Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   4 *
   5 * bitmap_create  - sets up the bitmap structure
   6 * bitmap_destroy - destroys the bitmap structure
   7 *
   8 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   9 * - added disk storage for bitmap
  10 * - changes to allow various bitmap chunk sizes
  11 */
  12
  13/*
  14 * Still to do:
  15 *
  16 * flush after percent set rather than just time based. (maybe both).
  17 */
  18
  19#include <linux/blkdev.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/slab.h>
  23#include <linux/init.h>
  24#include <linux/timer.h>
  25#include <linux/sched.h>
  26#include <linux/list.h>
  27#include <linux/file.h>
  28#include <linux/mount.h>
  29#include <linux/buffer_head.h>
  30#include <linux/seq_file.h>
  31#include <trace/events/block.h>
  32#include "md.h"
  33#include "md-bitmap.h"
  34
  35static inline char *bmname(struct bitmap *bitmap)
  36{
  37	return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  38}
  39
  40/*
  41 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
  42 *
  43 * 1) check to see if this page is allocated, if it's not then try to alloc
  44 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
  45 *    page pointer directly as a counter
  46 *
  47 * if we find our page, we increment the page's refcount so that it stays
  48 * allocated while we're using it
  49 */
  50static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
  51			       unsigned long page, int create, int no_hijack)
  52__releases(bitmap->lock)
  53__acquires(bitmap->lock)
  54{
  55	unsigned char *mappage;
  56
  57	WARN_ON_ONCE(page >= bitmap->pages);
 
 
 
 
 
 
 
  58	if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  59		return 0;
  60
  61	if (bitmap->bp[page].map) /* page is already allocated, just return */
  62		return 0;
  63
  64	if (!create)
  65		return -ENOENT;
  66
  67	/* this page has not been allocated yet */
  68
  69	spin_unlock_irq(&bitmap->lock);
  70	/* It is possible that this is being called inside a
  71	 * prepare_to_wait/finish_wait loop from raid5c:make_request().
  72	 * In general it is not permitted to sleep in that context as it
  73	 * can cause the loop to spin freely.
  74	 * That doesn't apply here as we can only reach this point
  75	 * once with any loop.
  76	 * When this function completes, either bp[page].map or
  77	 * bp[page].hijacked.  In either case, this function will
  78	 * abort before getting to this point again.  So there is
  79	 * no risk of a free-spin, and so it is safe to assert
  80	 * that sleeping here is allowed.
  81	 */
  82	sched_annotate_sleep();
  83	mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  84	spin_lock_irq(&bitmap->lock);
  85
  86	if (mappage == NULL) {
  87		pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  88		/* We don't support hijack for cluster raid */
  89		if (no_hijack)
  90			return -ENOMEM;
  91		/* failed - set the hijacked flag so that we can use the
  92		 * pointer as a counter */
  93		if (!bitmap->bp[page].map)
  94			bitmap->bp[page].hijacked = 1;
  95	} else if (bitmap->bp[page].map ||
  96		   bitmap->bp[page].hijacked) {
  97		/* somebody beat us to getting the page */
  98		kfree(mappage);
  99	} else {
 100
 101		/* no page was in place and we have one, so install it */
 102
 103		bitmap->bp[page].map = mappage;
 104		bitmap->missing_pages--;
 105	}
 106	return 0;
 107}
 108
 109/* if page is completely empty, put it back on the free list, or dealloc it */
 110/* if page was hijacked, unmark the flag so it might get alloced next time */
 111/* Note: lock should be held when calling this */
 112static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 113{
 114	char *ptr;
 115
 116	if (bitmap->bp[page].count) /* page is still busy */
 117		return;
 118
 119	/* page is no longer in use, it can be released */
 120
 121	if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 122		bitmap->bp[page].hijacked = 0;
 123		bitmap->bp[page].map = NULL;
 124	} else {
 125		/* normal case, free the page */
 126		ptr = bitmap->bp[page].map;
 127		bitmap->bp[page].map = NULL;
 128		bitmap->missing_pages++;
 129		kfree(ptr);
 130	}
 131}
 132
 133/*
 134 * bitmap file handling - read and write the bitmap file and its superblock
 135 */
 136
 137/*
 138 * basic page I/O operations
 139 */
 140
 141/* IO operations when bitmap is stored near all superblocks */
 142
 143/* choose a good rdev and read the page from there */
 144static int read_sb_page(struct mddev *mddev, loff_t offset,
 145		struct page *page, unsigned long index, int size)
 
 146{
 
 147
 148	sector_t sector = mddev->bitmap_info.offset + offset +
 149		index * (PAGE_SIZE / SECTOR_SIZE);
 150	struct md_rdev *rdev;
 
 151
 152	rdev_for_each(rdev, mddev) {
 153		u32 iosize = roundup(size, bdev_logical_block_size(rdev->bdev));
 154
 155		if (!test_bit(In_sync, &rdev->flags) ||
 156		    test_bit(Faulty, &rdev->flags) ||
 157		    test_bit(Bitmap_sync, &rdev->flags))
 158			continue;
 159
 160		if (sync_page_io(rdev, sector, iosize, page, REQ_OP_READ, true))
 
 
 
 
 
 161			return 0;
 
 162	}
 163	return -EIO;
 164}
 165
 166static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 167{
 168	/* Iterate the disks of an mddev, using rcu to protect access to the
 169	 * linked list, and raising the refcount of devices we return to ensure
 170	 * they don't disappear while in use.
 171	 * As devices are only added or removed when raid_disk is < 0 and
 172	 * nr_pending is 0 and In_sync is clear, the entries we return will
 173	 * still be in the same position on the list when we re-enter
 174	 * list_for_each_entry_continue_rcu.
 175	 *
 176	 * Note that if entered with 'rdev == NULL' to start at the
 177	 * beginning, we temporarily assign 'rdev' to an address which
 178	 * isn't really an rdev, but which can be used by
 179	 * list_for_each_entry_continue_rcu() to find the first entry.
 180	 */
 181	rcu_read_lock();
 182	if (rdev == NULL)
 183		/* start at the beginning */
 184		rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 185	else {
 186		/* release the previous rdev and start from there. */
 187		rdev_dec_pending(rdev, mddev);
 188	}
 189	list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 190		if (rdev->raid_disk >= 0 &&
 191		    !test_bit(Faulty, &rdev->flags)) {
 192			/* this is a usable devices */
 193			atomic_inc(&rdev->nr_pending);
 194			rcu_read_unlock();
 195			return rdev;
 196		}
 197	}
 198	rcu_read_unlock();
 199	return NULL;
 200}
 201
 202static unsigned int optimal_io_size(struct block_device *bdev,
 203				    unsigned int last_page_size,
 204				    unsigned int io_size)
 205{
 206	if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev))
 207		return roundup(last_page_size, bdev_io_opt(bdev));
 208	return io_size;
 209}
 210
 211static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size,
 212				   loff_t start, loff_t boundary)
 213{
 214	if (io_size != opt_size &&
 215	    start + opt_size / SECTOR_SIZE <= boundary)
 216		return opt_size;
 217	if (start + io_size / SECTOR_SIZE <= boundary)
 218		return io_size;
 219
 220	/* Overflows boundary */
 221	return 0;
 222}
 223
 224static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap,
 225			   unsigned long pg_index, struct page *page)
 226{
 
 227	struct block_device *bdev;
 228	struct mddev *mddev = bitmap->mddev;
 229	struct bitmap_storage *store = &bitmap->storage;
 230	loff_t sboff, offset = mddev->bitmap_info.offset;
 231	sector_t ps = pg_index * PAGE_SIZE / SECTOR_SIZE;
 232	unsigned int size = PAGE_SIZE;
 233	unsigned int opt_size = PAGE_SIZE;
 234	sector_t doff;
 235
 236	bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 237	if (pg_index == store->file_pages - 1) {
 238		unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1);
 239
 240		if (last_page_size == 0)
 241			last_page_size = PAGE_SIZE;
 242		size = roundup(last_page_size, bdev_logical_block_size(bdev));
 243		opt_size = optimal_io_size(bdev, last_page_size, size);
 244	}
 245
 246	sboff = rdev->sb_start + offset;
 247	doff = rdev->data_offset;
 248
 249	/* Just make sure we aren't corrupting data or metadata */
 250	if (mddev->external) {
 251		/* Bitmap could be anywhere. */
 252		if (sboff + ps > doff &&
 253		    sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE))
 254			return -EINVAL;
 255	} else if (offset < 0) {
 256		/* DATA  BITMAP METADATA  */
 257		size = bitmap_io_size(size, opt_size, offset + ps, 0);
 258		if (size == 0)
 259			/* bitmap runs in to metadata */
 260			return -EINVAL;
 261
 262		if (doff + mddev->dev_sectors > sboff)
 263			/* data runs in to bitmap */
 264			return -EINVAL;
 265	} else if (rdev->sb_start < rdev->data_offset) {
 266		/* METADATA BITMAP DATA */
 267		size = bitmap_io_size(size, opt_size, sboff + ps, doff);
 268		if (size == 0)
 269			/* bitmap runs in to data */
 270			return -EINVAL;
 271	} else {
 272		/* DATA METADATA BITMAP - no problems */
 
 
 
 
 
 
 
 
 
 
 
 
 
 273	}
 274
 275	md_super_write(mddev, rdev, sboff + ps, (int) size, page);
 
 276	return 0;
 
 
 
 277}
 278
 279static void write_sb_page(struct bitmap *bitmap, unsigned long pg_index,
 280			  struct page *page, bool wait)
 
 
 
 281{
 282	struct mddev *mddev = bitmap->mddev;
 283
 284	do {
 285		struct md_rdev *rdev = NULL;
 286
 287		while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 288			if (__write_sb_page(rdev, bitmap, pg_index, page) < 0) {
 289				set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 290				return;
 291			}
 292		}
 293	} while (wait && md_super_wait(mddev) < 0);
 294}
 295
 296static void md_bitmap_file_kick(struct bitmap *bitmap);
 297
 298#ifdef CONFIG_MD_BITMAP_FILE
 299static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 300{
 301	struct buffer_head *bh = page_buffers(page);
 
 
 
 302
 303	while (bh && bh->b_blocknr) {
 304		atomic_inc(&bitmap->pending_writes);
 305		set_buffer_locked(bh);
 306		set_buffer_mapped(bh);
 307		submit_bh(REQ_OP_WRITE | REQ_SYNC, bh);
 308		bh = bh->b_this_page;
 309	}
 310
 311	if (wait)
 312		wait_event(bitmap->write_wait,
 313			   atomic_read(&bitmap->pending_writes) == 0);
 314}
 315
 316static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 317{
 318	struct bitmap *bitmap = bh->b_private;
 319
 320	if (!uptodate)
 321		set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 322	if (atomic_dec_and_test(&bitmap->pending_writes))
 323		wake_up(&bitmap->write_wait);
 324}
 325
 326static void free_buffers(struct page *page)
 327{
 328	struct buffer_head *bh;
 329
 330	if (!PagePrivate(page))
 331		return;
 332
 333	bh = page_buffers(page);
 334	while (bh) {
 335		struct buffer_head *next = bh->b_this_page;
 336		free_buffer_head(bh);
 337		bh = next;
 338	}
 339	detach_page_private(page);
 340	put_page(page);
 341}
 342
 343/* read a page from a file.
 344 * We both read the page, and attach buffers to the page to record the
 345 * address of each block (using bmap).  These addresses will be used
 346 * to write the block later, completely bypassing the filesystem.
 347 * This usage is similar to how swap files are handled, and allows us
 348 * to write to a file with no concerns of memory allocation failing.
 349 */
 350static int read_file_page(struct file *file, unsigned long index,
 351		struct bitmap *bitmap, unsigned long count, struct page *page)
 
 
 352{
 353	int ret = 0;
 354	struct inode *inode = file_inode(file);
 355	struct buffer_head *bh;
 356	sector_t block, blk_cur;
 357	unsigned long blocksize = i_blocksize(inode);
 358
 359	pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 360		 (unsigned long long)index << PAGE_SHIFT);
 361
 362	bh = alloc_page_buffers(page, blocksize, false);
 363	if (!bh) {
 364		ret = -ENOMEM;
 365		goto out;
 366	}
 367	attach_page_private(page, bh);
 368	blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 369	while (bh) {
 370		block = blk_cur;
 371
 372		if (count == 0)
 373			bh->b_blocknr = 0;
 374		else {
 375			ret = bmap(inode, &block);
 376			if (ret || !block) {
 377				ret = -EINVAL;
 378				bh->b_blocknr = 0;
 379				goto out;
 380			}
 381
 382			bh->b_blocknr = block;
 383			bh->b_bdev = inode->i_sb->s_bdev;
 384			if (count < blocksize)
 385				count = 0;
 386			else
 387				count -= blocksize;
 388
 389			bh->b_end_io = end_bitmap_write;
 390			bh->b_private = bitmap;
 391			atomic_inc(&bitmap->pending_writes);
 392			set_buffer_locked(bh);
 393			set_buffer_mapped(bh);
 394			submit_bh(REQ_OP_READ, bh);
 395		}
 396		blk_cur++;
 397		bh = bh->b_this_page;
 398	}
 
 399
 400	wait_event(bitmap->write_wait,
 401		   atomic_read(&bitmap->pending_writes)==0);
 402	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 403		ret = -EIO;
 404out:
 405	if (ret)
 406		pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 407		       (int)PAGE_SIZE,
 408		       (unsigned long long)index << PAGE_SHIFT,
 409		       ret);
 410	return ret;
 411}
 412#else /* CONFIG_MD_BITMAP_FILE */
 413static void write_file_page(struct bitmap *bitmap, struct page *page, int wait)
 414{
 415}
 416static int read_file_page(struct file *file, unsigned long index,
 417		struct bitmap *bitmap, unsigned long count, struct page *page)
 418{
 419	return -EIO;
 420}
 421static void free_buffers(struct page *page)
 422{
 423	put_page(page);
 424}
 425#endif /* CONFIG_MD_BITMAP_FILE */
 426
 427/*
 428 * bitmap file superblock operations
 429 */
 430
 431/*
 432 * write out a page to a file
 433 */
 434static void filemap_write_page(struct bitmap *bitmap, unsigned long pg_index,
 435			       bool wait)
 436{
 437	struct bitmap_storage *store = &bitmap->storage;
 438	struct page *page = store->filemap[pg_index];
 439
 440	if (mddev_is_clustered(bitmap->mddev)) {
 441		pg_index += bitmap->cluster_slot *
 442			DIV_ROUND_UP(store->bytes, PAGE_SIZE);
 443	}
 444
 445	if (store->file)
 446		write_file_page(bitmap, page, wait);
 447	else
 448		write_sb_page(bitmap, pg_index, page, wait);
 449}
 450
 451/*
 452 * md_bitmap_wait_writes() should be called before writing any bitmap
 453 * blocks, to ensure previous writes, particularly from
 454 * md_bitmap_daemon_work(), have completed.
 455 */
 456static void md_bitmap_wait_writes(struct bitmap *bitmap)
 457{
 458	if (bitmap->storage.file)
 459		wait_event(bitmap->write_wait,
 460			   atomic_read(&bitmap->pending_writes)==0);
 461	else
 462		/* Note that we ignore the return value.  The writes
 463		 * might have failed, but that would just mean that
 464		 * some bits which should be cleared haven't been,
 465		 * which is safe.  The relevant bitmap blocks will
 466		 * probably get written again, but there is no great
 467		 * loss if they aren't.
 468		 */
 469		md_super_wait(bitmap->mddev);
 470}
 471
 472
 473/* update the event counter and sync the superblock to disk */
 474void md_bitmap_update_sb(struct bitmap *bitmap)
 475{
 476	bitmap_super_t *sb;
 477
 478	if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 479		return;
 480	if (bitmap->mddev->bitmap_info.external)
 481		return;
 482	if (!bitmap->storage.sb_page) /* no superblock */
 483		return;
 484	sb = kmap_atomic(bitmap->storage.sb_page);
 485	sb->events = cpu_to_le64(bitmap->mddev->events);
 486	if (bitmap->mddev->events < bitmap->events_cleared)
 487		/* rocking back to read-only */
 488		bitmap->events_cleared = bitmap->mddev->events;
 489	sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 490	/*
 491	 * clear BITMAP_WRITE_ERROR bit to protect against the case that
 492	 * a bitmap write error occurred but the later writes succeeded.
 493	 */
 494	sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 495	/* Just in case these have been changed via sysfs: */
 496	sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 497	sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 498	/* This might have been changed by a reshape */
 499	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 500	sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 501	sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 502	sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 503					   bitmap_info.space);
 504	kunmap_atomic(sb);
 505
 506	if (bitmap->storage.file)
 507		write_file_page(bitmap, bitmap->storage.sb_page, 1);
 508	else
 509		write_sb_page(bitmap, bitmap->storage.sb_index,
 510			      bitmap->storage.sb_page, 1);
 511}
 512EXPORT_SYMBOL(md_bitmap_update_sb);
 513
 514/* print out the bitmap file superblock */
 515void md_bitmap_print_sb(struct bitmap *bitmap)
 516{
 517	bitmap_super_t *sb;
 518
 519	if (!bitmap || !bitmap->storage.sb_page)
 520		return;
 521	sb = kmap_atomic(bitmap->storage.sb_page);
 522	pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 523	pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 524	pr_debug("       version: %u\n", le32_to_cpu(sb->version));
 525	pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 526		 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 527		 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 528		 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 529		 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 530	pr_debug("        events: %llu\n",
 531		 (unsigned long long) le64_to_cpu(sb->events));
 532	pr_debug("events cleared: %llu\n",
 533		 (unsigned long long) le64_to_cpu(sb->events_cleared));
 534	pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 535	pr_debug("     chunksize: %u B\n", le32_to_cpu(sb->chunksize));
 536	pr_debug("  daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep));
 537	pr_debug("     sync size: %llu KB\n",
 538		 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 539	pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind));
 540	kunmap_atomic(sb);
 541}
 542
 543/*
 544 * bitmap_new_disk_sb
 545 * @bitmap
 546 *
 547 * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 548 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 549 * This function verifies 'bitmap_info' and populates the on-disk bitmap
 550 * structure, which is to be written to disk.
 551 *
 552 * Returns: 0 on success, -Exxx on error
 553 */
 554static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 555{
 556	bitmap_super_t *sb;
 557	unsigned long chunksize, daemon_sleep, write_behind;
 558
 559	bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 560	if (bitmap->storage.sb_page == NULL)
 561		return -ENOMEM;
 562	bitmap->storage.sb_index = 0;
 563
 564	sb = kmap_atomic(bitmap->storage.sb_page);
 565
 566	sb->magic = cpu_to_le32(BITMAP_MAGIC);
 567	sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 568
 569	chunksize = bitmap->mddev->bitmap_info.chunksize;
 570	BUG_ON(!chunksize);
 571	if (!is_power_of_2(chunksize)) {
 572		kunmap_atomic(sb);
 573		pr_warn("bitmap chunksize not a power of 2\n");
 574		return -EINVAL;
 575	}
 576	sb->chunksize = cpu_to_le32(chunksize);
 577
 578	daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 579	if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 580		pr_debug("Choosing daemon_sleep default (5 sec)\n");
 581		daemon_sleep = 5 * HZ;
 582	}
 583	sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 584	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 585
 586	/*
 587	 * FIXME: write_behind for RAID1.  If not specified, what
 588	 * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 589	 */
 590	write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 591	if (write_behind > COUNTER_MAX)
 592		write_behind = COUNTER_MAX / 2;
 593	sb->write_behind = cpu_to_le32(write_behind);
 594	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 595
 596	/* keep the array size field of the bitmap superblock up to date */
 597	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 598
 599	memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 600
 601	set_bit(BITMAP_STALE, &bitmap->flags);
 602	sb->state = cpu_to_le32(bitmap->flags);
 603	bitmap->events_cleared = bitmap->mddev->events;
 604	sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 605	bitmap->mddev->bitmap_info.nodes = 0;
 606
 607	kunmap_atomic(sb);
 608
 609	return 0;
 610}
 611
 612/* read the superblock from the bitmap file and initialize some bitmap fields */
 613static int md_bitmap_read_sb(struct bitmap *bitmap)
 614{
 615	char *reason = NULL;
 616	bitmap_super_t *sb;
 617	unsigned long chunksize, daemon_sleep, write_behind;
 618	unsigned long long events;
 619	int nodes = 0;
 620	unsigned long sectors_reserved = 0;
 621	int err = -EINVAL;
 622	struct page *sb_page;
 623	loff_t offset = 0;
 624
 625	if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 626		chunksize = 128 * 1024 * 1024;
 627		daemon_sleep = 5 * HZ;
 628		write_behind = 0;
 629		set_bit(BITMAP_STALE, &bitmap->flags);
 630		err = 0;
 631		goto out_no_sb;
 632	}
 633	/* page 0 is the superblock, read it... */
 634	sb_page = alloc_page(GFP_KERNEL);
 635	if (!sb_page)
 636		return -ENOMEM;
 637	bitmap->storage.sb_page = sb_page;
 638
 639re_read:
 640	/* If cluster_slot is set, the cluster is setup */
 641	if (bitmap->cluster_slot >= 0) {
 642		sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 643
 644		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 645			   (bitmap->mddev->bitmap_info.chunksize >> 9));
 646		/* bits to bytes */
 647		bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 648		/* to 4k blocks */
 649		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 650		offset = bitmap->cluster_slot * (bm_blocks << 3);
 651		pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 652			bitmap->cluster_slot, offset);
 653	}
 654
 655	if (bitmap->storage.file) {
 656		loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 657		int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 658
 659		err = read_file_page(bitmap->storage.file, 0,
 660				bitmap, bytes, sb_page);
 661	} else {
 662		err = read_sb_page(bitmap->mddev, offset, sb_page, 0,
 663				   sizeof(bitmap_super_t));
 
