DirAsDSK.cc

Go to the documentation of this file.

#include "DirAsDSK.hh"
#include "DiskChanger.hh"
#include "Scheduler.hh"
#include "CliComm.hh"
#include "BootBlocks.hh"
#include "File.hh"
#include "FileException.hh"
#include "ReadDir.hh"
#include "StringOp.hh"
#include "ranges.hh"
#include "stl.hh"
#include <cassert>
#include <cstring>
#include <vector>

using std::string;
using std::vector;

namespace openmsx {

static const unsigned SECTOR_SIZE = sizeof(SectorBuffer);
static const unsigned SECTORS_PER_DIR = 7;
static const unsigned NUM_FATS = 2;
static const unsigned NUM_TRACKS = 80;
static const unsigned SECTORS_PER_CLUSTER = 2;
static const unsigned SECTORS_PER_TRACK = 9;
static const unsigned FIRST_FAT_SECTOR = 1;
static const unsigned DIR_ENTRIES_PER_SECTOR =
        SECTOR_SIZE / sizeof(MSXDirEntry);

// First valid regular cluster number.
static const unsigned FIRST_CLUSTER = 2;

static const unsigned FREE_FAT = 0x000;
static const unsigned BAD_FAT  = 0xFF7;
static const unsigned EOF_FAT  = 0xFFF; // actually 0xFF8-0xFFF


// Transform BAD_FAT (0xFF7) and EOF_FAT-range (0xFF8-0xFFF)
// to a single value: EOF_FAT (0xFFF).
static unsigned normalizeFAT(unsigned cluster)
{
        return (cluster < BAD_FAT) ? cluster : EOF_FAT;
}

unsigned DirAsDSK::readFATHelper(const SectorBuffer* fatBuf, unsigned cluster) const
{
        assert(FIRST_CLUSTER <= cluster);
        assert(cluster < maxCluster);
        auto* buf = fatBuf[0].raw;
        auto* p = &buf[(cluster * 3) / 2];
        unsigned result = (cluster & 1)
                        ? (p[0] >> 4) + (p[1] << 4)
                        : p[0] + ((p[1] & 0x0F) << 8);
        return normalizeFAT(result);
}

void DirAsDSK::writeFATHelper(SectorBuffer* fatBuf, unsigned cluster, unsigned val) const
{
        assert(FIRST_CLUSTER <= cluster);
        assert(cluster < maxCluster);
        auto* buf = fatBuf[0].raw;
        auto* p = &buf[(cluster * 3) / 2];
        if (cluster & 1) {
                p[0] = (p[0] & 0x0F) + (val << 4);
                p[1] = val >> 4;
        } else {
                p[0] = val;
                p[1] = (p[1] & 0xF0) + ((val >> 8) & 0x0F);
        }
}

SectorBuffer* DirAsDSK::fat()
{
        return &sectors[FIRST_FAT_SECTOR];
}
SectorBuffer* DirAsDSK::fat2()
{
        return &sectors[firstSector2ndFAT];
}

// Read entry from FAT.
unsigned DirAsDSK::readFAT(unsigned cluster)
{
        return readFATHelper(fat(), cluster);
}

// Write an entry to both FAT1 and FAT2.
void DirAsDSK::writeFAT12(unsigned cluster, unsigned val)
{
        writeFATHelper(fat (), cluster, val);
        writeFATHelper(fat2(), cluster, val);
        // An alternative is to copy FAT1 to FAT2 after changes have been made
        // to FAT1. This is probably more like what the real disk rom does.
}

// Returns maxCluster in case of no more free clusters
unsigned DirAsDSK::findNextFreeCluster(unsigned cluster)
{
        assert(cluster < maxCluster);
        do {
                ++cluster;
                assert(cluster >= FIRST_CLUSTER);
        } while ((cluster < maxCluster) && (readFAT(cluster) != FREE_FAT));
        return cluster;
}
unsigned DirAsDSK::findFirstFreeCluster()
{
        return findNextFreeCluster(FIRST_CLUSTER - 1);
}

// Throws when there are no more free clusters.
unsigned DirAsDSK::getFreeCluster()
{
        unsigned cluster = findFirstFreeCluster();
        if (cluster == maxCluster) {
                throw MSXException("disk full");
        }
        return cluster;
}

unsigned DirAsDSK::clusterToSector(unsigned cluster) const
{
        assert(cluster >= FIRST_CLUSTER);
        assert(cluster < maxCluster);
        return firstDataSector + SECTORS_PER_CLUSTER *
                    (cluster - FIRST_CLUSTER);
}

void DirAsDSK::sectorToCluster(unsigned sector, unsigned& cluster, unsigned& offset) const
{
        assert(sector >= firstDataSector);
        assert(sector < nofSectors);
        sector -= firstDataSector;
        cluster = (sector / SECTORS_PER_CLUSTER) + FIRST_CLUSTER;
        offset  = (sector % SECTORS_PER_CLUSTER) * SECTOR_SIZE;
}
unsigned DirAsDSK::sectorToCluster(unsigned sector) const
{
        unsigned cluster, offset;
        sectorToCluster(sector, cluster, offset);
        return cluster;
}

MSXDirEntry& DirAsDSK::msxDir(DirIndex dirIndex)
{
        assert(dirIndex.sector < nofSectors);
        assert(dirIndex.idx    < DIR_ENTRIES_PER_SECTOR);
        return sectors[dirIndex.sector].dirEntry[dirIndex.idx];
}

// Returns -1 when there are no more sectors for this directory.
unsigned DirAsDSK::nextMsxDirSector(unsigned sector)
{
        if (sector < firstDataSector) {
                // Root directory.
                assert(firstDirSector <= sector);
                ++sector;
                if (sector == firstDataSector) {
                        // Root directory has a fixed number of sectors.
                        return unsigned(-1);
                }
                return sector;
        } else {
                // Subdirectory.
                unsigned cluster, offset;
                sectorToCluster(sector, cluster, offset);
                if (offset < ((SECTORS_PER_CLUSTER - 1) * SECTOR_SIZE)) {
                        // Next sector still in same cluster.
                        return sector + 1;
                }
                unsigned nextCl = readFAT(cluster);
                if ((nextCl < FIRST_CLUSTER) || (maxCluster <= nextCl)) {
                        // No next cluster, end of directory reached.
                        return unsigned(-1);
                }
                return clusterToSector(nextCl);
        }
}

