exfat/libexfat/node.c

/*
	node.c (09.10.09)
	exFAT file system implementation library.

	Free exFAT implementation.
	Copyright (C) 2010-2015  Andrew Nayenko

	This program is free software; you can redistribute it and/or modify
	it under the terms of the GNU General Public License as published by
	the Free Software Foundation, either version 2 of the License, or
	(at your option) any later version.

	This program is distributed in the hope that it will be useful,
	but WITHOUT ANY WARRANTY; without even the implied warranty of
	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
	GNU General Public License for more details.

	You should have received a copy of the GNU General Public License along
	with this program; if not, write to the Free Software Foundation, Inc.,
	51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/

#include "exfat.h"
#include <errno.h>
#include <string.h>
#include <inttypes.h>

/* on-disk nodes iterator */
struct iterator
{
	cluster_t cluster;
	off_t offset;
	int contiguous;
	char* chunk;
};

struct exfat_node* exfat_get_node(struct exfat_node* node)
{
	/* if we switch to multi-threaded mode we will need atomic
	   increment here and atomic decrement in exfat_put_node() */
	node->references++;
	return node;
}

void exfat_put_node(struct exfat* ef, struct exfat_node* node)
{
	char buffer[UTF8_BYTES(EXFAT_NAME_MAX) + 1];

	--node->references;
	if (node->references < 0)
	{
		exfat_get_name(node, buffer, sizeof(buffer) - 1);
		exfat_bug("reference counter of '%s' is below zero", buffer);
	}
	else if (node->references == 0 && node != ef->root)
	{
		if (node->flags & EXFAT_ATTRIB_DIRTY)
		{
			exfat_get_name(node, buffer, sizeof(buffer) - 1);
			exfat_warn("dirty node '%s' with zero references", buffer);
		}
	}
}

/**
 * This function must be called on rmdir and unlink (after the last
 * exfat_put_node()) to free clusters.
 */
int exfat_cleanup_node(struct exfat* ef, struct exfat_node* node)
{
	int rc = 0;

	if (node->references != 0)
		exfat_bug("unable to cleanup a node with %d references",
				node->references);

	if (node->flags & EXFAT_ATTRIB_UNLINKED)
	{
		/* free all clusters and node structure itself */
		rc = exfat_truncate(ef, node, 0, true);
		/* free the node even in case of error or its memory will be lost */
		free(node);
	}
	return rc;
}

/**
 * Cluster + offset from the beginning of the directory to absolute offset.
 */
static off_t co2o(struct exfat* ef, cluster_t cluster, off_t offset)
{
	return exfat_c2o(ef, cluster) + offset % CLUSTER_SIZE(*ef->sb);
}

static int opendir(struct exfat* ef, const struct exfat_node* dir,
		struct iterator* it)
{
	if (!(dir->flags & EXFAT_ATTRIB_DIR))
		exfat_bug("not a directory");
	it->cluster = dir->start_cluster;
	it->offset = 0;
	it->contiguous = IS_CONTIGUOUS(*dir);
	it->chunk = malloc(CLUSTER_SIZE(*ef->sb));
	if (it->chunk == NULL)
	{
		exfat_error("out of memory");
		return -ENOMEM;
	}
	if (exfat_pread(ef->dev, it->chunk, CLUSTER_SIZE(*ef->sb),
			exfat_c2o(ef, it->cluster)) < 0)
	{
		exfat_error("failed to read directory cluster %#x", it->cluster);
		return -EIO;
	}
	return 0;
}

static void closedir(struct iterator* it)
{
	it->cluster = 0;
	it->offset = 0;
	it->contiguous = 0;
	free(it->chunk);
	it->chunk = NULL;
}

static bool fetch_next_entry(struct exfat* ef, const struct exfat_node* parent,
		struct iterator* it)
{
	/* move iterator to the next entry in the directory */
	it->offset += sizeof(struct exfat_entry);
	/* fetch the next cluster if needed */
	if ((it->offset & (CLUSTER_SIZE(*ef->sb) - 1)) == 0)
	{
		/* reached the end of directory; the caller should check this
		   condition too */
		if (it->offset >= parent->size)
			return true;
		it->cluster = exfat_next_cluster(ef, parent, it->cluster);
		if (CLUSTER_INVALID(it->cluster))
		{
			exfat_error("invalid cluster 0x%x while reading directory",
					it->cluster);
			return false;
		}
		if (exfat_pread(ef->dev, it->chunk, CLUSTER_SIZE(*ef->sb),
				exfat_c2o(ef, it->cluster)) < 0)
		{
			exfat_error("failed to read the next directory cluster %#x",
					it->cluster);
			return false;
		}
	}
	return true;
}

static struct exfat_node* allocate_node(void)
{
	struct exfat_node* node = malloc(sizeof(struct exfat_node));
	if (node == NULL)
	{
		exfat_error("failed to allocate node");
		return NULL;
	}
	memset(node, 0, sizeof(struct exfat_node));
	return node;
}

