/fs/ext4/mballoc.c

/*
 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
 * Written by Alex Tomas <alex@clusterfs.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public Licens
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
 */


/*
 * mballoc.c contains the multiblocks allocation routines
 */

#include "mballoc.h"
#include <linux/debugfs.h>
#include <linux/slab.h>
#include <trace/events/ext4.h>

/*
 * MUSTDO:
 *   - test ext4_ext_search_left() and ext4_ext_search_right()
 *   - search for metadata in few groups
 *
 * TODO v4:
 *   - normalization should take into account whether file is still open
 *   - discard preallocations if no free space left (policy?)
 *   - don't normalize tails
 *   - quota
 *   - reservation for superuser
 *
 * TODO v3:
 *   - bitmap read-ahead (proposed by Oleg Drokin aka green)
 *   - track min/max extents in each group for better group selection
 *   - mb_mark_used() may allocate chunk right after splitting buddy
 *   - tree of groups sorted by number of free blocks
 *   - error handling
 */

/*
 * The allocation request involve request for multiple number of blocks
 * near to the goal(block) value specified.
 *
 * During initialization phase of the allocator we decide to use the
 * group preallocation or inode preallocation depending on the size of
 * the file. The size of the file could be the resulting file size we
 * would have after allocation, or the current file size, which ever
 * is larger. If the size is less than sbi->s_mb_stream_request we
 * select to use the group preallocation. The default value of
 * s_mb_stream_request is 16 blocks. This can also be tuned via
 * /sys/fs/ext4/<partition>/mb_stream_req. The value is represented in
 * terms of number of blocks.
 *
 * The main motivation for having small file use group preallocation is to
 * ensure that we have small files closer together on the disk.
 *
 * First stage the allocator looks at the inode prealloc list,
 * ext4_inode_info->i_prealloc_list, which contains list of prealloc
 * spaces for this particular inode. The inode prealloc space is
 * represented as:
 *
 * pa_lstart -> the logical start block for this prealloc space
 * pa_pstart -> the physical start block for this prealloc space
 * pa_len    -> length for this prealloc space
 * pa_free   ->  free space available in this prealloc space
 *
 * The inode preallocation space is used looking at the _logical_ start
 * block. If only the logical file block falls within the range of prealloc
 * space we will consume the particular prealloc space. This makes sure that
 * we have contiguous physical blocks representing the file blocks
 *
 * The important thing to be noted in case of inode prealloc space is that
 * we don't modify the values associated to inode prealloc space except
 * pa_free.
 *
 * If we are not able to find blocks in the inode prealloc space and if we
 * have the group allocation flag set then we look at the locality group
 * prealloc space. These are per CPU prealloc list represented as
 *
 * ext4_sb_info.s_locality_groups[smp_processor_id()]
 *
 * The reason for having a per cpu locality group is to reduce the contention
 * between CPUs. It is possible to get scheduled at this point.
 *
 * The locality group prealloc space is used looking at whether we have
 * enough free space (pa_free) within the prealloc space.
 *
 * If we can't allocate blocks via inode prealloc or/and locality group
 * prealloc then we look at the buddy cache. The buddy cache is represented
 * by ext4_sb_info.s_buddy_cache (struct inode) whose file offset gets
 * mapped to the buddy and bitmap information regarding different
 * groups. The buddy information is attached to buddy cache inode so that
 * we can access them through the page cache. The information regarding
 * each group is loaded via ext4_mb_load_buddy.  The information involve
 * block bitmap and buddy information. The information are stored in the
 * inode as:
 *
 *  {                        page                        }
 *  [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
 *
 *
 * one block each for bitmap and buddy information.  So for each group we
 * take up 2 blocks. A page can contain blocks_per_page (PAGE_CACHE_SIZE /
 * blocksize) blocks.  So it can have information regarding groups_per_page
 * which is blocks_per_page/2
 *
 * The buddy cache inode is not stored on disk. The inode is thrown
 * away when the filesystem is unmounted.
 *
 * We look for count number of blocks in the buddy cache. If we were able
 * to locate that many free blocks we return with additional information
 * regarding rest of the contiguous physical block available
 *
 * Before allocating blocks via buddy cache we normalize the request
 * blocks. This ensure we ask for more blocks that we needed. The extra
 * blocks that we get after allocation is added to the respective prealloc
 * list. In case of inode preallocation we follow a list of heuristics
 * based on file size. This can be found in ext4_mb_normalize_request. If
 * we are doing a group prealloc we try to normalize the request to
 * sbi->s_mb_group_prealloc. Default value of s_mb_group_prealloc is
 * 512 blocks. This can be tuned via
 * /sys/fs/ext4/<partition>/mb_group_prealloc. The value is represented in
 * terms of number of blocks. If we have mounted the file system with -O
 * stripe=<value> option the group prealloc request is normalized to the
 * the smallest multiple of the stripe value (sbi->s_stripe) which is
 * greater than the default mb_group_prealloc.
 *
 * The regular allocator (using the buddy cache) supports a few tunables.
 *
 * /sys/fs/ext4/<partition>/mb_min_to_scan
 * /sys/fs/ext4/<partition>/mb_max_to_scan
 * /sys/fs/ext4/<partition>/mb_order2_req
 *
 * The regular allocator uses buddy scan only if the request len is power of
 * 2 blocks and the order of allocation is >= sbi->s_mb_order2_reqs. The
 * value of s_mb_order2_reqs can be tuned via
 * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
 * stripe size (sbi->s_stripe), we try to search for contiguous block in
 * stripe size. This should result in better allocation on RAID setups. If
 * not, we search in the specific group using bitmap for best extents. The
 * tunable min_to_scan and max_to_scan control the behaviour here.
 * min_to_scan indicate how long the mballoc __must__ look for a best
 * extent and max_to_scan indicates how long the mballoc __can__ look for a
 * best extent in the found extents. Searching for the blocks starts with
 * the group specified as the goal value in allocation context via
 * ac_g_ex. Each group is first checked based on the criteria whether it
 * can be used for allocation. ext4_mb_good_group explains how the groups are
 * checked.
 *
 * Both the prealloc space are getting populated as above. So for the first
 * request we will hit the buddy cache which will result in this prealloc
 * space getting filled. The prealloc space is then later used for the
 * subsequent request.
 */

