FILE
md-bitmap.c

File preview with syntax highlighting.
linux-6.19.14/drivers/md/md-bitmap.c C
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   4 *
   5 * bitmap_create  - sets up the bitmap structure
   6 * bitmap_destroy - destroys the bitmap structure
   7 *
   8 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   9 * - added disk storage for bitmap
  10 * - changes to allow various bitmap chunk sizes
  11 */
  12
  13/*
  14 * Still to do:
  15 *
  16 * flush after percent set rather than just time based. (maybe both).
  17 */
  18
  19#include <linux/blkdev.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/slab.h>
  23#include <linux/init.h>
  24#include <linux/timer.h>
  25#include <linux/sched.h>
  26#include <linux/list.h>
  27#include <linux/file.h>
  28#include <linux/mount.h>
  29#include <linux/buffer_head.h>
  30#include <linux/seq_file.h>
  31#include <trace/events/block.h>
  32
  33#include "md.h"
  34#include "md-bitmap.h"
  35#include "md-cluster.h"
  36
  37/*
  38 * in-memory bitmap:
  39 *
  40 * Use 16 bit block counters to track pending writes to each "chunk".
  41 * The 2 high order bits are special-purpose, the first is a flag indicating
  42 * whether a resync is needed.  The second is a flag indicating whether a
  43 * resync is active.
  44 * This means that the counter is actually 14 bits:
  45 *
  46 * +--------+--------+------------------------------------------------+
  47 * | resync | resync |               counter                          |
  48 * | needed | active |                                                |
  49 * |  (0-1) |  (0-1) |              (0-16383)                         |
  50 * +--------+--------+------------------------------------------------+
  51 *
  52 * The "resync needed" bit is set when:
  53 *    a '1' bit is read from storage at startup.
  54 *    a write request fails on some drives
  55 *    a resync is aborted on a chunk with 'resync active' set
  56 * It is cleared (and resync-active set) when a resync starts across all drives
  57 * of the chunk.
  58 *
  59 *
  60 * The "resync active" bit is set when:
  61 *    a resync is started on all drives, and resync_needed is set.
  62 *       resync_needed will be cleared (as long as resync_active wasn't already set).
  63 * It is cleared when a resync completes.
  64 *
  65 * The counter counts pending write requests, plus the on-disk bit.
  66 * When the counter is '1' and the resync bits are clear, the on-disk
  67 * bit can be cleared as well, thus setting the counter to 0.
  68 * When we set a bit, or in the counter (to start a write), if the fields is
  69 * 0, we first set the disk bit and set the counter to 1.
  70 *
  71 * If the counter is 0, the on-disk bit is clear and the stripe is clean
  72 * Anything that dirties the stripe pushes the counter to 2 (at least)
  73 * and sets the on-disk bit (lazily).
  74 * If a periodic sweep find the counter at 2, it is decremented to 1.
  75 * If the sweep find the counter at 1, the on-disk bit is cleared and the
  76 * counter goes to zero.
  77 *
  78 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
  79 * counters as a fallback when "page" memory cannot be allocated:
  80 *
  81 * Normal case (page memory allocated):
  82 *
  83 *     page pointer (32-bit)
  84 *
  85 *     [ ] ------+
  86 *               |
  87 *               +-------> [   ][   ]..[   ] (4096 byte page == 2048 counters)
  88 *                          c1   c2    c2048
  89 *
  90 * Hijacked case (page memory allocation failed):
  91 *
  92 *     hijacked page pointer (32-bit)
  93 *
  94 *     [		  ][		  ] (no page memory allocated)
  95 *      counter #1 (16-bit) counter #2 (16-bit)
  96 *
  97 */
  98
  99typedef __u16 bitmap_counter_t;
 100
 101#define PAGE_BITS (PAGE_SIZE << 3)
 102#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
 103
 104#define COUNTER_BITS 16
 105#define COUNTER_BIT_SHIFT 4
 106#define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
 107
 108#define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
 109#define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
 110#define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
 111
 112#define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
 113#define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
 114#define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
 115
 116/* how many counters per page? */
 117#define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
 118/* same, except a shift value for more efficient bitops */
 119#define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
 120/* same, except a mask value for more efficient bitops */
 121#define PAGE_COUNTER_MASK  (PAGE_COUNTER_RATIO - 1)
 122
 123#define BITMAP_BLOCK_SHIFT 9
 124
 125/*
 126 * bitmap structures:
 127 */
 128
 129/* the in-memory bitmap is represented by bitmap_pages */
 130struct bitmap_page {
 131	/*
 132	 * map points to the actual memory page
 133	 */
 134	char *map;
 135	/*
 136	 * in emergencies (when map cannot be alloced), hijack the map
 137	 * pointer and use it as two counters itself
 138	 */
 139	unsigned int hijacked:1;
 140	/*
 141	 * If any counter in this page is '1' or '2' - and so could be
 142	 * cleared then that page is marked as 'pending'
 143	 */
 144	unsigned int pending:1;
 145	/*
 146	 * count of dirty bits on the page
 147	 */
 148	unsigned int  count:30;
 149};
 150
 151/* the main bitmap structure - one per mddev */
 152struct bitmap {
 153
 154	struct bitmap_counts {
 155		spinlock_t lock;
 156		struct bitmap_page *bp;
 157		/* total number of pages in the bitmap */
 158		unsigned long pages;
 159		/* number of pages not yet allocated */
 160		unsigned long missing_pages;
 161		/* chunksize = 2^chunkshift (for bitops) */
 162		unsigned long chunkshift;
 163		/* total number of data chunks for the array */
 164		unsigned long chunks;
 165	} counts;
 166
 167	struct mddev *mddev; /* the md device that the bitmap is for */
 168
 169	__u64	events_cleared;
 170	int need_sync;
 171
 172	struct bitmap_storage {
 173		/* backing disk file */
 174		struct file *file;
 175		/* cached copy of the bitmap file superblock */
 176		struct page *sb_page;
 177		unsigned long sb_index;
 178		/* list of cache pages for the file */
 179		struct page **filemap;
 180		/* attributes associated filemap pages */
 181		unsigned long *filemap_attr;
 182		/* number of pages in the file */
 183		unsigned long file_pages;
 184		/* total bytes in the bitmap */
 185		unsigned long bytes;
 186	} storage;
 187
 188	unsigned long flags;
 189
 190	int allclean;
 191
 192	atomic_t behind_writes;
 193	/* highest actual value at runtime */
 194	unsigned long behind_writes_used;
 195
 196	/*
 197	 * the bitmap daemon - periodically wakes up and sweeps the bitmap
 198	 * file, cleaning up bits and flushing out pages to disk as necessary
 199	 */
 200	unsigned long daemon_lastrun; /* jiffies of last run */
 201	/*
 202	 * when we lasted called end_sync to update bitmap with resync
 203	 * progress.
 204	 */
 205	unsigned long last_end_sync;
 206
 207	/* pending writes to the bitmap file */
 208	atomic_t pending_writes;
 209	wait_queue_head_t write_wait;
 210	wait_queue_head_t overflow_wait;
 211	wait_queue_head_t behind_wait;
 212
 213	struct kernfs_node *sysfs_can_clear;
 214	/* slot offset for clustered env */
 215	int cluster_slot;
 216};
 217
 218static struct workqueue_struct *md_bitmap_wq;
 219
 220static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
 221			   int chunksize, bool init);
 222
 223static inline char *bmname(struct bitmap *bitmap)
 224{
 225	return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
 226}
 227
 228static bool bitmap_enabled(void *data, bool flush)
 229{
 230	struct bitmap *bitmap = data;
 231
 232	if (!flush)
 233		return true;
 234
 235	/*
 236	 * If caller want to flush bitmap pages to underlying disks, check if
 237	 * there are cached pages in filemap.
 238	 */
 239	return !test_bit(BITMAP_STALE, &bitmap->flags) &&
 240	       bitmap->storage.filemap != NULL;
 241}
 242
 243/*
 244 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
 245 *
 246 * 1) check to see if this page is allocated, if it's not then try to alloc
 247 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
 248 *    page pointer directly as a counter
 249 *
 250 * if we find our page, we increment the page's refcount so that it stays
 251 * allocated while we're using it
 252 */
 253static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
 254			       unsigned long page, int create, int no_hijack)
 255__releases(bitmap->lock)
 256__acquires(bitmap->lock)
 257{
 258	unsigned char *mappage;
 259
 260	WARN_ON_ONCE(page >= bitmap->pages);
 261	if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
 262		return 0;
 263
 264	if (bitmap->bp[page].map) /* page is already allocated, just return */
 265		return 0;
 266
 267	if (!create)
 268		return -ENOENT;
 269
 270	/* this page has not been allocated yet */
 271
 272	spin_unlock_irq(&bitmap->lock);
 273	/* It is possible that this is being called inside a
 274	 * prepare_to_wait/finish_wait loop from raid5c:make_request().
 275	 * In general it is not permitted to sleep in that context as it
 276	 * can cause the loop to spin freely.
 277	 * That doesn't apply here as we can only reach this point
 278	 * once with any loop.
 279	 * When this function completes, either bp[page].map or
 280	 * bp[page].hijacked.  In either case, this function will
 281	 * abort before getting to this point again.  So there is
 282	 * no risk of a free-spin, and so it is safe to assert
 283	 * that sleeping here is allowed.
 284	 */
 285	sched_annotate_sleep();
 286	mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
 287	spin_lock_irq(&bitmap->lock);
 288
 289	if (mappage == NULL) {
 290		pr_debug("md/bitmap: map page allocation failed, hijacking\n");
 291		/* We don't support hijack for cluster raid */
 292		if (no_hijack)
 293			return -ENOMEM;
 294		/* failed - set the hijacked flag so that we can use the
 295		 * pointer as a counter */
 296		if (!bitmap->bp[page].map)
 297			bitmap->bp[page].hijacked = 1;
 298	} else if (bitmap->bp[page].map ||
 299		   bitmap->bp[page].hijacked) {
 300		/* somebody beat us to getting the page */
 301		kfree(mappage);
 302	} else {
 303
 304		/* no page was in place and we have one, so install it */
 305
 306		bitmap->bp[page].map = mappage;
 307		bitmap->missing_pages--;
 308	}
 309	return 0;
 310}
 311
 312/* if page is completely empty, put it back on the free list, or dealloc it */
 313/* if page was hijacked, unmark the flag so it might get alloced next time */
 314/* Note: lock should be held when calling this */
 315static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 316{
 317	char *ptr;
 318
 319	if (bitmap->bp[page].count) /* page is still busy */
 320		return;
 321
 322	/* page is no longer in use, it can be released */
 323
 324	if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 325		bitmap->bp[page].hijacked = 0;
 326		bitmap->bp[page].map = NULL;
 327	} else {
 328		/* normal case, free the page */
 329		ptr = bitmap->bp[page].map;
 330		bitmap->bp[page].map = NULL;
 331		bitmap->missing_pages++;
 332		kfree(ptr);
 333	}
 334}
 335
 336/*
 337 * bitmap file handling - read and write the bitmap file and its superblock
 338 */
 339
 340/*
 341 * basic page I/O operations
 342 */
 343
 344/* IO operations when bitmap is stored near all superblocks */
 345
 346/* choose a good rdev and read the page from there */
 347static int read_sb_page(struct mddev *mddev, loff_t offset,
 348		struct page *page, unsigned long index, int size)
 349{
 350
 351	sector_t sector = mddev->bitmap_info.offset + offset +
 352		index * (PAGE_SIZE / SECTOR_SIZE);
 353	struct md_rdev *rdev;
 354
 355	rdev_for_each(rdev, mddev) {
 356		u32 iosize = roundup(size, bdev_logical_block_size(rdev->bdev));
 357
 358		if (!test_bit(In_sync, &rdev->flags) ||
 359		    test_bit(Faulty, &rdev->flags) ||
 360		    test_bit(Bitmap_sync, &rdev->flags))
 361			continue;
 362
 363		if (sync_page_io(rdev, sector, iosize, page, REQ_OP_READ, true))
 364			return 0;
 365	}
 366	return -EIO;
 367}
 368
 369static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 370{
 371	/* Iterate the disks of an mddev, using rcu to protect access to the
 372	 * linked list, and raising the refcount of devices we return to ensure
 373	 * they don't disappear while in use.
 374	 * As devices are only added or removed when raid_disk is < 0 and
 375	 * nr_pending is 0 and In_sync is clear, the entries we return will
 376	 * still be in the same position on the list when we re-enter
 377	 * list_for_each_entry_continue_rcu.