// Check if a msx filename is used in a specific msx (sub)directory.
bool DirAsDSK::checkMSXFileExists(
        const string& msxFilename, unsigned msxDirSector)
{
        vector<bool> visited(nofSectors, false);
        do {
                if (visited[msxDirSector]) {
                        // cycle detected, invalid disk, but don't crash on it
                        return false;
                }
                visited[msxDirSector] = true;

                for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                        DirIndex dirIndex(msxDirSector, idx);
                        if (memcmp(msxDir(dirIndex).filename,
                                   msxFilename.data(), 8 + 3) == 0) {
                                return true;
                        }
                }
                msxDirSector = nextMsxDirSector(msxDirSector);
        } while (msxDirSector != unsigned(-1));

        // Searched through all sectors of this (sub)directory.
        return false;
}

// Returns msx directory entry for the given host file. Or -1 if the host file
// is not mapped in the virtual disk.
DirAsDSK::DirIndex DirAsDSK::findHostFileInDSK(const string& hostName)
{
        for (auto& p : mapDirs) {
                if (p.second.hostName == hostName) {
                        return p.first;
                }
        }
        return {unsigned(-1), unsigned(-1)};
}

// Check if a host file is already mapped in the virtual disk.
bool DirAsDSK::checkFileUsedInDSK(const string& hostName)
{
        DirIndex dirIndex = findHostFileInDSK(hostName);
        return dirIndex.sector != unsigned(-1);
}

static string hostToMsxName(string hostName)
{
        // Create an MSX filename 8.3 format. TODO use vfat-like abbreviation
        transform_in_place(hostName, [](char a) {
                return (a == ' ') ? '_' : ::toupper(a);
        });
        string_view file, ext;
        StringOp::splitOnLast(hostName, '.', file, ext);
        if (file.empty()) std::swap(file, ext);

        string result(8 + 3, ' ');
        // c++17: use result.data()
        memcpy(&*begin(result) + 0, file.data(), std::min<size_t>(8, file.size()));
        memcpy(&*begin(result) + 8, ext .data(), std::min<size_t>(3, ext .size()));
        ranges::replace(result, '.', '_');
        return result;
}

static string msxToHostName(const char* msxName)
{
        string result;
        for (unsigned i = 0; (i < 8) && (msxName[i] != ' '); ++i) {
                result += char(tolower(msxName[i]));
        }
        if (msxName[8] != ' ') {
                result += '.';
                for (unsigned i = 8; (i < (8 + 3)) && (msxName[i] != ' '); ++i) {
                        result += char(tolower(msxName[i]));
                }
        }
        return result;
}


DirAsDSK::DirAsDSK(DiskChanger& diskChanger_, CliComm& cliComm_,
                   const Filename& hostDir_, SyncMode syncMode_,
                   BootSectorType bootSectorType)
        : SectorBasedDisk(hostDir_)
        , diskChanger(diskChanger_)
        , cliComm(cliComm_)
        , hostDir(hostDir_.getResolved() + '/')
        , syncMode(syncMode_)
        , lastAccess(EmuTime::zero)
        , nofSectors((diskChanger_.isDoubleSidedDrive() ? 2 : 1) * SECTORS_PER_TRACK * NUM_TRACKS)
        , nofSectorsPerFat((((3 * nofSectors) / (2 * SECTORS_PER_CLUSTER)) + SECTOR_SIZE - 1) / SECTOR_SIZE)
        , firstSector2ndFAT(FIRST_FAT_SECTOR + nofSectorsPerFat)
        , firstDirSector(FIRST_FAT_SECTOR + NUM_FATS * nofSectorsPerFat)
        , firstDataSector(firstDirSector + SECTORS_PER_DIR)
        , maxCluster((nofSectors - firstDataSector) / SECTORS_PER_CLUSTER + FIRST_CLUSTER)
        , sectors(nofSectors)
{
        if (!FileOperations::isDirectory(hostDir)) {
                throw MSXException("Not a directory");
        }

        // First create structure for the virtual disk.
        byte numSides = diskChanger_.isDoubleSidedDrive() ? 2 : 1;
        setNbSectors(nofSectors);
        setSectorsPerTrack(SECTORS_PER_TRACK);
        setNbSides(numSides);

        // Initially the whole disk is filled with 0xE5 (at least on Philips
        // NMS8250).
        memset(sectors.data(), 0xE5, sizeof(SectorBuffer) * nofSectors);

        // Use selected bootsector, fill-in values.
        byte mediaDescriptor = (numSides == 2) ? 0xF9 : 0xF8;
        const auto& protoBootSector = bootSectorType == BOOTSECTOR_DOS1
                ? BootBlocks::dos1BootBlock
                : BootBlocks::dos2BootBlock;
        memcpy(&sectors[0], &protoBootSector, sizeof(protoBootSector));
        auto& bootSector = sectors[0].bootSector;
        bootSector.bpSector     = SECTOR_SIZE;
        bootSector.spCluster    = SECTORS_PER_CLUSTER;
        bootSector.nrFats       = NUM_FATS;
        bootSector.dirEntries   = SECTORS_PER_DIR * (SECTOR_SIZE / sizeof(MSXDirEntry));
        bootSector.nrSectors    = nofSectors;
        bootSector.descriptor   = mediaDescriptor;
        bootSector.sectorsFat   = nofSectorsPerFat;
        bootSector.sectorsTrack = SECTORS_PER_TRACK;
        bootSector.nrSides      = numSides;

        // Clear FAT1 + FAT2.
        memset(fat(), 0, SECTOR_SIZE * nofSectorsPerFat * NUM_FATS);
        // First 3 bytes are initialized specially:
        //  'cluster 0' contains the media descriptor
        //  'cluster 1' is marked as EOF_FAT
        //  So cluster 2 is the first usable cluster number
        fat ()->raw[0] = mediaDescriptor; fat ()->raw[1] = 0xFF; fat ()->raw[2] = 0xFF;
        fat2()->raw[0] = mediaDescriptor; fat2()->raw[1] = 0xFF; fat2()->raw[2] = 0xFF;

        // Assign empty directory entries.
        memset(&sectors[firstDirSector], 0, SECTOR_SIZE * SECTORS_PER_DIR);

        // No host files are mapped to this disk yet.
        assert(mapDirs.empty());

        // Import the host filesystem.
        syncWithHost();
}

bool DirAsDSK::isWriteProtectedImpl() const
{
        return syncMode == SYNC_READONLY;
}

void DirAsDSK::checkCaches()
{
        bool needSync;
        if (auto* scheduler = diskChanger.getScheduler()) {
                auto now = scheduler->getCurrentTime();
                auto delta = now - lastAccess;
                needSync = delta > EmuDuration::sec(1);
                // Do not update lastAccess because we don't actually call
                // syncWithHost().
        } else {
                // Happens when dirasdisk is used in virtual_drive.
                needSync = true;
        }
        if (needSync) {
                flushCaches();
        }
}

void DirAsDSK::readSectorImpl(size_t sector, SectorBuffer& buf)
{
        assert(sector < nofSectors);

