Skip to content

HighWall 0.12.0 (Alpha)

Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 02 Nov 01:39
· 27 commits to main since this release
v0.12.0
821048d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

HighWall 0.12.0 Alpha 预发布

English | 简体中文

状态：GitHub pre-release（Alpha）。这是可供评审和验证的预发布，不是生产发布，不附带生产支持承诺。最新稳定版仍是 0.11.0，本次发布不会移动 latest。

本版本包含

  • 精确客户端 IP + 请求形状的执法键；/24、/64 只用于关联，不自动扩大封禁。
  • highwall ops preview 只读 Safe Ops 预演；R2/R3 默认拒绝，没有执行入口。
  • highwall capabilities --json 描述 Python H1 基线的 C0-C4 等级；原生核心能力单独按 H1/H2/H3 声明。
  • 独立 subprojects/l5wall 0.2.0：TLS 终止、SNI/ALPN/mTLS 策略、握手超时、回环 H1 桥接、SNI/Host 一致性、请求头/连接/字节上限。L5Wall 继续作为独立服务发布，不并入 Python wheel。
  • 阶段 1 的 l5wall serve 只允许回环监听和回环上游，只接受 http/1.1 ALPN；H2/H3 不会被伪装成已检查。
  • 回环桥接不传递原始客户端 IP；不能在其后启用按 IP 封禁、来源 ACL 或依赖真实来源的限速。
  • L4Wall 可选 send_proxy_v2，支持向上游发送标准 PROXY v2 TCP header；原生核心可要求 PROXY v2，并只信任显式配置的代理网段。
  • Rust 原生 B+D-4 核心：TLS/QUIC、H1/H2/H3 严格解析、H1 WebSocket、H2/H3 gRPC、压缩/trailers、有界正文 WAF/DLP、精确 IP 封禁、mTLS/CRL、QUIC Retry、默认禁 0-RTT/迁移和 QUIC 客户端网络 ACL。
  • TCP 和 UDP 是两条入口：H1/H2 可经 L4Wall 和 PROXY v2；H3 当前直达原生 QUIC listener，不经过 L4Wall。
  • Python 镜像保持 Python-only；原生核心使用独立 Dockerfile.native，以非 root 运行，不把 Rust 依赖塞入 Python wheel 或镜像。

发布资产

GitHub Release 预期包含：

  • highwall-0.12.0-py3-none-any.whl 与 highwall-0.12.0.tar.gz。
  • Linux x86_64 原生二进制 highwall-native-core-0.12.0-linux-x86_64。
  • Windows x86_64 原生二进制 highwall-native-core-0.12.0-windows-x86_64.exe。
  • 原生核心示例配置 highwall-native-config-0.12.0.example.json。
  • 覆盖上述资产的 SHA256SUMS。

发布工作流会为资产生成 GitHub build provenance attestation。容器工作流另外发布 Python-only 镜像 ghcr.io/fallen-leaves089/highwall:0.12.0（linux/amd64、linux/arm64），带 OCI provenance、SBOM 和 cosign 签名；它不是完整 B+D-4 镜像。Dockerfile.native 本轮未作为公开 GHCR 镜像发布，原生路径以 Release 中的独立二进制为准。

验证

  • 根 Python：384 项测试通过；coverage --branch 总覆盖率 81%，高于项目 80% 门槛；原生真实串联默认跳过，显式开启后单独通过。
  • L4Wall：71 项测试通过，包括 PROXY v2 编码和上游首包。
  • L5Wall：36 项测试通过；含真实回环 TLS→HighWall 串联、未知 ALPN、SNI/Host 不一致、异常 framing、超大请求头、转发字节上限，以及跨平台许可证一致性检查。
  • 原生核心：cargo test --all-targets --locked 144 项通过；cargo fmt --all -- --check、cargo clippy --all-targets --locked -- -D warnings 和锁定依赖的 release 构建通过。
  • ruff、mypy（30 个根源文件）、compileall、解析器 fuzz（2000 轮、8000 次检查）、示例/严格配置检查和 highwall diff 通过。
  • DSH 首轮独立验收：0 Critical、0 High、3 Medium、8 Low、3 信息项。r5 定向复测为 0 Critical、0 High：8 个指定项中 7 项定向关闭（HWA-01/02/03/07/08/09/10）；HWA-11 在 r6 窄范围独立复测中关闭。HWA-04/05/06 由维护者 r7 以仓库内可重复测试复核通过，见 REGRESSION-R7-20261102.md；r7 不是 DSH 独立验收，不能把这三个编号写成 DSH 已独立复测。
  • 维护者 R7 定向结果：HWA-04 的三个重复伪头部拒绝均为 H3_MESSAGE_ERROR 270、上游零命中；HWA-05 的四个 HEAD 路由均为 200、正确 Content-Length、0 B 正文、每个上游命中 1；HWA-06 在 8192 B 为 200/上游命中 1，在 8193 B 为 413 HW-NATIVE-PROTOCOL-001/上游命中 0。
  • 发布标签工作流会在 Python 3.10–3.14 上重复根测试，并在 Linux/Windows 原生 runner 上重复静态检查和测试。本地结果不能替代该远程结果；正式发布前以 Release/Container 工作流状态为准。