/*
 * mballoc operates on the following data:
 *  - on-disk bitmap
 *  - in-core buddy (actually includes buddy and bitmap)
 *  - preallocation descriptors (PAs)
 *
 * there are two types of preallocations:
 *  - inode
 *    assiged to specific inode and can be used for this inode only.
 *    it describes part of inode's space preallocated to specific
 *    physical blocks. any block from that preallocated can be used
 *    independent. the descriptor just tracks number of blocks left
 *    unused. so, before taking some block from descriptor, one must
 *    make sure corresponded logical block isn't allocated yet. this
 *    also means that freeing any block within descriptor's range
 *    must discard all preallocated blocks.
 *  - locality group
 *    assigned to specific locality group which does not translate to
 *    permanent set of inodes: inode can join and leave group. space
 *    from this type of preallocation can be used for any inode. thus
 *    it's consumed from the beginning to the end.
 *
 * relation between them can be expressed as:
 *    in-core buddy = on-disk bitmap + preallocation descriptors
 *
 * this mean blocks mballoc considers used are:
 *  - allocated blocks (persistent)
 *  - preallocated blocks (non-persistent)
 *
 * consistency in mballoc world means that at any time a block is either
 * free or used in ALL structures. notice: "any time" should not be read
 * literally -- time is discrete and delimited by locks.
 *
 *  to keep it simple, we don't use block numbers, instead we count number of
 *  blocks: how many blocks marked used/free in on-disk bitmap, buddy and PA.
 *
 * all operations can be expressed as:
 *  - init buddy:			buddy = on-disk + PAs
 *  - new PA:				buddy += N; PA = N
 *  - use inode PA:			on-disk += N; PA -= N
 *  - discard inode PA			buddy -= on-disk - PA; PA = 0
 *  - use locality group PA		on-disk += N; PA -= N
 *  - discard locality group PA		buddy -= PA; PA = 0
 *  note: 'buddy -= on-disk - PA' is used to show that on-disk bitmap
 *        is used in real operation because we can't know actual used
 *        bits from PA, only from on-disk bitmap
 *
 * if we follow this strict logic, then all operations above should be atomic.
 * given some of them can block, we'd have to use something like semaphores
 * killing performance on high-end SMP hardware. let's try to relax it using
 * the following knowledge:
 *  1) if buddy is referenced, it's already initialized
 *  2) while block is used in buddy and the buddy is referenced,
 *     nobody can re-allocate that block
 *  3) we work on bitmaps and '+' actually means 'set bits'. if on-disk has
 *     bit set and PA claims same block, it's OK. IOW, one can set bit in
 *     on-disk bitmap if buddy has same bit set or/and PA covers corresponded
 *     block
 *
 * so, now we're building a concurrency table:
 *  - init buddy vs.
 *    - new PA
 *      blocks for PA are allocated in the buddy, buddy must be referenced
 *      until PA is linked to allocation group to avoid concurrent buddy init
 *    - use inode PA
 *      we need to make sure that either on-disk bitmap or PA has uptodate data
 *      given (3) we care that PA-=N operation doesn't interfere with init
 *    - discard inode PA
 *      the simplest way would be to have buddy initialized by the discard
 *    - use locality group PA
 *      again PA-=N must be serialized with init
 *    - discard locality group PA
 *      the simplest way would be to have buddy initialized by the discard
 *  - new PA vs.
 *    - use inode PA
 *      i_data_sem serializes them
 *    - discard inode PA
 *      discard process must wait until PA isn't used by another process
 *    - use locality group PA
 *      some mutex should serialize them
 *    - discard locality group PA
 *      discard process must wait until PA isn't used by another process
 *  - use inode PA
 *    - use inode PA
 *      i_data_sem or another mutex should serializes them
 *    - discard inode PA
 *      discard process must wait until PA isn't used by another process
 *    - use locality group PA
 *      nothing wrong here -- they're different PAs covering different blocks
 *    - discard locality group PA
 *      discard process must wait until PA isn't used by another process
 *
 * now we're ready to make few consequences:
 *  - PA is referenced and while it is no discard is possible
 *  - PA is referenced until block isn't marked in on-disk bitmap
 *  - PA changes only after on-disk bitmap
 *  - discard must not compete with init. either init is done before
 *    any discard or they're serialized somehow
 *  - buddy init as sum of on-disk bitmap and PAs is done atomically
 *
 * a special case when we've used PA to emptiness. no need to modify buddy
 * in this case, but we should care about concurrent init
 *
 */

 /*
 * Logic in few words:
 *
 *  - allocation:
 *    load group
 *    find blocks
 *    mark bits in on-disk bitmap
 *    release group
 *
 *  - use preallocation:
 *    find proper PA (per-inode or group)
 *    load group
 *    mark bits in on-disk bitmap
 *    release group
 *    release PA
 *
 *  - free:
 *    load group
 *    mark bits in on-disk bitmap
 *    release group
 *
 *  - discard preallocations in group:
 *    mark PAs deleted
 *    move them onto local list
 *    load on-disk bitmap
 *    load group
 *    remove PA from object (inode or locality group)
 *    mark free blocks in-core
 *
 *  - discard inode's preallocations:
 */

/*
 * Locking rules
 *
 * Locks:
 *  - bitlock on a group	(group)
 *  - object (inode/locality)	(object)
 *  - per-pa lock		(pa)
 *
 * Paths:
 *  - new pa
 *    object
 *    group
 *
 *  - find and use pa:
 *    pa
 *
 *  - release consumed pa:
 *    pa
 *    group
 *    object
 *
 *  - generate in-core bitmap:
 *    group
 *        pa
 *
 *  - discard all for given object (inode, locality group):
 *    object
 *        pa
 *    group
 *
 *  - discard all for given group:
 *    group
 *        pa
 *    group
 *        object
 *
 */
static struct kmem_cache *ext4_pspace_cachep;
static struct kmem_cache *ext4_ac_cachep;
static struct kmem_cache *ext4_free_ext_cachep;