 378	 *
 379	 * Note that if entered with 'rdev == NULL' to start at the
 380	 * beginning, we temporarily assign 'rdev' to an address which
 381	 * isn't really an rdev, but which can be used by
 382	 * list_for_each_entry_continue_rcu() to find the first entry.
 383	 */
 384	rcu_read_lock();
 385	if (rdev == NULL)
 386		/* start at the beginning */
 387		rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 388	else {
 389		/* release the previous rdev and start from there. */
 390		rdev_dec_pending(rdev, mddev);
 391	}
 392	list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 393		if (rdev->raid_disk >= 0 &&
 394		    !test_bit(Faulty, &rdev->flags)) {
 395			/* this is a usable devices */
 396			atomic_inc(&rdev->nr_pending);
 397			rcu_read_unlock();
 398			return rdev;
 399		}
 400	}
 401	rcu_read_unlock();
 402	return NULL;
 403}
 404
 405static unsigned int optimal_io_size(struct block_device *bdev,
 406				    unsigned int last_page_size,
 407				    unsigned int io_size)
 408{
 409	if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
 410		return roundup(last_page_size, bdev_io_opt(bdev));
 411	return io_size;
 412}
 413
 414static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
 415				   loff_t start, loff_t boundary)
 416{
 417	if (io_size != opt_size &&
 418	    start + opt_size / SECTOR_SIZE <= boundary)
 419		return opt_size;
 420	if (start + io_size / SECTOR_SIZE <= boundary)
 421		return io_size;
 422
 423	/* Overflows boundary */
 424	return 0;
 425}
 426
 427static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 428			   unsigned long pg_index, struct page *page)
 429{
 430	struct block_device *bdev;
 431	struct mddev *mddev = bitmap->mddev;
 432	struct bitmap_storage *store = &bitmap->storage;
 433	unsigned long num_pages = bitmap->storage.file_pages;
 434	unsigned int bitmap_limit = (num_pages - pg_index % num_pages) << PAGE_SHIFT;
 435	loff_t sboff, offset = mddev->bitmap_info.offset;
 436	sector_t ps = pg_index * PAGE_SIZE / SECTOR_SIZE;
 437	unsigned int size = PAGE_SIZE;
 438	unsigned int opt_size = PAGE_SIZE;
 439	sector_t doff;
 440
 441	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 442	/* we compare length (page numbers), not page offset. */
 443	if ((pg_index - store->sb_index) == num_pages - 1) {
 444		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 445
 446		if (last_page_size == 0)
 447			last_page_size = PAGE_SIZE;
 448		size = roundup(last_page_size, bdev_logical_block_size(bdev));
 449		opt_size = optimal_io_size(bdev, last_page_size, size);
 450	}
 451
 452	sboff = rdev->sb_start + offset;
 453	doff = rdev->data_offset;
 454
 455	/* Just make sure we aren't corrupting data or metadata */
 456	if (mddev->external) {
 457		/* Bitmap could be anywhere. */
 458		if (sboff + ps > doff &&
 459		    sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE))
 460			return -EINVAL;
 461	} else if (offset < 0) {
 462		/* DATA  BITMAP METADATA  */
 463		size = bitmap_io_size(size, opt_size, offset + ps, 0);
 464		if (size == 0)
 465			/* bitmap runs in to metadata */
 466			return -EINVAL;
 467
 468		if (doff + mddev->dev_sectors > sboff)
 469			/* data runs in to bitmap */
 470			return -EINVAL;
 471	} else if (rdev->sb_start < rdev->data_offset) {
 472		/* METADATA BITMAP DATA */
 473		size = bitmap_io_size(size, opt_size, sboff + ps, doff);
 474		if (size == 0)
 475			/* bitmap runs in to data */
 476			return -EINVAL;
 477	}
 478
 479	md_write_metadata(mddev, rdev, sboff + ps, (int)min(size, bitmap_limit),
 480			  page, 0);
 481	return 0;
 482}
 483
 484static void write_sb_page(struct bitmap *bitmap, unsigned long pg_index,
 485			  struct page *page, bool wait)
 486{
 487	struct mddev *mddev = bitmap->mddev;
 488
 489	do {
 490		struct md_rdev *rdev = NULL;
 491
 492		while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 493			if (__write_sb_page(rdev, bitmap, pg_index, page) < 0) {
 494				set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 495				return;
 496			}
 497		}
 498	} while (wait && md_super_wait(mddev) < 0);
 499}
 500
 501static void md_bitmap_file_kick(struct bitmap *bitmap);
 502
 503#ifdef CONFIG_MD_BITMAP_FILE
 504static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 505{
 506	struct buffer_head *bh = page_buffers(page);
 507
 508	while (bh && bh->b_blocknr) {
 509		atomic_inc(&bitmap->pending_writes);
 510		set_buffer_locked(bh);
 511		set_buffer_mapped(bh);
 512		submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
 513		bh = bh->b_this_page;
 514	}
 515
 516	if (wait)
 517		wait_event(bitmap->write_wait,
 518			   atomic_read(&bitmap->pending_writes) == 0);
 519}
 520
 521static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 522{
 523	struct bitmap *bitmap = bh->b_private;
 524
 525	if (!uptodate)
 526		set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 527	if (atomic_dec_and_test(&bitmap->pending_writes))
 528		wake_up(&bitmap->write_wait);
 529}
 530
 531static void free_buffers(struct page *page)
 532{
 533	struct buffer_head *bh;
 534
 535	if (!PagePrivate(page))
 536		return;
 537
 538	bh = page_buffers(page);
 539	while (bh) {
 540		struct buffer_head *next = bh->b_this_page;
 541		free_buffer_head(bh);
 542		bh = next;
 543	}
 544	detach_page_private(page);
 545	put_page(page);
 546}
 547
 548/* read a page from a file.
 549 * We both read the page, and attach buffers to the page to record the
 550 * address of each block (using bmap).  These addresses will be used
 551 * to write the block later, completely bypassing the filesystem.
 552 * This usage is similar to how swap files are handled, and allows us
 553 * to write to a file with no concerns of memory allocation failing.
 554 */
 555static int read_file_page(struct file *file, unsigned long index,
 556		struct bitmap *bitmap, unsigned long count, struct page *page)
 557{
 558	int ret = 0;
 559	struct inode *inode = file_inode(file);
 560	struct buffer_head *bh;
 561	sector_t block, blk_cur;
 562	unsigned long blocksize = i_blocksize(inode);
 563
 564	pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 565		 (unsigned long long)index << PAGE_SHIFT);
 566
 567	bh = alloc_page_buffers(page, blocksize);
 568	if (!bh) {
 569		ret = -ENOMEM;
 570		goto out;
 571	}
 572	attach_page_private(page, bh);
 573	blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 574	while (bh) {
 575		block = blk_cur;
 576
 577		if (count == 0)
 578			bh->b_blocknr = 0;
 579		else {
 580			ret = bmap(inode, &block);
 581			if (ret || !block) {
 582				ret = -EINVAL;
 583				bh->b_blocknr = 0;
 584				goto out;
 585			}
 586
 587			bh->b_blocknr = block;
 588			bh->b_bdev = inode->i_sb->s_bdev;
 589			if (count < blocksize)
 590				count = 0;
 591			else
 592				count -= blocksize;
 593
 594			bh->b_end_io = end_bitmap_write;
 595			bh->b_private = bitmap;
 596			atomic_inc(&bitmap->pending_writes);
 597			set_buffer_locked(bh);
 598			set_buffer_mapped(bh);
 599			submit_bh(REQ_OP_READ, bh);
 600		}
 601		blk_cur++;
 602		bh = bh->b_this_page;
 603	}
 604
 605	wait_event(bitmap->write_wait,
 606		   atomic_read(&bitmap->pending_writes)==0);
 607	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 608		ret = -EIO;
 609out:
 610	if (ret)
 611		pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 612		       (int)PAGE_SIZE,
 613		       (unsigned long long)index << PAGE_SHIFT,
 614		       ret);
 615	return ret;
 616}
 617#else /* CONFIG_MD_BITMAP_FILE */
 618static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 619{
 620}
 621static int read_file_page(struct file *file, unsigned long index,
 622		struct bitmap *bitmap, unsigned long count, struct page *page)
 623{
 624	return -EIO;
 625}
 626static void free_buffers(struct page *page)
 627{
 628	put_page(page);
 629}
 630#endif /* CONFIG_MD_BITMAP_FILE */
 631
 632/*
 633 * bitmap file superblock operations
 634 */
 635
 636/*
 637 * write out a page to a file
 638 */
 639static void filemap_write_page(struct bitmap *bitmap, unsigned long pg_index,
 640			       bool wait)
 641{
 642	struct bitmap_storage *store = &bitmap->storage;
 643	struct page *page = store->filemap[pg_index];
 644
 645	if (mddev_is_clustered(bitmap->mddev)) {
 646		/* go to node bitmap area starting point */
 647		pg_index += store->sb_index;
 648	}
 649
 650	if (store->file)
 651		write_file_page(bitmap, page, wait);
 652	else
 653		write_sb_page(bitmap, pg_index, page, wait);
 654}
 655
 656/*
 657 * md_bitmap_wait_writes() should be called before writing any bitmap
 658 * blocks, to ensure previous writes, particularly from
 659 * md_bitmap_daemon_work(), have completed.
 660 */
 661static void md_bitmap_wait_writes(struct bitmap *bitmap)
 662{
 663	if (bitmap->storage.file)
 664		wait_event(bitmap->write_wait,
 665			   atomic_read(&bitmap->pending_writes)==0);
 666	else
 667		/* Note that we ignore the return value.  The writes
 668		 * might have failed, but that would just mean that
 669		 * some bits which should be cleared haven't been,
 670		 * which is safe.  The relevant bitmap blocks will
 671		 * probably get written again, but there is no great
 672		 * loss if they aren't.
 673		 */
 674		md_super_wait(bitmap->mddev);
 675}
 676
 677
 678/* update the event counter and sync the superblock to disk */
 679static void bitmap_update_sb(void *data)
 680{
 681	bitmap_super_t *sb;
 682	struct bitmap *bitmap = data;
 683
 684	if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 685		return;
 686	if (bitmap->mddev->bitmap_info.external)
 687		return;
 688	if (!bitmap->storage.sb_page) /* no superblock */
 689		return;
 690	sb = kmap_local_page(bitmap->storage.sb_page);
 691	sb->events = cpu_to_le64(bitmap->mddev->events);
 692	if (bitmap->mddev->events < bitmap->events_cleared)
 693		/* rocking back to read-only */
 694		bitmap->events_cleared = bitmap->mddev->events;
 695	sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 696	/*
 697	 * clear BITMAP_WRITE_ERROR bit to protect against the case that
 698	 * a bitmap write error occurred but the later writes succeeded.
 699	 */
 700	sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 701	/* Just in case these have been changed via sysfs: */
 702	sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 703	sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 704	/* This might have been changed by a reshape */
 705	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 706	sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 707	sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 708	sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 709					   bitmap_info.space);
 710	kunmap_local(sb);
 711
 712	if (bitmap->storage.file)
 713		write_file_page(bitmap, bitmap->storage.sb_page, 1);
 714	else
 715		write_sb_page(bitmap, bitmap->storage.sb_index,
 716			      bitmap->storage.sb_page, 1);
 717}
 718
 719static void bitmap_print_sb(struct bitmap *bitmap)
 720{
 721	bitmap_super_t *sb;
 722
 723	if (!bitmap || !bitmap->storage.sb_page)
 724		return;
 725	sb = kmap_local_page(bitmap->storage.sb_page);
 726	pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 727	pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 728	pr_debug("       version: %u\n", le32_to_cpu(sb->version));
 729	pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 730		 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 731		 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 732		 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 733		 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 734	pr_debug("        events: %llu\n",
 735		 (unsigned long long) le64_to_cpu(sb->events));
 736	pr_debug("events cleared: %llu\n",
 737		 (unsigned long long) le64_to_cpu(sb->events_cleared));
 738	pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 739	pr_debug("     chunksize: %u B\n", le32_to_cpu(sb->chunksize));
 740	pr_debug("  daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
 741	pr_debug("     sync size: %llu KB\n",
 742		 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 743	pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
 744	kunmap_local(sb);
 745}
 746
 747/*
 748 * bitmap_new_disk_sb
 749 * @bitmap
 750 *
 751 * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 752 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 753 * This function verifies 'bitmap_info' and populates the on-disk bitmap
 754 * structure, which is to be written to disk.
