libafl-fuzz: fix not loading seeds recursively from directories #2532
Conversation
|
Do we want the corpus traversing as function in the main lib/bolts? |
|
we already have it StdState::next_file() i think you should export some of them as public functions and use them instead of writing this from scratch |
| Err(e) => return Err(e), | ||
| } | ||
| } | ||
| self.reset_initial_files_state(); |
There was a problem hiding this comment.
Isn't this the exact opposite of what we want?
We want to add all initial files in order to then load them - not clear them(?)
Keep in mind most fuzzers can restart during input loading
There was a problem hiding this comment.
@domenukk we are popping them anyways. For libafl-fuzz, we want to copy the corpora first to the queue directory before loading them, and thus the path will be different already and we cannot re-use remaining_initial_files etc.
There was a problem hiding this comment.
I don't fully get it sorry, is remaining_initial_files refilled after this point then? Maybe not reset inside this method?
There was a problem hiding this comment.
@domenukk remaining initial files will be empty by the time the walking the corpus or loading the corpus is done but do_no_reenter may still be populated and this is why we need to reset the state.
There was a problem hiding this comment.
I guess I don't have to fully get it - but still, reset_initial_files_state shoudn't be in a walk_initial_inputs method, they are unrelated.
There was a problem hiding this comment.
@domenukk why not? I'm thinking from a usage perspective. what's the point of having files in do_not_reenter after we're done walking? we won't traverse those symlinks again in load_initial_inputs without resetting the state. So IMO it makes sense to reset it after anyways.
There was a problem hiding this comment.
Ok I'm not 100% convinced this is the cleanest implementation, but it is the one we have :)
No description provided.