Keynote by @halvarflake at #FUZZING'24 reflecting on the Reasons for the Unreasonable Success of Fuzzing.
Hacking culture in the 90's had very strong values. It had a value system outside of and different from normal society. Fuzzing was for the dumb kids.
"No one can argue with a root shell". Fuzzing was done via Perl scripts and pretty much manually. Bugs where super abundant. Even the CNN website run on a server you could find vulns in within 2 days.
OpenSSH CRC32 compensation CVE-2001-0144.
Michal Zalewski (!) piped random data into OpenSSL for fun and found a crash. Probably the only remotely exploitable OpenSSL bug found via fuzzing.
Going from a fuzzer bug in Acrobat Reader to an actual working exploit took way longer than finding the bug, but I learned so much that I wrote the weird machine paper 6 years later as a consequence.
Programs get more complex, human auditors must get "smarter" but fuzzing doesn't get much slower or less effective.
Fuzzing is embarrassingly parallel that it can exploit Moore's law for horizontal scalability.
Fuzzing is always a winner in the hardware lottery. Program analysis is in software while fuzzing is perf in hardware.
Lack of false positives. This is huge in industry! More important even than "absence of bugs".
AFL had outside impact because it harnessed these strengths. Designed to just work. Smart design decisions better than a dumb strategy.
Also, fuzzers are really simple to implement. Anyone can do it.
Perhaps fuzzing is "the bitter lesson", but applied to program analysis, not ML.
The AI community are all-in on compute now. This seems to be the same in program.analysis. "The two methods that scale arbitrarily are search and fuzzing".
Fuzzing community should look into research in RL. Novelty search.
We should work on our reward functions. Coverage is blind to the state machine. Similar to the RL game-playing issue?
But maybe we need to be faster rather than smarter? Let's look at fuzzer performance bottlenecks. Let's look at dedicated hardware.
Before we announce the exciting keynotes for #FUZZING'24, we found some time to upload the recordings for the last two years by Abhishek Arya (@infernosec), @AndreasZeller, Cristian Cadar (@c_cadar), and Kostya Serebryany (@kayseesee).
//@lszekeres, @baishakhir, @yannicnoller.
FUZZING'22 Keynote by Abhishek Arya (Google) on "The Evolution of Fuzzing in Finding the Unknowns"
FUZZING'22 Keynote by Andreas Zeller (CISPA & Saarland U) on "Fuzzing: A Tale of Two Cultures"
Recently modified code and sanitizer instrumentation seem to be among the most effective heuristics for target selection in directed #fuzzing according to this recent SoK by Weissberg et al. LLMs show much promise for target selection, too.
But in an interesting twist, the authors find that choosing functions by their complexity might be even better at retrieving functions that contained vulnerabilities in the past.
- Human artifacts (documentation) as oracles.
- How to infer oracles, e.g. from JavaDoc comments? What about false pos? Consider them as signal for user.
- Oracle problem impacts how good deduplication works.
- Metamorphic testing. Explore in other domains, e.g. perf. testing!
- Mine assertions and use them in a fuzzer feedback loop
- Assertions are the best way to build oracles into the code
- hyperproperties are free oracles (differential testing)
- ML to detect vuln patterns. Use as oracles
- Bugs as deviant behavior (Dawson)
- Bi-abductive symbolic execution
- Infer ran "symbolic execution" on changed part of every commit/diff
- Post-land analysis versus diff-time analysis changed fix rate from 0% to 70%. Why?
* Cost of context switch
* Relevance to developer
- Deploying a static analysis tool is an interaction with the developers.
- Devs would accept false positives and work with the team to "fit" the tool to the project rather.
- Audience matters!
* Dev vs SecEng
* Speed tolerance
* FP/FN tolerance
Security tooling
- ideal solution mitigates entire classes of bugs
- performance is important.
- adoption is critical!
- works with the ecosystem
Rewriting in memory-safe language (e.g. Swift)
- View new code as green islands in a blue ocean of memory-unsafe code.
- Objective: Turn blue to green.
- We need solutions with low adoption cost.