static void init_node_meta1(struct exfat_node* node,
		const struct exfat_entry_meta1* meta1)
{
	node->flags = le16_to_cpu(meta1->attrib);
	node->mtime = exfat_exfat2unix(meta1->mdate, meta1->mtime,
			meta1->mtime_cs);
	/* there is no centiseconds field for atime */
	node->atime = exfat_exfat2unix(meta1->adate, meta1->atime, 0);
}

static void init_node_meta2(struct exfat_node* node,
		const struct exfat_entry_meta2* meta2)
{
	node->size = le64_to_cpu(meta2->size);
	node->start_cluster = le32_to_cpu(meta2->start_cluster);
	node->fptr_cluster = node->start_cluster;
	if (meta2->flags & EXFAT_FLAG_CONTIGUOUS)
		node->flags |= EXFAT_ATTRIB_CONTIGUOUS;
}

static const struct exfat_entry* get_entry_ptr(const struct exfat* ef,
		const struct iterator* it)
{
	return (const struct exfat_entry*)
			(it->chunk + it->offset % CLUSTER_SIZE(*ef->sb));
}

static bool check_node(const struct exfat_node* node, uint16_t actual_checksum,
		uint16_t reference_checksum, uint64_t valid_size)
{
	char buffer[UTF8_BYTES(EXFAT_NAME_MAX) + 1];

	/*
	   Validate checksum first. If it's invalid all other fields probably
	   contain just garbage.
	*/
	if (actual_checksum != reference_checksum)
	{
		exfat_get_name(node, buffer, sizeof(buffer) - 1);
		exfat_error("'%s' has invalid checksum (%#hx != %#hx)", buffer,
				actual_checksum, reference_checksum);
		return false;
	}

	/*
	   exFAT does not support sparse files but allows files with uninitialized
	   clusters. For such files valid_size means initialized data size and
	   cannot be greater than file size. See SetFileValidData() function
	   description in MSDN.
	*/
	if (valid_size > node->size)
	{
		exfat_get_name(node, buffer, sizeof(buffer) - 1);
		exfat_error("'%s' has valid size (%"PRIu64") greater than size "
				"(%"PRIu64")", buffer, valid_size, node->size);
		return false;
	}

	return true;
}

/*
 * Reads one entry in directory at position pointed by iterator and fills
 * node structure.
 */
static int readdir(struct exfat* ef, const struct exfat_node* parent,
		struct exfat_node** node, struct iterator* it)
{
	int rc = -EIO;
	const struct exfat_entry* entry;
	const struct exfat_entry_meta1* meta1;
	const struct exfat_entry_meta2* meta2;
	const struct exfat_entry_name* file_name;
	const struct exfat_entry_upcase* upcase;
	const struct exfat_entry_bitmap* bitmap;
	const struct exfat_entry_label* label;
	uint8_t continuations = 0;
	le16_t* namep = NULL;
	uint16_t reference_checksum = 0;
	uint16_t actual_checksum = 0;
	uint64_t valid_size = 0;

	*node = NULL;

	for (;;)
	{
		if (it->offset >= parent->size)
		{
			if (continuations != 0)
			{
				exfat_error("expected %hhu continuations", continuations);
				goto error;
			}
			return -ENOENT; /* that's OK, means end of directory */
		}

		entry = get_entry_ptr(ef, it);
		switch (entry->type)
		{
		case EXFAT_ENTRY_FILE:
			if (continuations != 0)
			{
				exfat_error("expected %hhu continuations before new entry",
						continuations);
				goto error;
			}
			meta1 = (const struct exfat_entry_meta1*) entry;
			continuations = meta1->continuations;
			/* each file entry must have at least 2 continuations:
			   info and name */
			if (continuations < 2)
			{
				exfat_error("too few continuations (%hhu)", continuations);
				goto error;
			}
			if (continuations > 1 +
					DIV_ROUND_UP(EXFAT_NAME_MAX, EXFAT_ENAME_MAX))
			{
				exfat_error("too many continuations (%hhu)", continuations);
				goto error;
			}
			reference_checksum = le16_to_cpu(meta1->checksum);
			actual_checksum = exfat_start_checksum(meta1);
			*node = allocate_node();
			if (*node == NULL)
			{
				rc = -ENOMEM;
				goto error;
			}
			/* new node has zero reference counter */
			(*node)->entry_cluster = it->cluster;
			(*node)->entry_offset = it->offset;
			init_node_meta1(*node, meta1);
			namep = (*node)->name;
			break;

		case EXFAT_ENTRY_FILE_INFO:
			if (continuations < 2)
			{
				exfat_error("unexpected continuation (%hhu)",
						continuations);
				goto error;
			}
			meta2 = (const struct exfat_entry_meta2*) entry;
			if (meta2->flags & ~(EXFAT_FLAG_ALWAYS1 | EXFAT_FLAG_CONTIGUOUS))
			{
				exfat_error("unknown flags in meta2 (0x%hhx)", meta2->flags);
				goto error;
			}
			init_node_meta2(*node, meta2);
			actual_checksum = exfat_add_checksum(entry, actual_checksum);
			valid_size = le64_to_cpu(meta2->valid_size);
			/* empty files must be marked as non-contiguous */
			if ((*node)->size == 0 && (meta2->flags & EXFAT_FLAG_CONTIGUOUS))
			{
				exfat_error("empty file marked as contiguous (0x%hhx)",
						meta2->flags);
				goto error;
			}
			/* directories must be aligned on at cluster boundary */
			if (((*node)->flags & EXFAT_ATTRIB_DIR) &&
				(*node)->size % CLUSTER_SIZE(*ef->sb) != 0)
			{
				exfat_error("directory has invalid size %"PRIu64" bytes",
						(*node)->size);
				goto error;
			}
			--continuations;
			break;

