v0.0.14

v0.0.14 fix: ignore SIGHUP to prevent unexpected termination on daemon reload

• register SIG_IGN for SIGHUP alongside existing SIGPIPE handling
• prevent process stop when OpenWrt service reload or terminal disconnect triggers SIGHUP

v0.0.14 fix: 忽略 SIGHUP 信号，防止守护进程重载时意外退出

• 在已有 SIGPIPE 忽略基础上增加 SIGHUP 信号忽略注册
• 避免 OpenWrt 服务重载或终端断开时 SIGHUP 导致进程退出

v0.0.13

v0.0.13 feat: add DNS resolver system with Fake-IP bypass, TLS SNI support, and connection tracking

• add complete DNS resolver system: Fake-IP detection (198.18.0.0/15), DoH via curl (AliDNS, Cloudflare), UDP DNS (223.5.5.5, 119.29.29.29, 1.1.1.1, 8.8.8.8), 12 hardcoded Baidu proxy fallback IPs
• add baidu_resolver config field with custom resolver spec support (doh://IP/path?host=HOST, udp://IP)
• enhance resolve_host_ips() with Fake-IP detection and automatic fallback chain
• add TLS SNI support: build_tls_client_hello, tls_probe_host, read_tls_client_hello, parse_tls_sni
• add kqueue EV_CLEAR flag for macOS (edge-triggered, matching Linux EPOLLET)
• add connection tracking: conn_id, client_addr in ConnCtx, global atomic g_conn_seq counter
• refactor Baidu proxy pool: build_baidu_pool_from_ips(), carrier_init_baidu_pool(), carrier_check_baidu_ipv6_support()
• add SNI-aware IP selection: carrier_tls_sni_check_ip(), carrier_choose_ip_for_sni()
• add TLS-aware health check: port 443 uses tls_probe() instead of http_probe()
• enhance connection accept logging with sequential conn_id and client_addr

v0.0.13 feat: 新增 DNS 解析系统支持 Fake-IP 绕过、TLS SNI 支持与连接追踪

• 新增完整 DNS 解析系统：Fake-IP 检测 (198.18.0.0/15)、DoH via curl (AliDNS、Cloudflare)、UDP DNS (223.5.5.5、119.29.29.29、1.1.1.1、8.8.8.8)、12 个硬编码百度代理备用 IP
• 新增 baidu_resolver 配置字段，支持自定义解析器规格 (doh://IP/path?host=HOST, udp://IP)
• 增强 resolve_host_ips()，支持 Fake-IP 检测与自动回退链
• 新增 TLS SNI 支持：build_tls_client_hello、tls_probe_host、read_tls_client_hello、parse_tls_sni
• macOS kqueue 增加 EV_CLEAR 标志 (边缘触发，与 Linux EPOLLET 一致)
• 新增连接追踪：ConnCtx 增加 conn_id、client_addr，全局原子计数器 g_conn_seq
• 重构百度代理池：build_baidu_pool_from_ips()、carrier_init_baidu_pool()、carrier_check_baidu_ipv6_support()
• 新增 SNI 感知 IP 选择：carrier_tls_sni_check_ip()、carrier_choose_ip_for_sni()
• 新增 TLS 感知健康检查：443 端口使用 tls_probe() 替代 http_probe()
• 增强连接接收日志，包含顺序 conn_id 与 client_addr

v0.0.12

v0.0.12 feat: add support for interface binding, IPv6, TLS SNI, and connection tracking

• add -bind-if interface binding support with cross-platform implementation
• add IPv6 tool functions: rand_u64, ipv6_cidr_sample, is_ip_literal, format_host_literal, format_addr_port
• add send_all_timeout for reliable timed data transmission
• add TLS SNI feature: ClientHello construction, SNI parsing, tls_probe, recv_more_timeout
• add conn_msg connection-level logging and extend ConnCtx with client_addr, runtime, conn_id
• upgrade connection_thread with SNI-based routing, send_all_timeout, fast-fail detection
• enhance scan_worker HTTP probe with alternating Host headers
• make carrier_probe_and_check TLS-aware (tls_probe for port 443, http_probe otherwise)
• add carrier_tls_sni_check_ip and carrier_choose_ip_for_sni for SNI-aware IP selection
• add carrier_note_connection_result for fast-disconnect tracking (<3000ms)
• update carrier_accept_thread to populate client_addr, runtime, conn_id

