[linux-2.6-block.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/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_path.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_path.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) {
			pgoff_t index = map_bh.b_blocknr >>
					(PAGE_CACHE_SHIFT - inode->i_blkbits);
			if (!ra_has_index(&filp->f_ra, index))
				page_cache_sync_readahead(
					sb->s_bdev->bd_inode->i_mapping,
					&filp->f_ra, filp,
					index, 1);
			filp->f_ra.prev_index = index;
			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);
			/* corrupt size?  Maybe no more blocks to read */
			if (filp->f_pos > inode->i_blocks << 9)
				break;
			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_path.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_path.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	27	#include <linux/buffer_head.h>
ac27a0ec DK	28	#include <linux/slab.h>
	29	#include <linux/rbtree.h>
	30
617ba13b	31	static unsigned char ext4_filetype_table[] = {
ac27a0ec DK	32	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
	33	};
	34
617ba13b MC	35	static int ext4_readdir(struct file , void , filldir_t);
617ba13b MC	36	static int ext4_dx_readdir(struct file * filp,
ac27a0ec	37	void * dirent, filldir_t filldir);
617ba13b	38	static int ext4_release_dir (struct inode * inode,
ac27a0ec DK	39	struct file * filp);
ac27a0ec DK	40
617ba13b	41	const struct file_operations ext4_dir_operations = {
ac27a0ec DK	42	.llseek = generic_file_llseek,
ac27a0ec DK	43	.read = generic_read_dir,
617ba13b MC	44	.readdir = ext4_readdir, /* we take BKL. needed?*/
617ba13b MC	45	.ioctl = ext4_ioctl, /* BKL held */
ac27a0ec	46	#ifdef CONFIG_COMPAT
617ba13b	47	.compat_ioctl = ext4_compat_ioctl,
ac27a0ec	48	#endif
617ba13b MC	49	.fsync = ext4_sync_file, /* BKL held */
	50	#ifdef CONFIG_EXT4_INDEX
	51	.release = ext4_release_dir,
ac27a0ec DK	52	#endif
	53	};
	54
	55
	56	static unsigned char get_dtype(struct super_block *sb, int filetype)
	57	{
617ba13b MC	58	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) \|\|
617ba13b MC	59	(filetype >= EXT4_FT_MAX))
ac27a0ec DK	60	return DT_UNKNOWN;
ac27a0ec DK	61
617ba13b	62	return (ext4_filetype_table[filetype]);
ac27a0ec DK	63	}
	64
	65
617ba13b MC	66	int ext4_check_dir_entry (const char * function, struct inode * dir,
617ba13b MC	67	struct ext4_dir_entry_2 * de,
ac27a0ec DK	68	struct buffer_head * bh,
	69	unsigned long offset)
	70	{
	71	const char * error_msg = NULL;
	72	const int rlen = le16_to_cpu(de->rec_len);
	73
617ba13b	74	if (rlen < EXT4_DIR_REC_LEN(1))
ac27a0ec DK	75	error_msg = "rec_len is smaller than minimal";
	76	else if (rlen % 4 != 0)
	77	error_msg = "rec_len % 4 != 0";
617ba13b	78	else if (rlen < EXT4_DIR_REC_LEN(de->name_len))
ac27a0ec DK	79	error_msg = "rec_len is too small for name_len";
	80	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
	81	error_msg = "directory entry across blocks";
	82	else if (le32_to_cpu(de->inode) >
617ba13b	83	le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))
ac27a0ec DK	84	error_msg = "inode out of bounds";
	85
	86	if (error_msg != NULL)
617ba13b	87	ext4_error (dir->i_sb, function,
ac27a0ec DK	88	"bad entry in directory #%lu: %s - "
	89	"offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
	90	dir->i_ino, error_msg, offset,
	91	(unsigned long) le32_to_cpu(de->inode),
	92	rlen, de->name_len);
	93	return error_msg == NULL ? 1 : 0;
	94	}
	95
617ba13b	96	static int ext4_readdir(struct file * filp,
ac27a0ec DK	97	void * dirent, filldir_t filldir)
	98	{
	99	int error = 0;
	100	unsigned long offset;
	101	int i, stored;
617ba13b	102	struct ext4_dir_entry_2 *de;
ac27a0ec DK	103	struct super_block *sb;
ac27a0ec DK	104	int err;
9d549890	105	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	106	int ret = 0;
	107
	108	sb = inode->i_sb;
	109
617ba13b MC	110	#ifdef CONFIG_EXT4_INDEX
	111	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
	112	EXT4_FEATURE_COMPAT_DIR_INDEX) &&
	113	((EXT4_I(inode)->i_flags & EXT4_INDEX_FL) \|\|
ac27a0ec	114	((inode->i_size >> sb->s_blocksize_bits) == 1))) {
617ba13b	115	err = ext4_dx_readdir(filp, dirent, filldir);
ac27a0ec DK	116	if (err != ERR_BAD_DX_DIR) {
	117	ret = err;
	118	goto out;
	119	}
	120	/*
	121	* We don't set the inode dirty flag since it's not
	122	* critical that it get flushed back to the disk.
