Skip to content

HighWall v0.11.0 + L4Wall 0.1.0

Latest

Choose a tag to compare

@Fallen-leaves089 Fallen-leaves089 released this 30 Oct 04:53
18e9f7e

HighWall v0.11.0 + L4Wall 0.1.0

L4Wall 独立发布(2026-11-03):L4Wall 0.1.0 已在 l4wall-v0.1.0 单独发布。本页原有 L4Wall wheel/sdist 为兼容旧链接保留,字节未变;新下载请优先使用独立 Release。该调整不改变 HighWall 0.11.0 的代码或附件。

Standalone L4Wall release (2026-11-03): L4Wall 0.1.0 is now published separately at l4wall-v0.1.0. The L4Wall wheel and sdist on this page remain unchanged for link compatibility. Prefer the standalone release for new downloads. This update does not change the HighWall 0.11.0 code or assets.
English release notes | 简体中文

发布日期:2026-10-30

概览

v0.11.0 把仓库从单一 HTTP/1.1 防护网关扩展为可组合的两层防护基线:

  • L4Wall 0.1.0:失败关闭的 TCP 连接级网关,负责准入、并发/速率限制、超时、字节上限、行为评分、封禁和审计。
  • HighWall 0.11.0:继续负责 HTTP/1.1 的 Host/方法/路径策略、WAF、DLP、响应完整性、L7 限速和蜜罐。
  • 复合部署:提供 lite 串联形态和 standard 边缘代理骨架;两者都不把“部署了组件”等同于“已经覆盖所有入口”。

新增

L4Wall 子项目

  • 连接级来源白名单和默认拒绝。
  • 单 IP 并发上限、新建连接速率和连接突发控制。
  • 首字节、空闲、总连接时长和单方向字节上限。
  • 多 listener 扇出、短连接、零字节探测、拒绝风暴和固定节奏等行为信号。
  • 两阶段评分、限时遏制、确认封禁和 offenders.json 持久化。
  • JSONL 审计、回环管理面、健康检查和 Prometheus 文本指标。
  • 可选私网 Redis 共享封禁读取端:版本化 JSON、严格字段校验、容量限制、TTL 截断、本地优先和故障隔离。

复合部署与攻防验收

  • scripts/attack_range.py:本机回环靶场,覆盖 L4Wall -> HighWall -> 模拟上游的 28 项矩阵;支持独立 WSL 攻击实例和常驻模式。
  • deploy/standard/:HAProxy、Compose、TLS 变体、上游回声服务和旁路验收脚本。
  • docs/STACK.md:信任边界、真实客户端 IP、直连封堵、封禁同步、故障策略和验收清单。
  • docs/ATTACK_TESTING.md:自动化矩阵、实机结果、已修复问题和未验证项。

修复与加固

  • 全局在途请求超限时,503 响应现在带 Retry-After: 1 和 X-Request-ID;审计同步记录 request_id、HW-SERVER-1001、status=503、detail=inflight_limit 和真实响应字节数。
  • l4wall inspect 不再输出密码或管理令牌的环境变量名,并增加脱敏回归测试。
  • 靶场把 L4 单方向字节上限调整到 2 MiB,使 HighWall 的 413 在临界区间可观测。
  • 发布包补充 subprojects/,sdist 不会再漏掉 L4Wall 源码。

验证

  • HighWall 全量:354 项测试通过。
  • L4Wall:65 项测试通过。
  • 跨层攻防:26 项端到端测试通过。
  • ruff、mypy、compileall、配置检查和干净虚拟环境安装通过。
  • PR CI:Ubuntu/Windows x Python 3.10/3.11/3.12、CodeQL、静态分析全部通过。
  • 实机矩阵:28/28 通过,0 失败;2 项已知边界风险已记录,未计作通过项。
  • 40 并发复测:31 个 200、1 个可追溯 503、8 个无 HTTP 响应;L4 侧 32 个允许、8 个 ip_capacity 拒绝。
  • 审计对账:在途请求允许/拒绝记录的 request_id 与客户端 X-Request-ID 32/32 对齐。
  • 字节边界:1,048,491 字节返回 200,1,048,577 字节返回 413(HW-PROTO-1106)。

下载

下载文件在 v0.11.0 Release 的 Assets 区域。下载后先用 SHA256SUMS 校验,再安装:

python -m pip install highwall-0.11.0-py3-none-any.whl
python -m pip install l4wall-0.1.0-py3-none-any.whl

容器镜像

v0.11.0 标签会触发多架构镜像构建并发布到 GHCR:

docker pull ghcr.io/fallen-leaves089/highwall:0.11.0

兼容性

  • HighWall 的 HTTP/1.1 配置 schema 未破坏性变更;v0.10.0 配置可继续使用。
  • L4Wall 是独立进程和独立 0.1.0 包,不要求 HighWall 进程内加载。
  • lite 形态下 HighWall 看到的来源是 L4Wall 地址;需要真实客户端归属时使用 standard 骨架。

边界与未覆盖项

  • HighWall 0.11.0 和 L4Wall 0.1.0 都是 alpha 阶段,未经过独立第三方安全审计。
  • L4Wall 不处理带宽型 DDoS、SYN flood、UDP、QUIC/HTTP3,也不替代应用认证、授权或主机 EDR。
  • standard 骨架当前只保证入口、真实 IP、网络闭环、L7 防护和验收;L4Wall Controller、动态封禁下发和多节点 HA 仍是后续里程碑。
  • 共享封禁目前只有读取端,自动发布、冲突合并和 Controller 尚未实现。
  • Docker/HAProxy 的完整容器拓扑仍需在目标环境执行联调;本地回环通过不等于生产拓扑通过。
  • 直连 HighWall 或上游会绕过 L4 ACL;必须用网络 ACL 和监听地址保证攻击者不可达。
  • 本版本不声称“绝对无漏”。未验证项和残余风险以 docs/ATTACK_TESTING.md 与 docs/STACK.md 为准。

