[deliverable/linux.git] / fs / ext4 / dir.c

/*
 *  linux/fs/ext4/dir.c
 *
 * Copyright (C) 1992, 1993, 1994, 1995
 * Remy Card (card@masi.ibp.fr)
 * Laboratoire MASI - Institut Blaise Pascal
 * Universite Pierre et Marie Curie (Paris VI)
 *
 *  from
 *
 *  linux/fs/minix/dir.c
 *
 *  Copyright (C) 1991, 1992  Linus Torvalds
 *
 *  ext4 directory handling functions
 *
 *  Big-endian to little-endian byte-swapping/bitmaps by
 *        David S. Miller (davem@caip.rutgers.edu), 1995
 *
 * Hash Tree Directory indexing (c) 2001  Daniel Phillips
 *
 */

#include <linux/fs.h>
#include <linux/jbd2.h>
#include <linux/ext4_fs.h>
#include <linux/buffer_head.h>
#include <linux/smp_lock.h>
#include <linux/slab.h>
#include <linux/rbtree.h>

static unsigned char ext4_filetype_table[] = {
	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
};

static int ext4_readdir(struct file *, void *, filldir_t);
static int ext4_dx_readdir(struct file * filp,
			   void * dirent, filldir_t filldir);
static int ext4_release_dir (struct inode * inode,
				struct file * filp);

const struct file_operations ext4_dir_operations = {
	.llseek		= generic_file_llseek,
	.read		= generic_read_dir,
	.readdir	= ext4_readdir,		/* we take BKL. needed?*/
	.ioctl		= ext4_ioctl,		/* BKL held */
#ifdef CONFIG_COMPAT
	.compat_ioctl	= ext4_compat_ioctl,
#endif
	.fsync		= ext4_sync_file,	/* BKL held */
#ifdef CONFIG_EXT4_INDEX
	.release	= ext4_release_dir,
#endif
};


static unsigned char get_dtype(struct super_block *sb, int filetype)
{
	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
	    (filetype >= EXT4_FT_MAX))
		return DT_UNKNOWN;

	return (ext4_filetype_table[filetype]);
}


int ext4_check_dir_entry (const char * function, struct inode * dir,
			  struct ext4_dir_entry_2 * de,
			  struct buffer_head * bh,
			  unsigned long offset)
{
	const char * error_msg = NULL;
	const int rlen = le16_to_cpu(de->rec_len);

	if (rlen < EXT4_DIR_REC_LEN(1))
		error_msg = "rec_len is smaller than minimal";
	else if (rlen % 4 != 0)
		error_msg = "rec_len % 4 != 0";
	else if (rlen < EXT4_DIR_REC_LEN(de->name_len))
		error_msg = "rec_len is too small for name_len";
	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
		error_msg = "directory entry across blocks";
	else if (le32_to_cpu(de->inode) >
			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))
		error_msg = "inode out of bounds";

	if (error_msg != NULL)
		ext4_error (dir->i_sb, function,
			"bad entry in directory #%lu: %s - "
			"offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
			dir->i_ino, error_msg, offset,
			(unsigned long) le32_to_cpu(de->inode),
			rlen, de->name_len);
	return error_msg == NULL ? 1 : 0;
}

static int ext4_readdir(struct file * filp,
			 void * dirent, filldir_t filldir)
{
	int error = 0;
	unsigned long offset;
	int i, stored;
	struct ext4_dir_entry_2 *de;
	struct super_block *sb;
	int err;
	struct inode *inode = filp->f_dentry->d_inode;
	int ret = 0;

	sb = inode->i_sb;

#ifdef CONFIG_EXT4_INDEX
	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
				    EXT4_FEATURE_COMPAT_DIR_INDEX) &&
	    ((EXT4_I(inode)->i_flags & EXT4_INDEX_FL) ||
	     ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
		err = ext4_dx_readdir(filp, dirent, filldir);
		if (err != ERR_BAD_DX_DIR) {
			ret = err;
			goto out;
		}
		/*
		 * We don't set the inode dirty flag since it's not
		 * critical that it get flushed back to the disk.
		 */
		EXT4_I(filp->f_dentry->d_inode)->i_flags &= ~EXT4_INDEX_FL;
	}
#endif
	stored = 0;
	offset = filp->f_pos & (sb->s_blocksize - 1);

	while (!error && !stored && filp->f_pos < inode->i_size) {
		unsigned long blk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
		struct buffer_head map_bh;
		struct buffer_head *bh = NULL;

		map_bh.b_state = 0;
		err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh, 0, 0);
		if (err > 0) {
			page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
				&filp->f_ra,
				filp,
				map_bh.b_blocknr >>
					(PAGE_CACHE_SHIFT - inode->i_blkbits),
				1);
			bh = ext4_bread(NULL, inode, blk, 0, &err);
		}