 755 *
 756 * Returns: 0 on success, -Exxx on error
 757 */
 758static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 759{
 760	bitmap_super_t *sb;
 761	unsigned long chunksize, daemon_sleep, write_behind;
 762
 763	bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 764	if (bitmap->storage.sb_page == NULL)
 765		return -ENOMEM;
 766	bitmap->storage.sb_index = 0;
 767
 768	sb = kmap_local_page(bitmap->storage.sb_page);
 769
 770	sb->magic = cpu_to_le32(BITMAP_MAGIC);
 771	sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 772
 773	chunksize = bitmap->mddev->bitmap_info.chunksize;
 774	BUG_ON(!chunksize);
 775	if (!is_power_of_2(chunksize)) {
 776		kunmap_local(sb);
 777		pr_warn("bitmap chunksize not a power of 2\n");
 778		return -EINVAL;
 779	}
 780	sb->chunksize = cpu_to_le32(chunksize);
 781
 782	daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 783	if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 784		pr_debug("Choosing daemon_sleep default (5 sec)\n");
 785		daemon_sleep = 5 * HZ;
 786	}
 787	sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 788	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 789
 790	/*
 791	 * FIXME: write_behind for RAID1.  If not specified, what
 792	 * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 793	 */
 794	write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 795	if (write_behind > COUNTER_MAX / 2)
 796		write_behind = COUNTER_MAX / 2;
 797	sb->write_behind = cpu_to_le32(write_behind);
 798	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 799
 800	/* keep the array size field of the bitmap superblock up to date */
 801	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 802
 803	memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 804
 805	set_bit(BITMAP_STALE, &bitmap->flags);
 806	sb->state = cpu_to_le32(bitmap->flags);
 807	bitmap->events_cleared = bitmap->mddev->events;
 808	sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 809	bitmap->mddev->bitmap_info.nodes = 0;
 810
 811	kunmap_local(sb);
 812
 813	return 0;
 814}
 815
 816/* read the superblock from the bitmap file and initialize some bitmap fields */
 817static int md_bitmap_read_sb(struct bitmap *bitmap)
 818{
 819	char *reason = NULL;
 820	bitmap_super_t *sb;
 821	unsigned long chunksize, daemon_sleep, write_behind;
 822	unsigned long long events;
 823	int nodes = 0;
 824	unsigned long sectors_reserved = 0;
 825	int err = -EINVAL;
 826	struct page *sb_page;
 827	loff_t offset = 0;
 828
 829	if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 830		chunksize = 128 * 1024 * 1024;
 831		daemon_sleep = 5 * HZ;
 832		write_behind = 0;
 833		set_bit(BITMAP_STALE, &bitmap->flags);
 834		err = 0;
 835		goto out_no_sb;
 836	}
 837	/* page 0 is the superblock, read it... */
 838	sb_page = alloc_page(GFP_KERNEL);
 839	if (!sb_page)
 840		return -ENOMEM;
 841	bitmap->storage.sb_page = sb_page;
 842
 843re_read:
 844	/* If cluster_slot is set, the cluster is setup */
 845	if (bitmap->cluster_slot >= 0) {
 846		sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 847
 848		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 849			   (bitmap->mddev->bitmap_info.chunksize >> 9));
 850		/* bits to bytes */
 851		bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 852		/* to 4k blocks */
 853		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 854		offset = bitmap->cluster_slot * (bm_blocks << 3);
 855		pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 856			bitmap->cluster_slot, offset);
 857	}
 858
 859	if (bitmap->storage.file) {
 860		loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 861		int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 862
 863		err = read_file_page(bitmap->storage.file, 0,
 864				bitmap, bytes, sb_page);
 865	} else {
 866		err = read_sb_page(bitmap->mddev, offset, sb_page, 0,
 867				   sizeof(bitmap_super_t));
 868	}
 869	if (err)
 870		return err;
 871
 872	err = -EINVAL;
 873	sb = kmap_local_page(sb_page);
 874
 875	chunksize = le32_to_cpu(sb->chunksize);
 876	daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 877	write_behind = le32_to_cpu(sb->write_behind);
 878	sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 879
 880	/* verify that the bitmap-specific fields are valid */
 881	if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 882		reason = "bad magic";
 883	else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 884		 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 885		reason = "unrecognized superblock version";
 886	else if (chunksize < 512)
 887		reason = "bitmap chunksize too small";
 888	else if (!is_power_of_2(chunksize))
 889		reason = "bitmap chunksize not a power of 2";
 890	else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 891		reason = "daemon sleep period out of range";
 892	else if (write_behind > COUNTER_MAX)
 893		reason = "write-behind limit out of range (0 - 16383)";
 894	if (reason) {
 895		pr_warn("%s: invalid bitmap file superblock: %s\n",
 896			bmname(bitmap), reason);
 897		goto out;
 898	}
 899
 900	/*
 901	 * Setup nodes/clustername only if bitmap version is
 902	 * cluster-compatible
 903	 */
 904	if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 905		nodes = le32_to_cpu(sb->nodes);
 906		strscpy(bitmap->mddev->bitmap_info.cluster_name,
 907				sb->cluster_name, 64);
 908	}
 909
 910	/* keep the array size field of the bitmap superblock up to date */
 911	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 912
 913	if (bitmap->mddev->persistent) {
 914		/*
 915		 * We have a persistent array superblock, so compare the
 916		 * bitmap's UUID and event counter to the mddev's
 917		 */
 918		if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 919			pr_warn("%s: bitmap superblock UUID mismatch\n",
 920				bmname(bitmap));
 921			goto out;
 922		}
 923		events = le64_to_cpu(sb->events);
 924		if (!nodes && (events < bitmap->mddev->events)) {
 925			pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 926				bmname(bitmap), events,
 927				(unsigned long long) bitmap->mddev->events);
 928			set_bit(BITMAP_STALE, &bitmap->flags);
 929		}
 930	}
 931
 932	/* assign fields using values from superblock */
 933	bitmap->flags |= le32_to_cpu(sb->state);
 934	if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 935		set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 936	bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 937	err = 0;
 938
 939out:
 940	kunmap_local(sb);
 941	if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 942		/* Assigning chunksize is required for "re_read" */
 943		bitmap->mddev->bitmap_info.chunksize = chunksize;
 944		err = md_setup_cluster(bitmap->mddev, nodes);
 945		if (err) {
 946			pr_warn("%s: Could not setup cluster service (%d)\n",
 947				bmname(bitmap), err);
 948			goto out_no_sb;
 949		}
 950		bitmap->cluster_slot = bitmap->mddev->cluster_ops->slot_number(bitmap->mddev);
 951		goto re_read;
 952	}
 953
 954out_no_sb:
 955	if (err == 0) {
 956		if (test_bit(BITMAP_STALE, &bitmap->flags))
 957			bitmap->events_cleared = bitmap->mddev->events;
 958		bitmap->mddev->bitmap_info.chunksize = chunksize;
 959		bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 960		bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 961		bitmap->mddev->bitmap_info.nodes = nodes;
 962		if (bitmap->mddev->bitmap_info.space == 0 ||
 963			bitmap->mddev->bitmap_info.space > sectors_reserved)
 964			bitmap->mddev->bitmap_info.space = sectors_reserved;
 965	} else {
 966		bitmap_print_sb(bitmap);
 967		if (bitmap->cluster_slot < 0)
 968			md_cluster_stop(bitmap->mddev);
 969	}
 970	return err;
 971}
 972
 973/*
 974 * general bitmap file operations
 975 */
 976
 977/*
 978 * on-disk bitmap:
 979 *
 980 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 981 * file a page at a time. There's a superblock at the start of the file.
 982 */
 983/* calculate the index of the page that contains this bit */
 984static inline unsigned long file_page_index(struct bitmap_storage *store,
 985					    unsigned long chunk)
 986{
 987	if (store->sb_page)
 988		chunk += sizeof(bitmap_super_t) << 3;
 989	return chunk >> PAGE_BIT_SHIFT;
 990}
 991
 992/* calculate the (bit) offset of this bit within a page */
 993static inline unsigned long file_page_offset(struct bitmap_storage *store,
 994					     unsigned long chunk)
 995{
 996	if (store->sb_page)
 997		chunk += sizeof(bitmap_super_t) << 3;
 998	return chunk & (PAGE_BITS - 1);
 999}
1000
1001/*
1002 * return a pointer to the page in the filemap that contains the given bit
1003 *
1004 */
1005static inline struct page *filemap_get_page(struct bitmap_storage *store,
1006					    unsigned long chunk)
1007{
1008	if (file_page_index(store, chunk) >= store->file_pages)
1009		return NULL;
1010	return store->filemap[file_page_index(store, chunk)];
1011}
1012
1013static int md_bitmap_storage_alloc(struct bitmap_storage *store,
1014				   unsigned long chunks, int with_super,
1015				   int slot_number)
1016{
1017	int pnum, offset = 0;
1018	unsigned long num_pages;
1019	unsigned long bytes;
1020
1021	bytes = DIV_ROUND_UP(chunks, 8);
1022	if (with_super)
1023		bytes += sizeof(bitmap_super_t);
1024
1025	num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
1026	offset = slot_number * num_pages;
1027
1028	store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
1029				       GFP_KERNEL);
1030	if (!store->filemap)
1031		return -ENOMEM;
1032
1033	if (with_super && !store->sb_page) {
1034		store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
1035		if (store->sb_page == NULL)
1036			return -ENOMEM;
1037	}
1038
1039	pnum = 0;
1040	if (store->sb_page) {
1041		store->filemap[0] = store->sb_page;
1042		pnum = 1;
1043		store->sb_index = offset;
1044	}
1045
1046	for ( ; pnum < num_pages; pnum++) {
1047		store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
1048		if (!store->filemap[pnum]) {
1049			store->file_pages = pnum;
1050			return -ENOMEM;
1051		}
1052	}
1053	store->file_pages = pnum;
1054
1055	/* We need 4 bits per page, rounded up to a multiple
1056	 * of sizeof(unsigned long) */
1057	store->filemap_attr = kzalloc(
1058		roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
1059		GFP_KERNEL);
1060	if (!store->filemap_attr)
1061		return -ENOMEM;
1062
1063	store->bytes = bytes;
1064
1065	return 0;
1066}
1067
1068static void md_bitmap_file_unmap(struct bitmap_storage *store)
1069{
1070	struct file *file = store->file;
1071	struct page *sb_page = store->sb_page;
1072	struct page **map = store->filemap;
1073	int pages = store->file_pages;
1074
1075	while (pages--)
1076		if (map[pages] != sb_page) /* 0 is sb_page, release it below */
1077			free_buffers(map[pages]);
1078	kfree(map);
1079	kfree(store->filemap_attr);
1080
1081	if (sb_page)
1082		free_buffers(sb_page);
1083
1084	if (file) {
1085		struct inode *inode = file_inode(file);
1086		invalidate_mapping_pages(inode->i_mapping, 0, -1);
1087		fput(file);
1088	}
1089}
1090
1091/*
1092 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
1093 * then it is no longer reliable, so we stop using it and we mark the file
1094 * as failed in the superblock
1095 */
1096static void md_bitmap_file_kick(struct bitmap *bitmap)
1097{
1098	if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
1099		bitmap_update_sb(bitmap);
1100
1101		if (bitmap->storage.file) {
1102			pr_warn("%s: kicking failed bitmap file %pD4 from array!\n",
1103				bmname(bitmap), bitmap->storage.file);
1104
1105		} else
1106			pr_warn("%s: disabling internal bitmap due to errors\n",
1107				bmname(bitmap));
1108	}
1109}
1110
1111enum bitmap_page_attr {
1112	BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
1113	BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
1114				    * i.e. counter is 1 or 2. */
1115	BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
1116};
1117
1118static inline void set_page_attr(struct bitmap *bitmap, int pnum,
1119				 enum bitmap_page_attr attr)
1120{
1121	set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1122}
1123
1124static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
1125				   enum bitmap_page_attr attr)
1126{
1127	clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1128}
1129
1130static inline int test_page_attr(struct bitmap *bitmap, int pnum,
1131				 enum bitmap_page_attr attr)
1132{
