-
Notifications
You must be signed in to change notification settings - Fork 972
Code Similarity Detector
slither-simil
uses state-of-the-art machine learning to detect similar (vulnerable) Solidity functions. We have provided a pretrained model from etherscan_verified_contracts with 60,000 contracts and more than 850,000 functions to get you started quickly. We included the capability to easily train new models if you have access to larger or different datasets.
slither-simil
uses FastText, a vector embedding technique, to generate compact numerical representations of every function. We used FastText because it:
- implements several state-of-the-art techniques such as nbow and skipgrams,
- has high performance (it is C++ code with Python bindings),
- has MIT license, and
- is well maintained (by Facebook).
Install the required packages before using slither-simil
:
$ pip3 install pybind11 --user
$ pip3 install https://github.com/facebookresearch/fastText/archive/0.2.0.zip --user
$ pip3 install sklearn matplotlib --user # for plot mode
Make sure that you are using pip3.6
or later. If you are running from inside a virtualenv, remove the --user
parameter.
Note that these examples will use the following files:
slither-simil
has three modes:
-
test
- finds similar functions to your own in a dataset of contracts -
plot
- provide a visual representation of similarity of multiple sampled functions -
train
- builds new models of large datasets of contracts -
info
- inspects the internal information of the pre-trained model or the assessed code
This mode transforms a function into a vector and uses it to find similar functions.
Test mode requires the following parameters:
- A pre-trained model: this file will be used to transform every function into a vector, you can train your own or use our pre-trained one (etherscan_verified_contracts.bin).
- A contract filename: this file will contain the code that you want to compare,
- A function name (e.g.
SafeMath.add
oradd
), - An input directory or file: this can be either a directory with contracts or a cache file with a pre-computed list of vectors for every contract (cache.npz).
Use the cache to avoid long processing times to compile and vectorize the input contracts.
Here's an example that finds functions similar to sendCoin
in MetaCoin
(compiled with solc-0.4.25
). Searching for similar functions among more than 800,000 functions takes only 20 seconds.
$ slither-simil test etherscan_verified_contracts.bin --filename MetaCoin.sol --fname MetaCoin.sendCoin --input cache.npz --ntop 25 --solc solc-0.4.25
INFO:Slither-simil:Reviewed 825062 functions, listing the 25 most similar ones:
INFO:Slither-simil:filename contract function score
INFO:Slither-simil:0x954b5de09a55e59755acbda29e1eb74a45d30175_Fluz.sol Fluz transfer 1.0
INFO:Slither-simil:0x55648de19836338549130b1af587f16bea46f66b_Pebbles.sol Pebbles transfer 1.0
INFO:Slither-simil:0x3fcee23add6e86dde3c4d395cbce1cae7f16d06d_SnipCoin.sol SnipCoin sendCoin 1.0
INFO:Slither-simil:0x000000005fbe2cc9b1b684ec445caf176042348e_ProperProposal.sol Vote transfer 1.0
INFO:Slither-simil:0x000000002bb43c83ece652d161ad0fa862129a2c_AccountRegistry.sol Vote transfer 1.0
INFO:Slither-simil:0x4e84e9e5fb0a972628cf4568c403167ef1d40431_Fluzcoin.sol Fluzcoin transfer 1.0
INFO:Slither-simil:0x334eec1482109bd802d9e72a447848de3bcc1063_AirDropToken.sol AirDropToken transfer 1.0
INFO:Slither-simil:0x28ccdda197d319a241005b9c9f01bac48b90f556_AirDropToken.sol AirDropToken transfer 1.0
INFO:Slither-simil:0x000000002647e16d9bab9e46604d75591d289277_Vote.sol Vote transfer 1.0
INFO:Slither-simil:0xc6c4c7826D44ABF22c711E8E86bDC3f5242d2182_token.sol token sendCoin 1.0
INFO:Slither-simil:0x22033df1d104736ff4c2b23a28affe52863ca9c8_AtmOnlyFluzcoin.sol AtmOnlyFluzcoin transfer 1.0
INFO:Slither-simil:0xcad796d6a2c0bb1de7f24262819be96fb08c1c3a_Love.sol Love transfer 1.0
INFO:Slither-simil:0x6cb2b8dc6a508c9a21db9683d1a729715969a6ee_TokenEscrow.sol TokenEscrow transferFromOwner 0.996
INFO:Slither-simil:0xd75fefe3cdb647281eec3f8fc738e3bc9658f9e4_ProofOfReadToken.sol ProofOfReadToken transfer 0.996
INFO:Slither-simil:0x7A8Ef7E8c8f16B9D6F39069ce03d752Af23b46d6_OBS_V1.sol MyObs transfer 0.996
INFO:Slither-simil:0x5ac0197c944c961f58bb02f3d0df58a74fdc15b6_TokenEscrow.sol TokenEscrow transferFromOwner 0.996
INFO:Slither-simil:0x69719c8c207036bdfc3632ccc24b290fb7240f4a_BitPayToken.sol BitPayToken transfer 0.996
INFO:Slither-simil:0x3d8a10ce3228cb428cb56baa058d4432464ea25d_TestToken.sol TestToken transfer 0.993
INFO:Slither-simil:0x69719c8c207036bdfc3632ccc24b290fb7240f4a_BitPayToken.sol BitPayToken transferFrom 0.992
INFO:Slither-simil:0x486e1f44b2a85150a6dd2de5aab87df375cd8880_CAIRToken.sol StandardToken transfer 0.991
INFO:Slither-simil:0x2bec16b164725efc192b7ec0296f838c61317514_eda.sol StandardToken transfer 0.991
INFO:Slither-simil:0xb7cb1c96db6b22b0d3d9536e0108d062bd488f74_WaltonToken.sol StandardToken transfer 0.991
INFO:Slither-simil:0x346c3be6aebEBaF5Cb766a75aDc9827EfbB7E41A_DelphiToken.sol StandardToken transfer 0.991
INFO:Slither-simil:0xf5068761511594c82328102f4fde4650ed9ea6c4_WHP.sol WHP transfer 0.991
INFO:Slither-simil:0x5f9f2ae7150d0beef3bb50ac8d8f4b43e6a6cc57_NABC.sol NABC transfer 0.991
Train mode trains new models used to vectorize functions. You will need a large amount of contracts/functions if you plan to train a new model.
$ slither-simil train model.bin --input contracts
INFO:Slither-simil:Saving extracted data into last_data_train.txt
INFO:Slither-simil:Starting training
Read 0M words
Number of words: 348
Number of labels: 0
Progress: 100.0% words/sec/thread: 53124 lr: 0.000000 loss: 2.066949 ETA: 0h 0m
INFO:Slither-simil:Training complete
INFO:Slither-simil:Saving model
INFO:Slither-simil:Saving cache in cache.npz
INFO:Slither-simil:Done!
After it runs, the slither-simil
will output the the trained model in model.bin
, a cache of every function for use in test mode in cache.npz
, and the SlithIR of every function for debugging in last_data_train.txt
.
Plot mode plots sets of functions to visually detect clusters of similar ones.
Here's an example to plot all the functions named add
from contracts named SafeMath
sampling from 500 random contracts:
$ slither-simil plot etherscan_verified_contracts.bin --fname SafeMath.add --input cache.npz --nsamples 500
INFO:Slither-simil:Loading data..
INFO:Slither-simil:Procesing data..
INFO:Slither-simil:Plotting data..
INFO:Slither-simil:Saving figure to plot.png..
This mode performs dimensionality reduction using PCA, so the axes you see here are not associated with any particular unit.
It can can be also used to plot sets of functions using only a name from any contract (e.g. burn
) .
This mode has two features. You can inspect the internal information about a pre-trained model. Info mode is typically used for debugging.
$ slither-simil info etherscan_verified_contracts.bin
INFO:Slither-simil:etherscan_verified_contracts.bin uses the following words:
INFO:Slither-simil:</s>
INFO:Slither-simil:index(uint256)
INFO:Slither-simil:return
INFO:Slither-simil:condition(temporary_variable)
INFO:Slither-simil:member
INFO:Slither-simil:solidity_call(require(bool))
INFO:Slither-simil:library_call
INFO:Slither-simil:binary(+)
INFO:Slither-simil:event
INFO:Slither-simil:(local_solc_variable(default)):=(temporary_variable)
...
... or examine the internal representation of function:
$ slither-simil info etherscan_verified_contracts.bin --filename MetaCoin.sol --fname MetaCoin.sendCoin --solc solc-0.4.25
INFO:Slither-simil:Function sendCoin in contract MetaCoin is encoded as:
INFO:Slither-simil:index(uint256) binary(<) condition(temporary_variable) return index(uint256) binary(-) index(uint256) binary(+) event return
INFO:Slither-simil:[ 0.00689753 -0.05349572 -0.06854086 -0.01667773 0.1259813 -0.05974023
0.06719872 -0.04520541 0.13745852 0.14690697 -0.03721125 0.00579037
0.06865194 -0.03804035 0.01224702 -0.1014601 -0.02655532 -0.15334933
...