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

/*
 *  linux/fs/ext4/dir.c
 *
 * Copyright (C) 1992, 1993, 1994, 1995
 * Remy Card ([email protected])
 * 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 ([email protected]), 1995
 *
 * Hash Tree Directory indexing (c) 2001  Daniel Phillips
 *
 */

#include <linux/fs.h>
#include <linux/jbd2.h>
#include <linux/buffer_head.h>
#include <linux/slab.h>
#include <linux/rbtree.h>
#include "ext4.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?*/
	.unlocked_ioctl = ext4_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl	= ext4_compat_ioctl,
#endif
	.fsync		= ext4_sync_file,
	.release	= ext4_release_dir,
};


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 = ext4_rec_len_from_disk(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;

	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;
	}
	stored = 0;
	offset = filp->f_pos & (sb->s_blocksize - 1);

	while (!error && !stored && filp->f_pos < inode->i_size) {
		ext4_lblk_t 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, 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_pos = (loff_t)index << PAGE_CACHE_SHIFT;
			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 (ext4_rec_len_from_disk(de->rec_len)
						< EXT4_DIR_REC_LEN(1))
					break;
				i += ext4_rec_len_from_disk(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 += ext4_rec_len_from_disk(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.
				 */
				u64 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 += ext4_rec_len_from_disk(de->rec_len);
		}
		offset = 0;
		brelse (bh);
	}
out:
	return ret;
}

/*
 * 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;
	}
}


static struct dir_private_info *ext4_htree_create_dir_info(loff_t pos)
{
	struct dir_private_info *p;

	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
	if (!p)
		return NULL;
	p->curr_hash = pos2maj_hash(pos);
	p->curr_minor_hash = pos2min_hash(pos);
	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 = ext4_htree_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;
}
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 ([email protected])
	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 ([email protected]), 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>
ac27a0ec	26	#include <linux/buffer_head.h>
ac27a0ec DK	27	#include <linux/slab.h>
ac27a0ec DK	28	#include <linux/rbtree.h>
3dcf5451	29	#include "ext4.h"
ac27a0ec	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	44	.readdir = ext4_readdir, /* we take BKL. needed?*/
5cdd7b2d	45	.unlocked_ioctl = ext4_ioctl,
ac27a0ec	46	#ifdef CONFIG_COMPAT
617ba13b	47	.compat_ioctl = ext4_compat_ioctl,
ac27a0ec	48	#endif
642be6ec	49	.fsync = ext4_sync_file,
617ba13b	50	.release = ext4_release_dir,
ac27a0ec DK	51	};
	52
	53
	54	static unsigned char get_dtype(struct super_block *sb, int filetype)
	55	{
617ba13b MC	56	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) \|\|
617ba13b MC	57	(filetype >= EXT4_FT_MAX))
ac27a0ec DK	58	return DT_UNKNOWN;
ac27a0ec DK	59
617ba13b	60	return (ext4_filetype_table[filetype]);
ac27a0ec DK	61	}
	62
	63
617ba13b MC	64	int ext4_check_dir_entry (const char * function, struct inode * dir,
617ba13b MC	65	struct ext4_dir_entry_2 * de,
ac27a0ec DK	66	struct buffer_head * bh,
	67	unsigned long offset)
	68	{
	69	const char * error_msg = NULL;
a72d7f83	70	const int rlen = ext4_rec_len_from_disk(de->rec_len);
ac27a0ec	71
617ba13b	72	if (rlen < EXT4_DIR_REC_LEN(1))
ac27a0ec DK	73	error_msg = "rec_len is smaller than minimal";
	74	else if (rlen % 4 != 0)
	75	error_msg = "rec_len % 4 != 0";
