/*
 * Copyright (c) 2013, 2014, 2015, 2018, 2023 Jonas 'Sortie' Termansen.
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 *
 * inode.cpp
 * Filesystem inode.
 */

#define __STDC_CONSTANT_MACROS
#define __STDC_LIMIT_MACROS

#include <sys/stat.h>
#include <sys/types.h>

#include <assert.h>
#include <errno.h>
#include <endian.h>
#include <fcntl.h>
#include <limits.h>
#include <pthread.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <time.h>

#include "ext-constants.h"
#include "ext-structs.h"
#include "fat.h"

#include "block.h"
#include "blockgroup.h"
#include "device.h"
#include "fatfs.h"
#include "filesystem.h"
#include "inode.h"
#include "util.h"

#ifndef S_SETABLE
#define S_SETABLE 02777
#endif
#ifndef O_WRITE
#define O_WRITE (O_WRONLY | O_RDWR)
#endif

Inode::Inode(Filesystem* filesystem, uint32_t inode_id)
{
	this->prev_inode = NULL;
	this->next_inode = NULL;
	this->prev_hashed = NULL;
	this->next_hashed = NULL;
	this->prev_dirty = NULL;
	this->next_dirty = NULL;
	this->data_block = NULL;
	this->data = NULL;
	this->filesystem = filesystem;
	this->reference_count = 1;
	this->remote_reference_count = 0;
	this->inode_id = inode_id;
	this->dirty = false;
}

Inode::~Inode()
{
	Sync();
	if ( data_block )
		data_block->Unref();
	Unlink();
}

uint32_t Inode::Mode()
{
	if ( inode_id == filesystem->root_inode_id )
		return filesystem->mode_dir;
	mode_t mode = dirent->attributes & FAT_ATTRIBUTE_DIRECTORY ?
	              filesystem->mode_dir : filesystem->mode_reg;
	if ( dirent->attributes & FAT_ATTRIBUTE_READ_ONLY )
		mode &= ~0222;
	return mode;
}

void Inode::SetMode(uint32_t mode)
{
	(void) mode;
	assert(filesystem->device->write);
	assert(false); // TODO: Unsupported.
}

uint32_t Inode::UserId()
{
	return filesystem->uid;
}

void Inode::SetUserId(uint32_t user)
{
	(void) user;
	assert(filesystem->device->write);
	assert(false); // TODO: Unsupported.
}

uint32_t Inode::GroupId()
{
	return filesystem->gid;
}

void Inode::SetGroupId(uint32_t group)
{
	(void) group;
	assert(filesystem->device->write);
	assert(false); // TODO: Unsupported.
}

uint64_t Inode::Size()
{
	if ( inode_id == filesystem->root_inode_id )
		return 0;
	if ( dirent->attributes & FAT_ATTRIBUTE_DIRECTORY )
		return 0;
	return dirent->size;
}

void Inode::SetSize(uint64_t new_size)
{
	assert(filesystem->device->write);
	assert(false); // TODO: Not implemented.
	bool largefile = filesystem->sb->s_feature_ro_compat &
	                 FAT_FEATURE_RO_COMPAT_LARGE_FILE;
	uint32_t lower = new_size & 0xFFFFFFFF;
	uint32_t upper = new_size >> 32;
	// TODO: Enforce filesize limit with no largefile support or non-files.
	data->i_size = lower;

	// TODO: Figure out these i_blocks semantics and how stuff is reserved.
	const uint64_t ENTRIES = filesystem->block_size / sizeof(uint32_t);
	uint64_t block_direct = sizeof(data->i_block) / sizeof(uint32_t) - 3;
	uint64_t block_singly = ENTRIES;
	uint64_t block_doubly = block_singly * block_singly;
	uint64_t max_direct = block_direct;
	uint64_t max_singly = max_direct + block_singly;
	uint64_t max_doubly = max_singly + block_doubly;
	uint64_t logical_blocks = divup(new_size, (uint64_t) filesystem->block_size);
	uint64_t actual_blocks  = logical_blocks;
	if ( max_direct <= logical_blocks )
		actual_blocks += divup(logical_blocks - max_direct, ENTRIES);
	if ( max_singly <= logical_blocks )
		actual_blocks += divup(logical_blocks - max_singly, ENTRIES * ENTRIES);
	if ( max_doubly <= logical_blocks )
		actual_blocks += divup(logical_blocks - max_doubly, ENTRIES * ENTRIES * ENTRIES);

	BeginWrite();
	data->i_blocks = (actual_blocks * filesystem->block_size) / 512;
	if ( FAT_S_ISREG(data->i_mode) && largefile )
		data->i_dir_acl = upper;
	FinishWrite();

	Modified();
}

Block* Inode::GetBlockFromTable(Block* table, uint32_t index)
{
	if ( uint32_t block_id = ((uint32_t*) table->block_data)[index] )
		return filesystem->device->GetBlock(block_id);
	// TODO: If in read only mode, then perhaps return a zero block here.
	if ( !filesystem->device->write )
		return NULL;
	uint32_t group_id = (inode_id - 1) / filesystem->sb->s_inodes_per_group;
	assert(group_id < filesystem->num_groups);
	BlockGroup* block_group = filesystem->GetBlockGroup(group_id);
	uint32_t block_id = filesystem->AllocateBlock(block_group);
	block_group->Unref();
	if ( block_id )
	{
		Block* block = filesystem->device->GetBlockZeroed(block_id);
		table->BeginWrite();
		((uint32_t*) table->block_data)[index] = block_id;
		table->FinishWrite();
		return block;
	}
	return NULL;
}