		/*
		 * We ignore I/O errors on directories so users have a chance
		 * of recovering data when there's a bad sector
		 */
		if (!bh) {
			ext4_error (sb, "ext4_readdir",
				"directory #%lu contains a hole at offset %lu",
				inode->i_ino, (unsigned long)filp->f_pos);
			filp->f_pos += sb->s_blocksize - offset;
			continue;
		}

revalidate:
		/* If the dir block has changed since the last call to
		 * readdir(2), then we might be pointing to an invalid
		 * dirent right now.  Scan from the start of the block
		 * to make sure. */
		if (filp->f_version != inode->i_version) {
			for (i = 0; i < sb->s_blocksize && i < offset; ) {
				de = (struct ext4_dir_entry_2 *)
					(bh->b_data + i);
				/* It's too expensive to do a full
				 * dirent test each time round this
				 * loop, but we do have to test at
				 * least that it is non-zero.  A
				 * failure will be detected in the
				 * dirent test below. */
				if (le16_to_cpu(de->rec_len) <
						EXT4_DIR_REC_LEN(1))
					break;
				i += le16_to_cpu(de->rec_len);
			}
			offset = i;
			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
				| offset;
			filp->f_version = inode->i_version;
		}

		while (!error && filp->f_pos < inode->i_size
		       && offset < sb->s_blocksize) {
			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
			if (!ext4_check_dir_entry ("ext4_readdir", inode, de,
						   bh, offset)) {
				/* On error, skip the f_pos to the
                                   next block. */
				filp->f_pos = (filp->f_pos |
						(sb->s_blocksize - 1)) + 1;
				brelse (bh);
				ret = stored;
				goto out;
			}
			offset += le16_to_cpu(de->rec_len);
			if (le32_to_cpu(de->inode)) {
				/* We might block in the next section
				 * if the data destination is
				 * currently swapped out.  So, use a
				 * version stamp to detect whether or
				 * not the directory has been modified
				 * during the copy operation.
				 */
				unsigned long version = filp->f_version;

				error = filldir(dirent, de->name,
						de->name_len,
						filp->f_pos,
						le32_to_cpu(de->inode),
						get_dtype(sb, de->file_type));
				if (error)
					break;
				if (version != filp->f_version)
					goto revalidate;
				stored ++;
			}
			filp->f_pos += le16_to_cpu(de->rec_len);
		}
		offset = 0;
		brelse (bh);
	}
out:
	return ret;
}

#ifdef CONFIG_EXT4_INDEX
/*
 * These functions convert from the major/minor hash to an f_pos
 * value.
 *
 * Currently we only use major hash numer.  This is unfortunate, but
 * on 32-bit machines, the same VFS interface is used for lseek and
 * llseek, so if we use the 64 bit offset, then the 32-bit versions of
 * lseek/telldir/seekdir will blow out spectacularly, and from within
 * the ext2 low-level routine, we don't know if we're being called by
 * a 64-bit version of the system call or the 32-bit version of the
 * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
 * cookie.  Sigh.
 */
#define hash2pos(major, minor)	(major >> 1)
#define pos2maj_hash(pos)	((pos << 1) & 0xffffffff)
#define pos2min_hash(pos)	(0)

/*
 * This structure holds the nodes of the red-black tree used to store
 * the directory entry in hash order.
 */
struct fname {
	__u32		hash;
	__u32		minor_hash;
	struct rb_node	rb_hash;
	struct fname	*next;
	__u32		inode;
	__u8		name_len;
	__u8		file_type;
	char		name[0];
};

