Saturday, August 5, 2017

Theano Deep Learning Framework on Fedora 25

 [This article is rich in hyperlinks. I chose these to do a better job of explaining things than I can normally do. Please visit them.]

A few articles ago we covered Torch 7 and how to set it up. There are several other frameworks that are important each having advantages in one area or another. Its important to have access to all of them because you never know when a killer app lands on any one of them. Today we will show how to setup theano. Theano is one of the older frameworks and takes a unique approach to GPU acceleration. When you run a program that uses GPU acceleration, it generates and compiles CUDA code based on what your program describes.


THEANO
In the last article about AI, I mentioned that you can setup an account specifically to run AI programs. This is because most of the frameworks install things to your home directory. Sometimes they want versions of things that clash with other frameworks. Sounds like a classic use case for containers. But I wanted to set this up on bare metal so let's dive in.

Theano is python based. It typically wants things that are newer than the system python libraries. So, I'll show you how to set all this up. If you want to create a new ai account, go ahead and do that and/or log in under the account you want to set this up in.

The first step is to download miniconda which is a scaled back version of anaconda which is a package installer used by Continuum Analytics. (There is some overlap in names with Anaconda the Fedora and Red Hat package installer. They are not the same.) They have lots of scientific computing packages ready to install. Look over this list to get a feel for it.

To install miniconda, do this:

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh


Click through the license and accept the default locations.

Once it's done, source .bashrc to update your variables.

source ~/.bashrc

Now, let's install theano. First we need to tell it where our CUDA libraries are installed. If you need information on how to setup a CUDA development environment, see this blog post.

export CUDA_ROOT="/usr/local/cuda"
conda install theano pygpu python=3

This will download and install theano for python 3 and all its dependencies. Contiuum is shipping python 3.6 which is ahead of Fedora's python 3.5. Next create a .theanorc file in the homedir. In it, put this:

[cuda]
root = /usr/local/cuda

[nvcc]
flags = -std=c++11


This fixes the nvidia compiler to not choke on the gcc/glibc headers and in a more permanent way where to find the CUDA environment. It is also important at this point that you have fixed /usr/local/cuda/include/math_functions.h as I explained in the article about setting up your CUDA development environment. Theano is the one that chokes on that bad code.

Next, we should test the setup to see if it works. We will start with the bottom layer, pygpu. If this is not working, then something went wrong and nothing else will work. I took the following from this article: http://blog.mdda.net/oss/2015/07/07/nvidia-on-fedora-22. You don't have to make this a program. Just use the python shell.

$ python3
>>> import pygpu
>>> pygpu.init('cuda0')


If its working, you should see
<pygpu.gpuarray.GpuContext object at 0x7f1547e79550

Good. Let's exit.

>>> quit()


Now let's test theano itself. The idea here is to make sure it works with simple apps before you jump into a complex AI program and then find trouble. Let's make a program. Copy this into a file we'll call gpu_check.py in the homedir.


from theano import function, config, shared, sandbox
import theano.tensor as T
import numpy
import time

vlen = 10 * 30 * 768  # 10 x #cores x # threads per core
iters = 1000

rng = numpy.random.RandomState(22)
x = shared(numpy.asarray(rng.rand(vlen), config.floatX))
f = function([], T.exp(x))
print(f.maker.fgraph.toposort())
t0 = time.time()
for i in range(iters):
    r = f()
t1 = time.time()
print('Looping %d times took' % iters, t1 - t0, 'seconds')
print('Result is', r)
if numpy.any([isinstance(x.op, T.Elemwise) for x in f.maker.fgraph.toposort()]):
    print('Used the cpu')
else:
    print('Used the gpu')


We will run 2 tests. One to check that the CPU is working and one to see that the GPU is working.


# THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=cpu python3 gpu_check.py
# THEANO_FLAGS=mode=FAST_RUN,floatX=float32,device=gpu python3 gpu_check.py


When you test the gpu, if you see an errors like:

miniconda3/lib/python3.6/site-packages/numpy/core/include/numpy/ndarraytypes.h(84): error: expected a "}"
/home/ai3/miniconda3/lib/python3.6/site-packages/numpy/core/include/numpy/ndarraytypes.h(446): error: identifier "NPY_NTYPES_ABI_COMPATIBLE" is undefined
...
WARNING (theano.sandbox.cuda): CUDA is installed, but device gpu is not available  (error: cuda unavailable)
...
Used the cpu


This is normal the first time. You need to edit ~/miniconda3/lib/python3.6/site-packages/numpy/core/include/numpy/ndarraytypes.h.

On line 84, put the whole NPY_ATTR_DEPRECATE line in comments /* */ including the ending comma, save, and retest.

When you see:

Using gpu device 0: GeForce GTX 1080 Ti (CNMeM is disabled, cuDNN 5005)
...
Used the gpu


you are ready for theano...

Next blog post I'll show you something really cool that uses theano.

Wednesday, July 12, 2017

Interactive R programs

In the past, we have looked at using R to analyze audit data. The programs are kind of like batch processing. Whatever they do is predefined and you can't tell it to change without modifying the source code. Today we are going to take a look at how to make R application that respond to user input.


Shiny
The developers at RStudio created a way to marry web programming with R so that you have a web presentation layer and an R backend that responds to the changes. This brings a much needed capability because sometimes you want to see the data differently right away.

The shiny interface does bring with it a number of controls like Radio Buttons, drop down text boxes, sliders, charts, and boxes for grouping. You can take a look at a gallery of controls here.

To create a basic shiny app, open RStudio. Click on File|New File and then select "Shiny Web App". That brings up a dialog asking some basic questions. It asks what the application's name is. I put in Test. Then it asks if you want 1 file or 2. I select 1. If you choose 2, then it makes one file for the UI and one file for the back end. The last thing is to select the directory for the file. When you click on Create, it will open a file fully populated with a simple working app.

If you click "Run App", then you should have a program that looks something like this:




Moving the slider causes the histogram to change. Let's look at the code.

library(shiny)

# Define UI for application that draws a histogram
ui <- fluidPage(

   # Application title
   titlePanel("Old Faithful Geyser Data"),

   # Sidebar with a slider input for number of bins
   sidebarLayout(
      sidebarPanel(
         sliderInput("bins",
                     "Number of bins:",
                     min = 1,
                     max = 50,
                     value = 30)
      ),

      # Show a plot of the generated distribution
      mainPanel(
         plotOutput("distPlot")
      )
   )
)

# Define server logic required to draw a histogram
server <- function(input, output) {

   output$distPlot <- renderPlot({
      # generate bins based on input$bins from ui.R
      x    <- faithful[, 2]
      bins <- seq(min(x), max(x), length.out = input$bins + 1)

      # draw the histogram with the specified number of bins
      hist(x, breaks = bins, col = 'darkgray', border = 'white')
   })
}

# Run the application
shinyApp(ui = ui, server = server)



There are 2 parts to this program. The first part is the GUI. There is a call to fluid page that takes an undefined number of arguments that describe the widgets on the page. Each widget is itself a function call that takes parameters or other objects created by other functions. In the basic design, we have a title, a slider, and a plot.

On the server side we have a server object created by a function that has input and output objects. To make the GUI change, we define a distPlot sub-variable to output. We can call this anything. It just has to match what's on the GUI side. This variable is initialized by a renderPlot function which takes a few parameters to describe what to plot. It knows what to plot based on a sub-variable from the input argument, bins. This could be named anything but it has to match what the slide control has or nothing will happen.

The server side and GUI side are tied together with a function call to ShinyApp at the bottom. This is what runs the program. Under the hood, RStudio starts up a little web server that runs a cgi-bin application with an R environment that your app gets loaded into. On the front end it opens a little web browser and connects to the web server on localhost. The cgi-bin starts your session and sends a web page to draw. When you change anything in the web page, it sends a post to the cgi-bin with a new copy of all the variables in the GUI. This  immediately triggers the server code and it responds with an updated web page.