	123	*/
9d549890	124	EXT4_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT4_INDEX_FL;
ac27a0ec DK	125	}
	126	#endif
	127	stored = 0;
	128	offset = filp->f_pos & (sb->s_blocksize - 1);
	129
	130	while (!error && !stored && filp->f_pos < inode->i_size) {
617ba13b	131	unsigned long blk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec DK	132	struct buffer_head map_bh;
	133	struct buffer_head *bh = NULL;
	134
	135	map_bh.b_state = 0;
a86c6181	136	err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh, 0, 0);
ac27a0ec	137	if (err > 0) {
dc7868fc FW	138	pgoff_t index = map_bh.b_blocknr >>
	139	(PAGE_CACHE_SHIFT - inode->i_blkbits);
	140	if (!ra_has_index(&filp->f_ra, index))
cf914a7d	141	page_cache_sync_readahead(
dc7868fc FW	142	sb->s_bdev->bd_inode->i_mapping,
dc7868fc FW	143	&filp->f_ra, filp,
cf914a7d	144	index, 1);
dc7868fc	145	filp->f_ra.prev_index = index;
617ba13b	146	bh = ext4_bread(NULL, inode, blk, 0, &err);
ac27a0ec DK	147	}
	148
	149	/*
	150	* We ignore I/O errors on directories so users have a chance
	151	* of recovering data when there's a bad sector
	152	*/
	153	if (!bh) {
617ba13b	154	ext4_error (sb, "ext4_readdir",
ac27a0ec DK	155	"directory #%lu contains a hole at offset %lu",
ac27a0ec DK	156	inode->i_ino, (unsigned long)filp->f_pos);
e6c40211 ES	157	/* corrupt size? Maybe no more blocks to read */
	158	if (filp->f_pos > inode->i_blocks << 9)
	159	break;
ac27a0ec DK	160	filp->f_pos += sb->s_blocksize - offset;
	161	continue;
	162	}
	163
	164	revalidate:
	165	/* If the dir block has changed since the last call to
	166	* readdir(2), then we might be pointing to an invalid
	167	* dirent right now. Scan from the start of the block
	168	* to make sure. */
	169	if (filp->f_version != inode->i_version) {
	170	for (i = 0; i < sb->s_blocksize && i < offset; ) {
617ba13b	171	de = (struct ext4_dir_entry_2 *)
ac27a0ec DK	172	(bh->b_data + i);
	173	/* It's too expensive to do a full
	174	* dirent test each time round this
	175	* loop, but we do have to test at
	176	* least that it is non-zero. A
	177	* failure will be detected in the
	178	* dirent test below. */
	179	if (le16_to_cpu(de->rec_len) <
617ba13b	180	EXT4_DIR_REC_LEN(1))
ac27a0ec DK	181	break;
	182	i += le16_to_cpu(de->rec_len);
	183	}
	184	offset = i;
	185	filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
	186	\| offset;
	187	filp->f_version = inode->i_version;
	188	}
	189
	190	while (!error && filp->f_pos < inode->i_size
	191	&& offset < sb->s_blocksize) {
617ba13b MC	192	de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
617ba13b MC	193	if (!ext4_check_dir_entry ("ext4_readdir", inode, de,
ac27a0ec	194	bh, offset)) {
63f57933 AM	195	/*
	196	* On error, skip the f_pos to the next block
	197	*/
ac27a0ec DK	198	filp->f_pos = (filp->f_pos \|
	199	(sb->s_blocksize - 1)) + 1;
	200	brelse (bh);
	201	ret = stored;
	202	goto out;
	203	}
	204	offset += le16_to_cpu(de->rec_len);
	205	if (le32_to_cpu(de->inode)) {
	206	/* We might block in the next section
	207	* if the data destination is
	208	* currently swapped out. So, use a
	209	* version stamp to detect whether or
	210	* not the directory has been modified
	211	* during the copy operation.