 
 664	}
 665	if (err)
 666		return err;
 667
 668	err = -EINVAL;
 669	sb = kmap_atomic(sb_page);
 670
 671	chunksize = le32_to_cpu(sb->chunksize);
 672	daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 673	write_behind = le32_to_cpu(sb->write_behind);
 674	sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 
 
 
 
 
 
 
 
 675
 676	/* verify that the bitmap-specific fields are valid */
 677	if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 678		reason = "bad magic";
 679	else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 680		 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 681		reason = "unrecognized superblock version";
 682	else if (chunksize < 512)
 683		reason = "bitmap chunksize too small";
 684	else if (!is_power_of_2(chunksize))
 685		reason = "bitmap chunksize not a power of 2";
 686	else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 687		reason = "daemon sleep period out of range";
 688	else if (write_behind > COUNTER_MAX)
 689		reason = "write-behind limit out of range (0 - 16383)";
 690	if (reason) {
 691		pr_warn("%s: invalid bitmap file superblock: %s\n",
 692			bmname(bitmap), reason);
 693		goto out;
 694	}
 695
 696	/*
 697	 * Setup nodes/clustername only if bitmap version is
 698	 * cluster-compatible
 699	 */
 700	if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 701		nodes = le32_to_cpu(sb->nodes);
 702		strscpy(bitmap->mddev->bitmap_info.cluster_name,
 703				sb->cluster_name, 64);
 704	}
 705
 706	/* keep the array size field of the bitmap superblock up to date */
 707	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 708
 709	if (bitmap->mddev->persistent) {
 710		/*
 711		 * We have a persistent array superblock, so compare the
 712		 * bitmap's UUID and event counter to the mddev's
 713		 */
 714		if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 715			pr_warn("%s: bitmap superblock UUID mismatch\n",
 716				bmname(bitmap));
 717			goto out;
 718		}
 719		events = le64_to_cpu(sb->events);
 720		if (!nodes && (events < bitmap->mddev->events)) {
 721			pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 722				bmname(bitmap), events,
 723				(unsigned long long) bitmap->mddev->events);
 724			set_bit(BITMAP_STALE, &bitmap->flags);
 725		}
 726	}
 727
 728	/* assign fields using values from superblock */
 729	bitmap->flags |= le32_to_cpu(sb->state);
 730	if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 731		set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 732	bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 
 733	err = 0;
 734
 735out:
 736	kunmap_atomic(sb);
 
 
 737	if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 738		/* Assigning chunksize is required for "re_read" */
 739		bitmap->mddev->bitmap_info.chunksize = chunksize;
 740		err = md_setup_cluster(bitmap->mddev, nodes);
 741		if (err) {
 742			pr_warn("%s: Could not setup cluster service (%d)\n",
 743				bmname(bitmap), err);
 744			goto out_no_sb;
 745		}
 746		bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 747		goto re_read;
 748	}
 749
 
 750out_no_sb:
 751	if (err == 0) {
 752		if (test_bit(BITMAP_STALE, &bitmap->flags))
 753			bitmap->events_cleared = bitmap->mddev->events;
 754		bitmap->mddev->bitmap_info.chunksize = chunksize;
 755		bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 756		bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 757		bitmap->mddev->bitmap_info.nodes = nodes;
 758		if (bitmap->mddev->bitmap_info.space == 0 ||
 759			bitmap->mddev->bitmap_info.space > sectors_reserved)
 760			bitmap->mddev->bitmap_info.space = sectors_reserved;
 761	} else {
 762		md_bitmap_print_sb(bitmap);
 763		if (bitmap->cluster_slot < 0)
 764			md_cluster_stop(bitmap->mddev);
 765	}
 766	return err;
 767}
 768
 769/*
 770 * general bitmap file operations
 771 */
 772
 773/*
 774 * on-disk bitmap:
 775 *
 776 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 777 * file a page at a time. There's a superblock at the start of the file.
 778 */
 779/* calculate the index of the page that contains this bit */
 780static inline unsigned long file_page_index(struct bitmap_storage *store,
 781					    unsigned long chunk)
 782{
 783	if (store->sb_page)
 784		chunk += sizeof(bitmap_super_t) << 3;
 785	return chunk >> PAGE_BIT_SHIFT;
 786}
 787
 788/* calculate the (bit) offset of this bit within a page */
 789static inline unsigned long file_page_offset(struct bitmap_storage *store,
 790					     unsigned long chunk)
 791{
 792	if (store->sb_page)
 793		chunk += sizeof(bitmap_super_t) << 3;
 794	return chunk & (PAGE_BITS - 1);
 795}
 796
 797/*
 798 * return a pointer to the page in the filemap that contains the given bit
 799 *
 800 */
 801static inline struct page *filemap_get_page(struct bitmap_storage *store,
 802					    unsigned long chunk)
 803{
 804	if (file_page_index(store, chunk) >= store->file_pages)
 805		return NULL;
 806	return store->filemap[file_page_index(store, chunk)];
 807}
 808
 809static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 810				   unsigned long chunks, int with_super,
 811				   int slot_number)
 812{
 813	int pnum, offset = 0;
 814	unsigned long num_pages;
 815	unsigned long bytes;
 816
 817	bytes = DIV_ROUND_UP(chunks, 8);
 818	if (with_super)
 819		bytes += sizeof(bitmap_super_t);
 820
 821	num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 822	offset = slot_number * num_pages;
 823
 824	store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 825				       GFP_KERNEL);
 826	if (!store->filemap)
 827		return -ENOMEM;
 828
 829	if (with_super && !store->sb_page) {
 830		store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 831		if (store->sb_page == NULL)
 832			return -ENOMEM;
 833	}
 834
 835	pnum = 0;
 836	if (store->sb_page) {
 837		store->filemap[0] = store->sb_page;
 838		pnum = 1;
 839		store->sb_index = offset;
 840	}
 841
 842	for ( ; pnum < num_pages; pnum++) {
 843		store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 844		if (!store->filemap[pnum]) {
 845			store->file_pages = pnum;
 846			return -ENOMEM;
 847		}
 
 848	}
 849	store->file_pages = pnum;
 850
 851	/* We need 4 bits per page, rounded up to a multiple
 852	 * of sizeof(unsigned long) */
 853	store->filemap_attr = kzalloc(
 854		roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 855		GFP_KERNEL);
 856	if (!store->filemap_attr)
 857		return -ENOMEM;
 858
 859	store->bytes = bytes;
 860
 861	return 0;
 862}
 863
 864static void md_bitmap_file_unmap(struct bitmap_storage *store)
 865{
 866	struct file *file = store->file;
 867	struct page *sb_page = store->sb_page;
 868	struct page **map = store->filemap;
 869	int pages = store->file_pages;
 
 
 
 
 870
 871	while (pages--)
 872		if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 873			free_buffers(map[pages]);
 874	kfree(map);
 875	kfree(store->filemap_attr);
 876
 877	if (sb_page)
 878		free_buffers(sb_page);
 879
 880	if (file) {
 881		struct inode *inode = file_inode(file);
 882		invalidate_mapping_pages(inode->i_mapping, 0, -1);
 883		fput(file);
 884	}
 885}
 886
 887/*
 888 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 889 * then it is no longer reliable, so we stop using it and we mark the file
 890 * as failed in the superblock
 891 */
 892static void md_bitmap_file_kick(struct bitmap *bitmap)
 893{
 
 
 894	if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 895		md_bitmap_update_sb(bitmap);
 896
 897		if (bitmap->storage.file) {
 898			pr_warn("%s: kicking failed bitmap file %pD4 from array!\n",
 899				bmname(bitmap), bitmap->storage.file);
 
 
 
 
 
 900
 
 901		} else
 902			pr_warn("%s: disabling internal bitmap due to errors\n",
 903				bmname(bitmap));
 904	}
 905}
 906
 907enum bitmap_page_attr {
 908	BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 909	BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 910				    * i.e. counter is 1 or 2. */
 911	BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 912};
 913
 914static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 915				 enum bitmap_page_attr attr)
 916{
 917	set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 918}
 919
 920static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 921				   enum bitmap_page_attr attr)
 922{
 923	clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 924}
 925
 926static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 927				 enum bitmap_page_attr attr)
 928{
 929	return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 930}
 931
 932static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 933					   enum bitmap_page_attr attr)
 934{
 935	return test_and_clear_bit((pnum<<2) + attr,
 936				  bitmap->storage.filemap_attr);
 937}
 938/*
 939 * bitmap_file_set_bit -- called before performing a write to the md device
 940 * to set (and eventually sync) a particular bit in the bitmap file
 941 *
 942 * we set the bit immediately, then we record the page number so that
 943 * when an unplug occurs, we can flush the dirty pages out to disk
 944 */
 945static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 946{
 947	unsigned long bit;
 948	struct page *page;
 949	void *kaddr;
 950	unsigned long chunk = block >> bitmap->counts.chunkshift;
 951	struct bitmap_storage *store = &bitmap->storage;
 952	unsigned long index = file_page_index(store, chunk);
 953	unsigned long node_offset = 0;
 954
 955	if (mddev_is_clustered(bitmap->mddev))
 956		node_offset = bitmap->cluster_slot * store->file_pages;
 957
 958	page = filemap_get_page(&bitmap->storage, chunk);
 959	if (!page)
 960		return;
 961	bit = file_page_offset(&bitmap->storage, chunk);
 962
 963	/* set the bit */
 964	kaddr = kmap_atomic(page);
 965	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 966		set_bit(bit, kaddr);
 967	else
 968		set_bit_le(bit, kaddr);
 969	kunmap_atomic(kaddr);
 970	pr_debug("set file bit %lu page %lu\n", bit, index);
 971	/* record page number so it gets flushed to disk when unplug occurs */
 972	set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_DIRTY);
 973}
 974
 975static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 976{
 977	unsigned long bit;
 978	struct page *page;
 979	void *paddr;
 980	unsigned long chunk = block >> bitmap->counts.chunkshift;
 981	struct bitmap_storage *store = &bitmap->storage;
 982	unsigned long index = file_page_index(store, chunk);
 983	unsigned long node_offset = 0;
 984
 985	if (mddev_is_clustered(bitmap->mddev))
 986		node_offset = bitmap->cluster_slot * store->file_pages;
 987
 988	page = filemap_get_page(&bitmap->storage, chunk);
 989	if (!page)
 990		return;
 991	bit = file_page_offset(&bitmap->storage, chunk);
 992	paddr = kmap_atomic(page);
 993	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 994		clear_bit(bit, paddr);
 995	else
 996		clear_bit_le(bit, paddr);
 997	kunmap_atomic(paddr);
 998	if (!test_page_attr(bitmap, index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
 999		set_page_attr(bitmap, index - node_offset, BITMAP_PAGE_PENDING);
1000		bitmap->allclean = 0;
1001	}
1002}
1003
1004static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
1005{
1006	unsigned long bit;
1007	struct page *page;
1008	void *paddr;
1009	unsigned long chunk = block >> bitmap->counts.chunkshift;
1010	int set = 0;
1011
1012	page = filemap_get_page(&bitmap->storage, chunk);
1013	if (!page)
1014		return -EINVAL;
1015	bit = file_page_offset(&bitmap->storage, chunk);
1016	paddr = kmap_atomic(page);
1017	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1018		set = test_bit(bit, paddr);
1019	else
1020		set = test_bit_le(bit, paddr);
1021	kunmap_atomic(paddr);
1022	return set;
1023}
1024
 
1025/* this gets called when the md device is ready to unplug its underlying
1026 * (slave) device queues -- before we let any writes go down, we need to
1027 * sync the dirty pages of the bitmap file to disk */
1028void md_bitmap_unplug(struct bitmap *bitmap)
1029{
1030	unsigned long i;
1031	int dirty, need_write;
1032	int writing = 0;
1033
1034	if (!md_bitmap_enabled(bitmap))
 
1035		return;
1036
1037	/* look at each page to see if there are any set bits that need to be
1038	 * flushed out to disk */
1039	for (i = 0; i < bitmap->storage.file_pages; i++) {
1040		dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1041		need_write = test_and_clear_page_attr(bitmap, i,
1042						      BITMAP_PAGE_NEEDWRITE);
1043		if (dirty || need_write) {
1044			if (!writing) {
1045				md_bitmap_wait_writes(bitmap);
1046				if (bitmap->mddev->queue)
1047					blk_add_trace_msg(bitmap->mddev->queue,
1048							  "md bitmap_unplug");
1049			}
1050			clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1051			filemap_write_page(bitmap, i, false);
1052			writing = 1;
1053		}
1054	}
1055	if (writing)
1056		md_bitmap_wait_writes(bitmap);
1057
1058	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1059		md_bitmap_file_kick(bitmap);
1060}
1061EXPORT_SYMBOL(md_bitmap_unplug);
1062
1063struct bitmap_unplug_work {
1064	struct work_struct work;
1065	struct bitmap *bitmap;
1066	struct completion *done;
1067};
1068
1069static void md_bitmap_unplug_fn(struct work_struct *work)
1070{
1071	struct bitmap_unplug_work *unplug_work =
1072		container_of(work, struct bitmap_unplug_work, work);
1073
1074	md_bitmap_unplug(unplug_work->bitmap);
1075	complete(unplug_work->done);
1076}
1077
1078void md_bitmap_unplug_async(struct bitmap *bitmap)
1079{
1080	DECLARE_COMPLETION_ONSTACK(done);
1081	struct bitmap_unplug_work unplug_work;
1082
1083	INIT_WORK_ONSTACK(&unplug_work.work, md_bitmap_unplug_fn);
1084	unplug_work.bitmap = bitmap;
1085	unplug_work.done = &done;
1086
1087	queue_work(md_bitmap_wq, &unplug_work.work);
1088	wait_for_completion(&done);
1089}
1090EXPORT_SYMBOL(md_bitmap_unplug_async);
1091
1092static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1093
1094/*
1095 * Initialize the in-memory bitmap from the on-disk bitmap and set up the memory
1096 * mapping of the bitmap file.
1097 *
1098 * Special case: If there's no bitmap file, or if the bitmap file had been
1099 * previously kicked from the array, we mark all the bits as 1's in order to
1100 * cause a full resync.
1101 *
1102 * We ignore all bits for sectors that end earlier than 'start'.
1103 * This is used when reading an out-of-date bitmap.
1104 */
1105static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1106{
1107	bool outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1108	struct mddev *mddev = bitmap->mddev;
1109	unsigned long chunks = bitmap->counts.chunks;
1110	struct bitmap_storage *store = &bitmap->storage;
1111	struct file *file = store->file;
1112	unsigned long node_offset = 0;
1113	unsigned long bit_cnt = 0;
1114	unsigned long i;
1115	int ret;
 
 
 
 
1116
1117	if (!file && !mddev->bitmap_info.offset) {
 
 
 
1118		/* No permanent bitmap - fill with '1s'. */
1119		store->filemap = NULL;
1120		store->file_pages = 0;
1121		for (i = 0; i < chunks ; i++) {
1122			/* if the disk bit is set, set the memory bit */
1123			int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1124				      >= start);
1125			md_bitmap_set_memory_bits(bitmap,
1126						  (sector_t)i << bitmap->counts.chunkshift,
1127						  needed);
1128		}
1129		return 0;
1130	}
1131
 
 
 
 
1132	if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1133		pr_warn("%s: bitmap file too short %lu < %lu\n",
1134			bmname(bitmap),
1135			(unsigned long) i_size_read(file->f_mapping->host),
1136			store->bytes);
1137		ret = -ENOSPC;
1138		goto err;
1139	}
1140
1141	if (mddev_is_clustered(mddev))
 
 
 
 
 
1142		node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1143
1144	for (i = 0; i < store->file_pages; i++) {
1145		struct page *page = store->filemap[i];
1146		int count;
1147
1148		/* unmap the old page, we're done with it */
1149		if (i == store->file_pages - 1)
1150			count = store->bytes - i * PAGE_SIZE;
1151		else
1152			count = PAGE_SIZE;
 
 
 
 
 
 
 
 
 
 
 
 
1153
1154		if (file)
1155			ret = read_file_page(file, i, bitmap, count, page);
1156		else
1157			ret = read_sb_page(mddev, 0, page, i + node_offset,
1158					   count);
1159		if (ret)
1160			goto err;
1161	}
1162
1163	if (outofdate) {
1164		pr_warn("%s: bitmap file is out of date, doing full recovery\n",
1165			bmname(bitmap));
1166
1167		for (i = 0; i < store->file_pages; i++) {
1168			struct page *page = store->filemap[i];
1169			unsigned long offset = 0;
1170			void *paddr;
1171
1172			if (i == 0 && !mddev->bitmap_info.external)
1173				offset = sizeof(bitmap_super_t);
1174
1175			/*
1176			 * If the bitmap is out of date, dirty the whole page
1177			 * and write it out
1178			 */
1179			paddr = kmap_atomic(page);
1180			memset(paddr + offset, 0xff, PAGE_SIZE - offset);
1181			kunmap_atomic(paddr);
1182
1183			filemap_write_page(bitmap, i, true);
1184			if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) {
1185				ret = -EIO;
1186				goto err;
 
 
1187			}
1188		}
1189	}
1190
1191	for (i = 0; i < chunks; i++) {
1192		struct page *page = filemap_get_page(&bitmap->storage, i);
1193		unsigned long bit = file_page_offset(&bitmap->storage, i);
1194		void *paddr;
1195		bool was_set;
1196
1197		paddr = kmap_atomic(page);
1198		if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1199			was_set = test_bit(bit, paddr);
1200		else
1201			was_set = test_bit_le(bit, paddr);
1202		kunmap_atomic(paddr);
1203
1204		if (was_set) {
1205			/* if the disk bit is set, set the memory bit */
1206			int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1207				      >= start);
1208			md_bitmap_set_memory_bits(bitmap,
1209						  (sector_t)i << bitmap->counts.chunkshift,
1210						  needed);
1211			bit_cnt++;
1212		}
 
1213	}
1214
1215	pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1216		 bmname(bitmap), store->file_pages,
1217		 bit_cnt, chunks);
1218
1219	return 0;
1220
1221 err:
1222	pr_warn("%s: bitmap initialisation failed: %d\n",
1223		bmname(bitmap), ret);
1224	return ret;
1225}
1226
1227void md_bitmap_write_all(struct bitmap *bitmap)
1228{
1229	/* We don't actually write all bitmap blocks here,
1230	 * just flag them as needing to be written
1231	 */
1232	int i;
1233
1234	if (!bitmap || !bitmap->storage.filemap)
1235		return;
1236	if (bitmap->storage.file)
1237		/* Only one copy, so nothing needed */
1238		return;
1239
1240	for (i = 0; i < bitmap->storage.file_pages; i++)
1241		set_page_attr(bitmap, i,
1242			      BITMAP_PAGE_NEEDWRITE);
1243	bitmap->allclean = 0;
1244}
1245
1246static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1247				 sector_t offset, int inc)
1248{
1249	sector_t chunk = offset >> bitmap->chunkshift;
1250	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1251	bitmap->bp[page].count += inc;
1252	md_bitmap_checkfree(bitmap, page);
1253}
1254
1255static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1256{
1257	sector_t chunk = offset >> bitmap->chunkshift;
1258	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1259	struct bitmap_page *bp = &bitmap->bp[page];
1260
1261	if (!bp->pending)
1262		bp->pending = 1;
1263}
1264
1265static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1266					       sector_t offset, sector_t *blocks,
1267					       int create);
1268
1269static void mddev_set_timeout(struct mddev *mddev, unsigned long timeout,
1270			      bool force)
1271{
1272	struct md_thread *thread;
1273
1274	rcu_read_lock();
1275	thread = rcu_dereference(mddev->thread);
1276
1277	if (!thread)
1278		goto out;
1279
1280	if (force || thread->timeout < MAX_SCHEDULE_TIMEOUT)
1281		thread->timeout = timeout;
1282
1283out:
1284	rcu_read_unlock();
1285}
1286
1287/*
1288 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1289 *			out to disk
1290 */
 