        // 'Peek-mode' is used to periodically calculate a sha1sum for the
        // whole disk (used by reverse). We don't want this calculation to
        // interfer with the access time we use for normal read/writes. So in
        // peek-mode we skip the whole sync-step.
        if (!isPeekMode()) {
                bool needSync;
                if (auto* scheduler = diskChanger.getScheduler()) {
                        auto now = scheduler->getCurrentTime();
                        auto delta = now - lastAccess;
                        lastAccess = now;
                        needSync = delta > EmuDuration::sec(1);
                } else {
                        // Happens when dirasdisk is used in virtual_drive.
                        needSync = true;
                }
                if (needSync) {
                        syncWithHost();
                        flushCaches(); // e.g. sha1sum
                        // Let the diskdrive report the disk has been ejected.
                        // E.g. a turbor machine uses this to flush its
                        // internal disk caches.
                        diskChanger.forceDiskChange();
                }
        }

        // Simply return the sector from our virtual disk image.
        memcpy(&buf, &sectors[sector], sizeof(buf));
}

void DirAsDSK::syncWithHost()
{
        // Check for removed host files. This frees up space in the virtual
        // disk. Do this first because otherwise later actions may fail (run
        // out of virtual disk space) for no good reason.
        checkDeletedHostFiles();

        // Next update existing files. This may enlarge or shrink virtual
        // files. In case not all host files fit on the virtual disk it's
        // better to update the existing files than to (partly) add a too big
        // new file and have no space left to enlarge the existing files.
        checkModifiedHostFiles();

        // Last add new host files (this can only consume virtual disk space).
        addNewHostFiles({}, firstDirSector);
}

void DirAsDSK::checkDeletedHostFiles()
{
        // This handles both host files and directories.
        auto copy = mapDirs;
        for (auto& p : copy) {
                if (!mapDirs.count(p.first)) { // c++20 contains()
                        // While iterating over (the copy of) mapDirs we delete
                        // entries of mapDirs (when we delete files only the
                        // current entry is deleted, when we delete
                        // subdirectories possibly many entries are deleted).
                        // At this point in the code we've reached such an
                        // entry that's still in the original (copy of) mapDirs
                        // but has already been deleted from the current
                        // mapDirs. Ignore it.
                        continue;
                }
                const DirIndex& dirIndex = p.first;
                MapDir& mapDir = p.second;
                string fullHostName = hostDir + mapDir.hostName;
                bool isMSXDirectory = (msxDir(dirIndex).attrib &
                                       MSXDirEntry::ATT_DIRECTORY) != 0;
                FileOperations::Stat fst;
                if ((!FileOperations::getStat(fullHostName, fst)) ||
                    (FileOperations::isDirectory(fst) != isMSXDirectory)) {
                        // TODO also check access permission
                        // Error stat-ing file, or directory/file type is not
                        // the same on the msx and host side (e.g. a host file
                        // has been removed and a host directory with the same
                        // name has been created). In both cases delete the msx
                        // entry (if needed it will be recreated soon).
                        deleteMSXFile(dirIndex);
                }
        }
}

void DirAsDSK::deleteMSXFile(DirIndex dirIndex)
{
        // Remove mapping between host and msx file (if any).
        mapDirs.erase(dirIndex);

        char c = msxDir(dirIndex).filename[0];
        if (c == 0 || c == char(0xE5)) {
                // Directory entry not in use, don't need to do anything.
                return;
        }

        if (msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY) {
                // If we're deleting a directory then also (recursively)
                // delete the files/directories in this directory.
                const char* msxName = msxDir(dirIndex).filename;
                if ((memcmp(msxName, ".          ", 11) == 0) ||
                    (memcmp(msxName, "..         ", 11) == 0)) {
                        // But skip the "." and ".." entries.
                        return;
                }
                // Sanity check on cluster range.
                unsigned cluster = msxDir(dirIndex).startCluster;
                if ((FIRST_CLUSTER <= cluster) && (cluster < maxCluster)) {
                        // Recursively delete all files in this subdir.
                        deleteMSXFilesInDir(clusterToSector(cluster));
                }
        }

        // At this point we have a regular file or an empty subdirectory.
        // Delete it by marking the first filename char as 0xE5.
        msxDir(dirIndex).filename[0] = char(0xE5);

        // Clear the FAT chain to free up space in the virtual disk.
        freeFATChain(msxDir(dirIndex).startCluster);
}

void DirAsDSK::deleteMSXFilesInDir(unsigned msxDirSector)
{
        vector<bool> visited(nofSectors, false);
        do {
                if (visited[msxDirSector]) {
                        // cycle detected, invalid disk, but don't crash on it
                        return;
                }
                visited[msxDirSector] = true;

                for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                        deleteMSXFile(DirIndex(msxDirSector, idx));
                }
                msxDirSector = nextMsxDirSector(msxDirSector);
        } while (msxDirSector != unsigned(-1));
}

void DirAsDSK::freeFATChain(unsigned cluster)
{
        // Follow a FAT chain and mark all clusters on this chain as free.
        while ((FIRST_CLUSTER <= cluster) && (cluster < maxCluster)) {
                unsigned nextCl = readFAT(cluster);
                writeFAT12(cluster, FREE_FAT);
                cluster = nextCl;
        }
}

void DirAsDSK::checkModifiedHostFiles()
{
        auto copy = mapDirs;
        for (auto& p : copy) {
                if (!mapDirs.count(p.first)) { // c++20 contains()
                        // See comment in checkDeletedHostFiles().
                        continue;
                }
                const DirIndex& dirIndex = p.first;
                MapDir& mapDir = p.second;
                string fullHostName = hostDir + mapDir.hostName;
                bool isMSXDirectory = (msxDir(dirIndex).attrib &
                                       MSXDirEntry::ATT_DIRECTORY) != 0;
                FileOperations::Stat fst;
                if (FileOperations::getStat(fullHostName, fst) &&
                    (FileOperations::isDirectory(fst) == isMSXDirectory)) {
                        // Detect changes in host file.
                        // Heuristic: we use filesize and modification time to detect
                        // changes in file content.
                        //  TODO do we need both filesize and mtime or is mtime alone
                        //       enough?
                        // We ignore time/size changes in directories,
                        // typically such a change indicates one of the files
                        // in that directory is changed/added/removed. But such
                        // changes are handled elsewhere.
                        if (!isMSXDirectory &&
                            ((mapDir.mtime    != fst.st_mtime) ||
                             (mapDir.filesize != size_t(fst.st_size)))) {
                                importHostFile(dirIndex, fst);
                        }
                } else {
                        // Only very rarely happens (because checkDeletedHostFiles()
                        // checked this just recently).
                        deleteMSXFile(dirIndex);
                }
        }
}

void DirAsDSK::importHostFile(DirIndex dirIndex, FileOperations::Stat& fst)
{
        assert(!(msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY));
        assert(mapDirs.count(dirIndex)); // c++20 contains()