English

Release date: 2026-10-30

Overview

v0.11.0 expands the repository from a single HTTP/1.1 gateway into a composable two-layer defense baseline:

  • L4Wall 0.1.0: a fail-closed TCP connection-level gateway for admission, concurrency and rate limits, timeouts, byte caps, behavior scoring, bans, and audit.
  • HighWall 0.11.0: continues to handle HTTP/1.1 Host/method/path policy, WAF, DLP, response integrity, L7 rate limits, and honeypots.
  • Composite deployment: provides the lite chain and a standard edge-proxy skeleton. Neither treats “components deployed” as “all ingress covered.”

Added

L4Wall subproject

  • Connection-level source allowlists and default deny.
  • Per-IP concurrency caps, new-connection rates, and burst control.
  • First-byte, idle, total-duration, and per-direction byte limits.
  • Behavior signals for multi-listener fan-out, short connections, zero-byte probes, rejection storms, and regular timing.
  • Two-stage scoring, timed containment, confirmed bans, and offenders.json persistence.
  • JSONL audit, a loopback admin plane, health checks, and Prometheus text metrics.
  • An optional private-network Redis shared-ban reader with versioned JSON, strict validation, capacity limits, TTL clamping, local-first merge, and failure isolation.

Composite deployment and attack validation

  • scripts/attack_range.py: a local loopback range with a 28-item matrix across L4Wall -> HighWall -> a stub upstream; supports an isolated WSL attacker instance and a long-running mode.
  • deploy/standard/: HAProxy, Compose, a TLS variant, an upstream echo service, and a bypass verification script.
  • docs/STACK.md: trust boundaries, trusted client IPs, direct-access blocking, ban synchronization, failure policy, and an acceptance checklist.
  • docs/ATTACK_TESTING.md: the automated matrix, live results, fixed issues, and unverified items.

Fixes and hardening

  • When the global in-flight limit is exceeded, the 503 response now includes Retry-After: 1 and X-Request-ID; the audit record includes request_id, HW-SERVER-1001, status=503, detail=inflight_limit, and the real response byte count.
  • l4wall inspect no longer prints password or admin-token environment variable names, with a redaction regression test.
  • The range raises the L4 per-direction byte cap to 2 MiB so the HighWall 413 boundary is observable.
  • Release packaging includes subprojects/, so the sdist no longer omits L4Wall sources.

Verification

  • HighWall: 354 tests passed.
  • L4Wall: 65 tests passed.
  • Cross-layer attack/defense: 26 end-to-end tests passed.
  • ruff, mypy, compileall, configuration checks, and installation in a clean virtual environment passed.
  • PR CI: Ubuntu and Windows on Python 3.10/3.11/3.12, CodeQL, and static analysis all passed.
  • Live matrix: 28/28 passed, 0 failed; 2 known boundary risks are recorded separately and are not counted as passes.
  • 40-connection retest: 31 200 responses, 1 traceable 503, and 8 connections without an HTTP response; L4 admitted 32 and rejected 8 with ip_capacity.
  • Audit correlation: request IDs for allowed and blocked in-flight requests matched the client X-Request-ID in 32/32 cases.
  • Byte boundary: 1,048,491 bytes returned 200; 1,048,577 bytes returned 413 (HW-PROTO-1106).

Downloads

The files are in the Assets section of the v0.11.0 release. Verify them against SHA256SUMS before installation:

python -m pip install highwall-0.11.0-py3-none-any.whl
python -m pip install l4wall-0.1.0-py3-none-any.whl

Container image

The v0.11.0 tag builds a multi-architecture image and publishes it to GHCR:

docker pull ghcr.io/fallen-leaves089/highwall:0.11.0
  • Digest (linux/amd64 + linux/arm64 OCI index): sha256:efedb3b9cca32a65f086c1a2ba89ae3c268a45e7a94a4158acfa5e8cda6c2dfc
  • Build workflow: https://github.com/Fallen-leaves089/highwall/actions/runs/41005104183
  • The image includes an SBOM, a GitHub provenance attestation, and a cosign keyless signature. Pin an immutable version in production instead of relying on latest.

Compatibility

  • The HighWall HTTP/1.1 configuration schema has no breaking changes; v0.10.0 configurations remain usable.
  • L4Wall is a separate process and a separate 0.1.0 package; HighWall does not load it in-process.
  • In the lite topology, HighWall sees L4Wall as its peer address. Use the standard skeleton when real client identity is required.

Boundaries and unverified items

  • HighWall 0.11.0 and L4Wall 0.1.0 are alpha releases and have not received an independent third-party security audit.
  • L4Wall does not handle bandwidth DDoS, SYN floods, UDP, or QUIC/HTTP3, and does not replace application authentication, authorization, or host EDR.
  • The standard skeleton currently covers ingress, trusted client IPs, network closure, L7 protection, and acceptance checks. The L4Wall Controller, dynamic ban publishing, and multi-node HA remain future milestones.
  • Shared bans currently have a reader only. Automatic publishing, conflict merging, and the Controller are not implemented.
  • The full Docker/HAProxy container topology still needs integration testing in the target environment; loopback success does not prove production-topology success.
  • Direct access to HighWall or the upstream bypasses the L4 ACL; network ACLs and listener addresses must make those paths unreachable to attackers.
  • This release does not claim “no bypasses.” See docs/ATTACK_TESTING.md and docs/STACK.md for unverified items and residual risk.