		case EXFAT_ENTRY_FILE_NAME:
			if (continuations == 0)
			{
				exfat_error("unexpected continuation");
				goto error;
			}
			file_name = (const struct exfat_entry_name*) entry;
			actual_checksum = exfat_add_checksum(entry, actual_checksum);

			memcpy(namep, file_name->name,
					MIN(EXFAT_ENAME_MAX,
						((*node)->name + EXFAT_NAME_MAX - namep)) *
					sizeof(le16_t));
			namep += EXFAT_ENAME_MAX;
			if (--continuations == 0)
			{
				if (!check_node(*node, actual_checksum, reference_checksum,
						valid_size))
					goto error;
				if (!fetch_next_entry(ef, parent, it))
					goto error;
				return 0; /* entry completed */
			}
			break;

		case EXFAT_ENTRY_UPCASE:
			if (ef->upcase != NULL)
				break;
			upcase = (const struct exfat_entry_upcase*) entry;
			if (CLUSTER_INVALID(le32_to_cpu(upcase->start_cluster)))
			{
				exfat_error("invalid cluster 0x%x in upcase table",
						le32_to_cpu(upcase->start_cluster));
				goto error;
			}
			if (le64_to_cpu(upcase->size) == 0 ||
				le64_to_cpu(upcase->size) > 0xffff * sizeof(uint16_t) ||
				le64_to_cpu(upcase->size) % sizeof(uint16_t) != 0)
			{
				exfat_error("bad upcase table size (%"PRIu64" bytes)",
						le64_to_cpu(upcase->size));
				goto error;
			}
			ef->upcase = malloc(le64_to_cpu(upcase->size));
			if (ef->upcase == NULL)
			{
				exfat_error("failed to allocate upcase table (%"PRIu64" bytes)",
						le64_to_cpu(upcase->size));
				rc = -ENOMEM;
				goto error;
			}
			ef->upcase_chars = le64_to_cpu(upcase->size) / sizeof(le16_t);

			if (exfat_pread(ef->dev, ef->upcase, le64_to_cpu(upcase->size),
					exfat_c2o(ef, le32_to_cpu(upcase->start_cluster))) < 0)
			{
				exfat_error("failed to read upper case table "
						"(%"PRIu64" bytes starting at cluster %#x)",
						le64_to_cpu(upcase->size),
						le32_to_cpu(upcase->start_cluster));
				goto error;
			}
			break;

		case EXFAT_ENTRY_BITMAP:
			bitmap = (const struct exfat_entry_bitmap*) entry;
			ef->cmap.start_cluster = le32_to_cpu(bitmap->start_cluster);
			if (CLUSTER_INVALID(ef->cmap.start_cluster))
			{
				exfat_error("invalid cluster 0x%x in clusters bitmap",
						ef->cmap.start_cluster);
				goto error;
			}
			ef->cmap.size = le32_to_cpu(ef->sb->cluster_count) -
				EXFAT_FIRST_DATA_CLUSTER;
			if (le64_to_cpu(bitmap->size) < DIV_ROUND_UP(ef->cmap.size, 8))
			{
				exfat_error("invalid clusters bitmap size: %"PRIu64
						" (expected at least %u)",
						le64_to_cpu(bitmap->size),
						DIV_ROUND_UP(ef->cmap.size, 8));
				goto error;
			}
			/* FIXME bitmap can be rather big, up to 512 MB */
			ef->cmap.chunk_size = ef->cmap.size;
			ef->cmap.chunk = malloc(BMAP_SIZE(ef->cmap.chunk_size));
			if (ef->cmap.chunk == NULL)
			{
				exfat_error("failed to allocate clusters bitmap chunk "
						"(%"PRIu64" bytes)", le64_to_cpu(bitmap->size));
				rc = -ENOMEM;
				goto error;
			}

			if (exfat_pread(ef->dev, ef->cmap.chunk,
					BMAP_SIZE(ef->cmap.chunk_size),
					exfat_c2o(ef, ef->cmap.start_cluster)) < 0)
			{
				exfat_error("failed to read clusters bitmap "
						"(%"PRIu64" bytes starting at cluster %#x)",
						le64_to_cpu(bitmap->size), ef->cmap.start_cluster);
				goto error;
			}
			break;

		case EXFAT_ENTRY_LABEL:
			label = (const struct exfat_entry_label*) entry;
			if (label->length > EXFAT_ENAME_MAX)
			{
				exfat_error("too long label (%hhu chars)", label->length);
				goto error;
			}
			if (utf16_to_utf8(ef->label, label->name,
						sizeof(ef->label) - 1, EXFAT_ENAME_MAX) != 0)
				goto error;
			break;