1133	return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
1134}
1135
1136static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
1137					   enum bitmap_page_attr attr)
1138{
1139	return test_and_clear_bit((pnum<<2) + attr,
1140				  bitmap->storage.filemap_attr);
1141}
1142/*
1143 * bitmap_file_set_bit -- called before performing a write to the md device
1144 * to set (and eventually sync) a particular bit in the bitmap file
1145 *
1146 * we set the bit immediately, then we record the page number so that
1147 * when an unplug occurs, we can flush the dirty pages out to disk
1148 */
1149static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
1150{
1151	unsigned long bit;
1152	struct page *page;
1153	void *kaddr;
1154	unsigned long chunk = block >> bitmap->counts.chunkshift;
1155	struct bitmap_storage *store = &bitmap->storage;
1156	unsigned long index = file_page_index(store, chunk);
1157	unsigned long node_offset = 0;
1158
1159	index += store->sb_index;
1160	if (mddev_is_clustered(bitmap->mddev))
1161		node_offset = bitmap->cluster_slot * store->file_pages;
1162
1163	page = filemap_get_page(&bitmap->storage, chunk);
1164	if (!page)
1165		return;
1166	bit = file_page_offset(&bitmap->storage, chunk);
1167
1168	/* set the bit */
1169	kaddr = kmap_local_page(page);
1170	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1171		set_bit(bit, kaddr);
1172	else
1173		set_bit_le(bit, kaddr);
1174	kunmap_local(kaddr);
1175	pr_debug("set file bit %lu page %lu\n", bit, index);
1176	/* record page number so it gets flushed to disk when unplug occurs */
1177	set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_DIRTY);
1178}
1179
1180static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
1181{
1182	unsigned long bit;
1183	struct page *page;
1184	void *paddr;
1185	unsigned long chunk = block >> bitmap->counts.chunkshift;
1186	struct bitmap_storage *store = &bitmap->storage;
1187	unsigned long index = file_page_index(store, chunk);
1188	unsigned long node_offset = 0;
1189
1190	index += store->sb_index;
1191	if (mddev_is_clustered(bitmap->mddev))
1192		node_offset = bitmap->cluster_slot * store->file_pages;
1193
1194	page = filemap_get_page(&bitmap->storage, chunk);
1195	if (!page)
1196		return;
1197	bit = file_page_offset(&bitmap->storage, chunk);
1198	paddr = kmap_local_page(page);
1199	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1200		clear_bit(bit, paddr);
1201	else
1202		clear_bit_le(bit, paddr);
1203	kunmap_local(paddr);
1204	if (!test_page_attr(bitmap, index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
1205		set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_PENDING);
1206		bitmap->allclean = 0;
1207	}
1208}
1209
1210static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
1211{
1212	unsigned long bit;
1213	struct page *page;
1214	void *paddr;
1215	unsigned long chunk = block >> bitmap->counts.chunkshift;
1216	int set = 0;
1217
1218	page = filemap_get_page(&bitmap->storage, chunk);
1219	if (!page)
1220		return -EINVAL;
1221	bit = file_page_offset(&bitmap->storage, chunk);
1222	paddr = kmap_local_page(page);
1223	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1224		set = test_bit(bit, paddr);
1225	else
1226		set = test_bit_le(bit, paddr);
1227	kunmap_local(paddr);
1228	return set;
1229}
1230
1231/* this gets called when the md device is ready to unplug its underlying
1232 * (slave) device queues -- before we let any writes go down, we need to
1233 * sync the dirty pages of the bitmap file to disk */
1234static void __bitmap_unplug(struct bitmap *bitmap)
1235{
1236	unsigned long i;
1237	int dirty, need_write;
1238	int writing = 0;
1239
1240	if (!bitmap_enabled(bitmap, true))
1241		return;
1242
1243	/* look at each page to see if there are any set bits that need to be
1244	 * flushed out to disk */
1245	for (i = 0; i < bitmap->storage.file_pages; i++) {
1246		dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1247		need_write = test_and_clear_page_attr(bitmap, i,
1248						      BITMAP_PAGE_NEEDWRITE);
1249		if (dirty || need_write) {
1250			if (!writing) {
1251				md_bitmap_wait_writes(bitmap);
1252				mddev_add_trace_msg(bitmap->mddev,
1253					"md bitmap_unplug");
1254			}
1255			clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1256			filemap_write_page(bitmap, i, false);
1257			writing = 1;
1258		}
1259	}
1260	if (writing)
1261		md_bitmap_wait_writes(bitmap);
1262
1263	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1264		md_bitmap_file_kick(bitmap);
1265}
1266
1267struct bitmap_unplug_work {
1268	struct work_struct work;
1269	struct bitmap *bitmap;
1270	struct completion *done;
1271};
1272
1273static void md_bitmap_unplug_fn(struct work_struct *work)
1274{
1275	struct bitmap_unplug_work *unplug_work =
1276		container_of(work, struct bitmap_unplug_work, work);
1277
1278	__bitmap_unplug(unplug_work->bitmap);
1279	complete(unplug_work->done);
1280}
1281
1282static void bitmap_unplug_async(struct bitmap *bitmap)
1283{
1284	DECLARE_COMPLETION_ONSTACK(done);
1285	struct bitmap_unplug_work unplug_work;
1286
1287	INIT_WORK_ONSTACK(&unplug_work.work, md_bitmap_unplug_fn);
1288	unplug_work.bitmap = bitmap;
1289	unplug_work.done = &done;
1290
1291	queue_work(md_bitmap_wq, &unplug_work.work);
1292	wait_for_completion(&done);
1293	destroy_work_on_stack(&unplug_work.work);
1294}
1295
1296static void bitmap_unplug(struct mddev *mddev, bool sync)
1297{
1298	struct bitmap *bitmap = mddev->bitmap;
1299
1300	if (!bitmap)
1301		return;
1302
1303	if (sync)
1304		__bitmap_unplug(bitmap);
1305	else
1306		bitmap_unplug_async(bitmap);
1307}
1308
1309static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1310
1311/*
1312 * Initialize the in-memory bitmap from the on-disk bitmap and set up the memory
1313 * mapping of the bitmap file.
1314 *
1315 * Special case: If there's no bitmap file, or if the bitmap file had been
1316 * previously kicked from the array, we mark all the bits as 1's in order to
1317 * cause a full resync.
1318 *
1319 * We ignore all bits for sectors that end earlier than 'start'.
1320 * This is used when reading an out-of-date bitmap.
1321 */
1322static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1323{
1324	bool outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1325	struct mddev *mddev = bitmap->mddev;
1326	unsigned long chunks = bitmap->counts.chunks;
1327	struct bitmap_storage *store = &bitmap->storage;
1328	struct file *file = store->file;
1329	unsigned long node_offset = 0;
1330	unsigned long bit_cnt = 0;
1331	unsigned long i;
1332	int ret;
1333
1334	if (!file && !mddev->bitmap_info.offset) {
1335		/* No permanent bitmap - fill with '1s'. */
1336		store->filemap = NULL;
1337		store->file_pages = 0;
1338		for (i = 0; i < chunks ; i++) {
1339			/* if the disk bit is set, set the memory bit */
1340			int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1341				      >= start);
1342			md_bitmap_set_memory_bits(bitmap,
1343						  (sector_t)i << bitmap->counts.chunkshift,
1344						  needed);
1345		}
1346		return 0;
1347	}
1348
1349	if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1350		pr_warn("%s: bitmap file too short %lu < %lu\n",
1351			bmname(bitmap),
1352			(unsigned long) i_size_read(file->f_mapping->host),
1353			store->bytes);
1354		ret = -ENOSPC;
1355		goto err;
1356	}
1357
1358	if (mddev_is_clustered(mddev))
1359		node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1360
1361	for (i = 0; i < store->file_pages; i++) {
1362		struct page *page = store->filemap[i];
1363		int count;
1364
1365		/* unmap the old page, we're done with it */
1366		if (i == store->file_pages - 1)
1367			count = store->bytes - i * PAGE_SIZE;
1368		else
1369			count = PAGE_SIZE;
1370
1371		if (file)
1372			ret = read_file_page(file, i, bitmap, count, page);
1373		else
1374			ret = read_sb_page(mddev, 0, page, i + node_offset,
1375					   count);
1376		if (ret)
1377			goto err;
1378	}
1379
1380	if (outofdate) {
1381		pr_warn("%s: bitmap file is out of date, doing full recovery\n",
1382			bmname(bitmap));
1383
1384		for (i = 0; i < store->file_pages; i++) {
1385			struct page *page = store->filemap[i];
1386			unsigned long offset = 0;
1387			void *paddr;
1388
1389			if (i == 0 && !mddev->bitmap_info.external)
1390				offset = sizeof(bitmap_super_t);
1391
1392			/*
1393			 * If the bitmap is out of date, dirty the whole page
1394			 * and write it out
1395			 */
1396			paddr = kmap_local_page(page);
1397			memset(paddr + offset, 0xff, PAGE_SIZE - offset);
1398			kunmap_local(paddr);
1399
1400			filemap_write_page(bitmap, i, true);
1401			if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) {
1402				ret = -EIO;
1403				goto err;
1404			}
1405		}
1406	}
1407
1408	for (i = 0; i < chunks; i++) {
1409		struct page *page = filemap_get_page(&bitmap->storage, i);
1410		unsigned long bit = file_page_offset(&bitmap->storage, i);
1411		void *paddr;
1412		bool was_set;
1413
1414		paddr = kmap_local_page(page);
1415		if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1416			was_set = test_bit(bit, paddr);
1417		else
1418			was_set = test_bit_le(bit, paddr);
1419		kunmap_local(paddr);
1420
1421		if (was_set) {
1422			/* if the disk bit is set, set the memory bit */
1423			int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1424				      >= start);
1425			md_bitmap_set_memory_bits(bitmap,
1426						  (sector_t)i << bitmap->counts.chunkshift,
1427						  needed);
1428			bit_cnt++;
1429		}
1430	}
1431
1432	pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1433		 bmname(bitmap), store->file_pages,
1434		 bit_cnt, chunks);
1435
1436	return 0;
1437
1438 err:
1439	pr_warn("%s: bitmap initialisation failed: %d\n",
1440		bmname(bitmap), ret);
1441	return ret;
1442}
1443
1444/* just flag bitmap pages as needing to be written. */
1445static void bitmap_write_all(struct mddev *mddev)
1446{
1447	int i;
1448	struct bitmap *bitmap = mddev->bitmap;
1449
1450	if (!bitmap || !bitmap->storage.filemap)
1451		return;
1452
1453	/* Only one copy, so nothing needed */
1454	if (bitmap->storage.file)
1455		return;
1456
1457	for (i = 0; i < bitmap->storage.file_pages; i++)
1458		set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
1459	bitmap->allclean = 0;
1460}
1461
1462static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1463				 sector_t offset, int inc)
1464{
1465	sector_t chunk = offset >> bitmap->chunkshift;
1466	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1467	bitmap->bp[page].count += inc;
1468	md_bitmap_checkfree(bitmap, page);
1469}
1470
1471static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1472{
1473	sector_t chunk = offset >> bitmap->chunkshift;
1474	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1475	struct bitmap_page *bp = &bitmap->bp[page];
1476
1477	if (!bp->pending)
1478		bp->pending = 1;
1479}
1480
1481static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1482					       sector_t offset, sector_t *blocks,
1483					       int create);
1484
1485static void mddev_set_timeout(struct mddev *mddev, unsigned long timeout,
1486			      bool force)
1487{
1488	struct md_thread *thread;
1489
1490	rcu_read_lock();
1491	thread = rcu_dereference(mddev->thread);
1492
1493	if (!thread)
1494		goto out;
1495
1496	if (force || thread->timeout < MAX_SCHEDULE_TIMEOUT)
1497		thread->timeout = timeout;
1498
1499out:
1500	rcu_read_unlock();
1501}
1502
1503/*
1504 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1505 *			out to disk
1506 */
1507static void bitmap_daemon_work(struct mddev *mddev)
1508{
1509	struct bitmap *bitmap;
1510	unsigned long j;
1511	unsigned long nextpage;
1512	sector_t blocks;
1513	struct bitmap_counts *counts;
1514
1515	/* Use a mutex to guard daemon_work against
1516	 * bitmap_destroy.
1517	 */
1518	mutex_lock(&mddev->bitmap_info.mutex);
1519	bitmap = mddev->bitmap;
1520	if (bitmap == NULL) {
1521		mutex_unlock(&mddev->bitmap_info.mutex);
1522		return;
1523	}
1524	if (time_before(jiffies, bitmap->daemon_lastrun
1525			+ mddev->bitmap_info.daemon_sleep))
1526		goto done;
1527
1528	bitmap->daemon_lastrun = jiffies;
1529	if (bitmap->allclean) {
1530		mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
1531		goto done;
1532	}
1533	bitmap->allclean = 1;
1534
1535	mddev_add_trace_msg(bitmap->mddev, "md bitmap_daemon_work");
1536
1537	/* Any file-page which is PENDING now needs to be written.
