Archive

Archive for the ‘Tips and tricks’ Category

How not to protect your app

20/10/2018 Leave a comment

So, I went to Droidcon this week.

And to be honest, it disappointed me almost in every parameter: from content to catering.

I don’t go to many conventions, but compared to August penguin that costs about one tenth for a ticket, Droidcon was surprisingly low quality.

The agenda had not one but two presentations on how to “protect your app from hackers”, and unfortunately, both could be summed up in one word:   obfuscate.

Obfuscation is the worst way to secure any software, and Android applications are no different in this regard. If anything, since smartphones today often contain more sensitive and valuable user data then PCs, it is even more important to use real security in mobile apps!

Obfuscation is bad because:

  • It is difficult and costly to implement but relatively easy to break.
  • It prevents, or at least significantly reduces your ability to debug your app and resolve user reported issues.
  • It only needs to be broken once to be broken everywhere, always for any user.
  • Fixing it once broken is almost impossible.

Basically, obfuscation is what security experts call “security by obscurity” which is considered very insecure.

Consider this: the entire Internet, including most banking and financial sites runs almost completely on open source software and standard, open, and documented protocols: Apache, NGINX, OpenSSL, Firefox, Chrome.

OpenSSL is particularly interesting, because it provides encryption that is good enough for the most sensitive information on the net, yet it is completely open.

Even the infamous “Dark Web” or more specifically the Tor network, is completely open source.

Truly secure software design does not rely on others not understanding what your code does.

Even malware writers, who’s entire bread and butter depends on hiding what their apps are doing, no longer rely on obfuscation but instead moved on to full blown code encryption and delivering code on demand from a server.

This is because obfuscated code was too easy for security researchers, and even fully automated malware scanners to detect.

If you want to learn more about this, follow security company blogs such as Semantek, Kaspersky and Checkpoint. Their researchers sometimes publish very interesting malware analysis showing the tricks malware writers use to hide their evil code.

Specifically bad advice

Now I would like to go over some specific advice given in one presentation, that I consider to be particularly bad, some of it to the point where I would call it “anti-advice”:

  1. Use reflection
  2. Hide things in “native”
  3. Hide data with protobufs or similar
  4. Hide code with ProGuard or similar

I listed these “recommendations” from most to least harmful.

If you are considering using any of these technics in your app, please read the following explanation before doing so, and reconsider.

1. Using reflection

Reflection is a powerful tool, but it was not designed to hide code.

If you use reflection to call a method in a class, anyone looking at a disassembled version of your app, or even running simple “strings” utility on it will still see all the method and class names.

You will gain nothing but loose any protection from bugs and crashes that compiler checks and lint tools normally provide.

You could go as far as scramble (or even fully encrypt) the strings holding the names of the classes and methods you call with reflection and only decipher them at run time.

But this takes long time to implement, is very error prone, and will make your app slower since it has a lot more work to do for a single, simple function call.

Stop and think carefully: what would a “hacker” analyzing your app gain from knowing what API you are calling?

The answer will always be: nothing, unless your app is very badly written!

Most security apps, like password managers, advertise exactly what encryption they are using so their customers would know how secure they are.

In a truly secure app, knowing what the app does, will not help a bad guy break it in any way.

I challenge anyone to give me an example in the comments of an API call that is worth hiding in a legitimate app.

2. Hide things in “native”

If you are not familiar with JNI and writing C and C++ code for Android, go read up on it, but not in order to protect your app!

Because Java (and Kotlin) still have performance issues when it comes to certain types of tasks (specifically, games and graphics related code), developers at Google created the Native Development Kit – NDK, to let you include native C and C++ code in your app that will run directly on the device processor and not the JVM (Dalvik / ART).

But just as with reflection, the NDK was not designed to hide your code from prying eyes.

And thus, it will not hide anything!

To many Java developers, particularly ones with no experiences or knowledge developing in native (to the hardware) languages like C,  a SO binary file will look like complete gibberish even in a hex editor.

But, just because you don’t understand what you are looking at, does not mean someone trying to crack your app will have a hard time understanding it!

A so file is a standard library used on Linux systems for decades (remember: Android is running on top of the Linux kernel), so there are lots of tools out there to decompile and analyze them.