/*
 * This functoin implements a non-recursive way of freeing all of the
 * nodes in the red-black tree.
 */
static void free_rb_tree_fname(struct rb_root *root)
{
	struct rb_node	*n = root->rb_node;
	struct rb_node	*parent;
	struct fname	*fname;

	while (n) {
		/* Do the node's children first */
		if ((n)->rb_left) {
			n = n->rb_left;
			continue;
		}
		if (n->rb_right) {
			n = n->rb_right;
			continue;
		}
		/*
		 * The node has no children; free it, and then zero
		 * out parent's link to it.  Finally go to the
		 * beginning of the loop and try to free the parent
		 * node.
		 */
		parent = rb_parent(n);
		fname = rb_entry(n, struct fname, rb_hash);
		while (fname) {
			struct fname * old = fname;
			fname = fname->next;
			kfree (old);
		}
		if (!parent)
			root->rb_node = NULL;
		else if (parent->rb_left == n)
			parent->rb_left = NULL;
		else if (parent->rb_right == n)
			parent->rb_right = NULL;
		n = parent;
	}
	root->rb_node = NULL;
}


static struct dir_private_info *create_dir_info(loff_t pos)
{
	struct dir_private_info *p;

	p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
	if (!p)
		return NULL;
	p->root.rb_node = NULL;
	p->curr_node = NULL;
	p->extra_fname = NULL;
	p->last_pos = 0;
	p->curr_hash = pos2maj_hash(pos);
	p->curr_minor_hash = pos2min_hash(pos);
	p->next_hash = 0;
	return p;
}

void ext4_htree_free_dir_info(struct dir_private_info *p)
{
	free_rb_tree_fname(&p->root);
	kfree(p);
}

/*
 * Given a directory entry, enter it into the fname rb tree.
 */
int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
			     __u32 minor_hash,
			     struct ext4_dir_entry_2 *dirent)
{
	struct rb_node **p, *parent = NULL;
	struct fname * fname, *new_fn;
	struct dir_private_info *info;
	int len;

	info = (struct dir_private_info *) dir_file->private_data;
	p = &info->root.rb_node;

	/* Create and allocate the fname structure */
	len = sizeof(struct fname) + dirent->name_len + 1;
	new_fn = kzalloc(len, GFP_KERNEL);
	if (!new_fn)
		return -ENOMEM;
	new_fn->hash = hash;
	new_fn->minor_hash = minor_hash;
	new_fn->inode = le32_to_cpu(dirent->inode);
	new_fn->name_len = dirent->name_len;
	new_fn->file_type = dirent->file_type;
	memcpy(new_fn->name, dirent->name, dirent->name_len);
	new_fn->name[dirent->name_len] = 0;

	while (*p) {
		parent = *p;
		fname = rb_entry(parent, struct fname, rb_hash);

		/*
		 * If the hash and minor hash match up, then we put
		 * them on a linked list.  This rarely happens...
		 */
		if ((new_fn->hash == fname->hash) &&
		    (new_fn->minor_hash == fname->minor_hash)) {
			new_fn->next = fname->next;
			fname->next = new_fn;
			return 0;
		}

		if (new_fn->hash < fname->hash)
			p = &(*p)->rb_left;
		else if (new_fn->hash > fname->hash)
			p = &(*p)->rb_right;
		else if (new_fn->minor_hash < fname->minor_hash)
			p = &(*p)->rb_left;
		else /* if (new_fn->minor_hash > fname->minor_hash) */
			p = &(*p)->rb_right;
	}

	rb_link_node(&new_fn->rb_hash, parent, p);
	rb_insert_color(&new_fn->rb_hash, &info->root);
	return 0;
}


/*
 * This is a helper function for ext4_dx_readdir.  It calls filldir
 * for all entres on the fname linked list.  (Normally there is only
 * one entry on the linked list, unless there are 62 bit hash collisions.)
 */
static int call_filldir(struct file * filp, void * dirent,
			filldir_t filldir, struct fname *fname)
{
	struct dir_private_info *info = filp->private_data;
	loff_t	curr_pos;
	struct inode *inode = filp->f_dentry->d_inode;
	struct super_block * sb;
	int error;

	sb = inode->i_sb;

	if (!fname) {
		printk("call_filldir: called with null fname?!?\n");
		return 0;
	}
	curr_pos = hash2pos(fname->hash, fname->minor_hash);
	while (fname) {
		error = filldir(dirent, fname->name,
				fname->name_len, curr_pos,
				fname->inode,
				get_dtype(sb, fname->file_type));
		if (error) {
			filp->f_pos = curr_pos;
			info->extra_fname = fname->next;
			return error;
		}
		fname = fname->next;
	}
	return 0;
}

static int ext4_dx_readdir(struct file * filp,
			 void * dirent, filldir_t filldir)
{
	struct dir_private_info *info = filp->private_data;
	struct inode *inode = filp->f_dentry->d_inode;
	struct fname *fname;
	int	ret;

	if (!info) {
		info = create_dir_info(filp->f_pos);
		if (!info)
			return -ENOMEM;
		filp->private_data = info;
	}

	if (filp->f_pos == EXT4_HTREE_EOF)
		return 0;	/* EOF */

	/* Some one has messed with f_pos; reset the world */
	if (info->last_pos != filp->f_pos) {
		free_rb_tree_fname(&info->root);
		info->curr_node = NULL;
		info->extra_fname = NULL;
		info->curr_hash = pos2maj_hash(filp->f_pos);
		info->curr_minor_hash = pos2min_hash(filp->f_pos);
	}

