Next-gen Rowhammer fuzzer that uses non-uniform, frequency-based patterns.

Overview

Blacksmith Rowhammer Fuzzer

Academic Code Language Badge License: MIT contributions welcome

Preprint: arXiv Paper Funding

This repository provides the code accompanying the paper Blacksmith: Scalable Rowhammering in the Frequency Domain that is to appear in the IEEE conference Security & Privacy (S&P) 2022.

This is the implementation of our Blacksmith Rowhammer fuzzer. This fuzzer crafts novel non-uniform Rowhammer access patterns based on the concepts of frequency, phase, and amplitude. Our evaluation on 40 DIMMs showed that it is able to bypass recent Target Row Refresh (TRR) in-DRAM mitigations effectively and as such can could trigger bit flips on all 40 tested DIMMs.

Getting Started

Following, we quickly describe how to build and run Blacksmith.

Prerequisites

Blacksmith has been tested on Ubuntu 18.04 LTS with Linux kernel 4.15.0. As the CMakeLists we ship with Blacksmith downloads all required dependencies at compile time, there is no need to install any package other than g++ (>= 8) and cmake (>= 3.14).

To facilitate the development, we also provide a Docker container (see Dockerfile) where all required tools and libraries are installed. This container can be configured, for example, as remote host in the CLion IDE, which automatically transfers the files via SSH to the Docker container (i.e., no manual mapping required).

Building Blacksmith

You can build Blacksmith with its supplied CMakeLists.txt in a new build directory:

mkdir build \ 
  && cd build \
  && cmake .. \
  && make -j$(nproc)

Now we can run Blacksmith. For example, we can run Blacksmith in fuzzing mode by passing a random DIMM ID (e.g., --dimm-id 1; only used internally for logging into stdout.log), we limit the fuzzing to 6 hours (--runtime-limit 21600), pass the number of ranks of our current DIMM (--ranks 1) to select the proper bank/rank functions, and tell Blacksmith to do a sweep with the best found pattern after fuzzing finished (--sweeping):

sudo ./blacksmith --dimm-id 1 --runtime-limit 21600 --ranks 1 --sweeping  

While Blacksmith is running, you can use tail -f stdout.log to keep track of the current progress (e.g., patterns, found bit flips). You will see a line like

[!] Flip 0x2030486dcc, row 3090, page offset: 3532, from 8f to 8b, detected after 0 hours 6 minutes 6 seconds.

in case that a bit flip was found. After finishing the Blacksmith run, you can find a fuzz-summary.json that contains the information found in the stdout.log in a machine-processable format. In case you passed the --sweeping flag, you can additionally find a sweep-summary-*.json file that contains the information of the sweeping pass.

Supported Parameters

Blacksmith supports the command-line arguments listed in the following. Except for the parameters --dimm-id and --ranks all other parameters are optional.

    -h, --help
        shows this help message

==== Mandatory Parameters ==================================

    -d, --dimm-id
        internal identifier of the currently inserted DIMM (default: 0)
    -r, --ranks
        number of ranks on the DIMM, used to determine bank/rank/row functions, assumes Intel Coffe Lake CPU (default: None)
    
==== Execution Modes ==============================================

    -f, --fuzzing
        perform a fuzzing run (default program mode)        
    -g, --generate-patterns
        generates N patterns, but does not perform hammering; used by ARM port
    -y, --replay-patterns <csv-list>
        replays patterns given as comma-separated list of pattern IDs

==== Replaying-Specific Configuration =============================

    -j, --load-json
        loads the specified JSON file generated in a previous fuzzer run, required for --replay-patterns
        
==== Fuzzing-Specific Configuration =============================

    -s, --sync
        synchronize with REFRESH while hammering (default: 1)
    -w, --sweeping
        sweep the best pattern over a contig. memory area after fuzzing (default: 0)
    -t, --runtime-limit
        number of seconds to run the fuzzer before sweeping/terminating (default: 120)
    -a, --acts-per-ref
        number of activations in a tREF interval, i.e., 7.8us (default: None)
    -p, --probes
        number of different DRAM locations to try each pattern on (default: NUM_BANKS/4)

The default values of the parameters can be found in the struct ProgramArguments.