1291void md_bitmap_daemon_work(struct mddev *mddev)
1292{
1293	struct bitmap *bitmap;
1294	unsigned long j;
1295	unsigned long nextpage;
1296	sector_t blocks;
1297	struct bitmap_counts *counts;
1298
1299	/* Use a mutex to guard daemon_work against
1300	 * bitmap_destroy.
1301	 */
1302	mutex_lock(&mddev->bitmap_info.mutex);
1303	bitmap = mddev->bitmap;
1304	if (bitmap == NULL) {
1305		mutex_unlock(&mddev->bitmap_info.mutex);
1306		return;
1307	}
1308	if (time_before(jiffies, bitmap->daemon_lastrun
1309			+ mddev->bitmap_info.daemon_sleep))
1310		goto done;
1311
1312	bitmap->daemon_lastrun = jiffies;
1313	if (bitmap->allclean) {
1314		mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
1315		goto done;
1316	}
1317	bitmap->allclean = 1;
1318
1319	if (bitmap->mddev->queue)
1320		blk_add_trace_msg(bitmap->mddev->queue,
1321				  "md bitmap_daemon_work");
1322
1323	/* Any file-page which is PENDING now needs to be written.
1324	 * So set NEEDWRITE now, then after we make any last-minute changes
1325	 * we will write it.
1326	 */
1327	for (j = 0; j < bitmap->storage.file_pages; j++)
1328		if (test_and_clear_page_attr(bitmap, j,
1329					     BITMAP_PAGE_PENDING))
1330			set_page_attr(bitmap, j,
1331				      BITMAP_PAGE_NEEDWRITE);
1332
1333	if (bitmap->need_sync &&
1334	    mddev->bitmap_info.external == 0) {
1335		/* Arrange for superblock update as well as
1336		 * other changes */
1337		bitmap_super_t *sb;
1338		bitmap->need_sync = 0;
1339		if (bitmap->storage.filemap) {
1340			sb = kmap_atomic(bitmap->storage.sb_page);
1341			sb->events_cleared =
1342				cpu_to_le64(bitmap->events_cleared);
1343			kunmap_atomic(sb);
1344			set_page_attr(bitmap, 0,
1345				      BITMAP_PAGE_NEEDWRITE);
1346		}
1347	}
1348	/* Now look at the bitmap counters and if any are '2' or '1',
1349	 * decrement and handle accordingly.
1350	 */
1351	counts = &bitmap->counts;
1352	spin_lock_irq(&counts->lock);
1353	nextpage = 0;
1354	for (j = 0; j < counts->chunks; j++) {
1355		bitmap_counter_t *bmc;
1356		sector_t  block = (sector_t)j << counts->chunkshift;
1357
1358		if (j == nextpage) {
1359			nextpage += PAGE_COUNTER_RATIO;
1360			if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1361				j |= PAGE_COUNTER_MASK;
1362				continue;
1363			}
1364			counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1365		}
1366
1367		bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1368		if (!bmc) {
1369			j |= PAGE_COUNTER_MASK;
1370			continue;
1371		}
1372		if (*bmc == 1 && !bitmap->need_sync) {
1373			/* We can clear the bit */
1374			*bmc = 0;
1375			md_bitmap_count_page(counts, block, -1);
1376			md_bitmap_file_clear_bit(bitmap, block);
1377		} else if (*bmc && *bmc <= 2) {
1378			*bmc = 1;
1379			md_bitmap_set_pending(counts, block);
1380			bitmap->allclean = 0;
1381		}
1382	}
1383	spin_unlock_irq(&counts->lock);
1384
1385	md_bitmap_wait_writes(bitmap);
1386	/* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1387	 * DIRTY pages need to be written by bitmap_unplug so it can wait
1388	 * for them.
1389	 * If we find any DIRTY page we stop there and let bitmap_unplug
1390	 * handle all the rest.  This is important in the case where
1391	 * the first blocking holds the superblock and it has been updated.
1392	 * We mustn't write any other blocks before the superblock.
1393	 */
1394	for (j = 0;
1395	     j < bitmap->storage.file_pages
1396		     && !test_bit(BITMAP_STALE, &bitmap->flags);
1397	     j++) {
1398		if (test_page_attr(bitmap, j,
1399				   BITMAP_PAGE_DIRTY))
1400			/* bitmap_unplug will handle the rest */
1401			break;
1402		if (bitmap->storage.filemap &&
1403		    test_and_clear_page_attr(bitmap, j,
1404					     BITMAP_PAGE_NEEDWRITE))
1405			filemap_write_page(bitmap, j, false);
 
1406	}
1407
1408 done:
1409	if (bitmap->allclean == 0)
1410		mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
 
1411	mutex_unlock(&mddev->bitmap_info.mutex);
1412}
1413
1414static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1415					       sector_t offset, sector_t *blocks,
1416					       int create)
1417__releases(bitmap->lock)
1418__acquires(bitmap->lock)
1419{
1420	/* If 'create', we might release the lock and reclaim it.
1421	 * The lock must have been taken with interrupts enabled.
1422	 * If !create, we don't release the lock.
1423	 */
1424	sector_t chunk = offset >> bitmap->chunkshift;
1425	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1426	unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1427	sector_t csize;
1428	int err;
1429
1430	if (page >= bitmap->pages) {
1431		/*
1432		 * This can happen if bitmap_start_sync goes beyond
1433		 * End-of-device while looking for a whole page or
1434		 * user set a huge number to sysfs bitmap_set_bits.
1435		 */
1436		return NULL;
1437	}
1438	err = md_bitmap_checkpage(bitmap, page, create, 0);
1439
1440	if (bitmap->bp[page].hijacked ||
1441	    bitmap->bp[page].map == NULL)
1442		csize = ((sector_t)1) << (bitmap->chunkshift +
1443					  PAGE_COUNTER_SHIFT);
1444	else
1445		csize = ((sector_t)1) << bitmap->chunkshift;
1446	*blocks = csize - (offset & (csize - 1));
1447
1448	if (err < 0)
1449		return NULL;
1450
1451	/* now locked ... */
1452
1453	if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1454		/* should we use the first or second counter field
1455		 * of the hijacked pointer? */
1456		int hi = (pageoff > PAGE_COUNTER_MASK);
1457		return  &((bitmap_counter_t *)
1458			  &bitmap->bp[page].map)[hi];
1459	} else /* page is allocated */
1460		return (bitmap_counter_t *)
1461			&(bitmap->bp[page].map[pageoff]);
1462}
1463
1464int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1465{
1466	if (!bitmap)
1467		return 0;
1468
1469	if (behind) {
1470		int bw;
1471		atomic_inc(&bitmap->behind_writes);
1472		bw = atomic_read(&bitmap->behind_writes);
1473		if (bw > bitmap->behind_writes_used)
1474			bitmap->behind_writes_used = bw;
1475
1476		pr_debug("inc write-behind count %d/%lu\n",
1477			 bw, bitmap->mddev->bitmap_info.max_write_behind);
1478	}
1479
1480	while (sectors) {
1481		sector_t blocks;
1482		bitmap_counter_t *bmc;
1483
1484		spin_lock_irq(&bitmap->counts.lock);
1485		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1486		if (!bmc) {
1487			spin_unlock_irq(&bitmap->counts.lock);
1488			return 0;
1489		}
1490
1491		if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1492			DEFINE_WAIT(__wait);
1493			/* note that it is safe to do the prepare_to_wait
1494			 * after the test as long as we do it before dropping
1495			 * the spinlock.
1496			 */
1497			prepare_to_wait(&bitmap->overflow_wait, &__wait,
1498					TASK_UNINTERRUPTIBLE);
1499			spin_unlock_irq(&bitmap->counts.lock);
1500			schedule();
1501			finish_wait(&bitmap->overflow_wait, &__wait);
1502			continue;
1503		}
1504
1505		switch (*bmc) {
1506		case 0:
1507			md_bitmap_file_set_bit(bitmap, offset);
1508			md_bitmap_count_page(&bitmap->counts, offset, 1);
1509			fallthrough;
1510		case 1:
1511			*bmc = 2;
1512		}
1513
1514		(*bmc)++;
1515
1516		spin_unlock_irq(&bitmap->counts.lock);
1517
1518		offset += blocks;
1519		if (sectors > blocks)
1520			sectors -= blocks;
1521		else
1522			sectors = 0;
1523	}
1524	return 0;
1525}
1526EXPORT_SYMBOL(md_bitmap_startwrite);
1527
1528void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1529			unsigned long sectors, int success, int behind)
1530{
1531	if (!bitmap)
1532		return;
1533	if (behind) {
1534		if (atomic_dec_and_test(&bitmap->behind_writes))
1535			wake_up(&bitmap->behind_wait);
1536		pr_debug("dec write-behind count %d/%lu\n",
1537			 atomic_read(&bitmap->behind_writes),
1538			 bitmap->mddev->bitmap_info.max_write_behind);
1539	}
1540
1541	while (sectors) {
1542		sector_t blocks;
1543		unsigned long flags;
1544		bitmap_counter_t *bmc;
1545
1546		spin_lock_irqsave(&bitmap->counts.lock, flags);
1547		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1548		if (!bmc) {
1549			spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1550			return;
1551		}
1552
1553		if (success && !bitmap->mddev->degraded &&
1554		    bitmap->events_cleared < bitmap->mddev->events) {
1555			bitmap->events_cleared = bitmap->mddev->events;
1556			bitmap->need_sync = 1;
1557			sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1558		}
1559
1560		if (!success && !NEEDED(*bmc))
1561			*bmc |= NEEDED_MASK;
1562
1563		if (COUNTER(*bmc) == COUNTER_MAX)
1564			wake_up(&bitmap->overflow_wait);
1565
1566		(*bmc)--;
1567		if (*bmc <= 2) {
1568			md_bitmap_set_pending(&bitmap->counts, offset);
1569			bitmap->allclean = 0;
1570		}
1571		spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1572		offset += blocks;
1573		if (sectors > blocks)
1574			sectors -= blocks;
1575		else
1576			sectors = 0;
1577	}
1578}
1579EXPORT_SYMBOL(md_bitmap_endwrite);
1580
1581static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1582			       int degraded)
1583{
1584	bitmap_counter_t *bmc;
1585	int rv;
1586	if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1587		*blocks = 1024;
1588		return 1; /* always resync if no bitmap */
1589	}
1590	spin_lock_irq(&bitmap->counts.lock);
1591	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1592	rv = 0;
1593	if (bmc) {
1594		/* locked */
1595		if (RESYNC(*bmc))
1596			rv = 1;
1597		else if (NEEDED(*bmc)) {
1598			rv = 1;
1599			if (!degraded) { /* don't set/clear bits if degraded */
1600				*bmc |= RESYNC_MASK;
1601				*bmc &= ~NEEDED_MASK;
1602			}
1603		}
1604	}
1605	spin_unlock_irq(&bitmap->counts.lock);
1606	return rv;
1607}
1608
1609int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1610			 int degraded)
1611{
1612	/* bitmap_start_sync must always report on multiples of whole
1613	 * pages, otherwise resync (which is very PAGE_SIZE based) will
1614	 * get confused.
1615	 * So call __bitmap_start_sync repeatedly (if needed) until
1616	 * At least PAGE_SIZE>>9 blocks are covered.
1617	 * Return the 'or' of the result.
1618	 */
1619	int rv = 0;
1620	sector_t blocks1;
1621
1622	*blocks = 0;
1623	while (*blocks < (PAGE_SIZE>>9)) {
1624		rv |= __bitmap_start_sync(bitmap, offset,
1625					  &blocks1, degraded);
1626		offset += blocks1;
1627		*blocks += blocks1;
1628	}
1629	return rv;
1630}
1631EXPORT_SYMBOL(md_bitmap_start_sync);
1632
1633void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1634{
1635	bitmap_counter_t *bmc;
1636	unsigned long flags;
1637
1638	if (bitmap == NULL) {
1639		*blocks = 1024;
1640		return;
1641	}
1642	spin_lock_irqsave(&bitmap->counts.lock, flags);
1643	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1644	if (bmc == NULL)
1645		goto unlock;
1646	/* locked */
1647	if (RESYNC(*bmc)) {
1648		*bmc &= ~RESYNC_MASK;
1649
1650		if (!NEEDED(*bmc) && aborted)
1651			*bmc |= NEEDED_MASK;
1652		else {
1653			if (*bmc <= 2) {
1654				md_bitmap_set_pending(&bitmap->counts, offset);
1655				bitmap->allclean = 0;
1656			}
1657		}
1658	}
1659 unlock:
1660	spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1661}
1662EXPORT_SYMBOL(md_bitmap_end_sync);
1663
1664void md_bitmap_close_sync(struct bitmap *bitmap)
1665{
1666	/* Sync has finished, and any bitmap chunks that weren't synced
1667	 * properly have been aborted.  It remains to us to clear the
1668	 * RESYNC bit wherever it is still on
1669	 */
1670	sector_t sector = 0;
1671	sector_t blocks;
1672	if (!bitmap)
1673		return;
1674	while (sector < bitmap->mddev->resync_max_sectors) {
1675		md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1676		sector += blocks;
1677	}
1678}
1679EXPORT_SYMBOL(md_bitmap_close_sync);
1680
1681void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1682{
1683	sector_t s = 0;
1684	sector_t blocks;
1685
1686	if (!bitmap)
1687		return;
1688	if (sector == 0) {
1689		bitmap->last_end_sync = jiffies;
1690		return;
1691	}
1692	if (!force && time_before(jiffies, (bitmap->last_end_sync
1693				  + bitmap->mddev->bitmap_info.daemon_sleep)))
1694		return;
1695	wait_event(bitmap->mddev->recovery_wait,
1696		   atomic_read(&bitmap->mddev->recovery_active) == 0);
1697
1698	bitmap->mddev->curr_resync_completed = sector;
1699	set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1700	sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1701	s = 0;
1702	while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1703		md_bitmap_end_sync(bitmap, s, &blocks, 0);
1704		s += blocks;
1705	}
1706	bitmap->last_end_sync = jiffies;
1707	sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1708}
1709EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1710
1711void md_bitmap_sync_with_cluster(struct mddev *mddev,
1712			      sector_t old_lo, sector_t old_hi,
1713			      sector_t new_lo, sector_t new_hi)
1714{
1715	struct bitmap *bitmap = mddev->bitmap;
1716	sector_t sector, blocks = 0;
1717
1718	for (sector = old_lo; sector < new_lo; ) {
1719		md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1720		sector += blocks;
1721	}
1722	WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1723
1724	for (sector = old_hi; sector < new_hi; ) {
1725		md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1726		sector += blocks;
1727	}
1728	WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1729}
1730EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1731
1732static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1733{
1734	/* For each chunk covered by any of these sectors, set the
1735	 * counter to 2 and possibly set resync_needed.  They should all
1736	 * be 0 at this point
1737	 */
1738
1739	sector_t secs;
1740	bitmap_counter_t *bmc;
1741	spin_lock_irq(&bitmap->counts.lock);
1742	bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1743	if (!bmc) {
1744		spin_unlock_irq(&bitmap->counts.lock);
1745		return;
1746	}
1747	if (!*bmc) {
1748		*bmc = 2;
1749		md_bitmap_count_page(&bitmap->counts, offset, 1);
1750		md_bitmap_set_pending(&bitmap->counts, offset);
1751		bitmap->allclean = 0;
1752	}
1753	if (needed)
1754		*bmc |= NEEDED_MASK;
1755	spin_unlock_irq(&bitmap->counts.lock);
1756}
1757
1758/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1759void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1760{
1761	unsigned long chunk;
1762
1763	for (chunk = s; chunk <= e; chunk++) {
1764		sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1765		md_bitmap_set_memory_bits(bitmap, sec, 1);
1766		md_bitmap_file_set_bit(bitmap, sec);
1767		if (sec < bitmap->mddev->recovery_cp)
1768			/* We are asserting that the array is dirty,
1769			 * so move the recovery_cp address back so
1770			 * that it is obvious that it is dirty
1771			 */
1772			bitmap->mddev->recovery_cp = sec;
1773	}
1774}
1775
1776/*
1777 * flush out any pending updates
1778 */
1779void md_bitmap_flush(struct mddev *mddev)
1780{
1781	struct bitmap *bitmap = mddev->bitmap;
1782	long sleep;
1783
1784	if (!bitmap) /* there was no bitmap */
1785		return;
1786
1787	/* run the daemon_work three time to ensure everything is flushed
1788	 * that can be
1789	 */
1790	sleep = mddev->bitmap_info.daemon_sleep * 2;
1791	bitmap->daemon_lastrun -= sleep;
1792	md_bitmap_daemon_work(mddev);
1793	bitmap->daemon_lastrun -= sleep;
1794	md_bitmap_daemon_work(mddev);
1795	bitmap->daemon_lastrun -= sleep;
1796	md_bitmap_daemon_work(mddev);
1797	if (mddev->bitmap_info.external)
1798		md_super_wait(mddev);
1799	md_bitmap_update_sb(bitmap);
1800}
1801
1802/*
1803 * free memory that was allocated
1804 */
1805void md_bitmap_free(struct bitmap *bitmap)
1806{
1807	unsigned long k, pages;
1808	struct bitmap_page *bp;
1809
1810	if (!bitmap) /* there was no bitmap */
1811		return;
1812
1813	if (bitmap->sysfs_can_clear)
1814		sysfs_put(bitmap->sysfs_can_clear);
1815
1816	if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1817		bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1818		md_cluster_stop(bitmap->mddev);
1819
1820	/* Shouldn't be needed - but just in case.... */
1821	wait_event(bitmap->write_wait,
1822		   atomic_read(&bitmap->pending_writes) == 0);
1823
1824	/* release the bitmap file  */
1825	md_bitmap_file_unmap(&bitmap->storage);
1826
1827	bp = bitmap->counts.bp;
1828	pages = bitmap->counts.pages;
1829
1830	/* free all allocated memory */
1831
1832	if (bp) /* deallocate the page memory */
1833		for (k = 0; k < pages; k++)
1834			if (bp[k].map && !bp[k].hijacked)
1835				kfree(bp[k].map);
1836	kfree(bp);
1837	kfree(bitmap);
1838}
1839EXPORT_SYMBOL(md_bitmap_free);
1840
1841void md_bitmap_wait_behind_writes(struct mddev *mddev)
1842{
1843	struct bitmap *bitmap = mddev->bitmap;
1844
1845	/* wait for behind writes to complete */
1846	if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1847		pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1848			 mdname(mddev));
1849		/* need to kick something here to make sure I/O goes? */
1850		wait_event(bitmap->behind_wait,
1851			   atomic_read(&bitmap->behind_writes) == 0);
1852	}
1853}
1854
1855void md_bitmap_destroy(struct mddev *mddev)
1856{
1857	struct bitmap *bitmap = mddev->bitmap;
1858
1859	if (!bitmap) /* there was no bitmap */
1860		return;
1861
1862	md_bitmap_wait_behind_writes(mddev);
1863	if (!mddev->serialize_policy)
1864		mddev_destroy_serial_pool(mddev, NULL);
1865
1866	mutex_lock(&mddev->bitmap_info.mutex);
1867	spin_lock(&mddev->lock);
1868	mddev->bitmap = NULL; /* disconnect from the md device */
1869	spin_unlock(&mddev->lock);
1870	mutex_unlock(&mddev->bitmap_info.mutex);
1871	mddev_set_timeout(mddev, MAX_SCHEDULE_TIMEOUT, true);
 
