A 0-click exploit chain for the Pixel 10
448 points
• 4 days ago
• Article
Link
Project Zero 的 Seth Jenkins 详述了他们开发的一条针对 Google Pixel 10 的新型零点击漏洞利用链，该工作建立在此前攻破 Pixel 9 的成果之上。原始的 Dolby UDC 漏洞（CVE-2025-54957）经少量调整后被移植到 Pixel 10，但由于 Pixel 10 使用 RET PAC 而非 -fstack-protector，利用中需要覆盖 dap_cpdp_init 而不是 __stack_chk_fail 。该更新后的利用链针对安全补丁级别为 2025 年 12 月或更早的未修补设备有效。
利用链中的本地权限提升部分无法直接移植，因为 Pixel 10 上不存在 BigWave 驱动。不过，研究人员发现了 Tensor G5 上 Chips&Media Wave677DV VPU 的新驱动。该驱动由与 BigWave 相同的团队开发，直接将芯片硬件接口暴露给用户空间，允许用户态映射芯片的 MMIO 寄存器接口，而不是通过标准的 V4L2 API 。
在 VPU 驱动的 vpu_mmap 处理函数中发现了一个严重漏洞：该处理函数仅根据 VMA 的大小调用 remap_pfn_range，而没有将映射范围限定在实际寄存器区域的大小内。这导致用户态可以从 VPU 寄存器区域的物理地址起映射任意物理内存。由于内核镜像位于更高的物理地址，攻击者可以访问并修改内核。结合 Pixel 设备上固定的内核物理地址，这几乎是一种任意内核读写原语——实现基本原语只需五行代码，完成完整利用通常不到一天时间。
该漏洞于 2025 年 11 月 24 日被报告，Android VRP 将其评为高危（High），比类似的 BigWave 漏洞获得了更高的严重性评级。漏洞在 71 天后的 2 月 Pixel 安全公告中得到修补，这是作者报告的 Android 驱动漏洞首次在 90 天内被修复，反映了 Android 漏洞分类与修补流程的改进与加速。
尽管这是积极进展，研究仍强调 Android 驱动代码需更强的安全意识与更严格的审计。 VPU 漏洞在 BigWave 问题曝光仅五个月后被发现，表明相关驱动并未得到充分的横向审计。加强驱动安全仍是当务之急，建议厂商采取更主动的软件安全策略，强化代码审计与漏洞修补流程，防止类似问题流向终端用户。
Seth Jenkins of Project Zero details the development of a new zero-click exploit chain targeting the Google Pixel 10, building on prior work that compromised the Pixel 9. The original Dolby UDC exploit (CVE-2025-54957) was adapted for the Pixel 10 with minor adjustments, though the Pixel 10's use of RET PAC instead of `-fstack-protector` required overwriting `dap_cpdp_init` instead of `__stack_chk_fail`. This updated exploit targets unpatched devices with a security patch level of December 2025 or earlier.
The local privilege escalation component of the chain could not be ported because the BigWave driver is absent on the Pixel 10. However, a new driver for the Chips&Media Wave677DV VPU on the Tensor G5 chip was identified. This driver, developed by the same team behind BigWave, directly exposes the chip's hardware interface to userspace, including allowing userspace to map the chip's MMIO register interface, rather than integrating with the standard V4L2 API.
A critical vulnerability was discovered in the VPU driver's `vpu_mmap` handler. The handler calls `remap_pfn_range` based solely on the size of the VMA, without bounding it to the actual size of the register region. This allows userspace to map arbitrary physical memory starting from the VPU register region's physical address. Since the kernel image resides at a higher physical address, it can be accessed and modified. Combined with the kernel's fixed physical address on Pixel devices, this provides an arbitrary read-write primitive on the kernel with minimal effort, requiring only five lines of code for the basic primitive and less than a day for a full exploit.
The vulnerability was reported on November 24, 2025, and rated High severity by Android VRP, an improvement over the Moderate rating given to the similar BigWave bug. It was patched 71 days later in the February Pixel security bulletin, marking the first time an Android driver bug reported by the author was patched within 90 days. This demonstrates progress in Android's triage pipeline and efforts to patch serious vulnerabilities efficiently.
Despite this positive step, the research highlights the ongoing need for more robust and security-aware code in Android drivers. The VPU vulnerability was found just five months after the BigWave bugs were reported, suggesting that developers did not adequately audit their other drivers for similar issues. Strengthening driver security remains a crucial priority, and vendors are encouraged to adopt proactive approaches to software security, code auditing, and vulnerability patching to prevent such issues from reaching end-users.