But your attacker probably won’t even need to decompile your so file. If all they are looking for is some string, like a hard-coded password or API token you put in your app, they can still see it with a simple “strings” utility, same as they would in your Java or Kotlin code. There is no magic here – all strings remain intact when compiling to native.

Also, any external functions or methods your native code calls will appear in the binary file as plain text strings – your OS needs to find them to call them, so they will be there, exposed just like in any other language!

But things can get worse: lets say you have some valuable business logic in your app, and you want to make it harder for hackers to decompile your code and see this logic.

It is true that when you compile Java source, a lot more information about the original code is preserved than when compiling C or C++.

But don’t be tempted by this, because you may leave your self even more exposed!

If a hacker really wants to run your code for his (or hers) nefarious purposes, then wrapping it up in a library that can easily be called from any app is like gift wrapping it for them.

And this is exactly what you will be doing if you put your code in a native SO file: you are putting it in an easy to use library!

Instead of decompiling your code and rewriting it in their app, the hacker can just take your SO file, and call its functions (that must be exposed to work!) from their own app.

They don’t need to know what your code does, they can just feed it parameters and get the result, which is what they really wanted in the first place.

So instead of hampering hacking, you make it easier by using the wrong tool, all the while giving your self a false sense of security.

4. Hide data with protobufs or similar

By now you should have noticed a pattern: one thing all these bad advice have in common is recommending the wrong tool for the job.

Protobufs is an excellent open source tool for data serialization.

It is not a security tool!

The actual advice given in the presentation was to replace JSON in server responses with protobufs in order to make the information sent by the server less readable.

But what security do you gain from this? If your server sends a reply like this:

{
"first_name" : "Jhon",
"last_name" : "Smith",
"phone" : "555-12345",
"email" : "jhon@email.com"
}

converting this structure to a protobuf will look something like this:
xxxxJhonxxxxSmithxxxx555-12345xxxjhon@email.com
Is that really hiding anything?

Protobufs are more compact then JSON, and they can be deserialized faster and easier than JSON but they also have some disadvantages: they are not as flexible as JSON.

It is hard to support optional fields with protobuffs and even harder to create dynamic or self describing objects.

If your app needs flexibility in parsing server replies, or if you have other clients, particularly web clients written with JS that access the same server API, JSON may be the better choice for you.

When deciding whether to use JSON or protobuffs, consider their advantages and disadvantages for your use case, DO NOT CONSIDER SECURITY!

They are both equally insecure, and you will need encryption (always use SSL!) and proper access validation (passwords, tokens, client certificates) if you want to keep your data safe.

4. Hide code with ProGuard or similar

This advice actually talks about the right tool for a change: ProGuard.

This is a tool Google ships with Android Studio, and it does two things: reduces the size of your code and resources after compilation, and slightly obfuscates your code.

This is not a bad tool, but it comes with a cost, and it won’t really give you protection from hackers.

It will rename your methods like getMySecretPassword() to a() but will that really stop anyone from doing anything bad?

At best, it will slow them down, but keep in mind that it will also cost you:

ProGuard has the side effect of rendering all stack traces useless and making debugging the app extremely difficult.

There is a way to mitigate this: you need to keep a special translation file for every single build of your app (because ProGuard randomizes its name mangling).

If you need to support users in production and don’t want to be helpless or work extra hard when they report a crash, you might want to give up on ProGuard.

Also keep in mind that you need to carefully tell ProGuard what not to obfuscate, since you must keep any external API calls, components declared in the manifest and some third party library calls intact, or your app will not run.

Remember – ProGuard will:

  • Not keep any hardcoded strings safe.
  • Not keep your user password safe if you store it as plain text in your app data folder.
  • Not keep your communication safe if you do not use SSL.
  • Not protect you from MITM attacks if you do not use certificate pinning.

ProGuard might make your final APK file smaller by getting rid of unused code and reducing length of class and methods names, but you should carefully consider the cost of this reduction you will pay when dealing with bugs and crashes.

I find it is usually just not worth the hassle.

And there are better tools now for reducing download size, such as App Bundles.

Summary

Messing with your code will never make your app more secure. It will not protect you from hackers.

Even if you do not want, or can not, release your app as open source, you still need to remember that trying to hide its code with obfuscation will cost you more then having your app reverse engineered.

The development, debugging, and user support costs can be as devastating as any hacks!

