Crafter.DB/implementations/Crafter.DB-File.cpp

// SPDX-License-Identifier: LGPL-3.0-only
// SPDX-FileCopyrightText: Copyright (C) 2026 Catcrafts®

module;

#define _GNU_SOURCE 1
#include <cerrno>
#include <cstdio>
#include <cstring>
#include <fcntl.h>
#include <sys/file.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>

module Crafter.DB:File_impl;
import :File;
import :Error;
import std;

using namespace Crafter::DB;
namespace fs = std::filesystem;

namespace {
    std::error_code Errno() noexcept {
        return std::error_code(errno, std::generic_category());
    }

    [[noreturn]] void Fail(std::string_view what, const fs::path& path) {
        throw IOError(what, Errno(), path);
    }
}

// --- File -------------------------------------------------------------------

File::File(File&& other) noexcept : descriptor(other.descriptor), path(std::move(other.path)) {
    other.descriptor = -1;
}

File& File::operator=(File&& other) noexcept {
    if (this != &other) {
        Close();
        descriptor = other.descriptor;
        path = std::move(other.path);
        other.descriptor = -1;
    }
    return *this;
}

File::~File() {
    Close();
}

File File::Open(const fs::path& path, bool writable) {
    File file;
    file.path = path;
    // O_NOFOLLOW: a private cache file must not be swapped for a symlink to
    // somewhere we would then happily write 0600 data into.
    const std::int32_t flags = (writable ? O_RDWR : O_RDONLY) | O_CLOEXEC | O_NOFOLLOW;
    file.descriptor = ::open(path.c_str(), flags);
    if (file.descriptor == -1) Fail("cannot open", path);
    return file;
}

File File::CreateExclusive(const fs::path& path, fs::perms permissions) {
    File file;
    file.path = path;
    file.descriptor = ::open(path.c_str(), O_RDWR | O_CREAT | O_EXCL | O_CLOEXEC | O_NOFOLLOW, static_cast<mode_t>(permissions));
    if (file.descriptor == -1) Fail("cannot create", path);
    return file;
}

void File::LockExclusive() {
    if (::flock(descriptor, LOCK_EX | LOCK_NB) == -1) {
        if (errno == EWOULDBLOCK) throw LockedError(std::format("{}: another Table holds this file open for writing", path.string()));
        Fail("cannot lock", path);
    }
}

void File::LockShared() {
    if (::flock(descriptor, LOCK_SH | LOCK_NB) == -1) {
        if (errno == EWOULDBLOCK) throw LockedError(std::format("{}: another Table holds this file open for writing", path.string()));
        Fail("cannot lock", path);
    }
}

std::uint64_t File::Size() const {
    struct stat info{};
    if (::fstat(descriptor, &info) == -1) Fail("cannot stat", path);
    return static_cast<std::uint64_t>(info.st_size);
}

void File::Truncate(std::uint64_t length) {
    if (::ftruncate(descriptor, static_cast<off_t>(length)) == -1) Fail("cannot truncate", path);
}

void File::Allocate(std::uint64_t offset, std::uint64_t length) {
    if (::fallocate(descriptor, 0, static_cast<off_t>(offset), static_cast<off_t>(length)) == 0) return;
    if (errno != EOPNOTSUPP && errno != ENOSYS) Fail("cannot allocate space", path);
    // Filesystems without fallocate (FUSE, 9p, exFAT): posix_fallocate falls
    // back to writing zeros, which equally guarantees the blocks exist.
    const std::int32_t result = ::posix_fallocate(descriptor, static_cast<off_t>(offset), static_cast<off_t>(length));
    if (result != 0) throw IOError("cannot allocate space", std::error_code(result, std::generic_category()), path);
}