240 comments • Comments Link
• 移动设备上的 AI 功能在用户打开消息前就对消息媒体进行解码，从而扩大了"0-click"攻击面，带来了新的安全风险。
• Google 在 90 天内修复了一个 Android 驱动漏洞，这一速度明显快于其他 Android 厂商和 Apple，引发了人们对其他厂商补丁速度的质疑。
• 由于设备制造商对 Android UI 的分叉导致碎片化，Android 厂商更新缓慢，补丁难以迁移。
• Apple 对安全漏洞的历史响应时间大约为 6 个月，尽管有报告称近年来有所改善。
• Apple 设备上的 Lockdown 模式被视为记者等高风险人群的重要工具，但由于缺乏证据以及零点击漏洞的高价值，其对国家级攻击者的有效性仍受质疑。
• 已发布的 CVE 数量显著上升，翻倍周期已从 4–4.5 年缩短到大约两年，尽管这些 CVE 的质量和合法性存在争议。
• AI 一方面通过引入带有安全漏洞的新功能扩大了攻击面，另一方面又为安全研究人员提供了更高效的漏洞发现工具。
• 关于开发者对安全漏洞的责任存在争论，一些人认为追究编写不安全代码的个人责任能够提高软件安全性。
• Rust 提供了整数溢出的检测选项：在调试构建中默认开启，在发布构建中可选择开启，这可以防止某些类型的漏洞。
• 安全社区对 KASLR 等缓解措施的有效性存在分歧，一些人认为由于信息泄露普遍，这类措施只提供有限的边际收益。
• GrapheneOS 因其减少攻击面和加强加固而受到认可，但也有人认为其缓解措施表面化，需要更实质性的改进。
• 讨论强调了新增功能与维护安全性之间的紧张关系：有人认为移除功能是提升安全性的简单方法，但这会以牺牲功能性为代价。
• 呼吁将驱动更好地上游合并到 Linux 内核，以提升所有 Android 设备的安全性，而不仅仅是那些有开源 BSP 的设备。
总体讨论揭示了移动安全的复杂局面：AI 功能的快速整合正在创造新漏洞，而现有安全措施的有效性常被质疑。尽管在补丁速度和安全研究方面有所进展，但攻击面总体扩大以及许多厂商响应缓慢仍是重大问题。 AI 在制造与缓解安全问题方面的双重作用反复出现：它既可能引入新的漏洞，也能帮助更快地发现漏洞。关于开发者责任、 Lockdown 模式和 KASLR 等防护手段有效性的争论，凸显了在安全性、可用性与创新之间寻求平衡的持续挑战。 • AI-powered features on mobile devices are increasing the "0-click" attack surface by requiring message media to be decoded before the user opens the message, creating new security vulnerabilities.
• Google patched an Android driver bug within 90 days, which is notably fast and raises questions about the patching speed of other Android vendors and Apple.
• Android vendors have been slow to provide updates due to the fragmentation caused by manufacturers forking the Android UI, making it difficult to migrate updates.
• Apple's response time for security bugs has historically been around 6 months, though there are reports of improvement in recent years.
• Lockdown mode on Apple devices is seen as a necessary tool for high-risk individuals like journalists, but its effectiveness against state-level actors is debated due to a lack of evidence and the high value of zero-click exploits.
• The rate of published CVEs has increased significantly, with a doubling interval that has decreased from 4-4.5 years to approximately two years, though the quality and legitimacy of these CVEs are questioned.
• AI is both increasing the attack surface by introducing new features with security holes and providing security researchers with tools to find vulnerabilities more efficiently.
• There is a debate around the responsibility of developers for security vulnerabilities, with some suggesting that personal liability for writing insecure code could improve software security.
• Rust programming language offers choices regarding integer overflow checks, with default checks in debug builds and optional checks in release builds, which can prevent certain types of vulnerabilities.
• The security community is divided on the effectiveness of mitigations like KASLR, with some arguing that they provide only marginal utility due to the prevalence of info leaks.
• GrapheneOS is recognized for its attack surface reduction and security hardening, though some argue that its mitigations are superficial and that more significant improvements are needed.
• The discussion highlights the tension between adding new features and maintaining security, with some suggesting that removing features is an easy way to improve security but at the cost of functionality.
• There is a call for better upstreaming of drivers into the Linux kernel to improve security across all Android devices, not just those with open-source BSPs.
The discussion reveals a complex landscape of mobile security, where the rapid integration of AI features is creating new vulnerabilities, and the effectiveness of security measures is often debated. While some progress is made in patching speeds and security research, the overall increase in attack surfaces and the slow response times of many vendors remain significant concerns. The role of AI in both creating and mitigating security issues is a recurring theme, with its potential to both introduce new vulnerabilities and aid in their discovery. The debate around developer liability and the effectiveness of security mitigations like Lockdown mode and KASLR underscores the ongoing challenges in balancing security with usability and innovation.