1872
1873	md_bitmap_free(bitmap);
1874}
1875
1876/*
1877 * initialize the bitmap structure
1878 * if this returns an error, bitmap_destroy must be called to do clean up
1879 * once mddev->bitmap is set
1880 */
1881struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1882{
1883	struct bitmap *bitmap;
1884	sector_t blocks = mddev->resync_max_sectors;
1885	struct file *file = mddev->bitmap_info.file;
1886	int err;
1887	struct kernfs_node *bm = NULL;
1888
1889	BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1890
1891	BUG_ON(file && mddev->bitmap_info.offset);
1892
1893	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1894		pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1895			  mdname(mddev));
1896		return ERR_PTR(-EBUSY);
1897	}
1898
1899	bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1900	if (!bitmap)
1901		return ERR_PTR(-ENOMEM);
1902
1903	spin_lock_init(&bitmap->counts.lock);
1904	atomic_set(&bitmap->pending_writes, 0);
1905	init_waitqueue_head(&bitmap->write_wait);
1906	init_waitqueue_head(&bitmap->overflow_wait);
1907	init_waitqueue_head(&bitmap->behind_wait);
1908
1909	bitmap->mddev = mddev;
1910	bitmap->cluster_slot = slot;
1911
1912	if (mddev->kobj.sd)
1913		bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1914	if (bm) {
1915		bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1916		sysfs_put(bm);
1917	} else
1918		bitmap->sysfs_can_clear = NULL;
1919
1920	bitmap->storage.file = file;
1921	if (file) {
1922		get_file(file);
1923		/* As future accesses to this file will use bmap,
1924		 * and bypass the page cache, we must sync the file
1925		 * first.
1926		 */
1927		vfs_fsync(file, 1);
1928	}
1929	/* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1930	if (!mddev->bitmap_info.external) {
1931		/*
1932		 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1933		 * instructing us to create a new on-disk bitmap instance.
1934		 */
1935		if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1936			err = md_bitmap_new_disk_sb(bitmap);
1937		else
1938			err = md_bitmap_read_sb(bitmap);
1939	} else {
1940		err = 0;
1941		if (mddev->bitmap_info.chunksize == 0 ||
1942		    mddev->bitmap_info.daemon_sleep == 0)
1943			/* chunksize and time_base need to be
1944			 * set first. */
1945			err = -EINVAL;
1946	}
1947	if (err)
1948		goto error;
1949
1950	bitmap->daemon_lastrun = jiffies;
1951	err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1952	if (err)
1953		goto error;
1954
1955	pr_debug("created bitmap (%lu pages) for device %s\n",
1956		 bitmap->counts.pages, bmname(bitmap));
1957
1958	err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1959	if (err)
1960		goto error;
1961
1962	return bitmap;
1963 error:
1964	md_bitmap_free(bitmap);
1965	return ERR_PTR(err);
1966}
1967
1968int md_bitmap_load(struct mddev *mddev)
1969{
1970	int err = 0;
1971	sector_t start = 0;
1972	sector_t sector = 0;
1973	struct bitmap *bitmap = mddev->bitmap;
1974	struct md_rdev *rdev;
1975
1976	if (!bitmap)
1977		goto out;
1978
1979	rdev_for_each(rdev, mddev)
1980		mddev_create_serial_pool(mddev, rdev);
1981
1982	if (mddev_is_clustered(mddev))
1983		md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1984
1985	/* Clear out old bitmap info first:  Either there is none, or we
1986	 * are resuming after someone else has possibly changed things,
1987	 * so we should forget old cached info.
1988	 * All chunks should be clean, but some might need_sync.
1989	 */
1990	while (sector < mddev->resync_max_sectors) {
1991		sector_t blocks;
1992		md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1993		sector += blocks;
1994	}
1995	md_bitmap_close_sync(bitmap);
1996
1997	if (mddev->degraded == 0
1998	    || bitmap->events_cleared == mddev->events)
1999		/* no need to keep dirty bits to optimise a
2000		 * re-add of a missing device */
2001		start = mddev->recovery_cp;
2002
2003	mutex_lock(&mddev->bitmap_info.mutex);
2004	err = md_bitmap_init_from_disk(bitmap, start);
2005	mutex_unlock(&mddev->bitmap_info.mutex);
2006
2007	if (err)
2008		goto out;
2009	clear_bit(BITMAP_STALE, &bitmap->flags);
2010
2011	/* Kick recovery in case any bits were set */
2012	set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
2013
2014	mddev_set_timeout(mddev, mddev->bitmap_info.daemon_sleep, true);
2015	md_wakeup_thread(mddev->thread);
2016
2017	md_bitmap_update_sb(bitmap);
2018
2019	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
2020		err = -EIO;
2021out:
2022	return err;
2023}
2024EXPORT_SYMBOL_GPL(md_bitmap_load);
2025
2026/* caller need to free returned bitmap with md_bitmap_free() */
2027struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
2028{
2029	int rv = 0;
2030	struct bitmap *bitmap;
2031
2032	bitmap = md_bitmap_create(mddev, slot);
2033	if (IS_ERR(bitmap)) {
2034		rv = PTR_ERR(bitmap);
2035		return ERR_PTR(rv);
2036	}
2037
2038	rv = md_bitmap_init_from_disk(bitmap, 0);
2039	if (rv) {
2040		md_bitmap_free(bitmap);
2041		return ERR_PTR(rv);
2042	}
2043
2044	return bitmap;
2045}
2046EXPORT_SYMBOL(get_bitmap_from_slot);
2047
2048/* Loads the bitmap associated with slot and copies the resync information
2049 * to our bitmap
2050 */
2051int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
2052		sector_t *low, sector_t *high, bool clear_bits)
2053{
2054	int rv = 0, i, j;
2055	sector_t block, lo = 0, hi = 0;
2056	struct bitmap_counts *counts;
2057	struct bitmap *bitmap;
2058
2059	bitmap = get_bitmap_from_slot(mddev, slot);
2060	if (IS_ERR(bitmap)) {
2061		pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
2062		return -1;
2063	}
2064
2065	counts = &bitmap->counts;
2066	for (j = 0; j < counts->chunks; j++) {
2067		block = (sector_t)j << counts->chunkshift;
2068		if (md_bitmap_file_test_bit(bitmap, block)) {
2069			if (!lo)
2070				lo = block;
2071			hi = block;
2072			md_bitmap_file_clear_bit(bitmap, block);
2073			md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2074			md_bitmap_file_set_bit(mddev->bitmap, block);
2075		}
2076	}
2077
2078	if (clear_bits) {
2079		md_bitmap_update_sb(bitmap);
2080		/* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2081		 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2082		for (i = 0; i < bitmap->storage.file_pages; i++)
2083			if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2084				set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2085		md_bitmap_unplug(bitmap);
2086	}
2087	md_bitmap_unplug(mddev->bitmap);
2088	*low = lo;
2089	*high = hi;
2090	md_bitmap_free(bitmap);
2091
2092	return rv;
2093}
2094EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2095
2096
2097void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2098{
2099	unsigned long chunk_kb;
2100	struct bitmap_counts *counts;
2101
2102	if (!bitmap)
2103		return;
2104
2105	counts = &bitmap->counts;
2106
2107	chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2108	seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2109		   "%lu%s chunk",
2110		   counts->pages - counts->missing_pages,
2111		   counts->pages,
2112		   (counts->pages - counts->missing_pages)
2113		   << (PAGE_SHIFT - 10),
2114		   chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2115		   chunk_kb ? "KB" : "B");
2116	if (bitmap->storage.file) {
2117		seq_printf(seq, ", file: ");
2118		seq_file_path(seq, bitmap->storage.file, " \t\n");
2119	}
2120
2121	seq_printf(seq, "\n");
2122}
2123
2124int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2125		  int chunksize, int init)
2126{
2127	/* If chunk_size is 0, choose an appropriate chunk size.
2128	 * Then possibly allocate new storage space.
2129	 * Then quiesce, copy bits, replace bitmap, and re-start
2130	 *
2131	 * This function is called both to set up the initial bitmap
2132	 * and to resize the bitmap while the array is active.
2133	 * If this happens as a result of the array being resized,
2134	 * chunksize will be zero, and we need to choose a suitable
2135	 * chunksize, otherwise we use what we are given.
2136	 */
2137	struct bitmap_storage store;
2138	struct bitmap_counts old_counts;
2139	unsigned long chunks;
2140	sector_t block;
2141	sector_t old_blocks, new_blocks;
2142	int chunkshift;
2143	int ret = 0;
2144	long pages;
2145	struct bitmap_page *new_bp;
2146
2147	if (bitmap->storage.file && !init) {
2148		pr_info("md: cannot resize file-based bitmap\n");
2149		return -EINVAL;
2150	}
2151
2152	if (chunksize == 0) {
2153		/* If there is enough space, leave the chunk size unchanged,
2154		 * else increase by factor of two until there is enough space.
2155		 */
2156		long bytes;
2157		long space = bitmap->mddev->bitmap_info.space;
2158
2159		if (space == 0) {
2160			/* We don't know how much space there is, so limit
2161			 * to current size - in sectors.
2162			 */
2163			bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2164			if (!bitmap->mddev->bitmap_info.external)
2165				bytes += sizeof(bitmap_super_t);
2166			space = DIV_ROUND_UP(bytes, 512);
2167			bitmap->mddev->bitmap_info.space = space;
2168		}
2169		chunkshift = bitmap->counts.chunkshift;
2170		chunkshift--;
2171		do {
2172			/* 'chunkshift' is shift from block size to chunk size */
2173			chunkshift++;
2174			chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2175			bytes = DIV_ROUND_UP(chunks, 8);
2176			if (!bitmap->mddev->bitmap_info.external)
2177				bytes += sizeof(bitmap_super_t);
2178		} while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) <
2179			(BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1));
2180	} else
2181		chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2182
2183	chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2184	memset(&store, 0, sizeof(store));
2185	if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2186		ret = md_bitmap_storage_alloc(&store, chunks,
2187					      !bitmap->mddev->bitmap_info.external,
2188					      mddev_is_clustered(bitmap->mddev)
2189					      ? bitmap->cluster_slot : 0);
2190	if (ret) {
2191		md_bitmap_file_unmap(&store);
2192		goto err;
2193	}
2194
2195	pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2196
2197	new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2198	ret = -ENOMEM;
2199	if (!new_bp) {
2200		md_bitmap_file_unmap(&store);
2201		goto err;
2202	}
2203
2204	if (!init)
2205		bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2206
2207	store.file = bitmap->storage.file;
2208	bitmap->storage.file = NULL;
2209
2210	if (store.sb_page && bitmap->storage.sb_page)
2211		memcpy(page_address(store.sb_page),
2212		       page_address(bitmap->storage.sb_page),
2213		       sizeof(bitmap_super_t));
2214	spin_lock_irq(&bitmap->counts.lock);
2215	md_bitmap_file_unmap(&bitmap->storage);
2216	bitmap->storage = store;
2217
2218	old_counts = bitmap->counts;
2219	bitmap->counts.bp = new_bp;
2220	bitmap->counts.pages = pages;
2221	bitmap->counts.missing_pages = pages;
2222	bitmap->counts.chunkshift = chunkshift;
2223	bitmap->counts.chunks = chunks;
2224	bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift +
2225						     BITMAP_BLOCK_SHIFT);
2226
2227	blocks = min(old_counts.chunks << old_counts.chunkshift,
2228		     chunks << chunkshift);
2229
2230	/* For cluster raid, need to pre-allocate bitmap */
2231	if (mddev_is_clustered(bitmap->mddev)) {
2232		unsigned long page;
2233		for (page = 0; page < pages; page++) {
2234			ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2235			if (ret) {
2236				unsigned long k;
2237
2238				/* deallocate the page memory */
2239				for (k = 0; k < page; k++) {
2240					kfree(new_bp[k].map);
2241				}
2242				kfree(new_bp);
2243
2244				/* restore some fields from old_counts */
2245				bitmap->counts.bp = old_counts.bp;
2246				bitmap->counts.pages = old_counts.pages;
2247				bitmap->counts.missing_pages = old_counts.pages;
2248				bitmap->counts.chunkshift = old_counts.chunkshift;
2249				bitmap->counts.chunks = old_counts.chunks;
2250				bitmap->mddev->bitmap_info.chunksize =
2251					1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT);
2252				blocks = old_counts.chunks << old_counts.chunkshift;
2253				pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2254				break;
2255			} else
2256				bitmap->counts.bp[page].count += 1;
2257		}
2258	}
2259
2260	for (block = 0; block < blocks; ) {
2261		bitmap_counter_t *bmc_old, *bmc_new;
2262		int set;
2263
2264		bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2265		set = bmc_old && NEEDED(*bmc_old);
2266
2267		if (set) {
2268			bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2269			if (bmc_new) {
2270				if (*bmc_new == 0) {
2271					/* need to set on-disk bits too. */
2272					sector_t end = block + new_blocks;
2273					sector_t start = block >> chunkshift;
2274
2275					start <<= chunkshift;
2276					while (start < end) {
2277						md_bitmap_file_set_bit(bitmap, block);
2278						start += 1 << chunkshift;
2279					}
2280					*bmc_new = 2;
2281					md_bitmap_count_page(&bitmap->counts, block, 1);
2282					md_bitmap_set_pending(&bitmap->counts, block);
2283				}
2284				*bmc_new |= NEEDED_MASK;
 
 
2285			}
 
2286			if (new_blocks < old_blocks)
2287				old_blocks = new_blocks;
2288		}
2289		block += old_blocks;
2290	}
2291
2292	if (bitmap->counts.bp != old_counts.bp) {
2293		unsigned long k;
2294		for (k = 0; k < old_counts.pages; k++)
2295			if (!old_counts.bp[k].hijacked)
2296				kfree(old_counts.bp[k].map);
2297		kfree(old_counts.bp);
2298	}
2299
2300	if (!init) {
2301		int i;
2302		while (block < (chunks << chunkshift)) {
2303			bitmap_counter_t *bmc;
2304			bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2305			if (bmc) {
2306				/* new space.  It needs to be resynced, so
2307				 * we set NEEDED_MASK.
2308				 */
2309				if (*bmc == 0) {
2310					*bmc = NEEDED_MASK | 2;
2311					md_bitmap_count_page(&bitmap->counts, block, 1);
2312					md_bitmap_set_pending(&bitmap->counts, block);
2313				}
2314			}
2315			block += new_blocks;
2316		}
2317		for (i = 0; i < bitmap->storage.file_pages; i++)
2318			set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2319	}
2320	spin_unlock_irq(&bitmap->counts.lock);
2321
2322	if (!init) {
2323		md_bitmap_unplug(bitmap);
2324		bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2325	}
2326	ret = 0;
2327err:
2328	return ret;
2329}
2330EXPORT_SYMBOL_GPL(md_bitmap_resize);
2331
2332static ssize_t
2333location_show(struct mddev *mddev, char *page)
2334{
2335	ssize_t len;
2336	if (mddev->bitmap_info.file)
2337		len = sprintf(page, "file");
2338	else if (mddev->bitmap_info.offset)
2339		len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2340	else
2341		len = sprintf(page, "none");
2342	len += sprintf(page+len, "\n");
2343	return len;
2344}
2345
2346static ssize_t
2347location_store(struct mddev *mddev, const char *buf, size_t len)
2348{
2349	int rv;
2350
2351	rv = mddev_suspend_and_lock(mddev);
2352	if (rv)
2353		return rv;
2354
2355	if (mddev->pers) {
 
 
 
 
2356		if (mddev->recovery || mddev->sync_thread) {
2357			rv = -EBUSY;
2358			goto out;
2359		}
2360	}
2361
2362	if (mddev->bitmap || mddev->bitmap_info.file ||
2363	    mddev->bitmap_info.offset) {
2364		/* bitmap already configured.  Only option is to clear it */
2365		if (strncmp(buf, "none", 4) != 0) {
2366			rv = -EBUSY;
2367			goto out;
2368		}
2369
2370		md_bitmap_destroy(mddev);
 
 
 
2371		mddev->bitmap_info.offset = 0;
2372		if (mddev->bitmap_info.file) {
2373			struct file *f = mddev->bitmap_info.file;
2374			mddev->bitmap_info.file = NULL;
2375			fput(f);
2376		}
2377	} else {
2378		/* No bitmap, OK to set a location */
2379		long long offset;
2380		struct bitmap *bitmap;
2381
2382		if (strncmp(buf, "none", 4) == 0)
2383			/* nothing to be done */;
2384		else if (strncmp(buf, "file:", 5) == 0) {
2385			/* Not supported yet */
2386			rv = -EINVAL;
2387			goto out;
2388		} else {
2389			if (buf[0] == '+')
2390				rv = kstrtoll(buf+1, 10, &offset);
2391			else
2392				rv = kstrtoll(buf, 10, &offset);
2393			if (rv)
2394				goto out;
2395			if (offset == 0) {
2396				rv = -EINVAL;
2397				goto out;
2398			}
2399			if (mddev->bitmap_info.external == 0 &&
2400			    mddev->major_version == 0 &&
2401			    offset != mddev->bitmap_info.default_offset) {
2402				rv = -EINVAL;
2403				goto out;
2404			}
2405
2406			mddev->bitmap_info.offset = offset;
2407			bitmap = md_bitmap_create(mddev, -1);
2408			if (IS_ERR(bitmap)) {
2409				rv = PTR_ERR(bitmap);
2410				goto out;
2411			}
2412
2413			mddev->bitmap = bitmap;
2414			rv = md_bitmap_load(mddev);
2415			if (rv) {
2416				mddev->bitmap_info.offset = 0;
2417				md_bitmap_destroy(mddev);
2418				goto out;
 
 
 
 
 
 
2419			}
2420		}
2421	}
2422	if (!mddev->external) {
2423		/* Ensure new bitmap info is stored in
2424		 * metadata promptly.
2425		 */
2426		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2427		md_wakeup_thread(mddev->thread);
2428	}
2429	rv = 0;
2430out:
2431	mddev_unlock_and_resume(mddev);
2432	if (rv)
2433		return rv;
2434	return len;
2435}
2436
2437static struct md_sysfs_entry bitmap_location =
2438__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2439
2440/* 'bitmap/space' is the space available at 'location' for the
2441 * bitmap.  This allows the kernel to know when it is safe to
2442 * resize the bitmap to match a resized array.
2443 */
2444static ssize_t
2445space_show(struct mddev *mddev, char *page)
2446{
2447	return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2448}
2449
2450static ssize_t
2451space_store(struct mddev *mddev, const char *buf, size_t len)
2452{
2453	unsigned long sectors;
2454	int rv;
2455
2456	rv = kstrtoul(buf, 10, &sectors);
2457	if (rv)
2458		return rv;
2459
2460	if (sectors == 0)
2461		return -EINVAL;
2462
2463	if (mddev->bitmap &&
2464	    sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2465		return -EFBIG; /* Bitmap is too big for this small space */
2466
2467	/* could make sure it isn't too big, but that isn't really
2468	 * needed - user-space should be careful.
2469	 */
2470	mddev->bitmap_info.space = sectors;
2471	return len;
2472}
2473
2474static struct md_sysfs_entry bitmap_space =
2475__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2476
2477static ssize_t
2478timeout_show(struct mddev *mddev, char *page)
2479{
2480	ssize_t len;
2481	unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2482	unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2483
2484	len = sprintf(page, "%lu", secs);
2485	if (jifs)
2486		len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2487	len += sprintf(page+len, "\n");
2488	return len;
2489}
2490
2491static ssize_t
2492timeout_store(struct mddev *mddev, const char *buf, size_t len)
2493{
2494	/* timeout can be set at any time */
2495	unsigned long timeout;
2496	int rv = strict_strtoul_scaled(buf, &timeout, 4);
2497	if (rv)
2498		return rv;
2499
2500	/* just to make sure we don't overflow... */
2501	if (timeout >= LONG_MAX / HZ)
2502		return -EINVAL;
2503
2504	timeout = timeout * HZ / 10000;
2505
2506	if (timeout >= MAX_SCHEDULE_TIMEOUT)
2507		timeout = MAX_SCHEDULE_TIMEOUT-1;
2508	if (timeout < 1)
2509		timeout = 1;
2510
2511	mddev->bitmap_info.daemon_sleep = timeout;
2512	mddev_set_timeout(mddev, timeout, false);
2513	md_wakeup_thread(mddev->thread);
2514
 
 
 
 
 
 
 
2515	return len;
2516}
2517
2518static struct md_sysfs_entry bitmap_timeout =
2519__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2520
2521static ssize_t
2522backlog_show(struct mddev *mddev, char *page)
2523{
2524	return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2525}
2526
2527static ssize_t
2528backlog_store(struct mddev *mddev, const char *buf, size_t len)
2529{
2530	unsigned long backlog;
2531	unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
2532	struct md_rdev *rdev;
2533	bool has_write_mostly = false;
2534	int rv = kstrtoul(buf, 10, &backlog);
2535	if (rv)
2536		return rv;
2537	if (backlog > COUNTER_MAX)
2538		return -EINVAL;
2539
2540	rv = mddev_suspend_and_lock(mddev);
2541	if (rv)
2542		return rv;
2543
2544	/*
2545	 * Without write mostly device, it doesn't make sense to set
2546	 * backlog for max_write_behind.
2547	 */
2548	rdev_for_each(rdev, mddev) {
2549		if (test_bit(WriteMostly, &rdev->flags)) {
2550			has_write_mostly = true;
2551			break;
2552		}
2553	}
2554	if (!has_write_mostly) {
2555		pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n",
2556				    mdname(mddev));
2557		mddev_unlock(mddev);
2558		return -EINVAL;
2559	}
2560
2561	mddev->bitmap_info.max_write_behind = backlog;
2562	if (!backlog && mddev->serial_info_pool) {
2563		/* serial_info_pool is not needed if backlog is zero */
2564		if (!mddev->serialize_policy)
2565			mddev_destroy_serial_pool(mddev, NULL);
2566	} else if (backlog && !mddev->serial_info_pool) {
2567		/* serial_info_pool is needed since backlog is not zero */
 
