QuickFuzz, a tool written in Haskell designed for testing un- expected inputs of common file formats on third-party software, taking advantage of off-the-shelf, well known fuzzers. Unlike other generational fuzzers, QuickFuzz does not require to write specifications for the file formats in question since it relies on existing file-format-handling libraries available on the Haskell code repository. There is more information in its website.
In this example, we uncover a null pointer dereference in gif2webp from libwebp 0.5:
$ QuickFuzz gentest gif "./gif2webp @@ -o /dev/null" -l 1 -u 10 -f radamsa
...
Test case number 4481 has failed.
Moving to outdir/QuickFuzz.68419739009.4481.3692945303624111961.1.gif
...
We found a crash. We can inspect it manually to verify it is a null pointer issue:
$ ./gif2webp outdir/QuickFuzz.68419739009.4481.3692945303624111961.1.gif
==10953== ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000
(pc 0x000000403ff9 sp 0x7fffffffd6e0 bp 0x7fffffffded0 T0)
AddressSanitizer can not provide additional info.
#0 0x403ff8 (examples/gif2webp+0x403ff8)
#1 0x7ffff437af44 (/lib/x86_64-linux-gnu/libc-2.19.so+0x21f44)
#2 0x401b18 (examples/gif2webp+0x401b18)
==10953== ABORTING
Finally, we can shrink the crashing input to obtain a smaller file:
$ QuickFuzz gentest gif "./gif2webp @@ -o /dev/null" -l 1 -s 3692945303624111961 -f radamsa -r
Test case number 1 has failed.
Moving to outdir/QuickFuzz.68997856397.1.3692945303624111961.1.gif
Shrinking over bytes has begun...
Testing shrink of size 48
Testing shrink of size 47
...
Testing shrink of size 15
Shrinking finished
Reduced from 48 bytes to 16 bytes
After executing 554 shrinks with 33 failing shrinks.
Saving to outdir/QuickFuzz.68997856397.1.3692945303624111961.1.gif.reduced
Finished!
We support Stack to compile and install QuickFuzz. Before starting with it, make sure you have libgmp-dev installed otherwise ghc will fail to compile. Also, zlib.h is required to compile QuickFuzz (some packages require it). For instance, in Ubuntu/Debian:
# apt-get install zlib1g-dev libgmp-dev libtinfo-dev
After installing stack, you should:
$ git clone https://github.com/CIFASIS/QuickFuzz --depth 1
$ cd QuickFuzz
$ install_fuzzers.sh
$ stack setup
Because QuickFuzz generates a lot of dependencies that may not be necessary to test an specific category of files, we modularized the project with different activation flags. Currently we have 7 flags:
| Flag | Supported formats |
|---|---|
| image | svg, png, gif, tiff, jpeg |
| arch | tar, zip |
| doc | html, css, pdf, ps, eps, xml |
| code | c, js, py, go, lua, evm |
| media | wav |
| net | http |
| pki | asn1, crl, x509 |
| bnfc | cf, grammer format |
For instance, to compile only with image generation (Bmp, Gif, Png, Ico, ..):
$ stack install --flag QuickFuzz:image
Because of a Stack issue, you must install alex and happy manually before enabling the code flag:
$ stack install alex happy
Direct cabal installation is not recommended nor supported.
- Pablo Buiras (Chalmers University of Technology)
- Martín Ceresa (CIFASIS-Conicet)
- Gustavo Grieco (CIFASIS-Conicet and VERIMAG)
- Agustín Mista (Universidad Nacional de Rosario)
- Franco Costantini
- Lucas Salvatore
- Martín Escarrá (Universidad Nacional de Rosario)
You can join the QuickFuzz mailing group to get notifications of new features and releases. To join, you can send an empty email to [email protected].