v0.0.12 feat: 新增接口绑定、IPv6、TLS SNI、连接追踪功能

• 新增 -bind-if 出站接口绑定功能，跨平台实现
• 新增 IPv6 工具函数：rand_u64、ipv6_cidr_sample、is_ip_literal、format_host_literal、format_addr_port
• 新增 send_all_timeout 可靠超时发送
• 新增 TLS SNI 功能：ClientHello 构造、SNI 解析、tls_probe、recv_more_timeout
• 新增 conn_msg 连接级日志，扩展 ConnCtx 增加 client_addr、runtime、conn_id 字段
• 升级 connection_thread：支持 SNI 路由、send_all_timeout、快速断开检测
• 增强 scan_worker HTTP 探测：交替使用不同 Host 头
• carrier_probe_and_check 支持 TLS 感知探测（443 端口使用 tls_probe）
• 新增 carrier_tls_sni_check_ip 和 carrier_choose_ip_for_sni SNI 路由
• 新增 carrier_note_connection_result 快速断开追踪（<3000ms）
• 更新 carrier_accept_thread 填充 client_addr、runtime、conn_id

v0.0.11

v0.0.11 fix: improve EventLoop header receiving, colo selection, and startup retry on empty scan

• add platform guard and fd validation to recv_headers_ev, fallback to blocking read on unsupported platforms
• fix colo selection to use <= latency comparison so same-latency colos are also recorded
• retry scan with 3s delay when carrier mode finds no candidates or health checks all fail
• free stale IP list and reload IP range before each retry
• skip candidate cache print in single-scan mode when result set is empty

v0.0.11 fix: 改进 EventLoop 响应头接收、colo 选择逻辑及空扫描启动重试

• recv_headers_ev 增加平台守卫和 fd 校验，不支持的平台回退阻塞读取
• 修复 colo 选择条件，使用 <= 比较延迟，同等延迟的 colo 也能被记录
• 监听模式扫描不到候选或健康检查全部失败时，等待 3 秒后重试
• 每次重试前释放旧 IP 列表并重新加载 IP 段
• 单次扫描模式结果为空时跳过候选池打印

v0.0.10

v0.0.10 perf: rewrite scan engine with EventLoop, CIDR lazy expansion, and Keep-Alive

• replace select() with epoll/kqueue/select conditional compilation for I/O
• replace BatchIterator with CidrList (prefix-sum + binary search, O(log n))
• reuse HTTP Keep-Alive connections across probes, reconnect on disconnect
• add EWMA scoring (α=0.125) with jitter smoothing and exponential fail penalty
• add 5-second proxy node selection cache to BaiduProxyPool

v0.0.10 perf: 重写扫描引擎，引入 EventLoop、CIDR 惰性展开与 Keep-Alive

• select() 替换为 epoll/kqueue/select 条件编译事件循环
• BatchIterator 替换为 CidrList（前缀和 + 二分查找，O(log n)）
• 复用 HTTP Keep-Alive 连接探测，断连自动重连
• 新增 EWMA 评分（α=0.125），含抖动平滑与指数级失败惩罚
• 百度前置代理池新增 5 秒节点选择缓存

v0.0.9

refactor: release v0.0.9 — EventLoop abstraction, BatchIterator, cache-valid optimization, and race fix

• add cross-platform EventLoop I/O framework (epoll / kqueue / IOCP) replacing select()
• add BatchIterator for lazy CIDR expansion (1024 IPs per batch), reducing memory usage
• add cache_valid flag to CandidatePool, skipping redundant health checks on known-good IPs
• add recv_timeout() with configurable timeout, replace blocking recv in pipe_worker and connection_thread
• use Linux splice() for zero-copy forwarding in pipe_worker; unify all platforms on recv_timeout(300s)
• fix race between carrier_rescan_and_select_ip free(items) and carrier_choose_ip_for_connection items traversal by protecting with candidates.mu
• remove global candidate state (g_candidates, g_current_ip, etc.) in favor of per-pool management
• add version string (CFNAT_VERSION "0.0.9") and -V / -version flag