        // Set _msx_ modification time.
        setMSXTimeStamp(dirIndex, fst);

        // Set _host_ modification time (and filesize)
        // Note: this is the _only_ place where we update the mapdir.mtime
        // field. We do _not_ update it when the msx writes to the file. So in
        // case the msx does write a data sector, it writes to the correct
        // offset in the host file, but mtime is not updated. On the next
        // host->emu sync we will resync the full file content and only then
        // update mtime. This may seem inefficient, but it makes sure that we
        // never miss any file changes performed by the host. E.g. the host
        // changed part of the file short before the msx wrote to the same
        // file. We can never guarantee that such race-scenarios work
        // correctly, but at least with this mtime-update convention the file
        // content will be the same on the host and the msx side after every
        // sync.
        size_t hostSize = fst.st_size;
        auto& mapDir = mapDirs[dirIndex];
        mapDir.filesize = hostSize;
        mapDir.mtime = fst.st_mtime;

        bool moreClustersInChain = true;
        unsigned curCl = msxDir(dirIndex).startCluster;
        // If there is no cluster assigned yet to this file (curCl == 0), then
        // find a free cluster. Treat invalid cases in the same way (curCl == 1
        // or curCl >= maxCluster).
        if ((curCl < FIRST_CLUSTER) || (curCl >= maxCluster)) {
                moreClustersInChain = false;
                curCl = findFirstFreeCluster(); // maxCluster in case of disk-full
        }

        auto remainingSize = hostSize;
        unsigned prevCl = 0;
        try {
                string fullHostName = hostDir + mapDir.hostName;
                File file(fullHostName, "rb"); // don't uncompress

                while (remainingSize && (curCl < maxCluster)) {
                        unsigned logicalSector = clusterToSector(curCl);
                        for (unsigned i = 0; i < SECTORS_PER_CLUSTER; ++i) {
                                unsigned sector = logicalSector + i;
                                assert(sector < nofSectors);
                                auto* buf = &sectors[sector];
                                memset(buf, 0, SECTOR_SIZE); // in case (end of) file only fills partial sector
                                auto sz = std::min(remainingSize, SECTOR_SIZE);
                                file.read(buf, sz);
                                remainingSize -= sz;
                                if (remainingSize == 0) {
                                        // Don't fill next sectors in this cluster
                                        // if there is no data left.
                                        break;
                                }
                        }

                        if (prevCl) {
                                writeFAT12(prevCl, curCl);
                        } else {
                                msxDir(dirIndex).startCluster = curCl;
                        }
                        prevCl = curCl;

                        // Check if we can follow the existing FAT chain or
                        // need to allocate a free cluster.
                        if (moreClustersInChain) {
                                curCl = readFAT(curCl);
                                if ((curCl == EOF_FAT)      || // normal end
                                    (curCl < FIRST_CLUSTER) || // invalid
                                    (curCl >= maxCluster)) {  // invalid
                                        // Treat invalid FAT chain the same as
                                        // a normal EOF_FAT.
                                        moreClustersInChain = false;
                                        curCl = findFirstFreeCluster();
                                }
                        } else {
                                curCl = findNextFreeCluster(curCl);
                        }
                }
                if (remainingSize != 0) {
                        cliComm.printWarning("Virtual diskimage full: ",
                                             mapDir.hostName, " truncated.");
                }
        } catch (FileException& e) {
                // Error opening or reading host file.
                cliComm.printWarning("Error reading host file: ",
                                     mapDir.hostName, ": ", e.getMessage(),
                                     " Truncated file on MSX disk.");
        }

        // In all cases (no error / image full / host read error) we need to
        // properly terminate the FAT chain.
        if (prevCl) {
                writeFAT12(prevCl, EOF_FAT);
        } else {
                // Filesize zero: don't allocate any cluster, write zero
                // cluster number (checked on a MSXTurboR, DOS2 mode).
                msxDir(dirIndex).startCluster = FREE_FAT;
        }

        // Clear remains of FAT if needed.
        if (moreClustersInChain) {
                freeFATChain(curCl);
        }

        // Write (possibly truncated) file size.
        msxDir(dirIndex).size = uint32_t(hostSize - remainingSize);

        // TODO in case of an error (disk image full, or host file read error),
        // wouldn't it be better to remove the (half imported) msx file again?
        // Sometimes when I'm using DirAsDSK I have one file that is too big
        // (e.g. a core file, or a vim .swp file) and that one prevents
        // DirAsDSK from importing the other (small) files in my directory.
}

void DirAsDSK::setMSXTimeStamp(DirIndex dirIndex, FileOperations::Stat& fst)
{
        // Use intermediate param to prevent compilation error for Android
        time_t mtime = fst.st_mtime;
        auto* mtim = localtime(&mtime);
        int t1 = mtim ? (mtim->tm_sec >> 1) + (mtim->tm_min << 5) +
                        (mtim->tm_hour << 11)
                      : 0;
        msxDir(dirIndex).time = t1;
        int t2 = mtim ? mtim->tm_mday + ((mtim->tm_mon + 1) << 5) +
                        ((mtim->tm_year + 1900 - 1980) << 9)
                      : 0;
        msxDir(dirIndex).date = t2;
}