There is a nice and detailed tutorial video created by the RStudio developers if you wanted to learn more. I found it very helpful when learning Shiny. You can also browse around the widget gallery mentioned earlier. In it you can see the source code for all of these little examples.

Now let's do a simple program that does something with audit data. A long time ago, we learned how to do bar charts. That was a pretty simple program. Let's re-fit that code to run as a shiny app so that we tell it how to group the audit data.

library(shiny)
library(ggplot2)

# Read in the data and don't let strings become factors
audit <<- read.csv("~/R/audit-data/audit.csv", header=TRUE, stringsAsFactors = FALSE)
fnames <<- colnames(audit)
fnames[5] <<- "HOUR" # Change serial number to HOUR
audit$one <<- rep(1,nrow(audit))
# Create time series data frame for aggregating
audit$posixDate=as.POSIXct(paste(audit$DATE, audit$TIME), format="%m/%d/%Y %H:%M:%S")
# Create a column of hour and date to aggregate an hourly total.
audit$HOUR <- format(audit$posixDate, format = '%Y-%m-%d %H')
ourColors <<- c("red", "blue", "green", "cyan", "yellow", "orange", "black", "gray", "purple" )

# Define UI for application
ui <- shinyUI(fluidPage(
  # Application title
  titlePanel("Audit Barcharts"),

  sidebarLayout(
    sidebarPanel(
      selectInput("groupBy", "Group By", fnames, selected = "HOUR"),
      selectInput("lowColor", "Low Color", ourColors, selected = "blue"),
      selectInput("highColor", "High Color", ourColors, selected = "red"),
      width = 3
    ),
    # Show a plot of the generated distribution
    mainPanel(
      plotOutput("barPlot", width = "auto", height = "600px"),
      width = 9
    )
  )
))


# Define our server side code

server <- shinyServer(function(input, output) {
  observeEvent(c(input$groupBy, input$lowColor, input$highColor), {
    # Now summarize it
    grp <- input$groupBy

    temp <- aggregate(audit$one, by = audit[grp], FUN = length)
    temp$t <- as.character(temp[,grp])

    if (grp == "HOUR") {
      # Time based needs special handling
      final = data.frame(date=as.POSIXct(temp$t, format="%Y-%m-%d %H", tz="GMT"))
      final$num <- temp$x
      final$day <- weekdays(as.Date(final$date))
      final$oday <- factor(final$day, levels = unique(final$day))
      final$hour <- as.numeric(format(final$date, "%H"))

      output$barPlot<-renderPlot({
        pl <- ggplot(final, aes(x=final[,1], y=final$num, fill=final$num)) +
          geom_bar(stat="identity") + ggtitle(paste("Events by", grp)) +
          scale_x_datetime() + xlab("") + labs(x=grp, y="Number of Events") +
          scale_fill_gradient(low=input$lowColor, high = input$highColor, name=paste("Events/", grp, sep=""))
        print(pl)
      })
    } else {
      # non-time conversion branch
      final <- temp[,1:2]
      colnames(final) = c("factors", "num")
      final$factors <- abbreviate(final$factors, minlength = 20, strict = TRUE)

      # We will rotate based on how dense the labels are
      rot <- 90
      if (nrow(final) < 20)
        rot <- 60
      if (nrow(final) < 10)
        rot <- 45

      # Plot it
      output$barPlot<-renderPlot({
        pl <- ggplot(final, aes(x=final[,1], y=final$num, fill=final$num)) +
          geom_bar(stat="identity") + ggtitle(paste("Events by", grp)) +
          scale_x_discrete() + xlab("") + labs(x=grp, y="Number of Events") +
          scale_fill_gradient(low=input$lowColor, high = input$highColor, name=paste("Events/", grp, sep="")) +
          theme(axis.text.x = element_text(angle = rot, hjust = 1, size = 18))
        print(pl)
      })
    }
  })
})

# Run the application
shinyApp(ui = ui, server = server)



Make sure you have ~/R/audit-data/audit.csv filled with audit data. Save the above code as app.R and run it. You should see something like this:




Also notice that you can change the selection in the text drop downs and the chart is immediately redrawn. Briefly, the way this works is we setup some global data in the R environment. Next we define a GUI that has 3 selector inputs. All of the hard work is in the server function. What it does is wait for either of the 3 variables to change and if so re-draws the screen. We split the charting into 2 branches, time and everything else. The main difference is time variables need special handling. Basically we format the data to what's expected by the plotting function and pass it in. On the non-time side of things, we can get very dense groups. So what we do is rotate the text labels on the bottom if we start running out of room to fit more in.