refactor: 发布 v0.0.9 — 引入 EventLoop 抽象层、BatchIterator、缓存有效优化与竞态修复

• 新增跨平台 EventLoop 事件驱动 I/O 框架（epoll / kqueue / IOCP），替代 select() 模型
• 新增 BatchIterator 惰性 CIDR 批量迭代（每批 1024 个 IP），降低内存占用
• CandidatePool 新增 cache_valid 标志，健康 IP 跳过冗余检测，直接复用
• 新增 recv_timeout() 可配置超时接收，替换 pipe_worker 与 connection_thread 中的阻塞 recv
• pipe_worker Linux 平台使用 splice() 零拷贝转发；全平台统一 recv_timeout(300s)
• 修复 carrier_rescan_and_select_ip 中 free(items) 与 carrier_choose_ip_for_connection 遍历 items 的竞态，使用 candidates.mu 互斥锁保护
• 移除全局候选状态（g_candidates、g_current_ip 等），状态归入 CandidatePool 管理
• 新增版本号（CFNAT_VERSION "0.0.9"）及 -V / -version 命令行参数

v0.0.8

fix: v0.0.8 align health check timeout with user-configured delay_ms

• replace hardcoded 2000ms timeout in health_check_ip and carrier_health_check_ip with g_cfg.delay_ms
• fix mismatch where a low -delay (e.g. 500ms) kept an IP alive in health checks while client connections timed out against the same IP
• apply the fix to all 8 dial + probe calls across Baidu and Carrier health check paths

fix: v0.0.8 健康检查超时改为与用户设置的 -delay 一致

• 将 health_check_ip 和 carrier_health_check_ip 中硬编码的 2000ms 超时替换为 g_cfg.delay_ms
• 修复当 -delay 设为较低值（如 500ms）时，健康检查用 2000ms 判定 IP 可用，但客户端连接用 500ms 超时而连不上的问题
• 同步修复百度直连模式和运营商模式共 8 处 dial + probe 调用

v0.0.7

feat: release v0.0.7 with dual listening mode (direct + baidu proxy)

• support -direct-listen and -baidu-listen parameters
• add mixed mode when direct-listen and baidu-listen are the same address
• refactor carrier listening logic and health check for better clarity and stability
• simplify Config structure and remove legacy multi-carrier resolver code
• improve connection selection and fallback between direct and baidu proxy

feat: 发布 v0.0.7，支持双监听模式（直连 + 百度前置代理）

• 支持 -direct-listen 和 -baidu-listen 参数
• 新增两个监听地址相同时的 mixed 混合模式
• 重构运营商监听逻辑与健康检查，提升代码清晰度和稳定性
• 精简 Config 结构体，移除遗留的多运营商解析代码
• 优化直连与百度前置代理的连接选择和回退机制

v0.0.6

feat: release v0.0.6 with carrier listen priority over Baidu proxy config

• force -baidu-proxy=true automatically when -carrier-listens is configured
• make carrier split-port mode take priority over explicit -baidu-proxy=false
• add startup notice when carrier listen mode enables Baidu proxy automatically
• simplify carrier listen examples by removing the need to set -baidu-proxy=true
• update README to document -carrier-listens priority and automatic Baidu proxy behavior

feat: 发布 v0.0.6，让运营商监听优先于百度代理配置

• 配置 -carrier-listens 时自动强制启用 -baidu-proxy=true
• 让运营商分端口模式优先于显式设置的 -baidu-proxy=false
• 添加启动提示，说明运营商监听模式会自动启用百度前置代理
• 简化运营商分池示例，不再需要手动设置 -baidu-proxy=true
• 更新 README，说明 -carrier-listens 的优先级和自动启用百度前置代理行为

v0.0.5

feat: release v0.0.5 with unified C source and simplified cache startup

• merge Linux, macOS, and Windows implementations into cfnat.c
• replace split source files with platform-specific conditional compilation
• keep Windows support for Winsock2, WinINet, DNS API, and Unicode console output
• keep Linux and macOS support for POSIX sockets, pthreads, and built-in DNS TXT lookup
• update GitHub Actions to build all platforms from cfnat.c
• update release packaging to include the unified source file
• refresh README with single-source build commands and platform notes

feat: 发布 v0.0.5，合并 C 源码并简化缓存启动逻辑

• 将 Linux、macOS、Windows 三个平台实现合并到 cfnat.c
• 使用平台条件编译替代分裂源码维护
• 保留 Windows 的 Winsock2、WinINet、DNS API 和 Unicode 控制台输出支持
• 保留 Linux 与 macOS 的 POSIX socket、pthread 和内置 DNS TXT 查询支持
• 更新 GitHub Actions，所有平台统一从 cfnat.c 构建
• 更新 release 打包，只附带统一后的源码文件
• 更新 README，补充单源码构建命令和平台说明