Javantura v4 - Security architecture of the Java platform - Martin Toshev

1. Security architecture
of the Java platform
Martin Toshev
@martin_fmi

2. Who am I
Software consultant (CoffeeCupConsulting)
BG JUG board member (http://jug.bg)
OpenJDK and Oracle RBDMS enthusiast
Twitter: @martin_fmi

4. Agenda
• Evolution of the Java security model
• Outside the sandbox: APIs for secure coding
• Designing and coding with security in mind

5. Evolution of the Java security model

6. Evolution of the
Java security model
• Traditionally - companies protect they assets using strict physical
and network access policies
• Tools such as anti-virus software, firewalls, IPS/IDS systems
facilitate this approach

7. Evolution of the
Java security model
• With the introduction of various technologies for loading and
executing code on the client machine from the browser (such as
Applets) - a new range of concerns emerge related to client
security – this is when the Java security sandbox starts to evolve
…

8. Evolution of the
Java security model
• The goal of the Java security sandbox is to allow untrusted code
from applets to be executed in a trusted environment such as the
user's browser

9. Evolution of the
Java security model
• JDK 1.0 (when it all started …) – the original sandbox model was
introduced
Applet
(untrusted)
System code
(trusted)
JVM
Browser
http://javantura.com/demoapplet

10. Evolution of the
Java security model
• Code executed by the JVM is divided in two domains – trusted
and untrusted
• Strict restriction are applied by default on the security model of
applets such as denial to read/write data from disk, connect to
the network and so on

11. Evolution of the
Java security model
• JDK 1.1 (gaining trust …) – applet signing introduced
Applet
(untrusted)
System code
(trusted)
JVM
Browser
Signed Applet
(trusted)
http://javantura.com/demoapplet
http://javantura.com/trustedapplet

12. Evolution of the
Java security model
• Local code (as in JDK 1.0) and signed applet code (as of JDK 1.1)
are trusted
• Unsigned remote code (as in JDK 1.0) is not trusted

13. Evolution of the
Java security model
• Steps needed to sign and run an applet:
• Compile the applet
• Create a JAR file for the applet
• Generate a pair of public/private keys
• Sign the applet JAR with the private key
• Export a certificate for the public key
• Import the Certificate as a Trusted Certificate
• Create the policy file
• Load and run the applet

14. Evolution of the
Java security model
• JDK 1.2 (gaining more trust …) – fine-grained access control
Applet
System code
JVM
Browser
grant codeBase http://javantura.com/demoapplet {
permission java.io.FilePermisions “C:Windows” “delete”
}
security.policy
SecurityManager.checkPermission(…)
AccessController.checkPermission(…)
http://javantura.com/demoapplet

15. Evolution of the
Java security model
• The security model becomes code-centric
• Additional access control decisions are specified in a security
policy
• No more notion of trusted and untrusted code

16. Evolution of the
Java security model
• The notion of protection domain introduced – determined by the
security policy
• Two types of protection domains – system and application

17. Evolution of the
Java security model
• The protection domain is set during classloading and contains the
code source and the list of permissions for the class
applet.getClass().getProtectionDomain();

18. Evolution of the
Java security model
• One permission can imply another permission
java.io.FilePermissions “C:Windows” “delete”
implies
java.io.FilePermissions “C:Windowssystem32” “delete”

19. Evolution of the
Java security model
• One code source can imply another code source
codeBase http://javantura.com/
implies
codeBase http://javantura.com/demoapplet

20. Evolution of the
Java security model
• Since an execution thread may pass through classes loaded by
different classloaders (and hence – have different protection
domains) the following rule of thumb applies:
The permission set of an execution thread is considered to be the
intersection of the permissions of all protection domains traversed by the
execution thread

21. Evolution of the
Java security model
• JDK 1.3, 1,4 (what about entities running the code … ?) – JAAS
Applet
System code
JVM
Browser
http://javantura.com/demoapplet
grant principal javax.security.auth.x500.X500Principal "cn=Tom"
{ permission java.io.FilePermissions “C:Windows” “delete” }
security.policy

22. Evolution of the
Java security model
• JAAS (Java Authentication and Authorization Service) extends the
security model with role-based permissions
• The protection domain of a class now may contain not only the
code source and the permissions but a list of principals

23. Evolution of the
Java security model
• The authentication component of JAAS is independent of the
security sandbox in Java and hence is typically used in more wider
context (such as JavaEE application servers)
• The authorization component is the one that extends the Java
security policy

24. Evolution of the
Java security model
• Core classes of JAAS:
• javax.security.auth.Subject - an authenticated subject
• java.security.Principal - identifying characteristic of a subject
• javax.security.auth.spi.LoginModule - interface for implementors of login (PAM)
modules
• javax.security.auth.login.LoginContext - creates objects used for authentication

25. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
1) upon system startup a security policy is set and a security manager is
installed
Policy.setPolicy(…)
System.setSecurityManager(…)

26. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
2) during classloading (e.g. of a remote applet) bytecode verification is done
and the protection domain is set for the current classloader (along with
the code source, the set of permissions and the set of JAAS principals)

27. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
3) when system code is invoked from the remote code the SecurityManager
is used to check against the intersection of protection domains based on
the chain of threads and their call stacks

28. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
SocketPermission permission = new
SocketPermission(“javantura.com:8000-9000","connect,accept");
SecurityManager sm = System.getSecurityManager();
if (sm != null) sm.checkPermission(permission);

29. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
4) application code can also do permission checking against remote code
using a SecurityManager or an AccessController

30. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
SocketPermission permission = new
SocketPermission(“javantura.com:8000-9000", "connect,accept");
AccessController.checkPermission(permission)

31. Evolution of the
Java security model
• Up to JDK 1.4 the following is a typical flow for permission
checking:
5) application code can also do permission checking with all permissions of
the calling domain or a particular JAAS subject
AccessController.doPrivileged(…)
Subject.doAs(…)
Subject.doAsPrivileged(…)

32. Evolution of the
Java security model
• The security model defined by java.lang.SecurityManager is
customizable
• For example: Oracle JVM uses a custom SecurityManager with
additional permission classes where the code source is a database
schema (containing e.g. Java stored procedures)

33. Evolution of the
Java security model
• JDK 1.5, 1.6 (enhancing the model …) – new additions to the
sandbox model (e.g. LDAP support for JAAS)

34. Evolution of the
Java security model
• JDK 1.7, 1.8 (further enhancing the model …) – enhancements to
the sandbox model (e.g. AccessController.doPrivileged() for
checking against a subset of permissions)

35. Evolution of the
Java security model
• JDK 1.9 and beyond … (applying the model to modules …)
application module
system
module 1
JVM
Browser
http://javantura.com/appmodule
security.policy
system
module 2

36. Evolution of the
Java security model
• By modules we understand modules in JDK as defined by project
Jigsaw
• Modules must conform to the same security model as applets –
each module is loaded by a particular classloader (bootstrap,
extension or system)

37. Evolution of the
Java security model
• Modularization of the JDK system classes allows further to define
fine-grained access control permissions for classes in the system
domain
• This is not currently allowed due to the monolithic nature of the
JDK

38. Outside the sandbox:
APIs for secure coding

39. Outside the sandbox:
APIs for secure coding
• The security sandbox defines a strict model for execution of
remote code in the JVM
• The other side of the coin are the security APIs that provide
utilities for implementing the different aspects of application
security …

40. Outside the sandbox:
APIs for secure coding
• The additional set of APIs includes:
• JCA (Java Cryptography Architecture)
• PKI (Public Key Infrastructure) utilities
• JSSE (Java Secure Socket Extension)
• Java GSS API (Java Generic Security Services)
• Java SASL API (Java Simple Authentication and Security Layer)

41. Designing and coding
with security in mind

42. Designing and coding
with security in mind
• First of all - follow programing guidelines and best practices -
most are not bound to the Java programming language (input
validation, error handling, type safety, access modifiers, resource
cleanup, prepared SQL queries and whatever you can think of …)

43. Designing and coding
with security in mind
• Respect the SecurityManager - design libraries so that they work
in environments with installed SecurityManager
• Example: GSON library does not respect the SecurityManager and
cannot be used without additional reflective permissions in some
scenarios

44. Designing and coding
with security in mind
• Grant minimal permissions to code that requires them - the
principle of "least privilege"
• Copy-pasting, of course, increases the risk of security flows (if the
copied code is flawed)

45. Designing and coding
with security in mind
• Sanitize exception messages from sensitive information - often
this results in an unintended exposal of exploitable information
• Let alone exception stacktraces … in many cases they convey a
wealth of information about the system

46. Thank you

47. References
• Java Security Overview (white paper)
http://www.oracle.com/technetwork/java/js-white-paper-149932.pdf
• Java SE Platform Security Architecture Spec
http://docs.oracle.com/javase/7/docs/technotes/guides/security/spec/s
ecurity-spec.doc.html
• Inside Java 2 Platform Security, 2nd edition
http://www.amazon.com/Inside-Java%C2%BF-Platform-Security-
Implementation/dp/0201787911

48. References
• Java Security, 2nd edition, Scott Oaks
http://shop.oreilly.com/product/9780596001575.do
• Securing Java, Gary McGraw, Ed Felden
http://www.securingjava.com
• Secure Coding Guidelines for Java SE
http://www.oracle.com/technetwork/java/seccodeguide-139067.html#0

49. References
• Java 2 Network Security
http://www.amazon.com/JAVA-Network-Security-2nd-
Edition/dp/0130155926
• Java Security Documentation
http://docs.oracle.com/javase/8/docs/technotes/guides/security/index.
html

50. References
• Core Java Security: Class Loaders, Security Managers and
Encryption
http://www.informit.com/articles/article.aspx?p=1187967
• Overview of Java Security Models
http://docs.oracle.com/cd/E12839_01/core.1111/e10043/introjps.htm#
CHDCEJGH