Conclusion
This shows the basics of how a shiny app works. You can create very elaborate and complicate programs using this API. Now that we've been over Shiny basics, I'll talk about Audit Explorer next time.

Wednesday, July 5, 2017

Getting Torch running on Fedora 25

In this blog post we will setup the Torch AI framework so that it can be used on Fedora. This builds on the previous blog post which shows you how to setup a CUDA development environment for Fedora.


Torch
Torch is a Deep Learning AI framework that is written in LUA. This makes it very fast because there is little between the script and the pure C code that is performing the work. Both Facebook and Twitter are major contributors to this and have probably derived their in-house version from the open source version.

The first thing I would do is setup an account just for AI. The reason I suggest this is because we are going to be installing a bunch of software without rpm. All of this will be going into the home directory. So, if one day you want to delete it all, its as simple as deleting the account and home directory. Assuming you made the account and logged into it...

$ git clone https://github.com/torch/distro.git ~/torch --recursive
$ cd torch/
$ export CMAKE_CXX_FLAGS="-std=c++03"
$ ./install.sh


The Torch community say that they only support Torch built this way. I have tried to package Torch in rpm and it simply does not work. I get some strange errors related to math. There are probably compile options that fix this but I'm done with hunting this down. It's easier to use their method from an account just for this. But I digress...

After about 25 minutes, the build asks "Do you want to automatically prepend the Torch install location to PATH and LD_LIBRARY_PATH in your /home/ai/.bashrc? (yes/no)"

I typed "yes" to have it update ~/.bashrc. I logged out and back in. Test to see if the GPU based Torch is working:

luajit -lcutorch
luajit -lcunn


This should produce errors if its not working. To exit the shell, type:

os.exit()


At this point only one last thing is needed. We may want to play with machine vision at some point so get the camera module. And a lot of models seem to be trained using the Caffe Deep Learning framework. This means we need load it from that format so let's grab the loadcaffe module.

During the build of Torch, you got a copy of luarocks which is a package manager for LUA modules. We can use this to pull in the modules so that Torch can use them.

$ luarocks install camera
$ luarocks install loadcaffe


If you run the webcam from another account that is not your login account, then you need to go into /etc/group and find the video group and add the ai account as a supplemental group.


Quick Art Test
OK. Now lets see if Torch is working right. There is a famous project that can take a picture and transfer the artistic style of a work of art onto your picture. Its really quite astonishing to see. Let's use that as our test for Torch.

The project page is here:

https://github.com/jcjohnson/neural-style


To download it:

$ git clone https://github.com/jcjohnson/neural-style.git


Now download the caffe models:

$ cd neural-style/models
$ sh ./download_models.sh
$ cd ..


We need a picture and a work of art. I have a picture of a circuit board:




Let's see if we can make art from it. The boxiness of the circuit kind of suggests cubism to me. There is a web site called wikiart that curates a collection of art by style and genre. Let's grab a cubist style painting and see how well it works.

$ wget https://uploads7.wikiart.org/images/albert-gleizes/portrait-de-jacques-nayral-1911.jpg
$ mv portrait-de-jacques-nayral-1911.jpg cubist.jpg


To render the art:

$ th neural_style.lua -backend cudnn -style_image cubist.jpg -content_image circuit.jpg -output_image art.jpg


Using a 1050Ti GPU, it takes about 4 minutes and this is the results:




One thing you have to pay attention to is that if the picture is too big, you will run out of GPU memory. The video card only has so much working memory. You can use any image editing tool to re-scale the picture. The number of pixels is what matters rather than the size of the file. Something in the 512 - 1080 pixel range usually fits in a 4Gb video card.