 
2568		rdev_for_each(rdev, mddev)
2569			mddev_create_serial_pool(mddev, rdev);
2570	}
2571	if (old_mwb != backlog)
2572		md_bitmap_update_sb(mddev->bitmap);
2573
2574	mddev_unlock_and_resume(mddev);
2575	return len;
2576}
2577
2578static struct md_sysfs_entry bitmap_backlog =
2579__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2580
2581static ssize_t
2582chunksize_show(struct mddev *mddev, char *page)
2583{
2584	return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2585}
2586
2587static ssize_t
2588chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2589{
2590	/* Can only be changed when no bitmap is active */
2591	int rv;
2592	unsigned long csize;
2593	if (mddev->bitmap)
2594		return -EBUSY;
2595	rv = kstrtoul(buf, 10, &csize);
2596	if (rv)
2597		return rv;
2598	if (csize < 512 ||
2599	    !is_power_of_2(csize))
2600		return -EINVAL;
2601	if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE *
2602		sizeof(((bitmap_super_t *)0)->chunksize))))
2603		return -EOVERFLOW;
2604	mddev->bitmap_info.chunksize = csize;
2605	return len;
2606}
2607
2608static struct md_sysfs_entry bitmap_chunksize =
2609__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2610
2611static ssize_t metadata_show(struct mddev *mddev, char *page)
2612{
2613	if (mddev_is_clustered(mddev))
2614		return sprintf(page, "clustered\n");
2615	return sprintf(page, "%s\n", (mddev->bitmap_info.external
2616				      ? "external" : "internal"));
2617}
2618
2619static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2620{
2621	if (mddev->bitmap ||
2622	    mddev->bitmap_info.file ||
2623	    mddev->bitmap_info.offset)
2624		return -EBUSY;
2625	if (strncmp(buf, "external", 8) == 0)
2626		mddev->bitmap_info.external = 1;
2627	else if ((strncmp(buf, "internal", 8) == 0) ||
2628			(strncmp(buf, "clustered", 9) == 0))
2629		mddev->bitmap_info.external = 0;
2630	else
2631		return -EINVAL;
2632	return len;
2633}
2634
2635static struct md_sysfs_entry bitmap_metadata =
2636__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2637
2638static ssize_t can_clear_show(struct mddev *mddev, char *page)
2639{
2640	int len;
2641	spin_lock(&mddev->lock);
2642	if (mddev->bitmap)
2643		len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2644					     "false" : "true"));
2645	else
2646		len = sprintf(page, "\n");
2647	spin_unlock(&mddev->lock);
2648	return len;
2649}
2650
2651static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2652{
2653	if (mddev->bitmap == NULL)
2654		return -ENOENT;
2655	if (strncmp(buf, "false", 5) == 0)
2656		mddev->bitmap->need_sync = 1;
2657	else if (strncmp(buf, "true", 4) == 0) {
2658		if (mddev->degraded)
2659			return -EBUSY;
2660		mddev->bitmap->need_sync = 0;
2661	} else
2662		return -EINVAL;
2663	return len;
2664}
2665
2666static struct md_sysfs_entry bitmap_can_clear =
2667__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2668
2669static ssize_t
2670behind_writes_used_show(struct mddev *mddev, char *page)
2671{
2672	ssize_t ret;
2673	spin_lock(&mddev->lock);
2674	if (mddev->bitmap == NULL)
2675		ret = sprintf(page, "0\n");
2676	else
2677		ret = sprintf(page, "%lu\n",
2678			      mddev->bitmap->behind_writes_used);
2679	spin_unlock(&mddev->lock);
2680	return ret;
2681}
2682
2683static ssize_t
2684behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2685{
2686	if (mddev->bitmap)
2687		mddev->bitmap->behind_writes_used = 0;
2688	return len;
2689}
2690
2691static struct md_sysfs_entry max_backlog_used =
2692__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2693       behind_writes_used_show, behind_writes_used_reset);
2694
2695static struct attribute *md_bitmap_attrs[] = {
2696	&bitmap_location.attr,
2697	&bitmap_space.attr,
2698	&bitmap_timeout.attr,
2699	&bitmap_backlog.attr,
2700	&bitmap_chunksize.attr,
2701	&bitmap_metadata.attr,
2702	&bitmap_can_clear.attr,
2703	&max_backlog_used.attr,
2704	NULL
2705};
2706const struct attribute_group md_bitmap_group = {
2707	.name = "bitmap",
2708	.attrs = md_bitmap_attrs,
2709};
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
   4 *
   5 * bitmap_create  - sets up the bitmap structure
   6 * bitmap_destroy - destroys the bitmap structure
   7 *
   8 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
   9 * - added disk storage for bitmap
  10 * - changes to allow various bitmap chunk sizes
  11 */
  12
  13/*
  14 * Still to do:
  15 *
  16 * flush after percent set rather than just time based. (maybe both).
  17 */
  18
  19#include <linux/blkdev.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/slab.h>
  23#include <linux/init.h>
  24#include <linux/timer.h>
  25#include <linux/sched.h>
  26#include <linux/list.h>
  27#include <linux/file.h>
  28#include <linux/mount.h>
  29#include <linux/buffer_head.h>
  30#include <linux/seq_file.h>
  31#include <trace/events/block.h>
  32#include "md.h"
  33#include "md-bitmap.h"
  34
  35static inline char *bmname(struct bitmap *bitmap)
  36{
  37	return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
  38}
  39
  40/*
  41 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
  42 *
  43 * 1) check to see if this page is allocated, if it's not then try to alloc
  44 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
  45 *    page pointer directly as a counter
  46 *
  47 * if we find our page, we increment the page's refcount so that it stays
  48 * allocated while we're using it
  49 */
  50static int md_bitmap_checkpage(struct bitmap_counts *bitmap,
  51			       unsigned long page, int create, int no_hijack)
  52__releases(bitmap->lock)
  53__acquires(bitmap->lock)
  54{
  55	unsigned char *mappage;
  56
  57	if (page >= bitmap->pages) {
  58		/* This can happen if bitmap_start_sync goes beyond
  59		 * End-of-device while looking for a whole page.
  60		 * It is harmless.
  61		 */
  62		return -EINVAL;
  63	}
  64
  65	if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
  66		return 0;
  67
  68	if (bitmap->bp[page].map) /* page is already allocated, just return */
  69		return 0;
  70
  71	if (!create)
  72		return -ENOENT;
  73
  74	/* this page has not been allocated yet */
  75
  76	spin_unlock_irq(&bitmap->lock);
  77	/* It is possible that this is being called inside a
  78	 * prepare_to_wait/finish_wait loop from raid5c:make_request().
  79	 * In general it is not permitted to sleep in that context as it
  80	 * can cause the loop to spin freely.
  81	 * That doesn't apply here as we can only reach this point
  82	 * once with any loop.
  83	 * When this function completes, either bp[page].map or
  84	 * bp[page].hijacked.  In either case, this function will
  85	 * abort before getting to this point again.  So there is
  86	 * no risk of a free-spin, and so it is safe to assert
  87	 * that sleeping here is allowed.
  88	 */
  89	sched_annotate_sleep();
  90	mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
  91	spin_lock_irq(&bitmap->lock);
  92
  93	if (mappage == NULL) {
  94		pr_debug("md/bitmap: map page allocation failed, hijacking\n");
  95		/* We don't support hijack for cluster raid */
  96		if (no_hijack)
  97			return -ENOMEM;
  98		/* failed - set the hijacked flag so that we can use the
  99		 * pointer as a counter */
 100		if (!bitmap->bp[page].map)
 101			bitmap->bp[page].hijacked = 1;
 102	} else if (bitmap->bp[page].map ||
 103		   bitmap->bp[page].hijacked) {
 104		/* somebody beat us to getting the page */
 105		kfree(mappage);
 106	} else {
 107
 108		/* no page was in place and we have one, so install it */
 109
 110		bitmap->bp[page].map = mappage;
 111		bitmap->missing_pages--;
 112	}
 113	return 0;
 114}
 115
 116/* if page is completely empty, put it back on the free list, or dealloc it */
 117/* if page was hijacked, unmark the flag so it might get alloced next time */
 118/* Note: lock should be held when calling this */
 119static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
 120{
 121	char *ptr;
 122
 123	if (bitmap->bp[page].count) /* page is still busy */
 124		return;
 125
 126	/* page is no longer in use, it can be released */
 127
 128	if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
 129		bitmap->bp[page].hijacked = 0;
 130		bitmap->bp[page].map = NULL;
 131	} else {
 132		/* normal case, free the page */
 133		ptr = bitmap->bp[page].map;
 134		bitmap->bp[page].map = NULL;
 135		bitmap->missing_pages++;
 136		kfree(ptr);
 137	}
 138}
 139
 140/*
 141 * bitmap file handling - read and write the bitmap file and its superblock
 142 */
 143
 144/*
 145 * basic page I/O operations
 146 */
 147
 148/* IO operations when bitmap is stored near all superblocks */
 
 
 149static int read_sb_page(struct mddev *mddev, loff_t offset,
 150			struct page *page,
 151			unsigned long index, int size)
 152{
 153	/* choose a good rdev and read the page from there */
 154
 
 
 155	struct md_rdev *rdev;
 156	sector_t target;
 157
 158	rdev_for_each(rdev, mddev) {
 159		if (! test_bit(In_sync, &rdev->flags)
 160		    || test_bit(Faulty, &rdev->flags)
 161		    || test_bit(Bitmap_sync, &rdev->flags))
 
 
 162			continue;
 163
 164		target = offset + index * (PAGE_SIZE/512);
 165
 166		if (sync_page_io(rdev, target,
 167				 roundup(size, bdev_logical_block_size(rdev->bdev)),
 168				 page, REQ_OP_READ, 0, true)) {
 169			page->index = index;
 170			return 0;
 171		}
 172	}
 173	return -EIO;
 174}
 175
 176static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
 177{
 178	/* Iterate the disks of an mddev, using rcu to protect access to the
 179	 * linked list, and raising the refcount of devices we return to ensure
 180	 * they don't disappear while in use.
 181	 * As devices are only added or removed when raid_disk is < 0 and
 182	 * nr_pending is 0 and In_sync is clear, the entries we return will
 183	 * still be in the same position on the list when we re-enter
 184	 * list_for_each_entry_continue_rcu.
 185	 *
 186	 * Note that if entered with 'rdev == NULL' to start at the
 187	 * beginning, we temporarily assign 'rdev' to an address which
 188	 * isn't really an rdev, but which can be used by
 189	 * list_for_each_entry_continue_rcu() to find the first entry.
 190	 */
 191	rcu_read_lock();
 192	if (rdev == NULL)
 193		/* start at the beginning */
 194		rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
 195	else {
 196		/* release the previous rdev and start from there. */
 197		rdev_dec_pending(rdev, mddev);
 198	}
 199	list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
 200		if (rdev->raid_disk >= 0 &&
 201		    !test_bit(Faulty, &rdev->flags)) {
 202			/* this is a usable devices */
 203			atomic_inc(&rdev->nr_pending);
 204			rcu_read_unlock();
 205			return rdev;
 206		}
 207	}
 208	rcu_read_unlock();
 209	return NULL;
 210}
 211
 212static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 213{
 214	struct md_rdev *rdev;
 215	struct block_device *bdev;
 216	struct mddev *mddev = bitmap->mddev;
 217	struct bitmap_storage *store = &bitmap->storage;
 218
 219restart:
 220	rdev = NULL;
 221	while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
 222		int size = PAGE_SIZE;
 223		loff_t offset = mddev->bitmap_info.offset;
 224
 225		bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
 226
 227		if (page->index == store->file_pages-1) {
 228			int last_page_size = store->bytes & (PAGE_SIZE-1);
 229			if (last_page_size == 0)
 230				last_page_size = PAGE_SIZE;
 231			size = roundup(last_page_size,
 232				       bdev_logical_block_size(bdev));
 233		}
 234		/* Just make sure we aren't corrupting data or
 235		 * metadata
 236		 */
 237		if (mddev->external) {
 238			/* Bitmap could be anywhere. */
 239			if (rdev->sb_start + offset + (page->index
 240						       * (PAGE_SIZE/512))
 241			    > rdev->data_offset
 242			    &&
 243			    rdev->sb_start + offset
 244			    < (rdev->data_offset + mddev->dev_sectors
 245			     + (PAGE_SIZE/512)))
 246				goto bad_alignment;
 247		} else if (offset < 0) {
 248			/* DATA  BITMAP METADATA  */
 249			if (offset
 250			    + (long)(page->index * (PAGE_SIZE/512))
 251			    + size/512 > 0)
 252				/* bitmap runs in to metadata */
 253				goto bad_alignment;
 254			if (rdev->data_offset + mddev->dev_sectors
 255			    > rdev->sb_start + offset)
 256				/* data runs in to bitmap */
 257				goto bad_alignment;
 258		} else if (rdev->sb_start < rdev->data_offset) {
 259			/* METADATA BITMAP DATA */
 260			if (rdev->sb_start
 261			    + offset
 262			    + page->index*(PAGE_SIZE/512) + size/512
 263			    > rdev->data_offset)
 264				/* bitmap runs in to data */
 265				goto bad_alignment;
 266		} else {
 267			/* DATA METADATA BITMAP - no problems */
 268		}
 269		md_super_write(mddev, rdev,
 270			       rdev->sb_start + offset
 271			       + page->index * (PAGE_SIZE/512),
 272			       size,
 273			       page);
 274	}
 275
 276	if (wait && md_super_wait(mddev) < 0)
 277		goto restart;
 278	return 0;
 279
 280 bad_alignment:
 281	return -EINVAL;
 282}
 283
 284static void md_bitmap_file_kick(struct bitmap *bitmap);
 285/*
 286 * write out a page to a file
 287 */
 288static void write_page(struct bitmap *bitmap, struct page *page, int wait)
 289{
 290	struct buffer_head *bh;
 
 
 
 291
 292	if (bitmap->storage.file == NULL) {
 293		switch (write_sb_page(bitmap, page, wait)) {
 294		case -EINVAL:
 295			set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 
 296		}
 297	} else {
 
 298
 299		bh = page_buffers(page);
 300
 301		while (bh && bh->b_blocknr) {
 302			atomic_inc(&bitmap->pending_writes);
 303			set_buffer_locked(bh);
 304			set_buffer_mapped(bh);
 305			submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
 306			bh = bh->b_this_page;
 307		}
 308
 309		if (wait)
 310			wait_event(bitmap->write_wait,
 311				   atomic_read(&bitmap->pending_writes)==0);
 
 
 
 312	}
 313	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 314		md_bitmap_file_kick(bitmap);
 
 
 315}
 316
 317static void end_bitmap_write(struct buffer_head *bh, int uptodate)
 318{
 319	struct bitmap *bitmap = bh->b_private;
 320
 321	if (!uptodate)
 322		set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
 323	if (atomic_dec_and_test(&bitmap->pending_writes))
 324		wake_up(&bitmap->write_wait);
 325}
 326
 327static void free_buffers(struct page *page)
 328{
 329	struct buffer_head *bh;
 330
 331	if (!PagePrivate(page))
 332		return;
 333
 334	bh = page_buffers(page);
 335	while (bh) {
 336		struct buffer_head *next = bh->b_this_page;
 337		free_buffer_head(bh);
 338		bh = next;
 339	}
 340	detach_page_private(page);
 341	put_page(page);
 342}
 343
 344/* read a page from a file.
 345 * We both read the page, and attach buffers to the page to record the
 346 * address of each block (using bmap).  These addresses will be used
 347 * to write the block later, completely bypassing the filesystem.
 348 * This usage is similar to how swap files are handled, and allows us
 349 * to write to a file with no concerns of memory allocation failing.
 350 */
 351static int read_page(struct file *file, unsigned long index,
 352		     struct bitmap *bitmap,
 353		     unsigned long count,
 354		     struct page *page)
 355{
 356	int ret = 0;
 357	struct inode *inode = file_inode(file);
 358	struct buffer_head *bh;
 359	sector_t block, blk_cur;
 360	unsigned long blocksize = i_blocksize(inode);
 361
 362	pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
 363		 (unsigned long long)index << PAGE_SHIFT);
 364
 365	bh = alloc_page_buffers(page, blocksize, false);
 366	if (!bh) {
 367		ret = -ENOMEM;
 368		goto out;
 369	}
 370	attach_page_private(page, bh);
 371	blk_cur = index << (PAGE_SHIFT - inode->i_blkbits);
 372	while (bh) {
 373		block = blk_cur;
 374
 375		if (count == 0)
 376			bh->b_blocknr = 0;
 377		else {
 378			ret = bmap(inode, &block);
 379			if (ret || !block) {
 380				ret = -EINVAL;
 381				bh->b_blocknr = 0;
 382				goto out;
 383			}
 384
 385			bh->b_blocknr = block;
 386			bh->b_bdev = inode->i_sb->s_bdev;
 387			if (count < blocksize)
 388				count = 0;
 389			else
 390				count -= blocksize;
 391
 392			bh->b_end_io = end_bitmap_write;
 393			bh->b_private = bitmap;
 394			atomic_inc(&bitmap->pending_writes);
 395			set_buffer_locked(bh);
 396			set_buffer_mapped(bh);
 397			submit_bh(REQ_OP_READ, 0, bh);
 398		}
 399		blk_cur++;
 400		bh = bh->b_this_page;
 401	}
 402	page->index = index;
 403
 404	wait_event(bitmap->write_wait,
 405		   atomic_read(&bitmap->pending_writes)==0);
 406	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
 407		ret = -EIO;
 408out:
 409	if (ret)
 410		pr_err("md: bitmap read error: (%dB @ %llu): %d\n",
 411		       (int)PAGE_SIZE,
 412		       (unsigned long long)index << PAGE_SHIFT,
 413		       ret);
 414	return ret;
 415}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 416
 417/*
 418 * bitmap file superblock operations
 419 */
 420
 421/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 422 * md_bitmap_wait_writes() should be called before writing any bitmap
 423 * blocks, to ensure previous writes, particularly from
 424 * md_bitmap_daemon_work(), have completed.
 425 */
 426static void md_bitmap_wait_writes(struct bitmap *bitmap)
 427{
 428	if (bitmap->storage.file)
 429		wait_event(bitmap->write_wait,
 430			   atomic_read(&bitmap->pending_writes)==0);
 431	else
 432		/* Note that we ignore the return value.  The writes
 433		 * might have failed, but that would just mean that
 434		 * some bits which should be cleared haven't been,
 435		 * which is safe.  The relevant bitmap blocks will
 436		 * probably get written again, but there is no great
 437		 * loss if they aren't.
 438		 */
 439		md_super_wait(bitmap->mddev);
 440}
 441
 442
 443/* update the event counter and sync the superblock to disk */
 444void md_bitmap_update_sb(struct bitmap *bitmap)
 445{
 446	bitmap_super_t *sb;
 447
 448	if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
 449		return;
 450	if (bitmap->mddev->bitmap_info.external)
 451		return;
 452	if (!bitmap->storage.sb_page) /* no superblock */
 453		return;
 454	sb = kmap_atomic(bitmap->storage.sb_page);
 455	sb->events = cpu_to_le64(bitmap->mddev->events);
 456	if (bitmap->mddev->events < bitmap->events_cleared)
 457		/* rocking back to read-only */
 458		bitmap->events_cleared = bitmap->mddev->events;
 459	sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
 460	/*
 461	 * clear BITMAP_WRITE_ERROR bit to protect against the case that
 462	 * a bitmap write error occurred but the later writes succeeded.
 463	 */
 464	sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR));
 465	/* Just in case these have been changed via sysfs: */
 466	sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
 467	sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
 468	/* This might have been changed by a reshape */
 469	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 470	sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
 471	sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
 472	sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
 473					   bitmap_info.space);
 474	kunmap_atomic(sb);
 475	write_page(bitmap, bitmap->storage.sb_page, 1);
 
 
 
 
 
 476}
 477EXPORT_SYMBOL(md_bitmap_update_sb);
 478
 479/* print out the bitmap file superblock */
 480void md_bitmap_print_sb(struct bitmap *bitmap)
 481{
 482	bitmap_super_t *sb;
 483
 484	if (!bitmap || !bitmap->storage.sb_page)
 485		return;
 486	sb = kmap_atomic(bitmap->storage.sb_page);
 487	pr_debug("%s: bitmap file superblock:\n", bmname(bitmap));
 488	pr_debug("         magic: %08x\n", le32_to_cpu(sb->magic));
 489	pr_debug("       version: %d\n", le32_to_cpu(sb->version));
 490	pr_debug("          uuid: %08x.%08x.%08x.%08x\n",
 491		 le32_to_cpu(*(__le32 *)(sb->uuid+0)),
 492		 le32_to_cpu(*(__le32 *)(sb->uuid+4)),
 493		 le32_to_cpu(*(__le32 *)(sb->uuid+8)),
 494		 le32_to_cpu(*(__le32 *)(sb->uuid+12)));
 495	pr_debug("        events: %llu\n",
 496		 (unsigned long long) le64_to_cpu(sb->events));
 497	pr_debug("events cleared: %llu\n",
 498		 (unsigned long long) le64_to_cpu(sb->events_cleared));
 499	pr_debug("         state: %08x\n", le32_to_cpu(sb->state));
 500	pr_debug("     chunksize: %d B\n", le32_to_cpu(sb->chunksize));
 501	pr_debug("  daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
 502	pr_debug("     sync size: %llu KB\n",
 503		 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
 504	pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind));
 505	kunmap_atomic(sb);
 506}
 507
 508/*
 509 * bitmap_new_disk_sb
 510 * @bitmap
 511 *
 512 * This function is somewhat the reverse of bitmap_read_sb.  bitmap_read_sb
 513 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
 514 * This function verifies 'bitmap_info' and populates the on-disk bitmap
 515 * structure, which is to be written to disk.
 516 *
 517 * Returns: 0 on success, -Exxx on error
 518 */
 519static int md_bitmap_new_disk_sb(struct bitmap *bitmap)
 520{
 521	bitmap_super_t *sb;
 522	unsigned long chunksize, daemon_sleep, write_behind;
 523
 524	bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 525	if (bitmap->storage.sb_page == NULL)
 526		return -ENOMEM;
 527	bitmap->storage.sb_page->index = 0;
 528
 529	sb = kmap_atomic(bitmap->storage.sb_page);
 530
 531	sb->magic = cpu_to_le32(BITMAP_MAGIC);
 532	sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
 533
 534	chunksize = bitmap->mddev->bitmap_info.chunksize;
 535	BUG_ON(!chunksize);
 536	if (!is_power_of_2(chunksize)) {
 537		kunmap_atomic(sb);
 538		pr_warn("bitmap chunksize not a power of 2\n");
 539		return -EINVAL;
 540	}
 541	sb->chunksize = cpu_to_le32(chunksize);
 542
 543	daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
 544	if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
 545		pr_debug("Choosing daemon_sleep default (5 sec)\n");
 546		daemon_sleep = 5 * HZ;
 547	}
 548	sb->daemon_sleep = cpu_to_le32(daemon_sleep);
 549	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 550
 551	/*
 552	 * FIXME: write_behind for RAID1.  If not specified, what
 553	 * is a good choice?  We choose COUNTER_MAX / 2 arbitrarily.
 554	 */
 555	write_behind = bitmap->mddev->bitmap_info.max_write_behind;
 556	if (write_behind > COUNTER_MAX)
 557		write_behind = COUNTER_MAX / 2;
 558	sb->write_behind = cpu_to_le32(write_behind);
 559	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 560
 561	/* keep the array size field of the bitmap superblock up to date */
 562	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 563
 564	memcpy(sb->uuid, bitmap->mddev->uuid, 16);
 565
 566	set_bit(BITMAP_STALE, &bitmap->flags);
 567	sb->state = cpu_to_le32(bitmap->flags);
 568	bitmap->events_cleared = bitmap->mddev->events;
 569	sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
 570	bitmap->mddev->bitmap_info.nodes = 0;
 571
 572	kunmap_atomic(sb);
 573
 574	return 0;
 575}
 576
 577/* read the superblock from the bitmap file and initialize some bitmap fields */
 578static int md_bitmap_read_sb(struct bitmap *bitmap)
 579{
 580	char *reason = NULL;
 581	bitmap_super_t *sb;
 582	unsigned long chunksize, daemon_sleep, write_behind;
 583	unsigned long long events;
 584	int nodes = 0;
 585	unsigned long sectors_reserved = 0;
 586	int err = -EINVAL;
 587	struct page *sb_page;
 588	loff_t offset = bitmap->mddev->bitmap_info.offset;
 589
 590	if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
 591		chunksize = 128 * 1024 * 1024;
 592		daemon_sleep = 5 * HZ;
 593		write_behind = 0;
 594		set_bit(BITMAP_STALE, &bitmap->flags);
 595		err = 0;
 596		goto out_no_sb;
 597	}
 598	/* page 0 is the superblock, read it... */
 599	sb_page = alloc_page(GFP_KERNEL);
 600	if (!sb_page)
 601		return -ENOMEM;
 602	bitmap->storage.sb_page = sb_page;
 603
 604re_read:
 605	/* If cluster_slot is set, the cluster is setup */
 606	if (bitmap->cluster_slot >= 0) {
 607		sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
 608
 609		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks,
 610			   (bitmap->mddev->bitmap_info.chunksize >> 9));
 611		/* bits to bytes */
 612		bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
 613		/* to 4k blocks */
 614		bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
 615		offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
 616		pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
 617			bitmap->cluster_slot, offset);
 618	}
 619
 620	if (bitmap->storage.file) {
 621		loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
 622		int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
 623
 624		err = read_page(bitmap->storage.file, 0,
 625				bitmap, bytes, sb_page);
 626	} else {
 627		err = read_sb_page(bitmap->mddev,
 628				   offset,
 629				   sb_page,
 630				   0, sizeof(bitmap_super_t));
 631	}
 632	if (err)
 633		return err;
 634
 635	err = -EINVAL;
 636	sb = kmap_atomic(sb_page);
 637
 638	chunksize = le32_to_cpu(sb->chunksize);
 639	daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
 640	write_behind = le32_to_cpu(sb->write_behind);
 641	sectors_reserved = le32_to_cpu(sb->sectors_reserved);
 642	/* Setup nodes/clustername only if bitmap version is
 643	 * cluster-compatible
 644	 */
 645	if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
 646		nodes = le32_to_cpu(sb->nodes);
 647		strlcpy(bitmap->mddev->bitmap_info.cluster_name,
 648				sb->cluster_name, 64);
 649	}
 650
 651	/* verify that the bitmap-specific fields are valid */
 652	if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
 653		reason = "bad magic";
 654	else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
 655		 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
 656		reason = "unrecognized superblock version";
 657	else if (chunksize < 512)
 658		reason = "bitmap chunksize too small";
 659	else if (!is_power_of_2(chunksize))
 660		reason = "bitmap chunksize not a power of 2";
 661	else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
 662		reason = "daemon sleep period out of range";
 663	else if (write_behind > COUNTER_MAX)
 664		reason = "write-behind limit out of range (0 - 16383)";
 665	if (reason) {
 666		pr_warn("%s: invalid bitmap file superblock: %s\n",
 667			bmname(bitmap), reason);
 668		goto out;
 669	}
 670
 
