Improve the libafl_libfuzzer corpus #1539
Conversation
|
@domenukk fairly certain the CI failure here is not my fault lol |
| res => res?, | ||
| } | ||
|
|
||
| // we DO NOT save metadata! |
There was a problem hiding this comment.
why we don't save metadata?
(maybe because it's never evicted so no need to save it on the disk?)
There was a problem hiding this comment.
There's no point for libafl_libfuzzer -- all of the metadata will be recovered at runtime for effectively no cost.
|
Let's make this a general corpus then? |
|
but addison said it doesn't work with remove/replace |
|
It should work if we just always create new (and remove keeps the guy around on disk, who cares) |
This might be a problem because the entry will be reloaded on corpus load next time the fuzzer is started. |
addisoncrump
force-pushed
the
libfuzzer-corpus
branch
from
0ec1dbc to
45da9e9
Compare
addisoncrump
force-pushed
the
libfuzzer-corpus
branch
from
45da9e9 to
4e46c27
Compare
| _idx: CorpusId, | ||
| _testcase: Testcase<Self::Input>, | ||
| ) -> Result<Testcase<Self::Input>, Error> { | ||
| unimplemented!("It is unsafe to use this corpus variant with replace!"); |
There was a problem hiding this comment.
Why not? Just create a new file and link to the new file, keep the old one around?
There was a problem hiding this comment.
This would mean that the old corpus entry would be loaded at start time. Better in this case (since this will only ever be used by libafl_libfuzzer_runtime) to keep it tight to requirements and fail fast if we do anything new 🙂
This libafl_libfuzzer corpus had several annoying issues previously, esp. with duplication of existing inputs. This should fix it :)