/* We create slab caches for groupinfo data structures based on the
 * superblock block size.  There will be one per mounted filesystem for
 * each unique s_blocksize_bits */
#define NR_GRPINFO_CACHES 8
static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];

static const char *ext4_groupinfo_slab_names[NR_GRPINFO_CACHES] = {
	"ext4_groupinfo_1k", "ext4_groupinfo_2k", "ext4_groupinfo_4k",
	"ext4_groupinfo_8k", "ext4_groupinfo_16k", "ext4_groupinfo_32k",
	"ext4_groupinfo_64k", "ext4_groupinfo_128k"
};

static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
					ext4_group_t group);
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group);
static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);

static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
{
#if BITS_PER_LONG == 64
	*bit += ((unsigned long) addr & 7UL) << 3;
	addr = (void *) ((unsigned long) addr & ~7UL);
#elif BITS_PER_LONG == 32
	*bit += ((unsigned long) addr & 3UL) << 3;
	addr = (void *) ((unsigned long) addr & ~3UL);
#else
#error "how many bits you are?!"
#endif
	return addr;
}

static inline int mb_test_bit(int bit, void *addr)
{
	/*
	 * ext4_test_bit on architecture like powerpc
	 * needs unsigned long aligned address
	 */
	addr = mb_correct_addr_and_bit(&bit, addr);
	return ext4_test_bit(bit, addr);
}

static inline void mb_set_bit(int bit, void *addr)
{
	addr = mb_correct_addr_and_bit(&bit, addr);
	ext4_set_bit(bit, addr);
}

static inline void mb_clear_bit(int bit, void *addr)
{
	addr = mb_correct_addr_and_bit(&bit, addr);
	ext4_clear_bit(bit, addr);
}

static inline int mb_find_next_zero_bit(void *addr, int max, int start)
{
	int fix = 0, ret, tmpmax;
	addr = mb_correct_addr_and_bit(&fix, addr);
	tmpmax = max + fix;
	start += fix;

	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
}

static inline int mb_find_next_bit(void *addr, int max, int start)
{
	int fix = 0, ret, tmpmax;
	addr = mb_correct_addr_and_bit(&fix, addr);
	tmpmax = max + fix;
	start += fix;

	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
}

static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
{
	char *bb;

	BUG_ON(EXT4_MB_BITMAP(e4b) == EXT4_MB_BUDDY(e4b));
	BUG_ON(max == NULL);

	if (order > e4b->bd_blkbits + 1) {
		*max = 0;
		return NULL;
	}

	/* at order 0 we see each particular block */
	if (order == 0) {
		*max = 1 << (e4b->bd_blkbits + 3);
		return EXT4_MB_BITMAP(e4b);
	}

	bb = EXT4_MB_BUDDY(e4b) + EXT4_SB(e4b->bd_sb)->s_mb_offsets[order];
	*max = EXT4_SB(e4b->bd_sb)->s_mb_maxs[order];

	return bb;
}

#ifdef DOUBLE_CHECK
static void mb_free_blocks_double(struct inode *inode, struct ext4_buddy *e4b,
			   int first, int count)
{
	int i;
	struct super_block *sb = e4b->bd_sb;

	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
		return;
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
	for (i = 0; i < count; i++) {
		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
			ext4_fsblk_t blocknr;

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
			blocknr += first + i;
			ext4_grp_locked_error(sb, e4b->bd_group,
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing block already freed "
					      "(bit %u)",
					      first + i);
		}
		mb_clear_bit(first + i, e4b->bd_info->bb_bitmap);
	}
}

static void mb_mark_used_double(struct ext4_buddy *e4b, int first, int count)
{
	int i;

	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
		return;
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
	for (i = 0; i < count; i++) {
		BUG_ON(mb_test_bit(first + i, e4b->bd_info->bb_bitmap));
		mb_set_bit(first + i, e4b->bd_info->bb_bitmap);
	}
}

static void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
{
	if (memcmp(e4b->bd_info->bb_bitmap, bitmap, e4b->bd_sb->s_blocksize)) {
		unsigned char *b1, *b2;
		int i;
		b1 = (unsigned char *) e4b->bd_info->bb_bitmap;
		b2 = (unsigned char *) bitmap;
		for (i = 0; i < e4b->bd_sb->s_blocksize; i++) {
			if (b1[i] != b2[i]) {
				ext4_msg(e4b->bd_sb, KERN_ERR,
					 "corruption in group %u "
					 "at byte %u(%u): %x in copy != %x "
					 "on disk/prealloc",
					 e4b->bd_group, i, i * 8, b1[i], b2[i]);
				BUG();
			}
		}
	}
}

#else
static inline void mb_free_blocks_double(struct inode *inode,
				struct ext4_buddy *e4b, int first, int count)
{
	return;
}
static inline void mb_mark_used_double(struct ext4_buddy *e4b,
						int first, int count)
{
	return;
}
static inline void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
{
	return;
}
#endif

#ifdef AGGRESSIVE_CHECK

#define MB_CHECK_ASSERT(assert)						\
do {									\
	if (!(assert)) {						\
		printk(KERN_EMERG					\
			"Assertion failure in %s() at %s:%d: \"%s\"\n",	\
			function, file, line, # assert);		\
		BUG();							\
	}								\
} while (0)

static int __mb_check_buddy(struct ext4_buddy *e4b, char *file,
				const char *function, int line)
{
	struct super_block *sb = e4b->bd_sb;
	int order = e4b->bd_blkbits + 1;
	int max;
	int max2;
	int i;
	int j;
	int k;
	int count;
	struct ext4_group_info *grp;
	int fragments = 0;
	int fstart;
	struct list_head *cur;
	void *buddy;
	void *buddy2;

	{
		static int mb_check_counter;
		if (mb_check_counter++ % 100 != 0)
			return 0;
	}

	while (order > 1) {
		buddy = mb_find_buddy(e4b, order, &max);
		MB_CHECK_ASSERT(buddy);
		buddy2 = mb_find_buddy(e4b, order - 1, &max2);
		MB_CHECK_ASSERT(buddy2);
		MB_CHECK_ASSERT(buddy != buddy2);
		MB_CHECK_ASSERT(max * 2 == max2);

		count = 0;
		for (i = 0; i < max; i++) {

			if (mb_test_bit(i, buddy)) {
				/* only single bit in buddy2 may be 1 */
				if (!mb_test_bit(i << 1, buddy2)) {
					MB_CHECK_ASSERT(
						mb_test_bit((i<<1)+1, buddy2));
				} else if (!mb_test_bit((i << 1) + 1, buddy2)) {
					MB_CHECK_ASSERT(
						mb_test_bit(i << 1, buddy2));
				}
				continue;
			}