Conclusion
At some point we may come back to Torch to do some experimenting on security data. But I find it to be fun to play around with the art programs written for it. If you like this, look around. There are a number of apps written for Torch. The main point, though, is to show how to leverage the CUDA development environment we previously setup to get one of the main Deep Learning frameworks installed and running on a modern Fedora system.

Thursday, June 29, 2017

Setting up a CUDA development environment on Fedora 25

The aim of this blog is to explore Linux security topics using a data science approach to things. Many people don't like the idea of putting proprietary blobs of code on their nice open source system. But I am pragmatic about things and have to admit that Nvidia is the king of GPU right now. And GPU is the approach to accelerate Deep Learning for the last few years. So, today I'll go over what it takes to correctly setup a CUDA development environment for Fedora 25. This is a continuation of the earlier post about how to get an Nvidia GPU card setup in Fedora. That step is a prerequisite to this blog post.

CUDA
CUDA is the name that NVidia has given to a development environment for creating high performance GPU-accelerated applications. CUDA libraries enable acceleration across multiple domains such as linear algebra, image and video processing, deep learning and graph analytics.These libraries offload work normally done on a CPU to the GPU. And any program created by the CUDA toolkit  is tied to the Nvidia family of GPU's.


Setting it up
The first step is to go get the toolkit. This is not shipped by any distribution. You have to get it directly from Nvidia. You can find the toolkit here:

https://developer.nvidia.com/cuda-downloads

Below is a screenshot of the web site. All the dark boxes are the options that I selected. I like the local rpm option because that installs all CUDA rpms in a local repo that you can then install as you need.



Download it. Even though it says F23, it still works fine on F25.

The day I downloaded it, 8.0.44 was the current release. Today its different. So, I'll continue by using my version numbers and you'll have to make the appropriate substitutions. So, let's continue the setup as root...

rpm -ivh ~/Download/cuda-repo-fedora23-8-0-local-8.0.44-1.x86_64.rpm



This installs a local repo of cuda developer rpms. The repo is located in /var/cuda-repo-8-0-local/. You can list the directory to see all the rpms. Let's install the core libraries that are necessary for Deep Learning:

dnf install /var/cuda-repo-8-0-local/cuda-misc-headers-8-0-8.0.44-1.x86_64.rpm
dnf install /var/cuda-repo-8-0-local/cuda-core-8-0-8.0.44-1.x86_64.rpm
dnf install /var/cuda-repo-8-0-local/cuda-samples-8-0-8.0.44-1.x86_64.rpm


Next, we need to make sure that utilities provided such as the GPU software compiler, nvcc, are in our path and that the libraries can be found. The easiest way to do this by creating a bash profile file that gets included when you start a shell.

edit /etc/profile.d/cuda.sh (which is a new file you are creating now):

export PATH="/usr/local/cuda-8.0/bin${PATH:+:${PATH}}"
export LD_LIBRARY_PATH="/usr/local/cuda/lib64 ${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"
export EXTRA_NVCCFLAGS="-Xcompiler -std=c++03"


The reason CUDA is aimed at F23 rather than 25 is that NVidia is not testing against the newest gcc. So, they put something in the headers to make it fail.

I spoke with people from Nvidia at the GTC conference about why they don't support new gcc. Off the record they said they do extensive testing on everything they support and that its just not something they developed with when creating CUDA 8, but newer gcc will probably be support in CUDA 9.

Its easy enough to fix by altering one line in the header to test for the gcc version. Since we have gcc-6.3, we can fix the header to test for gcc 7 or later and then fail. To do this:

edit /usr/local/cuda-8.0/targets/x86_64-linux/include/host_config.h

On line 119 change from:

#if __GNUC__ > 5

to:

#if __GNUC__ > 6


This will allow things to compile with current gcc. There is one more thing that we need to fix in the headers so that Theano can compile GPU code later. The error looks like this:

math_functions.h(8901): error: cannot overload functions distinguished by return type alone