	212	*/
	213	unsigned long version = filp->f_version;
	214
	215	error = filldir(dirent, de->name,
	216	de->name_len,
	217	filp->f_pos,
	218	le32_to_cpu(de->inode),
	219	get_dtype(sb, de->file_type));
	220	if (error)
	221	break;
	222	if (version != filp->f_version)
	223	goto revalidate;
	224	stored ++;
	225	}
	226	filp->f_pos += le16_to_cpu(de->rec_len);
	227	}
	228	offset = 0;
	229	brelse (bh);
	230	}
	231	out:
	232	return ret;
	233	}
	234
617ba13b	235	#ifdef CONFIG_EXT4_INDEX
ac27a0ec DK	236	/*
	237	* These functions convert from the major/minor hash to an f_pos
	238	* value.
	239	*
	240	* Currently we only use major hash numer. This is unfortunate, but
	241	* on 32-bit machines, the same VFS interface is used for lseek and
	242	* llseek, so if we use the 64 bit offset, then the 32-bit versions of
	243	* lseek/telldir/seekdir will blow out spectacularly, and from within
	244	* the ext2 low-level routine, we don't know if we're being called by
	245	* a 64-bit version of the system call or the 32-bit version of the
	246	* system call. Worse yet, NFSv2 only allows for a 32-bit readdir
	247	* cookie. Sigh.
	248	*/
	249	#define hash2pos(major, minor) (major >> 1)
	250	#define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
	251	#define pos2min_hash(pos) (0)
	252
	253	/*
	254	* This structure holds the nodes of the red-black tree used to store
	255	* the directory entry in hash order.
	256	*/
	257	struct fname {
	258	__u32 hash;
	259	__u32 minor_hash;
	260	struct rb_node rb_hash;
	261	struct fname *next;
	262	__u32 inode;
	263	__u8 name_len;
	264	__u8 file_type;
	265	char name[0];
	266	};
	267
	268	/*
	269	* This functoin implements a non-recursive way of freeing all of the
	270	* nodes in the red-black tree.
	271	*/
	272	static void free_rb_tree_fname(struct rb_root *root)
	273	{
	274	struct rb_node *n = root->rb_node;
	275	struct rb_node *parent;
	276	struct fname *fname;
	277
	278	while (n) {
	279	/* Do the node's children first */
	280	if ((n)->rb_left) {
	281	n = n->rb_left;
	282	continue;
	283	}
	284	if (n->rb_right) {
	285	n = n->rb_right;
	286	continue;
	287	}
	288	/*
	289	* The node has no children; free it, and then zero
	290	* out parent's link to it. Finally go to the
	291	* beginning of the loop and try to free the parent
	292	* node.