 
 
 
 
 
 
 
 
 
 671	/* keep the array size field of the bitmap superblock up to date */
 672	sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
 673
 674	if (bitmap->mddev->persistent) {
 675		/*
 676		 * We have a persistent array superblock, so compare the
 677		 * bitmap's UUID and event counter to the mddev's
 678		 */
 679		if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
 680			pr_warn("%s: bitmap superblock UUID mismatch\n",
 681				bmname(bitmap));
 682			goto out;
 683		}
 684		events = le64_to_cpu(sb->events);
 685		if (!nodes && (events < bitmap->mddev->events)) {
 686			pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n",
 687				bmname(bitmap), events,
 688				(unsigned long long) bitmap->mddev->events);
 689			set_bit(BITMAP_STALE, &bitmap->flags);
 690		}
 691	}
 692
 693	/* assign fields using values from superblock */
 694	bitmap->flags |= le32_to_cpu(sb->state);
 695	if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
 696		set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
 697	bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
 698	strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
 699	err = 0;
 700
 701out:
 702	kunmap_atomic(sb);
 703	/* Assigning chunksize is required for "re_read" */
 704	bitmap->mddev->bitmap_info.chunksize = chunksize;
 705	if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
 
 
 706		err = md_setup_cluster(bitmap->mddev, nodes);
 707		if (err) {
 708			pr_warn("%s: Could not setup cluster service (%d)\n",
 709				bmname(bitmap), err);
 710			goto out_no_sb;
 711		}
 712		bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
 713		goto re_read;
 714	}
 715
 716
 717out_no_sb:
 718	if (test_bit(BITMAP_STALE, &bitmap->flags))
 719		bitmap->events_cleared = bitmap->mddev->events;
 720	bitmap->mddev->bitmap_info.chunksize = chunksize;
 721	bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
 722	bitmap->mddev->bitmap_info.max_write_behind = write_behind;
 723	bitmap->mddev->bitmap_info.nodes = nodes;
 724	if (bitmap->mddev->bitmap_info.space == 0 ||
 725	    bitmap->mddev->bitmap_info.space > sectors_reserved)
 726		bitmap->mddev->bitmap_info.space = sectors_reserved;
 727	if (err) {
 
 728		md_bitmap_print_sb(bitmap);
 729		if (bitmap->cluster_slot < 0)
 730			md_cluster_stop(bitmap->mddev);
 731	}
 732	return err;
 733}
 734
 735/*
 736 * general bitmap file operations
 737 */
 738
 739/*
 740 * on-disk bitmap:
 741 *
 742 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 743 * file a page at a time. There's a superblock at the start of the file.
 744 */
 745/* calculate the index of the page that contains this bit */
 746static inline unsigned long file_page_index(struct bitmap_storage *store,
 747					    unsigned long chunk)
 748{
 749	if (store->sb_page)
 750		chunk += sizeof(bitmap_super_t) << 3;
 751	return chunk >> PAGE_BIT_SHIFT;
 752}
 753
 754/* calculate the (bit) offset of this bit within a page */
 755static inline unsigned long file_page_offset(struct bitmap_storage *store,
 756					     unsigned long chunk)
 757{
 758	if (store->sb_page)
 759		chunk += sizeof(bitmap_super_t) << 3;
 760	return chunk & (PAGE_BITS - 1);
 761}
 762
 763/*
 764 * return a pointer to the page in the filemap that contains the given bit
 765 *
 766 */
 767static inline struct page *filemap_get_page(struct bitmap_storage *store,
 768					    unsigned long chunk)
 769{
 770	if (file_page_index(store, chunk) >= store->file_pages)
 771		return NULL;
 772	return store->filemap[file_page_index(store, chunk)];
 773}
 774
 775static int md_bitmap_storage_alloc(struct bitmap_storage *store,
 776				   unsigned long chunks, int with_super,
 777				   int slot_number)
 778{
 779	int pnum, offset = 0;
 780	unsigned long num_pages;
 781	unsigned long bytes;
 782
 783	bytes = DIV_ROUND_UP(chunks, 8);
 784	if (with_super)
 785		bytes += sizeof(bitmap_super_t);
 786
 787	num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
 788	offset = slot_number * num_pages;
 789
 790	store->filemap = kmalloc_array(num_pages, sizeof(struct page *),
 791				       GFP_KERNEL);
 792	if (!store->filemap)
 793		return -ENOMEM;
 794
 795	if (with_super && !store->sb_page) {
 796		store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
 797		if (store->sb_page == NULL)
 798			return -ENOMEM;
 799	}
 800
 801	pnum = 0;
 802	if (store->sb_page) {
 803		store->filemap[0] = store->sb_page;
 804		pnum = 1;
 805		store->sb_page->index = offset;
 806	}
 807
 808	for ( ; pnum < num_pages; pnum++) {
 809		store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
 810		if (!store->filemap[pnum]) {
 811			store->file_pages = pnum;
 812			return -ENOMEM;
 813		}
 814		store->filemap[pnum]->index = pnum + offset;
 815	}
 816	store->file_pages = pnum;
 817
 818	/* We need 4 bits per page, rounded up to a multiple
 819	 * of sizeof(unsigned long) */
 820	store->filemap_attr = kzalloc(
 821		roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
 822		GFP_KERNEL);
 823	if (!store->filemap_attr)
 824		return -ENOMEM;
 825
 826	store->bytes = bytes;
 827
 828	return 0;
 829}
 830
 831static void md_bitmap_file_unmap(struct bitmap_storage *store)
 832{
 833	struct page **map, *sb_page;
 834	int pages;
 835	struct file *file;
 836
 837	file = store->file;
 838	map = store->filemap;
 839	pages = store->file_pages;
 840	sb_page = store->sb_page;
 841
 842	while (pages--)
 843		if (map[pages] != sb_page) /* 0 is sb_page, release it below */
 844			free_buffers(map[pages]);
 845	kfree(map);
 846	kfree(store->filemap_attr);
 847
 848	if (sb_page)
 849		free_buffers(sb_page);
 850
 851	if (file) {
 852		struct inode *inode = file_inode(file);
 853		invalidate_mapping_pages(inode->i_mapping, 0, -1);
 854		fput(file);
 855	}
 856}
 857
 858/*
 859 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
 860 * then it is no longer reliable, so we stop using it and we mark the file
 861 * as failed in the superblock
 862 */
 863static void md_bitmap_file_kick(struct bitmap *bitmap)
 864{
 865	char *path, *ptr = NULL;
 866
 867	if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
 868		md_bitmap_update_sb(bitmap);
 869
 870		if (bitmap->storage.file) {
 871			path = kmalloc(PAGE_SIZE, GFP_KERNEL);
 872			if (path)
 873				ptr = file_path(bitmap->storage.file,
 874					     path, PAGE_SIZE);
 875
 876			pr_warn("%s: kicking failed bitmap file %s from array!\n",
 877				bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
 878
 879			kfree(path);
 880		} else
 881			pr_warn("%s: disabling internal bitmap due to errors\n",
 882				bmname(bitmap));
 883	}
 884}
 885
 886enum bitmap_page_attr {
 887	BITMAP_PAGE_DIRTY = 0,     /* there are set bits that need to be synced */
 888	BITMAP_PAGE_PENDING = 1,   /* there are bits that are being cleaned.
 889				    * i.e. counter is 1 or 2. */
 890	BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
 891};
 892
 893static inline void set_page_attr(struct bitmap *bitmap, int pnum,
 894				 enum bitmap_page_attr attr)
 895{
 896	set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 897}
 898
 899static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
 900				   enum bitmap_page_attr attr)
 901{
 902	clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 903}
 904
 905static inline int test_page_attr(struct bitmap *bitmap, int pnum,
 906				 enum bitmap_page_attr attr)
 907{
 908	return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
 909}
 910
 911static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
 912					   enum bitmap_page_attr attr)
 913{
 914	return test_and_clear_bit((pnum<<2) + attr,
 915				  bitmap->storage.filemap_attr);
 916}
 917/*
 918 * bitmap_file_set_bit -- called before performing a write to the md device
 919 * to set (and eventually sync) a particular bit in the bitmap file
 920 *
 921 * we set the bit immediately, then we record the page number so that
 922 * when an unplug occurs, we can flush the dirty pages out to disk
 923 */
 924static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
 925{
 926	unsigned long bit;
 927	struct page *page;
 928	void *kaddr;
 929	unsigned long chunk = block >> bitmap->counts.chunkshift;
 930	struct bitmap_storage *store = &bitmap->storage;
 
 931	unsigned long node_offset = 0;
 932
 933	if (mddev_is_clustered(bitmap->mddev))
 934		node_offset = bitmap->cluster_slot * store->file_pages;
 935
 936	page = filemap_get_page(&bitmap->storage, chunk);
 937	if (!page)
 938		return;
 939	bit = file_page_offset(&bitmap->storage, chunk);
 940
 941	/* set the bit */
 942	kaddr = kmap_atomic(page);
 943	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 944		set_bit(bit, kaddr);
 945	else
 946		set_bit_le(bit, kaddr);
 947	kunmap_atomic(kaddr);
 948	pr_debug("set file bit %lu page %lu\n", bit, page->index);
 949	/* record page number so it gets flushed to disk when unplug occurs */
 950	set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY);
 951}
 952
 953static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
 954{
 955	unsigned long bit;
 956	struct page *page;
 957	void *paddr;
 958	unsigned long chunk = block >> bitmap->counts.chunkshift;
 959	struct bitmap_storage *store = &bitmap->storage;
 
 960	unsigned long node_offset = 0;
 961
 962	if (mddev_is_clustered(bitmap->mddev))
 963		node_offset = bitmap->cluster_slot * store->file_pages;
 964
 965	page = filemap_get_page(&bitmap->storage, chunk);
 966	if (!page)
 967		return;
 968	bit = file_page_offset(&bitmap->storage, chunk);
 969	paddr = kmap_atomic(page);
 970	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 971		clear_bit(bit, paddr);
 972	else
 973		clear_bit_le(bit, paddr);
 974	kunmap_atomic(paddr);
 975	if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) {
 976		set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING);
 977		bitmap->allclean = 0;
 978	}
 979}
 980
 981static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
 982{
 983	unsigned long bit;
 984	struct page *page;
 985	void *paddr;
 986	unsigned long chunk = block >> bitmap->counts.chunkshift;
 987	int set = 0;
 988
 989	page = filemap_get_page(&bitmap->storage, chunk);
 990	if (!page)
 991		return -EINVAL;
 992	bit = file_page_offset(&bitmap->storage, chunk);
 993	paddr = kmap_atomic(page);
 994	if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
 995		set = test_bit(bit, paddr);
 996	else
 997		set = test_bit_le(bit, paddr);
 998	kunmap_atomic(paddr);
 999	return set;
1000}
1001
1002
1003/* this gets called when the md device is ready to unplug its underlying
1004 * (slave) device queues -- before we let any writes go down, we need to
1005 * sync the dirty pages of the bitmap file to disk */
1006void md_bitmap_unplug(struct bitmap *bitmap)
1007{
1008	unsigned long i;
1009	int dirty, need_write;
1010	int writing = 0;
1011
1012	if (!bitmap || !bitmap->storage.filemap ||
1013	    test_bit(BITMAP_STALE, &bitmap->flags))
1014		return;
1015
1016	/* look at each page to see if there are any set bits that need to be
1017	 * flushed out to disk */
1018	for (i = 0; i < bitmap->storage.file_pages; i++) {
1019		dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1020		need_write = test_and_clear_page_attr(bitmap, i,
1021						      BITMAP_PAGE_NEEDWRITE);
1022		if (dirty || need_write) {
1023			if (!writing) {
1024				md_bitmap_wait_writes(bitmap);
1025				if (bitmap->mddev->queue)
1026					blk_add_trace_msg(bitmap->mddev->queue,
1027							  "md bitmap_unplug");
1028			}
1029			clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
1030			write_page(bitmap, bitmap->storage.filemap[i], 0);
1031			writing = 1;
1032		}
1033	}
1034	if (writing)
1035		md_bitmap_wait_writes(bitmap);
1036
1037	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1038		md_bitmap_file_kick(bitmap);
1039}
1040EXPORT_SYMBOL(md_bitmap_unplug);
1041
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1042static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1043/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1044 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1045 * memory mapping of the bitmap file
1046 * Special cases:
1047 *   if there's no bitmap file, or if the bitmap file had been
1048 *   previously kicked from the array, we mark all the bits as
1049 *   1's in order to cause a full resync.
 
1050 *
1051 * We ignore all bits for sectors that end earlier than 'start'.
1052 * This is used when reading an out-of-date bitmap...
1053 */
1054static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1055{
1056	unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1057	struct page *page = NULL;
 
 
 
 
1058	unsigned long bit_cnt = 0;
1059	struct file *file;
1060	unsigned long offset;
1061	int outofdate;
1062	int ret = -ENOSPC;
1063	void *paddr;
1064	struct bitmap_storage *store = &bitmap->storage;
1065
1066	chunks = bitmap->counts.chunks;
1067	file = store->file;
1068
1069	if (!file && !bitmap->mddev->bitmap_info.offset) {
1070		/* No permanent bitmap - fill with '1s'. */
1071		store->filemap = NULL;
1072		store->file_pages = 0;
1073		for (i = 0; i < chunks ; i++) {
1074			/* if the disk bit is set, set the memory bit */
1075			int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1076				      >= start);
1077			md_bitmap_set_memory_bits(bitmap,
1078						  (sector_t)i << bitmap->counts.chunkshift,
1079						  needed);
1080		}
1081		return 0;
1082	}
1083
1084	outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1085	if (outofdate)
1086		pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap));
1087
1088	if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1089		pr_warn("%s: bitmap file too short %lu < %lu\n",
1090			bmname(bitmap),
1091			(unsigned long) i_size_read(file->f_mapping->host),
1092			store->bytes);
 
1093		goto err;
1094	}
1095
1096	oldindex = ~0L;
1097	offset = 0;
1098	if (!bitmap->mddev->bitmap_info.external)
1099		offset = sizeof(bitmap_super_t);
1100
1101	if (mddev_is_clustered(bitmap->mddev))
1102		node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1103
1104	for (i = 0; i < chunks; i++) {
1105		int b;
1106		index = file_page_index(&bitmap->storage, i);
1107		bit = file_page_offset(&bitmap->storage, i);
1108		if (index != oldindex) { /* this is a new page, read it in */
1109			int count;
1110			/* unmap the old page, we're done with it */
1111			if (index == store->file_pages-1)
1112				count = store->bytes - index * PAGE_SIZE;
1113			else
1114				count = PAGE_SIZE;
1115			page = store->filemap[index];
1116			if (file)
1117				ret = read_page(file, index, bitmap,
1118						count, page);
1119			else
1120				ret = read_sb_page(
1121					bitmap->mddev,
1122					bitmap->mddev->bitmap_info.offset,
1123					page,
1124					index + node_offset, count);
1125
1126			if (ret)
1127				goto err;
1128
1129			oldindex = index;
1130
1131			if (outofdate) {
1132				/*
1133				 * if bitmap is out of date, dirty the
1134				 * whole page and write it out
1135				 */
1136				paddr = kmap_atomic(page);
1137				memset(paddr + offset, 0xff,
1138				       PAGE_SIZE - offset);
1139				kunmap_atomic(paddr);
1140				write_page(bitmap, page, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
1141
 
 
1142				ret = -EIO;
1143				if (test_bit(BITMAP_WRITE_ERROR,
1144					     &bitmap->flags))
1145					goto err;
1146			}
1147		}
 
 
 
 
 
 
 
 
1148		paddr = kmap_atomic(page);
1149		if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1150			b = test_bit(bit, paddr);
1151		else
1152			b = test_bit_le(bit, paddr);
1153		kunmap_atomic(paddr);
1154		if (b) {
 
1155			/* if the disk bit is set, set the memory bit */
1156			int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1157				      >= start);
1158			md_bitmap_set_memory_bits(bitmap,
1159						  (sector_t)i << bitmap->counts.chunkshift,
1160						  needed);
1161			bit_cnt++;
1162		}
1163		offset = 0;
1164	}
1165
1166	pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n",
1167		 bmname(bitmap), store->file_pages,
1168		 bit_cnt, chunks);
1169
1170	return 0;
1171
1172 err:
1173	pr_warn("%s: bitmap initialisation failed: %d\n",
1174		bmname(bitmap), ret);
1175	return ret;
1176}
1177
1178void md_bitmap_write_all(struct bitmap *bitmap)
1179{
1180	/* We don't actually write all bitmap blocks here,
1181	 * just flag them as needing to be written
1182	 */
1183	int i;
1184
1185	if (!bitmap || !bitmap->storage.filemap)
1186		return;
1187	if (bitmap->storage.file)
1188		/* Only one copy, so nothing needed */
1189		return;
1190
1191	for (i = 0; i < bitmap->storage.file_pages; i++)
1192		set_page_attr(bitmap, i,
1193			      BITMAP_PAGE_NEEDWRITE);
1194	bitmap->allclean = 0;
1195}
1196
1197static void md_bitmap_count_page(struct bitmap_counts *bitmap,
1198				 sector_t offset, int inc)
1199{
1200	sector_t chunk = offset >> bitmap->chunkshift;
1201	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1202	bitmap->bp[page].count += inc;
1203	md_bitmap_checkfree(bitmap, page);
1204}
1205
1206static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1207{
1208	sector_t chunk = offset >> bitmap->chunkshift;
1209	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1210	struct bitmap_page *bp = &bitmap->bp[page];
1211
1212	if (!bp->pending)
1213		bp->pending = 1;
1214}
1215
1216static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1217					       sector_t offset, sector_t *blocks,
1218					       int create);
1219
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1220/*
1221 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1222 *			out to disk
1223 */
1224
1225void md_bitmap_daemon_work(struct mddev *mddev)
1226{
1227	struct bitmap *bitmap;
1228	unsigned long j;
1229	unsigned long nextpage;
1230	sector_t blocks;
1231	struct bitmap_counts *counts;
1232
1233	/* Use a mutex to guard daemon_work against
1234	 * bitmap_destroy.
1235	 */
1236	mutex_lock(&mddev->bitmap_info.mutex);
1237	bitmap = mddev->bitmap;
1238	if (bitmap == NULL) {
1239		mutex_unlock(&mddev->bitmap_info.mutex);
1240		return;
1241	}
1242	if (time_before(jiffies, bitmap->daemon_lastrun
1243			+ mddev->bitmap_info.daemon_sleep))
1244		goto done;
1245
1246	bitmap->daemon_lastrun = jiffies;
1247	if (bitmap->allclean) {
1248		mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1249		goto done;
1250	}
1251	bitmap->allclean = 1;
1252
1253	if (bitmap->mddev->queue)
1254		blk_add_trace_msg(bitmap->mddev->queue,
1255				  "md bitmap_daemon_work");
1256
1257	/* Any file-page which is PENDING now needs to be written.
1258	 * So set NEEDWRITE now, then after we make any last-minute changes
1259	 * we will write it.
1260	 */
1261	for (j = 0; j < bitmap->storage.file_pages; j++)
1262		if (test_and_clear_page_attr(bitmap, j,
1263					     BITMAP_PAGE_PENDING))
1264			set_page_attr(bitmap, j,
1265				      BITMAP_PAGE_NEEDWRITE);
1266
1267	if (bitmap->need_sync &&
1268	    mddev->bitmap_info.external == 0) {
1269		/* Arrange for superblock update as well as
1270		 * other changes */
1271		bitmap_super_t *sb;
1272		bitmap->need_sync = 0;
1273		if (bitmap->storage.filemap) {
1274			sb = kmap_atomic(bitmap->storage.sb_page);
1275			sb->events_cleared =
1276				cpu_to_le64(bitmap->events_cleared);
1277			kunmap_atomic(sb);
1278			set_page_attr(bitmap, 0,
1279				      BITMAP_PAGE_NEEDWRITE);
1280		}
1281	}
1282	/* Now look at the bitmap counters and if any are '2' or '1',
1283	 * decrement and handle accordingly.
1284	 */
1285	counts = &bitmap->counts;
1286	spin_lock_irq(&counts->lock);
1287	nextpage = 0;
1288	for (j = 0; j < counts->chunks; j++) {
1289		bitmap_counter_t *bmc;
1290		sector_t  block = (sector_t)j << counts->chunkshift;
1291
1292		if (j == nextpage) {
1293			nextpage += PAGE_COUNTER_RATIO;
1294			if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1295				j |= PAGE_COUNTER_MASK;
1296				continue;
1297			}
1298			counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1299		}
1300
1301		bmc = md_bitmap_get_counter(counts, block, &blocks, 0);
1302		if (!bmc) {
1303			j |= PAGE_COUNTER_MASK;
1304			continue;
1305		}
1306		if (*bmc == 1 && !bitmap->need_sync) {
1307			/* We can clear the bit */
1308			*bmc = 0;
1309			md_bitmap_count_page(counts, block, -1);
1310			md_bitmap_file_clear_bit(bitmap, block);
1311		} else if (*bmc && *bmc <= 2) {
1312			*bmc = 1;
1313			md_bitmap_set_pending(counts, block);
1314			bitmap->allclean = 0;
1315		}
1316	}
1317	spin_unlock_irq(&counts->lock);
1318
1319	md_bitmap_wait_writes(bitmap);
1320	/* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1321	 * DIRTY pages need to be written by bitmap_unplug so it can wait
1322	 * for them.
1323	 * If we find any DIRTY page we stop there and let bitmap_unplug
1324	 * handle all the rest.  This is important in the case where
1325	 * the first blocking holds the superblock and it has been updated.
1326	 * We mustn't write any other blocks before the superblock.
1327	 */
1328	for (j = 0;
1329	     j < bitmap->storage.file_pages
1330		     && !test_bit(BITMAP_STALE, &bitmap->flags);
1331	     j++) {
1332		if (test_page_attr(bitmap, j,
1333				   BITMAP_PAGE_DIRTY))
1334			/* bitmap_unplug will handle the rest */
1335			break;
1336		if (bitmap->storage.filemap &&
1337		    test_and_clear_page_attr(bitmap, j,
1338					     BITMAP_PAGE_NEEDWRITE)) {
1339			write_page(bitmap, bitmap->storage.filemap[j], 0);
1340		}
1341	}
1342
1343 done:
1344	if (bitmap->allclean == 0)
1345		mddev->thread->timeout =
1346			mddev->bitmap_info.daemon_sleep;
1347	mutex_unlock(&mddev->bitmap_info.mutex);
1348}
1349
1350static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap,
1351					       sector_t offset, sector_t *blocks,
1352					       int create)
1353__releases(bitmap->lock)
1354__acquires(bitmap->lock)
1355{
1356	/* If 'create', we might release the lock and reclaim it.
1357	 * The lock must have been taken with interrupts enabled.
1358	 * If !create, we don't release the lock.
1359	 */
1360	sector_t chunk = offset >> bitmap->chunkshift;
1361	unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1362	unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1363	sector_t csize;
1364	int err;
1365
 