void File::ReadAt(std::uint64_t offset, std::span<std::byte> out) const {
    while (!out.empty()) {
        const ssize_t got = ::pread(descriptor, out.data(), out.size(), static_cast<off_t>(offset));
        if (got == -1) {
            if (errno == EINTR) continue;
            Fail("read failed", path);
        }
        if (got == 0) throw CorruptError(std::format("{}: file ends {} bytes short", path.string(), out.size()));
        out = out.subspan(static_cast<std::size_t>(got));
        offset += static_cast<std::uint64_t>(got);
    }
}

void File::WriteAt(std::uint64_t offset, std::span<const std::byte> in) {
    while (!in.empty()) {
        const ssize_t put = ::pwrite(descriptor, in.data(), in.size(), static_cast<off_t>(offset));
        if (put == -1) {
            if (errno == EINTR) continue;
            Fail("write failed", path);
        }
        in = in.subspan(static_cast<std::size_t>(put));
        offset += static_cast<std::uint64_t>(put);
    }
}

void File::SyncData() {
    if (::fdatasync(descriptor) == -1) Fail("fdatasync failed", path);
}

void File::Sync() {
    if (::fsync(descriptor) == -1) Fail("fsync failed", path);
}

void File::Close() noexcept {
    if (descriptor != -1) {
        ::close(descriptor);
        descriptor = -1;
    }
}

// --- MappedRegion -----------------------------------------------------------

MappedRegion::MappedRegion(MappedRegion&& other) noexcept : base(other.base), reserved(other.reserved), mapped(other.mapped), dontDump(other.dontDump) {
    other.base = nullptr;
    other.reserved = 0;
    other.mapped = 0;
}

MappedRegion& MappedRegion::operator=(MappedRegion&& other) noexcept {
    if (this != &other) {
        Release();
        base = other.base;
        reserved = other.reserved;
        mapped = other.mapped;
        dontDump = other.dontDump;
        other.base = nullptr;
        other.reserved = 0;
        other.mapped = 0;
    }
    return *this;
}

MappedRegion::~MappedRegion() {
    Release();
}

void MappedRegion::Reserve(std::uint64_t bytes) {
    Release();
    // PROT_NONE + MAP_NORESERVE costs no memory and no commit charge: it only
    // claims the addresses, so later MAP_FIXED calls cannot collide with
    // anything else the process maps.
    void* address = ::mmap(nullptr, bytes, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
    if (address == MAP_FAILED) throw IOError(std::format("cannot reserve {} bytes of address space", bytes), Errno(), fs::path());
    base = static_cast<std::byte*>(address);
    reserved = bytes;
    mapped = 0;
}

void MappedRegion::Map(const File& file, std::uint64_t fileOffset, std::uint64_t length, bool writable) {
    if (mapped + length > reserved) {
        throw IOError(std::format("table needs more than the {} bytes of address space reserved per region; reopen with a larger OpenOptions::reserveBytes", reserved), std::make_error_code(std::errc::not_enough_memory), file.Path());
    }
    void* address = ::mmap(base + mapped, length, writable ? PROT_READ | PROT_WRITE : PROT_READ, MAP_SHARED | MAP_FIXED, file.Descriptor(), static_cast<off_t>(fileOffset));
    if (address == MAP_FAILED) Fail("cannot map", file.Path());
    if (dontDump) ::madvise(address, length, MADV_DONTDUMP);
    mapped += length;
}

// MADV_DONTDUMP only. MADV_DONTFORK was considered and rejected: it turns any
// touch of an inherited table in a forked child into SIGSEGV. The mapping is
// MAP_SHARED, so a child reads consistently; writes are refused by the
// owner-process check in TableFile.
void MappedRegion::ExcludeFromCoreDumps() {
    dontDump = true;
    if (mapped != 0) ::madvise(base, mapped, MADV_DONTDUMP);
}

void MappedRegion::WillNeed(std::uint64_t length) {
    if (length != 0) ::madvise(base, std::min(length, mapped), MADV_WILLNEED);
}

void MappedRegion::Release() noexcept {
    if (base != nullptr) {
        ::munmap(base, reserved);
        base = nullptr;
        reserved = 0;
        mapped = 0;
    }
}