This is because gcc defines the function also and conflicts with the one NVidia ships. The solution as best I can tell is simply to:

edit /usr/local/cuda-8.0/targets/x86_64-linux/include/math_functions.h

and around lines 8897 and 8901 you will find:

/* GCC 6.1 uses ::isnan(double x) for isnan(double x) */
__DEVICE_FUNCTIONS_DECL__ __cudart_builtin__ int isnan(double x) throw();
__DEVICE_FUNCTIONS_DECL__ __cudart_builtin__ constexpr bool isnan(long double x);
__DEVICE_FUNCTIONS_DECL__ __cudart_builtin__ constexpr bool isinf(float x);
/* GCC 6.1 uses ::isinf(double x) for isinf(double x) */
__DEVICE_FUNCTIONS_DECL__ __cudart_builtin__ int isinf(double x) throw();

__DEVICE_FUNCTIONS_DECL__ __cudart_builtin__ constexpr bool isinf(long double x);

What I did is to comment out both lines that immediately follow the comment about gcc 6.1.

OK. Next we need to fix the cuda install paths just a bit. As root:

# cd /usr/local/
# ln -s /usr/local/cuda-8.0/targets/x86_64-linux/ cuda
# cd cuda
# ln -s /usr/local/cuda-8.0/targets/x86_64-linux/lib/ lib64



cuDNN setup
One of the goals of this blog is to explore Deep Learning. You will need the cuDNN libraries for that. So, let's put that in place while we are setting up the system. For some reason this is not shipped in an rpm and this leads to a manual installation that I don't like.

You'll need cuDNN version 5. Go to:

https://developer.nvidia.com/cudnn

To get this you have to have a membership in the Nvidia Developer Program. Its free to join.

Look for "Download cuDNN v5 (May 27, 2016), for CUDA 8.0". Get the Linux one. I moved it to /var/cuda-repo-8-0-local. Assuming you did, too...as root:

# cd /var/cuda-repo-8-0-local
# tar -xzvf cudnn-8.0-linux-x64-v5.0-ga.tgz
# cp cuda/include/cudnn.h /usr/local/cuda/include/
# cp cuda/lib64/libcudnn.so.5.0.5 /usr/local/cuda/lib
# cd /usr/local/cuda/lib
# ln -s /usr/local/cuda/lib/libcudnn.so.5.0.5 libcudnn.so.5
# ln -s /usr/local/cuda/lib/libcudnn.so.5.0.5 libcudnn.so



Testing it
To verify setup, we will make some sample program shipped with the toolkit. I had you to install them quite a few steps ago. The following instructions assume that you have used my recipe for a rpm build environment. As a normal user:

cd working/BUILD
mkdir cuda-samples
cd cuda-samples
cp -rp /usr/local/cuda-8.0/samples/* .
make


When its done (and hopefully its successful):

cd 1_Utilities/deviceQuery
./deviceQuery


You should get something like:

  CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "GeForce GTX 1050 Ti"
  CUDA Driver Version / Runtime Version          8.0 / 8.0
  CUDA Capability Major/Minor version number:    6.1
  Total amount of global memory:                 4038 MBytes (4234608640 bytes)
  ( 6) Multiprocessors, (128) CUDA Cores/MP:     768 CUDA Cores
  GPU Max Clock rate:                            1468 MHz (1.47 GHz)
  Memory Clock rate:                             3504 Mhz
  Memory Bus Width:                              128-bit
  L2 Cache Size:                                 1048576 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024


<snip>

 You can also check the device bandwidth as follows:

cd ../bandwidthTest
./bandwidthTest



You should see something like:

[CUDA Bandwidth Test] - Starting...
Running on...

 Device 0: GeForce GTX 1050 Ti
 Quick Mode

 Host to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            6354.8

 Device to Host Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            6421.6

 Device to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            94113.5

Result = PASS


At this point you are done. I will refer back to these instructions in the future. If you see anything wrong or needs updating, please comment on this article.