// Used to add 'regular' files before 'derived' files. E.g. when editing a file
// in a host editor, you often get backup/swap files like this:
//   myfile.txt  myfile.txt~  .myfile.txt.swp
// Currently the 1st and 2nd are mapped to the same MSX filename. If more
// host files map to the same MSX file then (currently) one of the two is
// ignored. Which one is ignored depends on the order in which they are added
// to the virtual disk. This routine/heuristic tries to add 'regular' files
// before derived files.
static size_t weight(const string& hostName)
{
        // TODO this weight function can most likely be improved
        size_t result = 0;
        string_view file, ext;
        StringOp::splitOnLast(hostName, '.', file, ext);
        // too many '.' characters
        result += ranges::count(file, '.') * 100;
        // too long extension
        result += ext.size() * 10;
        // too long file
        result += file.size();
        return result;
}

void DirAsDSK::addNewHostFiles(const string& hostSubDir, unsigned msxDirSector)
{
        assert(!StringOp::startsWith(hostSubDir, '/'));
        assert(hostSubDir.empty() || StringOp::endsWith(hostSubDir, '/'));

        vector<string> hostNames;
        {
                ReadDir dir(hostDir + hostSubDir);
                while (auto* d = dir.getEntry()) {
                        hostNames.emplace_back(d->d_name);
                }
        }
        ranges::sort(hostNames,
             [](const string& l, const string& r) { return weight(l) < weight(r); });

        for (auto& hostName : hostNames) {
                try {
                        if (StringOp::startsWith(hostName, '.')) {
                                // skip '.' and '..'
                                // also skip hidden files on unix
                                continue;
                        }
                        string fullHostName = strCat(hostDir, hostSubDir, hostName);
                        FileOperations::Stat fst;
                        if (!FileOperations::getStat(fullHostName, fst)) {
                                throw MSXException("Error accessing ", fullHostName);
                        }
                        if (FileOperations::isDirectory(fst)) {
                                addNewDirectory(hostSubDir, hostName, msxDirSector, fst);
                        } else if (FileOperations::isRegularFile(fst)) {
                                addNewHostFile(hostSubDir, hostName, msxDirSector, fst);
                        } else {
                                throw MSXException("Not a regular file: ", fullHostName);
                        }
                } catch (MSXException& e) {
                        cliComm.printWarning(e.getMessage());
                }
        }
}

void DirAsDSK::addNewDirectory(const string& hostSubDir, const string& hostName,
                               unsigned msxDirSector, FileOperations::Stat& fst)
{
        string hostPath = hostSubDir + hostName;
        DirIndex dirIndex = findHostFileInDSK(hostPath);
        unsigned newMsxDirSector;
        if (dirIndex.sector == unsigned(-1)) {
                // MSX directory doesn't exist yet, create it.
                // Allocate a cluster to hold the subdirectory entries.
                unsigned cluster = getFreeCluster();
                writeFAT12(cluster, EOF_FAT);

                // Allocate and fill in directory entry.
                try {
                        dirIndex = fillMSXDirEntry(hostSubDir, hostName, msxDirSector);
                } catch (...) {
                        // Rollback allocation of directory cluster.
                        writeFAT12(cluster, FREE_FAT);
                        throw;
                }
                setMSXTimeStamp(dirIndex, fst);
                msxDir(dirIndex).attrib = MSXDirEntry::ATT_DIRECTORY;
                msxDir(dirIndex).startCluster = cluster;

                // Initialize the new directory.
                newMsxDirSector = clusterToSector(cluster);
                for (unsigned i = 0; i < SECTORS_PER_CLUSTER; ++i) {
                        memset(&sectors[newMsxDirSector + i], 0, SECTOR_SIZE);
                }
                DirIndex idx0(newMsxDirSector, 0); // entry for "."
                DirIndex idx1(newMsxDirSector, 1); //           ".."
                memset(msxDir(idx0).filename, ' ', 11);
                memset(msxDir(idx1).filename, ' ', 11);
                memset(msxDir(idx0).filename, '.', 1);
                memset(msxDir(idx1).filename, '.', 2);
                msxDir(idx0).attrib = MSXDirEntry::ATT_DIRECTORY;
                msxDir(idx1).attrib = MSXDirEntry::ATT_DIRECTORY;
                setMSXTimeStamp(idx0, fst);
                setMSXTimeStamp(idx1, fst);
                msxDir(idx0).startCluster = cluster;
                msxDir(idx1).startCluster = msxDirSector == firstDirSector
                                          ? 0 : sectorToCluster(msxDirSector);
        } else {
                if (!(msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY)) {
                        // Should rarely happen because checkDeletedHostFiles()
                        // recently checked this. (It could happen when a host
                        // directory is *just*recently* created with the same
                        // name as an existing msx file). Ignore, it will be
                        // corrected in the next sync.
                        return;
                }
                unsigned cluster = msxDir(dirIndex).startCluster;
                if ((cluster < FIRST_CLUSTER) || (cluster >= maxCluster)) {
                        // Sanity check on cluster range.
                        return;
                }
                newMsxDirSector = clusterToSector(cluster);
        }

        // Recursively process this directory.
        addNewHostFiles(strCat(hostSubDir, hostName, '/'), newMsxDirSector);
}

void DirAsDSK::addNewHostFile(const string& hostSubDir, const string& hostName,
                              unsigned msxDirSector, FileOperations::Stat& fst)
{
        if (checkFileUsedInDSK(hostSubDir + hostName)) {
                // File is already present in the virtual disk, do nothing.
                return;
        }
        string hostPath = hostSubDir + hostName;
        string fullHostName = hostDir + hostPath;

        // TODO check for available free space on disk instead of max free space
        static const int DISK_SPACE = (nofSectors - firstDataSector) * SECTOR_SIZE;
        if (fst.st_size > DISK_SPACE) {
                cliComm.printWarning("File too large: ", fullHostName);
                return;
        }

        DirIndex dirIndex = fillMSXDirEntry(hostSubDir, hostName, msxDirSector);
        importHostFile(dirIndex, fst);
}

DirAsDSK::DirIndex DirAsDSK::fillMSXDirEntry(
        const string& hostSubDir, const string& hostName, unsigned msxDirSector)
{
        string hostPath = hostSubDir + hostName;
        try {
                // Get empty dir entry (possibly extends subdirectory).
                DirIndex dirIndex = getFreeDirEntry(msxDirSector);

                // Create correct MSX filename.
                string msxFilename = hostToMsxName(hostName);
                if (checkMSXFileExists(msxFilename, msxDirSector)) {
                        // TODO: actually should increase vfat abrev if possible!!
                        throw MSXException(
                                "MSX name ", msxToHostName(msxFilename.c_str()),
                                " already exists");
                }

                // Fill in hostName / msx filename.
                assert(!StringOp::endsWith(hostPath, '/'));
                mapDirs[dirIndex].hostName = hostPath;
                memset(&msxDir(dirIndex), 0, sizeof(MSXDirEntry)); // clear entry
                memcpy(msxDir(dirIndex).filename, msxFilename.data(), 8 + 3);
                return dirIndex;
        } catch (MSXException& e) {
                throw MSXException("Couldn't add ", hostPath, ": ",
                                   e.getMessage());
        }
}