Block* Inode::GetBlock(uint64_t offset)
{
	const uint64_t ENTRIES = filesystem->block_size / sizeof(uint32_t);
	uint64_t block_direct = sizeof(data->i_block) / sizeof(uint32_t) - 3;
	uint64_t block_singly = ENTRIES;
	uint64_t block_doubly = block_singly * block_singly;
	uint64_t block_triply = block_doubly * block_singly;
	uint64_t max_direct = block_direct;
	uint64_t max_singly = max_direct + block_singly;
	uint64_t max_doubly = max_singly + block_doubly;
	uint64_t max_triply = max_doubly + block_triply;
	uint32_t index;

	Block* table = data_block; table->Refer();
	Block* block;

	uint32_t inode_offset = (uintptr_t) data - (uintptr_t) data_block->block_data;
	uint32_t table_offset = offsetof(struct ext_inode, i_block);
	uint32_t inode_block_table_offset = (inode_offset + table_offset) / sizeof(uint32_t);

	// TODO: It would seem that somebody needs a lesson in induction. :-)
	if ( offset < max_direct )
	{
		offset -= 0;
		offset += inode_block_table_offset * 1;
	read_direct:
		index = offset;
		offset %= 1;
		block = GetBlockFromTable(table, index);
		table->Unref();
		if ( !block )
			return NULL;
		return block;
	}
	else if ( offset < max_singly )
	{
		offset -= max_direct;
		offset += (inode_block_table_offset + 12) * ENTRIES;
	read_singly:
		index = offset / ENTRIES;
		offset = offset % ENTRIES;
		block = GetBlockFromTable(table, index);
		table->Unref();
		if ( !block )
			return NULL;
		table = block;
		goto read_direct;
	}
	else if ( offset < max_doubly )
	{
		offset -= max_singly;
		offset += (inode_block_table_offset + 13) * ENTRIES * ENTRIES;
	read_doubly:
		index = offset / (ENTRIES * ENTRIES);
		offset = offset % (ENTRIES * ENTRIES);
		block = GetBlockFromTable(table, index);
		table->Unref();
		if ( !block )
			return NULL;
		table = block;
		goto read_singly;
	}
	else if ( offset < max_triply )
	{
		offset -= max_doubly;
		offset += (inode_block_table_offset + 14) * ENTRIES * ENTRIES * ENTRIES;
	/*read_triply:*/
		index = offset / (ENTRIES * ENTRIES * ENTRIES);
		offset = offset % (ENTRIES * ENTRIES * ENTRIES);
		block = GetBlockFromTable(table, index);
		table->Unref();
		if ( !block )
			return NULL;
		table = block;
		goto read_doubly;
	}

	table->Unref();
	return NULL;
}

Block* Inode::GetClusterSector(uint32_t cluster, uint8_t sector)
{
	uint32_t block_id;
	if ( inode_id == filesystem->root_inode_id && filesystem->fat_type != 32 )
		block_id = filesystem->root_sector + cluster;
	else
		block_id = filesystem->data_sector +
		           (cluster - 2) * filesystem->bpb->sectors_per_cluster +
		           sector;
	return filesystem->device->GetBlock(block_id);
}

// TODO: Yes. Do review this function carefully.
bool Inode::Iterate(Block** block_ptr, uint32_t* cluster_ptr,
                    uint8_t* sector_ptr, uint16_t* offset_ptr)
{
	// TODO: Restructure to cache this.
	if ( *block_ptr )
	{
		(*block_ptr)->Unref();
		*block_ptr = NULL;
	}
	if ( *offset_ptr == filesystem->bytes_per_sector )
	{
		*offset_ptr = 0;
		if ( inode_id == filesystem->root_inode_id &&
		     filesystem->fat_type != 32 )
		{
			// TODO: This kinda assumes the root directory is sector sized.
			uint32_t end = filesystem->root_dirent_count *
			               sizeof(struct fat_dirent);
			uint32_t end_lba = end / filesystem->bytes_per_sector;
			if ( end_lba <= *cluster_ptr )
				return errno = 0, false;
			(*cluster_ptr)++;
		}
		else
		{
			(*sector_ptr)++;
			if ( *sector_ptr == filesystem->bpb->sectors_per_cluster )
			{
				*sector_ptr = 0;
				*cluster_ptr = filesystem->ReadFAT(*cluster_ptr);
			}
		}
	}
	if ( inode_id != filesystem->root_inode_id || filesystem->fat_type == 32 )
	{
		if ( *cluster_ptr < 2 )
			return errno = EIO, false;
		if ( filesystem->eof_cluster <= *cluster_ptr )
			return errno = 0, false;
		if ( filesystem->eio_cluster <= *cluster_ptr )
			return errno = EIO, false;
	}
	if ( !(*block_ptr = GetClusterSector(*cluster_ptr, *sector_ptr)) )
		return false;
	return errno = 0, true;
}

uint32_t Inode::SeekCluster(uint32_t cluster_id)
{
	// TODO: Cache.
	uint32_t cluster = first_cluster;
	while ( cluster_id-- )
	{
		cluster = filesystem->ReadFAT(cluster);
		if ( cluster < 2 )
			return errno = EIO, filesystem->eio_cluster;
		if ( filesystem->eof_cluster <= cluster )
			return errno = EIO, filesystem->eio_cluster;
	}
	return cluster;
}