Wednesday, June 28, 2017

Updated Rstudio srpm available


Due to the unexpected update to R 3.4 on Fedora 25 which is incompatible with the version of RStudio that I wrote about in this blog, I have spent the time to create a new srpm with an updated RStudio which runs on the new R 3.4. The release notes are here:

https://www.rstudio.com/products/rstudio/release-notes/

If you had previously built the version I blogged about, that would correspond with the 0.99a release. So, you can see in the release notes what new things have been added since then.

The source  (updated 08/11/2017)
https://people.redhat.com/sgrubb/files/Rstudio/

Building
The build process is very similar to the original instructions. Please review them if you are new to building rpms. In essence you download the srpm. Then:

rpm -ivh R-studio-desktop-1.0.146-1.fc25.src.rpm
rpmbuild -bb working/R-studio-desktop/R-studio-desktop.spec

Then install. This assumes you followed the directory layout suggested in an earlier post.

RStudio picked up one new dependency for qt5-qtwebchannel-devel. You may need to install it first.

This version seems to work with R-3.4 and I've had some time to do limited testing. The only issue I see so far is that audit-explorer (which I'm yet to blog about) seems to have a bug that needs fixing.

One note about R upgrades...you have to re-install all of your packages. So, if you have upgraded R and RStudio, you'll need to start running the install.packages("") command in the console portion of RStudio prior to running any programs.

Tuesday, June 27, 2017

PSA: R3.4 upgrade

If you have built your own version of RStudio from my instructions and srpm, do not upgrade to R 3.4. If you do, you will see a message like this:


R graphics engine version 12 is not supported by this version of RStudio. The Plots tab will be disabled until a newer version of RStudio is installed.

At some point I need to create a newer build of RStudio to take care of this problem. But in the mean time you might want to put an exclude statement in /etc/yum.conf or /etc/dnf/dnf.conf to prevent "R" from updating.

Update June 29, 2017. You can upgrade to the new R 3.4 if you then update your RStudio package as I mention in my next blog post.

Monday, June 26, 2017

Using auparse in python

A while back we took a look at how to write a basic auparse program. The audit libraries have python bindings so that can let you write scripts that do things with audit events. Today, we will take a look at previously given example programs for "C" and see how to recreate them in python. I will avoid the lengthy discussion of the how's and why's from the original article, please refer back to it if explanation is needed.

Now in Python
I was going to publish this blog post about 2 weeks ago. In writing the code, I discovered that the python bindings for auparse had bugs and outright errors in them. These were all corrected in the last release, audit-2.7.7. I held up publishing this to give time for various distributions to get this update pushed out. The following code is not guaranteed to work unless you are on 2.7.7 or later.

We started the article off by showing the basic application construct to loop through all the logs. This is the equivalent of the first example:

#!/usr/bin/env python3

import sys
import auparse
import audit

aup = auparse.AuParser(auparse.AUSOURCE_LOGS);
aup.first_record()
while True:
    while True:
        while True:
            aup.get_field_name()
            if not aup.next_field(): break
        if not aup.next_record(): break
    if not aup.parse_next_event(): break
aup = None
sys.exit(0)


Just as stated in the original article...it's not too useful but it shows the basic structure of how to iterate through logs. We start by importing both audit libraries. Then we call the equivalent of auparse_init which is auparse.AuParser. The auparse state is caught in the variable aup. After that, all functions in auparse are called similarly to the C version except you do not need the auparse_ part of the function name. When done with the state variable, it is destroyed by setting it to None.

Now let's recreate example 2 which is a small program that loops through the logs and prints the record type and the field names contained in each record that follows:

#!/usr/bin/env python3

import sys
import auparse
import audit

aup = auparse.AuParser(auparse.AUSOURCE_LOGS);
aup.first_record()
while True:
    while True:
        mytype = aup.get_type_name()
        print("Record type: %s" % mytype, "- ", end='')
        while True:
            print("%s," % aup.get_field_name(), end='')
            if not aup.next_field(): break
        print("\b")
        if not aup.next_record(): break
    if not aup.parse_next_event(): break
aup = None
sys.exit(0)



I don't think there is anything new to mention here. Running it should give some output such as:

Record type: PROCTITLE - type,proctitle,
Record type: SYSCALL - type,arch,syscall,success,exit,a0,a1,a2,a3,items,ppid,pid,auid,uid,gid,euid,suid,fsuid,egid,sgid,fsgid,tty,ses,comm,exe,subj,key,
Record type: CWD - type,cwd,
Record type: PATH - type,item,name,inode,dev,mode,ouid,ogid,rdev,obj,nametype,
Record type: PROCTITLE - type,proctitle,
Record type: SYSCALL - type,arch,syscall,success,exit,a0,a1,a2,a3,items,ppid,pid,auid,uid,gid,euid,suid,fsuid,egid,sgid,fsgid,tty,ses,comm,exe,subj,key,


Now, let's take a quick look at how to use output from the auparse normalizer. I will not repeat the explanation of how auparse_normalize works. Please refer to the original article for a deeper explanation. The next program takes its input from stdin. So, run ausearch --raw and pipe that into the following program.


#!/usr/bin/env python3

import sys
import auparse
import audit

aup = auparse.AuParser(auparse.AUSOURCE_DESCRIPTOR, 0);
if not aup:
    print("Error initializing")
    sys.exit(1)

while aup.parse_next_event():
    print("---")
    mytype = aup.get_type_name()
    print("event: ", mytype)

    if aup.aup_normalize(auparse.NORM_OPT_NO_ATTRS):
        print("Error normalizing")
        continue

    try:
        evkind = aup.aup_normalize_get_event_kind()
    except RuntimeError:
        evkind = ""
    print("  event-kind:", evkind)

    if aup.aup_normalize_session():
        print("  session:", aup.interpret_field())

    if aup.aup_normalize_subject_primary():
        subj = aup.interpret_field()
        field = aup.get_field_name()
        if subj == "unset":
            subj = "system"
        print("  subject.primary:", field, "=", subj)

    if aup.aup_normalize_subject_secondary():
        subj = aup.interpret_field()
        field = aup.get_field_name()
        print("  subject.secondary:", field, "=", subj)

    try:
        action = aup.aup_normalize_get_action()
    except RuntimeError:
        action = ""
    print("  action:", action)

    if aup.aup_normalize_object_primary():
        field = aup.get_field_name()
        print("  object.primary:", field, "=", aup.interpret_field())

    if aup.aup_normalize_object_secondary():
        field = aup.get_field_name()
        print("  object.secondary:", field, "=", aup.interpret_field())

    try:
        str = aup.aup_normalize_object_kind()
    except RuntimeError:
       str = ""
    print("  object-kind:", str)

    try:
        how = aup.aup_normalize_how()
    except RuntimeError:
        how = ""
    print("  how:", how)

aup = None
sys.exit(0)



There is one thing about the function names that I wanted to point out. The auparse_normalizer functions are all prefixed with aup_. There were some unfortunate naming collisions that necessitated the change in names.

Another thing to notice is that the normalizer metadata functions can throw exceptions. They are always a RuntimeError whenever the function would have returned NULL as a C function. The above program also shows how to read a file from stdin which is descriptor 0. Below is some sample output:

ausearch --start today --raw | ./test3.py

---
event:  SYSCALL
  event-kind: audit-rule
  session: 4
  subject.primary: auid = sgrubb
  subject.secondary: uid = sgrubb
  action: opened-file
  object.primary: name = /etc/audit/auditd.conf
  object-kind: file
  how: /usr/sbin/ausearch
---
event:  SYSCALL
  event-kind: audit-rule
  session: 4
  subject.primary: auid = sgrubb
  subject.secondary: uid = sgrubb
  action: opened-file
  object.primary: name = /etc/audit/auditd.conf
  object-kind: file
  how: /usr/sbin/ausearch



Conclusion
The auparse python bindings can be used whenever you want to manipulate audit data via python. This might be preferable in some cases where you want to create a Jupyter notebook with some reports inside. Another possibility is that you can go straight to Keras, Theano, or TensorFlow in the same application. We will eventually cover machine learning and the audit logs. It'll take some time to get there because there are a lot of prerequisite setups that you would need to do.