			/* both bits in buddy2 must be 0 */
			MB_CHECK_ASSERT(mb_test_bit(i << 1, buddy2));
			MB_CHECK_ASSERT(mb_test_bit((i << 1) + 1, buddy2));

			for (j = 0; j < (1 << order); j++) {
				k = (i * (1 << order)) + j;
				MB_CHECK_ASSERT(
					!mb_test_bit(k, EXT4_MB_BITMAP(e4b)));
			}
			count++;
		}
		MB_CHECK_ASSERT(e4b->bd_info->bb_counters[order] == count);
		order--;
	}

	fstart = -1;
	buddy = mb_find_buddy(e4b, 0, &max);
	for (i = 0; i < max; i++) {
		if (!mb_test_bit(i, buddy)) {
			MB_CHECK_ASSERT(i >= e4b->bd_info->bb_first_free);
			if (fstart == -1) {
				fragments++;
				fstart = i;
			}
			continue;
		}
		fstart = -1;
		/* check used bits only */
		for (j = 0; j < e4b->bd_blkbits + 1; j++) {
			buddy2 = mb_find_buddy(e4b, j, &max2);
			k = i >> j;
			MB_CHECK_ASSERT(k < max2);
			MB_CHECK_ASSERT(mb_test_bit(k, buddy2));
		}
	}
	MB_CHECK_ASSERT(!EXT4_MB_GRP_NEED_INIT(e4b->bd_info));
	MB_CHECK_ASSERT(e4b->bd_info->bb_fragments == fragments);

	grp = ext4_get_group_info(sb, e4b->bd_group);
	list_for_each(cur, &grp->bb_prealloc_list) {
		ext4_group_t groupnr;
		struct ext4_prealloc_space *pa;
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
		for (i = 0; i < pa->pa_len; i++)
			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
	}
	return 0;
}
#undef MB_CHECK_ASSERT
#define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
					__FILE__, __func__, __LINE__)
#else
#define mb_check_buddy(e4b)
#endif

/*
 * Divide blocks started from @first with length @len into
 * smaller chunks with power of 2 blocks.
 * Clear the bits in bitmap which the blocks of the chunk(s) covered,
 * then increase bb_counters[] for corresponded chunk size.
 */
static void ext4_mb_mark_free_simple(struct super_block *sb,
				void *buddy, ext4_grpblk_t first, ext4_grpblk_t len,
					struct ext4_group_info *grp)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	ext4_grpblk_t min;
	ext4_grpblk_t max;
	ext4_grpblk_t chunk;
	unsigned short border;

	BUG_ON(len > EXT4_BLOCKS_PER_GROUP(sb));

	border = 2 << sb->s_blocksize_bits;

	while (len > 0) {
		/* find how many blocks can be covered since this position */
		max = ffs(first | border) - 1;

		/* find how many blocks of power 2 we need to mark */
		min = fls(len) - 1;

		if (max < min)
			min = max;
		chunk = 1 << min;

		/* mark multiblock chunks only */
		grp->bb_counters[min]++;
		if (min > 0)
			mb_clear_bit(first >> min,
				     buddy + sbi->s_mb_offsets[min]);

		len -= chunk;
		first += chunk;
	}
}

/*
 * Cache the order of the largest free extent we have available in this block
 * group.
 */
static void
mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
{
	int i;
	int bits;

	grp->bb_largest_free_order = -1; /* uninit */

	bits = sb->s_blocksize_bits + 1;
	for (i = bits; i >= 0; i--) {
		if (grp->bb_counters[i] > 0) {
			grp->bb_largest_free_order = i;
			break;
		}
	}
}

static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
				void *buddy, void *bitmap, ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
	ext4_grpblk_t max = EXT4_BLOCKS_PER_GROUP(sb);
	ext4_grpblk_t i = 0;
	ext4_grpblk_t first;
	ext4_grpblk_t len;
	unsigned free = 0;
	unsigned fragments = 0;
	unsigned long long period = get_cycles();

	/* initialize buddy from bitmap which is aggregation
	 * of on-disk bitmap and preallocations */
	i = mb_find_next_zero_bit(bitmap, max, 0);
	grp->bb_first_free = i;
	while (i < max) {
		fragments++;
		first = i;
		i = mb_find_next_bit(bitmap, max, i);
		len = i - first;
		free += len;
		if (len > 1)
			ext4_mb_mark_free_simple(sb, buddy, first, len, grp);
		else
			grp->bb_counters[0]++;
		if (i < max)
			i = mb_find_next_zero_bit(bitmap, max, i);
	}
	grp->bb_fragments = fragments;

	if (free != grp->bb_free) {
		ext4_grp_locked_error(sb, group, 0, 0,
				      "%u blocks in bitmap, %u in gd",
				      free, grp->bb_free);
		/*
		 * If we intent to continue, we consider group descritor
		 * corrupt and update bb_free using bitmap value
		 */
		grp->bb_free = free;
	}
	mb_set_largest_free_order(sb, grp);

	clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));

	period = get_cycles() - period;
	spin_lock(&EXT4_SB(sb)->s_bal_lock);
	EXT4_SB(sb)->s_mb_buddies_generated++;
	EXT4_SB(sb)->s_mb_generation_time += period;
	spin_unlock(&EXT4_SB(sb)->s_bal_lock);
}

/* The buddy information is attached the buddy cache inode
 * for convenience. The information regarding each group
 * is loaded via ext4_mb_load_buddy. The information involve
 * block bitmap and buddy information. The information are
 * stored in the inode as
 *
 * {                        page                        }
 * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
 *
 *
 * one block each for bitmap and buddy information.
 * So for each group we take up 2 blocks. A page can
 * contain blocks_per_page (PAGE_CACHE_SIZE / blocksize)  blocks.
 * So it can have information regarding groups_per_page which
 * is blocks_per_page/2
 *
 * Locking note:  This routine takes the block group lock of all groups
 * for this page; do not hold this lock when calling this routine!
 */

static int ext4_mb_init_cache(struct page *page, char *incore)
{
	ext4_group_t ngroups;
	int blocksize;
	int blocks_per_page;
	int groups_per_page;
	int err = 0;
	int i;
	ext4_group_t first_group;
	int first_block;
	struct super_block *sb;
	struct buffer_head *bhs;
	struct buffer_head **bh;
	struct inode *inode;
	char *data;
	char *bitmap;
	struct ext4_group_info *grinfo;

	mb_debug(1, "init page %lu\n", page->index);

	inode = page->mapping->host;
	sb = inode->i_sb;
	ngroups = ext4_get_groups_count(sb);
	blocksize = 1 << inode->i_blkbits;
	blocks_per_page = PAGE_CACHE_SIZE / blocksize;

	groups_per_page = blocks_per_page >> 1;
	if (groups_per_page == 0)
		groups_per_page = 1;