		default:
			if (!(entry->type & EXFAT_ENTRY_VALID))
				break; /* deleted entry, ignore it */
			if (!(entry->type & EXFAT_ENTRY_OPTIONAL))
			{
				exfat_error("unknown entry type %#hhx", entry->type);
				goto error;
			}
			/* optional entry, warn and skip */
			exfat_warn("unknown entry type %#hhx", entry->type);
			if (continuations == 0)
			{
				exfat_error("unexpected continuation");
				goto error;
			}
			--continuations;
			break;
		}

		if (!fetch_next_entry(ef, parent, it))
			goto error;
	}
	/* we never reach here */

error:
	free(*node);
	*node = NULL;
	return rc;
}

int exfat_cache_directory(struct exfat* ef, struct exfat_node* dir)
{
	struct iterator it;
	int rc;
	struct exfat_node* node;
	struct exfat_node* current = NULL;

	if (dir->flags & EXFAT_ATTRIB_CACHED)
		return 0; /* already cached */

	rc = opendir(ef, dir, &it);
	if (rc != 0)
		return rc;
	while ((rc = readdir(ef, dir, &node, &it)) == 0)
	{
		node->parent = dir;
		if (current != NULL)
		{
			current->next = node;
			node->prev = current;
		}
		else
			dir->child = node;

		current = node;
	}
	closedir(&it);

	if (rc != -ENOENT)
	{
		/* rollback */
		for (current = dir->child; current; current = node)
		{
			node = current->next;
			free(current);
		}
		dir->child = NULL;
		return rc;
	}

	dir->flags |= EXFAT_ATTRIB_CACHED;
	return 0;
}

static void tree_attach(struct exfat_node* dir, struct exfat_node* node)
{
	node->parent = dir;
	if (dir->child)
	{
		dir->child->prev = node;
		node->next = dir->child;
	}
	dir->child = node;
}

static void tree_detach(struct exfat_node* node)
{
	if (node->prev)
		node->prev->next = node->next;
	else /* this is the first node in the list */
		node->parent->child = node->next;
	if (node->next)
		node->next->prev = node->prev;
	node->parent = NULL;
	node->prev = NULL;
	node->next = NULL;
}

static void reset_cache(struct exfat* ef, struct exfat_node* node)
{
	char buffer[UTF8_BYTES(EXFAT_NAME_MAX) + 1];

	while (node->child)
	{
		struct exfat_node* p = node->child;
		reset_cache(ef, p);
		tree_detach(p);
		free(p);
	}
	node->flags &= ~EXFAT_ATTRIB_CACHED;
	if (node->references != 0)
	{
		exfat_get_name(node, buffer, sizeof(buffer) - 1);
		exfat_warn("non-zero reference counter (%d) for '%s'",
				node->references, buffer);
	}
	if (node != ef->root && (node->flags & EXFAT_ATTRIB_DIRTY))
	{
		exfat_get_name(node, buffer, sizeof(buffer) - 1);
		exfat_bug("node '%s' is dirty", buffer);
	}
	while (node->references)
		exfat_put_node(ef, node);
}

void exfat_reset_cache(struct exfat* ef)
{
	reset_cache(ef, ef->root);
}

static bool next_entry(struct exfat* ef, const struct exfat_node* parent,
		cluster_t* cluster, off_t* offset)
{
	*offset += sizeof(struct exfat_entry);
	if (*offset % CLUSTER_SIZE(*ef->sb) == 0)
	{
		*cluster = exfat_next_cluster(ef, parent, *cluster);
		if (CLUSTER_INVALID(*cluster))
		{
			exfat_error("invalid cluster %#x while getting next entry",
					*cluster);
			return false;
		}
	}
	return true;
}

int exfat_flush_node(struct exfat* ef, struct exfat_node* node)
{
	cluster_t cluster;
	off_t offset;
	off_t meta1_offset, meta2_offset;
	struct exfat_entry_meta1 meta1;
	struct exfat_entry_meta2 meta2;

	if (!(node->flags & EXFAT_ATTRIB_DIRTY))
		return 0; /* no need to flush */

	if (ef->ro)
		exfat_bug("unable to flush node to read-only FS");

	if (node->parent == NULL)
		return 0; /* do not flush unlinked node */

	cluster = node->entry_cluster;
	offset = node->entry_offset;
	meta1_offset = co2o(ef, cluster, offset);
	if (!next_entry(ef, node->parent, &cluster, &offset))
		return -EIO;
	meta2_offset = co2o(ef, cluster, offset);

	if (exfat_pread(ef->dev, &meta1, sizeof(meta1), meta1_offset) < 0)
	{
		exfat_error("failed to read meta1 entry on flush");
		return -EIO;
	}
	if (meta1.type != EXFAT_ENTRY_FILE)
		exfat_bug("invalid type of meta1: 0x%hhx", meta1.type);
	meta1.attrib = cpu_to_le16(node->flags);
	exfat_unix2exfat(node->mtime, &meta1.mdate, &meta1.mtime, &meta1.mtime_cs);
	exfat_unix2exfat(node->atime, &meta1.adate, &meta1.atime, NULL);