1538	 * So set NEEDWRITE now, then after we make any last-minute changes
1539	 * we will write it.
1540	 */
1541	for (j = 0; j < bitmap->storage.file_pages; j++)
1542		if (test_and_clear_page_attr(bitmap, j,
1543					     BITMAP_PAGE_PENDING))
1544			set_page_attr(bitmap, j,
1545				      BITMAP_PAGE_NEEDWRITE);
1546
1547	if (bitmap->need_sync &&
1548	    mddev->bitmap_info.external == 0) {
1549		/* Arrange for superblock update as well as
1550		 * other changes */
1551		bitmap_super_t *sb;
1552		bitmap->need_sync = 0;
1553		if (bitmap->storage.filemap) {
1554			sb = kmap_local_page(bitmap->storage.sb_page);
1555			sb->events_cleared =
1556				cpu_to_le64(bitmap->events_cleared);
1557			kunmap_local(sb);
1558			set_page_attr(bitmap, 0,
1559				      BITMAP_PAGE_NEEDWRITE);
1560		}
1561	}
1562	/* Now look at the bitmap counters and if any are '2' or '1',
1563	 * decrement and handle accordingly.
1564	 */
1565	counts = &bitmap->counts;
1566	spin_lock_irq(&counts->lock);
1567	nextpage = 0;
1568	for (j = 0; j < counts->chunks; j++) {
1569		bitmap_counter_t *bmc;
1570		sector_t  block = (sector_t)j << counts->chunkshift;
1571
1572		if (j == nextpage) {
1573			nextpage += PAGE_COUNTER_RATIO;
1574			if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1575				j |= PAGE_COUNTER_MASK;
1576				continue;
1577			}
1578			counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1579		}
1580
1581		bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1582		if (!bmc) {
1583			j |= PAGE_COUNTER_MASK;
1584			continue;
1585		}
1586		if (*bmc == 1 && !bitmap->need_sync) {
1587			/* We can clear the bit */
1588			*bmc = 0;
1589			md_bitmap_count_page(counts, block, -1);
1590			md_bitmap_file_clear_bit(bitmap, block);
1591		} else if (*bmc && *bmc <= 2) {
1592			*bmc = 1;
1593			md_bitmap_set_pending(counts, block);
1594			bitmap->allclean = 0;
1595		}
1596	}
1597	spin_unlock_irq(&counts->lock);
1598
1599	md_bitmap_wait_writes(bitmap);
1600	/* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1601	 * DIRTY pages need to be written by bitmap_unplug so it can wait
1602	 * for them.
1603	 * If we find any DIRTY page we stop there and let bitmap_unplug
1604	 * handle all the rest.  This is important in the case where
1605	 * the first blocking holds the superblock and it has been updated.
1606	 * We mustn't write any other blocks before the superblock.
1607	 */
1608	for (j = 0;
1609	     j < bitmap->storage.file_pages
1610		     && !test_bit(BITMAP_STALE, &bitmap->flags);
1611	     j++) {
1612		if (test_page_attr(bitmap, j,
1613				   BITMAP_PAGE_DIRTY))
1614			/* bitmap_unplug will handle the rest */
1615			break;
1616		if (bitmap->storage.filemap &&
1617		    test_and_clear_page_attr(bitmap, j,
1618					     BITMAP_PAGE_NEEDWRITE))
1619			filemap_write_page(bitmap, j, false);
1620	}
1621
1622 done:
1623	if (bitmap->allclean == 0)
1624		mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
1625	mutex_unlock(&mddev->bitmap_info.mutex);
1626}
1627
1628static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1629					       sector_t offset, sector_t *blocks,
1630					       int create)
1631__releases(bitmap->lock)
1632__acquires(bitmap->lock)
1633{
1634	/* If 'create', we might release the lock and reclaim it.
1635	 * The lock must have been taken with interrupts enabled.
1636	 * If !create, we don't release the lock.
1637	 */
1638	sector_t chunk = offset >> bitmap->chunkshift;
1639	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1640	unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1641	sector_t csize = ((sector_t)1) << bitmap->chunkshift;
1642	int err;
1643
1644	if (page >= bitmap->pages) {
1645		/*
1646		 * This can happen if bitmap_start_sync goes beyond
1647		 * End-of-device while looking for a whole page or
1648		 * user set a huge number to sysfs bitmap_set_bits.
1649		 */
1650		*blocks = csize - (offset & (csize - 1));
1651		return NULL;
1652	}
1653	err = md_bitmap_checkpage(bitmap, page, create, 0);
1654
1655	if (bitmap->bp[page].hijacked ||
1656	    bitmap->bp[page].map == NULL)
1657		csize = ((sector_t)1) << (bitmap->chunkshift +
1658					  PAGE_COUNTER_SHIFT);
1659
1660	*blocks = csize - (offset & (csize - 1));
1661
1662	if (err < 0)
1663		return NULL;
1664
1665	/* now locked ... */
1666
1667	if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1668		/* should we use the first or second counter field
1669		 * of the hijacked pointer? */
1670		int hi = (pageoff > PAGE_COUNTER_MASK);
1671		return  &((bitmap_counter_t *)
1672			  &bitmap->bp[page].map)[hi];
1673	} else /* page is allocated */
1674		return (bitmap_counter_t *)
1675			&(bitmap->bp[page].map[pageoff]);
1676}
1677
1678static void bitmap_start_write(struct mddev *mddev, sector_t offset,
1679			       unsigned long sectors)
1680{
1681	struct bitmap *bitmap = mddev->bitmap;
1682
1683	if (!bitmap)
1684		return;
1685
1686	while (sectors) {
1687		sector_t blocks;
1688		bitmap_counter_t *bmc;
1689
1690		spin_lock_irq(&bitmap->counts.lock);
1691		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1692		if (!bmc) {
1693			spin_unlock_irq(&bitmap->counts.lock);
1694			return;
1695		}
1696
1697		if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1698			DEFINE_WAIT(__wait);
1699			/* note that it is safe to do the prepare_to_wait
1700			 * after the test as long as we do it before dropping
1701			 * the spinlock.
1702			 */
1703			prepare_to_wait(&bitmap->overflow_wait, &__wait,
1704					TASK_UNINTERRUPTIBLE);
1705			spin_unlock_irq(&bitmap->counts.lock);
1706			schedule();
1707			finish_wait(&bitmap->overflow_wait, &__wait);
1708			continue;
1709		}
1710
1711		switch (*bmc) {
1712		case 0:
1713			md_bitmap_file_set_bit(bitmap, offset);
1714			md_bitmap_count_page(&bitmap->counts, offset, 1);
1715			fallthrough;
1716		case 1:
1717			*bmc = 2;
1718		}
1719
1720		(*bmc)++;
1721
1722		spin_unlock_irq(&bitmap->counts.lock);
1723
1724		offset += blocks;
1725		if (sectors > blocks)
1726			sectors -= blocks;
1727		else
1728			sectors = 0;
1729	}
1730}
1731
1732static void bitmap_end_write(struct mddev *mddev, sector_t offset,
1733			     unsigned long sectors)
1734{
1735	struct bitmap *bitmap = mddev->bitmap;
1736
1737	if (!bitmap)
1738		return;
1739
1740	while (sectors) {
1741		sector_t blocks;
1742		unsigned long flags;
1743		bitmap_counter_t *bmc;
1744
1745		spin_lock_irqsave(&bitmap->counts.lock, flags);
1746		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1747		if (!bmc) {
1748			spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1749			return;
1750		}
1751
1752		if (!bitmap->mddev->degraded) {
1753			if (bitmap->events_cleared < bitmap->mddev->events) {
1754				bitmap->events_cleared = bitmap->mddev->events;
1755				bitmap->need_sync = 1;
1756				sysfs_notify_dirent_safe(
1757						bitmap->sysfs_can_clear);
1758			}
1759		} else if (!NEEDED(*bmc)) {
1760			*bmc |= NEEDED_MASK;
1761		}
1762
1763		if (COUNTER(*bmc) == COUNTER_MAX)
1764			wake_up(&bitmap->overflow_wait);
1765
1766		(*bmc)--;
1767		if (*bmc <= 2) {
1768			md_bitmap_set_pending(&bitmap->counts, offset);
1769			bitmap->allclean = 0;
1770		}
1771		spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1772		offset += blocks;
1773		if (sectors > blocks)
1774			sectors -= blocks;
1775		else
1776			sectors = 0;
1777	}
1778}
1779
1780static bool __bitmap_start_sync(struct bitmap *bitmap, sector_t offset,
1781				sector_t *blocks, bool degraded)
1782{
1783	bitmap_counter_t *bmc;
1784	bool rv = false;
1785
1786	spin_lock_irq(&bitmap->counts.lock);
1787	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1788	if (bmc) {
1789		/* locked */
1790		if (RESYNC(*bmc)) {
1791			rv = true;
1792		} else if (NEEDED(*bmc)) {
1793			rv = true;
1794			if (!degraded) { /* don't set/clear bits if degraded */
1795				*bmc |= RESYNC_MASK;
1796				*bmc &= ~NEEDED_MASK;
1797			}
1798		}
1799	}
1800	spin_unlock_irq(&bitmap->counts.lock);
1801
1802	return rv;
1803}
1804
1805static bool bitmap_start_sync(struct mddev *mddev, sector_t offset,
1806			      sector_t *blocks, bool degraded)
1807{
1808	/* bitmap_start_sync must always report on multiples of whole
1809	 * pages, otherwise resync (which is very PAGE_SIZE based) will
1810	 * get confused.
1811	 * So call __bitmap_start_sync repeatedly (if needed) until
1812	 * At least PAGE_SIZE>>9 blocks are covered.
1813	 * Return the 'or' of the result.