 
 
 
 
 
 
 
1366	err = md_bitmap_checkpage(bitmap, page, create, 0);
1367
1368	if (bitmap->bp[page].hijacked ||
1369	    bitmap->bp[page].map == NULL)
1370		csize = ((sector_t)1) << (bitmap->chunkshift +
1371					  PAGE_COUNTER_SHIFT);
1372	else
1373		csize = ((sector_t)1) << bitmap->chunkshift;
1374	*blocks = csize - (offset & (csize - 1));
1375
1376	if (err < 0)
1377		return NULL;
1378
1379	/* now locked ... */
1380
1381	if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1382		/* should we use the first or second counter field
1383		 * of the hijacked pointer? */
1384		int hi = (pageoff > PAGE_COUNTER_MASK);
1385		return  &((bitmap_counter_t *)
1386			  &bitmap->bp[page].map)[hi];
1387	} else /* page is allocated */
1388		return (bitmap_counter_t *)
1389			&(bitmap->bp[page].map[pageoff]);
1390}
1391
1392int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1393{
1394	if (!bitmap)
1395		return 0;
1396
1397	if (behind) {
1398		int bw;
1399		atomic_inc(&bitmap->behind_writes);
1400		bw = atomic_read(&bitmap->behind_writes);
1401		if (bw > bitmap->behind_writes_used)
1402			bitmap->behind_writes_used = bw;
1403
1404		pr_debug("inc write-behind count %d/%lu\n",
1405			 bw, bitmap->mddev->bitmap_info.max_write_behind);
1406	}
1407
1408	while (sectors) {
1409		sector_t blocks;
1410		bitmap_counter_t *bmc;
1411
1412		spin_lock_irq(&bitmap->counts.lock);
1413		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1414		if (!bmc) {
1415			spin_unlock_irq(&bitmap->counts.lock);
1416			return 0;
1417		}
1418
1419		if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1420			DEFINE_WAIT(__wait);
1421			/* note that it is safe to do the prepare_to_wait
1422			 * after the test as long as we do it before dropping
1423			 * the spinlock.
1424			 */
1425			prepare_to_wait(&bitmap->overflow_wait, &__wait,
1426					TASK_UNINTERRUPTIBLE);
1427			spin_unlock_irq(&bitmap->counts.lock);
1428			schedule();
1429			finish_wait(&bitmap->overflow_wait, &__wait);
1430			continue;
1431		}
1432
1433		switch (*bmc) {
1434		case 0:
1435			md_bitmap_file_set_bit(bitmap, offset);
1436			md_bitmap_count_page(&bitmap->counts, offset, 1);
1437			fallthrough;
1438		case 1:
1439			*bmc = 2;
1440		}
1441
1442		(*bmc)++;
1443
1444		spin_unlock_irq(&bitmap->counts.lock);
1445
1446		offset += blocks;
1447		if (sectors > blocks)
1448			sectors -= blocks;
1449		else
1450			sectors = 0;
1451	}
1452	return 0;
1453}
1454EXPORT_SYMBOL(md_bitmap_startwrite);
1455
1456void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
1457			unsigned long sectors, int success, int behind)
1458{
1459	if (!bitmap)
1460		return;
1461	if (behind) {
1462		if (atomic_dec_and_test(&bitmap->behind_writes))
1463			wake_up(&bitmap->behind_wait);
1464		pr_debug("dec write-behind count %d/%lu\n",
1465			 atomic_read(&bitmap->behind_writes),
1466			 bitmap->mddev->bitmap_info.max_write_behind);
1467	}
1468
1469	while (sectors) {
1470		sector_t blocks;
1471		unsigned long flags;
1472		bitmap_counter_t *bmc;
1473
1474		spin_lock_irqsave(&bitmap->counts.lock, flags);
1475		bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1476		if (!bmc) {
1477			spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1478			return;
1479		}
1480
1481		if (success && !bitmap->mddev->degraded &&
1482		    bitmap->events_cleared < bitmap->mddev->events) {
1483			bitmap->events_cleared = bitmap->mddev->events;
1484			bitmap->need_sync = 1;
1485			sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1486		}
1487
1488		if (!success && !NEEDED(*bmc))
1489			*bmc |= NEEDED_MASK;
1490
1491		if (COUNTER(*bmc) == COUNTER_MAX)
1492			wake_up(&bitmap->overflow_wait);
1493
1494		(*bmc)--;
1495		if (*bmc <= 2) {
1496			md_bitmap_set_pending(&bitmap->counts, offset);
1497			bitmap->allclean = 0;
1498		}
1499		spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1500		offset += blocks;
1501		if (sectors > blocks)
1502			sectors -= blocks;
1503		else
1504			sectors = 0;
1505	}
1506}
1507EXPORT_SYMBOL(md_bitmap_endwrite);
1508
1509static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1510			       int degraded)
1511{
1512	bitmap_counter_t *bmc;
1513	int rv;
1514	if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1515		*blocks = 1024;
1516		return 1; /* always resync if no bitmap */
1517	}
1518	spin_lock_irq(&bitmap->counts.lock);
1519	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1520	rv = 0;
1521	if (bmc) {
1522		/* locked */
1523		if (RESYNC(*bmc))
1524			rv = 1;
1525		else if (NEEDED(*bmc)) {
1526			rv = 1;
1527			if (!degraded) { /* don't set/clear bits if degraded */
1528				*bmc |= RESYNC_MASK;
1529				*bmc &= ~NEEDED_MASK;
1530			}
1531		}
1532	}
1533	spin_unlock_irq(&bitmap->counts.lock);
1534	return rv;
1535}
1536
1537int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1538			 int degraded)
1539{
1540	/* bitmap_start_sync must always report on multiples of whole
1541	 * pages, otherwise resync (which is very PAGE_SIZE based) will
1542	 * get confused.
1543	 * So call __bitmap_start_sync repeatedly (if needed) until
1544	 * At least PAGE_SIZE>>9 blocks are covered.
1545	 * Return the 'or' of the result.
1546	 */
1547	int rv = 0;
1548	sector_t blocks1;
1549
1550	*blocks = 0;
1551	while (*blocks < (PAGE_SIZE>>9)) {
1552		rv |= __bitmap_start_sync(bitmap, offset,
1553					  &blocks1, degraded);
1554		offset += blocks1;
1555		*blocks += blocks1;
1556	}
1557	return rv;
1558}
1559EXPORT_SYMBOL(md_bitmap_start_sync);
1560
1561void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1562{
1563	bitmap_counter_t *bmc;
1564	unsigned long flags;
1565
1566	if (bitmap == NULL) {
1567		*blocks = 1024;
1568		return;
1569	}
1570	spin_lock_irqsave(&bitmap->counts.lock, flags);
1571	bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1572	if (bmc == NULL)
1573		goto unlock;
1574	/* locked */
1575	if (RESYNC(*bmc)) {
1576		*bmc &= ~RESYNC_MASK;
1577
1578		if (!NEEDED(*bmc) && aborted)
1579			*bmc |= NEEDED_MASK;
1580		else {
1581			if (*bmc <= 2) {
1582				md_bitmap_set_pending(&bitmap->counts, offset);
1583				bitmap->allclean = 0;
1584			}
1585		}
1586	}
1587 unlock:
1588	spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1589}
1590EXPORT_SYMBOL(md_bitmap_end_sync);
1591
1592void md_bitmap_close_sync(struct bitmap *bitmap)
1593{
1594	/* Sync has finished, and any bitmap chunks that weren't synced
1595	 * properly have been aborted.  It remains to us to clear the
1596	 * RESYNC bit wherever it is still on
1597	 */
1598	sector_t sector = 0;
1599	sector_t blocks;
1600	if (!bitmap)
1601		return;
1602	while (sector < bitmap->mddev->resync_max_sectors) {
1603		md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1604		sector += blocks;
1605	}
1606}
1607EXPORT_SYMBOL(md_bitmap_close_sync);
1608
1609void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1610{
1611	sector_t s = 0;
1612	sector_t blocks;
1613
1614	if (!bitmap)
1615		return;
1616	if (sector == 0) {
1617		bitmap->last_end_sync = jiffies;
1618		return;
1619	}
1620	if (!force && time_before(jiffies, (bitmap->last_end_sync
1621				  + bitmap->mddev->bitmap_info.daemon_sleep)))
1622		return;
1623	wait_event(bitmap->mddev->recovery_wait,
1624		   atomic_read(&bitmap->mddev->recovery_active) == 0);
1625
1626	bitmap->mddev->curr_resync_completed = sector;
1627	set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags);
1628	sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1629	s = 0;
1630	while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1631		md_bitmap_end_sync(bitmap, s, &blocks, 0);
1632		s += blocks;
1633	}
1634	bitmap->last_end_sync = jiffies;
1635	sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed);
1636}
1637EXPORT_SYMBOL(md_bitmap_cond_end_sync);
1638
1639void md_bitmap_sync_with_cluster(struct mddev *mddev,
1640			      sector_t old_lo, sector_t old_hi,
1641			      sector_t new_lo, sector_t new_hi)
1642{
1643	struct bitmap *bitmap = mddev->bitmap;
1644	sector_t sector, blocks = 0;
1645
1646	for (sector = old_lo; sector < new_lo; ) {
1647		md_bitmap_end_sync(bitmap, sector, &blocks, 0);
1648		sector += blocks;
1649	}
1650	WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n");
1651
1652	for (sector = old_hi; sector < new_hi; ) {
1653		md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1654		sector += blocks;
1655	}
1656	WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n");
1657}
1658EXPORT_SYMBOL(md_bitmap_sync_with_cluster);
1659
1660static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1661{
1662	/* For each chunk covered by any of these sectors, set the
1663	 * counter to 2 and possibly set resync_needed.  They should all
1664	 * be 0 at this point
1665	 */
1666
1667	sector_t secs;
1668	bitmap_counter_t *bmc;
1669	spin_lock_irq(&bitmap->counts.lock);
1670	bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1671	if (!bmc) {
1672		spin_unlock_irq(&bitmap->counts.lock);
1673		return;
1674	}
1675	if (!*bmc) {
1676		*bmc = 2;
1677		md_bitmap_count_page(&bitmap->counts, offset, 1);
1678		md_bitmap_set_pending(&bitmap->counts, offset);
1679		bitmap->allclean = 0;
1680	}
1681	if (needed)
1682		*bmc |= NEEDED_MASK;
1683	spin_unlock_irq(&bitmap->counts.lock);
1684}
1685
1686/* dirty the memory and file bits for bitmap chunks "s" to "e" */
1687void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1688{
1689	unsigned long chunk;
1690
1691	for (chunk = s; chunk <= e; chunk++) {
1692		sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1693		md_bitmap_set_memory_bits(bitmap, sec, 1);
1694		md_bitmap_file_set_bit(bitmap, sec);
1695		if (sec < bitmap->mddev->recovery_cp)
1696			/* We are asserting that the array is dirty,
1697			 * so move the recovery_cp address back so
1698			 * that it is obvious that it is dirty
1699			 */
1700			bitmap->mddev->recovery_cp = sec;
1701	}
1702}
1703
1704/*
1705 * flush out any pending updates
1706 */
1707void md_bitmap_flush(struct mddev *mddev)
1708{
1709	struct bitmap *bitmap = mddev->bitmap;
1710	long sleep;
1711
1712	if (!bitmap) /* there was no bitmap */
1713		return;
1714
1715	/* run the daemon_work three time to ensure everything is flushed
1716	 * that can be
1717	 */
1718	sleep = mddev->bitmap_info.daemon_sleep * 2;
1719	bitmap->daemon_lastrun -= sleep;
1720	md_bitmap_daemon_work(mddev);
1721	bitmap->daemon_lastrun -= sleep;
1722	md_bitmap_daemon_work(mddev);
1723	bitmap->daemon_lastrun -= sleep;
1724	md_bitmap_daemon_work(mddev);
1725	if (mddev->bitmap_info.external)
1726		md_super_wait(mddev);
1727	md_bitmap_update_sb(bitmap);
1728}
1729
1730/*
1731 * free memory that was allocated
1732 */
1733void md_bitmap_free(struct bitmap *bitmap)
1734{
1735	unsigned long k, pages;
1736	struct bitmap_page *bp;
1737
1738	if (!bitmap) /* there was no bitmap */
1739		return;
1740
1741	if (bitmap->sysfs_can_clear)
1742		sysfs_put(bitmap->sysfs_can_clear);
1743
1744	if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1745		bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1746		md_cluster_stop(bitmap->mddev);
1747
1748	/* Shouldn't be needed - but just in case.... */
1749	wait_event(bitmap->write_wait,
1750		   atomic_read(&bitmap->pending_writes) == 0);
1751
1752	/* release the bitmap file  */
1753	md_bitmap_file_unmap(&bitmap->storage);
1754
1755	bp = bitmap->counts.bp;
1756	pages = bitmap->counts.pages;
1757
1758	/* free all allocated memory */
1759
1760	if (bp) /* deallocate the page memory */
1761		for (k = 0; k < pages; k++)
1762			if (bp[k].map && !bp[k].hijacked)
1763				kfree(bp[k].map);
1764	kfree(bp);
1765	kfree(bitmap);
1766}
1767EXPORT_SYMBOL(md_bitmap_free);
1768
1769void md_bitmap_wait_behind_writes(struct mddev *mddev)
1770{
1771	struct bitmap *bitmap = mddev->bitmap;
1772
1773	/* wait for behind writes to complete */
1774	if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1775		pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
1776			 mdname(mddev));
1777		/* need to kick something here to make sure I/O goes? */
1778		wait_event(bitmap->behind_wait,
1779			   atomic_read(&bitmap->behind_writes) == 0);
1780	}
1781}
1782
1783void md_bitmap_destroy(struct mddev *mddev)
1784{
1785	struct bitmap *bitmap = mddev->bitmap;
1786
1787	if (!bitmap) /* there was no bitmap */
1788		return;
1789
1790	md_bitmap_wait_behind_writes(mddev);
1791	if (!mddev->serialize_policy)
1792		mddev_destroy_serial_pool(mddev, NULL, true);
1793
1794	mutex_lock(&mddev->bitmap_info.mutex);
1795	spin_lock(&mddev->lock);
1796	mddev->bitmap = NULL; /* disconnect from the md device */
1797	spin_unlock(&mddev->lock);
1798	mutex_unlock(&mddev->bitmap_info.mutex);
1799	if (mddev->thread)
1800		mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1801
1802	md_bitmap_free(bitmap);
1803}
1804
1805/*
1806 * initialize the bitmap structure
1807 * if this returns an error, bitmap_destroy must be called to do clean up
1808 * once mddev->bitmap is set
1809 */
1810struct bitmap *md_bitmap_create(struct mddev *mddev, int slot)
1811{
1812	struct bitmap *bitmap;
1813	sector_t blocks = mddev->resync_max_sectors;
1814	struct file *file = mddev->bitmap_info.file;
1815	int err;
1816	struct kernfs_node *bm = NULL;
1817
1818	BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1819
1820	BUG_ON(file && mddev->bitmap_info.offset);
1821
1822	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1823		pr_notice("md/raid:%s: array with journal cannot have bitmap\n",
1824			  mdname(mddev));
1825		return ERR_PTR(-EBUSY);
1826	}
1827
1828	bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1829	if (!bitmap)
1830		return ERR_PTR(-ENOMEM);
1831
1832	spin_lock_init(&bitmap->counts.lock);
1833	atomic_set(&bitmap->pending_writes, 0);
1834	init_waitqueue_head(&bitmap->write_wait);
1835	init_waitqueue_head(&bitmap->overflow_wait);
1836	init_waitqueue_head(&bitmap->behind_wait);
1837
1838	bitmap->mddev = mddev;
1839	bitmap->cluster_slot = slot;
1840
1841	if (mddev->kobj.sd)
1842		bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1843	if (bm) {
1844		bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1845		sysfs_put(bm);
1846	} else
1847		bitmap->sysfs_can_clear = NULL;
1848
1849	bitmap->storage.file = file;
1850	if (file) {
1851		get_file(file);
1852		/* As future accesses to this file will use bmap,
1853		 * and bypass the page cache, we must sync the file
1854		 * first.
1855		 */
1856		vfs_fsync(file, 1);
1857	}
1858	/* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1859	if (!mddev->bitmap_info.external) {
1860		/*
1861		 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1862		 * instructing us to create a new on-disk bitmap instance.
1863		 */
1864		if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1865			err = md_bitmap_new_disk_sb(bitmap);
1866		else
1867			err = md_bitmap_read_sb(bitmap);
1868	} else {
1869		err = 0;
1870		if (mddev->bitmap_info.chunksize == 0 ||
1871		    mddev->bitmap_info.daemon_sleep == 0)
1872			/* chunksize and time_base need to be
1873			 * set first. */
1874			err = -EINVAL;
1875	}
1876	if (err)
1877		goto error;
1878
1879	bitmap->daemon_lastrun = jiffies;
1880	err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1881	if (err)
1882		goto error;
1883
1884	pr_debug("created bitmap (%lu pages) for device %s\n",
1885		 bitmap->counts.pages, bmname(bitmap));
1886
1887	err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1888	if (err)
1889		goto error;
1890
1891	return bitmap;
1892 error:
1893	md_bitmap_free(bitmap);
1894	return ERR_PTR(err);
1895}
1896
1897int md_bitmap_load(struct mddev *mddev)
1898{
1899	int err = 0;
1900	sector_t start = 0;
1901	sector_t sector = 0;
1902	struct bitmap *bitmap = mddev->bitmap;
1903	struct md_rdev *rdev;
1904
1905	if (!bitmap)
1906		goto out;
1907
1908	rdev_for_each(rdev, mddev)
1909		mddev_create_serial_pool(mddev, rdev, true);
1910
1911	if (mddev_is_clustered(mddev))
1912		md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes);
1913
1914	/* Clear out old bitmap info first:  Either there is none, or we
1915	 * are resuming after someone else has possibly changed things,
1916	 * so we should forget old cached info.
1917	 * All chunks should be clean, but some might need_sync.
1918	 */
1919	while (sector < mddev->resync_max_sectors) {
1920		sector_t blocks;
1921		md_bitmap_start_sync(bitmap, sector, &blocks, 0);
1922		sector += blocks;
1923	}
1924	md_bitmap_close_sync(bitmap);
1925
1926	if (mddev->degraded == 0
1927	    || bitmap->events_cleared == mddev->events)
1928		/* no need to keep dirty bits to optimise a
1929		 * re-add of a missing device */
1930		start = mddev->recovery_cp;
1931
1932	mutex_lock(&mddev->bitmap_info.mutex);
1933	err = md_bitmap_init_from_disk(bitmap, start);
1934	mutex_unlock(&mddev->bitmap_info.mutex);
1935
1936	if (err)
1937		goto out;
1938	clear_bit(BITMAP_STALE, &bitmap->flags);
1939
1940	/* Kick recovery in case any bits were set */
1941	set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1942
1943	mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1944	md_wakeup_thread(mddev->thread);
1945
1946	md_bitmap_update_sb(bitmap);
1947
1948	if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1949		err = -EIO;
1950out:
1951	return err;
1952}
1953EXPORT_SYMBOL_GPL(md_bitmap_load);
1954
1955/* caller need to free returned bitmap with md_bitmap_free() */
1956struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot)
1957{
1958	int rv = 0;
1959	struct bitmap *bitmap;
1960
1961	bitmap = md_bitmap_create(mddev, slot);
1962	if (IS_ERR(bitmap)) {
1963		rv = PTR_ERR(bitmap);
1964		return ERR_PTR(rv);
1965	}
1966
1967	rv = md_bitmap_init_from_disk(bitmap, 0);
1968	if (rv) {
1969		md_bitmap_free(bitmap);
1970		return ERR_PTR(rv);
1971	}
1972
1973	return bitmap;
1974}
1975EXPORT_SYMBOL(get_bitmap_from_slot);
1976
1977/* Loads the bitmap associated with slot and copies the resync information
1978 * to our bitmap
1979 */
1980int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
1981		sector_t *low, sector_t *high, bool clear_bits)
1982{
1983	int rv = 0, i, j;
1984	sector_t block, lo = 0, hi = 0;
1985	struct bitmap_counts *counts;
1986	struct bitmap *bitmap;
1987
1988	bitmap = get_bitmap_from_slot(mddev, slot);
1989	if (IS_ERR(bitmap)) {
1990		pr_err("%s can't get bitmap from slot %d\n", __func__, slot);
1991		return -1;
1992	}
1993
1994	counts = &bitmap->counts;
1995	for (j = 0; j < counts->chunks; j++) {
1996		block = (sector_t)j << counts->chunkshift;
1997		if (md_bitmap_file_test_bit(bitmap, block)) {
1998			if (!lo)
1999				lo = block;
2000			hi = block;
2001			md_bitmap_file_clear_bit(bitmap, block);
2002			md_bitmap_set_memory_bits(mddev->bitmap, block, 1);
2003			md_bitmap_file_set_bit(mddev->bitmap, block);
2004		}
2005	}
2006
2007	if (clear_bits) {
2008		md_bitmap_update_sb(bitmap);
2009		/* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs
2010		 * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */
2011		for (i = 0; i < bitmap->storage.file_pages; i++)
2012			if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING))
2013				set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE);
2014		md_bitmap_unplug(bitmap);
2015	}
2016	md_bitmap_unplug(mddev->bitmap);
2017	*low = lo;
2018	*high = hi;
2019	md_bitmap_free(bitmap);
2020
2021	return rv;
2022}
2023EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot);
2024
2025
2026void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
2027{
2028	unsigned long chunk_kb;
2029	struct bitmap_counts *counts;
2030
2031	if (!bitmap)
2032		return;
2033
2034	counts = &bitmap->counts;
2035
2036	chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
2037	seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
2038		   "%lu%s chunk",
2039		   counts->pages - counts->missing_pages,
2040		   counts->pages,
2041		   (counts->pages - counts->missing_pages)
2042		   << (PAGE_SHIFT - 10),
2043		   chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
2044		   chunk_kb ? "KB" : "B");
2045	if (bitmap->storage.file) {
2046		seq_printf(seq, ", file: ");
2047		seq_file_path(seq, bitmap->storage.file, " \t\n");
2048	}
2049
2050	seq_printf(seq, "\n");
2051}
2052
2053int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
2054		  int chunksize, int init)
2055{
2056	/* If chunk_size is 0, choose an appropriate chunk size.
2057	 * Then possibly allocate new storage space.
2058	 * Then quiesce, copy bits, replace bitmap, and re-start
2059	 *
2060	 * This function is called both to set up the initial bitmap
2061	 * and to resize the bitmap while the array is active.
2062	 * If this happens as a result of the array being resized,
2063	 * chunksize will be zero, and we need to choose a suitable
2064	 * chunksize, otherwise we use what we are given.
2065	 */
2066	struct bitmap_storage store;
2067	struct bitmap_counts old_counts;
2068	unsigned long chunks;
2069	sector_t block;
2070	sector_t old_blocks, new_blocks;
2071	int chunkshift;
2072	int ret = 0;
2073	long pages;
2074	struct bitmap_page *new_bp;
2075
2076	if (bitmap->storage.file && !init) {
2077		pr_info("md: cannot resize file-based bitmap\n");
2078		return -EINVAL;
2079	}
2080
2081	if (chunksize == 0) {
2082		/* If there is enough space, leave the chunk size unchanged,
2083		 * else increase by factor of two until there is enough space.
2084		 */
2085		long bytes;
2086		long space = bitmap->mddev->bitmap_info.space;
2087
2088		if (space == 0) {
2089			/* We don't know how much space there is, so limit
2090			 * to current size - in sectors.
2091			 */
2092			bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
2093			if (!bitmap->mddev->bitmap_info.external)
2094				bytes += sizeof(bitmap_super_t);
2095			space = DIV_ROUND_UP(bytes, 512);
2096			bitmap->mddev->bitmap_info.space = space;
2097		}
2098		chunkshift = bitmap->counts.chunkshift;
2099		chunkshift--;
2100		do {
2101			/* 'chunkshift' is shift from block size to chunk size */
2102			chunkshift++;
2103			chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2104			bytes = DIV_ROUND_UP(chunks, 8);
2105			if (!bitmap->mddev->bitmap_info.external)
2106				bytes += sizeof(bitmap_super_t);
2107		} while (bytes > (space << 9));
 