	if (exfat_pread(ef->dev, &meta2, sizeof(meta2), meta2_offset) < 0)
	{
		exfat_error("failed to read meta2 entry on flush");
		return -EIO;
	}
	if (meta2.type != EXFAT_ENTRY_FILE_INFO)
		exfat_bug("invalid type of meta2: 0x%hhx", meta2.type);
	meta2.size = meta2.valid_size = cpu_to_le64(node->size);
	meta2.start_cluster = cpu_to_le32(node->start_cluster);
	meta2.flags = EXFAT_FLAG_ALWAYS1;
	/* empty files must not be marked as contiguous */
	if (node->size != 0 && IS_CONTIGUOUS(*node))
		meta2.flags |= EXFAT_FLAG_CONTIGUOUS;
	/* name hash remains unchanged, no need to recalculate it */

	meta1.checksum = exfat_calc_checksum(&meta1, &meta2, node->name);

	if (exfat_pwrite(ef->dev, &meta1, sizeof(meta1), meta1_offset) < 0)
	{
		exfat_error("failed to write meta1 entry on flush");
		return -EIO;
	}
	if (exfat_pwrite(ef->dev, &meta2, sizeof(meta2), meta2_offset) < 0)
	{
		exfat_error("failed to write meta2 entry on flush");
		return -EIO;
	}

	node->flags &= ~EXFAT_ATTRIB_DIRTY;
	return exfat_flush(ef);
}

static bool erase_entry(struct exfat* ef, struct exfat_node* node)
{
	cluster_t cluster = node->entry_cluster;
	off_t offset = node->entry_offset;
	int name_entries = DIV_ROUND_UP(utf16_length(node->name), EXFAT_ENAME_MAX);
	uint8_t entry_type;

	entry_type = EXFAT_ENTRY_FILE & ~EXFAT_ENTRY_VALID;
	if (exfat_pwrite(ef->dev, &entry_type, 1, co2o(ef, cluster, offset)) < 0)
	{
		exfat_error("failed to erase meta1 entry");
		return false;
	}

	if (!next_entry(ef, node->parent, &cluster, &offset))
		return false;
	entry_type = EXFAT_ENTRY_FILE_INFO & ~EXFAT_ENTRY_VALID;
	if (exfat_pwrite(ef->dev, &entry_type, 1, co2o(ef, cluster, offset)) < 0)
	{
		exfat_error("failed to erase meta2 entry");
		return false;
	}

	while (name_entries--)
	{
		if (!next_entry(ef, node->parent, &cluster, &offset))
			return false;
		entry_type = EXFAT_ENTRY_FILE_NAME & ~EXFAT_ENTRY_VALID;
		if (exfat_pwrite(ef->dev, &entry_type, 1,
				co2o(ef, cluster, offset)) < 0)
		{
			exfat_error("failed to erase name entry");
			return false;
		}
	}
	return true;
}

static int shrink_directory(struct exfat* ef, struct exfat_node* dir,
		off_t deleted_offset)
{
	const struct exfat_node* node;
	const struct exfat_node* last_node;
	uint64_t entries = 0;
	uint64_t new_size;

	if (!(dir->flags & EXFAT_ATTRIB_DIR))
		exfat_bug("attempted to shrink a file");
	if (!(dir->flags & EXFAT_ATTRIB_CACHED))
		exfat_bug("attempted to shrink uncached directory");

	for (last_node = node = dir->child; node; node = node->next)
	{
		if (deleted_offset < node->entry_offset)
		{
			/* there are other entries after the removed one, no way to shrink
			   this directory */
			return 0;
		}
		if (last_node->entry_offset < node->entry_offset)
			last_node = node;
	}

	if (last_node)
	{
		/* offset of the last entry */
		entries += last_node->entry_offset / sizeof(struct exfat_entry);
		/* two subentries with meta info */
		entries += 2;
		/* subentries with file name */
		entries += DIV_ROUND_UP(utf16_length(last_node->name),
				EXFAT_ENAME_MAX);
	}

	new_size = DIV_ROUND_UP(entries * sizeof(struct exfat_entry),
				 CLUSTER_SIZE(*ef->sb)) * CLUSTER_SIZE(*ef->sb);
	if (new_size == 0) /* directory always has at least 1 cluster */
		new_size = CLUSTER_SIZE(*ef->sb);
	if (new_size == dir->size)
		return 0;
	return exfat_truncate(ef, dir, new_size, true);
}

static int delete(struct exfat* ef, struct exfat_node* node)
{
	struct exfat_node* parent = node->parent;
	off_t deleted_offset = node->entry_offset;
	int rc;

	exfat_get_node(parent);
	if (!erase_entry(ef, node))
	{
		exfat_put_node(ef, parent);
		return -EIO;
	}
	exfat_update_mtime(parent);
	tree_detach(node);
	rc = shrink_directory(ef, parent, deleted_offset);
	node->flags |= EXFAT_ATTRIB_UNLINKED;
	if (rc != 0)
	{
		exfat_flush_node(ef, parent);
		exfat_put_node(ef, parent);
		return rc;
	}
	rc = exfat_flush_node(ef, parent);
	exfat_put_node(ef, parent);
	return rc;
}

int exfat_unlink(struct exfat* ef, struct exfat_node* node)
{
	if (node->flags & EXFAT_ATTRIB_DIR)
		return -EISDIR;
	return delete(ef, node);
}

int exfat_rmdir(struct exfat* ef, struct exfat_node* node)
{
	int rc;

	if (!(node->flags & EXFAT_ATTRIB_DIR))
		return -ENOTDIR;
	/* check that directory is empty */
	rc = exfat_cache_directory(ef, node);
	if (rc != 0)
		return rc;
	if (node->child)
		return -ENOTEMPTY;
	return delete(ef, node);
}

static int grow_directory(struct exfat* ef, struct exfat_node* dir,
		uint64_t asize, uint32_t difference)
{
	return exfat_truncate(ef, dir,
			DIV_ROUND_UP(asize + difference, CLUSTER_SIZE(*ef->sb))
				* CLUSTER_SIZE(*ef->sb), true);
}

static int find_slot(struct exfat* ef, struct exfat_node* dir,
		cluster_t* cluster, off_t* offset, int subentries)
{
	struct iterator it;
	int rc;
	const struct exfat_entry* entry;
	int contiguous = 0;

	rc = opendir(ef, dir, &it);
	if (rc != 0)
		return rc;
	for (;;)
	{
		if (contiguous == 0)
		{
			*cluster = it.cluster;
			*offset = it.offset;
		}
		entry = get_entry_ptr(ef, &it);
		if (entry->type & EXFAT_ENTRY_VALID)
			contiguous = 0;
		else
			contiguous++;
		if (contiguous == subentries)
			break;	/* suitable slot is found */
		if (it.offset + sizeof(struct exfat_entry) >= dir->size)
		{
			rc = grow_directory(ef, dir, dir->size,
					(subentries - contiguous) * sizeof(struct exfat_entry));
			if (rc != 0)
			{
				closedir(&it);
				return rc;
			}
		}
		if (!fetch_next_entry(ef, dir, &it))
		{
			closedir(&it);
			return -EIO;
		}
	}
	closedir(&it);
	return 0;
}

static int write_entry(struct exfat* ef, struct exfat_node* dir,
		const le16_t* name, cluster_t cluster, off_t offset, uint16_t attrib)
{
	struct exfat_node* node;
	struct exfat_entry_meta1 meta1;
	struct exfat_entry_meta2 meta2;
	const size_t name_length = utf16_length(name);
	const int name_entries = DIV_ROUND_UP(name_length, EXFAT_ENAME_MAX);
	int i;

	node = allocate_node();
	if (node == NULL)
		return -ENOMEM;
	node->entry_cluster = cluster;
	node->entry_offset = offset;
	memcpy(node->name, name, name_length * sizeof(le16_t));

	memset(&meta1, 0, sizeof(meta1));
	meta1.type = EXFAT_ENTRY_FILE;
	meta1.continuations = 1 + name_entries;
	meta1.attrib = cpu_to_le16(attrib);
	exfat_unix2exfat(time(NULL), &meta1.crdate, &meta1.crtime,
			&meta1.crtime_cs);
	meta1.adate = meta1.mdate = meta1.crdate;
	meta1.atime = meta1.mtime = meta1.crtime;
	meta1.mtime_cs = meta1.crtime_cs; /* there is no atime_cs */

	memset(&meta2, 0, sizeof(meta2));
	meta2.type = EXFAT_ENTRY_FILE_INFO;
	meta2.flags = EXFAT_FLAG_ALWAYS1;
	meta2.name_length = name_length;
	meta2.name_hash = exfat_calc_name_hash(ef, node->name);
	meta2.start_cluster = cpu_to_le32(EXFAT_CLUSTER_FREE);

	meta1.checksum = exfat_calc_checksum(&meta1, &meta2, node->name);

	if (exfat_pwrite(ef->dev, &meta1, sizeof(meta1),
			co2o(ef, cluster, offset)) < 0)
	{
		exfat_error("failed to write meta1 entry");
		return -EIO;
	}
	if (!next_entry(ef, dir, &cluster, &offset))
		return -EIO;
	if (exfat_pwrite(ef->dev, &meta2, sizeof(meta2),
			co2o(ef, cluster, offset)) < 0)
	{
		exfat_error("failed to write meta2 entry");
		return -EIO;
	}
	for (i = 0; i < name_entries; i++)
	{
		struct exfat_entry_name name_entry = {EXFAT_ENTRY_FILE_NAME, 0};
		memcpy(name_entry.name, node->name + i * EXFAT_ENAME_MAX,
				MIN(EXFAT_ENAME_MAX, EXFAT_NAME_MAX - i * EXFAT_ENAME_MAX) *
				sizeof(le16_t));
		if (!next_entry(ef, dir, &cluster, &offset))
			return -EIO;
		if (exfat_pwrite(ef->dev, &name_entry, sizeof(name_entry),
				co2o(ef, cluster, offset)) < 0)
		{
			exfat_error("failed to write name entry");
			return -EIO;
		}
	}