Configuration parameters of Blacksmith that we did not need to modify frequently, and thus are not runtime parameters, can be found in the GlobalDefines.hpp file.

Citing our Work

To cite Blacksmith in academic papers, please use the following BibTeX entry:

@inproceedings{jattke2021blacksmith,
  title = {{{BLACKSMITH}}: Rowhammering in the {{Frequency Domain}}},
  shorttitle = {Blacksmith},
  booktitle = {{{IEEE S}}\&{{P}} '22},
  author = {Jattke, Patrick and {van der Veen}, Victor and Frigo, Pietro and Gunter, Stijn and Razavi, Kaveh},
  year = {2021},
  month = nov,
  note = {\url{https://comsec.ethz.ch/wp-content/files/blacksmith_sp22.pdf}}
}
Comments
  • mmap: Invalid argument

    mmap: Invalid argument

    after installing blacksmith successfully and setting up the hugepage to 1 GB I tested the following : sudo ./blacksmith --dimm-id 1 --runtime-limit 120 --ranks 1 and i get this error message: mmap: Invalid argument how can I check if my --dimm-id is valid or not? I think its the argument that creates this issue! my OS : Linux ubuntu 5.11.0-27-generic 64-bit

    opened by AnaMazda 6
  • Blacksmith terminated: Illegal instruction

    Blacksmith terminated: Illegal instruction

    After setting the hugepage size to 1G and build the blacksmith successfully, the program ends with the output "Illegal Instructions" and there is no content in the stdout.log

    opened by HxJi 5
  • Blacksmith not running: mmap: Cannot allocate memory

    Blacksmith not running: mmap: Cannot allocate memory

    Hello,

    I wanted to try your fuzzer on various computers but I always end up with the mmap: Cannot allocate memory error. I thought this would come from my configuration so I tried to increase the number of available huge pages.

    I currently have the following memory configuration regarding huge pages:

    ▶ cat /proc/meminfo|grep Huge                         
    AnonHugePages:         0 kB
    ShmemHugePages:        0 kB
    FileHugePages:         0 kB
    HugePages_Total:     535
    HugePages_Free:      535
    HugePages_Rsvd:        0
    HugePages_Surp:        0
    Hugepagesize:       2048 kB
    Hugetlb:         1095680 kB
    

    On other devices I could even reach more than 1000 free huge pages, which I believe is enough for allocating 1GB of memory with huge pages. However the issue seems to come from somewhere else. I tried the execution on two different devices with ArchLinux, Debian 11 and Ubuntu 18.04 LTS with no success.

    Am I missing something ?

    opened by T-TROUCHKINE 5
  • Does this work on WSL2?

    Does this work on WSL2?

    Hi. I got it working on my gen3 I7 build, but I was wondering if this works on WSL2?

    [email protected]:~/blacksmith-public/build$ sudo ./blacksmith --dimm-id 2 --runtime-limit 21600 --ranks 1 --sweeping Writing into logfile stdout.log [email protected]:~/blacksmith-public/build$ sudo ./blacksmith --dimm-id 1 --runtime-limit 21600 --ranks 1 --sweeping

    [+] General information about this fuzzing run: Start timestamp:: 1637689072 Hostname: PSTEJSKA03-PC Commit SHA: NO_REPOSITORY Run time limit: 21600 (6 hours 0 minutes 0 seconds) [+] Printing run configuration (GlobalDefines.hpp): DRAMA_ROUNDS: 1000 CACHELINE_SIZE: 64 HAMMER_ROUNDS: 1000000 THRESH: 495 NUM_TARGETS: 10 MAX_ROWS: 30 NUM_BANKS: 16 DIMM: 1 CHANNEL: 1 MEM_SIZE: 1073741824 PAGE_SIZE: 4096

    [+] Initializing memory with pseudorandom sequence. [-] Could not find conflicting address sets. Is the number of banks (16) defined correctly?

    opened by MrObvious 4
  • blacksmith doesnt work :  /mnt/huge/buff not found

    blacksmith doesnt work : /mnt/huge/buff not found

    after running blacksmith with default param as mentioned in the description, it stopped immediately with the following erreur in the logfile :

