1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-05-14 14:24:57 +00:00
serenity/Kernel/FileSystem/FATFS/Inode.cpp

383 lines
13 KiB
C++

/*
* Copyright (c) 2022, Undefine <undefine@undefine.pl>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/ByteReader.h>
#include <AK/Endian.h>
#include <AK/Time.h>
#include <Kernel/Debug.h>
#include <Kernel/FileSystem/FATFS/Inode.h>
#include <Kernel/Library/KBufferBuilder.h>
namespace Kernel {
ErrorOr<NonnullRefPtr<FATInode>> FATInode::create(FATFS& fs, FATEntry entry, Vector<FATLongFileNameEntry> const& lfn_entries)
{
auto filename = TRY(compute_filename(entry, lfn_entries));
return adopt_nonnull_ref_or_enomem(new (nothrow) FATInode(fs, entry, move(filename)));
}
FATInode::FATInode(FATFS& fs, FATEntry entry, NonnullOwnPtr<KString> filename)
: Inode(fs, first_cluster(fs.m_fat_version))
, m_entry(entry)
, m_filename(move(filename))
{
dbgln_if(FAT_DEBUG, "FATFS: Creating inode {} with filename \"{}\"", index(), m_filename);
m_metadata = {
.inode = identifier(),
.size = m_entry.file_size,
.mode = static_cast<mode_t>((has_flag(m_entry.attributes, FATAttributes::Directory) ? S_IFDIR : S_IFREG) | 0777),
.uid = 0,
.gid = 0,
.link_count = 0,
.atime = time_from_packed_dos(m_entry.last_accessed_date, { 0 }),
.ctime = time_from_packed_dos(m_entry.creation_date, m_entry.creation_time),
.mtime = time_from_packed_dos(m_entry.modification_date, m_entry.modification_time),
.dtime = {},
.block_count = 0,
.block_size = 0,
.major_device = 0,
.minor_device = 0,
};
}
ErrorOr<Vector<BlockBasedFileSystem::BlockIndex>> FATInode::compute_block_list()
{
VERIFY(m_inode_lock.is_locked());
dbgln_if(FAT_DEBUG, "FATFS: computing block list for inode {}", index());
u32 cluster = first_cluster();
Vector<BlockBasedFileSystem::BlockIndex> block_list;
auto fat_sector = TRY(KBuffer::try_create_with_size("FATFS: FAT read buffer"sv, fs().m_device_block_size));
auto fat_sector_buffer = UserOrKernelBuffer::for_kernel_buffer(fat_sector->data());
while (cluster < end_of_chain_marker()) {
dbgln_if(FAT_DEBUG, "FATFS: Appending cluster {} to inode {}'s cluster chain", cluster, index());
auto first_block_and_length = fs().first_block_of_cluster(cluster);
for (u8 i = 0; i < first_block_and_length.number_of_sectors; i++)
block_list.append(BlockBasedFileSystem::BlockIndex { first_block_and_length.start_block.value() + i });
// Clusters 0 and 1 are reserved in the FAT, and their entries in the FAT will
// not point to another valid cluster in the chain (Cluster 0 typically holds
// the "FAT ID" field with some flags, Cluster 1 should be the end of chain
// marker).
// Internally, we use `cluster == 0` to represent the root directory Inode,
// which is a signal to read the root directory region blocks on FAT12/16
// file systems. (`fs().first_block_of_cluster` will return the appropriate
// block/sectors to read given cluster == 0).
// Therefore, we read one set of sectors for these invalud cluster numbers,
// and then terminate the loop becuase the FAT entry at `cluster` for these
// values does not represent the next step in the chain (because there is
// nothing else to read).
if (cluster <= 1) {
break;
}
u32 fat_offset = cluster * cluster_size();
u32 fat_sector_index = fs().m_parameter_block->common_bpb()->reserved_sector_count + (fat_offset / fs().m_device_block_size);
u32 entry_offset = fat_offset % fs().m_device_block_size;
TRY(fs().raw_read(fat_sector_index, fat_sector_buffer));
cluster = cluster_number(*fat_sector, entry_offset);
}
return block_list;
}
u32 FATInode::end_of_chain_marker() const
{
// Returns the end of chain entry for the given file system.