But, if you treat your code as though it is meant to be open, and make sure that even if a bad person can read and understand everything your app does they still can not get your users data or exploit your web server, then, and only then, will your app be truly secure.

And doing that is often easier and cheaper than trying to obfuscate your code or data.

P. S.

One of the presentations mentioned a phenomenon I was not familiar with: “App cloning”.

Apparently, if you publish an ad supported app, some bad people can take your app without your permission, replace your advertisement API keys with their own, and release the app to some unofficial app stores like the ones that are common in China (because Google Play is blocked there by the government).

This way, they will get ad revenue from your app instead of you.

But consider this: would you publish your app to these stores?

If your answer is “no”, then you are not losing anything!

You will never get any money from these users because they will never be able to install your original app, so any effort you put in defending against “cloning” will be a net financial loss to you.

Remember – as a developer, your time is money!

P. S. 2

Someone in the audience asked about Google API keys like the Google Maps API key.

Usually, it is bad practice to put API keys in plain text in the manifest of your app, because anyone can get them from there and use a paid API at your expense.

But this is not the case with Google API keys!

The reason Google tells you to put the key in the manifest, is because Google designed these API keys in such a way, that they will be useless to anyone but you, so stealing them is pointless.

This is a great example of a good security design: instead of relying on app developers to figure out how to distribute an API key to millions of users but keep it safe from hackers, Google tide the key to your signing certificate and your app id (package name).

When you create the API key, you must enter your certificate fingerprint and your package name.

Your private key – the one you use to sign your apps for release, is something most developers already keep very safe. There is never a reason to send it anywhere and it would never be included in the app itself.

It will stay safe on the developers computer.

And without this private key, the public API key will not work.

If it is used in an app signed by anyone else, even if that app fakes your app’s id, the API key will still be invalid.

This is how you secure apps!

Beware Java’s half baked generics

13/10/2016 Leave a comment

Usually I don’t badmouth Java. I think its a very good programming language.

In fact, I tend to defend it in arguments on various forums.

Sure, it lacks features compared to some other languages, but then again throwing everything including a kitchen sink in to a language is not necessarily a good idea. Just look at how easy it is to get a horrible mess of code in C++ with single operator doing different things depending on context. Is &some_var trying to get address of a variable or a reference? And what does &&some_var do? It has nothing to do with the boolean AND operator!

So here we have a simple language friendly to new developers, which is good because there are lots of those using it on the popular Android platform.

Unfortunately, even the best languages have some implementation detail that will make you want to lynch their creators or just reap out your hair, depending on whether you externalize your violent tendencies or not.

Here is a short code example that demonstrates a bug that for about 5 minutes made me think I was high on something:

HashMap<Integer, String> map = new HashMap<>();

byte a = 42;
int b = a;

map.put(b, "The answer!");

if (map.containsKey(a))
	System.out.println("The answer is: " + map.get(a));
else
	System.out.println("What was the question?");

What do you expect this code to print?

Will it even compile?

Apparently it will, but the result will surprise anyone who is not well familiar with Java’s generic types.

Yes folks – the key will not be found and the message What was the question? will be printed.

Here is why:

The generic types in Java are not fully parameterized. Unlike a proper C++ template, some methods of generic containers take parameters of type Object, instead of the type the container instantiation was defined with.

For HashMap, even though it’s add is properly parameterized and will raise a compiler error if the wrong type key is used, the get and containsKey methods take a parameter of type Object and will not even throw a runtime exception if the wrong type is provided. They will simply return null or false respectively as if the key was simply not there.

The other part of the problem is that primitive types such as byte and int are second class citizens in Java. They are not objects like everything else and can not be used to parameterize generics.

They do have object equivalents named Byte and Integer but those don’t have proper operator overloading so are not convenient for all use cases.

Thus in the code sample above the variable a gets autoboxed to Byte, which as far as Java is concerned a completely different type that has nothing to do with Integer and therefore there is no way to search for Byte keys in Integer map.

A language that implements proper generics would have parameterized these methods so either a compilation error occurred or an implicit cast was made.

In Java, it is up to you as a programmer to keep you key type straight even between seemingly compatible types like various size integers.

In my case I was working with a binary protocol received from external device and the function filling up the map was not the same one reading from it, so it was not straight forward to align types everywhere. But in the end I did it and learned my lesson.

