ClusterFuzz: Fuzzing at Google Scale
summary
ClusterFuzz runs fuzzing at Google scale: a public infrastructure powering OSS-Fuzz and Google fuzzing across 25,000 cores, about 2,500 targets, and finds over 8,000 security vulnerabilities (40,000 total bugs reported, ~91% fixed). It uses libFuzzer/AFL, sanitizers (ASan/UBSan/LSan), protobuf mutators, corpus distillation, preemptible VMs and automated triage (dedupe, grouping, bisection, variant analysis) to minimize testcases, prioritize targets, file bugs automatically and verify fixes.
AI-written summary · about
chapters (53)
- 0:00Introduction
- 0:43History
- 2:33Fuzzing Myths
- 4:24How to Scale
- 5:52Ideal Fuzzing Workflow
- 7:11ClusterFuzz Overview
- 8:32What to Fuzz
- 10:11Blackbox Fuzzing
- 11:16Chrome Blackbox Fuzzing
- 12:21AFL
- 13:25Fuzzer
- 16:48Example
- 17:39Educating Developers
- 18:37Build Pipeline
- 19:20Optimization
- 20:55Fuzzy Machines
- 21:54Fuzzing Targets
- 23:09Corpus
- 24:52Search
- 25:32Strategy Selection
- 26:39Deduplication
- 27:33Deduplication example
- 28:17Grouping
- 28:59Minimize
- 30:09Continuous Version
- 31:00Variant Analysis
- 33:08Unique Crashes
- 34:14Examples
- 34:33Test Case Report
- 35:11Prioritization
- 36:29Verification
- 37:22External Reports
- 38:33Reporters
- 38:46Execution Speed
- 39:22Crash Statistics
- 39:43Code Coverage
- 40:46More Applications
- 43:25Chrome
- 44:53OSS Security
- 45:29Whats next
- 47:22Key takeaways
- 49:05Questions
- 49:25concurrency issues
- 51:00bugs
- 53:03Corpus Sharing
- 54:04Other Questions
- 54:16Mutator
- 55:02Hybrid Fighting
- 55:27Future Plans
- 56:38Corrupted Stack
- 57:10Address Sanitizer
- 57:52The biggest challenge
- 58:23How we made it simple
related talks
44mOSS-Fuzz - Google's continuous fuzzing service for open source software
39mMaking Fuzzing Part of Your Software Development Lifecycle
53mFuzzing 101
39mFuzzing Session: Finding Bugs and Vulnerabilities Automatically
31m“Fuzz or lose..."
25mHow fuzzing helps to find bugs
36mFuzzing the CNCF Landscape
01h 15mMake your programs more reliable with Fuzzing
