Differences

This shows you the differences between two versions of the page.

--- use_john_the_ripper_to_crack_password_hashes [31.01.2021 01:52] – created Pascal Suter
+++ use_john_the_ripper_to_crack_password_hashes [31.01.2021 22:11] – [compile with OpenCL support to run on NVIDIA (and other) GPU's (tested on CentOS)] Pascal Suter
@@ Line 9: / Line 9: @@
   ../run/unshadow ~/tmp/passwd ~/tmp/shadow > ~/tmp/unshadow
 ===== use john with openMP on multiple cores =====
+  export OMP_NUM_THREADS=8
+  john ~/tmp/unshadow
+===== use john with multiple cores by forking the process =====
   john --fork=8 ~/tmp/unshadow
@@ Line 15: / Line 19: @@
 ===== compile john with MPI support to run on clusters =====
+ubuntu dependencies
   sudo apt install libopenmpi-dev openmpi-bin
+centos dependencies / preparation:
+  yum install openmpi-devel openmpi openssl-devel
+  module load mpi/openmpi-x86_64
+get source and compile:
   wget https://github.com/openwall/john/archive/1.9.0-Jumbo-1.tar.gz
   tar xvf 1.9.0-Jumbo-1.tar.gz
@@ Line 26: / Line 35: @@
 to check the progress run
   kill -USR1 $(pidof mpirun)
+===== compile with OpenCL support to run on NVIDIA (and other) GPU's (tested on CentOS) =====
+download and install cuda if you haven't already
+  wget https://developer.download.nvidia.com/compute/cuda/11.2.0/local_installers/cuda_11.2.0_460.27.04_linux.run
+  sudo sh cuda_11.2.0_460.27.04_linux.run
+  wget https://github.com/openwall/john/archive/1.9.0-Jumbo-1.tar.gz
+  tar xvf 1.9.0-Jumbo-1.tar.gz
+  cd john-1.9.0-Jumbo-1/src
+  ./configure LDFLAGS=-L/usr/local/cuda/targets/x86_64-linux/lib CPPFLAGS=-I/usr/local/cuda/targets/x86_64-linux/include
+the summary should show that OpenCL support is now enabled (yes)
+  make -s clean && make -sj4
+now let's run it :) here is an example where I ran john the ripper on a server with 8x NVIDIA GeForce RTX 2080 Ti:
+  run/john --format=sha512crypt-opencl -dev=gpu -fork=8  ../root.unshadow
+the ''--format'' option needs to be specified in order to use the GPU at all. without specifying a format, john will always default to the CPU implementation of the crypt algo. to figure out which format to use, start john without a format parameter and then look at the output to find the crypt that was used. now run
+  run/john --list=formats --format=opencl
+to get a list of all crypts that support opencl. if you are lucky, the one you are looking for is in there as well :)
+the ''-dev=gpu -fork=8'' options are there to use all cards in parallel. this will fork 8 individual processes each working on their own range of passwords at a time and each on a different GPU. If you have multiple hosts with GPU you may use MPI for that.
+===== continue an interrupted session =====
+John saves its status as it's working, so in case it crashes or you have to abort it because you need to work with your pc and don't want the cpu load on it for example, you can always resume the session and continue where John has left off.
+BUT.. it is **important** that you specify the ''--restore'' option, otherwise john will start over again! To avoid accidently starting over and by doing so loosing all the previous compute time, it is advisable to give each session its own **session name** by using the ''--session:<name>'' parameter. By doing so, you can avoid overwriting the session in case you start another john session in the mean time.
+so you would start john like this:
+  john --session:sess1 unshadow.txt
+and then resume the session like so:
+  john --restore:sess1
+also the other commands like for example ''status'' accept a session name parameter:
+  john --status:sess1
+===== Performance examples =====
+if you press any key during the run, you will get a status showing you c/s (crypts per second). here are a few numbers from the systems i had access to at the moment of writing this article:
+^ CPU / GPU ^ c/s rate ^ method used ^
+| 8-core i7-8809G CPU| 7'500 | OpenMP |
+| 128-core (2 socket) AMD ROME 7742| 150'000 | OpenMP |
+| 8 x GeForce RTX 2080 Ti | 8x190'000 | Fork |