balloc.c

linux 内核源代码

/*
 *  linux/fs/ext3/balloc.c
 *
 * Copyright (C) 1992, 1993, 1994, 1995
 * Remy Card ([e-mail])
 * Laboratoire MASI - Institut Blaise Pascal
 * Universite Pierre et Marie Curie (Paris VI)
 *
 *  Enhanced block allocation by Stephen Tweedie ([e-mail]), 1993
 *  Big-endian to little-endian byte-swapping/bitmaps by
 *        David S. Miller ([e-mail]), 1995
 */

#include <linux/time.h>
#include <linux/capability.h>
#include <linux/fs.h>
#include <linux/jbd.h>
#include <linux/ext3_fs.h>
#include <linux/ext3_jbd.h>
#include <linux/quotaops.h>
#include <linux/buffer_head.h>

/*
 * balloc.c contains the blocks allocation and deallocation routines
 */

/*
 * The free blocks are managed by bitmaps.  A file system contains several
 * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 * block for inodes, N blocks for the inode table and data blocks.
 *
 * The file system contains group descriptors which are located after the
 * super block.  Each descriptor contains the number of the bitmap block and
 * the free blocks count in the block.  The descriptors are loaded in memory
 * when a file system is mounted (see ext3_fill_super).
 */


#define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)

/**
 * ext3_get_group_desc() -- load group descriptor from disk
 * @sb:			super block
 * @block_group:	given block group
 * @bh:			pointer to the buffer head to store the block
 *			group descriptor
 */
struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
					     unsigned int block_group,
					     struct buffer_head ** bh)
{
	unsigned long group_desc;
	unsigned long offset;
	struct ext3_group_desc * desc;
	struct ext3_sb_info *sbi = EXT3_SB(sb);

	if (block_group >= sbi->s_groups_count) {
		ext3_error (sb, "ext3_get_group_desc",
			    "block_group >= groups_count - "
			    "block_group = %d, groups_count = %lu",
			    block_group, sbi->s_groups_count);

		return NULL;
	}
	smp_rmb();

	group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
	offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
	if (!sbi->s_group_desc[group_desc]) {
		ext3_error (sb, "ext3_get_group_desc",
			    "Group descriptor not loaded - "
			    "block_group = %d, group_desc = %lu, desc = %lu",
			     block_group, group_desc, offset);
		return NULL;
	}

	desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
	if (bh)
		*bh = sbi->s_group_desc[group_desc];
	return desc + offset;
}

/**
 * read_block_bitmap()
 * @sb:			super block
 * @block_group:	given block group
 *
 * Read the bitmap for a given block_group, reading into the specified
 * slot in the superblock's bitmap cache.
 *
 * Return buffer_head on success or NULL in case of failure.
 */
static struct buffer_head *
read_block_bitmap(struct super_block *sb, unsigned int block_group)
{
	struct ext3_group_desc * desc;
	struct buffer_head * bh = NULL;

	desc = ext3_get_group_desc (sb, block_group, NULL);
	if (!desc)
		goto error_out;
	bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
	if (!bh)
		ext3_error (sb, "read_block_bitmap",
			    "Cannot read block bitmap - "
			    "block_group = %d, block_bitmap = %u",
			    block_group, le32_to_cpu(desc->bg_block_bitmap));
error_out:
	return bh;
}
/*
 * The reservation window structure operations
 * --------------------------------------------
 * Operations include:
 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
 *
 * We use a red-black tree to represent per-filesystem reservation
 * windows.
 *
 */

/**
 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
 * @rb_root:		root of per-filesystem reservation rb tree
 * @verbose:		verbose mode
 * @fn:			function which wishes to dump the reservation map
 *
 * If verbose is turned on, it will print the whole block reservation
 * windows(start, end).	Otherwise, it will only print out the "bad" windows,
 * those windows that overlap with their immediate neighbors.
 */
#if 1
static void __rsv_window_dump(struct rb_root *root, int verbose,
			      const char *fn)
{
	struct rb_node *n;
	struct ext3_reserve_window_node *rsv, *prev;
	int bad;

restart:
	n = rb_first(root);
	bad = 0;
	prev = NULL;