DirAsDSK::DirIndex DirAsDSK::getFreeDirEntry(unsigned msxDirSector)
{
        vector<bool> visited(nofSectors, false);
        while (true) {
                if (visited[msxDirSector]) {
                        // cycle detected, invalid disk, but don't crash on it
                        throw MSXException("cycle in FAT");
                }
                visited[msxDirSector] = true;

                for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                        DirIndex dirIndex(msxDirSector, idx);
                        const char* msxName = msxDir(dirIndex).filename;
                        if ((msxName[0] == char(0x00)) ||
                            (msxName[0] == char(0xE5))) {
                                // Found an unused msx entry. There shouldn't
                                // be any hostfile mapped to this entry.
                                assert(!mapDirs.count(dirIndex)); // c++20 contains()
                                return dirIndex;
                        }
                }
                unsigned sector = nextMsxDirSector(msxDirSector);
                if (sector == unsigned(-1)) break;
                msxDirSector = sector;
        }

        // No free space in existing directory.
        if (msxDirSector == (firstDataSector - 1)) {
                // Can't extend root directory.
                throw MSXException("root directory full");
        }

        // Extend sub-directory: allocate and clear a new cluster, add this
        // cluster in the existing FAT chain.
        unsigned cluster = sectorToCluster(msxDirSector);
        unsigned newCluster = getFreeCluster(); // throws if disk full
        unsigned sector = clusterToSector(newCluster);
        memset(&sectors[sector], 0, SECTORS_PER_CLUSTER * SECTOR_SIZE);
        writeFAT12(cluster, newCluster);
        writeFAT12(newCluster, EOF_FAT);

        // First entry in this newly allocated cluster is free. Return it.
        return {sector, 0};
}

void DirAsDSK::writeSectorImpl(size_t sector_, const SectorBuffer& buf)
{
        assert(sector_ < nofSectors);
        assert(syncMode != SYNC_READONLY);
        auto sector = unsigned(sector_);

        // Update last access time.
        if (auto* scheduler = diskChanger.getScheduler()) {
                lastAccess = scheduler->getCurrentTime();
        }

        DirIndex dirDirIndex;
        if (sector == 0) {
                // Ignore. We don't allow writing to the bootsector. It would
                // be very bad if the MSX tried to format this disk using other
                // disk parameters than this code assumes. It's also not useful
                // to write a different bootprogram to this disk because it
                // will be lost when this virtual disk is ejected.
        } else if (sector < firstSector2ndFAT) {
                writeFATSector(sector, buf);
        } else if (sector < firstDirSector) {
                // Write to 2nd FAT, only buffer it. Don't interpret the data
                // in FAT2 in any way (nor trigger any action on this write).
                memcpy(&sectors[sector], &buf, sizeof(buf));
        } else if (isDirSector(sector, dirDirIndex)) {
                // Either root- or sub-directory.
                writeDIRSector(sector, dirDirIndex, buf);
        } else {
                writeDataSector(sector, buf);
        }
}

void DirAsDSK::writeFATSector(unsigned sector, const SectorBuffer& buf)
{
        // Create copy of old FAT (to be able to detect changes).
        vector<SectorBuffer> oldFAT(nofSectorsPerFat);
        memcpy(&oldFAT[0], fat(), SECTOR_SIZE * nofSectorsPerFat);

        // Update current FAT with new data.
        memcpy(&sectors[sector], &buf, sizeof(buf));

        // Look for changes.
        for (unsigned i = FIRST_CLUSTER; i < maxCluster; ++i) {
                if (readFAT(i) != readFATHelper(oldFAT.data(), i)) {
                        exportFileFromFATChange(i, oldFAT.data());
                }
        }
        // At this point there should be no more differences.
        // Note: we can't use
        //   assert(memcmp(fat(), oldFAT, sizeof(oldFAT)) == 0);
        // because exportFileFromFATChange() only updates the part of the FAT
        // that actually contains FAT info. E.g. not the media ID at the
        // beginning nor the unsused part at the end. And for example the 'CALL
        // FORMAT' routine also writes these parts of the FAT.
        for (unsigned i = FIRST_CLUSTER; i < maxCluster; ++i) {
                assert(readFAT(i) == readFATHelper(oldFAT.data(), i));
        }
}

void DirAsDSK::exportFileFromFATChange(unsigned cluster, SectorBuffer* oldFAT)
{
        // Get first cluster in the FAT chain that contains 'cluster'.
        unsigned chainLength; // not used
        unsigned startCluster = getChainStart(cluster, chainLength);

        // Copy this whole chain from FAT1 to FAT2 (so that the loop in
        // writeFATSector() sees this part is already handled).
        vector<bool> visited(maxCluster, false);
        unsigned tmp = startCluster;
        while ((FIRST_CLUSTER <= tmp) && (tmp < maxCluster)) {
                if (visited[tmp]) {
                        // detected cycle, don't export file
                        return;
                }
                visited[tmp] = true;

                unsigned next = readFAT(tmp);
                writeFATHelper(oldFAT, tmp, next);
                tmp = next;
        }

        // Find the corresponding direntry and (if found) export file based on
        // new cluster chain.
        DirIndex dirIndex, dirDirIndex;
        if (getDirEntryForCluster(startCluster, dirIndex, dirDirIndex)) {
                exportToHost(dirIndex, dirDirIndex);
        }
}

unsigned DirAsDSK::getChainStart(unsigned cluster, unsigned& chainLength)
{
        // Search for the first cluster in the chain that contains 'cluster'
        // Note: worst case (this implementation of) the search is O(N^2), but
        // because usually FAT chains are allocated in ascending order, this
        // search is fast O(N).
        chainLength = 0;
        for (unsigned i = FIRST_CLUSTER; i < maxCluster; ++i) {
                if (readFAT(i) == cluster) {
                        // Found a predecessor.
                        cluster = i;
                        ++chainLength;
                        i = FIRST_CLUSTER - 1; // restart search
                }
        }
        return cluster;
}

// Generic helper function that walks over the whole MSX directory tree
// (possibly it stops early so it doesn't always walk over the whole tree).
// The action that is performed while walking depends on the functor parameter.
template<typename FUNC> bool DirAsDSK::scanMsxDirs(FUNC func, unsigned sector)
{
        size_t rdIdx = 0;
        vector<unsigned> dirs;  // TODO make vector of struct instead of
        vector<DirIndex> dirs2; //      2 parallel vectors.
        while (true) {
                do {
                        // About to process a new directory sector.
                        if (func.onDirSector(sector)) return true;

                        for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                                // About to process a new directory entry.
                                DirIndex dirIndex(sector, idx);
                                const MSXDirEntry& entry = msxDir(dirIndex);
                                if (func.onDirEntry(dirIndex, entry)) return true;