	/*
	 * If there are any leftover names on the hash collision
	 * chain, return them first.
	 */
	if (info->extra_fname &&
	    call_filldir(filp, dirent, filldir, info->extra_fname))
		goto finished;

	if (!info->curr_node)
		info->curr_node = rb_first(&info->root);

	while (1) {
		/*
		 * Fill the rbtree if we have no more entries,
		 * or the inode has changed since we last read in the
		 * cached entries.
		 */
		if ((!info->curr_node) ||
		    (filp->f_version != inode->i_version)) {
			info->curr_node = NULL;
			free_rb_tree_fname(&info->root);
			filp->f_version = inode->i_version;
			ret = ext4_htree_fill_tree(filp, info->curr_hash,
						   info->curr_minor_hash,
						   &info->next_hash);
			if (ret < 0)
				return ret;
			if (ret == 0) {
				filp->f_pos = EXT4_HTREE_EOF;
				break;
			}
			info->curr_node = rb_first(&info->root);
		}

		fname = rb_entry(info->curr_node, struct fname, rb_hash);
		info->curr_hash = fname->hash;
		info->curr_minor_hash = fname->minor_hash;
		if (call_filldir(filp, dirent, filldir, fname))
			break;

		info->curr_node = rb_next(info->curr_node);
		if (!info->curr_node) {
			if (info->next_hash == ~0) {
				filp->f_pos = EXT4_HTREE_EOF;
				break;
			}
			info->curr_hash = info->next_hash;
			info->curr_minor_hash = 0;
		}
	}
finished:
	info->last_pos = filp->f_pos;
	return 0;
}

static int ext4_release_dir (struct inode * inode, struct file * filp)
{
       if (filp->private_data)
		ext4_htree_free_dir_info(filp->private_data);