	293	*/
	294	parent = rb_parent(n);
	295	fname = rb_entry(n, struct fname, rb_hash);
	296	while (fname) {
	297	struct fname * old = fname;
	298	fname = fname->next;
	299	kfree (old);
300	}
301	if (!parent)
302	root->rb_node = NULL;
303	else if (parent->rb_left == n)
304	parent->rb_left = NULL;
305	else if (parent->rb_right == n)
306	parent->rb_right = NULL;
307	n = parent;
308	}
309	root->rb_node = NULL;
310	}
311
312
313	static struct dir_private_info *create_dir_info(loff_t pos)
314	{
315	struct dir_private_info *p;
316
317	p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
318	if (!p)
319	return NULL;
320	p->root.rb_node = NULL;
321	p->curr_node = NULL;
322	p->extra_fname = NULL;
323	p->last_pos = 0;
324	p->curr_hash = pos2maj_hash(pos);
325	p->curr_minor_hash = pos2min_hash(pos);
326	p->next_hash = 0;
327	return p;
328	}
329
617ba13b	330	void ext4_htree_free_dir_info(struct dir_private_info *p)
ac27a0ec DK	331	{
	332	free_rb_tree_fname(&p->root);
	333	kfree(p);
	334	}
	335
	336	/*
	337	* Given a directory entry, enter it into the fname rb tree.
	338	*/
617ba13b	339	int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
ac27a0ec	340	__u32 minor_hash,
617ba13b	341	struct ext4_dir_entry_2 *dirent)
ac27a0ec DK	342	{
	343	struct rb_node *p, parent = NULL;
	344	struct fname * fname, *new_fn;
	345	struct dir_private_info *info;
	346	int len;
	347
	348	info = (struct dir_private_info *) dir_file->private_data;
	349	p = &info->root.rb_node;
	350
	351	/* Create and allocate the fname structure */
	352	len = sizeof(struct fname) + dirent->name_len + 1;
	353	new_fn = kzalloc(len, GFP_KERNEL);
	354	if (!new_fn)
	355	return -ENOMEM;
	356	new_fn->hash = hash;
	357	new_fn->minor_hash = minor_hash;
	358	new_fn->inode = le32_to_cpu(dirent->inode);
	359	new_fn->name_len = dirent->name_len;
	360	new_fn->file_type = dirent->file_type;
	361	memcpy(new_fn->name, dirent->name, dirent->name_len);
	362	new_fn->name[dirent->name_len] = 0;
	363
	364	while (*p) {
	365	parent = *p;
	366	fname = rb_entry(parent, struct fname, rb_hash);
	367
	368	/*
	369	* If the hash and minor hash match up, then we put
	370	* them on a linked list. This rarely happens...
	371	*/
	372	if ((new_fn->hash == fname->hash) &&
	373	(new_fn->minor_hash == fname->minor_hash)) {
	374	new_fn->next = fname->next;
	375	fname->next = new_fn;
	376	return 0;
	377	}
	378
	379	if (new_fn->hash < fname->hash)
	380	p = &(*p)->rb_left;
	381	else if (new_fn->hash > fname->hash)
	382	p = &(*p)->rb_right;
	383	else if (new_fn->minor_hash < fname->minor_hash)
	384	p = &(*p)->rb_left;
	385	else /* if (new_fn->minor_hash > fname->minor_hash) */
	386	p = &(*p)->rb_right;
	387	}
	388
	389	rb_link_node(&new_fn->rb_hash, parent, p);
	390	rb_insert_color(&new_fn->rb_hash, &info->root);
	391	return 0;
	392	}
	393
	394
	395
	396	/*
617ba13b	397	* This is a helper function for ext4_dx_readdir. It calls filldir
ac27a0ec DK	398	* for all entres on the fname linked list. (Normally there is only
	399	* one entry on the linked list, unless there are 62 bit hash collisions.)