Maybe this long rant will help you too. At least until a version of Java gets this part right…

Android, Busybox and the GNU project

12/11/2012 2 comments

Richard Stallman, the father of the Free Software movement and the GNU project, always insists that people refer to some Linux based operating systems as “GNU/Linux”. This point is so important to him, he will refuse to grant an interview to anyone not willing to use the correct term.

There are people who don’t like this attitude. Some have even tried to “scientifically prove” that GNU project code comprises such a small part of a modern Linux distribution that it does not deserved to be mentioned in the name of such distributions.

 

Personally, I used to think that the GNU project deserved recognition for it’s crucial historical role in building freedom respecting operating systems, even if it was only a small part of a modern system.

But a recent experience proved to me that it is not about the amount of code lines or number of packages. And it is not a historical issue. There really is a huge distinction between Linux and GNU/Linux, but to notice it you have to work with a different kind of Linux. One that is not only stripped of GNU components, but of its approach to system design and user interface.

Say hello to Android. Or should I say Android/Linux…

 

Many people forget, it seems, that Linux is just a kernel. And as such, it is invisible to all users, advanced and novice alike. To interact with it, you need an interface, be it a text based shell or a graphical desktop.

So what happens when someone slaps a completely different user-space with a completely different set of interfaces on top of the Linux kernel?

 

Here is the story that prompted me to write this half rant half tip post:

My boss wanted to backup his personal data on his Android phone. This sounds like it should be simple enough to do, but the reality is quite the opposite.

In the Android security model, every application is isolated by having its own user (they are created sequentially and have names like app_123).

An application is given its own folder in the devices data partition where it is supposed to store its data such as configuration, user progress (for games) etc.

No application can access the folder of another application and read its data.

 

This makes sense from the security perspective, except for one major flaw: no 3rd party backup utility can ever be made. And there is no backup utility provided as part of the system.

Some device makers provide their own backup utilities, and starting with Android 4.0 there is a way to perform a backup through ADB (which is part of Android SDK), but this method is not designed for the average user and has several issues.

 

There is one way, an application on the device can create a proper backup: by gaining root privileges.

But Android is so “secure” it has no mechanism to allow the user to grant such privileges to an application, no matter how much he wants or needs to.

The solution of course, is to change the OS to add the needed capability, but how?

Usually, the owner of a stock Android device would look for a tool that exploits a security flaw in the system to gain root privileges. Some devices can be officially unlocked so a modified version of Android can be installed on them with root access already open.

 

The phone my boss has is somewhat unusual: it has a version of the OS designed for development and testing, so it has root but the applications on it do not have root.

What this confusing statement means is, that the ADB daemon is running with root privileges on the device allowing you to get a root shell on the phone from the PC and even remount the system partition as writable.

But, there is still no way for an application running on the device to gain root privileges, so when my boss tried to use Titanium Backup, he got a message that his device is not “rooted” and therefore the application will not work.

 

Like other “root” applications for Android, Titanium Backup needs the su binary to function. But stock Android does not have a su binary. In fact, it does not even have the cp command. Thats right – you can get a shell interface on Android that might look a little bit like the “regular Linux”, but if you want to copy a file you have to use cat.

This is something you will not see on a GNU/Linux OS, not even other Linux based OSs designed for phones such as Maemo or SHR.

 

Google wanted to avoid any GPL covered code in the user-space (i.e. anywhere they could get away with it), so not only did they not use a “real” shell (such as BASH) they didn’t even use Busybox which is the usual shell replacement in small and embedded systems. Instead, they created their own very limited (or as I call it neutered) version called “Toolbox”.

 

Fortunately, a lot of work has been done to remedy this, so it is not hard to find a Busybox binary ready made to run on Android powered ARM based device.

The trick is installing it. Instructions vary slightly from site to site, but I believe the following will work in most cases:

adb remount
adb push busybox /system/bin
adb shell chmod 6755 /system/bin/busybox
adb shell busybox --install /system/bin

Note that your ADB must run as root on the device side!

The important part to notice here is line 3: you must set gid and uid bits on the busybox binary if you want it to function properly as su.

And no – I didn’t write the permissions parameter to chmod as digits to make my self look like a “1337 hax0r”. Android’s version of chmod does not accept letter parameters for permissions.

