AI Security AI安全 7h ago Updated 1h ago 更新于 1小时前 52

New OVSwrap Linux Kernel Flaw Lets Local Users Gain Root via Open vSwitch 新发现的OVSwrap Linux内核漏洞允许本地用户通过Open vSwitch获取root权限

CVE-2026-64531 (OVSwrap) is a memory corruption flaw in the Linux kernel's Open vSwitch datapath that allows unprivileged local users to escalate to root on broadly configured distributions The bug stems from a 16-bit nla_len overflow: a March 2025 change removed a 32 KiB cap on generated action streams, exposing a 13-year-old truncation bug that enables deterministic buffer wraparound A public PoC with pre-built records for ~800 x86-64 kernel builds chains three primitives (kernel pointer leak, Linux内核Open vSwitch数据路径存在内存损坏漏洞(CVE-2026-64531,CVSS 7.8),普通本地用户无需任何特殊权限即可获取root 漏洞利用链包含内核指针泄露、任意内核读取和目标内存修改三个原语,PoC已公开且包含约800个x86-64内核构建的预构建记录 2025年3月移除的32KiB动作流限制暴露了存在13年的截断bug,导致16位nla_len字段溢出回绕 上游修复已于2026年7月24日发布,覆盖Linux 5.15.212、6.1.178、6.6.145、6.12.97、6.18.40和7.1.5版本 在AlmaLinux、Debian、Fedora、Ub

82
Hot 热度
72
Quality 质量
65
Impact 影响力

Analysis 深度分析

TL;DR

  • CVE-2026-64531 (OVSwrap) is a memory corruption flaw in the Linux kernel's Open vSwitch datapath that allows unprivileged local users to escalate to root on broadly configured distributions
  • The bug stems from a 16-bit nla_len overflow: a March 2025 change removed a 32 KiB cap on generated action streams, exposing a 13-year-old truncation bug that enables deterministic buffer wraparound
  • A public PoC with pre-built records for ~800 x86-64 kernel builds chains three primitives (kernel pointer leak, arbitrary kernel read, targeted decrement) to zero out fsuid/fsgid
  • The upstream fix shipped July 24 across Linux 5.15.212, 6.1.178, 6.6.145, 6.12.97, 6.18.40, and 7.1.5; EOL series will not receive patches
  • Exploitation requires no existing OVS bridge, no running ovs-vswitchd, and no host-level CAP_NET_ADMIN—only an installed openvswitch module and unprivileged user namespaces

Why It Matters

This vulnerability demonstrates how a long-standing kernel logic bug, dormant for 13 years, can be weaponized by a single reliability-focused upstream change that overlooked security implications—a pattern increasingly common as kernel codebases grow in complexity. For AI practitioners and infrastructure teams, it underscores the critical risk of shared-host multi-tenant environments where a compromise of one workload can chain through OVSwrap into full root access, and it highlights the importance of monitoring vendor-specific kernel trackers rather than relying solely on upstream version numbers.

Technical Details

  • Root cause: Open vSwitch stores flow actions as Netlink attributes with a 16-bit nla_len field (max 65,535 bytes). An attacker crafts a CLONE action containing hundreds of conntrack sub-actions; on x86-64 each expands to 164 bytes, exceeding the 16-bit limit and causing the length to wrap around. Subsequent parsing resumes from attacker-controlled data within the same contiguous buffer, enabling arbitrary memory corruption without heap grooming.
  • Enabling change: A March 2025 commit removed a 32 KiB cap on the total generated action stream to fix unpredictable failures in large OpenStack deployments. The review discussed reliability but did not address the security consequence of exposing the older truncation bug.
  • Exploit chain: Three primitives derived from the wraparound—a kernel pointer leak via a fake OUTPUT action, an arbitrary kernel read via a forged tunnel SET action, and a targeted decrement via teardown of a forged tun_dst pointer—are chained to locate a host process's credentials and decrement fsuid/fsgid to zero on modern kernels.
  • Attack prerequisites: The openvswitch module must be installable (auto-loads on Generic Netlink family resolution; lsmod may show it as unloaded), unprivileged user namespaces must be enabled, and the PoC additionally requires OVS conntrack support, the FTP conntrack helper, and sudo.
  • Affected and unaffected systems: Default-config exploitation confirmed on AlmaLinux 9/10, Alpine 3.22–3.24, Amazon Linux 2023, Arch, CentOS Stream 9/10, Debian 12/13, Fedora 42–44, Gentoo, Kali 2026.1, Linux Mint 22.3, NixOS, openSUSE Tumbleweed, Pop!_OS, Rocky Linux 9/10, and Ubuntu 22.04. Ubuntu 24.04 is partially mitigated by AppArmor but bypassable via aa-exec -p trinity. Older systems (Amazon Linux 2, Debian 11, Rocky Linux 8, Ubuntu 20.04) retained older code paths and were not exploitable through this route.