	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
	while (n) {
		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
		if (verbose)
			printk("reservation window 0x%p "
			       "start:  %lu, end:  %lu\n",
			       rsv, rsv->rsv_start, rsv->rsv_end);
		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
			printk("Bad reservation %p (start >= end)\n",
			       rsv);
			bad = 1;
		}
		if (prev && prev->rsv_end >= rsv->rsv_start) {
			printk("Bad reservation %p (prev->end >= start)\n",
			       rsv);
			bad = 1;
		}
		if (bad) {
			if (!verbose) {
				printk("Restarting reservation walk in verbose mode\n");
				verbose = 1;
				goto restart;
			}
		}
		n = rb_next(n);
		prev = rsv;
	}
	printk("Window map complete.\n");
	if (bad)
		BUG();
}
#define rsv_window_dump(root, verbose) \
	__rsv_window_dump((root), (verbose), __FUNCTION__)
#else
#define rsv_window_dump(root, verbose) do {} while (0)
#endif

/**
 * goal_in_my_reservation()
 * @rsv:		inode's reservation window
 * @grp_goal:		given goal block relative to the allocation block group
 * @group:		the current allocation block group
 * @sb:			filesystem super block
 *
 * Test if the given goal block (group relative) is within the file's
 * own block reservation window range.
 *
 * If the reservation window is outside the goal allocation group, return 0;
 * grp_goal (given goal block) could be -1, which means no specific
 * goal block. In this case, always return 1.
 * If the goal block is within the reservation window, return 1;
 * otherwise, return 0;
 */
static int
goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
			unsigned int group, struct super_block * sb)
{
	ext3_fsblk_t group_first_block, group_last_block;

	group_first_block = ext3_group_first_block_no(sb, group);
	group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);

	if ((rsv->_rsv_start > group_last_block) ||
	    (rsv->_rsv_end < group_first_block))
		return 0;
	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
		|| (grp_goal + group_first_block > rsv->_rsv_end)))
		return 0;
	return 1;
}

/**
 * search_reserve_window()
 * @rb_root:		root of reservation tree
 * @goal:		target allocation block
 *
 * Find the reserved window which includes the goal, or the previous one
 * if the goal is not in any window.
 * Returns NULL if there are no windows or if all windows start after the goal.
 */
static struct ext3_reserve_window_node *
search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
{
	struct rb_node *n = root->rb_node;
	struct ext3_reserve_window_node *rsv;

	if (!n)
		return NULL;

	do {
		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);

		if (goal < rsv->rsv_start)
			n = n->rb_left;
		else if (goal > rsv->rsv_end)
			n = n->rb_right;
		else
			return rsv;
	} while (n);
	/*
	 * We've fallen off the end of the tree: the goal wasn't inside
	 * any particular node.  OK, the previous node must be to one
	 * side of the interval containing the goal.  If it's the RHS,
	 * we need to back up one.
	 */
	if (rsv->rsv_start > goal) {
		n = rb_prev(&rsv->rsv_node);
		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
	}
	return rsv;
}

/**
 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
 * @sb:			super block
 * @rsv:		reservation window to add
 *
 * Must be called with rsv_lock hold.
 */
void ext3_rsv_window_add(struct super_block *sb,
		    struct ext3_reserve_window_node *rsv)
{
	struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
	struct rb_node *node = &rsv->rsv_node;
	ext3_fsblk_t start = rsv->rsv_start;

	struct rb_node ** p = &root->rb_node;
	struct rb_node * parent = NULL;
	struct ext3_reserve_window_node *this;

	while (*p)
	{
		parent = *p;
		this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);

		if (start < this->rsv_start)
			p = &(*p)->rb_left;
		else if (start > this->rsv_end)
			p = &(*p)->rb_right;
		else {
			rsv_window_dump(root, 1);
			BUG();
		}
	}

	rb_link_node(node, parent, p);
	rb_insert_color(node, root);
}

/**
 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
 * @sb:			super block
 * @rsv:		reservation window to remove
 *
 * Mark the block reservation window as not allocated, and unlink it
 * from the filesystem reservation window rb tree. Must be called with
 * rsv_lock hold.
 */
static void rsv_window_remove(struct super_block *sb,
			      struct ext3_reserve_window_node *rsv)
{
	rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
	rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
	rsv->rsv_alloc_hit = 0;
	rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
}