bool Inode::FreeIndirect(uint64_t from, uint64_t offset, uint32_t block_id,
                         int indirection, uint64_t entry_span)
{
	assert(filesystem->device->write);
	const uint64_t ENTRIES = filesystem->block_size / sizeof(uint32_t);
	Block* block = filesystem->device->GetBlock(block_id);
	uint32_t* table = (uint32_t*) block->block_data;
	bool any_children = false;
	bool begun_writing = false;
	for ( uint64_t i = 0; i < ENTRIES; i++ )
	{
		if ( !table[i] )
			continue;
		uint64_t entry_offset = offset + entry_span * i;
		if ( entry_offset < from ||
		     (indirection &&
		      FreeIndirect(from, entry_offset, table[i], indirection-1,
		                   entry_span / ENTRIES)) )
		{
			any_children = true;
			continue;
		}
		filesystem->FreeBlock(table[i]);
		if ( !begun_writing )
		{
			block->BeginWrite();
			begun_writing = true;
		}
		table[i] = 0;
	}
	if ( begun_writing )
		block->FinishWrite();
	return any_children;
}

bool Inode::Truncate(uint64_t new_size_64)
{
	assert(filesystem->device->write);
	assert(S_ISREG(Mode()));
	uint32_t new_size = (uint32_t) new_size_64;
	if ( new_size_64 != new_size )
		return errno = E2BIG, false;
	uint32_t old_size = dirent->size;
	uint32_t pos = old_size < new_size ? old_size : new_size;
	uint32_t bytes_per_sector = filesystem->bytes_per_sector;
	uint32_t cluster_id = pos / filesystem->cluster_size;
	uint32_t cluster_offset = pos % filesystem->cluster_size;
	if ( cluster_id && !cluster_offset )
	{
		cluster_id--;
		cluster_offset = filesystem->cluster_size;
	}
	uint32_t cluster = SeekCluster(cluster_id);
	if ( cluster_id == filesystem->eio_cluster )
		return errno = EIO, false;
	if ( old_size < new_size )
	{
		while ( old_size < new_size )
		{
			if ( cluster_offset == filesystem->cluster_size )
			{
				// TODO: Zero the new sectors since the old contents may leak
				//       if we were to implement mmap.
				uint32_t next_cluster = filesystem->AllocateCluster();
				if ( !next_cluster )
					return false;
				filesystem->WriteFAT(next_cluster, filesystem->eof_cluster);
				filesystem->WriteFAT(cluster, next_cluster);
				cluster_offset = 0;
				cluster = next_cluster;
			}
			uint8_t sector = cluster_offset / bytes_per_sector;
			uint16_t sector_offset = cluster_offset % bytes_per_sector;
			Block* block = GetClusterSector(cluster, sector);
			if ( !block )
				return false;
			size_t left = new_size - old_size;
			size_t available = bytes_per_sector - sector_offset;
			size_t amount = left < available ? left : available;
			block->BeginWrite();
			memset(block->block_data + sector_offset, 0, amount);
			block->FinishWrite();
			old_size += amount;
			cluster_offset += amount;
			block->Unref();
		}
	}
	else if ( new_size < old_size )
	{
		uint32_t marker = filesystem->eof_cluster;
		while ( true )
		{
			uint32_t next_cluster = filesystem->ReadFAT(cluster);
			if ( next_cluster < 2 || filesystem->eio_cluster == next_cluster )
				return errno = EIO, false;
			if ( next_cluster != marker )
			{
				filesystem->WriteFAT(cluster, marker);
				filesystem->FreeCluster(next_cluster);
			}
			if ( filesystem->eof_cluster <= next_cluster )
				break;
			cluster = next_cluster;
			cluster_id++;
			marker = 0;
		}
	}
	else
		return true;
	data_block->BeginWrite();
	dirent->size = new_size;
	data_block->FinishWrite();
	return true;
}

Inode* Inode::Open(const char* elem, int flags, mode_t mode)
{
	if ( !S_ISDIR(Mode()) )
		return errno = ENOTDIR, (Inode*) NULL;
	size_t elem_length = strlen(elem);
	if ( elem_length == 0 )
		return errno = ENOENT, (Inode*) NULL;
	if ( inode_id == filesystem->root_inode_id )
	{
		if ( !strcmp(elem, ".") || !strcmp(elem, "..") )
		{
			if ( (flags & O_CREAT) && (flags & O_EXCL) )
				return errno = EEXIST, (Inode*) NULL;
			if ( flags & O_WRITE && !filesystem->device->write )
				return errno = EROFS, (Inode*) NULL;
			Refer();
			return this;
			// TODO: Reopen the same inode.
		}
	}
	uint32_t cluster = first_cluster;
	uint8_t sector = 0;
	uint16_t offset = 0;
	Block* block = NULL;
	bool found_free = false;
	uint32_t free_cluster = 0;
	uint8_t free_sector = 0;
	uint16_t free_offset = 0;
	uint32_t last_cluster = 0;
	while ( Iterate(&block, &cluster, &sector, &offset) )
	{
		last_cluster = cluster;
		uint8_t* block_data = block->block_data + offset;
		struct fat_dirent* entry = (struct fat_dirent*) block_data;
		if ( !found_free &&
		     (!entry->name[0] || (unsigned char) entry->name[0] == 0xE5) )
		{
			found_free = true;
			free_cluster = cluster;
			free_sector = sector;
			free_offset = offset;
		}
		if ( !entry->name[0] )
			break;
		char name[8 + 1 + 3 + 1];
		if ( (unsigned char) entry->name[0] != 0xE5 &&
		     !(entry->attributes & FAT_ATTRIBUTE_VOLUME_ID) &&
		     (decode_8_3(entry->name, name), !strcmp(elem, name)) )
		{
			// TODO: Opening .. ends up with EIO failures.
			if ( (flags & O_CREAT) && (flags & O_EXCL) )
				return block->Unref(), errno = EEXIST, (Inode*) NULL;
			if ( (flags & O_DIRECTORY) &&
			     !(entry->attributes & FAT_ATTRIBUTE_DIRECTORY) )
				return block->Unref(), errno = ENOTDIR, (Inode*) NULL;
			uint32_t inode_id = entry->cluster_low | entry->cluster_high << 16;
			// TODO: If the inode is a directory, keep a reference open to this
			// parent directory so it can find the .. path back and keep track
			// of where the directory records with metadata is.
			Inode* inode = filesystem->GetInode(inode_id, block, entry);
			block->Unref();
			if ( !inode )
				return (Inode*) NULL;
			if ( flags & O_WRITE && !filesystem->device->write )
				return inode->Unref(), errno = EROFS, (Inode*) NULL;
			if ( S_ISREG(inode->Mode()) && (flags & O_WRITE) &&
			    (flags & O_TRUNC) && !inode->Truncate(0) )
				return (Inode*) NULL;
			return inode;
		}
		offset += sizeof(struct fat_dirent);
	}
	if ( block )
		block->Unref();
	if ( errno )
		return (Inode*) NULL;
	if ( flags & O_CREAT )
	{
		if ( !filesystem->device->write )
			return errno = EROFS, (Inode*) NULL;
		// TODO: Protect against . and ..
		if ( !is_8_3(elem) )
			return errno = ENAMETOOLONG, (Inode*) NULL;
		// TODO: Root directory support.
		if ( inode_id == filesystem->root_inode_id &&
		     filesystem->fat_type != 32 )
			return errno = ENOTSUP, (Inode*) NULL;
		if ( !found_free )
		{
			uint32_t new_cluster = filesystem->AllocateCluster();
			if ( !new_cluster )
				return (Inode*) NULL;
			for ( size_t i = 0; i < filesystem->bpb->sectors_per_cluster; i++ )
			{
				Block* block = GetClusterSector(new_cluster, i);
				if ( !block )
				{
					filesystem->FreeCluster(new_cluster);
					return (Inode*) NULL;
				}
				block->BeginWrite();
				memset(block->block_data, 0, filesystem->bytes_per_sector);
				block->FinishWrite();
				block->Unref();
			}
			filesystem->WriteFAT(last_cluster, new_cluster);
			found_free = true;
			free_cluster = new_cluster;
			free_sector = 0;
			free_offset = 0;
		}
		// TODO: Avoid shadowing with the class member.
		uint32_t inode_id = filesystem->AllocateCluster();
		// TODO: Actually zero this cluster entirely.
		if ( !inode_id )
			return (Inode*) NULL;
		mode_t attributes = mode & 0200 ? 0 : FAT_ATTRIBUTE_READ_ONLY;
		if ( S_ISDIR(mode) )
		{
			attributes |= FAT_ATTRIBUTE_DIRECTORY;
			Block* block = GetClusterSector(inode_id, 0);
			if ( !block )
				return filesystem->FreeCluster(inode_id), (Inode*) NULL;
			block->BeginWrite();
			memset(block->block_data, 0, filesystem->bytes_per_sector);
			struct fat_dirent* dirent = (struct fat_dirent*) block->block_data;
			// TODO: Mirror modified times in here.
			memcpy(dirent->name, ".", 1);
			dirent->attributes = attributes;
			dirent->cluster_high = htole16(inode_id >> 16);
			dirent->cluster_low = htole16(inode_id & 0xFFFF);
			dirent++;
			memcpy(dirent->name, "..", 2);
			dirent->attributes = FAT_ATTRIBUTE_DIRECTORY;
			if ( this->inode_id == filesystem->root_inode_id )
			{
				dirent->cluster_high = htole16(0);
				dirent->cluster_low = htole16(0);
			}
			else
			{
				dirent->cluster_high = htole16(this->inode_id >> 16);
				dirent->cluster_low = htole16(this->inode_id & 0xFFFF);
			}
			block->FinishWrite();
			block->Unref();
		}
		Block* block = GetClusterSector(free_cluster, free_sector);
		if ( !block )
			return filesystem->FreeCluster(inode_id), (Inode*) NULL;
		filesystem->WriteFAT(inode_id, filesystem->eof_cluster);
		struct timespec now;
		clock_gettime(CLOCK_REALTIME, &now);
		block->BeginWrite();
		struct fat_dirent* dirent =
			(struct fat_dirent*) (block->block_data + free_offset);
		encode_8_3(elem, dirent->name);
		dirent->attributes = attributes;
		dirent->creation_tenths = timespec_to_fat_tenths(&now);
		dirent->creation_time = htole16(timespec_to_fat_time(&now));
		dirent->creation_date = htole16(timespec_to_fat_date(&now));
		dirent->access_date = dirent->creation_date;
		dirent->cluster_high = htole16(inode_id >> 16);
		dirent->modified_time = dirent->creation_time;
		dirent->modified_date = dirent->creation_date;
		dirent->cluster_low = htole16(inode_id & 0xFFFF);
		dirent->size = htole16(0);
		block->FinishWrite();
		Inode* inode = filesystem->GetInode(inode_id, block, dirent);
		block->Unref();
		return inode;
	}
	return errno = ENOENT, (Inode*) NULL;
}

bool Inode::Link(const char* elem, Inode* dest, bool directories)
{
	if ( !FAT_S_ISDIR(Mode()) )
		return errno = ENOTDIR, false;
	if ( directories && !FAT_S_ISDIR(dest->Mode()) )
		return errno = ENOTDIR, false;
	if ( !directories && FAT_S_ISDIR(dest->Mode()) )
		return errno = EISDIR, false;

	// Search for a hole in which we can store the new directory entry and stop
	// if we meet an existing link with the requested name.
	size_t elem_length = strlen(elem);
	if ( elem_length == 0 )
		return errno = ENOENT, false;
	size_t new_entry_size = roundup(sizeof(struct ext_dirent) + elem_length, (size_t) 4);
	uint64_t filesize = Size();
	uint64_t offset = 0;
	Block* block = NULL;
	uint64_t block_id = 0;
	bool found_hole = false;
	bool splitting = false;
	uint64_t hole_block_id = 0;
	uint64_t hole_block_offset = 0;
	while ( offset < filesize )
	{
		uint64_t entry_block_id = offset / filesystem->block_size;
		uint64_t entry_block_offset = offset % filesystem->block_size;
		if ( block && block_id != entry_block_id )
			block->Unref(),
			block = NULL;
		if ( !block && !(block = GetBlock(block_id = entry_block_id)) )
			return NULL;
		const uint8_t* block_data = block->block_data + entry_block_offset;
		const struct ext_dirent* entry = (const struct ext_dirent*) block_data;
		if ( entry->name_len == elem_length &&
		     memcmp(elem, entry->name, elem_length) == 0 )
		{
			block->Unref();
			return errno = EEXIST, false;
		}
		if ( !found_hole )
		{
			size_t entry_size = roundup(sizeof(struct ext_dirent) + entry->name_len, (size_t) 4);
			if ( (!entry->inode || !entry->name[0]) && new_entry_size <= entry->reclen )
			{
				hole_block_id = entry_block_id;
				hole_block_offset = entry_block_offset;
				new_entry_size = entry->reclen;
				found_hole = true;
			}
			else if ( new_entry_size <= entry->reclen - entry_size )
			{
				hole_block_id = entry_block_id;
				hole_block_offset = entry_block_offset;
				new_entry_size = entry->reclen - entry_size;
				splitting = true;
				found_hole = true;
			}
		}
		offset += entry->reclen;
	}

	if ( !filesystem->device->write )
		return errno = EROFS, false;

	if ( UINT16_MAX <= dest->data->i_links_count )
		return errno = EMLINK, false;

	if ( 255 < elem_length )
		return errno = ENAMETOOLONG, false;

	// We'll append another block if we failed to find a suitable hole.
	if ( !found_hole )
	{
		hole_block_id = filesize / filesystem->block_size;
		hole_block_offset = filesize % filesystem->block_size;
		new_entry_size = filesystem->block_size;
	}

	if ( block && block_id != hole_block_id )
		block->Unref(),
		block = NULL;
	if ( !block && !(block = GetBlock(block_id = hole_block_id)) )
		return NULL;

	Modified();

	block->BeginWrite();

	uint8_t* block_data = block->block_data + hole_block_offset;
	struct ext_dirent* entry = (struct ext_dirent*) block_data;

	// If we found a directory entry with room at the end, split it!
	if ( splitting )
	{
		entry->reclen = roundup(sizeof(struct ext_dirent) + entry->name_len, (size_t) 4);
		assert(entry->reclen);
		block_data += entry->reclen;
		entry = (struct ext_dirent*) block_data;
	}

	// Write the new directory entry.
	entry->inode = dest->inode_id;
	entry->reclen = new_entry_size;
	entry->name_len = elem_length;
	if ( filesystem->sb->s_feature_incompat & FAT_FEATURE_INCOMPAT_FILETYPE )
		entry->file_type = FAT_FT_OF_MODE(dest->Mode());
	else
		entry->file_type = FAT_FT_UNKNOWN;
	strncpy(entry->name, elem, new_entry_size - sizeof(struct ext_dirent));

	block->FinishWrite();
	block->Unref();

	dest->BeginWrite();
	dest->data->i_links_count++;
	dest->FinishWrite();

	if ( !found_hole )
		SetSize(Size() + filesystem->block_size);

	return true;
}

Inode* Inode::UnlinkKeep(const char* elem, bool directories, bool force)
{
	if ( !FAT_S_ISDIR(Mode()) )
		return errno = ENOTDIR, (Inode*) NULL;
	size_t elem_length = strlen(elem);
	if ( elem_length == 0 )
		return errno = ENOENT, (Inode*) NULL;
	uint32_t block_size = filesystem->block_size;
	uint64_t filesize = Size();
	uint64_t num_blocks = divup(filesize, (uint64_t) block_size);
	uint64_t offset = 0;
	Block* block = NULL;
	uint64_t block_id = 0;
	struct ext_dirent* last_entry = NULL;
	while ( offset < filesize )
	{
		uint64_t entry_block_id = offset / block_size;
		uint64_t entry_block_offset = offset % block_size;
		if ( block && block_id != entry_block_id )
			last_entry = NULL,
			block->Unref(),
			block = NULL;
		if ( !block && !(block = GetBlock(block_id = entry_block_id)) )
			return NULL;
		uint8_t* block_data = block->block_data + entry_block_offset;
		struct ext_dirent* entry = (struct ext_dirent*) block_data;
		if ( entry->inode &&
		     entry->name_len == elem_length &&
		     memcmp(elem, entry->name, elem_length) == 0 )
		{
			Inode* inode = filesystem->GetInode(entry->inode);

			if ( !force && directories && !FAT_S_ISDIR(inode->Mode()) )
			{
				inode->Unref();
				block->Unref();
				return errno = ENOTDIR, (Inode*) NULL;
			}

			if ( !force && directories && !inode->IsEmptyDirectory() )
			{
				inode->Unref();
				block->Unref();
				return errno = ENOTEMPTY, (Inode*) NULL;
			}

			if ( !force && !directories && FAT_S_ISDIR(inode->Mode()) )
			{
				inode->Unref();
				block->Unref();
				return errno = EISDIR, (Inode*) NULL;
			}

			if ( !filesystem->device->write )
			{
				inode->Unref();
				block->Unref();
				return errno = EROFS, (Inode*) NULL;
			}

			Modified();

			inode->BeginWrite();
			inode->data->i_links_count--;
			inode->FinishWrite();

			block->BeginWrite();

			entry->inode = 0;
			entry->name_len = 0;
			entry->file_type = 0;

			// Merge the current entry with the previous if any.
			if ( last_entry )
			{
				last_entry->reclen += entry->reclen;
				memset(entry, 0, entry->reclen);
				entry = last_entry;
			}

			strncpy(entry->name + entry->name_len, "",
			        entry->reclen - sizeof(struct ext_dirent) - entry->name_len);

			// If the entire block is empty, we'll need to remove it.
			if ( !entry->name[0] && entry->reclen == block_size )
			{
				// If this is not the last block, we'll make it. This is faster
				// than shifting the entire directory a single block. We don't
				// actually copy this block to the end, since we'll truncate it
				// regardless.
				if ( entry_block_id + 1 != num_blocks )
				{
					Block* last_block = GetBlock(num_blocks-1);
					if ( last_block )
					{
						memcpy(block->block_data, last_block->block_data, block_size);
						last_block->Unref();
						Truncate(filesize - block_size);
					}
				}
				else
				{
					Truncate(filesize - block_size);
				}
			}

			block->FinishWrite();

			block->Unref();

			return inode;
		}
		offset += entry->reclen;
		last_entry = entry;
	}
	if ( block )
		block->Unref();
	return errno = ENOENT, (Inode*) NULL;
}

bool Inode::Unlink(const char* elem, bool directories, bool force)
{
	Inode* result = UnlinkKeep(elem, directories, force);
	if ( !result )
		return false;
	result->Unref();
	return true;
}

ssize_t Inode::ReadAt(uint8_t* buf, size_t s_count, off_t o_offset)
{
	if ( !FAT_S_ISREG(Mode()) )
		return errno = EISDIR, -1;
	if ( o_offset < 0 )
		return errno = EINVAL, -1;
	if ( SSIZE_MAX < s_count )
		s_count = SSIZE_MAX;
	// TODO: Downgrade to 32-bit.
	uint64_t sofar = 0;
	uint64_t count = (uint64_t) s_count;
	uint64_t offset = (uint64_t) o_offset;
	uint32_t file_size = Size();
	if ( file_size <= offset )
		return 0;
	if ( file_size - offset < count )
		count = file_size - offset;
	if ( !count )
		return 0;
	uint32_t cluster_id = offset / filesystem->cluster_size;
	uint32_t cluster_offset = offset % filesystem->cluster_size;
	uint32_t cluster = SeekCluster(cluster_id);
	if ( filesystem->eio_cluster <= cluster )
		return -1;
	while ( sofar < count )
	{
		if ( filesystem->cluster_size <= cluster_offset )
		{
			cluster = filesystem->ReadFAT(cluster);
			// TODO: Better checks.
			if ( filesystem->eio_cluster <= cluster )
				return sofar ? sofar : (errno = EIO, -1);
			cluster_offset = 0;
			cluster_id++;
		}
		uint8_t sector = cluster_offset / filesystem->bytes_per_sector;
		uint16_t block_offset = cluster_offset % filesystem->bytes_per_sector;
		uint32_t block_left = filesystem->bytes_per_sector - block_offset;
		Block* block = GetClusterSector(cluster, sector);
		if ( !block )
			return sofar ? sofar : -1;
		size_t amount = count - sofar < block_left ? count - sofar : block_left;
		memcpy(buf + sofar, block->block_data + block_offset, amount);
		sofar += amount;
		cluster_offset += amount;
		block->Unref();
	}
	return (ssize_t) sofar;
}

