8);
2432			if (!bitmap->mddev->bitmap_info.external)
2433				bytes += sizeof(bitmap_super_t);
2434		} while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2435			(BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2436	} else
2437		chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2438
2439	chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2440	memset(&store, 0, sizeof(store));
2441	if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2442		ret = md_bitmap_storage_alloc(&store, chunks,
2443					      !bitmap->mddev->bitmap_info.external,
2444					      mddev_is_clustered(bitmap->mddev)
2445					      ? bitmap->cluster_slot : 0);
2446	if (ret) {
2447		md_bitmap_file_unmap(&store);
2448		goto err;
2449	}
2450
2451	pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2452
2453	new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2454	ret = -ENOMEM;
2455	if (!new_bp) {
2456		md_bitmap_file_unmap(&store);
2457		goto err;
2458	}
2459
2460	if (!init)
2461		bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2462
2463	store.file = bitmap->storage.file;
2464	bitmap->storage.file = NULL;
2465
2466	if (store.sb_page && bitmap->storage.sb_page)
2467		memcpy(page_address(store.sb_page),
2468		       page_address(bitmap->storage.sb_page),
2469		       sizeof(bitmap_super_t));
2470	spin_lock_irq(&bitmap->counts.lock);
2471	md_bitmap_file_unmap(&bitmap->storage);
2472	bitmap->storage = store;
2473
2474	old_counts = bitmap->counts;
2475	bitmap->counts.bp = new_bp;
2476	bitmap->counts.pages = pages;
2477	bitmap->counts.missing_pages = pages;
2478	bitmap->counts.chunkshift = chunkshift;
2479	bitmap->counts.chunks = chunks;
2480	bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2481						     BITMAP_BLOCK_SHIFT);
2482
2483	blocks = min(old_counts.chunks << old_counts.chunkshift,
2484		     chunks << chunkshift);
2485
2486	/* For cluster raid, need to pre-allocate bitmap */
2487	if (mddev_is_clustered(bitmap->mddev)) {
2488		unsigned long page;
2489		for (page = 0; page < pages; page++) {
2490			ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2491			if (ret) {
2492				unsigned long k;
2493
2494				/* deallocate the page memory */
2495				for (k = 0; k < page; k++) {
2496					kfree(new_bp[k].map);
2497				}
2498				kfree(new_bp);
2499
2500				/* restore some fields from old_counts */
2501				bitmap->counts.bp = old_counts.bp;
2502				bitmap->counts.pages = old_counts.pages;
2503				bitmap->counts.missing_pages = old_counts.pages;
2504				bitmap->counts.chunkshift = old_counts.chunkshift;
2505				bitmap->counts.chunks = old_counts.chunks;
2506				bitmap->mddev->bitmap_info.chunksize =
2507					1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2508				blocks = old_counts.chunks << old_counts.chunkshift;
2509				pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2510				break;
2511			} else
2512				bitmap->counts.bp[page].count += 1;
2513		}
2514	}
2515
2516	for (block = 0; block < blocks; ) {
2517		bitmap_counter_t *bmc_old, *bmc_new;
2518		int set;
2519
2520		bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2521		set = bmc_old && NEEDED(*bmc_old);
2522
2523		if (set) {
2524			bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2525			if (bmc_new) {
2526				if (*bmc_new == 0) {
2527					/* need to set on-disk bits too. */
2528					sector_t end = block + new_blocks;
2529					sector_t start = block >> chunkshift;
2530
2531					start <<= chunkshift;
2532					while (start < end) {
2533						md_bitmap_file_set_bit(bitmap, block);
2534						start += 1 << chunkshift;
2535					}
2536					*bmc_new = 2;
2537					md_bitmap_count_page(&bitmap->counts, block, 1);
2538					md_bitmap_set_pending(&bitmap->counts, block);
2539				}
2540				*bmc_new |= NEEDED_MASK;
2541			}
2542			if (new_blocks < old_blocks)
2543				old_blocks = new_blocks;
2544		}
2545		block += old_blocks;
2546	}
2547
2548	if (bitmap->counts.bp != old_counts.bp) {
2549		unsigned long k;
2550		for (k = 0; k < old_counts.pages; k++)
2551			if (!old_counts.bp[k].hijacked)
2552				kfree(old_counts.bp[k].map);
2553		kfree(old_counts.bp);
2554	}
2555
2556	if (!init) {
2557		int i;
2558		while (block < (chunks << chunkshift)) {
2559			bitmap_counter_t *bmc;
2560			bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2561			if (bmc) {
2562				/* new space.  It needs to be resynced, so
2563				 * we set NEEDED_MASK.