	return 0;
}

#endif
Commit	Line	Data
ac27a0ec	1	/*
617ba13b	2	* linux/fs/ext4/dir.c
ac27a0ec DK	3	*
	4	* Copyright (C) 1992, 1993, 1994, 1995
	5	* Remy Card (card@masi.ibp.fr)
	6	* Laboratoire MASI - Institut Blaise Pascal
	7	* Universite Pierre et Marie Curie (Paris VI)
	8	*
	9	* from
	10	*
	11	* linux/fs/minix/dir.c
	12	*
	13	* Copyright (C) 1991, 1992 Linus Torvalds
	14	*
617ba13b	15	* ext4 directory handling functions
ac27a0ec DK	16	*
	17	* Big-endian to little-endian byte-swapping/bitmaps by
	18	* David S. Miller (davem@caip.rutgers.edu), 1995
	19	*
	20	* Hash Tree Directory indexing (c) 2001 Daniel Phillips
	21	*
	22	*/
	23
	24	#include <linux/fs.h>
dab291af	25	#include <linux/jbd2.h>
617ba13b	26	#include <linux/ext4_fs.h>
ac27a0ec DK	27	#include <linux/buffer_head.h>
	28	#include <linux/smp_lock.h>
	29	#include <linux/slab.h>
	30	#include <linux/rbtree.h>
	31
617ba13b	32	static unsigned char ext4_filetype_table[] = {
ac27a0ec DK	33	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
	34	};
	35
617ba13b MC	36	static int ext4_readdir(struct file , void , filldir_t);
617ba13b MC	37	static int ext4_dx_readdir(struct file * filp,
ac27a0ec	38	void * dirent, filldir_t filldir);
617ba13b	39	static int ext4_release_dir (struct inode * inode,
ac27a0ec DK	40	struct file * filp);
ac27a0ec DK	41
617ba13b	42	const struct file_operations ext4_dir_operations = {
ac27a0ec DK	43	.llseek = generic_file_llseek,
ac27a0ec DK	44	.read = generic_read_dir,
617ba13b MC	45	.readdir = ext4_readdir, /* we take BKL. needed?*/
617ba13b MC	46	.ioctl = ext4_ioctl, /* BKL held */
ac27a0ec	47	#ifdef CONFIG_COMPAT
617ba13b	48	.compat_ioctl = ext4_compat_ioctl,
ac27a0ec	49	#endif
617ba13b MC	50	.fsync = ext4_sync_file, /* BKL held */
	51	#ifdef CONFIG_EXT4_INDEX
	52	.release = ext4_release_dir,
ac27a0ec DK	53	#endif
	54	};
	55
	56
	57	static unsigned char get_dtype(struct super_block *sb, int filetype)
	58	{
617ba13b MC	59	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) \|\|
617ba13b MC	60	(filetype >= EXT4_FT_MAX))
ac27a0ec DK	61	return DT_UNKNOWN;
ac27a0ec DK	62
617ba13b	63	return (ext4_filetype_table[filetype]);
ac27a0ec DK	64	}
	65
	66
617ba13b MC	67	int ext4_check_dir_entry (const char * function, struct inode * dir,
617ba13b MC	68	struct ext4_dir_entry_2 * de,
ac27a0ec DK	69	struct buffer_head * bh,
	70	unsigned long offset)
	71	{
	72	const char * error_msg = NULL;
	73	const int rlen = le16_to_cpu(de->rec_len);
	74
617ba13b	75	if (rlen < EXT4_DIR_REC_LEN(1))
ac27a0ec DK	76	error_msg = "rec_len is smaller than minimal";
	77	else if (rlen % 4 != 0)
	78	error_msg = "rec_len % 4 != 0";
617ba13b	79	else if (rlen < EXT4_DIR_REC_LEN(de->name_len))
ac27a0ec DK	80	error_msg = "rec_len is too small for name_len";
	81	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
	82	error_msg = "directory entry across blocks";
	83	else if (le32_to_cpu(de->inode) >
617ba13b	84	le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))
ac27a0ec DK	85	error_msg = "inode out of bounds";
	86
	87	if (error_msg != NULL)
617ba13b	88	ext4_error (dir->i_sb, function,
ac27a0ec DK	89	"bad entry in directory #%lu: %s - "
	90	"offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
	91	dir->i_ino, error_msg, offset,
	92	(unsigned long) le32_to_cpu(de->inode),
	93	rlen, de->name_len);
	94	return error_msg == NULL ? 1 : 0;
	95	}
	96
617ba13b	97	static int ext4_readdir(struct file * filp,
ac27a0ec DK	98	void * dirent, filldir_t filldir)
	99	{
	100	int error = 0;
	101	unsigned long offset;
	102	int i, stored;
617ba13b	103	struct ext4_dir_entry_2 *de;
ac27a0ec DK	104	struct super_block *sb;
	105	int err;
	106	struct inode *inode = filp->f_dentry->d_inode;
	107	int ret = 0;
	108
	109	sb = inode->i_sb;
	110
617ba13b MC	111	#ifdef CONFIG_EXT4_INDEX
	112	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
	113	EXT4_FEATURE_COMPAT_DIR_INDEX) &&
	114	((EXT4_I(inode)->i_flags & EXT4_INDEX_FL) \|\|
ac27a0ec	115	((inode->i_size >> sb->s_blocksize_bits) == 1))) {