	/* allocate buffer_heads to read bitmaps */
	if (groups_per_page > 1) {
		err = -ENOMEM;
		i = sizeof(struct buffer_head *) * groups_per_page;
		bh = kzalloc(i, GFP_NOFS);
		if (bh == NULL)
			goto out;
	} else
		bh = &bhs;

	first_group = page->index * blocks_per_page / 2;

	/* read all groups the page covers into the cache */
	for (i = 0; i < groups_per_page; i++) {
		struct ext4_group_desc *desc;

		if (first_group + i >= ngroups)
			break;

		grinfo = ext4_get_group_info(sb, first_group + i);
		/*
		 * If page is uptodate then we came here after online resize
		 * which added some new uninitialized group info structs, so
		 * we must skip all initialized uptodate buddies on the page,
		 * which may be currently in use by an allocating task.
		 */
		if (PageUptodate(page) && !EXT4_MB_GRP_NEED_INIT(grinfo)) {
			bh[i] = NULL;
			continue;
		}

		err = -EIO;
		desc = ext4_get_group_desc(sb, first_group + i, NULL);
		if (desc == NULL)
			goto out;

		err = -ENOMEM;
		bh[i] = sb_getblk(sb, ext4_block_bitmap(sb, desc));
		if (bh[i] == NULL)
			goto out;

		if (bitmap_uptodate(bh[i]))
			continue;

		lock_buffer(bh[i]);
		if (bitmap_uptodate(bh[i])) {
			unlock_buffer(bh[i]);
			continue;
		}
		ext4_lock_group(sb, first_group + i);
		if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
			ext4_init_block_bitmap(sb, bh[i],
						first_group + i, desc);
			set_bitmap_uptodate(bh[i]);
			set_buffer_uptodate(bh[i]);
			ext4_unlock_group(sb, first_group + i);
			unlock_buffer(bh[i]);
			continue;
		}
		ext4_unlock_group(sb, first_group + i);
		if (buffer_uptodate(bh[i])) {
			/*
			 * if not uninit if bh is uptodate,
			 * bitmap is also uptodate
			 */
			set_bitmap_uptodate(bh[i]);
			unlock_buffer(bh[i]);
			continue;
		}
		get_bh(bh[i]);
		/*
		 * submit the buffer_head for read. We can
		 * safely mark the bitmap as uptodate now.
		 * We do it here so the bitmap uptodate bit
		 * get set with buffer lock held.
		 */
		set_bitmap_uptodate(bh[i]);
		bh[i]->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh[i]);
		mb_debug(1, "read bitmap for group %u\n", first_group + i);
	}

	/* wait for I/O completion */
	for (i = 0; i < groups_per_page; i++)
		if (bh[i])
			wait_on_buffer(bh[i]);

	err = -EIO;
	for (i = 0; i < groups_per_page; i++)
		if (bh[i] && !buffer_uptodate(bh[i]))
			goto out;

	err = 0;
	first_block = page->index * blocks_per_page;
	for (i = 0; i < blocks_per_page; i++) {
		int group;

		group = (first_block + i) >> 1;
		if (group >= ngroups)
			break;

		if (!bh[group - first_group])
			/* skip initialized uptodate buddy */
			continue;

		/*
		 * data carry information regarding this
		 * particular group in the format specified
		 * above
		 *
		 */
		data = page_address(page) + (i * blocksize);
		bitmap = bh[group - first_group]->b_data;

		/*
		 * We place the buddy block and bitmap block
		 * close together
		 */
		if ((first_block + i) & 1) {
			/* this is block of buddy */
			BUG_ON(incore == NULL);
			mb_debug(1, "put buddy for group %u in page %lu/%x\n",
				group, page->index, i * blocksize);
			trace_ext4_mb_buddy_bitmap_load(sb, group);
			grinfo = ext4_get_group_info(sb, group);
			grinfo->bb_fragments = 0;
			memset(grinfo->bb_counters, 0,
			       sizeof(*grinfo->bb_counters) *
				(sb->s_blocksize_bits+2));
			/*
			 * incore got set to the group block bitmap below
			 */
			ext4_lock_group(sb, group);
			/* init the buddy */
			memset(data, 0xff, blocksize);
			ext4_mb_generate_buddy(sb, data, incore, group);
			ext4_unlock_group(sb, group);
			incore = NULL;
		} else {
			/* this is block of bitmap */
			BUG_ON(incore != NULL);
			mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
				group, page->index, i * blocksize);
			trace_ext4_mb_bitmap_load(sb, group);

			/* see comments in ext4_mb_put_pa() */
			ext4_lock_group(sb, group);
			memcpy(data, bitmap, blocksize);

			/* mark all preallocated blks used in in-core bitmap */
			ext4_mb_generate_from_pa(sb, data, group);
			ext4_mb_generate_from_freelist(sb, data, group);
			ext4_unlock_group(sb, group);

			/* set incore so that the buddy information can be
			 * generated using this
			 */
			incore = data;
		}
	}
	SetPageUptodate(page);

out:
	if (bh) {
		for (i = 0; i < groups_per_page; i++)
			brelse(bh[i]);
		if (bh != &bhs)
			kfree(bh);
	}
	return err;
}

/*
 * Lock the buddy and bitmap pages. This make sure other parallel init_group
 * on the same buddy page doesn't happen whild holding the buddy page lock.
 * Return locked buddy and bitmap pages on e4b struct. If buddy and bitmap
 * are on the same page e4b->bd_buddy_page is NULL and return value is 0.
 */
static int ext4_mb_get_buddy_page_lock(struct super_block *sb,
		ext4_group_t group, struct ext4_buddy *e4b)
{
	struct inode *inode = EXT4_SB(sb)->s_buddy_cache;
	int block, pnum, poff;
	int blocks_per_page;
	struct page *page;

	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (!page)
		return -EIO;
	BUG_ON(page->mapping != inode->i_mapping);
	e4b->bd_bitmap_page = page;
	e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);

	if (blocks_per_page >= 2) {
		/* buddy and bitmap are on the same page */
		return 0;
	}

	block++;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (!page)
		return -EIO;
	BUG_ON(page->mapping != inode->i_mapping);
	e4b->bd_buddy_page = page;
	return 0;
}

static void ext4_mb_put_buddy_page_lock(struct ext4_buddy *e4b)
{
	if (e4b->bd_bitmap_page) {
		unlock_page(e4b->bd_bitmap_page);
		page_cache_release(e4b->bd_bitmap_page);
	}
	if (e4b->bd_buddy_page) {
		unlock_page(e4b->bd_buddy_page);
		page_cache_release(e4b->bd_buddy_page);
	}
}