617ba13b	76	else if (rlen < EXT4_DIR_REC_LEN(de->name_len))
ac27a0ec DK	77	error_msg = "rec_len is too small for name_len";
	78	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
	79	error_msg = "directory entry across blocks";
	80	else if (le32_to_cpu(de->inode) >
617ba13b	81	le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))
ac27a0ec DK	82	error_msg = "inode out of bounds";
	83
	84	if (error_msg != NULL)
617ba13b	85	ext4_error (dir->i_sb, function,
ac27a0ec DK	86	"bad entry in directory #%lu: %s - "
	87	"offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
	88	dir->i_ino, error_msg, offset,
	89	(unsigned long) le32_to_cpu(de->inode),
	90	rlen, de->name_len);
	91	return error_msg == NULL ? 1 : 0;
	92	}
	93
617ba13b	94	static int ext4_readdir(struct file * filp,
ac27a0ec DK	95	void * dirent, filldir_t filldir)
	96	{
	97	int error = 0;
	98	unsigned long offset;
	99	int i, stored;
617ba13b	100	struct ext4_dir_entry_2 *de;
ac27a0ec DK	101	struct super_block *sb;
ac27a0ec DK	102	int err;
9d549890	103	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	104	int ret = 0;
	105
	106	sb = inode->i_sb;
	107
617ba13b MC	108	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
	109	EXT4_FEATURE_COMPAT_DIR_INDEX) &&
	110	((EXT4_I(inode)->i_flags & EXT4_INDEX_FL) \|\|
ac27a0ec	111	((inode->i_size >> sb->s_blocksize_bits) == 1))) {
617ba13b	112	err = ext4_dx_readdir(filp, dirent, filldir);
ac27a0ec DK	113	if (err != ERR_BAD_DX_DIR) {
	114	ret = err;
	115	goto out;
	116	}
	117	/*
	118	* We don't set the inode dirty flag since it's not
	119	* critical that it get flushed back to the disk.
	120	*/
9d549890	121	EXT4_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT4_INDEX_FL;
ac27a0ec	122	}
ac27a0ec DK	123	stored = 0;
	124	offset = filp->f_pos & (sb->s_blocksize - 1);
	125
	126	while (!error && !stored && filp->f_pos < inode->i_size) {
725d26d3	127	ext4_lblk_t blk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec DK	128	struct buffer_head map_bh;
	129	struct buffer_head *bh = NULL;
	130
	131	map_bh.b_state = 0;
d2a17637 MC	132	err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh,
d2a17637 MC	133	0, 0, 0);
ac27a0ec	134	if (err > 0) {
dc7868fc FW	135	pgoff_t index = map_bh.b_blocknr >>
	136	(PAGE_CACHE_SHIFT - inode->i_blkbits);
	137	if (!ra_has_index(&filp->f_ra, index))
cf914a7d	138	page_cache_sync_readahead(
dc7868fc FW	139	sb->s_bdev->bd_inode->i_mapping,
dc7868fc FW	140	&filp->f_ra, filp,
cf914a7d	141	index, 1);
f4e6b498	142	filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
617ba13b	143	bh = ext4_bread(NULL, inode, blk, 0, &err);
ac27a0ec DK	144	}
	145
	146	/*
	147	* We ignore I/O errors on directories so users have a chance
	148	* of recovering data when there's a bad sector
	149	*/
	150	if (!bh) {
617ba13b	151	ext4_error (sb, "ext4_readdir",
ac27a0ec DK	152	"directory #%lu contains a hole at offset %lu",
ac27a0ec DK	153	inode->i_ino, (unsigned long)filp->f_pos);
e6c40211 ES	154	/* corrupt size? Maybe no more blocks to read */
	155	if (filp->f_pos > inode->i_blocks << 9)
	156	break;
ac27a0ec DK	157	filp->f_pos += sb->s_blocksize - offset;
	158	continue;
	159	}
	160
	161	revalidate:
	162	/* If the dir block has changed since the last call to
	163	* readdir(2), then we might be pointing to an invalid
	164	* dirent right now. Scan from the start of the block
	165	* to make sure. */
	166	if (filp->f_version != inode->i_version) {
	167	for (i = 0; i < sb->s_blocksize && i < offset; ) {
617ba13b	168	de = (struct ext4_dir_entry_2 *)
ac27a0ec DK	169	(bh->b_data + i);
	170	/* It's too expensive to do a full
	171	* dirent test each time round this