/*
 * rsv_is_empty() -- Check if the reservation window is allocated.
 * @rsv:		given reservation window to check
 *
 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
 */
static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
{
	/* a valid reservation end block could not be 0 */
	return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
}

/**
 * ext3_init_block_alloc_info()
 * @inode:		file inode structure
 *
 * Allocate and initialize the	reservation window structure, and
 * link the window to the ext3 inode structure at last
 *
 * The reservation window structure is only dynamically allocated
 * and linked to ext3 inode the first time the open file
 * needs a new block. So, before every ext3_new_block(s) call, for
 * regular files, we should check whether the reservation window
 * structure exists or not. In the latter case, this function is called.
 * Fail to do so will result in block reservation being turned off for that
 * open file.
 *
 * This function is called from ext3_get_blocks_handle(), also called
 * when setting the reservation window size through ioctl before the file
 * is open for write (needs block allocation).
 *
 * Needs truncate_mutex protection prior to call this function.
 */
void ext3_init_block_alloc_info(struct inode *inode)
{
	struct ext3_inode_info *ei = EXT3_I(inode);
	struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
	struct super_block *sb = inode->i_sb;

	block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
	if (block_i) {
		struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;

		rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
		rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;

		/*
		 * if filesystem is mounted with NORESERVATION, the goal
		 * reservation window size is set to zero to indicate
		 * block reservation is off
		 */
		if (!test_opt(sb, RESERVATION))
			rsv->rsv_goal_size = 0;
		else
			rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
		rsv->rsv_alloc_hit = 0;
		block_i->last_alloc_logical_block = 0;
		block_i->last_alloc_physical_block = 0;
	}
	ei->i_block_alloc_info = block_i;
}

/**
 * ext3_discard_reservation()
 * @inode:		inode
 *
 * Discard(free) block reservation window on last file close, or truncate
 * or at last iput().
 *
 * It is being called in three cases:
 *	ext3_release_file(): last writer close the file
 *	ext3_clear_inode(): last iput(), when nobody link to this file.
 *	ext3_truncate(): when the block indirect map is about to change.
 *
 */
void ext3_discard_reservation(struct inode *inode)
{
	struct ext3_inode_info *ei = EXT3_I(inode);
	struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
	struct ext3_reserve_window_node *rsv;
	spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;

	if (!block_i)
		return;

	rsv = &block_i->rsv_window_node;
	if (!rsv_is_empty(&rsv->rsv_window)) {
		spin_lock(rsv_lock);
		if (!rsv_is_empty(&rsv->rsv_window))
			rsv_window_remove(inode->i_sb, rsv);
		spin_unlock(rsv_lock);
	}
}

/**
 * ext3_free_blocks_sb() -- Free given blocks and update quota
 * @handle:			handle to this transaction
 * @sb:				super block
 * @block:			start physcial block to free
 * @count:			number of blocks to free
 * @pdquot_freed_blocks:	pointer to quota
 */
void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
			 ext3_fsblk_t block, unsigned long count,
			 unsigned long *pdquot_freed_blocks)
{
	struct buffer_head *bitmap_bh = NULL;
	struct buffer_head *gd_bh;
	unsigned long block_group;
	ext3_grpblk_t bit;
	unsigned long i;
	unsigned long overflow;
	struct ext3_group_desc * desc;
	struct ext3_super_block * es;
	struct ext3_sb_info *sbi;
	int err = 0, ret;
	ext3_grpblk_t group_freed;

	*pdquot_freed_blocks = 0;
	sbi = EXT3_SB(sb);
	es = sbi->s_es;
	if (block < le32_to_cpu(es->s_first_data_block) ||
	    block + count < block ||
	    block + count > le32_to_cpu(es->s_blocks_count)) {
		ext3_error (sb, "ext3_free_blocks",
			    "Freeing blocks not in datazone - "
			    "block = "E3FSBLK", count = %lu", block, count);
		goto error_return;
	}

	ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);

do_more:
	overflow = 0;
	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
		      EXT3_BLOCKS_PER_GROUP(sb);
	bit = (block - le32_to_cpu(es->s_first_data_block)) %
		      EXT3_BLOCKS_PER_GROUP(sb);
	/*
	 * Check to see if we are freeing blocks across a group
	 * boundary.
	 */
	if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {