Tuesday, March 26, 2019

Prevent and Mitigate Stack Overflow

When programming micro-controllers you usually have to be more careful with your memory usage and also for smaller devices you often don't get any help from the hardware.

On ARM micro-controllers the memory layout is such that the stack is placed at the higher address range of the memory and writes 'downwards' towards the lower memory addresses. While the heap fills up memory from the lower to the higher addresses. In cases of high use of static/program memory use, heap use and stack use, the stack can overflow and overwrite heap memory. This may result in weird and undefined behavior. This is bad.

Some methods that can be used to prevent this:
  • Use the HW memory protection unit: HW unit that causes an interrupt or fault if stack is written passed a defined address.
  • Canary: Write a unique value to address designated as top of stack and check that this value is still written to top of stack address in main loop.
  • Add stack protection code: GCC can add functionality to check wether writes to stack would overflow. Results in larger program binary and higher overall stack usage.
  • Stack depth analysis: Compile time or static analysis of code to determine how much stack each function will use and combinations that would result in stack overflow. Some tools:
    • GCC -fstack-usage, doesn't work in combination with -lto (link time optimization).
    • ARM compiler, with Keil microvision (uvision), but gives only one number for each interrupt level.
  • Water mark: Fill stack with pattern for each main loop to, check stack usage after each automatic test. Create tests that you think will cause the most high stack usage functions to execute (preferably simultaneously/interrupted).

Well known case of this: https://developers.slashdot.org/story/14/02/21/2349204/stack-overflow-could-explain-toyota-vehicles-unintended-acceleration

Monday, April 18, 2016

Kerberos with groups on Tomcat 7

To provide Single-SignOn (SSO) to your web service on a linux server in a windows environment (Active Directory) you can add a login filter to tomcat to accept Kerberos tokens. It's really quite simple, if you set it up correctly, which is not simple to figure out.

Ack: Compiled from http://portlandlanguagecraft.com/ and https://pixabay.com/en/chain-gold-power-connection-rights-307886/

I did it using a custom SPNEGO filter to also extract AD groups from the kerberos tokens.

Mostly follow https://tomcat.apache.org/tomcat-7.0-doc/windows-auth-howto.html but with many additional tweaks.

You will need to create/edit the following files (On Ubuntu):


(These files are also found in /var/lib/tomcat7/conf).

Step-by-step (First section just for login and second section for getting groups too):


Download the Spnego HTTP filter:
Available from https://sourceforge.net/projects/spnego/. Put the file under /usr/share/tomcat7/libs/spnego-r7.jar to make tomcat load it on startup.

On the Active Directory (AD) / Kerberos Key Distribution Center (KDC) / Windows server:

Add service user:
Add a service user to let you linux server "log in" and validate kerberos tokens.

Link SPN to service user:
SPN (Service Principal Names) are identifiers for users or hosts. We need to add the ones representing our server. NOTE: The SPN is case sensitive and you must use the same case everywhere. The command is on the form:

So on your windows/AD server enter the following

setspn.exe -A HTTP/myserver DOMAIN\myserviceuser
setspn.exe -A HTTP/myserver.domain.local DOMAIN\myserviceuser

to link the SPN HTTP/myserver to the user myserviceuser.

Generate keytab:
The keytab is file which stores SPNs/usernames and password for them.

According to the apache tutorial you should do this on your windos/AD server using the ktpass tool. I found it was better to use the ktab.exe that comes with java on windows:
ktab -a HTTP/myserver.domain.local myservicepassword -k mykeytab.keytab -n 0

Note the n 0 flag which sets key version number, it needs to be 0 for tomcat/Spnego to find the key.

Put the mykeytab.keytab file under /etc/tomcat7/mykeytab.keytab on your linux server.

We also need to make the computers in the windows network trust our server, to do this we can use Group Policy on the AD server. But we'll get back to that later.

On the linux server (as sudo):

Add a filter to /etc/tomcat7/web.xml:

<!-- ======================== SPNEGO filter ==============================-->


You need to replace SERVICE_USER_NAME and SERVICE_USER_PASSWORD with the ones you use to create your keytab. spnego.allow.basic, spnego.prompt.ntlm are true to let users who haven't logged into windows to log in (WARNING: username and password are sent in cleartext/base64 to the linux server!). spnego.allow.unsecure.basic needs to be true if you don't use https, which you should do.


To apply filter to all files.

Create/edit the login.conf:

com.sun.security.jgss.krb5.initiate {
    com.sun.security.auth.module.Krb5LoginModule required;

com.sun.security.jgss.krb5.accept {
    com.sun.security.auth.module.Krb5LoginModule required

Differently from the apache tutorial, the initate object should only contain the module, otherwise tomcat will throw a parse error on startup.

