|0.1.3||Aug 26, 2021|
|0.1.2||Aug 25, 2021|
|0.1.1||Feb 23, 2021|
|0.1.0||Feb 22, 2021|
#117 in Testing
Siderophile finds the "most unsafe" functions in your Rust codebase, so you can fuzz them or refactor them out entirely. It checks the callgraph of each function in the codebase, estimates how many
unsafe expressions are called in an evalutation of that function, then produces a list sorted by this value. Here's what Siderophile's output format looks like:
Badness Function 092 <myProject::myThing as my_project::myThing>::tempt_fate 064 <myProject::myOtherThing::whatever as my_project::myThing>::defy_death [...]
"Badness" of a function is simply an approximation of how many unsafe expressions are evaluated during an evaluation of that function. For instance, marking unsafe functions with a
*, suppose your function
f calls functions
i*. Then the badness of
f is 2. Functions with high badness have a lot of opportunities to be memory unsafe.
Make sure that you have the following requirements:
cargomust be installed and in your
- LLVM 12 is required. Older versions may work (see https://crates.io/crates/llvm-ir) but require the
llvm-irpackage's features change in
Cargo.tomlbefore compiling Siderophile.
cargo build --release, and you'll have a Siderophile binary :)
Make sure that you followed the above steps, then do the following:
cdto the root directory of the crate you want to analyze
SIDEROPHILE_LOCATION/target/release/siderophile --crate-name CRATENAME, where
CRATENAMEis the name of the crate you want to analyze, and
SIDEROPHILE_LOCATIONis the location where you put the siderophile code (you know, normal running-rust-binary stuff).
Functions are written to stdout, ordered by their badness.
cargo-geiger, whose goal is to find unsafety at the crate-level.
First, the callgraph is created by having
cargo output the crate's bitcode, then parsing it to produce a callgraph and demangle the names into things that we can match with the source code.
siderophile finds all the sources of the current crate, finds every Rust file in the sources, and parses each file individually using the
syn crate. Each file is recursively combed through for unsafety occurring in functions, trait declarations, trait implementations, and submodules.
siderophile will output the path of these objects, along with an indication of what type of syntactic block they were found in. The list received from this step contains every unsafe block in every dependency of the crate, regardless of whether it's used. To narrow this down, we need to compare
siderophile's list to nodes in the callgraph of the crate.
Using the callgraph produced in the first step, we check which elements from the
siderophile output are actually executed from the crate in question. This step (implemented in
src/callgraph_matching) is not guaranteed to find everything, but it has shown good results against manual search. It is also not immune to false positives, although none have been found yet. The labels of the nodes that are found to be unsafe are used as input for the final step.
The final step is to trace these unsafe nodes in the callgraph. For each node in the list,
siderophile will find every upstream node in the callgraph, and increment their badness by one, thus indicating that they use unsafety at some point in their execution. At the end of this process, all the nodes with nonzero badness are printed out, sorted in descending order by badness.
Siderophile is not guaranteed to catch all the unsafety in a crate's deps. Since things are only tagged at a source-level, we do not have the ability to inspect macros or resolve dynamically dispatched methods. Accordingly, this tool should not be used to "prove" that a crate uses no unsafety.
To get debugging output from
siderophile, set the
RUST_LOG environment variable to
XXX can be
Siderophile is licensed and distributed under the AGPLv3 license. Contact us if you're looking for an exception to the terms.