617ba13b	116	err = ext4_dx_readdir(filp, dirent, filldir);
ac27a0ec DK	117	if (err != ERR_BAD_DX_DIR) {
	118	ret = err;
	119	goto out;
	120	}
	121	/*
	122	* We don't set the inode dirty flag since it's not
	123	* critical that it get flushed back to the disk.
	124	*/
617ba13b	125	EXT4_I(filp->f_dentry->d_inode)->i_flags &= ~EXT4_INDEX_FL;
ac27a0ec DK	126	}
	127	#endif
	128	stored = 0;
	129	offset = filp->f_pos & (sb->s_blocksize - 1);
	130
	131	while (!error && !stored && filp->f_pos < inode->i_size) {
617ba13b	132	unsigned long blk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec DK	133	struct buffer_head map_bh;
	134	struct buffer_head *bh = NULL;
	135
	136	map_bh.b_state = 0;
a86c6181	137	err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh, 0, 0);
ac27a0ec DK	138	if (err > 0) {
	139	page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
	140	&filp->f_ra,
	141	filp,
	142	map_bh.b_blocknr >>
	143	(PAGE_CACHE_SHIFT - inode->i_blkbits),
	144	1);
617ba13b	145	bh = ext4_bread(NULL, inode, blk, 0, &err);
ac27a0ec DK	146	}
	147
	148	/*
	149	* We ignore I/O errors on directories so users have a chance
	150	* of recovering data when there's a bad sector
	151	*/
	152	if (!bh) {
617ba13b	153	ext4_error (sb, "ext4_readdir",
ac27a0ec DK	154	"directory #%lu contains a hole at offset %lu",
	155	inode->i_ino, (unsigned long)filp->f_pos);
	156	filp->f_pos += sb->s_blocksize - offset;
	157	continue;
	158	}
	159
	160	revalidate:
	161	/* If the dir block has changed since the last call to
	162	* readdir(2), then we might be pointing to an invalid
	163	* dirent right now. Scan from the start of the block
	164	* to make sure. */
	165	if (filp->f_version != inode->i_version) {
	166	for (i = 0; i < sb->s_blocksize && i < offset; ) {
617ba13b	167	de = (struct ext4_dir_entry_2 *)
ac27a0ec DK	168	(bh->b_data + i);
	169	/* It's too expensive to do a full
	170	* dirent test each time round this
	171	* loop, but we do have to test at
	172	* least that it is non-zero. A
	173	* failure will be detected in the
	174	* dirent test below. */
	175	if (le16_to_cpu(de->rec_len) <
617ba13b	176	EXT4_DIR_REC_LEN(1))
ac27a0ec DK	177	break;
	178	i += le16_to_cpu(de->rec_len);
	179	}
	180	offset = i;
	181	filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
	182	\| offset;
	183	filp->f_version = inode->i_version;
	184	}
	185
	186	while (!error && filp->f_pos < inode->i_size
	187	&& offset < sb->s_blocksize) {
617ba13b MC	188	de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
617ba13b MC	189	if (!ext4_check_dir_entry ("ext4_readdir", inode, de,
ac27a0ec DK	190	bh, offset)) {
	191	/* On error, skip the f_pos to the
	192	next block. */
	193	filp->f_pos = (filp->f_pos \|
	194	(sb->s_blocksize - 1)) + 1;
	195	brelse (bh);
	196	ret = stored;
	197	goto out;
	198	}
	199	offset += le16_to_cpu(de->rec_len);
	200	if (le32_to_cpu(de->inode)) {
	201	/* We might block in the next section
	202	* if the data destination is
	203	* currently swapped out. So, use a
	204	* version stamp to detect whether or
	205	* not the directory has been modified
	206	* during the copy operation.
	207	*/
	208	unsigned long version = filp->f_version;
	209
	210	error = filldir(dirent, de->name,
	211	de->name_len,
	212	filp->f_pos,
	213	le32_to_cpu(de->inode),
	214	get_dtype(sb, de->file_type));
	215	if (error)
	216	break;
	217	if (version != filp->f_version)
	218	goto revalidate;
	219	stored ++;
	220	}
	221	filp->f_pos += le16_to_cpu(de->rec_len);
	222	}
	223	offset = 0;
	224	brelse (bh);
	225	}
	226	out:
	227	return ret;
	228	}
	229
617ba13b	230	#ifdef CONFIG_EXT4_INDEX
ac27a0ec DK	231	/*
	232	* These functions convert from the major/minor hash to an f_pos
	233	* value.
	234	*
	235	* Currently we only use major hash numer. This is unfortunate, but
	236	* on 32-bit machines, the same VFS interface is used for lseek and
	237	* llseek, so if we use the 64 bit offset, then the 32-bit versions of
	238	* lseek/telldir/seekdir will blow out spectacularly, and from within