ssize_t Inode::WriteAt(const uint8_t* buf, size_t s_count, off_t o_offset)
{
	if ( !FAT_S_ISREG(Mode()) )
		return errno = EISDIR, -1;
	if ( o_offset < 0 )
		return errno = EINVAL, -1;
	if ( !filesystem->device->write )
		return errno = EROFS, -1;
	if ( SSIZE_MAX < s_count )
		s_count = SSIZE_MAX;
	// TODO: Update modified time.
	//Modified();
	// TODO: Downgrade to 32-bit.
	uint64_t sofar = 0;
	uint64_t count = (uint64_t) s_count;
	uint64_t offset = (uint64_t) o_offset;
	uint32_t file_size = Size();
	uint64_t end_at = offset + count;
	if ( offset < end_at )
		/* TODO: Overflow! off_t overflow? */{};
	if ( file_size < end_at && !Truncate(end_at) )
		return -1;
	uint32_t cluster_id = offset / filesystem->cluster_size;
	uint32_t cluster_offset = offset % filesystem->cluster_size;
	uint32_t cluster = SeekCluster(cluster_id);
	if ( filesystem->eio_cluster <= cluster )
		return -1;
	while ( sofar < count )
	{
		if ( filesystem->cluster_size <= cluster_offset )
		{
			cluster = filesystem->ReadFAT(cluster);
			// TODO: Better checks.
			if ( filesystem->eio_cluster <= cluster )
				return sofar ? sofar : (errno = EIO, -1);
			cluster_offset = 0;
			cluster_id++;
		}
		uint8_t sector = cluster_offset / filesystem->bytes_per_sector;
		uint16_t block_offset = cluster_offset % filesystem->bytes_per_sector;
		uint32_t block_left = filesystem->bytes_per_sector - block_offset;
		Block* block = GetClusterSector(cluster, sector);
		if ( !block )
			return sofar ? sofar : -1;
		size_t amount = count - sofar < block_left ? count - sofar : block_left;
		block->BeginWrite();
		memcpy(block->block_data + block_offset, buf + sofar, amount);
		block->FinishWrite();
		sofar += amount;
		cluster_offset += amount;
		block->Unref();
	}
	return (ssize_t) sofar;
}

bool Inode::UnembedInInode()
{
	assert(data->i_blocks == 0 && 0 < data->i_size && data->i_size <= 60);
	unsigned char* block_data = (unsigned char*) &data->i_block[0];
	size_t content_size = Size();
	unsigned char content[60];
	memcpy(content, block_data, content_size);
	data_block->BeginWrite();
	memset(block_data, 0, 60);
	data->i_size = 0;
	data_block->FinishWrite();
	if ( WriteAt(content, content_size, 0) != (ssize_t) content_size )
	{
		Truncate(0);
		data_block->BeginWrite();
		memcpy(block_data, content, content_size);
		data->i_size = content_size;
		data->i_blocks = 0;
		data_block->FinishWrite();
		return false;
	}
	return true;
}

bool Inode::Rename(Inode* olddir, const char* oldname, const char* newname)
{
	if ( !strcmp(oldname, ".") || !strcmp(oldname, "..") ||
	     !strcmp(newname, ".") || !strcmp(newname, "..") )
		return errno = EPERM, false;
	Inode* src_inode = olddir->Open(oldname, O_RDONLY, 0);
	if ( !src_inode )
		return false;
	if ( Inode* dst_inode = Open(newname, O_RDONLY, 0) )
	{
		bool same_inode = src_inode->inode_id == dst_inode->inode_id;
		dst_inode->Unref();
		if ( same_inode )
			return src_inode->Unref(), true;
	}
	// TODO: Prove that this cannot fail and handle such a situation.
	if ( FAT_S_ISDIR(src_inode->Mode()) )
	{
		if ( !Unlink(newname, true) && errno != ENOENT )
			return src_inode->Unref(), false;
		Link(newname, src_inode, true);
		olddir->Unlink(oldname, true, true);
		if ( olddir != this )
		{
			src_inode->Unlink("..", true, true);
			src_inode->Link("..", this, true);
		}
	}
	else
	{
		if ( !Unlink(newname, false) && errno != ENOENT )
			return src_inode->Unref(), false;
		Link(newname, src_inode, false);
		olddir->Unlink(oldname, false);
	}

	src_inode->Unref();
	return true;
}

bool Inode::Symlink(const char* elem, const char* dest)
{
	if ( !filesystem->device->write )
		return errno = EROFS, false;

	// TODO: Preferred block group!
	uint32_t result_inode_id = filesystem->AllocateInode();
	if ( !result_inode_id )
		return NULL;

	Inode* result = filesystem->GetInode(result_inode_id);
	if ( !result )
		return filesystem->FreeInode(result_inode_id), false;
	struct timespec now;
	clock_gettime(CLOCK_REALTIME, &now);
	result->BeginWrite();
	memset(result->data, 0, sizeof(*result->data));
	result->data->i_atime = now.tv_sec;
	result->data->i_ctime = now.tv_sec;
	result->data->i_mtime = now.tv_sec;
	// TODO: Set all the other inode properties!
	result->FinishWrite();
	result->SetMode((0777 & S_SETABLE) | FAT_S_IFLNK);
	result->SetUserId(request_uid);
	result->SetGroupId(request_gid);

	size_t dest_length = strlen(dest);
	if ( SSIZE_MAX < dest_length )
		return errno = EFBIG, -1;
	if ( result->WriteAt((const uint8_t*) dest, dest_length, 0) < (ssize_t) dest_length )
		return result->Unref(), false;

	if ( !Link(elem, result, false) )
		return result->Truncate(0), result->Unref(), false;

	result->Unref();
	return true;
}

Inode* Inode::CreateDirectory(const char* path, mode_t mode)
{
	return Open(path, O_CREAT | O_EXCL, mode | S_IFDIR);
}

bool Inode::RemoveDirectory(const char* path)
{
	Inode* result = UnlinkKeep(path, true);
	if ( !result )
		return false;
	result->Unlink("..", true, true);
	result->Unlink(".", true, true);
	result->Truncate(0);