1814	 */
1815	bool rv = false;
1816	sector_t blocks1;
1817
1818	*blocks = 0;
1819	while (*blocks < (PAGE_SIZE>>9)) {
1820		rv |= __bitmap_start_sync(mddev->bitmap, offset,
1821					  &blocks1, degraded);
1822		offset += blocks1;
1823		*blocks += blocks1;
1824	}
1825
1826	return rv;
1827}
1828
1829static void __bitmap_end_sync(struct bitmap *bitmap, sector_t offset,
1830			      sector_t *blocks, bool aborted)
1831{
1832	bitmap_counter_t *bmc;
1833	unsigned long flags;
1834
1835	spin_lock_irqsave(&bitmap->counts.lock, flags);
1836	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1837	if (bmc == NULL)
1838		goto unlock;
1839	/* locked */
1840	if (RESYNC(*bmc)) {
1841		*bmc &= ~RESYNC_MASK;
1842
1843		if (!NEEDED(*bmc) && aborted)
1844			*bmc |= NEEDED_MASK;
1845		else {
1846			if (*bmc <= 2) {
1847				md_bitmap_set_pending(&bitmap->counts, offset);
1848				bitmap->allclean = 0;
1849			}
1850		}
1851	}
1852 unlock:
1853	spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1854}
1855
1856static void bitmap_end_sync(struct mddev *mddev, sector_t offset,
1857			    sector_t *blocks)
1858{
1859	__bitmap_end_sync(mddev->bitmap, offset, blocks, true);
1860}
1861
1862static void bitmap_close_sync(struct mddev *mddev)
1863{
1864	/* Sync has finished, and any bitmap chunks that weren't synced
1865	 * properly have been aborted.  It remains to us to clear the
1866	 * RESYNC bit wherever it is still on
1867	 */
1868	sector_t sector = 0;
1869	sector_t blocks;
1870	struct bitmap *bitmap = mddev->bitmap;
1871
1872	if (!bitmap)
1873		return;
1874
1875	while (sector < bitmap->mddev->resync_max_sectors) {
1876		__bitmap_end_sync(bitmap, sector, &blocks, false);
1877		sector += blocks;
1878	}
1879}
1880
1881static void bitmap_cond_end_sync(struct mddev *mddev, sector_t sector,
1882				 bool force)
1883{
1884	sector_t s = 0;
1885	sector_t blocks;
1886	struct bitmap *bitmap = mddev->bitmap;
1887
1888	if (!bitmap)
1889		return;
1890	if (sector == 0) {
1891		bitmap->last_end_sync = jiffies;
1892		return;
1893	}
1894	if (!force && time_before(jiffies, (bitmap->last_end_sync
1895				  + bitmap->mddev->bitmap_info.daemon_sleep)))
1896		return;
1897	wait_event(bitmap->mddev->recovery_wait,
1898		   atomic_read(&bitmap->mddev->recovery_active) == 0);
1899
1900	bitmap->mddev->curr_resync_completed = sector;
1901	set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1902	sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1903	s = 0;
1904	while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1905		__bitmap_end_sync(bitmap, s, &blocks, false);
1906		s += blocks;
1907	}
1908	bitmap->last_end_sync = jiffies;
1909	sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1910}
1911
1912static void bitmap_sync_with_cluster(struct mddev *mddev,
1913				     sector_t old_lo, sector_t old_hi,
1914				     sector_t new_lo, sector_t new_hi)
1915{
1916	struct bitmap *bitmap = mddev->bitmap;
1917	sector_t sector, blocks = 0;
1918
1919	for (sector = old_lo; sector < new_lo; ) {
1920		__bitmap_end_sync(bitmap, sector, &blocks, false);
1921		sector += blocks;
1922	}
1923	WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1924
1925	for (sector = old_hi; sector < new_hi; ) {
1926		bitmap_start_sync(mddev, sector, &blocks, false);
1927		sector += blocks;
1928	}
1929	WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1930}
1931
1932static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1933{
1934	/* For each chunk covered by any of these sectors, set the
1935	 * counter to 2 and possibly set resync_needed.  They should all
1936	 * be 0 at this point
1937	 */
1938
1939	sector_t secs;
1940	bitmap_counter_t *bmc;
1941	spin_lock_irq(&bitmap->counts.lock);
1942	bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1943	if (!bmc) {
1944		spin_unlock_irq(&bitmap->counts.lock);
1945		return;
1946	}
1947	if (!*bmc) {
1948		*bmc = 2;
1949		md_bitmap_count_page(&bitmap->counts, offset, 1);
1950		md_bitmap_set_pending(&bitmap->counts, offset);
1951		bitmap->allclean = 0;
1952	}
1953	if (needed)
1954		*bmc |= NEEDED_MASK;
1955	spin_unlock_irq(&bitmap->counts.lock);
1956}
1957
1958/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1959static void bitmap_dirty_bits(struct mddev *mddev, unsigned long s,
1960			      unsigned long e)
1961{
1962	unsigned long chunk;
1963	struct bitmap *bitmap = mddev->bitmap;
1964
1965	if (!bitmap)
1966		return;
1967
1968	for (chunk = s; chunk <= e; chunk++) {
1969		sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1970
1971		md_bitmap_set_memory_bits(bitmap, sec, 1);
1972		md_bitmap_file_set_bit(bitmap, sec);
1973		if (sec < bitmap->mddev->resync_offset)
1974			/* We are asserting that the array is dirty,
1975			 * so move the resync_offset address back so
1976			 * that it is obvious that it is dirty
1977			 */
1978			bitmap->mddev->resync_offset = sec;
1979	}
1980}
1981
1982static void bitmap_flush(struct mddev *mddev)
1983{
1984	struct bitmap *bitmap = mddev->bitmap;
1985	long sleep;
1986
1987	if (!bitmap) /* there was no bitmap */
1988		return;
1989
1990	/* run the daemon_work three time to ensure everything is flushed
1991	 * that can be
1992	 */
1993	sleep = mddev->bitmap_info.daemon_sleep * 2;
1994	bitmap->daemon_lastrun -= sleep;
1995	bitmap_daemon_work(mddev);
1996	bitmap->daemon_lastrun -= sleep;
1997	bitmap_daemon_work(mddev);
1998	bitmap->daemon_lastrun -= sleep;
1999	bitmap_daemon_work(mddev);
2000	if (mddev->bitmap_info.external)
2001		md_super_wait(mddev);
2002	bitmap_update_sb(bitmap);
2003}
2004
2005static void md_bitmap_free(void *data)
2006{
2007	unsigned long k, pages;
2008	struct bitmap_page *bp;
2009	struct bitmap *bitmap = data;
2010
2011	if (!bitmap) /* there was no bitmap */
2012		return;
2013
2014	if (bitmap->sysfs_can_clear)
2015		sysfs_put(bitmap->sysfs_can_clear);
2016
2017	if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
2018		bitmap->cluster_slot == bitmap->mddev->cluster_ops->slot_number(bitmap->mddev))
2019		md_cluster_stop(bitmap->mddev);
2020
2021	/* Shouldn't be needed - but just in case.... */
2022	wait_event(bitmap->write_wait,
2023		   atomic_read(&bitmap->pending_writes) == 0);
2024
2025	/* release the bitmap file  */
2026	md_bitmap_file_unmap(&bitmap->storage);
2027
2028	bp = bitmap->counts.bp;
2029	pages = bitmap->counts.pages;
2030
2031	/* free all allocated memory */
2032
2033	if (bp) /* deallocate the page memory */
2034		for (k = 0; k < pages; k++)
2035			if (bp[k].map && !bp[k].hijacked)
2036				kfree(bp[k].map);
2037	kfree(bp);
2038	kfree(bitmap);
2039}
2040
2041static void bitmap_start_behind_write(struct mddev *mddev)
2042{
2043	struct bitmap *bitmap = mddev->bitmap;
2044	int bw;
2045
2046	atomic_inc(&bitmap->behind_writes);
2047	bw = atomic_read(&bitmap->behind_writes);
2048	if (bw > bitmap->behind_writes_used)
2049		bitmap->behind_writes_used = bw;
2050
2051	pr_debug("inc write-behind count %d/%lu\n",
2052		 bw, bitmap->mddev->bitmap_info.max_write_behind);
2053}
2054
2055static void bitmap_end_behind_write(struct mddev *mddev)
2056{
2057	struct bitmap *bitmap = mddev->bitmap;
2058
2059	if (atomic_dec_and_test(&bitmap->behind_writes))
2060		wake_up(&bitmap->behind_wait);
2061	pr_debug("dec write-behind count %d/%lu\n",
2062		 atomic_read(&bitmap->behind_writes),
2063		 bitmap->mddev->bitmap_info.max_write_behind);
2064}
2065
2066static void bitmap_wait_behind_writes(struct mddev *mddev)
2067{
2068	struct bitmap *bitmap = mddev->bitmap;
2069
2070	/* wait for behind writes to complete */
2071	if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2072		pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
2073			 mdname(mddev));
2074		/* need to kick something here to make sure I/O goes? */
2075		wait_event(bitmap->behind_wait,
2076			   atomic_read(&bitmap->behind_writes) == 0);
2077	}
2078}
2079
2080static void bitmap_destroy(struct mddev *mddev)
2081{
2082	struct bitmap *bitmap = mddev->bitmap;
2083
2084	if (!bitmap) /* there was no bitmap */
2085		return;
2086
2087	bitmap_wait_behind_writes(mddev);
2088	if (!mddev->serialize_policy)
2089		mddev_destroy_serial_pool(mddev, NULL);
2090
2091	mutex_lock(&mddev->bitmap_info.mutex);
2092	spin_lock(&mddev->lock);
2093	mddev->bitmap = NULL; /* disconnect from the md device */
2094	spin_unlock(&mddev->lock);
2095	mutex_unlock(&mddev->bitmap_info.mutex);
2096	mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
2097
2098	md_bitmap_free(bitmap);
2099}
2100
2101/*
2102 * initialize the bitmap structure
2103 * if this returns an error, bitmap_destroy must be called to do clean up
2104 * once mddev->bitmap is set
2105 */
2106static struct bitmap *__bitmap_create(struct mddev *mddev, int slot)
2107{
2108	struct bitmap *bitmap;
2109	sector_t blocks = mddev->resync_max_sectors;
2110	struct file *file = mddev->bitmap_info.file;
2111	int err;
2112	struct kernfs_node *bm = NULL;
2113
2114	BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
2115
2116	BUG_ON(file && mddev->bitmap_info.offset);
2117
2118	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
2119		pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
2120			  mdname(mddev));
2121		return ERR_PTR(-EBUSY);
2122	}
2123
2124	bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
2125	if (!bitmap)
2126		return ERR_PTR(-ENOMEM);
2127
2128	spin_lock_init(&bitmap->counts.lock);
2129	atomic_set(&bitmap->pending_writes, 0);
2130	init_waitqueue_head(&bitmap->write_wait);
2131	init_waitqueue_head(&bitmap->overflow_wait);
2132	init_waitqueue_head(&bitmap->behind_wait);
2133
2134	bitmap->mddev = mddev;
2135	bitmap->cluster_slot = slot;
2136
2137	if (mddev->kobj.sd)
2138		bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
2139	if (bm) {
2140		bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
2141		sysfs_put(bm);
2142	} else
2143		bitmap->sysfs_can_clear = NULL;
2144
2145	bitmap->storage.file = file;
2146	if (file) {
2147		get_file(file);
2148		/* As future accesses to this file will use bmap,
2149		 * and bypass the page cache, we must sync the file
2150		 * first.
2151		 */
2152		vfs_fsync(file, 1);
2153	}
2154	/* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
2155	if (!mddev->bitmap_info.external) {
2156		/*
2157		 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
2158		 * instructing us to create a new on-disk bitmap instance.
2159		 */
2160		if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
2161			err = md_bitmap_new_disk_sb(bitmap);
2162		else
2163			err = md_bitmap_read_sb(bitmap);
2164	} else {
2165		err = 0;
2166		if (mddev->bitmap_info.chunksize == 0 ||
2167		    mddev->bitmap_info.daemon_sleep == 0)
2168			/* chunksize and time_base need to be
2169			 * set first. */
2170			err = -EINVAL;
2171	}
2172	if (err)
2173		goto error;
2174
2175	bitmap->daemon_lastrun = jiffies;
2176	err = __bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize,
2177			      true);
2178	if (err)
2179		goto error;
2180
2181	pr_debug("created bitmap (%lu pages) for device %s\n",
2182		 bitmap->counts.pages, bmname(bitmap));
2183
2184	err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
2185	if (err)
2186		goto error;
2187
2188	return bitmap;
2189 error:
2190	md_bitmap_free(bitmap);
2191	return ERR_PTR(err);
2192}
2193
2194static int bitmap_create(struct mddev *mddev)
2195{
2196	struct bitmap *bitmap = __bitmap_create(mddev, -1);
2197
2198	if (IS_ERR(bitmap))
2199		return PTR_ERR(bitmap);
2200
2201	mddev->bitmap = bitmap;
2202	return 0;
2203}
2204
2205static int bitmap_load(struct mddev *mddev)
2206{
2207	int err = 0;
2208	sector_t start = 0;
2209	sector_t sector = 0;
2210	struct bitmap *bitmap = mddev->bitmap;
2211	struct md_rdev *rdev;
2212
2213	if (!bitmap)
2214		goto out;
2215
2216	rdev_for_each(rdev, mddev)
2217		mddev_create_serial_pool(mddev, rdev);
2218
2219	if (mddev_is_clustered(mddev))
2220		mddev->cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
2221
2222	/* Clear out old bitmap info first:  Either there is none, or we
2223	 * are resuming after someone else has possibly changed things,
2224	 * so we should forget old cached info.
2225	 * All chunks should be clean, but some might need_sync.