    ` [+] General information about this fuzzing run: Start timestamp:: 1637072011 Hostname: 1cc27a1cdb50 Commit SHA: c8e65b709a83665f9528efdedcf064abdb04859f Run time limit: 120 (0 hours 2 minutes 0 seconds) [+] Printing run configuration (GlobalDefines.hpp): DRAMA_ROUNDS: 1000 CACHELINE_SIZE: 64 HAMMER_ROUNDS: 1000000 THRESH: 495 NUM_TARGETS: 10 MAX_ROWS: 30 NUM_BANKS: 16 DIMM: 1 CHANNEL: 1 MEM_SIZE: 1073741824 PAGE_SIZE: 4096

    [-] Instruction setpriority failed. [+] Could not mount superpage from /mnt/huge/buff. Error: `

    opened by AnaMazda 3
  • Blacksmith run hangs and log shows strange characters

    Blacksmith run hangs and log shows strange characters

    Hi

    I installed blacksmith on a i3-8350k (Coffee-Lake-S) System. Unfortunately the test hangs after a while and the stdout.log shows some strange characters. Does anyone have an idea what could be the reason of this?

    BR JKR stdout_2022_02_04_hangs.log

    opened by JKRde 2
  • Could not find conflicting address sets

    Could not find conflicting address sets

    I'm not able to get past this error even when recompiling with different NUM_BANKS value - I tried 4, 8, 16 and even 32. Always the same output. I'm not sure what other parameters to adjust as the error message doesn't suggest anything else.

    My output is:

    [+] General information about this fuzzing run:
    Start timestamp:: 1637333603
    Hostname: ubuntu
    Commit SHA: c8e65b709a83665f9528efdedcf064abdb04859f
    Run time limit: 21600 (6 hours 0 minutes 0 seconds)
    [+] Printing run configuration (GlobalDefines.hpp):
    DRAMA_ROUNDS: 1000
    CACHELINE_SIZE: 64
    HAMMER_ROUNDS: 1000000
    THRESH: 495
    NUM_TARGETS: 10
    MAX_ROWS: 30
    NUM_BANKS: 32
    DIMM: 1
    CHANNEL: 1
    MEM_SIZE: 1073741824
    PAGE_SIZE: 4096
    
    [+] Initializing memory with pseudorandom sequence.
    [-] Could not find conflicting address sets. Is the number of banks (32) defined correctly?
    

    My kernel is 5.13.0-19-generic on ubuntu 21.10

    Any help is appreciated.

    opened by DominikBucko 2
  • Unable to compile on ARM processor

    Unable to compile on ARM processor

    BlackSmith 0.0.2 has no support for ARM processors:

    [81%] Building CXX object CMakeFiles/bs.dir/src/Fuzzer/AggressorAccessPattern.cpp.o In file included from /home/parallels/blacksmith/include/Memory/DramAnalyzer.hpp:13, from /home/parallels/blacksmith/include/Memory/Memory.hpp:13, from /home/parallels/blacksmith/include/Forges/TraditionalHammerer.hpp:9, from /home/parallels/blacksmith/src/Forges/TraditionalHammerer.cpp:1: /home/parallels/blacksmith/include/Utilities/AsmPrimitives.hpp: In static member function ‘static void TraditionalHammerer::hammer_sync(std::vector<volatile char*>&, int, volatile char*, volatile char*)’: /home/parallels/blacksmith/include/Utilities/AsmPrimitives.hpp:56:3: error: unknown register name ‘%rcx’ in ‘asm’ 56 | asm volatile("rdtscp\n" | ^~~ /home/parallels/blacksmith/include/Utilities/AsmPrimitives.hpp:56:3: error: unknown register name ‘%rcx’ in ‘asm’ 56 | asm volatile("rdtscp\n" | ^~~ /home/parallels/blacksmith/include/Utilities/AsmPrimitives.hpp:56:3: error: unknown register name ‘%rcx’ in ‘asm’ 56 | asm volatile("rdtscp\n" | ^~~ /home/parallels/blacksmith/include/Utilities/AsmPrimitives.hpp:56:3: error: unknown register name ‘%rcx’ in ‘asm’ 56 | asm volatile("rdtscp\n" | ^~~ make[2]: *** [CMakeFiles/bs.dir/build.make:104: CMakeFiles/bs.dir/src/Forges/TraditionalHammerer.cpp.o] Error 1 make[2]: *** Waiting for unfinished jobs.... make[1]: *** [CMakeFiles/Makefile2:387: CMakeFiles/bs.dir/all] Error 2 make: *** [Makefile:136: all] Error 2

    opened by UkeraGan 1
  • fix THRESH comment

    fix THRESH comment

    In my opinion, THRESH is the threshold to distinguish row buffer miss rather than cache miss since in function measure_time() each memory access is followed by a clflushopt to flush it from cache.

    opened by Emoth97 0
  • Fuzzer unable to find patterns on some DIMMs

    Fuzzer unable to find patterns on some DIMMs

    Hi @pjattke ,

    I've used the Blacksmith fuzzer to find patterns that produce a large number of bit flips on some DIMMs. However, on other DIMMs from the same manufacturer and having similar geometry (same number of ranks and banks), I have not managed to produce even a single bit flip even after repeated invocations of the fuzzer (I've roughly run the fuzzer 6 different times, each fuzzing for a duration of 6 hours). I assume it is unlikely for these DIMMs to be completely robust to the Rowhammer exploit and exploring the search space further should produce bit flips? Did you also come across something similar in your experiments? Do you have any practical advice (perhaps alter the THRESH value defined in GlobalDefines.hpp or run the fuzzer on a particular CPU) so I can produce bit flips on these DIMMs too?

    Let me know if you would require further information and thanks again for your time! cc @kaustav-goswami and @dxaen

    opened by hariv 1
  • Some questions regarding the use of time-based side channels in blacksmith

    Some questions regarding the use of time-based side channels in blacksmith

    Hi @pjattke, I have some questions regarding the use of some time-based side channels in the blacksmith code.

    • If I understood correctly, the find_bank_conflicts() method of DramAnalyzer is using a timing side-channel to find addresses that map to each DRAM bank. However, since blacksmith also uses DRAMA to figure out the DRAM functions to map physical addresses to the DRAM geometry (channel, rank, bank, row, etc) what is the need for this side channel?

    • find_bank_conflicts() checks if the time that is taken to access 2 addresses is above a threshold to determine if the 2 addresses belong to the same bank. How did you determine this threshold? My understanding is that the code is looking for Row buffer misses when accessing the 2 addresses (which would take longer implying that they belong to the same bank), but how did you set a value to the threshold? Is the threshold dependent on each individual DIMM or does it depend on the microarchitecture? Also, why is it that the same pair of addresses is checked twice? Is this done to account for jitter?

    • Lastly, the hammer_sync() method of TraditionalHammerer uses a timing side-channel to detect the start of a refresh interval to synchronize hammering within the interval. The timing side-channel uses 2 addresses in the same bank in order to do the sync. Is there any reason as to why the method uses 2 addresses? Can detecting the start of a refresh be detected just by accessing a single address?

    Thanks for your time and wish you a happy new year. cc @kaustav-goswami and @dxaen.

    opened by hariv 3
  • Packaging of Blacksmith in Guix.

    Packaging of Blacksmith in Guix.

    @jgarte and I, with the help of other volunteers, are packaging Blacksmith in Guix. Once completed, Blacksmith can be deployed on any GNU+Linux distribution, with or without Guix, in a reproducible way.

    I am opening this thread so that we can update our progress, including any issues.

    opened by ghost 3
  • Blacksmith on non-Coffee Lake CPUs

    Blacksmith on non-Coffee Lake CPUs

    Did anyone try running blacksmith on CPUs other than Coffee Lake?

    I was able to run it successfully on Kaby Lake, but it didn't work on Comet Lake. It errors out immediately saying it could not find conflicting address sets and asks if the number of banks has been defined correctly (which I checked is correct).

    opened by hariv 14
Releases(0.0.2)
Owner
Computer Security Group @ ETH Zurich
Computer Security Group @ ETH Zurich
unet for image segmentation

Implementation of deep learning framework -- Unet, using Keras The architecture was inspired by U-Net: Convolutional Networks for Biomedical Image Seg

zhixuhao 4.1k Dec 31, 2022
Neural Re-rendering for Full-frame Video Stabilization

NeRViS: Neural Re-rendering for Full-frame Video Stabilization Project Page | Video | Paper | Google Colab Setup Setup environment for [Yu and Ramamoo

Yu-Lun Liu 9 Jun 17, 2022
Experiments with differentiable stacks and queues in PyTorch

Please use stacknn-core instead! StackNN This project implements differentiable stacks and queues in PyTorch. The data structures are implemented in s

Will Merrill 141 Oct 06, 2022
Perform Linear Classification with Multi-way Data

MultiwayClassification This is an R package to perform linear classification for data with multi-way structure. The distance-weighted discrimination (

Eric F. Lock 2 Dec 15, 2020
Rule Based Classification Project

Kural Tabanlı Sınıflandırma ile Potansiyel Müşteri Getirisi Hesaplama İş Problemi: Bir oyun şirketi müşterilerinin bazı özelliklerini kullanaraknseviy

Şafak 1 Jan 12, 2022
JORLDY an open-source Reinforcement Learning (RL) framework provided by KakaoEnterprise

Repository for Open Source Reinforcement Learning Framework JORLDY

Kakao Enterprise Corp. 330 Dec 30, 2022
SMORE: Knowledge Graph Completion and Multi-hop Reasoning in Massive Knowledge Graphs

SMORE: Knowledge Graph Completion and Multi-hop Reasoning in Massive Knowledge Graphs SMORE is a a versatile framework that scales multi-hop query emb

Google Research 135 Dec 27, 2022
Codebase for the solution that won first place and was awarded the most human-like agent in the 2021 NeurIPS Competition MineRL BASALT Challenge.

KAIROS MineRL BASALT Codebase for the solution that won first place and was awarded the most human-like agent in the 2021 NeurIPS Competition MineRL B

Vinicius G. Goecks 37 Oct 30, 2022
A framework to train language models to learn invariant representations.

Invariant Language Modeling Implementation of the training for invariant language models. Motivation Modern pretrained language models are critical co

6 Nov 16, 2022
Roach: End-to-End Urban Driving by Imitating a Reinforcement Learning Coach

CARLA-Roach This is the official code release of the paper End-to-End Urban Driving by Imitating a Reinforcement Learning Coach by Zhejun Zhang, Alexa

Zhejun Zhang 118 Dec 28, 2022
Self-Supervised Generative Style Transfer for One-Shot Medical Image Segmentation

Self-Supervised Generative Style Transfer for One-Shot Medical Image Segmentation This repository contains the Pytorch implementation of the proposed

Devavrat Tomar 19 Nov 10, 2022
A distributed deep learning framework that supports flexible parallelization strategies.

FlexFlow FlexFlow is a deep learning framework that accelerates distributed DNN training by automatically searching for efficient parallelization stra

528 Dec 25, 2022
Pytorch GUI(demo) for iVOS(interactive VOS) and GIS (Guided iVOS)

GUI for iVOS(interactive VOS) and GIS (Guided iVOS) GUI Implementation of CVPR2021 paper "Guided Interactive Video Object Segmentation Using Reliabili

Yuk Heo 13 Dec 09, 2022
Official implementation of the paper Vision Transformer with Progressive Sampling, ICCV 2021.

Vision Transformer with Progressive Sampling This is the official implementation of the paper Vision Transformer with Progressive Sampling, ICCV 2021.

yuexy 123 Jan 01, 2023
Pytorch implementation for the paper: Contrastive Learning for Cold-start Recommendation

Contrastive Learning for Cold-start Recommendation This is our Pytorch implementation for the paper: Yinwei Wei, Xiang Wang, Qi Li, Liqiang Nie, Yan L

45 Dec 13, 2022
Art Project "Schrödinger's Game of Life"

Repo of the project "Team Creative Quantum AI: Schrödinger's Game of Life" Installation new conda env: conda create --name qcml python=3.8 conda activ

ℍ◮ℕℕ◭ℍ ℝ∈ᛔ∈ℝ 2 Sep 15, 2022
Reproduces ResNet-V3 with pytorch

ResNeXt.pytorch Reproduces ResNet-V3 (Aggregated Residual Transformations for Deep Neural Networks) with pytorch. Tried on pytorch 1.6 Trains on Cifar

Pau Rodriguez 481 Dec 23, 2022
PyTorch Implementation of AnimeGANv2

PyTorch implementation of AnimeGANv2

4k Jan 07, 2023
LBK 20 Dec 02, 2022