	// Decrease the directory count statistics.
	uint32_t group_id = (result->inode_id - 1) / filesystem->sb->s_inodes_per_group;
	assert(group_id < filesystem->num_groups);
	BlockGroup* block_group = filesystem->GetBlockGroup(group_id);
	if ( block_group )
	{
		block_group->BeginWrite();
		block_group->data->bg_used_dirs_count--;
		block_group->FinishWrite();
		block_group->Unref();
	}

	result->Unref();

	return true;
}

bool Inode::IsEmptyDirectory()
{
	if ( !FAT_S_ISDIR(Mode()) )
		return errno = ENOTDIR, false;
	uint32_t block_size = filesystem->block_size;
	uint64_t filesize = Size();
	uint64_t offset = 0;
	Block* block = NULL;
	uint64_t block_id = 0;
	while ( offset < filesize )
	{
		uint64_t entry_block_id = offset / block_size;
		uint64_t entry_block_offset = offset % block_size;
		if ( block && block_id != entry_block_id )
			block->Unref(),
			block = NULL;
		if ( !block && !(block = GetBlock(block_id = entry_block_id)) )
			return false;
		uint8_t* block_data = block->block_data + entry_block_offset;
		struct ext_dirent* entry = (struct ext_dirent*) block_data;
		if ( entry->inode &&
		     !((entry->name_len == 1 && entry->name[0] == '.') ||
		       (entry->name_len == 2 && entry->name[0] == '.' &&
		                                entry->name[1] == '.' )) )
		{
			block->Unref();
			return false;
		}
		offset += entry->reclen;
	}
	if ( block )
		block->Unref();
	return true;
}

void Inode::Delete()
{
	assert(!data->i_links_count);
	assert(!reference_count);
	assert(!remote_reference_count);

	Truncate(0);

	uint32_t deleted_inode_id = inode_id;

	struct timespec now;
	clock_gettime(CLOCK_REALTIME, &now);

	BeginWrite();
	memset(data, 0, sizeof(*data));
	data->i_dtime = now.tv_sec;
	FinishWrite();

	filesystem->FreeInode(deleted_inode_id);
}

void Inode::Refer()
{
	reference_count++;
}

void Inode::Unref()
{
	assert(0 < reference_count);
	reference_count--;
	if ( !reference_count && !remote_reference_count )
	{
		// TODO: Can data be NULL here? Should the other drives do this?
		if ( data && !data->i_links_count )
			Delete();
		delete this;
	}
}

void Inode::RemoteRefer()
{
	remote_reference_count++;
}

void Inode::RemoteUnref()
{
	assert(0 < remote_reference_count);
	remote_reference_count--;
	if ( !reference_count && !remote_reference_count )
	{
		// TODO: Check if deleted.
		//Delete();
		delete this;
	}
}

void Inode::Modified()
{
	struct timespec now;
	clock_gettime(CLOCK_REALTIME, &now);
	BeginWrite();
	data->i_mtime = now.tv_sec;
	FinishWrite();
}

void Inode::BeginWrite()
{
	data_block->BeginWrite();
}

void Inode::FinishWrite()
{
	struct timespec now;
	clock_gettime(CLOCK_REALTIME, &now);
	data->i_ctime = now.tv_sec;
	if ( !dirty )
	{
		dirty = true;
		prev_dirty = NULL;
		next_dirty = filesystem->dirty_inode;
		if ( next_dirty )
			next_dirty->prev_dirty = this;
		filesystem->dirty_inode = this;
	}
	data_block->FinishWrite();
	Use();
}

void Inode::Sync()
{
	if ( !dirty )
		return;
	data_block->Sync();
	// TODO: The inode contents needs to be sync'd as well!
	(prev_dirty ? prev_dirty->next_dirty : filesystem->dirty_inode) = next_dirty;
	if ( next_dirty )
		next_dirty->prev_dirty = prev_dirty;
	prev_dirty = NULL;
	next_dirty = NULL;
	dirty = false;
}

void Inode::Use()
{
	data_block->Use();
	Unlink();
	Prelink();
}

void Inode::Unlink()
{
	(prev_inode ? prev_inode->next_inode : filesystem->mru_inode) = next_inode;
	(next_inode ? next_inode->prev_inode : filesystem->lru_inode) = prev_inode;
	size_t bin = inode_id % INODE_HASH_LENGTH;
	(prev_hashed ? prev_hashed->next_hashed : filesystem->hash_inodes[bin]) = next_hashed;
	if ( next_hashed ) next_hashed->prev_hashed = prev_hashed;
}

void Inode::Prelink()
{
	prev_inode = NULL;
	next_inode = filesystem->mru_inode;
	if ( filesystem->mru_inode )
		filesystem->mru_inode->prev_inode = this;
	filesystem->mru_inode = this;
	if ( !filesystem->lru_inode )
		filesystem->lru_inode = this;
	size_t bin = inode_id % INODE_HASH_LENGTH;
	prev_hashed = NULL;
	next_hashed = filesystem->hash_inodes[bin];
	filesystem->hash_inodes[bin] = this;
	if ( next_hashed )
		next_hashed->prev_hashed = this;
}