// --- MappedPages ------------------------------------------------------------

MappedPages::MappedPages(MappedPages&& other) noexcept : data(other.data), length(other.length) {
    other.data = nullptr;
    other.length = 0;
}

MappedPages& MappedPages::operator=(MappedPages&& other) noexcept {
    if (this != &other) {
        Release();
        data = other.data;
        length = other.length;
        other.data = nullptr;
        other.length = 0;
    }
    return *this;
}

MappedPages::~MappedPages() {
    Release();
}

void MappedPages::Map(const File& file, std::uint64_t fileOffset, std::uint64_t mapLength, bool writable) {
    Release();
    void* address = ::mmap(nullptr, mapLength, writable ? PROT_READ | PROT_WRITE : PROT_READ, MAP_SHARED, file.Descriptor(), static_cast<off_t>(fileOffset));
    if (address == MAP_FAILED) Fail("cannot map header", file.Path());
    data = static_cast<std::byte*>(address);
    length = mapLength;
}

void MappedPages::Sync() {
    if (data != nullptr && ::msync(data, length, MS_SYNC) == -1) throw IOError("msync failed", Errno(), fs::path());
}

void MappedPages::Release() noexcept {
    if (data != nullptr) {
        ::munmap(data, length);
        data = nullptr;
        length = 0;
    }
}

// --- Os ---------------------------------------------------------------------

void Os::CreateDirectories(const fs::path& directory, fs::perms permissions) {
    if (directory.empty()) return;
    fs::path current;
    for (const fs::path& part : directory) {
        current /= part;
        if (current == directory.root_path() || current.empty()) continue;
        if (::mkdir(current.c_str(), static_cast<mode_t>(permissions)) == -1 && errno != EEXIST) Fail("cannot create directory", current);
    }
}

void Os::Rename(const fs::path& from, const fs::path& to) {
    if (::rename(from.c_str(), to.c_str()) == -1) Fail(std::format("cannot rename to {}", to.string()), from);
}

bool Os::RenameNoReplace(const fs::path& from, const fs::path& to) {
    if (::renameat2(AT_FDCWD, from.c_str(), AT_FDCWD, to.c_str(), RENAME_NOREPLACE) == 0) return true;
    if (errno == EEXIST) return false;
    Fail(std::format("cannot rename to {}", to.string()), from);
}

void Os::HardLink(const fs::path& from, const fs::path& to) {
    if (::link(from.c_str(), to.c_str()) == -1) Fail(std::format("cannot link to {}", to.string()), from);
}

bool Os::Remove(const fs::path& path) noexcept {
    return ::unlink(path.c_str()) == 0;
}

void Os::SyncDirectory(const fs::path& directory) {
    const fs::path dir = directory.empty() ? fs::path(".") : directory;
    const std::int32_t fd = ::open(dir.c_str(), O_RDONLY | O_DIRECTORY | O_CLOEXEC);
    if (fd == -1) Fail("cannot open directory", dir);
    const std::int32_t result = ::fsync(fd);
    ::close(fd);
    if (result == -1) Fail("fsync of directory failed", dir);
}

void Os::RandomBytes(std::span<std::byte> out) {
    std::random_device device;
    for (std::size_t i = 0; i < out.size(); i += 4) {
        const std::uint32_t value = device();
        std::memcpy(out.data() + i, &value, std::min<std::size_t>(4, out.size() - i));
    }
}

std::uint64_t Os::UnixNanos() noexcept {
    return static_cast<std::uint64_t>(std::chrono::duration_cast<std::chrono::nanoseconds>(std::chrono::system_clock::now().time_since_epoch()).count());
}

std::uint64_t Os::PageSize() noexcept {
    const long size = ::sysconf(_SC_PAGESIZE);
    return size > 0 ? static_cast<std::uint64_t>(size) : 4096;
}

std::int64_t Os::ProcessId() noexcept {
    return static_cast<std::int64_t>(::getpid());
}