// Any FAT entry of this value or greater signifies the end
// of the chain has been reached for a given entry.
switch (fs().m_fat_version) {
case FATVersion::FAT12:
return 0xff8;
case FATVersion::FAT16:
return 0xfff8;
case FATVersion::FAT32:
return 0x0FFFFFF8;
default:
VERIFY_NOT_REACHED();
}
}
size_t FATInode::cluster_size() const
{
switch (fs().m_fat_version) {
case FATVersion::FAT12:
VERIFY(false);
break;
case FATVersion::FAT16:
return 2;
case FATVersion::FAT32:
return 4;
default:
VERIFY(false);
}
}
u32 FATInode::cluster_number(KBuffer const& fat_sector, u32 entry_offset) const
{
u32 cluster = 0;
switch (fs().m_fat_version) {
case FATVersion::FAT12:
VERIFY(false);
break;
case FATVersion::FAT16: {
u16 cluster_u16_le = 0;
ByteReader::load<u16>(fat_sector.bytes().offset(entry_offset), cluster_u16_le);
cluster = LittleEndian { cluster_u16_le };
break;
}
case FATVersion::FAT32: {
u32 cluster_u32_le = 0;
ByteReader::load<u32>(fat_sector.bytes().offset(entry_offset), cluster_u32_le);
cluster = LittleEndian { cluster_u32_le };
// FAT32 entries use 28-bits to represent the cluster number. The top 4 bits
// may contain flags or other data and must be masked off.
cluster &= 0x0FFFFFFF;
break;
}
default:
VERIFY(false);
}
return cluster;
}
ErrorOr<NonnullOwnPtr<KBuffer>> FATInode::read_block_list()
{
VERIFY(m_inode_lock.is_locked());
dbgln_if(FAT_DEBUG, "FATFS: reading block list for inode {} ({} blocks)", index(), m_block_list.size());
if (m_block_list.is_empty())
m_block_list = TRY(compute_block_list());
auto builder = TRY(KBufferBuilder::try_create());
u8 buffer[512];
VERIFY(fs().m_device_block_size <= sizeof(buffer));
auto buf = UserOrKernelBuffer::for_kernel_buffer(buffer);
for (BlockBasedFileSystem::BlockIndex block : m_block_list) {
dbgln_if(FAT_DEBUG, "FATFS: reading block: {}", block);
TRY(fs().raw_read(block, buf));
TRY(builder.append((char const*)buffer, fs().m_device_block_size));
}
auto blocks = builder.build();
if (!blocks)
return ENOMEM;
return blocks.release_nonnull();
}
ErrorOr<void> FATInode::replace_child(StringView, Inode&)
{
// TODO: Implement this once we have write support.
return Error::from_errno(EROFS);
}
ErrorOr<RefPtr<FATInode>> FATInode::traverse(Function<ErrorOr<bool>(RefPtr<FATInode>)> callback)
{
VERIFY(has_flag(m_entry.attributes, FATAttributes::Directory));
Vector<FATLongFileNameEntry> lfn_entries;
auto blocks = TRY(read_block_list());
for (u32 i = 0; i < blocks->size() / sizeof(FATEntry); i++) {
auto* entry = reinterpret_cast<FATEntry*>(blocks->data() + i * sizeof(FATEntry));
if (entry->filename[0] == end_entry_byte) {
dbgln_if(FAT_DEBUG, "FATFS: Found end entry");
return nullptr;
} else if (static_cast<u8>(entry->filename[0]) == unused_entry_byte) {
dbgln_if(FAT_DEBUG, "FATFS: Found unused entry");
lfn_entries.clear();
} else if (entry->attributes == FATAttributes::LongFileName) {
dbgln_if(FAT_DEBUG, "FATFS: Found LFN entry");
TRY(lfn_entries.try_append(*reinterpret_cast<FATLongFileNameEntry*>(entry)));
} else if ((entry->first_cluster_high << 16 | entry->first_cluster_low) <= 1 && entry->file_size > 0) {
// Because clusters 0 and 1 are reserved, only empty files (size == 0 files)
// should specify these clusters.
// This driver uses a cluster number == 0 to represent the root directory inode
// on FAT12/16 file systems (a signal to look in the root directory region),
// so we ensure that no entries read off the file system have a cluster number
// that would also point to this region.
dbgln_if(FAT_DEBUG, "FATFS: Invalid cluster for entry");
return EINVAL;
} else {
dbgln_if(FAT_DEBUG, "FATFS: Found 8.3 entry");
lfn_entries.reverse();
auto inode = TRY(FATInode::create(fs(), *entry, lfn_entries));
if (TRY(callback(inode)))
return inode;
lfn_entries.clear();
}
}
return EINVAL;
}
ErrorOr<NonnullOwnPtr<KString>> FATInode::compute_filename(FATEntry& entry, Vector<FATLongFileNameEntry> const& lfn_entries)
{
if (lfn_entries.is_empty()) {
StringBuilder filename;
filename.append(byte_terminated_string(StringView(entry.filename, normal_filename_length), ' '));
if (entry.extension[0] != ' ') {
filename.append('.');
filename.append(byte_terminated_string(StringView(entry.extension, normal_extension_length), ' '));
}
return TRY(KString::try_create(filename.string_view()));
} else {
StringBuilder filename;
for (auto& lfn_entry : lfn_entries) {
filename.append(lfn_entry.characters1[0]);
filename.append(lfn_entry.characters1[1]);
filename.append(lfn_entry.characters1[2]);
filename.append(lfn_entry.characters1[3]);
filename.append(lfn_entry.characters1[4]);
filename.append(lfn_entry.characters2[0]);
filename.append(lfn_entry.characters2[1]);
filename.append(lfn_entry.characters2[2]);
filename.append(lfn_entry.characters2[3]);
filename.append(lfn_entry.characters2[4]);
filename.append(lfn_entry.characters2[5]);
filename.append(lfn_entry.characters3[0]);
filename.append(lfn_entry.characters3[1]);
}
// Long Filenames have two terminators:
// 1. Completely unused "entries" (the `characterN` fields of
// `lfn_entry`) are filled with 0xFF (`lfn_entry_unused_byte`).