/*
 * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
 * block group lock of all groups for this page; do not hold the BG lock when
 * calling this routine!
 */
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
{

	struct ext4_group_info *this_grp;
	struct ext4_buddy e4b;
	struct page *page;
	int ret = 0;

	mb_debug(1, "init group %u\n", group);
	this_grp = ext4_get_group_info(sb, group);
	/*
	 * This ensures that we don't reinit the buddy cache
	 * page which map to the group from which we are already
	 * allocating. If we are looking at the buddy cache we would
	 * have taken a reference using ext4_mb_load_buddy and that
	 * would have pinned buddy page to page cache.
	 */
	ret = ext4_mb_get_buddy_page_lock(sb, group, &e4b);
	if (ret || !EXT4_MB_GRP_NEED_INIT(this_grp)) {
		/*
		 * somebody initialized the group
		 * return without doing anything
		 */
		goto err;
	}

	page = e4b.bd_bitmap_page;
	ret = ext4_mb_init_cache(page, NULL);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);

	if (e4b.bd_buddy_page == NULL) {
		/*
		 * If both the bitmap and buddy are in
		 * the same page we don't need to force
		 * init the buddy
		 */
		ret = 0;
		goto err;
	}
	/* init buddy cache */
	page = e4b.bd_buddy_page;
	ret = ext4_mb_init_cache(page, e4b.bd_bitmap);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);
err:
	ext4_mb_put_buddy_page_lock(&e4b);
	return ret;
}

/*
 * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
 * block group lock of all groups for this page; do not hold the BG lock when
 * calling this routine!
 */
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
					struct ext4_buddy *e4b)
{
	int blocks_per_page;
	int block;
	int pnum;
	int poff;
	struct page *page;
	int ret;
	struct ext4_group_info *grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;

	mb_debug(1, "load group %u\n", group);

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	grp = ext4_get_group_info(sb, group);

	e4b->bd_blkbits = sb->s_blocksize_bits;
	e4b->bd_info = grp;
	e4b->bd_sb = sb;
	e4b->bd_group = group;
	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;

	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
		/*
		 * we need full data about the group
		 * to make a good selection
		 */
		ret = ext4_mb_init_group(sb, group);
		if (ret)
			return ret;
	}

	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;

	/* we could use find_or_create_page(), but it locks page
	 * what we'd like to avoid in fast path ... */
	page = find_get_page(inode->i_mapping, pnum);
	if (page == NULL || !PageUptodate(page)) {
		if (page)
			/*
			 * drop the page reference and try
			 * to get the page with lock. If we
			 * are not uptodate that implies
			 * somebody just created the page but
			 * is yet to initialize the same. So
			 * wait for it to initialize.
			 */
			page_cache_release(page);
		page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
		if (page) {
			BUG_ON(page->mapping != inode->i_mapping);
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, NULL);
				if (ret) {
					unlock_page(page);
					goto err;
				}
				mb_cmp_bitmaps(e4b, page_address(page) +
					       (poff * sb->s_blocksize));
			}
			unlock_page(page);
		}
	}
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	e4b->bd_bitmap_page = page;
	e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);
	mark_page_accessed(page);

	block++;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;

	page = find_get_page(inode->i_mapping, pnum);
	if (page == NULL || !PageUptodate(page)) {
		if (page)
			page_cache_release(page);
		page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
		if (page) {
			BUG_ON(page->mapping != inode->i_mapping);
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
				if (ret) {
					unlock_page(page);
					goto err;
				}
			}
			unlock_page(page);
		}
	}
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	e4b->bd_buddy_page = page;
	e4b->bd_buddy = page_address(page) + (poff * sb->s_blocksize);
	mark_page_accessed(page);

	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

	return 0;

err:
	if (page)
		page_cache_release(page);
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
	e4b->bd_buddy = NULL;
	e4b->bd_bitmap = NULL;
	return ret;
}

static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
{
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
}


static int mb_find_order_for_block(struct ext4_buddy *e4b, int block)
{
	int order = 1;
	void *bb;

	BUG_ON(EXT4_MB_BITMAP(e4b) == EXT4_MB_BUDDY(e4b));
	BUG_ON(block >= (1 << (e4b->bd_blkbits + 3)));

	bb = EXT4_MB_BUDDY(e4b);
	while (order <= e4b->bd_blkbits + 1) {
		block = block >> 1;
		if (!mb_test_bit(block, bb)) {
			/* this block is part of buddy of order 'order' */
			return order;
		}
		bb += 1 << (e4b->bd_blkbits - order);
		order++;
	}
	return 0;
}

static void mb_clear_bits(void *bm, int cur, int len)
{
	__u32 *addr;

	len = cur + len;
	while (cur < len) {
		if ((cur & 31) == 0 && (len - cur) >= 32) {
			/* fast path: clear whole word at once */
			addr = bm + (cur >> 3);
			*addr = 0;
			cur += 32;
			continue;
		}
		mb_clear_bit(cur, bm);
		cur++;
	}
}

void ext4_set_bits(void *bm, int cur, int len)
{
	__u32 *addr;

	len = cur + len;
	while (cur < len) {
		if ((cur & 31) == 0 && (len - cur) >= 32) {
			/* fast path: set whole word at once */
			addr = bm + (cur >> 3);
			*addr = 0xffffffff;
			cur += 32;
			continue;
		}
		mb_set_bit(cur, bm);
		cur++;
	}
}

static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
			  int first, int count)
{
	int block = 0;
	int max = 0;
	int order;
	void *buddy;
	void *buddy2;
	struct super_block *sb = e4b->bd_sb;

	BUG_ON(first + count > (sb->s_blocksize << 3));
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
	mb_check_buddy(e4b);
	mb_free_blocks_double(inode, e4b, first, count);

	e4b->bd_info->bb_free += count;
	if (first < e4b->bd_info->bb_first_free)
		e4b->bd_info->bb_first_free = first;

	/* let's maintain fragments counter */
	if (first != 0)
		block = !mb_test_bit(first - 1, EXT4_MB_BITMAP(e4b));
	if (first + count < EXT4_SB(sb)->s_mb_maxs[0])
		max = !mb_test_bit(first + count, EXT4_MB_BITMAP(e4b));
	if (block && max)
		e4b->bd_info->bb_fragments--;
	else if (!block && !max)
		e4b->bd_info->bb_fragments++;