	400	*/
	401	static int call_filldir(struct file * filp, void * dirent,
	402	filldir_t filldir, struct fname *fname)
	403	{
	404	struct dir_private_info *info = filp->private_data;
	405	loff_t curr_pos;
9d549890	406	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	407	struct super_block * sb;
	408	int error;
	409
	410	sb = inode->i_sb;
	411
	412	if (!fname) {
	413	printk("call_filldir: called with null fname?!?\n");
	414	return 0;
	415	}
	416	curr_pos = hash2pos(fname->hash, fname->minor_hash);
	417	while (fname) {
	418	error = filldir(dirent, fname->name,
	419	fname->name_len, curr_pos,
	420	fname->inode,
	421	get_dtype(sb, fname->file_type));
	422	if (error) {
	423	filp->f_pos = curr_pos;
	424	info->extra_fname = fname->next;
	425	return error;
	426	}
	427	fname = fname->next;
	428	}
	429	return 0;
	430	}
	431
617ba13b	432	static int ext4_dx_readdir(struct file * filp,
ac27a0ec DK	433	void * dirent, filldir_t filldir)
	434	{
	435	struct dir_private_info *info = filp->private_data;
9d549890	436	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	437	struct fname *fname;
	438	int ret;
	439
	440	if (!info) {
	441	info = create_dir_info(filp->f_pos);
	442	if (!info)
	443	return -ENOMEM;
	444	filp->private_data = info;
	445	}
	446
617ba13b	447	if (filp->f_pos == EXT4_HTREE_EOF)
ac27a0ec DK	448	return 0; /* EOF */
	449
	450	/* Some one has messed with f_pos; reset the world */
	451	if (info->last_pos != filp->f_pos) {
	452	free_rb_tree_fname(&info->root);
	453	info->curr_node = NULL;
	454	info->extra_fname = NULL;
	455	info->curr_hash = pos2maj_hash(filp->f_pos);
	456	info->curr_minor_hash = pos2min_hash(filp->f_pos);
	457	}
	458
	459	/*
	460	* If there are any leftover names on the hash collision
	461	* chain, return them first.
	462	*/
	463	if (info->extra_fname &&
	464	call_filldir(filp, dirent, filldir, info->extra_fname))
	465	goto finished;
	466
	467	if (!info->curr_node)
	468	info->curr_node = rb_first(&info->root);
	469
	470	while (1) {
	471	/*
	472	* Fill the rbtree if we have no more entries,
	473	* or the inode has changed since we last read in the
	474	* cached entries.
	475	*/
	476	if ((!info->curr_node) \|\|
	477	(filp->f_version != inode->i_version)) {
	478	info->curr_node = NULL;
	479	free_rb_tree_fname(&info->root);
	480	filp->f_version = inode->i_version;
617ba13b	481	ret = ext4_htree_fill_tree(filp, info->curr_hash,
ac27a0ec DK	482	info->curr_minor_hash,
	483	&info->next_hash);
	484	if (ret < 0)
	485	return ret;
	486	if (ret == 0) {
617ba13b	487	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	488	break;
	489	}
	490	info->curr_node = rb_first(&info->root);
	491	}
	492
	493	fname = rb_entry(info->curr_node, struct fname, rb_hash);
	494	info->curr_hash = fname->hash;
	495	info->curr_minor_hash = fname->minor_hash;
	496	if (call_filldir(filp, dirent, filldir, fname))
	497	break;
	498
	499	info->curr_node = rb_next(info->curr_node);
	500	if (!info->curr_node) {
	501	if (info->next_hash == ~0) {
617ba13b	502	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	503	break;
	504	}
	505	info->curr_hash = info->next_hash;
	506	info->curr_minor_hash = 0;
	507	}
	508	}
	509	finished:
	510	info->last_pos = filp->f_pos;
	511	return 0;
	512	}
	513
617ba13b	514	static int ext4_release_dir (struct inode * inode, struct file * filp)
ac27a0ec	515	{
63f57933	516	if (filp->private_data)
617ba13b	517	ext4_htree_free_dir_info(filp->private_data);
ac27a0ec DK	518
	519	return 0;
	520	}
	521
	522	#endif