	239	* the ext2 low-level routine, we don't know if we're being called by
	240	* a 64-bit version of the system call or the 32-bit version of the
	241	* system call. Worse yet, NFSv2 only allows for a 32-bit readdir
	242	* cookie. Sigh.
	243	*/
	244	#define hash2pos(major, minor) (major >> 1)
	245	#define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
	246	#define pos2min_hash(pos) (0)
	247
	248	/*
	249	* This structure holds the nodes of the red-black tree used to store
	250	* the directory entry in hash order.
	251	*/
	252	struct fname {
	253	__u32 hash;
	254	__u32 minor_hash;
	255	struct rb_node rb_hash;
	256	struct fname *next;
	257	__u32 inode;
	258	__u8 name_len;
	259	__u8 file_type;
	260	char name[0];
	261	};
	262
	263	/*
	264	* This functoin implements a non-recursive way of freeing all of the
	265	* nodes in the red-black tree.
	266	*/
	267	static void free_rb_tree_fname(struct rb_root *root)
	268	{
	269	struct rb_node *n = root->rb_node;
	270	struct rb_node *parent;
	271	struct fname *fname;
	272
	273	while (n) {
	274	/* Do the node's children first */
	275	if ((n)->rb_left) {
	276	n = n->rb_left;
	277	continue;
	278	}
	279	if (n->rb_right) {
	280	n = n->rb_right;
	281	continue;
	282	}
	283	/*
	284	* The node has no children; free it, and then zero
	285	* out parent's link to it. Finally go to the
	286	* beginning of the loop and try to free the parent
	287	* node.
	288	*/
	289	parent = rb_parent(n);
	290	fname = rb_entry(n, struct fname, rb_hash);
	291	while (fname) {
	292	struct fname * old = fname;
	293	fname = fname->next;
	294	kfree (old);
295	}
296	if (!parent)
297	root->rb_node = NULL;
298	else if (parent->rb_left == n)
299	parent->rb_left = NULL;
300	else if (parent->rb_right == n)
301	parent->rb_right = NULL;
302	n = parent;
303	}
304	root->rb_node = NULL;
305	}
306
307
308	static struct dir_private_info *create_dir_info(loff_t pos)
309	{
310	struct dir_private_info *p;
311
312	p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
313	if (!p)
314	return NULL;
315	p->root.rb_node = NULL;
316	p->curr_node = NULL;
317	p->extra_fname = NULL;
318	p->last_pos = 0;
319	p->curr_hash = pos2maj_hash(pos);
320	p->curr_minor_hash = pos2min_hash(pos);
321	p->next_hash = 0;
322	return p;
323	}
324
617ba13b	325	void ext4_htree_free_dir_info(struct dir_private_info *p)
ac27a0ec DK	326	{
	327	free_rb_tree_fname(&p->root);
	328	kfree(p);
	329	}
	330
	331	/*
	332	* Given a directory entry, enter it into the fname rb tree.
	333	*/
617ba13b	334	int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
ac27a0ec	335	__u32 minor_hash,
617ba13b	336	struct ext4_dir_entry_2 *dirent)
ac27a0ec DK	337	{
	338	struct rb_node *p, parent = NULL;
	339	struct fname * fname, *new_fn;
	340	struct dir_private_info *info;
	341	int len;
	342
	343	info = (struct dir_private_info *) dir_file->private_data;
	344	p = &info->root.rb_node;
	345
	346	/* Create and allocate the fname structure */
	347	len = sizeof(struct fname) + dirent->name_len + 1;
	348	new_fn = kzalloc(len, GFP_KERNEL);
	349	if (!new_fn)
	350	return -ENOMEM;
	351	new_fn->hash = hash;
	352	new_fn->minor_hash = minor_hash;
	353	new_fn->inode = le32_to_cpu(dirent->inode);
	354	new_fn->name_len = dirent->name_len;
	355	new_fn->file_type = dirent->file_type;
	356	memcpy(new_fn->name, dirent->name, dirent->name_len);
	357	new_fn->name[dirent->name_len] = 0;
	358
	359	while (*p) {
	360	parent = *p;
	361	fname = rb_entry(parent, struct fname, rb_hash);
	362
	363	/*
	364	* If the hash and minor hash match up, then we put
	365	* them on a linked list. This rarely happens...
	366	*/
	367	if ((new_fn->hash == fname->hash) &&
	368	(new_fn->minor_hash == fname->minor_hash)) {
	369	new_fn->next = fname->next;
	370	fname->next = new_fn;
	371	return 0;
	372	}
	373
	374	if (new_fn->hash < fname->hash)
	375	p = &(*p)->rb_left;
	376	else if (new_fn->hash > fname->hash)
	377	p = &(*p)->rb_right;
	378	else if (new_fn->minor_hash < fname->minor_hash)