	172	* loop, but we do have to test at
	173	* least that it is non-zero. A
	174	* failure will be detected in the
	175	* dirent test below. */
a72d7f83 JK	176	if (ext4_rec_len_from_disk(de->rec_len)
a72d7f83 JK	177	< EXT4_DIR_REC_LEN(1))
ac27a0ec	178	break;
a72d7f83	179	i += ext4_rec_len_from_disk(de->rec_len);
ac27a0ec DK	180	}
	181	offset = i;
	182	filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
	183	\| offset;
	184	filp->f_version = inode->i_version;
	185	}
	186
	187	while (!error && filp->f_pos < inode->i_size
	188	&& offset < sb->s_blocksize) {
617ba13b MC	189	de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
617ba13b MC	190	if (!ext4_check_dir_entry ("ext4_readdir", inode, de,
ac27a0ec	191	bh, offset)) {
63f57933 AM	192	/*
	193	* On error, skip the f_pos to the next block
	194	*/
ac27a0ec DK	195	filp->f_pos = (filp->f_pos \|
	196	(sb->s_blocksize - 1)) + 1;
	197	brelse (bh);
	198	ret = stored;
	199	goto out;
	200	}
a72d7f83	201	offset += ext4_rec_len_from_disk(de->rec_len);
ac27a0ec DK	202	if (le32_to_cpu(de->inode)) {
	203	/* We might block in the next section
	204	* if the data destination is
	205	* currently swapped out. So, use a
	206	* version stamp to detect whether or
	207	* not the directory has been modified
	208	* during the copy operation.
	209	*/
2b47c361	210	u64 version = filp->f_version;
ac27a0ec DK	211
	212	error = filldir(dirent, de->name,
	213	de->name_len,
	214	filp->f_pos,
	215	le32_to_cpu(de->inode),
	216	get_dtype(sb, de->file_type));
	217	if (error)
	218	break;
	219	if (version != filp->f_version)
	220	goto revalidate;
	221	stored ++;
	222	}
a72d7f83	223	filp->f_pos += ext4_rec_len_from_disk(de->rec_len);
ac27a0ec DK	224	}
	225	offset = 0;
	226	brelse (bh);
	227	}
	228	out:
	229	return ret;
	230	}
	231
ac27a0ec DK	232	/*
	233	* These functions convert from the major/minor hash to an f_pos
	234	* value.
	235	*
	236	* Currently we only use major hash numer. This is unfortunate, but
	237	* on 32-bit machines, the same VFS interface is used for lseek and
	238	* llseek, so if we use the 64 bit offset, then the 32-bit versions of
	239	* lseek/telldir/seekdir will blow out spectacularly, and from within
	240	* the ext2 low-level routine, we don't know if we're being called by
	241	* a 64-bit version of the system call or the 32-bit version of the
	242	* system call. Worse yet, NFSv2 only allows for a 32-bit readdir
	243	* cookie. Sigh.
	244	*/
	245	#define hash2pos(major, minor) (major >> 1)
	246	#define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
	247	#define pos2min_hash(pos) (0)
	248
	249	/*
	250	* This structure holds the nodes of the red-black tree used to store
	251	* the directory entry in hash order.
	252	*/
	253	struct fname {
	254	__u32 hash;
	255	__u32 minor_hash;
	256	struct rb_node rb_hash;
	257	struct fname *next;
	258	__u32 inode;
	259	__u8 name_len;
	260	__u8 file_type;
	261	char name[0];
	262	};
	263
	264	/*
	265	* This functoin implements a non-recursive way of freeing all of the
	266	* nodes in the red-black tree.
	267	*/
	268	static void free_rb_tree_fname(struct rb_root *root)
	269	{
	270	struct rb_node *n = root->rb_node;
	271	struct rb_node *parent;
	272	struct fname *fname;
	273
	274	while (n) {
	275	/* Do the node's children first */
69baee06	276	if (n->rb_left) {
ac27a0ec DK	277	n = n->rb_left;
	278	continue;
	279	}
	280	if (n->rb_right) {
	281	n = n->rb_right;
	282	continue;
	283	}
	284	/*
	285	* The node has no children; free it, and then zero
	286	* out parent's link to it. Finally go to the
	287	* beginning of the loop and try to free the parent
	288	* node.