 

After doing the steps above I had a working busybox and a proper command shell on the phone, but the backup application still could not get root. When I installed a virtual terminal application on the phone and tried to run su manually I got the weirdest error: unknow user: root

How could this be? ls -l clearly showed files belonging to ‘root’ user. As GNU/Linux user I was used to more descriptive and helpful error messages.

I tried running ‘whoami’ from the ADB root shell, and got a similarly cryptic message: unknown uid 0

Clearly there was a root user with the proper UID 0 on the system, but busybox could not recognize it.

 

Googling showed that I was not the only one encountering this problem, but no solution was in sight. Some advised to reinstall busybox, others suggested playing with permissions.

Finally, something clicked: on a normal GNU/Linux system there is a file called passwd in etc folder. This file lists all the users on the system and some information for each user such as their home folder and login shell.

But Android does not use this file, and so it does not exist by default.

 

Yet another difference.

So I did the following:

adb shell
# echo 'root::0:0:root:/root:/system/sh' >/etc/passwd

This worked like a charm and finally solved the su problem for the backup application. My boss could finally backup and restore all his data on his own, directly on the phone and without any special trickery.

 

Some explanation of the “magic” line:

In the passwd file each line represents a single user, and has several ‘fields’ separated by colons (:). You can read in detail about it here.

I copied the line for the root user from my PC, with some slight changes:

The second field is the password field. I left it blank so the su command will not prompt for password.

This is a horrible practice in terms of security, but on Android there is no other choice, since applications attempting to use the su command do not prompt for password.

There are applications called SuperUser and SuperSU that try to ask user permission before granting root privileges, but they require a special version of the su binary which I was unable to install.

 

The last field is the “login shell” which on Android is /system/sh

The su binary must be able to start a shell for the application to execute its commands.

Note, this is actually a symlink to the /system/mksh binary, and you may want to redirect it to busybox.

 

So this is my story of making one Android/Linux device a little more GNU/Linux device.

I took me a lot of time, trial and error and of course googling to get this done, and reminded me again that the saying “Linux is Linux” has its limits and that we should not take the GNU for granted.

It is an important part of the OS I use both at home and at work, not only in terms of components but also in terms of structure and behavior.

 

And it deserves to be part of the OS classification, if for no other reason than to distinguish the truly different kinds of Linux that are out there.

 

Get XML element value in Python using minidom

29/07/2011 Leave a comment

Finally, a “development” post for my “developer” blog.

Recently, I’ve been working on some XML processing programs in Python.

The minidom module is great if you want your XML in a tree, and want tag names and attributes easily accessible, but, what happens if you want the text content inside a tag?

DOM, does not have a “tag value” concept. Instead, every bit of text in the XML, including the indentation is a “text node”, which is parsed as a separate tree element.

That means, that if you have something like this:


<name>John Smith</name>

You will get a tree with two levels: top level for “name” element, for which nodeValue will be None. This element will have a child node (second level of the tree) which will be of type TEXT_NODE an it’s values will be the text “John Smith”.

So far, so good, but, what if the value we want has some XML markup of its own?


<text>This text has <b>bold</b> and <i>italic</i> words.</text>

Now we have a complex tree on our hands with 3 levels and multiple branches.

It will look something like this:

<text>
   |______
          |-"This text has
          |-<b>
          |  |_________
          |            -"bold"
          |-"and"
          |-<i>
          |  |_________
          |            -"italic"
          --"words."

As you can see, this is a big mess, with the text split in to multiple parts on two separate tree levels.

There is no facility in minidom, to get the value of our <text> tag directly.

There is however, a way around it, that is simple but not obvious: you need to “flatten” the desired tag in to an XML string, then strip the tag it self from the string and you will have a clean value.

Here is the code:

def get_tag_value(node):
    """retrieves value of given XML node
    parameter:
    node - node object containing the tag element produced by minidom

    return:
    content of the tag element as string
    """

    xml_str = node.toxml() # flattens the element to string

    # cut off the base tag to get clean content:
    start = xml_str.find('>')
    if start == -1:
        return ''
    end = xml_str.rfind('<')
    if end < start:
        return ''

    return xml_str[start + 1:end]

Just pass the node you want the value of to the function and it will give you back the value as a string, including any internal markup.

I place this code in the public domain, which means you can use it anywhere any way you want with no strings attached.