2108	} else
2109		chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
2110
2111	chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
2112	memset(&store, 0, sizeof(store));
2113	if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
2114		ret = md_bitmap_storage_alloc(&store, chunks,
2115					      !bitmap->mddev->bitmap_info.external,
2116					      mddev_is_clustered(bitmap->mddev)
2117					      ? bitmap->cluster_slot : 0);
2118	if (ret) {
2119		md_bitmap_file_unmap(&store);
2120		goto err;
2121	}
2122
2123	pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2124
2125	new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL);
2126	ret = -ENOMEM;
2127	if (!new_bp) {
2128		md_bitmap_file_unmap(&store);
2129		goto err;
2130	}
2131
2132	if (!init)
2133		bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2134
2135	store.file = bitmap->storage.file;
2136	bitmap->storage.file = NULL;
2137
2138	if (store.sb_page && bitmap->storage.sb_page)
2139		memcpy(page_address(store.sb_page),
2140		       page_address(bitmap->storage.sb_page),
2141		       sizeof(bitmap_super_t));
2142	spin_lock_irq(&bitmap->counts.lock);
2143	md_bitmap_file_unmap(&bitmap->storage);
2144	bitmap->storage = store;
2145
2146	old_counts = bitmap->counts;
2147	bitmap->counts.bp = new_bp;
2148	bitmap->counts.pages = pages;
2149	bitmap->counts.missing_pages = pages;
2150	bitmap->counts.chunkshift = chunkshift;
2151	bitmap->counts.chunks = chunks;
2152	bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2153						     BITMAP_BLOCK_SHIFT);
2154
2155	blocks = min(old_counts.chunks << old_counts.chunkshift,
2156		     chunks << chunkshift);
2157
2158	/* For cluster raid, need to pre-allocate bitmap */
2159	if (mddev_is_clustered(bitmap->mddev)) {
2160		unsigned long page;
2161		for (page = 0; page < pages; page++) {
2162			ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1);
2163			if (ret) {
2164				unsigned long k;
2165
2166				/* deallocate the page memory */
2167				for (k = 0; k < page; k++) {
2168					kfree(new_bp[k].map);
2169				}
2170				kfree(new_bp);
2171
2172				/* restore some fields from old_counts */
2173				bitmap->counts.bp = old_counts.bp;
2174				bitmap->counts.pages = old_counts.pages;
2175				bitmap->counts.missing_pages = old_counts.pages;
2176				bitmap->counts.chunkshift = old_counts.chunkshift;
2177				bitmap->counts.chunks = old_counts.chunks;
2178				bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2179									     BITMAP_BLOCK_SHIFT);
2180				blocks = old_counts.chunks << old_counts.chunkshift;
2181				pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n");
2182				break;
2183			} else
2184				bitmap->counts.bp[page].count += 1;
2185		}
2186	}
2187
2188	for (block = 0; block < blocks; ) {
2189		bitmap_counter_t *bmc_old, *bmc_new;
2190		int set;
2191
2192		bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0);
2193		set = bmc_old && NEEDED(*bmc_old);
2194
2195		if (set) {
2196			bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2197			if (*bmc_new == 0) {
2198				/* need to set on-disk bits too. */
2199				sector_t end = block + new_blocks;
2200				sector_t start = block >> chunkshift;
2201				start <<= chunkshift;
2202				while (start < end) {
2203					md_bitmap_file_set_bit(bitmap, block);
2204					start += 1 << chunkshift;
 
 
 
 
 
 
2205				}
2206				*bmc_new = 2;
2207				md_bitmap_count_page(&bitmap->counts, block, 1);
2208				md_bitmap_set_pending(&bitmap->counts, block);
2209			}
2210			*bmc_new |= NEEDED_MASK;
2211			if (new_blocks < old_blocks)
2212				old_blocks = new_blocks;
2213		}
2214		block += old_blocks;
2215	}
2216
2217	if (bitmap->counts.bp != old_counts.bp) {
2218		unsigned long k;
2219		for (k = 0; k < old_counts.pages; k++)
2220			if (!old_counts.bp[k].hijacked)
2221				kfree(old_counts.bp[k].map);
2222		kfree(old_counts.bp);
2223	}
2224
2225	if (!init) {
2226		int i;
2227		while (block < (chunks << chunkshift)) {
2228			bitmap_counter_t *bmc;
2229			bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1);
2230			if (bmc) {
2231				/* new space.  It needs to be resynced, so
2232				 * we set NEEDED_MASK.
2233				 */
2234				if (*bmc == 0) {
2235					*bmc = NEEDED_MASK | 2;
2236					md_bitmap_count_page(&bitmap->counts, block, 1);
2237					md_bitmap_set_pending(&bitmap->counts, block);
2238				}
2239			}
2240			block += new_blocks;
2241		}
2242		for (i = 0; i < bitmap->storage.file_pages; i++)
2243			set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2244	}
2245	spin_unlock_irq(&bitmap->counts.lock);
2246
2247	if (!init) {
2248		md_bitmap_unplug(bitmap);
2249		bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2250	}
2251	ret = 0;
2252err:
2253	return ret;
2254}
2255EXPORT_SYMBOL_GPL(md_bitmap_resize);
2256
2257static ssize_t
2258location_show(struct mddev *mddev, char *page)
2259{
2260	ssize_t len;
2261	if (mddev->bitmap_info.file)
2262		len = sprintf(page, "file");
2263	else if (mddev->bitmap_info.offset)
2264		len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2265	else
2266		len = sprintf(page, "none");
2267	len += sprintf(page+len, "\n");
2268	return len;
2269}
2270
2271static ssize_t
2272location_store(struct mddev *mddev, const char *buf, size_t len)
2273{
2274	int rv;
2275
2276	rv = mddev_lock(mddev);
2277	if (rv)
2278		return rv;
 
2279	if (mddev->pers) {
2280		if (!mddev->pers->quiesce) {
2281			rv = -EBUSY;
2282			goto out;
2283		}
2284		if (mddev->recovery || mddev->sync_thread) {
2285			rv = -EBUSY;
2286			goto out;
2287		}
2288	}
2289
2290	if (mddev->bitmap || mddev->bitmap_info.file ||
2291	    mddev->bitmap_info.offset) {
2292		/* bitmap already configured.  Only option is to clear it */
2293		if (strncmp(buf, "none", 4) != 0) {
2294			rv = -EBUSY;
2295			goto out;
2296		}
2297		if (mddev->pers) {
2298			mddev_suspend(mddev);
2299			md_bitmap_destroy(mddev);
2300			mddev_resume(mddev);
2301		}
2302		mddev->bitmap_info.offset = 0;
2303		if (mddev->bitmap_info.file) {
2304			struct file *f = mddev->bitmap_info.file;
2305			mddev->bitmap_info.file = NULL;
2306			fput(f);
2307		}
2308	} else {
2309		/* No bitmap, OK to set a location */
2310		long long offset;
 
 
2311		if (strncmp(buf, "none", 4) == 0)
2312			/* nothing to be done */;
2313		else if (strncmp(buf, "file:", 5) == 0) {
2314			/* Not supported yet */
2315			rv = -EINVAL;
2316			goto out;
2317		} else {
2318			if (buf[0] == '+')
2319				rv = kstrtoll(buf+1, 10, &offset);
2320			else
2321				rv = kstrtoll(buf, 10, &offset);
2322			if (rv)
2323				goto out;
2324			if (offset == 0) {
2325				rv = -EINVAL;
2326				goto out;
2327			}
2328			if (mddev->bitmap_info.external == 0 &&
2329			    mddev->major_version == 0 &&
2330			    offset != mddev->bitmap_info.default_offset) {
2331				rv = -EINVAL;
2332				goto out;
2333			}
 
2334			mddev->bitmap_info.offset = offset;
2335			if (mddev->pers) {
2336				struct bitmap *bitmap;
2337				bitmap = md_bitmap_create(mddev, -1);
2338				mddev_suspend(mddev);
2339				if (IS_ERR(bitmap))
2340					rv = PTR_ERR(bitmap);
2341				else {
2342					mddev->bitmap = bitmap;
2343					rv = md_bitmap_load(mddev);
2344					if (rv)
2345						mddev->bitmap_info.offset = 0;
2346				}
2347				if (rv) {
2348					md_bitmap_destroy(mddev);
2349					mddev_resume(mddev);
2350					goto out;
2351				}
2352				mddev_resume(mddev);
2353			}
2354		}
2355	}
2356	if (!mddev->external) {
2357		/* Ensure new bitmap info is stored in
2358		 * metadata promptly.
2359		 */
2360		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2361		md_wakeup_thread(mddev->thread);
2362	}
2363	rv = 0;
2364out:
2365	mddev_unlock(mddev);
2366	if (rv)
2367		return rv;
2368	return len;
2369}
2370
2371static struct md_sysfs_entry bitmap_location =
2372__ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2373
2374/* 'bitmap/space' is the space available at 'location' for the
2375 * bitmap.  This allows the kernel to know when it is safe to
2376 * resize the bitmap to match a resized array.
2377 */
2378static ssize_t
2379space_show(struct mddev *mddev, char *page)
2380{
2381	return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2382}
2383
2384static ssize_t
2385space_store(struct mddev *mddev, const char *buf, size_t len)
2386{
2387	unsigned long sectors;
2388	int rv;
2389
2390	rv = kstrtoul(buf, 10, &sectors);
2391	if (rv)
2392		return rv;
2393
2394	if (sectors == 0)
2395		return -EINVAL;
2396
2397	if (mddev->bitmap &&
2398	    sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2399		return -EFBIG; /* Bitmap is too big for this small space */
2400
2401	/* could make sure it isn't too big, but that isn't really
2402	 * needed - user-space should be careful.
2403	 */
2404	mddev->bitmap_info.space = sectors;
2405	return len;
2406}
2407
2408static struct md_sysfs_entry bitmap_space =
2409__ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2410
2411static ssize_t
2412timeout_show(struct mddev *mddev, char *page)
2413{
2414	ssize_t len;
2415	unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2416	unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2417
2418	len = sprintf(page, "%lu", secs);
2419	if (jifs)
2420		len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2421	len += sprintf(page+len, "\n");
2422	return len;
2423}
2424
2425static ssize_t
2426timeout_store(struct mddev *mddev, const char *buf, size_t len)
2427{
2428	/* timeout can be set at any time */
2429	unsigned long timeout;
2430	int rv = strict_strtoul_scaled(buf, &timeout, 4);
2431	if (rv)
2432		return rv;
2433
2434	/* just to make sure we don't overflow... */
2435	if (timeout >= LONG_MAX / HZ)
2436		return -EINVAL;
2437
2438	timeout = timeout * HZ / 10000;
2439
2440	if (timeout >= MAX_SCHEDULE_TIMEOUT)
2441		timeout = MAX_SCHEDULE_TIMEOUT-1;
2442	if (timeout < 1)
2443		timeout = 1;
 
2444	mddev->bitmap_info.daemon_sleep = timeout;
2445	if (mddev->thread) {
2446		/* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2447		 * the bitmap is all clean and we don't need to
2448		 * adjust the timeout right now
2449		 */
2450		if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2451			mddev->thread->timeout = timeout;
2452			md_wakeup_thread(mddev->thread);
2453		}
2454	}
2455	return len;
2456}
2457
2458static struct md_sysfs_entry bitmap_timeout =
2459__ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2460
2461static ssize_t
2462backlog_show(struct mddev *mddev, char *page)
2463{
2464	return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2465}
2466
2467static ssize_t
2468backlog_store(struct mddev *mddev, const char *buf, size_t len)
2469{
2470	unsigned long backlog;
2471	unsigned long old_mwb = mddev->bitmap_info.max_write_behind;
 
 
2472	int rv = kstrtoul(buf, 10, &backlog);
2473	if (rv)
2474		return rv;
2475	if (backlog > COUNTER_MAX)
2476		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2477	mddev->bitmap_info.max_write_behind = backlog;
2478	if (!backlog && mddev->serial_info_pool) {
2479		/* serial_info_pool is not needed if backlog is zero */
2480		if (!mddev->serialize_policy)
2481			mddev_destroy_serial_pool(mddev, NULL, false);
2482	} else if (backlog && !mddev->serial_info_pool) {
2483		/* serial_info_pool is needed since backlog is not zero */
2484		struct md_rdev *rdev;
2485
2486		rdev_for_each(rdev, mddev)
2487			mddev_create_serial_pool(mddev, rdev, false);
2488	}
2489	if (old_mwb != backlog)
2490		md_bitmap_update_sb(mddev->bitmap);
 
 
2491	return len;
2492}
2493
2494static struct md_sysfs_entry bitmap_backlog =
2495__ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2496
2497static ssize_t
2498chunksize_show(struct mddev *mddev, char *page)
2499{
2500	return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2501}
2502
2503static ssize_t
2504chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2505{
2506	/* Can only be changed when no bitmap is active */
2507	int rv;
2508	unsigned long csize;
2509	if (mddev->bitmap)
2510		return -EBUSY;
2511	rv = kstrtoul(buf, 10, &csize);
2512	if (rv)
2513		return rv;
2514	if (csize < 512 ||
2515	    !is_power_of_2(csize))
2516		return -EINVAL;
 
 
 
2517	mddev->bitmap_info.chunksize = csize;
2518	return len;
2519}
2520
2521static struct md_sysfs_entry bitmap_chunksize =
2522__ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2523
2524static ssize_t metadata_show(struct mddev *mddev, char *page)
2525{
2526	if (mddev_is_clustered(mddev))
2527		return sprintf(page, "clustered\n");
2528	return sprintf(page, "%s\n", (mddev->bitmap_info.external
2529				      ? "external" : "internal"));
2530}
2531
2532static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2533{
2534	if (mddev->bitmap ||
2535	    mddev->bitmap_info.file ||
2536	    mddev->bitmap_info.offset)
2537		return -EBUSY;
2538	if (strncmp(buf, "external", 8) == 0)
2539		mddev->bitmap_info.external = 1;
2540	else if ((strncmp(buf, "internal", 8) == 0) ||
2541			(strncmp(buf, "clustered", 9) == 0))
2542		mddev->bitmap_info.external = 0;
2543	else
2544		return -EINVAL;
2545	return len;
2546}
2547
2548static struct md_sysfs_entry bitmap_metadata =
2549__ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2550
2551static ssize_t can_clear_show(struct mddev *mddev, char *page)
2552{
2553	int len;
2554	spin_lock(&mddev->lock);
2555	if (mddev->bitmap)
2556		len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2557					     "false" : "true"));
2558	else
2559		len = sprintf(page, "\n");
2560	spin_unlock(&mddev->lock);
2561	return len;
2562}
2563
2564static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2565{
2566	if (mddev->bitmap == NULL)
2567		return -ENOENT;
2568	if (strncmp(buf, "false", 5) == 0)
2569		mddev->bitmap->need_sync = 1;
2570	else if (strncmp(buf, "true", 4) == 0) {
2571		if (mddev->degraded)
2572			return -EBUSY;
2573		mddev->bitmap->need_sync = 0;
2574	} else
2575		return -EINVAL;
2576	return len;
2577}
2578
2579static struct md_sysfs_entry bitmap_can_clear =
2580__ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2581
2582static ssize_t
2583behind_writes_used_show(struct mddev *mddev, char *page)
2584{
2585	ssize_t ret;
2586	spin_lock(&mddev->lock);
2587	if (mddev->bitmap == NULL)
2588		ret = sprintf(page, "0\n");
2589	else
2590		ret = sprintf(page, "%lu\n",
2591			      mddev->bitmap->behind_writes_used);
2592	spin_unlock(&mddev->lock);
2593	return ret;
2594}
2595
2596static ssize_t
2597behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2598{
2599	if (mddev->bitmap)
2600		mddev->bitmap->behind_writes_used = 0;
2601	return len;
2602}
2603
2604static struct md_sysfs_entry max_backlog_used =
2605__ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2606       behind_writes_used_show, behind_writes_used_reset);
2607
2608static struct attribute *md_bitmap_attrs[] = {
2609	&bitmap_location.attr,
2610	&bitmap_space.attr,
2611	&bitmap_timeout.attr,
2612	&bitmap_backlog.attr,
2613	&bitmap_chunksize.attr,
2614	&bitmap_metadata.attr,
2615	&bitmap_can_clear.attr,
2616	&max_backlog_used.attr,
2617	NULL
2618};
2619const struct attribute_group md_bitmap_group = {
2620	.name = "bitmap",
2621	.attrs = md_bitmap_attrs,
2622};