部署边界

  • H1/H2/H3 仍只声明 C3，不声明 C4。C4 所需的完整流式背压饱和、QPACK 阻塞、真实 WebSocket 端到端和更完整的 slow-flow 覆盖仍未关闭。
  • H3/UDP 绕过 L4Wall；应用层 QUIC ACL 不能替代主机防火墙或安全组。部署必须另设 UDP 来源限制、速率限制和入口 ACL。
  • 被封禁来源仍可能先完成 TLS/QUIC 握手，再收到 HTTP 403。当前结论只证明应用层拒绝，不证明握手前零成本。
  • 原生封禁是进程内状态，带 TTL 和容量上限；重启清空，单写者持久状态服务尚未接入。
  • 运行中进程映像没有独立 attestation；DSH 的 runtime_process_image_evidence=false / process_image_independent_attestation=false 仍然成立。Release 资产 attestation 只证明构建产物来源，不证明某个运行进程就是该产物。
  • Docker/HAProxy 真实边缘、standard 容器拓扑、原生镜像公开分发、真实多节点状态服务和审计写失败拓扑未在本轮验收。Docker 是可选打包/部署层，不阻塞原生可执行文件路径。
  • 本轮未修改主机防火墙规则；Windows 自动规则可能绑定二进制绝对路径，部署必须显式创建并复验 TCP/UDP Allow 规则。
  • 本项目尚未完成独立第三方安全审计；DSH 结果是本轮安全验证输入，不等于外部审计证书或生产准入。

发布门槛与回滚

不得把本版本标记为生产发布，也不得把 C3 写成 C4 或完整防护。DSH 对生产发布的 no-push 建议保持不变；本次只发布 Alpha 预发布和脱敏、可复现的验证记录。

回滚策略：生产环境继续使用并固定 0.11.0 及对应摘要；不要依赖 latest。若需要评估 0.12.0，使用 Release 中的固定版本资产并先核对 SHA256SUMS 与 build provenance，再在隔离环境完成入口、真实来源、H3/UDP 网络层封堵和回滚演练。

HighWall 0.12.0 Alpha Pre-release

简体中文 | English

Status: GitHub pre-release (Alpha). This is a reviewable and testable pre-release, not a production release, and it carries no production support commitment. The latest stable version remains 0.11.0; this release does not move latest.

Included

  • Exact client-IP plus request-shape enforcement keys; /24 and /64 are correlation-only and do not automatically widen bans.
  • Read-only highwall ops preview; R2/R3 remain denied by default and no executor exists.
  • highwall capabilities --json for the Python H1 baseline with explicit C0-C4 levels; native-core coverage is declared separately for H1/H2/H3.
  • Standalone subprojects/l5wall 0.2.0: TLS termination, SNI/ALPN/mTLS policy, handshake timeout, loopback HTTP/1.1 bridge, SNI/Host consistency, and request-head/connection/byte bounds. L5Wall remains a standalone service release and is not folded into the Python wheel.
  • The phase-1 l5wall serve path accepts loopback listeners and upstreams only and accepts only the http/1.1 ALPN. H2/H3 are not presented as inspected.
  • The loopback bridge does not preserve the original client IP; do not enable per-IP bans, source ACLs, or source-dependent rate limits behind it.
  • Optional L4Wall send_proxy_v2 emits a standard PROXY v2 TCP header upstream; the native core can require PROXY v2 and trust only explicitly configured proxy networks.
  • Rust native B+D-4 core: TLS/QUIC, strict H1/H2/H3 parsing, H1 WebSocket, H2/H3 gRPC, compression/trailers, bounded-body WAF/DLP, exact-IP bans, mTLS/CRL, QUIC Retry, 0-RTT and migration disabled by default, and a QUIC client-network ACL.
  • TCP and UDP are separate ingress paths: H1/H2 can pass through L4Wall and PROXY v2; H3 currently reaches the native QUIC listener directly without L4Wall.
  • The Python image remains Python-only. The native core has a separate Dockerfile.native, runs as non-root, and does not pull Rust dependencies into the Python wheel or image.

Release assets

The GitHub Release is expected to contain:

  • highwall-0.12.0-py3-none-any.whl and highwall-0.12.0.tar.gz.
  • Linux x86_64 native binary highwall-native-core-0.12.0-linux-x86_64.
  • Windows x86_64 native binary highwall-native-core-0.12.0-windows-x86_64.exe.
  • Native example configuration highwall-native-config-0.12.0.example.json.
  • A SHA256SUMS file covering those assets.

The release workflow creates GitHub build provenance attestations for the assets. A separate container workflow publishes the Python-only image ghcr.io/fallen-leaves089/highwall:0.12.0 for linux/amd64 and linux/arm64 with OCI provenance, SBOM, and a cosign signature; it is not a full B+D-4 image. Dockerfile.native is not published as a public GHCR image in this release; use the standalone binary from the Release for the native path.

Verification

  • Root Python suite: 384 tests passed; coverage --branch reported 81% total coverage, above the project's 80% gate. The native real-chain test is skipped by default and passes separately when explicitly enabled.
  • L4Wall: 71 tests passed, including PROXY v2 encoding and upstream first bytes.
  • L5Wall: 36 tests passed, including a real loopback TLS-to-HighWall chain, unknown ALPN, SNI/Host mismatch, malformed framing, oversized request heads, relay byte limits, and a cross-platform license consistency check.
  • Native core: all 144 cargo test --all-targets --locked cases passed; cargo fmt --all -- --check, cargo clippy --all-targets --locked -- -D warnings, and the locked release build passed.
  • ruff, mypy (30 root source files), compileall, parser fuzzing (2,000 iterations and 8,000 checks), example and strict configuration checks, and highwall diff passed.
  • DSH first independent acceptance: 0 Critical, 0 High, 3 Medium, 8 Low, and 3 informational findings. The r5 directed retest found 0 Critical and 0 High: seven of eight scoped items closed (HWA-01/02/03/07/08/09/10); HWA-11 closed in the narrow r6 independent retest. HWA-04/05/06 were repeated by the maintainer's r7 repository tests, documented in REGRESSION-R7-20261102.md. The r7 run is not DSH independent acceptance and must not be described as a DSH independent retest of those three items.
  • Maintainer r7 directed results: all three HWA-04 duplicate-pseudo-header cases returned H3_MESSAGE_ERROR 270 with zero upstream hits; all four HWA-05 HEAD routes returned 200 with the correct Content-Length, a 0-byte body, and one upstream hit each; HWA-06 returned 200 with one upstream hit at 8192 bytes and 413 HW-NATIVE-PROTOCOL-001 with zero upstream hits at 8193 bytes.
  • The release-tag workflow repeats the root tests on Python 3.10-3.14 and repeats static analysis and tests on Linux and Windows native runners. Local results do not replace those remote results; check the Release and Container workflow runs before treating the pre-release as published.

Deployment boundaries

  • H1/H2/H3 remain declared at C3, not C4. Complete streaming backpressure saturation, QPACK blocking, real end-to-end WebSocket coverage, and broader slow-flow coverage remain incomplete.
  • H3/UDP bypasses L4Wall. An application-level QUIC ACL does not replace a host firewall or security group. Deployments must add separate UDP source restrictions, rate limits, and ingress ACLs.
  • A banned source can still complete a TLS/QUIC handshake before receiving HTTP 403. The current result proves application-layer rejection, not zero-cost pre-handshake blocking.
  • Native bans are process-local with TTL and capacity bounds; they are lost on restart and not yet backed by a single-writer persistent state service.
  • There is no independent attestation of the running process image; DSH's runtime_process_image_evidence=false / process_image_independent_attestation=false still applies. Release asset attestation proves build provenance, not that a particular running process is that build.
  • The real Docker/HAProxy edge, the standard container topology, public distribution of the native image, a real multi-node state service, and audit-write-failure topology were not validated in this round. Docker remains an optional packaging/deployment layer and does not block the native executable path.
  • Host firewall rules were not modified in this environment. Windows automatic rules can bind to an absolute binary path, so deployment must explicitly create and re-verify TCP/UDP Allow rules.
  • This project has not completed an independent third-party security audit. DSH results are security-validation inputs, not an external audit certificate or production admission.

Release gate and rollback

Do not label this version as a production release and do not describe C3 as C4 or complete protection. DSH's no-push recommendation for production release remains unchanged; this publication is limited to the Alpha pre-release and redacted, reproducible validation records.

Rollback: keep production on a pinned 0.11.0 artifact and its digest; do not depend on latest. To evaluate 0.12.0, use a fixed release asset, verify SHA256SUMS and build provenance, and complete an isolated ingress, real-source, H3/UDP network-blocking, and rollback exercise before any promotion.

Release assets include SHA-256 checksums and GitHub build provenance attestations. Verify the attached SHA256SUMS before installation.

Assets 8
Loading