// 2. Partially used entries (within `characterN`) are null-padded.
//
// `filename` is truncated first to eliminate unused entries, and
// then further truncated to remove any existing null padding characters.
//
// Page 8 of the Long Filename Specification
// (http://www.osdever.net/documents/LongFileName.pdf)
// details this encoding ("If the long name does not fill...").
return TRY(KString::try_create(
byte_terminated_string(
byte_terminated_string(filename.string_view(), lfn_entry_unused_byte), lfn_entry_character_termination)));
}
VERIFY_NOT_REACHED();
}
StringView FATInode::byte_terminated_string(StringView string, u8 fill_byte)
{
if (auto index = string.find_last_not(fill_byte); index.has_value())
return string.substring_view(0, index.value() + 1);
return string;
}
u32 FATInode::first_cluster() const
{
return first_cluster(fs().m_fat_version);
}
u32 FATInode::first_cluster(FATVersion const version) const
{
if (version == FATVersion::FAT32) {
return (static_cast<u32>(m_entry.first_cluster_high) << 16) | m_entry.first_cluster_low;
}
// The space occupied in a directory entry by `first_cluster_high` (0x14)
// is reserved in FAT12/16, and may be used to store file meta-data.
// As a result, do not include it.
return m_entry.first_cluster_low;
}
ErrorOr<size_t> FATInode::read_bytes_locked(off_t offset, size_t size, UserOrKernelBuffer& buffer, OpenFileDescription*) const
{
dbgln_if(FAT_DEBUG, "FATFS: Reading inode {}: size: {} offset: {}", identifier().index(), size, offset);
VERIFY(offset >= 0);
if (offset >= m_metadata.size)
return 0;
// FIXME: Read only the needed blocks instead of the whole file
auto blocks = TRY(const_cast<FATInode&>(*this).read_block_list());
TRY(buffer.write(blocks->data() + offset, min(size, m_block_list.size() * fs().m_device_block_size - offset)));
return min(size, m_block_list.size() * fs().m_device_block_size - offset);
}
InodeMetadata FATInode::metadata() const
{
return m_metadata;
}
ErrorOr<void> FATInode::traverse_as_directory(Function<ErrorOr<void>(FileSystem::DirectoryEntryView const&)> callback) const
{
MutexLocker locker(m_inode_lock);
VERIFY(has_flag(m_entry.attributes, FATAttributes::Directory));
[[maybe_unused]] auto inode = TRY(const_cast<FATInode&>(*this).traverse([&callback](auto inode) -> ErrorOr<bool> {
if (inode->m_filename->view() == "" || inode->m_filename->view() == "." || inode->m_filename->view() == "..")
return false;
TRY(callback({ inode->m_filename->view(), inode->identifier(), static_cast<u8>(inode->m_entry.attributes) }));
return false;
}));
return {};
}
ErrorOr<NonnullRefPtr<Inode>> FATInode::lookup(StringView name)
{
MutexLocker locker(m_inode_lock);
VERIFY(has_flag(m_entry.attributes, FATAttributes::Directory));
auto inode = TRY(traverse([name](auto child) -> ErrorOr<bool> {
return child->m_filename->view() == name;
}));
if (inode.is_null())
return ENOENT;
return inode.release_nonnull();
}
ErrorOr<size_t> FATInode::write_bytes_locked(off_t, size_t, UserOrKernelBuffer const&, OpenFileDescription*)
{
return EROFS;
}
ErrorOr<NonnullRefPtr<Inode>> FATInode::create_child(StringView, mode_t, dev_t, UserID, GroupID)
{
return EROFS;
}
ErrorOr<void> FATInode::add_child(Inode&, StringView, mode_t)
{
return EROFS;
}
ErrorOr<void> FATInode::remove_child(StringView)
{
return EROFS;
}
ErrorOr<void> FATInode::chmod(mode_t)
{
return EROFS;
}
ErrorOr<void> FATInode::chown(UserID, GroupID)
{
return EROFS;
}
ErrorOr<void> FATInode::flush_metadata()
{
return EROFS;
}
}