Create/edit the krb5.conf:

default_realm = DOMAIN.LOCAL
default_keytab_name = FILE:/etc/tomcat7/mykeytab.keytab
default_tkt_enctypes = rc4-hmac,aes256-cts-hmac-sha1-96,aes128-cts-hmac-sha1-96
default_tgs_enctypes = rc4-hmac,aes256-cts-hmac-sha1-96,aes128-cts-hmac-sha1-96
forwardable = true

        kdc =

domain.local = DOMAIN.LOCAL
.domain.local = DOMAIN.LOCAL

        krb4_convert = true
        krb4_get_tickets = false

The kdc parameter should support the hostname of the KDC/AD server e.g. kdc.domain.local, but mine had trouble with DNS lookup for it, luckily IP works fine.

NOTE: Make sure time is within a few minutes of the AD server, consider installing a NTP client to keep in sync.

AUTHORIZATION (roles/groups):

You will need to create/edit the following files (On Ubuntu):


Active Directory adds a blob to their kerberos tokens called PAC (Privilege Attribute Certificate), which includes a users roles. We can extract these roles from our ticket so we don't have to do an additional LDAP request (which is the normal way).

To do this we need a custom build of the spnego library by Ricardo Martín Camarero (rickyepoderi) (see http://blogs.nologin.es/rickyepoderi/index.php?/archives/73-SPNEGOKerberos-in-JavaEE-PAC.html) which utilizes JaasLounge and Bouncy Castle ASN1 to extract the PAC roles.
I've added support for fetching a users kerberos token when using Basic Auth as well as adding compressed PAC from another library.

The spnego-pac source code is available from github (https://github.com/asmund1/spnego-pac), and the final binaries used in this project from https://github.com/asmund1/spnego-pac/blob/master/jars/spnego-pac.jar and https://github.com/asmund1/spnego-pac/blob/master/jars/bcprov-jdk15on-147.jar (additional library needed).

On the linux server (as sudo):

Copy spnego-pac.jar and bcprov-jdk15on-147.jar to /usr/share/tomcat7/libs/ so that tomcat loads it on startup. NOTE: Remove the original spnego jar if you have it there already.

The PAC contains only the numerical representation for each role for the user, you can use this directly in your servlets, but I added some aliases for my roles. I did this in the webapp web.xml, but it should work in the global web.xml too (/etc/tomcat7/web.xml):

        <description>Alias for write access role</description>

To get the value for your roles, check your tomcat log (/var/lib/tomcat7/logs/catalina.out) after login using kerberos, the library prints the SIDs found for a user.
You might have to change the log level since they are printed at FINER level. Do this by appending
net.sourceforge.spnego.SpnegoAuthenticator = FINER
to the bottom of /etc/tomcat/logging.properties file and restarting tomcat.


The username/SPN of the logged in user and his/her roles are added to the request object, so to fetch them in Java servlets / JSP use the following lines of code:

For Java Servlet:
import javax.servlet.http.HttpServletRequest;
import javax.servlet.ServletContext;

public class MyServlet extends HttpServlet {

    public void doPost(HttpServletRequest req, HttpServletResponse resp) throws IOException, ServletException {

        // Fetches username of logged in user

         // Check if user has write access
        ServletContext context = req.getServletContext();
        if (!req.isUserInRole(context.getInitParameter("myserver_write_role"))) {

For JSP:
<%= request.getRemoteUser() %>
<%= request.isUserInRole(request.getServletContext().getInitParameter("myserver_write_role")) %>


To get the kerberos token from windows you need to be on the trusted server list. Chrome and IE use a common list while firefox and other browsers have their own.

To add you server as trusted in IE (and Chrome) open Internet Options -> Security -> Local intranet -> Sites button -> Advanced button
Enter the url for your server and press the Add button.

To do the same for firefox do the following:
1. Open Firefox, and type "about:config" in the Address Bar.
2. In the Search field, type "negotiate".
3. Set the following fields:
      network.negotiate-auth.trusted-uris  myserver.domain.local
      network.negotiate-auth.delegation-uris myserver.domain.local


To add  your server for IE and Chrome for all windows machines in the intranet, you can use Group Policy: https://www.serverknowledge.net/group-policy/adding-trusted-sites-internet-explorer-using-group-policy-gpo/


On the linux server (as sudo):
Start/Restart tomcat server to load changes in config and load jars:
service tomcat7 restart

Then tail the log output for any errors:
tail -f /var/lib/tomcat7/logs/catalina.out

Then navigate to your server in IE/Chrome. You should not be prompted for username or password. If you go to the server from a non-windows logged on computer (e.g. mobile phone) you should get a popup asking you to enter username and password. If you enter the incorrect password you should get a white page, with correct credentials you should see your content.






Friday, April 15, 2016

Share one file

Frequently I have two computers next to each other and I want to copy a file or directory from one to the other. What do you do? Mail it to yourself, sync it via some cloud service, set up samba? No just use woof (http://www.home.unix-ag.org/simon/woof.html).