	289	*/
	290	parent = rb_parent(n);
	291	fname = rb_entry(n, struct fname, rb_hash);
	292	while (fname) {
	293	struct fname * old = fname;
	294	fname = fname->next;
	295	kfree (old);
	296	}
	297	if (!parent)
	298	root->rb_node = NULL;
	299	else if (parent->rb_left == n)
	300	parent->rb_left = NULL;
	301	else if (parent->rb_right == n)
	302	parent->rb_right = NULL;
	303	n = parent;
	304	}
ac27a0ec DK	305	}
	306
	307
69baee06	308	static struct dir_private_info *ext4_htree_create_dir_info(loff_t pos)
ac27a0ec DK	309	{
	310	struct dir_private_info *p;
	311
69baee06	312	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
ac27a0ec DK	313	if (!p)
ac27a0ec DK	314	return NULL;
ac27a0ec DK	315	p->curr_hash = pos2maj_hash(pos);
ac27a0ec DK	316	p->curr_minor_hash = pos2min_hash(pos);
ac27a0ec DK	317	return p;
	318	}
	319
617ba13b	320	void ext4_htree_free_dir_info(struct dir_private_info *p)
ac27a0ec DK	321	{
	322	free_rb_tree_fname(&p->root);
	323	kfree(p);
	324	}
	325
	326	/*
	327	* Given a directory entry, enter it into the fname rb tree.
	328	*/
617ba13b	329	int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
ac27a0ec	330	__u32 minor_hash,
617ba13b	331	struct ext4_dir_entry_2 *dirent)
ac27a0ec DK	332	{
	333	struct rb_node *p, parent = NULL;
	334	struct fname * fname, *new_fn;
	335	struct dir_private_info *info;
	336	int len;
	337
	338	info = (struct dir_private_info *) dir_file->private_data;
	339	p = &info->root.rb_node;
	340
	341	/* Create and allocate the fname structure */
	342	len = sizeof(struct fname) + dirent->name_len + 1;
	343	new_fn = kzalloc(len, GFP_KERNEL);
	344	if (!new_fn)
	345	return -ENOMEM;
	346	new_fn->hash = hash;
	347	new_fn->minor_hash = minor_hash;
	348	new_fn->inode = le32_to_cpu(dirent->inode);
	349	new_fn->name_len = dirent->name_len;
	350	new_fn->file_type = dirent->file_type;
	351	memcpy(new_fn->name, dirent->name, dirent->name_len);
	352	new_fn->name[dirent->name_len] = 0;
	353
	354	while (*p) {
	355	parent = *p;
	356	fname = rb_entry(parent, struct fname, rb_hash);
	357
	358	/*
	359	* If the hash and minor hash match up, then we put
	360	* them on a linked list. This rarely happens...
	361	*/
	362	if ((new_fn->hash == fname->hash) &&
	363	(new_fn->minor_hash == fname->minor_hash)) {
	364	new_fn->next = fname->next;
	365	fname->next = new_fn;
	366	return 0;
	367	}
	368
	369	if (new_fn->hash < fname->hash)
	370	p = &(*p)->rb_left;
	371	else if (new_fn->hash > fname->hash)
	372	p = &(*p)->rb_right;
	373	else if (new_fn->minor_hash < fname->minor_hash)
	374	p = &(*p)->rb_left;
	375	else /* if (new_fn->minor_hash > fname->minor_hash) */
	376	p = &(*p)->rb_right;
	377	}
	378
	379	rb_link_node(&new_fn->rb_hash, parent, p);
	380	rb_insert_color(&new_fn->rb_hash, &info->root);
	381	return 0;
	382	}
	383
	384
	385
	386	/*
617ba13b	387	* This is a helper function for ext4_dx_readdir. It calls filldir
ac27a0ec DK	388	* for all entres on the fname linked list. (Normally there is only
	389	* one entry on the linked list, unless there are 62 bit hash collisions.)