2564				 */
2565				if (*bmc == 0) {
2566					*bmc = NEEDED_MASK | 2;
2567					md_bitmap_count_page(&bitmap->counts, block, 1);
2568					md_bitmap_set_pending(&bitmap->counts, block);
2569				}
2570			}
2571			block += new_blocks;
2572		}
2573		for (i = 0; i < bitmap->storage.file_pages; i++)
2574			set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2575	}
2576	spin_unlock_irq(&bitmap->counts.lock);
2577
2578	if (!init) {
2579		__bitmap_unplug(bitmap);
2580		bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2581	}
2582	ret = 0;
2583err:
2584	return ret;
2585}
2586
2587static int bitmap_resize(struct mddev *mddev, sector_t blocks, int chunksize,
2588			 bool init)
2589{
2590	struct bitmap *bitmap = mddev->bitmap;
2591
2592	if (!bitmap)
2593		return 0;
2594
2595	return __bitmap_resize(bitmap, blocks, chunksize, init);
2596}
2597
2598static ssize_t
2599location_show(struct mddev *mddev, char *page)
2600{
2601	ssize_t len;
2602	if (mddev->bitmap_info.file)
2603		len = sprintf(page, "file");
2604	else if (mddev->bitmap_info.offset)
2605		len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2606	else
2607		len = sprintf(page, "none");
2608	len += sprintf(page+len, "\n");
2609	return len;
2610}
2611
2612static ssize_t
2613location_store(struct mddev *mddev, const char *buf, size_t len)
2614{
2615	int rv;
2616
2617	rv = mddev_suspend_and_lock(mddev);
2618	if (rv)
2619		return rv;
2620
2621	if (mddev->pers) {
2622		if (mddev->recovery || mddev->sync_thread) {
2623			rv = -EBUSY;
2624			goto out;
2625		}
2626	}
2627
2628	if (mddev->bitmap || mddev->bitmap_info.file ||
2629	    mddev->bitmap_info.offset) {
2630		/* bitmap already configured.  Only option is to clear it */
2631		if (strncmp(buf, "none", 4) != 0) {
2632			rv = -EBUSY;
2633			goto out;
2634		}
2635
2636		bitmap_destroy(mddev);
2637		mddev->bitmap_info.offset = 0;
2638		if (mddev->bitmap_info.file) {
2639			struct file *f = mddev->bitmap_info.file;
2640			mddev->bitmap_info.file = NULL;
2641			fput(f);
2642		}
2643	} else {
2644		/* No bitmap, OK to set a location */
2645		long long offset;
2646
2647		if (strncmp(buf, "none", 4) == 0)
2648			/* nothing to be done */;
2649		else if (strncmp(buf, "file:", 5) == 0) {
2650			/* Not supported yet */
2651			rv = -EINVAL;
2652			goto out;
2653		} else {
2654			if (buf[0] == '+')
2655				rv = kstrtoll(buf+1, 10, &offset);
2656			else
2657				rv = kstrtoll(buf, 10, &offset);
2658			if (rv)
2659				goto out;
2660			if (offset == 0) {
2661				rv = -EINVAL;
2662				goto out;
2663			}
2664			if (mddev->bitmap_info.external == 0 &&
2665			    mddev->major_version == 0 &&
2666			    offset != mddev->bitmap_info.default_offset) {
2667				rv = -EINVAL;
2668				goto out;
2669			}
2670
2671			mddev->bitmap_info.offset = offset;
2672			rv = bitmap_create(mddev, -1);
2673			if (rv)
2674				goto out;
2675
2676			rv = bitmap_load(mddev);
2677			if (rv) {
2678				mddev->bitmap_info.offset = 0;
2679				bitmap_destroy(mddev);
2680				goto out;
2681			}
2682		}
2683	}
2684	if (!mddev->external) {
2685		/* Ensure new bitmap info is stored in
2686		 * metadata promptly.
2687		 */
2688		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2689		md_wakeup_thread(mddev->thread);
2690	}
2691	rv = 0;
2692out:
2693	mddev_unlock_and_resume(mddev);
2694	if (rv)
2695		return rv;
2696	return len;
2697}
2698
2699static struct md_sysfs_entry bitmap_location =
2700__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2701
2702/* 'bitmap/space' is the space available at 'location' for the
2703 * bitmap.  This allows the kernel to know when it is safe to
2704 * resize the bitmap to match a resized array.
2705 */
2706static ssize_t
2707space_show(struct mddev *mddev, char *page)
2708{
2709	return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2710}
2711
2712static ssize_t
2713space_store(struct mddev *mddev, const char *buf, size_t len)
2714{
2715	struct bitmap *bitmap;
2716	unsigned long sectors;
2717	int rv;
2718
2719	rv = kstrtoul(buf, 10, &sectors);
2720	if (rv)
2721		return rv;
2722
2723	if (sectors == 0)
2724		return -EINVAL;
2725
2726	bitmap = mddev->bitmap;
2727	if (bitmap && sectors < (bitmap->storage.bytes + 511) >> 9)
2728		return -EFBIG; /* Bitmap is too big for this small space */
2729
2730	/* could make sure it isn't too big, but that isn't really
2731	 * needed - user-space should be careful.
2732	 */
2733	mddev->bitmap_info.space = sectors;
2734	return len;
2735}
2736
2737static struct md_sysfs_entry bitmap_space =
2738__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2739
2740static ssize_t
2741timeout_show(struct mddev *mddev, char *page)
2742{
2743	ssize_t len;
2744	unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2745	unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2746
2747	len = sprintf(page, "%lu", secs);
2748	if (jifs)
2749		len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2750	len += sprintf(page+len, "\n");
2751	return len;
2752}
2753
2754static ssize_t
2755timeout_store(struct mddev *mddev, const char *buf, size_t len)
2756{
2757	/* timeout can be set at any time */
2758	unsigned long timeout;
2759	int rv = strict_strtoul_scaled(buf, &timeout, 4);
2760	if (rv)
2761		return rv;
2762
2763	/* just to make sure we don't overflow... */
2764	if (timeout >= LONG_MAX / HZ)
2765		return -EINVAL;
2766
2767	timeout = timeout * HZ / 10000;
2768
2769	if (timeout >= MAX_SCHEDULE_TIMEOUT)
2770		timeout = MAX_SCHEDULE_TIMEOUT-1;
2771	if (timeout < 1)
2772		timeout = 1;
2773
2774	mddev->bitmap_info.daemon_sleep = timeout;
2775	mddev_set_timeout(mddev, timeout, false);
2776	md_wakeup_thread(mddev->thread);
2777
2778	return len;
2779}
2780
2781static struct md_sysfs_entry bitmap_timeout =
2782__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2783
2784static ssize_t
2785backlog_show(struct mddev *mddev, char *page)
2786{
2787	return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2788}
2789
2790static ssize_t
2791backlog_store(struct mddev *mddev, const char *buf, size_t len)
2792{
2793	unsigned long backlog;
2794	unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2795	struct md_rdev *rdev;
2796	bool has_write_mostly = false;
2797	int rv = kstrtoul(buf, 10, &backlog);
2798	if (rv)
2799		return rv;
2800	if (backlog > COUNTER_MAX)
2801		return -EINVAL;
2802
2803	rv = mddev_suspend_and_lock(mddev);
2804	if (rv)
2805		return rv;
2806
2807	/*
2808	 * Without write mostly device, it doesn't make sense to set
2809	 * backlog for max_write_behind.
2810	 */
2811	rdev_for_each(rdev, mddev) {
2812		if (test_bit(WriteMostly, &rdev->flags)) {
2813			has_write_mostly = true;
2814			break;
2815		}
2816	}
2817	if (!has_write_mostly) {
2818		pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2819				    mdname(mddev));
2820		mddev_unlock(mddev);
2821		return -EINVAL;
2822	}
2823
2824	mddev->bitmap_info.max_write_behind = backlog;
2825	if (!backlog && mddev->serial_info_pool) {
2826		/* serial_info_pool is not needed if backlog is zero */
2827		if (!mddev->serialize_policy)
2828			mddev_destroy_serial_pool(mddev, NULL);
2829	} else if (backlog && !mddev->serial_info_pool) {
2830		/* serial_info_pool is needed since backlog is not zero */
2831		rdev_for_each(rdev, mddev)
2832			mddev_create_serial_pool(mddev, rdev);
2833	}
2834	if (old_mwb != backlog)
2835		bitmap_update_sb(mddev->bitmap);
2836
2837	mddev_unlock_and_resume(mddev);
2838	return len;
2839}
2840
2841static struct md_sysfs_entry bitmap_backlog =
2842__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2843
2844static ssize_t
2845chunksize_show(struct mddev *mddev, char *page)
2846{
2847	return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2848}
2849
2850static ssize_t
2851chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2852{
2853	/* Can only be changed when no bitmap is active */
2854	int rv;
2855	unsigned long csize;
2856	if (mddev->bitmap)
2857		return -EBUSY;
2858	rv = kstrtoul(buf, 10, &csize);
2859	if (rv)
2860		return rv;
2861	if (csize < 512 ||
2862	    !is_power_of_2(csize))
2863		return -EINVAL;
2864	if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2865		sizeof(((bitmap_super_t *)0)->chunksize))))
2866		return -EOVERFLOW;
2867	mddev->bitmap_info.chunksize = csize;
2868	return len;
2869}
2870
2871static struct md_sysfs_entry bitmap_chunksize =
2872__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2873
2874static ssize_t metadata_show(struct mddev *mddev, char *page)
2875{
2876	if (mddev_is_clustered(mddev))
2877		return sprintf(page, "clustered\n");
2878	return sprintf(page, "%s\n", (mddev->bitmap_info.external
2879				      ? "external" : "internal"));
2880}
2881
2882static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2883{
2884	if (mddev->bitmap ||
2885	    mddev->bitmap_info.file ||
2886	    mddev->bitmap_info.offset)
2887		return -EBUSY;
2888	if (strncmp(buf, "external", 8) == 0)
2889		mddev->bitmap_info.external = 1;
2890	else if ((strncmp(buf, "internal", 8) == 0) ||
2891			(strncmp(buf, "clustered", 9) == 0))
2892		mddev->bitmap_info.external = 0;
2893	else
2894		return -EINVAL;
2895	return len;
2896}
2897
2898static struct md_sysfs_entry bitmap_metadata =
2899__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2900
2901static ssize_t can_clear_show(struct mddev *mddev, char *page)
2902{
2903	int len;
2904	struct bitmap *bitmap;
2905
2906	spin_lock(&mddev->lock);
2907	bitmap = mddev->bitmap;
2908	if (bitmap)
2909		len = sprintf(page, "%s\n", (bitmap->need_sync ? "false" :
2910								 "true"));
2911	else