                                if ((entry.filename[0] == char(0x00)) ||
                                    (entry.filename[0] == char(0xE5)) ||
                                    !(entry.attrib & MSXDirEntry::ATT_DIRECTORY)) {
                                        // Not a directory.
                                        continue;
                                }
                                unsigned cluster = msxDir(dirIndex).startCluster;
                                if ((cluster < FIRST_CLUSTER) ||
                                    (cluster >= maxCluster)) {
                                        // Cluster=0 happens for ".." entries to
                                        // the root directory, also be robust for
                                        // bogus data.
                                        continue;
                                }
                                unsigned dir = clusterToSector(cluster);
                                if (contains(dirs, dir)) {
                                        // Already found this sector. Except
                                        // for the special "." and ".."
                                        // entries, loops should not occur in
                                        // valid disk images, but don't crash
                                        // on (intentionally?) invalid images.
                                        continue;
                                }
                                // Found a new directory, insert in the set of
                                // yet-to-be-processed directories.
                                dirs.push_back(dir);
                                dirs2.push_back(dirIndex);
                        }
                        sector = nextMsxDirSector(sector);
                } while (sector != unsigned(-1));

                // Scan next subdirectory (if any).
                if (rdIdx == dirs.size()) {
                        // Visited all directories.
                        return false;
                }
                // About to process a new subdirectory.
                func.onVisitSubDir(dirs2[rdIdx]);
                sector = dirs[rdIdx++];
        }
}

// Base class for functor objects to be used in scanMsxDirs().
// This implements all required methods with empty implementations.
struct NullScanner {
        // Called right before we enter a new subdirectory.
        void onVisitSubDir(DirAsDSK::DirIndex /*subdir*/) {}

        // Called when a new sector of a (sub)directory is being scanned.
        inline bool onDirSector(unsigned /*dirSector*/) {
                return false;
        }

        // Called for each directory entry (in a sector).
        inline bool onDirEntry(DirAsDSK::DirIndex /*dirIndex*/,
                               const MSXDirEntry& /*entry*/) {
                return false;
        }
};

// Base class for the IsDirSector and DirEntryForCluster scanner algorithms
// below. This class remembers the directory entry of the last visited subdir.
struct DirScanner : NullScanner {
        explicit DirScanner(DirAsDSK::DirIndex& dirDirIndex_)
                : dirDirIndex(dirDirIndex_)
        {
                dirDirIndex = DirAsDSK::DirIndex(0, 0); // represents entry for root dir
        }

        // Called right before we enter a new subdirectory.
        void onVisitSubDir(DirAsDSK::DirIndex subdir) {
                dirDirIndex = subdir;
        }

        DirAsDSK::DirIndex& dirDirIndex;
};

// Figure out whether a given sector is part of the msx directory structure.
struct IsDirSector : DirScanner {
        IsDirSector(unsigned sector_, DirAsDSK::DirIndex& dirDirIndex_)
                : DirScanner(dirDirIndex_)
                , sector(sector_) {}
        bool onDirSector(unsigned dirSector) {
                return sector == dirSector;
        }
        const unsigned sector;
};
bool DirAsDSK::isDirSector(unsigned sector, DirIndex& dirDirIndex)
{
        return scanMsxDirs(IsDirSector(sector, dirDirIndex), firstDirSector);
}

// Search for the directory entry that has the given startCluster.
struct DirEntryForCluster : DirScanner {
        DirEntryForCluster(unsigned cluster_,
                           DirAsDSK::DirIndex& dirIndex_,
                           DirAsDSK::DirIndex& dirDirIndex_)
                : DirScanner(dirDirIndex_)
                , cluster(cluster_)
                , result(dirIndex_) {}
        bool onDirEntry(DirAsDSK::DirIndex dirIndex, const MSXDirEntry& entry) {
                if (entry.startCluster == cluster) {
                        result = dirIndex;
                        return true;
                }
                return false;
        }
        const unsigned cluster;
        DirAsDSK::DirIndex& result;
};
bool DirAsDSK::getDirEntryForCluster(unsigned cluster,
                                     DirIndex& dirIndex, DirIndex& dirDirIndex)
{
        return scanMsxDirs(DirEntryForCluster(cluster, dirIndex, dirDirIndex),
                           firstDirSector);
}
DirAsDSK::DirIndex DirAsDSK::getDirEntryForCluster(unsigned cluster)
{
        DirIndex dirIndex, dirDirIndex;
        if (getDirEntryForCluster(cluster, dirIndex, dirDirIndex)) {
                return dirIndex;
        } else {
                return {unsigned(-1), unsigned(-1)}; // not found
        }
}

// Remove the mapping between the msx and host for all the files/dirs in the
// given msx directory (+ subdirectories).
struct UnmapHostFiles : NullScanner {
        explicit UnmapHostFiles(DirAsDSK::MapDirs& mapDirs_)
                : mapDirs(mapDirs_) {}
        bool onDirEntry(DirAsDSK::DirIndex dirIndex,
                        const MSXDirEntry& /*entry*/) {
                mapDirs.erase(dirIndex);
                return false;
        }
        DirAsDSK::MapDirs& mapDirs;
};
void DirAsDSK::unmapHostFiles(unsigned msxDirSector)
{
        scanMsxDirs(UnmapHostFiles(mapDirs), msxDirSector);
}

void DirAsDSK::exportToHost(DirIndex dirIndex, DirIndex dirDirIndex)
{
        // Handle both files and subdirectories.
        if (msxDir(dirIndex).attrib & MSXDirEntry::ATT_VOLUME) {
                // But ignore volume ID.
                return;
        }
        const char* msxName = msxDir(dirIndex).filename;
        string hostName;
        if (auto v = lookup(mapDirs, dirIndex)) {
                // Hostname is already known.
                hostName = v->hostName;
        } else {
                // Host file/dir does not yet exist, create hostname from
                // msx name.
                if ((msxName[0] == char(0x00)) || (msxName[0] == char(0xE5))) {
                        // Invalid MSX name, don't do anything.
                        return;
                }
                string hostSubDir;
                if (dirDirIndex.sector != 0) {
                        // Not the msx root directory.
                        auto v2 = lookup(mapDirs, dirDirIndex);
                        assert(v2);
                        hostSubDir = v2->hostName;
                        assert(!StringOp::endsWith(hostSubDir, '/'));
                        hostSubDir += '/';
                }
                hostName = hostSubDir + msxToHostName(msxName);
                mapDirs[dirIndex].hostName = hostName;
        }
        if (msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY) {
                if ((memcmp(msxName, ".          ", 11) == 0) ||
                    (memcmp(msxName, "..         ", 11) == 0)) {
                        // Don't export "." or "..".
                        return;
                }
                exportToHostDir(dirIndex, hostName);
        } else {
                exportToHostFile(dirIndex, hostName);
        }
}

void DirAsDSK::exportToHostDir(DirIndex dirIndex, const string& hostName)
{
        try {
                unsigned cluster = msxDir(dirIndex).startCluster;
                if ((cluster < FIRST_CLUSTER) || (cluster >= maxCluster)) {
                        // Sanity check on cluster range.
                        return;
                }
                unsigned msxDirSector = clusterToSector(cluster);

                // Create the host directory.
                string fullHostName = hostDir + hostName;
                FileOperations::mkdirp(fullHostName);

                // Export all the components in this directory.
                vector<bool> visited(nofSectors, false);
                do {
                        if (visited[msxDirSector]) {
                                // detected cycle, invalid disk, but don't crash on it
                                return;
                        }
                        visited[msxDirSector] = true;

                        if (readFAT(sectorToCluster(msxDirSector)) == FREE_FAT) {
                                // This happens e.g. on a TurboR when a directory
                                // is removed: first the FAT is cleared before
                                // the directory entry is updated.
                                return;
                        }
                        for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                                exportToHost(DirIndex(msxDirSector, idx), dirIndex);
                        }
                        msxDirSector = nextMsxDirSector(msxDirSector);
                } while (msxDirSector != unsigned(-1));
        } catch (FileException& e) {
                cliComm.printWarning("Error while syncing host directory: ",
                                     hostName, ": ", e.getMessage());
        }
}

void DirAsDSK::exportToHostFile(DirIndex dirIndex, const string& hostName)
{
        // We write a host file with length that is the minimum of:
        //  - Length indicated in msx directory entry.
        //  - Length of FAT-chain * cluster-size, this chain can have length=0
        //    if startCluster is not (yet) filled in.
        try {
                unsigned curCl = msxDir(dirIndex).startCluster;
                unsigned msxSize = msxDir(dirIndex).size;

                string fullHostName = hostDir + hostName;
                File file(fullHostName, File::TRUNCATE);
                unsigned offset = 0;
                vector<bool> visited(maxCluster, false);

                while ((FIRST_CLUSTER <= curCl) && (curCl < maxCluster)) {
                        if (visited[curCl]) {
                                // detected cycle, invalid disk, but don't crash on it
                                return;
                        }
                        visited[curCl] = true;

                        unsigned logicalSector = clusterToSector(curCl);
                        for (unsigned i = 0; i < SECTORS_PER_CLUSTER; ++i) {
                                if (offset >= msxSize) break;
                                unsigned sector = logicalSector + i;
                                assert(sector < nofSectors);
                                auto writeSize = std::min<size_t>(msxSize - offset, SECTOR_SIZE);
                                file.write(&sectors[sector], writeSize);
                                offset += SECTOR_SIZE;
                        }
                        if (offset >= msxSize) break;
                        curCl = readFAT(curCl);
                }
        } catch (FileException& e) {
                cliComm.printWarning("Error while syncing host file: ",
                                     hostName, ": ", e.getMessage());
        }
}

void DirAsDSK::writeDIRSector(unsigned sector, DirIndex dirDirIndex,
                              const SectorBuffer& buf)
{
        // Look for changed directory entries.
        for (unsigned idx = 0; idx < DIR_ENTRIES_PER_SECTOR; ++idx) {
                auto& newEntry = buf.dirEntry[idx];
                DirIndex dirIndex(sector, idx);
                if (memcmp(&msxDir(dirIndex), &newEntry, sizeof(newEntry)) != 0) {
                        writeDIREntry(dirIndex, dirDirIndex, newEntry);
                }
        }
        // At this point sector should be updated.
        assert(memcmp(&sectors[sector], &buf, sizeof(buf)) == 0);
}

void DirAsDSK::writeDIREntry(DirIndex dirIndex, DirIndex dirDirIndex,
                             const MSXDirEntry& newEntry)
{
        if (memcmp(msxDir(dirIndex).filename, newEntry.filename, 8 + 3) != 0 ||
            ((msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY) !=
             (        newEntry.attrib & MSXDirEntry::ATT_DIRECTORY))) {
                // Name or file-type in the direntry was changed.
                auto it = mapDirs.find(dirIndex);
                if (it != end(mapDirs)) {
                        // If there is an associated hostfile, then delete it
                        // (in case of a rename, the file will be recreated
                        // below).
                        string fullHostName = hostDir + it->second.hostName;
                        FileOperations::deleteRecursive(fullHostName); // ignore return value
                        // Remove mapping between msx and host file/dir.
                        mapDirs.erase(it);
                        if (msxDir(dirIndex).attrib & MSXDirEntry::ATT_DIRECTORY) {
                                // In case of a directory also unmap all
                                // sub-components.
                                unsigned cluster = msxDir(dirIndex).startCluster;
                                if ((FIRST_CLUSTER <= cluster) &&
                                    (cluster < maxCluster)) {
                                        unmapHostFiles(clusterToSector(cluster));
                                }
                        }
                }
        }

        // Copy the new msx directory entry.
        memcpy(&msxDir(dirIndex), &newEntry, sizeof(newEntry));

        // (Re-)export the full file/directory.
        exportToHost(dirIndex, dirDirIndex);
}

void DirAsDSK::writeDataSector(unsigned sector, const SectorBuffer& buf)
{
        assert(sector >= firstDataSector);
        assert(sector < nofSectors);

        // Buffer the write, whether the sector is mapped to a file or not.
        memcpy(&sectors[sector], &buf, sizeof(buf));

        // Get first cluster in the FAT chain that contains this sector.
        unsigned cluster, offset, chainLength;
        sectorToCluster(sector, cluster, offset);
        unsigned startCluster = getChainStart(cluster, chainLength);
        offset += (sizeof(buf) * SECTORS_PER_CLUSTER) * chainLength;

        // Get corresponding directory entry.
        DirIndex dirIndex = getDirEntryForCluster(startCluster);
        // no need to check for 'dirIndex.sector == unsigned(-1)'
        auto v = lookup(mapDirs, dirIndex);
        if (!v) {
                // This sector was not mapped to a file, nothing more to do.
                return;
        }

        // Actually write data to host file.
        string fullHostName = hostDir + v->hostName;
        try {
                File file(fullHostName, "rb+"); // don't uncompress
                file.seek(offset);
                unsigned msxSize = msxDir(dirIndex).size;
                if (msxSize > offset) {
                        auto writeSize = std::min<size_t>(msxSize - offset, sizeof(buf));
                        file.write(&buf, writeSize);
                }
        } catch (FileException& e) {
                cliComm.printWarning("Couldn't write to file ", fullHostName,
                                     ": ", e.getMessage());
        }
}

} // namespace openmsx