	/* let's maintain buddy itself */
	while (count-- > 0) {
		block = first++;
		order = 0;

		if (!mb_test_bit(block, EXT4_MB_BITMAP(e4b))) {
			ext4_fsblk_t blocknr;

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
			blocknr += block;
			ext4_grp_locked_error(sb, e4b->bd_group,
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing already freed block "
					      "(bit %u)", block);
		}
		mb_clear_bit(block, EXT4_MB_BITMAP(e4b));
		e4b->bd_info->bb_counters[order]++;

		/* start of the buddy */
		buddy = mb_find_buddy(e4b, order, &max);

		do {
			block &= ~1UL;
			if (mb_test_bit(block, buddy) ||
					mb_test_bit(block + 1, buddy))
				break;

			/* both the buddies are free, try to coalesce them */
			buddy2 = mb_find_buddy(e4b, order + 1, &max);

			if (!buddy2)
				break;

			if (order > 0) {
				/* for special purposes, we don't set
				 * free bits in bitmap */
				mb_set_bit(block, buddy);
				mb_set_bit(block + 1, buddy);
			}
			e4b->bd_info->bb_counters[order]--;
			e4b->bd_info->bb_counters[order]--;

			block = block >> 1;
			order++;
			e4b->bd_info->bb_counters[order]++;

			mb_clear_bit(block, buddy2);
			buddy = buddy2;
		} while (1);
	}
	mb_set_largest_free_order(sb, e4b->bd_info);
	mb_check_buddy(e4b);
}

static int mb_find_extent(struct ext4_buddy *e4b, int order, int block,
				int needed, struct ext4_free_extent *ex)
{
	int next = block;
	int max;
	int ord;
	void *buddy;

	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
	BUG_ON(ex == NULL);

	buddy = mb_find_buddy(e4b, order, &max);
	BUG_ON(buddy == NULL);
	BUG_ON(block >= max);
	if (mb_test_bit(block, buddy)) {
		ex->fe_len = 0;
		ex->fe_start = 0;
		ex->fe_group = 0;
		return 0;
	}

	/* FIXME dorp order completely ? */
	if (likely(order == 0)) {
		/* find actual order */
		order = mb_find_order_for_block(e4b, block);
		block = block >> order;
	}

	ex->fe_len = 1 << order;
	ex->fe_start = block << order;
	ex->fe_group = e4b->bd_group;

	/* calc difference from given start */
	next = next - ex->fe_start;
	ex->fe_len -= next;
	ex->fe_start += next;

	while (needed > ex->fe_len &&
	       (buddy = mb_find_buddy(e4b, order, &max))) {

		if (block + 1 >= max)
			break;

		next = (block + 1) * (1 << order);
		if (mb_test_bit(next, EXT4_MB_BITMAP(e4b)))
			break;

		ord = mb_find_order_for_block(e4b, next);

		order = ord;
		block = next >> order;
		ex->fe_len += 1 << order;
	}

	BUG_ON(ex->fe_start + ex->fe_len > (1 << (e4b->bd_blkbits + 3)));
	return ex->fe_len;
}

static int mb_mark_used(struct ext4_buddy *e4b, struct ext4_free_extent *ex)
{
	int ord;
	int mlen = 0;
	int max = 0;
	int cur;
	int start = ex->fe_start;
	int len = ex->fe_len;
	unsigned ret = 0;
	int len0 = len;
	void *buddy;

	BUG_ON(start + len > (e4b->bd_sb->s_blocksize << 3));
	BUG_ON(e4b->bd_group != ex->fe_group);
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
	mb_check_buddy(e4b);
	mb_mark_used_double(e4b, start, len);

	e4b->bd_info->bb_free -= len;
	if (e4b->bd_info->bb_first_free == start)
		e4b->bd_info->bb_first_free += len;

	/* let's maintain fragments counter */
	if (start != 0)
		mlen = !mb_test_bit(start - 1, EXT4_MB_BITMAP(e4b));
	if (start + len < EXT4_SB(e4b->bd_sb)->s_mb_maxs[0])
		max = !mb_test_bit(start + len, EXT4_MB_BITMAP(e4b));
	if (mlen && max)
		e4b->bd_info->bb_fragments++;
	else if (!mlen && !max)
		e4b->bd_info->bb_fragments--;

	/* let's maintain buddy itself */
	while (len) {
		ord = mb_find_order_for_block(e4b, start);

		if (((start >> ord) << ord) == start && len >= (1 << ord)) {
			/* the whole chunk may be allocated at once! */
			mlen = 1 << ord;
			buddy = mb_find_buddy(e4b, ord, &max);
			BUG_ON((start >> ord) >= max);
			mb_set_bit(start >> ord, buddy);
			e4b->bd_info->bb_counters[ord]--;
			start += mlen;
			len -= mlen;
			BUG_ON(len < 0);
			continue;
		}

		/* store for history */
		if (ret == 0)
			ret = len | (ord << 16);

		/* we have to split large buddy */
		BUG_ON(ord <= 0);
		buddy = mb_find_buddy(e4b, ord, &max);
		mb_set_bit(start >> ord, buddy);
		e4b->bd_info->bb_counters[ord]--;

		ord--;
		cur = (start >> ord) & ~1U;
		buddy = mb_find_buddy(e4b, ord, &max);
		mb_clear_bit(cur, buddy);
		mb_clear_bit(cur + 1, buddy);
		e4b->bd_info->bb_counters[ord]++;
		e4b->bd_info->bb_counters[ord]++;
	}
	mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);

	ext4_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
	mb_check_buddy(e4b);

	return ret;
}

/*
 * Must be called under group lock!
 */
static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	int ret;

	BUG_ON(ac->ac_b_ex.fe_group != e4b->bd_group);
	BUG_ON(ac->ac_status == AC_STATUS_FOUND);

	ac->ac_b_ex.fe_len = min(ac->ac_b_ex.fe_len, ac->ac_g_ex.fe_len);
	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
	ret = mb_mark_used(e4b, &ac->ac_b_ex);

	/* preallocation can change ac_b_ex, thus we store actually
	 * allocated blocks for history */
	ac->ac_f_ex = ac->ac_b_ex;

	ac->ac_status = AC_STATUS_FOUND;
	ac->ac_tail = ret & 0xffff;
	ac->ac_buddy = ret >> 16;