2912		len = sprintf(page, "\n");
2913	spin_unlock(&mddev->lock);
2914	return len;
2915}
2916
2917static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2918{
2919	struct bitmap *bitmap = mddev->bitmap;
2920
2921	if (!bitmap)
2922		return -ENOENT;
2923
2924	if (strncmp(buf, "false", 5) == 0) {
2925		bitmap->need_sync = 1;
2926		return len;
2927	}
2928
2929	if (strncmp(buf, "true", 4) == 0) {
2930		if (mddev->degraded)
2931			return -EBUSY;
2932		bitmap->need_sync = 0;
2933		return len;
2934	}
2935
2936	return -EINVAL;
2937}
2938
2939static struct md_sysfs_entry bitmap_can_clear =
2940__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2941
2942static ssize_t
2943behind_writes_used_show(struct mddev *mddev, char *page)
2944{
2945	ssize_t ret;
2946	struct bitmap *bitmap;
2947
2948	spin_lock(&mddev->lock);
2949	bitmap = mddev->bitmap;
2950	if (!bitmap)
2951		ret = sprintf(page, "0\n");
2952	else
2953		ret = sprintf(page, "%lu\n", bitmap->behind_writes_used);
2954	spin_unlock(&mddev->lock);
2955
2956	return ret;
2957}
2958
2959static ssize_t
2960behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2961{
2962	struct bitmap *bitmap = mddev->bitmap;
2963
2964	if (bitmap)
2965		bitmap->behind_writes_used = 0;
2966	return len;
2967}
2968
2969static struct md_sysfs_entry max_backlog_used =
2970__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2971       behind_writes_used_show, behind_writes_used_reset);
2972
2973static struct attribute *md_bitmap_attrs[] = {
2974	&bitmap_location.attr,
2975	&bitmap_space.attr,
2976	&bitmap_timeout.attr,
2977	&bitmap_backlog.attr,
2978	&bitmap_chunksize.attr,
2979	&bitmap_metadata.attr,
2980	&bitmap_can_clear.attr,
2981	&max_backlog_used.attr,
2982	NULL
2983};
2984const struct attribute_group md_bitmap_group = {
2985	.name = "bitmap",
2986	.attrs = md_bitmap_attrs,
2987};
2988
2989static struct bitmap_operations bitmap_ops = {
2990	.enabled		= bitmap_enabled,
2991	.create			= bitmap_create,
2992	.resize			= bitmap_resize,
2993	.load			= bitmap_load,
2994	.destroy		= bitmap_destroy,
2995	.flush			= bitmap_flush,
2996	.write_all		= bitmap_write_all,
2997	.dirty_bits		= bitmap_dirty_bits,
2998	.unplug			= bitmap_unplug,
2999	.daemon_work		= bitmap_daemon_work,
3000
3001	.start_behind_write	= bitmap_start_behind_write,
3002	.end_behind_write	= bitmap_end_behind_write,
3003	.wait_behind_writes	= bitmap_wait_behind_writes,
3004
3005	.startwrite		= bitmap_startwrite,
3006	.endwrite		= bitmap_endwrite,
3007	.start_sync		= bitmap_start_sync,
3008	.end_sync		= bitmap_end_sync,
3009	.cond_end_sync		= bitmap_cond_end_sync,
3010	.close_sync		= bitmap_close_sync,
3011
3012	.update_sb		= bitmap_update_sb,
3013	.get_stats		= bitmap_get_stats,
3014
3015	.sync_with_cluster	= bitmap_sync_with_cluster,
3016	.get_from_slot		= bitmap_get_from_slot,
3017	.copy_from_slot		= bitmap_copy_from_slot,
3018	.set_pages		= bitmap_set_pages,
3019	.free			= md_bitmap_free,
3020};
3021
3022void mddev_set_bitmap_ops(struct mddev *mddev)
3023{
3024	mddev->bitmap_ops = &bitmap_ops;
3025}