安全研究员Hyunwoo Kim发现了编号为CVE-2026-64561的Zapscape漏洞，这是KVM/x86虚拟化中的use-after-free漏洞，位于shadow MMU模拟的递归zap路径中[1]。攻击者可通过在guest端执行操作逃逸至host端，以kernel权限在host上执行任意命令[1]。这一漏洞特别威胁接受不受信任虚拟机的多租户x86公有云环境[1]。
该漏洞影响Linux 7.1.3内核及其他版本（内核版本范围f95eec9bed76到2abd5287f083）[1]。已发布的PoC代码针对基于QEMU TCG模拟的AMD SVM/NPT平台，可成功在host上创建/Zapscape文件且所有者为root（uid=0）[1]。利用该漏洞需要L1内核权限，或在无guest root权限场景下结合其他本地特权提升漏洞（如Dirty Frag）一起使用[1]。在Intel处理器上，仅当EPT page walk length 4和5同时暴露给L1时才可触发该漏洞[1]。
Security researcher Hyunwoo Kim has disclosed Zapscape, a use-after-free vulnerability in KVM/x86 virtualization identified as CVE-2026-64561 [1]. The flaw exists in the recursive zap path of the shadow MMU emulation, allowing attackers to escape from a guest virtual machine to the host system and execute commands with kernel privileges [1]. This poses a significant threat to multi-tenant x86 public cloud environments that accept untrusted guests [1].
The vulnerability affects Linux kernel versions from 7.1.3 through the range spanning f95eec9bed76 to 2abd5287f083 [1]. A proof-of-concept exploit has been released demonstrating the complete attack chain [1]. On AMD SVM/NPT systems, the vulnerability can be triggered with guest kernel-level access or potentially without root privileges when combined with local privilege escalation exploits such as Dirty Frag [1]. Intel systems are only vulnerable when EPT page walk lengths 4 and 5 are simultaneously exposed to Layer 1 [1]. Successful exploitation creates a file named /Zapscape on the host system with root ownership (uid=0), confirming arbitrary command execution with elevated privileges [1].