Industry Insight

  • Multi-tenant host hardening is now a priority: In cloud and shared-hosting environments, OVSwrap transforms a low-privilege compromise into full root. Operators should audit Open vSwitch module usage and disable unprivileged user namespaces where feasible, or apply the emergency BPF guard for environments requiring both features.
  • Vendor trackers trump upstream versions: Distribution kernels carry backports and downstream changes, so relying on upstream fix numbers alone is insufficient. Teams should monitor vendor-specific security advisories and apply patched kernels promptly, especially for EOL series that will not receive upstream fixes.
  • Security review gaps in reliability patches: The March 2025 change that enabled this exploit was motivated by legitimate reliability concerns in large deployments. This reinforces the need for security-focused review checkpoints when removing defensive bounds (caps, limits) in kernel code, particularly in network subsystems with long code histories.

TL;DR

  • Linux内核Open vSwitch数据路径存在内存损坏漏洞(CVE-2026-64531,CVSS 7.8),普通本地用户无需任何特殊权限即可获取root
  • 漏洞利用链包含内核指针泄露、任意内核读取和目标内存修改三个原语,PoC已公开且包含约800个x86-64内核构建的预构建记录
  • 2025年3月移除的32KiB动作流限制暴露了存在13年的截断bug,导致16位nla_len字段溢出回绕
  • 上游修复已于2026年7月24日发布,覆盖Linux 5.15.212、6.1.178、6.6.145、6.12.97、6.18.40和7.1.5版本
  • 在AlmaLinux、Debian、Fedora、Ubuntu等主流发行版默认配置下均可利用,云环境和共享主机风险尤为突出

为什么值得看

该漏洞直接影响运行容器化工作负载的Linux系统,Open vSwitch广泛用于Kubernetes网络插件和云基础设施,本地权限提升可导致容器逃逸和整个主机沦陷。对于AI从业者而言,训练集群和推理服务多部署在共享Linux主机上,此漏洞可能成为攻击者从单点突破到整个AI基础设施的跳板。

技术解析

  • 漏洞原理:OVS内核数据路径将生成的流动作存储为Netlink属性,其nla_len字段为16位宽,上限65,535字节。2025年3月移除的32KiB总动作流限制本可防止嵌套动作超过此阈值,移除后攻击者可提交包含数百个conntrack子动作的CLONE动作,在x86-64上每个展开为164字节,导致长度字段回绕。
  • 利用链:PoC通过伪造OUTPUT动作泄露内核指针,通过伪造tunnel SET动作实现任意内核读取,再通过teardown伪造的tun_dst指针进行目标递减,最终将fsuid和fsgid置零获得root。利用过程无需堆 grooming,因为着陆点在同一个连续缓冲区中是确定性的。
  • 触发条件:攻击者只需创建私有用户和网络命名空间(unshare -Urn),在命名空间内获得CAP_NET_ADMIN,即可访问漏洞流安装路径。即使openvswitch模块未加载,解析Generic Netlink族名也会自动加载它,lsmod为空不代表安全。
  • 修复与缓解:上游已发布修复,建议安装修补的vendor内核。临时缓解措施包括:添加echo 'install openvswitch /bin/false' > /etc/modprobe.d/ovswrap.conf阻止模块加载,或禁用非特权用户命名空间。对于必须同时使用OVS和命名空间的环境,PoC仓库提供了紧急BPF守卫。
  • 受影响系统:测试覆盖AlmaLinux 9/10、Alpine 3.22-3.24、Amazon Linux 2023、Arch、CentOS Stream 9/10、Debian 12/13、Fedora 42-44、Gentoo、Kali 2026.1、Linux Mint 22.3、NixOS、openSUSE Tumbleweed、Pop!_OS、Rocky Linux 9/10、Ubuntu 22.04等默认配置。Ubuntu 24.04/26.04默认AppArmor限制可阻止利用,但禁用后可被绕过。Amazon Linux 2、Debian 11、Rocky Linux 8、Ubuntu 20.04因保留旧代码路径不受影响。

行业启示

  • 云原生安全需重新评估:OVS是Kubernetes CNI的主流选择,此漏洞表明即使容器内用户无特殊权限,也可能通过命名空间机制实现主机root获取。云服务商和平台团队应优先更新内核,并在多租户环境中限制非特权用户命名空间。
  • 历史债务的安全代价:该bug存在13年,因32KiB限制被移除而暴露,说明代码演进中的性能/可靠性优化可能无意中移除安全护栏。建议在关键安全路径上建立自动化回归测试,确保此类限制移除时进行安全影响评估。
  • 应急响应优先级:对于AI训练集群和推理服务,建议立即检查openvswitch模块状态,优先更新到修补内核;无法立即更新的环境应实施模块阻止策略,并审查容器运行时是否授予CAP_NET_ADMIN权限,减少攻击面。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Security 安全 Research 科学研究