	/*
	 * take the page reference. We want the page to be pinned
	 * so that we don't get a ext4_mb_init_cache_call for this
	 * group until we update the bitmap. That would mean we
	 * double allocate blocks. The reference is dropped
	 * in ext4_mb_release_context
	 */
	ac->ac_bitmap_page = e4b->bd_bitmap_page;
	get_page(ac->ac_bitmap_page);
	ac->ac_buddy_page = e4b->bd_buddy_page;
	get_page(ac->ac_buddy_page);
	/* store last allocated for subsequent stream allocation */
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
		spin_lock(&sbi->s_md_lock);
		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
		spin_unlock(&sbi->s_md_lock);
	}
}

/*
 * regular allocator, for general purposes allocation
 */

static void ext4_mb_check_limits(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b,
					int finish_group)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	struct ext4_free_extent *bex = &ac->ac_b_ex;
	struct ext4_free_extent *gex = &ac->ac_g_ex;
	struct ext4_free_extent ex;
	int max;

	if (ac->ac_status == AC_STATUS_FOUND)
		return;
	/*
	 * We don't want to scan for a whole year
	 */
	if (ac->ac_found > sbi->s_mb_max_to_scan &&
			!(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
		ac->ac_status = AC_STATUS_BREAK;
		return;
	}

	/*
	 * Haven't found good chunk so far, let's continue
	 */
	if (bex->fe_len < gex->fe_len)
		return;

	if ((finish_group || ac->ac_found > sbi->s_mb_min_to_scan)
			&& bex->fe_group == e4b->bd_group) {
		/* recheck chunk's availability - we don't know
		 * when it was found (within this lock-unlock
		 * period or not) */
		max = mb_find_extent(e4b, 0, bex->fe_start, gex->fe_len, &ex);
		if (max >= gex->fe_len) {
			ext4_mb_use_best_found(ac, e4b);
			return;
		}
	}
}

/*
 * The routine checks whether found extent is good enough. If it is,
 * then the extent gets marked used and flag is set to the context
 * to stop scanning. Otherwise, the extent is compared with the
 * previous found extent and if new one is better, then it's stored
 * in the context. Later, the best found extent will be used, if
 * mballoc can't find good enough extent.
 *
 * FIXME: real allocation policy is to be designed yet!
 */
static void ext4_mb_measure_extent(struct ext4_allocation_context *ac,
					struct ext4_free_extent *ex,
					struct ext4_buddy *e4b)
{
	struct ext4_free_extent *bex = &ac->ac_b_ex;
	struct ext4_free_extent *gex = &ac->ac_g_ex;

	BUG_ON(ex->fe_len <= 0);
	BUG_ON(ex->fe_len > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
	BUG_ON(ex->fe_start >= EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
	BUG_ON(ac->ac_status != AC_STATUS_CONTINUE);

	ac->ac_found++;

	/*
	 * The special case - take what you catch first
	 */
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
		*bex = *ex;
		ext4_mb_use_best_found(ac, e4b);
		return;
	}

	/*
	 * Let's check whether the chuck is good enough
	 */
	if (ex->fe_len == gex->fe_len) {
		*bex = *ex;
		ext4_mb_use_best_found(ac, e4b);
		return;
	}

	/*
	 * If this is first found extent, just store it in the context
	 */
	if (bex->fe_len == 0) {
		*bex = *ex;
		return;
	}

	/*
	 * If new found extent is better, store it in the context
	 */
	if (bex->fe_len < gex->fe_len) {
		/* if the request isn't satisfied, any found extent
		 * larger than previous best one is better */
		if (ex->fe_len > bex->fe_len)
			*bex = *ex;
	} else if (ex->fe_len > gex->fe_len) {
		/* if the request is satisfied, then we try to find
		 * an extent that still satisfy the request, but is
		 * smaller than previous one */
		if (ex->fe_len < bex->fe_len)
			*bex = *ex;
	}

	ext4_mb_check_limits(ac, e4b, 0);
}

static noinline_for_stack
int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b)
{
	struct ext4_free_extent ex = ac->ac_b_ex;
	ext4_group_t group = ex.fe_group;
	int max;
	int err;

	BUG_ON(ex.fe_len <= 0);
	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
	if (err)
		return err;

	ext4_lock_group(ac->ac_sb, group);
	max = mb_find_extent(e4b, 0, ex.fe_start, ex.fe_len, &ex);

	if (max > 0) {
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	}

	ext4_unlock_group(ac->ac_sb, group);
	ext4_mb_unload_buddy(e4b);

	return 0;
}

static noinline_for_stack
int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
				struct ext4_buddy *e4b)
{
	ext4_group_t group = ac->ac_g_ex.fe_group;
	int max;
	int err;
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	struct ext4_free_extent ex;

	if (!(ac->ac_flags & EXT4_MB_HINT_TRY_GOAL))
		return 0;

	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
	if (err)
		return err;

	ext4_lock_group(ac->ac_sb, group);
	max = mb_find_extent(e4b, 0, ac->ac_g_ex.fe_start,
			     ac->ac_g_ex.fe_len, &ex);

	if (max >= ac->ac_g_ex.fe_len && ac->ac_g_ex.fe_len == sbi->s_stripe) {
		ext4_fsblk_t start;

		start = ext4_group_first_block_no(ac->ac_sb, e4b->bd_group) +
			ex.fe_start;
		/* use do_div to get remainder (would be 64-bit modulo) */
		if (do_div(start, sbi->s_stripe) == 0) {
			ac->ac_found++;
			ac->ac_b_ex = ex;
			ext4_mb_use_best_found(ac, e4b);
		}
	} else if (max >= ac->ac_g_ex.fe_len) {
		BUG_ON(ex.fe_len <= 0);
		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
		ac->ac_found++;
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	} else if (max > 0 && (ac->ac_flags & EXT4_MB_HINT_MERGE)) {
		/* Sometimes, caller may want to merge even small
		 * number of blocks to an existing extent */
		BUG_ON(ex.fe_len <= 0);
		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
		ac->ac_found++;
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	}
	ext4_unlock_group(ac->ac_sb, group);
	ext4_mb_unload_buddy(e4b);

	return 0;
}

/*
 * The routine scans buddy structures (not bitmap!) from given order
 * to max order and tries to find big enough chunk to satisfy the req
 */
static noinline_for_stack
void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_group_info *grp = e4b->bd_info;
	void *buddy;
	int i;
	int k;
	int max;

	BUG_ON