	390	*/
	391	static int call_filldir(struct file * filp, void * dirent,
	392	filldir_t filldir, struct fname *fname)
	393	{
	394	struct dir_private_info *info = filp->private_data;
	395	loff_t curr_pos;
9d549890	396	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	397	struct super_block * sb;
	398	int error;
	399
	400	sb = inode->i_sb;
	401
	402	if (!fname) {
	403	printk("call_filldir: called with null fname?!?\n");
	404	return 0;
	405	}
	406	curr_pos = hash2pos(fname->hash, fname->minor_hash);
	407	while (fname) {
	408	error = filldir(dirent, fname->name,
	409	fname->name_len, curr_pos,
	410	fname->inode,
	411	get_dtype(sb, fname->file_type));
	412	if (error) {
	413	filp->f_pos = curr_pos;
	414	info->extra_fname = fname->next;
	415	return error;
	416	}
	417	fname = fname->next;
	418	}
	419	return 0;
	420	}
	421
617ba13b	422	static int ext4_dx_readdir(struct file * filp,
ac27a0ec DK	423	void * dirent, filldir_t filldir)
	424	{
	425	struct dir_private_info *info = filp->private_data;
9d549890	426	struct inode *inode = filp->f_path.dentry->d_inode;
ac27a0ec DK	427	struct fname *fname;
	428	int ret;
	429
	430	if (!info) {
69baee06	431	info = ext4_htree_create_dir_info(filp->f_pos);
ac27a0ec DK	432	if (!info)
	433	return -ENOMEM;
	434	filp->private_data = info;
	435	}
	436
617ba13b	437	if (filp->f_pos == EXT4_HTREE_EOF)
ac27a0ec DK	438	return 0; /* EOF */
	439
	440	/* Some one has messed with f_pos; reset the world */
	441	if (info->last_pos != filp->f_pos) {
	442	free_rb_tree_fname(&info->root);
	443	info->curr_node = NULL;
	444	info->extra_fname = NULL;
	445	info->curr_hash = pos2maj_hash(filp->f_pos);
	446	info->curr_minor_hash = pos2min_hash(filp->f_pos);
	447	}
	448
	449	/*
	450	* If there are any leftover names on the hash collision
	451	* chain, return them first.
	452	*/
	453	if (info->extra_fname &&
	454	call_filldir(filp, dirent, filldir, info->extra_fname))
	455	goto finished;
	456
	457	if (!info->curr_node)
	458	info->curr_node = rb_first(&info->root);
	459
	460	while (1) {
	461	/*
	462	* Fill the rbtree if we have no more entries,
	463	* or the inode has changed since we last read in the
	464	* cached entries.
	465	*/
	466	if ((!info->curr_node) \|\|
	467	(filp->f_version != inode->i_version)) {
	468	info->curr_node = NULL;
	469	free_rb_tree_fname(&info->root);
	470	filp->f_version = inode->i_version;
617ba13b	471	ret = ext4_htree_fill_tree(filp, info->curr_hash,
ac27a0ec DK	472	info->curr_minor_hash,
	473	&info->next_hash);
	474	if (ret < 0)
	475	return ret;
	476	if (ret == 0) {
617ba13b	477	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	478	break;
	479	}
	480	info->curr_node = rb_first(&info->root);
	481	}
	482
	483	fname = rb_entry(info->curr_node, struct fname, rb_hash);
	484	info->curr_hash = fname->hash;
	485	info->curr_minor_hash = fname->minor_hash;
	486	if (call_filldir(filp, dirent, filldir, fname))
	487	break;
	488
	489	info->curr_node = rb_next(info->curr_node);
	490	if (!info->curr_node) {
	491	if (info->next_hash == ~0) {
617ba13b	492	filp->f_pos = EXT4_HTREE_EOF;
ac27a0ec DK	493	break;
	494	}
	495	info->curr_hash = info->next_hash;
	496	info->curr_minor_hash = 0;
	497	}
	498	}
	499	finished:
	500	info->last_pos = filp->f_pos;
	501	return 0;
	502	}
	503
617ba13b	504	static int ext4_release_dir (struct inode * inode, struct file * filp)
ac27a0ec	505	{
63f57933	506	if (filp->private_data)
617ba13b	507	ext4_htree_free_dir_info(filp->private_data);
ac27a0ec DK	508
	509	return 0;
	510	}