2226	 */
2227	while (sector < mddev->resync_max_sectors) {
2228		sector_t blocks;
2229		bitmap_start_sync(mddev, sector, &blocks, false);
2230		sector += blocks;
2231	}
2232	bitmap_close_sync(mddev);
2233
2234	if (mddev->degraded == 0
2235	    || bitmap->events_cleared == mddev->events)
2236		/* no need to keep dirty bits to optimise a
2237		 * re-add of a missing device */
2238		start = mddev->resync_offset;
2239
2240	mutex_lock(&mddev->bitmap_info.mutex);
2241	err = md_bitmap_init_from_disk(bitmap, start);
2242	mutex_unlock(&mddev->bitmap_info.mutex);
2243
2244	if (err)
2245		goto out;
2246	clear_bit(BITMAP_STALE, &bitmap->flags);
2247
2248	/* Kick recovery in case any bits were set */
2249	set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
2250
2251	mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
2252	md_wakeup_thread(mddev->thread);
2253
2254	bitmap_update_sb(bitmap);
2255
2256	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
2257		err = -EIO;
2258out:
2259	return err;
2260}
2261
2262/* caller need to free returned bitmap with md_bitmap_free() */
2263static void *bitmap_get_from_slot(struct mddev *mddev, int slot)
2264{
2265	int rv = 0;
2266	struct bitmap *bitmap;
2267
2268	bitmap = __bitmap_create(mddev, slot);
2269	if (IS_ERR(bitmap)) {
2270		rv = PTR_ERR(bitmap);
2271		return ERR_PTR(rv);
2272	}
2273
2274	rv = md_bitmap_init_from_disk(bitmap, 0);
2275	if (rv) {
2276		md_bitmap_free(bitmap);
2277		return ERR_PTR(rv);
2278	}
2279
2280	return bitmap;
2281}
2282
2283/* Loads the bitmap associated with slot and copies the resync information
2284 * to our bitmap
2285 */
2286static int bitmap_copy_from_slot(struct mddev *mddev, int slot, sector_t *low,
2287				 sector_t *high, bool clear_bits)
2288{
2289	int rv = 0, i, j;
2290	sector_t block, lo = 0, hi = 0;
2291	struct bitmap_counts *counts;
2292	struct bitmap *bitmap;
2293
2294	bitmap = bitmap_get_from_slot(mddev, slot);
2295	if (IS_ERR(bitmap)) {
2296		pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
2297		return -1;
2298	}
2299
2300	counts = &bitmap->counts;
2301	for (j = 0; j < counts->chunks; j++) {
2302		block = (sector_t)j << counts->chunkshift;
2303		if (md_bitmap_file_test_bit(bitmap, block)) {
2304			if (!lo)
2305				lo = block;
2306			hi = block;
2307			md_bitmap_file_clear_bit(bitmap, block);
2308			md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2309			md_bitmap_file_set_bit(mddev->bitmap, block);
2310		}
2311	}
2312
2313	if (clear_bits) {
2314		bitmap_update_sb(bitmap);
2315		/* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2316		 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2317		for (i = 0; i < bitmap->storage.file_pages; i++)
2318			if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2319				set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2320		__bitmap_unplug(bitmap);
2321	}
2322	__bitmap_unplug(mddev->bitmap);
2323	*low = lo;
2324	*high = hi;
2325	md_bitmap_free(bitmap);
2326
2327	return rv;
2328}
2329
2330static void bitmap_set_pages(void *data, unsigned long pages)
2331{
2332	struct bitmap *bitmap = data;
2333
2334	bitmap->counts.pages = pages;
2335}
2336
2337static int bitmap_get_stats(void *data, struct md_bitmap_stats *stats)
2338{
2339	struct bitmap_storage *storage;
2340	struct bitmap_counts *counts;
2341	struct bitmap *bitmap = data;
2342	bitmap_super_t *sb;
2343
2344	if (!bitmap)
2345		return -ENOENT;
2346	if (!bitmap->storage.sb_page)
2347		return -EINVAL;
2348	sb = kmap_local_page(bitmap->storage.sb_page);
2349	stats->sync_size = le64_to_cpu(sb->sync_size);
2350	kunmap_local(sb);
2351
2352	counts = &bitmap->counts;
2353	stats->missing_pages = counts->missing_pages;
2354	stats->pages = counts->pages;
2355
2356	storage = &bitmap->storage;
2357	stats->file_pages = storage->file_pages;
2358	stats->file = storage->file;
2359
2360	stats->behind_writes = atomic_read(&bitmap->behind_writes);
2361	stats->behind_wait = wq_has_sleeper(&bitmap->behind_wait);
2362	stats->events_cleared = bitmap->events_cleared;
2363	return 0;
2364}
2365
2366static int __bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2367			   int chunksize, bool init)
2368{
2369	/* If chunk_size is 0, choose an appropriate chunk size.
2370	 * Then possibly allocate new storage space.
2371	 * Then quiesce, copy bits, replace bitmap, and re-start
2372	 *
2373	 * This function is called both to set up the initial bitmap
2374	 * and to resize the bitmap while the array is active.
2375	 * If this happens as a result of the array being resized,
2376	 * chunksize will be zero, and we need to choose a suitable
2377	 * chunksize, otherwise we use what we are given.
2378	 */
2379	struct bitmap_storage store;
2380	struct bitmap_counts old_counts;
2381	unsigned long chunks;
2382	sector_t block;
2383	sector_t old_blocks, new_blocks;
2384	int chunkshift;
2385	int ret = 0;
2386	long pages;
2387	struct bitmap_page *new_bp;
2388
2389	if (bitmap->storage.file && !init) {
2390		pr_info("md: cannot resize file-based bitmap\n");
2391		return -EINVAL;
2392	}
2393
2394	if (chunksize == 0) {
2395		/* If there is enough space, leave the chunk size unchanged,
2396		 * else increase by factor of two until there is enough space.
2397		 */
2398		long bytes;
2399		long space = bitmap->mddev->bitmap_info.space;
2400
2401		if (space == 0) {
2402			/* We don't know how much space there is, so limit
2403			 * to current size - in sectors.
2404			 */
2405			bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2406			if (!bitmap->mddev->bitmap_info.external)
2407				bytes += sizeof(bitmap_super_t);
2408			space = DIV_ROUND_UP(bytes, 512);
2409			bitmap->mddev->bitmap_info.space = space;
2410		}
2411		chunkshift = bitmap->counts.chunkshift;
2412		chunkshift--;
2413		do {
2414			/* 'chunkshift' is shift from block size to chunk size */
2415			chunkshift++;
2416			chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2417			bytes = DIV_ROUND_UP(chunks, 8);
2418			if (!bitmap->mddev->bitmap_info.external)
2419				bytes += sizeof(bitmap_super_t);
2420		} while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2421			(BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2422	} else
2423		chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2424
2425	chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2426	memset(&store, 0, sizeof(store));
2427	if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2428		ret = md_bitmap_storage_alloc(&store, chunks,
2429					      !bitmap->mddev->bitmap_info.external,
2430					      mddev_is_clustered(bitmap->mddev)
2431					      ? bitmap->cluster_slot : 0);
2432	if (ret) {
2433		md_bitmap_file_unmap(&store);
2434		goto err;
2435	}
2436
2437	pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2438
2439	new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2440	ret = -ENOMEM;
2441	if (!new_bp) {
2442		md_bitmap_file_unmap(&store);
2443		goto err;
2444	}
2445
2446	if (!init)
2447		bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2448
2449	store.file = bitmap->storage.file;
2450	bitmap->storage.file = NULL;
2451
2452	if (store.sb_page && bitmap->storage.sb_page)
2453		memcpy(page_address(store.sb_page),
2454		       page_address(bitmap->storage.sb_page),
2455		       sizeof(bitmap_super_t));
2456	mutex_lock(&bitmap->mddev->bitmap_info.mutex);
2457	spin_lock_irq(&bitmap->counts.lock);
2458	md_bitmap_file_unmap(&bitmap->storage);
2459	bitmap->storage = store;
2460
2461	old_counts = bitmap->counts;
2462	bitmap->counts.bp = new_bp;
2463	bitmap->counts.pages = pages;
2464	bitmap->counts.missing_pages = pages;
2465	bitmap->counts.chunkshift = chunkshift;
2466	bitmap->counts.chunks = chunks;
2467	bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2468						     BITMAP_BLOCK_SHIFT);
2469
2470	blocks = min(old_counts.chunks << old_counts.chunkshift,
2471		     chunks << chunkshift);
2472
2473	/* For cluster raid, need to pre-allocate bitmap */
2474	if (mddev_is_clustered(bitmap->mddev)) {
2475		unsigned long page;
2476		for (page = 0; page < pages; page++) {
2477			ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2478			if (ret) {
2479				unsigned long k;
2480
2481				/* deallocate the page memory */
2482				for (k = 0; k < page; k++) {
2483					kfree(new_bp[k].map);
2484				}
2485				kfree(new_bp);
2486
2487				/* restore some fields from old_counts */
2488				bitmap->counts.bp = old_counts.bp;
2489				bitmap->counts.pages = old_counts.pages;
2490				bitmap->counts.missing_pages = old_counts.pages;
2491				bitmap->counts.chunkshift = old_counts.chunkshift;
2492				bitmap->counts.chunks = old_counts.chunks;
2493				bitmap->mddev->bitmap_info.chunksize =
2494					1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2495				blocks = old_counts.chunks << old_counts.chunkshift;
2496				pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2497				break;
2498			} else
2499				bitmap->counts.bp[page].count += 1;
2500		}
2501	}
2502
2503	for (block = 0; block < blocks; ) {
2504		bitmap_counter_t *bmc_old, *bmc_new;
2505		int set;
2506
2507		bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2508		set = bmc_old && NEEDED(*bmc_old);
2509
2510		if (set) {
2511			bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2512			if (bmc_new) {
2513				if (*bmc_new == 0) {
2514					/* need to set on-disk bits too. */
2515					sector_t end = block + new_blocks;
2516					sector_t start = block >> chunkshift;
2517
2518					start <<= chunkshift;
2519					while (start < end) {
2520						md_bitmap_file_set_bit(bitmap, block);
2521						start += 1 << chunkshift;
2522					}
2523					*bmc_new = 2;
2524					md_bitmap_count_page(&bitmap->counts, block, 1);
2525					md_bitmap_set_pending(&bitmap->counts, block);
2526				}
2527				*bmc_new |= NEEDED_MASK;
2528			}
2529			if (new_blocks < old_blocks)
2530				old_blocks = new_blocks;
2531		}
2532		block += old_blocks;
2533	}
2534
2535	if (bitmap->counts.bp != old_counts.bp) {
2536		unsigned long k;
2537		for (k = 0; k < old_counts.pages; k++)
2538			if (!old_counts.bp[k].hijacked)
2539				kfree(old_counts.bp[k].map);
2540		kfree(old_counts.bp);
2541	}
2542
2543	if (!init) {
2544		int i;
2545		while (block < (chunks << chunkshift)) {
2546			bitmap_counter_t *bmc;
2547			bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2548			if (bmc) {
2549				/* new space.  It needs to be resynced, so
2550				 * we set NEEDED_MASK.
2551				 */
2552				if (*bmc == 0) {
2553					*bmc = NEEDED_MASK | 2;
2554					md_bitmap_count_page(&bitmap->counts, block, 1);
2555					md_bitmap_set_pending(&bitmap->counts, block);
2556				}
2557			}
2558			block += new_blocks;
2559		}
2560		for (i = 0; i < bitmap->storage.file_pages; i++)
2561			set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2562	}
2563	spin_unlock_irq(&bitmap->counts.lock);
2564	mutex_unlock(&bitmap->mddev->bitmap_info.mutex);
2565	if (!init) {
2566		__bitmap_unplug(bitmap);
2567		bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2568	}
2569	ret = 0;
2570err:
2571	return ret;
2572}
2573
2574static int bitmap_resize(struct mddev *mddev, sector_t blocks, int chunksize)
2575{
2576	struct bitmap *bitmap = mddev->bitmap;
2577
2578	if (!bitmap)
2579		return 0;
2580
2581	return __bitmap_resize(bitmap, blocks, chunksize, false);
2582}
2583
2584static ssize_t
2585location_show(struct mddev *mddev, char *page)
2586{
2587	ssize_t len;
2588	if (mddev->bitmap_info.file)
2589		len = sprintf(page, "file");
2590	else if (mddev->bitmap_info.offset)
2591		len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2592	else
2593		len = sprintf(page, "none");
2594	len += sprintf(page+len, "\n");
2595	return len;
2596}
2597
2598static ssize_t
2599location_store(struct mddev *mddev, const char *buf, size_t len)
2600{
2601	int rv;
2602
2603	rv = mddev_suspend_and_lock(mddev);
2604	if (rv)
2605		return rv;
2606
2607	if (mddev->pers) {
2608		if (mddev->recovery || mddev->sync_thread) {
2609			rv = -EBUSY;
2610			goto out;
2611		}
2612	}
2613
2614	if (mddev->bitmap || mddev->bitmap_info.file ||
2615	    mddev->bitmap_info.offset) {
2616		/* bitmap already configured.  Only option is to clear it */
2617		if (strncmp(buf, "none", 4) != 0) {
2618			rv = -EBUSY;
2619			goto out;
2620		}
2621
2622		bitmap_destroy(mddev);
2623		mddev->bitmap_info.offset = 0;
2624		if (mddev->bitmap_info.file) {
2625			struct file *f = mddev->bitmap_info.file;
2626			mddev->bitmap_info.file = NULL;
2627			fput(f);
2628		}
2629	} else {
2630		/* No bitmap, OK to set a location */
2631		long long offset;
2632
2633		if (strncmp(buf, "none", 4) == 0)
2634			/* nothing to be done */;
2635		else if (strncmp(buf, "file:", 5) == 0) {
2636			/* Not supported yet */
2637			rv = -EINVAL;
2638			goto out;
2639		} else {
2640			if (buf[0] == '+')
2641				rv = kstrtoll(buf+1, 10, &offset);
2642			else
2643				rv = kstrtoll(buf, 10, &offset);
2644			if (rv)
2645				goto out;
2646			if (offset == 0) {
2647				rv = -EINVAL;
2648				goto out;
2649			}
2650			if (mddev->bitmap_info.external == 0 &&
2651			    mddev->major_version == 0 &&
2652			    offset != mddev->bitmap_info.default_offset) {
2653				rv = -EINVAL;
2654				goto out;
2655			}
2656
2657			mddev->bitmap_info.offset = offset;
2658			rv = bitmap_create(mddev);
2659			if (rv)
2660				goto out;
2661
2662			rv = bitmap_load(mddev);
2663			if (rv) {
2664				mddev->bitmap_info.offset = 0;
2665				bitmap_destroy(mddev);
2666				goto out;
2667			}
2668		}
2669	}
2670	if (!mddev->external) {
2671		/* Ensure new bitmap info is stored in
2672		 * metadata promptly.