	init_node_meta1(node, &meta1);
	init_node_meta2(node, &meta2);

	tree_attach(dir, node);
	exfat_update_mtime(dir);
	return 0;
}

static int create(struct exfat* ef, const char* path, uint16_t attrib)
{
	struct exfat_node* dir;
	struct exfat_node* existing;
	cluster_t cluster = EXFAT_CLUSTER_BAD;
	off_t offset = -1;
	le16_t name[EXFAT_NAME_MAX + 1];
	int rc;

	rc = exfat_split(ef, &dir, &existing, name, path);
	if (rc != 0)
		return rc;
	if (existing != NULL)
	{
		exfat_put_node(ef, existing);
		exfat_put_node(ef, dir);
		return -EEXIST;
	}

	rc = find_slot(ef, dir, &cluster, &offset,
			2 + DIV_ROUND_UP(utf16_length(name), EXFAT_ENAME_MAX));
	if (rc != 0)
	{
		exfat_put_node(ef, dir);
		return rc;
	}
	rc = write_entry(ef, dir, name, cluster, offset, attrib);
	if (rc != 0)
	{
		exfat_put_node(ef, dir);
		return rc;
	}
	rc = exfat_flush_node(ef, dir);
	exfat_put_node(ef, dir);
	return rc;
}

int exfat_mknod(struct exfat* ef, const char* path)
{
	return create(ef, path, EXFAT_ATTRIB_ARCH);
}

int exfat_mkdir(struct exfat* ef, const char* path)
{
	int rc;
	struct exfat_node* node;

	rc = create(ef, path, EXFAT_ATTRIB_DIR);
	if (rc != 0)
		return rc;
	rc = exfat_lookup(ef, &node, path);
	if (rc != 0)
		return 0;
	/* directories always have at least one cluster */
	rc = exfat_truncate(ef, node, CLUSTER_SIZE(*ef->sb), true);
	if (rc != 0)
	{
		delete(ef, node);
		exfat_put_node(ef, node);
		return rc;
	}
	rc = exfat_flush_node(ef, node);
	if (rc != 0)
	{
		delete(ef, node);
		exfat_put_node(ef, node);
		return rc;
	}
	exfat_put_node(ef, node);
	return 0;
}

static int rename_entry(struct exfat* ef, struct exfat_node* dir,
		struct exfat_node* node, const le16_t* name, cluster_t new_cluster,
		off_t new_offset)
{
	struct exfat_entry_meta1 meta1;
	struct exfat_entry_meta2 meta2;
	cluster_t old_cluster = node->entry_cluster;
	off_t old_offset = node->entry_offset;
	const size_t name_length = utf16_length(name);
	const int name_entries = DIV_ROUND_UP(name_length, EXFAT_ENAME_MAX);
	int i;

	if (exfat_pread(ef->dev, &meta1, sizeof(meta1),
			co2o(ef, old_cluster, old_offset)) < 0)
	{
		exfat_error("failed to read meta1 entry on rename");
		return -EIO;
	}
	if (!next_entry(ef, node->parent, &old_cluster, &old_offset))
		return -EIO;
	if (exfat_pread(ef->dev, &meta2, sizeof(meta2),
			co2o(ef, old_cluster, old_offset)) < 0)
	{
		exfat_error("failed to read meta2 entry on rename");
		return -EIO;
	}
	meta1.continuations = 1 + name_entries;
	meta2.name_hash = exfat_calc_name_hash(ef, name);
	meta2.name_length = name_length;
	meta1.checksum = exfat_calc_checksum(&meta1, &meta2, name);

	if (!erase_entry(ef, node))
		return -EIO;

	node->entry_cluster = new_cluster;
	node->entry_offset = new_offset;

	if (exfat_pwrite(ef->dev, &meta1, sizeof(meta1),
			co2o(ef, new_cluster, new_offset)) < 0)
	{
		exfat_error("failed to write meta1 entry on rename");
		return -EIO;
	}
	if (!next_entry(ef, dir, &new_cluster, &new_offset))
		return -EIO;
	if (exfat_pwrite(ef->dev, &meta2, sizeof(meta2),
			co2o(ef, new_cluster, new_offset)) < 0)
	{
		exfat_error("failed to write meta2 entry on rename");
		return -EIO;
	}

	for (i = 0; i < name_entries; i++)
	{
		struct exfat_entry_name name_entry = {EXFAT_ENTRY_FILE_NAME, 0};
		memcpy(name_entry.name, name + i * EXFAT_ENAME_MAX,
				EXFAT_ENAME_MAX * sizeof(le16_t));
		if (!next_entry(ef, dir, &new_cluster, &new_offset))
			return -EIO;
		if (exfat_pwrite(ef->dev, &name_entry, sizeof(name_entry),
				co2o(ef, new_cluster, new_offset)) < 0)
		{
			exfat_error("failed to write name entry on rename");
			return -EIO;
		}
	}

