Fuzzing generates a large number of crashing inputs, and given a constant influx of new crashes, it is impractical to perform root-cause analysis and exploitability assessment for each one.
🔍 拆解
- Fuzzing generates
👉 模糊测试会产生 - a large number of crashing inputs
👉 大量导致崩溃的输入 - given a constant influx of new crashes
👉 考虑到持续不断的新崩溃出现 - it is impractical to perform
👉 做…是不现实的 - root-cause analysis
👉 根因分析 - exploitability assessment
👉 可利用性评估
👉 模糊测试会产生大量导致崩溃的输入,而由于新的崩溃不断出现,对每一个崩溃都进行根因分析和可利用性评估是不现实的。
🔥 英语重点句型
👉it is impractical to do…= 做某事不现实
🔹
Prior work assesses severity based on reachability (the range that can be read or written via memory-safety bugs), but cannot distinguish cases where severity differs greatly despite having similar reachability.
🔍 拆解
- Prior work👉 以往研究
- assesses severity based on reachability
👉 基于可达性评估严重性 - the range that can be read or written
👉 可以被读/写的范围 - via memory-safety bugs
👉 通过内存安全漏洞 - cannot distinguish cases
👉 无法区分情况 - where severity differs greatly
👉 严重性差异很大 - despite having similar reachability
👉 尽管可达性相似
✅ 翻译
👉 以往研究基于可达性(即通过内存安全漏洞可读写的范围)来评估漏洞严重性,但无法区分那些虽然可达性相似但严重性差异很大的情况。
🔥 重点表达
👉despite + 名词 / 动名词= 尽管…
🔹 第三句(核心创新)
We instead measure controllability: how freely an attacker can steer the next invalid access target within the reachable range.
🔍 拆解
- We instead measure👉 我们改为衡量
- controllability👉 可控性
- how freely👉 多么自由地
- an attacker can steer👉 攻击者可以操控
- the next invalid access target👉 下一次非法访问目标
- within the reachable range👉 在可达范围内
✅ 翻译
👉 我们转而衡量“可控性”:即攻击者在可达范围内能够多自由地操控下一次非法访问目标。
🔥 重点表达
👉how + 副词= 表示程度
例:
- how easily → 多容易
- how freely → 多自由
From diversified crashing inputs, we build a first-order Markov chain over ASan-reported access offsets and score bugs by the trajectory entropy of the chain.
🔍 拆解
- From diversified crashing inputs
👉 从多样化的崩溃输入中 - we build a first-order Markov chain
👉 我们构建一阶马尔可夫链 - over ASan-reported access offsets
👉 基于 ASan 报告的访问偏移 - score bugs👉 对漏洞打分
- by the trajectory entropy
👉 根据轨迹熵 - of the chain👉 (马尔可夫链的)
👉 我们从多样化的崩溃输入中,基于 ASan 报告的访问偏移构建一阶马尔可夫链,并通过该链的轨迹熵对漏洞进行评分。
🔥 技术表达
👉build a model over X 基于 X 构建模型
In a preliminary study on known CVEs, the entropy score is higher for bugs with higher controllability and separates vulnerabilities that reachability-based metrics fail to distinguish.
🔍 拆解
- In a preliminary study👉 在初步研究中
- on known CVEs👉 针对已知 CVE
- entropy score is higher👉 熵评分更高
- for bugs with higher controllability
- 👉 对于可控性更高的漏洞
- separates vulnerabilities👉 区分漏洞
- that reachability-based metrics fail to distinguish
👉 可达性指标无法区分的
✅
👉 在对已知 CVE 的初步研究中,可控性更高的漏洞具有更高的熵评分,并且该方法能够区分那些基于可达性指标无法区分的漏洞。
Controllability provides a better measure of exploitability than reachability.
👉“可控性比可达性更能反映漏洞是否真的危险”
最常见 5 种用法
1️⃣ X-based metric👉 基于某种方法的指标
例:
- entropy-based metric
👉 基于熵的指标 - performance-based metrics
👉 基于性能的指标
2️⃣ evaluation metrics👉 评估指标
例:
We use several evaluation metrics.
👉 我们使用多个评估指标
3️⃣ metric for …👉 用于…的指标
例:a metric for exploitability
👉 用于评估可利用性的指标
4️⃣ metric measures …👉 指标衡量什么
例:This metric measures controllability.
👉 该指标衡量可控性
5️⃣ metrics fail to …👉 指标无法做到…(论文常用)
metrics fail to distinguish vulnerabilities
👉 指标无法区分漏洞
在安全/编程中会常见的 metrics
| 表达 | 含义 |
|---|
| performance metrics | 性能指标 |
| security metrics | 安全指标 |
| accuracy metrics | 准确率指标 |
| risk metrics | 风险指标 |
| coverage metrics | 覆盖率指标(fuzzing 常见🔥) |