search

google / fuzzbench

FuzzBench - Fuzzer benchmarking as a service.
https://google.github.io/fuzzbench/
Apache License 2.0
1.09k stars 266 forks source link

Unable to open files under crashes #885

Open Nirmit-Zinzu opened 3 years ago

Nirmit-Zinzu commented 3 years ago

I ran the 4 fuzzers for 24hr period on few of the benchmarks. Under the experiment_data, it generated corpus, and results for all the trials. It generated crashes folder for only few of the benchmarks. I suppose this is expected.

When I open the crashes folder, it includes a file like timeout-58f2b4df76dde5cca12d0d1f8738544639013e19. I am unable to open this file and analyze any information. What information does this contain and how to fix it?

I would like to capture information about the crash for further analysis. Any help is appreciated it.

Thanks

jonathanmetzman commented 3 years ago

I am unable to open this file and analyze any information

This file is a file that caused "a crash" when the output_corpus folder was measured using libFuzzer. In this case, it means that this file took longer than the timeout (15 seconds I think) to execute. Files in this folder that begin with timeout- or oom- are probably not so interesting. The ones that begin with crash- are probably more interesting. If you're interested in this stuff, I'd probably wait, since we are going to add some native crash analysis capabilities to Fuzzbench. If you wait, you won't need to do the analysis manually after the experiment runs

DerekFei commented 3 years ago

@jonathanmetzman Thank you for the explanation. Do you have any expected time-frame for this feature being released?

Also, if we wish to do the analysis manually after the experiment for crashes, what would be the best place to look at the data/logs relates to the total crashes, unique crashes, etc.

Currently, I can find the crashes file under the experiment-folders -.. -> trail XX -> crashes . Are these files indicate the unique crashes and the input? What's the best way to interpreter the meaning behind these crash files? Thank you :)

jonathanmetzman commented 3 years ago

Do you have any expected time-frame for this feature being released?

I would guess that within 2-3 weeks we should have more for you to work with.

Also, if we wish to do the analysis manually after the experiment for crashes, what would be the best place to look at the data/logs relates to the total crashes, unique crashes, etc.

Right now the only data we collect on crashes is in that folder. You can also look at the log file for each trial. Some fuzzers might report crashes there.

Are these files indicate the unique crashes and the input?

These files are simply the inputs that causes the target to crash. They are not unique. To make sense of them you probably need to rerun the targets on them.