2673		 */
2674		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2675		md_wakeup_thread(mddev->thread);
2676	}
2677	rv = 0;
2678out:
2679	mddev_unlock_and_resume(mddev);
2680	if (rv)
2681		return rv;
2682	return len;
2683}
2684
2685static struct md_sysfs_entry bitmap_location =
2686__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2687
2688/* 'bitmap/space' is the space available at 'location' for the
2689 * bitmap.  This allows the kernel to know when it is safe to
2690 * resize the bitmap to match a resized array.
2691 */
2692static ssize_t
2693space_show(struct mddev *mddev, char *page)
2694{
2695	return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2696}
2697
2698static ssize_t
2699space_store(struct mddev *mddev, const char *buf, size_t len)
2700{
2701	struct bitmap *bitmap;
2702	unsigned long sectors;
2703	int rv;
2704
2705	rv = kstrtoul(buf, 10, &sectors);
2706	if (rv)
2707		return rv;
2708
2709	if (sectors == 0)
2710		return -EINVAL;
2711
2712	bitmap = mddev->bitmap;
2713	if (bitmap && sectors < (bitmap->storage.bytes + 511) >> 9)
2714		return -EFBIG; /* Bitmap is too big for this small space */
2715
2716	/* could make sure it isn't too big, but that isn't really
2717	 * needed - user-space should be careful.
2718	 */
2719	mddev->bitmap_info.space = sectors;
2720	return len;
2721}
2722
2723static struct md_sysfs_entry bitmap_space =
2724__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2725
2726static ssize_t
2727timeout_show(struct mddev *mddev, char *page)
2728{
2729	ssize_t len;
2730	unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2731	unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2732
2733	len = sprintf(page, "%lu", secs);
2734	if (jifs)
2735		len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2736	len += sprintf(page+len, "\n");
2737	return len;
2738}
2739
2740static ssize_t
2741timeout_store(struct mddev *mddev, const char *buf, size_t len)
2742{
2743	/* timeout can be set at any time */
2744	unsigned long timeout;
2745	int rv = strict_strtoul_scaled(buf, &timeout, 4);
2746	if (rv)
2747		return rv;
2748
2749	/* just to make sure we don't overflow... */
2750	if (timeout >= LONG_MAX / HZ)
2751		return -EINVAL;
2752
2753	timeout = timeout * HZ / 10000;
2754
2755	if (timeout >= MAX_SCHEDULE_TIMEOUT)
2756		timeout = MAX_SCHEDULE_TIMEOUT-1;
2757	if (timeout < 1)
2758		timeout = 1;
2759
2760	mddev->bitmap_info.daemon_sleep = timeout;
2761	mddev_set_timeout(mddev, timeout, false);
2762	md_wakeup_thread(mddev->thread);
2763
2764	return len;
2765}
2766
2767static struct md_sysfs_entry bitmap_timeout =
2768__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2769
2770static ssize_t
2771backlog_show(struct mddev *mddev, char *page)
2772{
2773	return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2774}
2775
2776static ssize_t
2777backlog_store(struct mddev *mddev, const char *buf, size_t len)
2778{
2779	unsigned long backlog;
2780	unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2781	struct md_rdev *rdev;
2782	bool has_write_mostly = false;
2783	int rv = kstrtoul(buf, 10, &backlog);
2784	if (rv)
2785		return rv;
2786	if (backlog > COUNTER_MAX)
2787		return -EINVAL;
2788
2789	rv = mddev_suspend_and_lock(mddev);
2790	if (rv)
2791		return rv;
2792
2793	/*
2794	 * Without write mostly device, it doesn't make sense to set
2795	 * backlog for max_write_behind.
2796	 */
2797	rdev_for_each(rdev, mddev) {
2798		if (test_bit(WriteMostly, &rdev->flags)) {
2799			has_write_mostly = true;
2800			break;
2801		}
2802	}
2803	if (!has_write_mostly) {
2804		pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2805				    mdname(mddev));
2806		mddev_unlock(mddev);
2807		return -EINVAL;
2808	}
2809
2810	mddev->bitmap_info.max_write_behind = backlog;
2811	if (!backlog && mddev->serial_info_pool) {
2812		/* serial_info_pool is not needed if backlog is zero */
2813		if (!mddev->serialize_policy)
2814			mddev_destroy_serial_pool(mddev, NULL);
2815	} else if (backlog && !mddev->serial_info_pool) {
2816		/* serial_info_pool is needed since backlog is not zero */
2817		rdev_for_each(rdev, mddev)
2818			mddev_create_serial_pool(mddev, rdev);
2819	}
2820	if (old_mwb != backlog)
2821		bitmap_update_sb(mddev->bitmap);
2822
2823	mddev_unlock_and_resume(mddev);
2824	return len;
2825}
2826
2827static struct md_sysfs_entry bitmap_backlog =
2828__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2829
2830static ssize_t
2831chunksize_show(struct mddev *mddev, char *page)
2832{
2833	return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2834}
2835
2836static ssize_t
2837chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2838{
2839	/* Can only be changed when no bitmap is active */
2840	int rv;
2841	unsigned long csize;
2842	if (mddev->bitmap)
2843		return -EBUSY;
2844	rv = kstrtoul(buf, 10, &csize);
2845	if (rv)
2846		return rv;
2847	if (csize < 512 ||
2848	    !is_power_of_2(csize))
2849		return -EINVAL;
2850	if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2851		sizeof(((bitmap_super_t *)0)->chunksize))))
2852		return -EOVERFLOW;
2853	mddev->bitmap_info.chunksize = csize;
2854	return len;
2855}
2856
2857static struct md_sysfs_entry bitmap_chunksize =
2858__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2859
2860static ssize_t metadata_show(struct mddev *mddev, char *page)
2861{
2862	if (mddev_is_clustered(mddev))
2863		return sprintf(page, "clustered\n");
2864	return sprintf(page, "%s\n", (mddev->bitmap_info.external
2865				      ? "external" : "internal"));
2866}
2867
2868static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2869{
2870	if (mddev->bitmap ||
2871	    mddev->bitmap_info.file ||
2872	    mddev->bitmap_info.offset)
2873		return -EBUSY;
2874	if (strncmp(buf, "external", 8) == 0)
2875		mddev->bitmap_info.external = 1;
2876	else if ((strncmp(buf, "internal", 8) == 0) ||
2877			(strncmp(buf, "clustered", 9) == 0))
2878		mddev->bitmap_info.external = 0;
2879	else
2880		return -EINVAL;
2881	return len;
2882}
2883
2884static struct md_sysfs_entry bitmap_metadata =
2885__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2886
2887static ssize_t can_clear_show(struct mddev *mddev, char *page)
2888{
2889	int len;
2890	struct bitmap *bitmap;
2891
2892	spin_lock(&mddev->lock);
2893	bitmap = mddev->bitmap;
2894	if (bitmap)
2895		len = sprintf(page, "%s\n", (bitmap->need_sync ? "false" :
2896								 "true"));
2897	else
2898		len = sprintf(page, "\n");
2899	spin_unlock(&mddev->lock);
2900	return len;
2901}
2902
2903static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2904{
2905	struct bitmap *bitmap = mddev->bitmap;
2906
2907	if (!bitmap)
2908		return -ENOENT;
2909
2910	if (strncmp(buf, "false", 5) == 0) {
2911		bitmap->need_sync = 1;
2912		return len;
2913	}
2914
2915	if (strncmp(buf, "true", 4) == 0) {
2916		if (mddev->degraded)
2917			return -EBUSY;
2918		bitmap->need_sync = 0;
2919		return len;
2920	}
2921
2922	return -EINVAL;
2923}
2924
2925static struct md_sysfs_entry bitmap_can_clear =
2926__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2927
2928static ssize_t
2929behind_writes_used_show(struct mddev *mddev, char *page)
2930{
2931	ssize_t ret;
2932	struct bitmap *bitmap;
2933
2934	spin_lock(&mddev->lock);
2935	bitmap = mddev->bitmap;
2936	if (!bitmap)
2937		ret = sprintf(page, "0\n");
2938	else
2939		ret = sprintf(page, "%lu\n", bitmap->behind_writes_used);
2940	spin_unlock(&mddev->lock);
2941
2942	return ret;
2943}
2944
2945static ssize_t
2946behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2947{
2948	struct bitmap *bitmap = mddev->bitmap;
2949
2950	if (bitmap)
2951		bitmap->behind_writes_used = 0;
2952	return len;
2953}
2954
2955static struct md_sysfs_entry max_backlog_used =
2956__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2957       behind_writes_used_show, behind_writes_used_reset);
2958
2959static struct attribute *md_bitmap_attrs[] = {
2960	&bitmap_location.attr,
2961	&bitmap_space.attr,
2962	&bitmap_timeout.attr,
2963	&bitmap_backlog.attr,
2964	&bitmap_chunksize.attr,
2965	&bitmap_metadata.attr,
2966	&bitmap_can_clear.attr,
2967	&max_backlog_used.attr,
2968	NULL
2969};
2970
2971static struct attribute_group md_bitmap_group = {
2972	.name = "bitmap",
2973	.attrs = md_bitmap_attrs,
2974};
2975
2976static struct bitmap_operations bitmap_ops = {
2977	.head = {
2978		.type	= MD_BITMAP,
2979		.id	= ID_BITMAP,
2980		.name	= "bitmap",
2981	},
2982
2983	.enabled		= bitmap_enabled,
2984	.create			= bitmap_create,
2985	.resize			= bitmap_resize,
2986	.load			= bitmap_load,
2987	.destroy		= bitmap_destroy,
2988	.flush			= bitmap_flush,
2989	.write_all		= bitmap_write_all,
2990	.dirty_bits		= bitmap_dirty_bits,
2991	.unplug			= bitmap_unplug,
2992	.daemon_work		= bitmap_daemon_work,
2993
2994	.start_behind_write	= bitmap_start_behind_write,
2995	.end_behind_write	= bitmap_end_behind_write,
2996	.wait_behind_writes	= bitmap_wait_behind_writes,
2997
2998	.start_write		= bitmap_start_write,
2999	.end_write		= bitmap_end_write,
3000	.start_discard		= bitmap_start_write,
3001	.end_discard		= bitmap_end_write,
3002
3003	.start_sync		= bitmap_start_sync,
3004	.end_sync		= bitmap_end_sync,
3005	.cond_end_sync		= bitmap_cond_end_sync,
3006	.close_sync		= bitmap_close_sync,
3007
3008	.update_sb		= bitmap_update_sb,
3009	.get_stats		= bitmap_get_stats,
3010
3011	.sync_with_cluster	= bitmap_sync_with_cluster,
3012	.get_from_slot		= bitmap_get_from_slot,
3013	.copy_from_slot		= bitmap_copy_from_slot,
3014	.set_pages		= bitmap_set_pages,
3015	.free			= md_bitmap_free,
3016
3017	.group			= &md_bitmap_group,
3018};
3019
3020int md_bitmap_init(void)
3021{
3022	md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
3023				       0);
3024	if (!md_bitmap_wq)
3025		return -ENOMEM;
3026
3027	return register_md_submodule(&bitmap_ops.head);
3028}
3029
3030void md_bitmap_exit(void)
3031{
3032	destroy_workqueue(md_bitmap_wq);
3033	unregister_md_submodule(&bitmap_ops.head);
3034}
~/ Read-only file preview.