	379	p = &(*p)->rb_left;
	380	else /* if (new_fn->minor_hash > fname->minor_hash) */
	381	p = &(*p)->rb_right;
	382	}
	383
	384	rb_link_node(&new_fn->rb_hash, parent, p);
	385	rb_insert_color(&new_fn->rb_hash, &info->root);
	386	return 0;
	387	}
	388
	389
	390
	391	/*
617ba13b	392	* This is a helper function for ext4_dx_readdir. It calls filldir
ac27a0ec DK	393	* for all entres on the fname linked list. (Normally there is only
	394	* one entry on the linked list, unless there are 62 bit hash collisions.)
	395	*/
	396	static int call_filldir(struct file * filp, void * dirent,
	397	filldir_t filldir, struct fname *fname)
	398	{
	399	struct dir_private_info *info = filp->private_data;
	400	loff_t curr_pos;
	401	struct inode *inode = filp->f_dentry->d_inode;
	402	struct super_block * sb;
	403	int error;
	404
	405	sb = inode->i_sb;
	406
	407	if (!fname) {
	408	printk("call_filldir: called with null fname?!?\n");
	409	return 0;
	410	}
	411	curr_pos = hash2pos(fname->hash, fname->minor_hash);
	412	while (fname) {
	413	error = filldir(dirent, fname->name,
	414	fname->name_len, curr_pos,
	415	fname->inode,
	416	get_dtype(sb, fname->file_type));
	417	if (error) {
	418	filp->f_pos = curr_pos;
	419	info->extra_fname = fname->next;
	420	return error;
	421	}
	422	fname = fname->next;
	423	}
	424	return 0;
	425	}
	426
617ba13b	427	static int ext4_dx_readdir(struct file * filp,
ac27a0ec DK	428	void * dirent, filldir_t filldir)
	429	{
	430	struct dir_private_info *info = filp->private_data;
	431	struct inode *inode = filp->f_dentry->d_inode;
	432	struct fname *fname;
	433	int ret;
	434
	435	if (!info) {
	436	info = create_dir_info(filp->f_pos);
	437	if (!info)
	438	return -ENOMEM;
	439	filp->private_data = info;
	440	}
	441
617ba13b	442	if (filp->f_pos == EXT4_HTREE_EOF)
ac27a0ec DK	443	return 0; /* EOF */
	444
	445	/* Some one has messed with f_pos; reset the world */
	446	if (info->last_pos != filp->f_pos) {
	447	free_rb_tree_fname(&info->root);
	448	info->curr_node = NULL;
	449	info->extra_fname = NULL;
	450	info->curr_hash = pos2maj_hash(filp->f_pos);
	451	info->curr_minor_hash = pos2min_hash(filp->f_pos);
	452	}
	453
	454	/*
	455	* If there are any leftover names on the hash collision
	456	* chain, return them first.
	457	*/
	458	if (info->extra_fname &&
	459	call_filldir(filp, dirent, filldir, info->extra_fname))
	460	goto finished;
	461
	462	if (!info->curr_node)
	463	info->curr_node = rb_first(&info->root);
	464
	465	while (1) {
	466	/*
	467	* Fill the rbtree if we have no more entries,
	468	* or the inode has changed since we last read in the
	469	* cached entries.
	470	*/
	471	if ((!info->curr_node) \|\|
	472	(filp->f_version != inode->i_version)) {
	473	info->curr_node = NULL;
	474	free_rb_tree_fname(&info->root);
	475	filp->f_version = inode->i_version;
617ba13b	476	ret = ext4_htree_fill_tree(filp, info->curr_hash,
ac27a0ec DK	477	info->curr_minor_hash,
	478	&info->next_hash);
	479	if (ret < 0)
	480	return ret;
	481	if (ret == 0) {
617ba13b	482	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	483	break;
	484	}
	485	info->curr_node = rb_first(&info->root);
	486	}
	487
	488	fname = rb_entry(info->curr_node, struct fname, rb_hash);
	489	info->curr_hash = fname->hash;
	490	info->curr_minor_hash = fname->minor_hash;
	491	if (call_filldir(filp, dirent, filldir, fname))
	492	break;
	493
	494	info->curr_node = rb_next(info->curr_node);
	495	if (!info->curr_node) {
	496	if (info->next_hash == ~0) {
617ba13b	497	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	498	break;
	499	}
	500	info->curr_hash = info->next_hash;
	501	info->curr_minor_hash = 0;
	502	}
	503	}
	504	finished:
	505	info->last_pos = filp->f_pos;
	506	return 0;
	507	}
	508
617ba13b	509	static int ext4_release_dir (struct inode * inode, struct file * filp)
ac27a0ec DK	510	{
ac27a0ec DK	511	if (filp->private_data)
617ba13b	512	ext4_htree_free_dir_info(filp->private_data);
ac27a0ec DK	513
	514	return 0;
	515	}
	516
	517	#endif