	memcpy(node->name, name, (EXFAT_NAME_MAX + 1) * sizeof(le16_t));
	tree_detach(node);
	tree_attach(dir, node);
	return 0;
}

int exfat_rename(struct exfat* ef, const char* old_path, const char* new_path)
{
	struct exfat_node* node;
	struct exfat_node* existing;
	struct exfat_node* dir;
	cluster_t cluster = EXFAT_CLUSTER_BAD;
	off_t offset = -1;
	le16_t name[EXFAT_NAME_MAX + 1];
	int rc;

	rc = exfat_lookup(ef, &node, old_path);
	if (rc != 0)
		return rc;

	rc = exfat_split(ef, &dir, &existing, name, new_path);
	if (rc != 0)
	{
		exfat_put_node(ef, node);
		return rc;
	}

	/* check that target is not a subdirectory of the source */
	if (node->flags & EXFAT_ATTRIB_DIR)
	{
		struct exfat_node* p;

		for (p = dir; p; p = p->parent)
			if (node == p)
			{
				if (existing != NULL)
					exfat_put_node(ef, existing);
				exfat_put_node(ef, dir);
				exfat_put_node(ef, node);
				return -EINVAL;
			}
	}

	if (existing != NULL)
	{
		/* remove target if it's not the same node as source */
		if (existing != node)
		{
			if (existing->flags & EXFAT_ATTRIB_DIR)
			{
				if (node->flags & EXFAT_ATTRIB_DIR)
					rc = exfat_rmdir(ef, existing);
				else
					rc = -ENOTDIR;
			}
			else
			{
				if (!(node->flags & EXFAT_ATTRIB_DIR))
					rc = exfat_unlink(ef, existing);
				else
					rc = -EISDIR;
			}
			exfat_put_node(ef, existing);
			if (rc != 0)
			{
				exfat_put_node(ef, dir);
				exfat_put_node(ef, node);
				return rc;
			}
		}
		else
			exfat_put_node(ef, existing);
	}

	rc = find_slot(ef, dir, &cluster, &offset,
			2 + DIV_ROUND_UP(utf16_length(name), EXFAT_ENAME_MAX));
	if (rc != 0)
	{
		exfat_put_node(ef, dir);
		exfat_put_node(ef, node);
		return rc;
	}
	rc = rename_entry(ef, dir, node, name, cluster, offset);
	exfat_put_node(ef, dir);
	exfat_put_node(ef, node);
	return rc;
}

void exfat_utimes(struct exfat_node* node, const struct timespec tv[2])
{
	node->atime = tv[0].tv_sec;
	node->mtime = tv[1].tv_sec;
	node->flags |= EXFAT_ATTRIB_DIRTY;
}

void exfat_update_atime(struct exfat_node* node)
{
	node->atime = time(NULL);
	node->flags |= EXFAT_ATTRIB_DIRTY;
}

void exfat_update_mtime(struct exfat_node* node)
{
	node->mtime = time(NULL);
	node->flags |= EXFAT_ATTRIB_DIRTY;
}

const char* exfat_get_label(struct exfat* ef)
{
	return ef->label;
}

static int find_label(struct exfat* ef, cluster_t* cluster, off_t* offset)
{
	struct iterator it;
	int rc;

	rc = opendir(ef, ef->root, &it);
	if (rc != 0)
		return rc;

	for (;;)
	{
		if (it.offset >= ef->root->size)
		{
			closedir(&it);
			return -ENOENT;
		}

		if (get_entry_ptr(ef, &it)->type == EXFAT_ENTRY_LABEL)
		{
			*cluster = it.cluster;
			*offset = it.offset;
			closedir(&it);
			return 0;
		}

		if (!fetch_next_entry(ef, ef->root, &it))
		{
			closedir(&it);
			return -EIO;
		}
	}
}

int exfat_set_label(struct exfat* ef, const char* label)
{
	le16_t label_utf16[EXFAT_ENAME_MAX + 1];
	int rc;
	cluster_t cluster;
	off_t offset;
	struct exfat_entry_label entry;

	memset(label_utf16, 0, sizeof(label_utf16));
	rc = utf8_to_utf16(label_utf16, label, EXFAT_ENAME_MAX, strlen(label));
	if (rc != 0)
		return rc;

	rc = find_label(ef, &cluster, &offset);
	if (rc == -ENOENT)
		rc = find_slot(ef, ef->root, &cluster, &offset, 1);
	if (rc != 0)
		return rc;

	entry.type = EXFAT_ENTRY_LABEL;
	entry.length = utf16_length(label_utf16);
	memcpy(entry.name, label_utf16, sizeof(entry.name));
	if (entry.length == 0)
		entry.type ^= EXFAT_ENTRY_VALID;

	if (exfat_pwrite(ef->dev, &entry, sizeof(struct exfat_entry_label),
			co2o(ef, cluster, offset)) < 0)
	{
		exfat_error("failed to write label entry");
		return -EIO;
	}
	strcpy(ef->label, label);
	return 0;
}