A tiny python server that serves one file and then shuts down as soon as the file has been downloaded. (You can also share a directory as a tar archive).

Just do:
python woof.py 'file to share'

Enter the resulting URL e.g.:
in a web browser on the other machine
and voila.

NOTE: the computers need to be on the same network. + you need to be running linux/unix with python 2.7.x.

Wednesday, April 6, 2016

Framebuffer screen dumps to PNG and back

I've got a nice card with an Atmel AVR32 processor and a nice big 320x240 (monochrome!) screen. :)
I can put data on the screen from buildroot linux by writing directly to the framebuffer, /dev/fb0. However the data there is in raw format. Here is how to dump from screen. Then convert, using ffmpeg, the raw data to a PNG and then back to raw data after editing:

  1. Get a screenshot of the framebuffer using cat /dev/fb0 > screendump.bin.
  2. Convert to PNG using ffmpeg: ffmpeg   -vcodec rawvideo   -f rawvideo   -pix_fmt monow   -s 320x240   -i screendump.bin     -f image2   -vcodec png screendump.png
    This will read raw video from the screendump.bin, handle it as pixel monochrome data from a 320x240 screen and save as sreendump PNG file.
  3. Next, open in Gimp or similar to edit. Save as PNG afterwards.
  4. Convert to rawvideo binary again using ffmpeg: ffmpeg -vcodec png -i savedimage.png -vcodec rawvideo -f rawvideo -pix_fmt monow savedimage.bin
  5. Put the data on the screen: cat savedimage.bin > /dev/fb0
Use ffmpeg -pix_fmts to see other raw video types to convert from/to.

Sources: Check out these great sites for more:

Remote Desktop over the Internet using UltraVNC on Windows

Here is a HowTo for setting up a UltraVNC server on your computer, automatically open needed ports, and creating a configured UltraVNC single-click client to let you see someones desktop where all they have to do is start the client you send to them. No installations!
(A great alternative is TeamViewer which is free for personal use, but expensive for commercial use)

Here is how:

- Download UltraVNC and install:
See bottom of: http://www.uvnc.com/downloads/ultravnc.html

- Download UltraVNC SingleClick custom.zip package from:

- Extract helpdesk.txt file.

- Edit to look like you want to.
Example helpdesk.txt:

UltraVnc SC

My RDC connection
-connect -noregistry

Double click to open connection






UltraVNC web



Establishing connection ...

5 min connection attempt

If it fails, the software will close.


Connection active.

Warning, your desktop is visible remotely

You can break the connection any time

by using the close button

(remember to replace with your external IP, e.g. from https://www.whatismyip.com/)

- Create a ZIP file with only helpdesk.txt in it.

- Go to:

- Enter user: foo
- Enter password: foobar
- Select Zip file containing helpdesk.txt
- Click Upload
- Wait until you a link to download exe file appears. Then download the exe file.
- Start it to see what it looks like.

- Download UPnP PortMapper to open the needed port on your computer (if you havent done so manually on your router).

- Open PortMapper jar file (should need only double-click if you have Java installed).
- Under the section "Port Mapping presets", select Create button.
- Enter a description, e.g. UltraVNC.
- Then click Add button.
- Change port 1 to 5900 in both External Port and Internal Port columns.
- Not sure if you also need UDP port open, but might be useful so click Add again and select UDP with port 5900 internal and external.
- Click save.

- Click connect to connect to your router.
When successfully connected (you might get an error, but stuff seems to work regardless).
- Select your UltraVNC entry in the list and press "Use".
- Wait for ports to show up in the list above.

Now you have ports open. You need to start VNC listener:
- Open command line prompt (cmd in windows start menu) and navigate to where you installed UltraVNC and run it with flag -listen 5900 to tell it to listen on port 5900:

cd "C:\Program Files\uvnc bvba\UltraVnc"
vncviewer.exe -listen 5900

- Finally, send the exe file you generated from helpdesk.txt (not the zip file) to the person you want to connect to. Get the person to open the file and double click on the top option and you should get a message on your computer asking you to accept the connection.


Sites to read about AngularJS and Ionic

Top 10 angularjs mistakes:

Best practice dir structure angularjs/ionic:
I've tried both. Unsure which one I think is best. I think the most important point is to create a directory structure to group your files according to how they are used. So that other people can understand how you program is structured more quickly.

Tips for ionic:
http://www.betsmartmedia.com/what-i-learned-building-an-app-with-ionic-framework (https://web.archive.org/web/20150925034548/http://www.betsmartmedia.com/what-i-learned-building-an-app-with-ionic-framework)

Intro and how to use jasmine unit test framework:
Great test tool!

Promises in AngularJS


Introduction to AngularJS promises (as a cartoon):

Promise anti-patterns
Flatten chaining, clean up, and broken chains:

HTTP promise not like Q promise, use deferred:

AngularJS Q: