Threads preserve class invariants when they are allowed to exit normally. Programmers often attempt to terminate threads abruptly when they believe the task is complete, the request has been canceled, or the program or Java Virtual Machine (JVM) must shut down expeditiously.
Certain thread APIs were introduced to facilitate thread suspension, resumption, and termination but were later deprecated because of inherent design weaknesses. For example, the Thread.stop()
method causes the thread to immediately throw a ThreadDeath
exception, which usually stops the thread. More information about deprecated methods is available in MET02-J. Do not use deprecated or obsolete classes or methods.
Invoking Thread.stop()
results in the release of all locks a thread has acquired, potentially exposing the objects protected by those locks when those objects are in an inconsistent state. The thread might catch the ThreadDeath
exception and use a finally
block in an attempt to repair the inconsistent object or objects. However, doing so requires careful inspection of all synchronized methods and blocks because a ThreadDeath
exception can be thrown at any point during the thread's execution. Furthermore, code must be protected from ThreadDeath
exceptions that might occur while executing catch
or finally
blocks [Sun 1999]. Consequently, programs must not invoke Thread.stop()
.
Removing the java.lang.RuntimePermission stopThread
permission from the security policy file prevents threads from being stopped using the Thread.stop()
method. Although this approach guarantees that the program cannot use the Thread.stop()
method, it is nevertheless strongly discouraged. Existing trusted, custom-developed code that uses the Thread.stop()
method presumably depends on the ability of the system to perform this action. Furthermore, the system might fail to correctly handle the resulting security exception. Additionally, third-party libraries may also depend on use of the Thread.stop()
method.
Refer to ERR09-J. Do not allow untrusted code to terminate the JVM for information on preventing data corruption when the JVM is abruptly shut down.
Noncompliant Code Example (Deprecated Thread.stop()
)
This noncompliant code example shows a thread that fills a vector with pseudorandom numbers. The thread is forcefully stopped after a given amount of time.
public final class Container implements Runnable { private final Vector<Integer> vector = new Vector<Integer>(1000); public Vector<Integer> getVector() { return vector; } @Override public synchronized void run() { Random number = new Random(123L); int i = vector.capacity(); while (i > 0) { vector.add(number.nextInt(100)); i--; } } public static void main(String[] args) throws InterruptedException { Thread thread = new Thread(new Container()); thread.start(); Thread.sleep(5000); thread.stop(); } }
Because the Vector
class is thread-safe, operations performed by multiple threads on its shared instance are expected to leave it in a consistent state. For instance, the Vector.size()
method always returns the correct number of elements in the vector, even after concurrent changes to the vector, because the vector instance uses its own intrinsic lock to prevent other threads from accessing it while its state is temporarily inconsistent.
However, the Thread.stop()
method causes the thread to stop what it is doing and throw a ThreadDeath
exception. All acquired locks are subsequently released [API 2014]. If the thread were in the process of adding a new integer to the vector when it was stopped, the vector would become accessible while it is in an inconsistent state. For example, this could result in Vector.size()
returning an incorrect element count because the element count is incremented after adding the element.
Compliant Solution (volatile flag)
This compliant solution uses a volatile flag to request thread termination. The shutdown()
accessor method is used to set the flag to true. The thread's run()
method polls the done
flag and terminates when it is set.
public final class Container implements Runnable { private final Vector<Integer> vector = new Vector<Integer>(1000); private volatile boolean done = false; public Vector<Integer> getVector() { return vector; } public void shutdown() { done = true; } @Override public synchronized void run() { Random number = new Random(123L); int i = vector.capacity(); while (!done && i > 0) { vector.add(number.nextInt(100)); i--; } } public static void main(String[] args) throws InterruptedException { Container container = new Container(); Thread thread = new Thread(container); thread.start(); Thread.sleep(5000); container.shutdown(); } }
Compliant Solution (Interruptible)
In this compliant solution, the Thread.interrupt()
method is called from main()
to terminate the thread. Invoking Thread.interrupt()
sets an internal interrupt status flag. The thread polls that flag using the Thread.interrupted()
method, which both returns true if the current thread has been interrupted and clears the interrupt status flag.
public final class Container implements Runnable { private final Vector<Integer> vector = new Vector<Integer>(1000); public Vector<Integer> getVector() { return vector; } @Override public synchronized void run() { Random number = new Random(123L); int i = vector.capacity(); while (!Thread.interrupted() && i > 0) { vector.add(number.nextInt(100)); i--; } } public static void main(String[] args) throws InterruptedException { Container c = new Container(); Thread thread = new Thread(c); thread.start(); Thread.sleep(5000); thread.interrupt(); } }
A thread may use interruption for performing tasks other than cancellation and shutdown. Consequently, a thread should be interrupted only when its interruption policy is known in advance. Failure to do so can result in failed interruption requests.
Risk Assessment
Forcing a thread to stop can result in inconsistent object state. Critical resources could also leak if cleanup operations are not carried out as required.
Rule | Severity | Likelihood | Remediation Cost | Priority | Level |
---|---|---|---|---|---|
THI05-J | Low | Probable | Medium | P4 | L3 |
Automated Detection
Tool | Version | Checker | Description |
---|---|---|---|
Parasoft Jtest | 2024.1 | CERT.THI05.THRD | Avoid calling unsafe deprecated methods of 'Thread' and 'Runtime' |
Related Guidelines
POS47-C. Do not use threads that can be canceled asynchronously | |
CWE-705, Incorrect Control Flow Scoping |
Android Implementation Details
On Android, Thread.stop()
was deprecated in API level 1.
Bibliography
[API 2006] | Class |
Section 24.3, "Stopping a Thread" | |
Chapter 7, "Cancellation and Shutdown" | |
Section 2.4, "Two Approaches to Stopping a Thread" | |
Concurrency Utilities, More information: Java Thread Primitive Deprecation | |
[JPL 2006] | Section 14.12.1, "Don't Stop" |
[Sun 1999] |
11 Comments
Dhruv Mohindra
Here is some text from the JDK7 docs -
We could use this as a separate rule that warns against such silent exceptions if there are more cases like this. If anyone remembers any from the top of the head (or otherwise), please do write.
Thomas Hawtin
Off the top of my head, undeclared checked exceptions can also be thrown by:
Dhruv Mohindra
Thanks! These have been grouped together under EXC10-J. Do not let code throw undeclared checked exceptions now.
Dhruv Mohindra
Throwing a
ThreadDeath
error to stop a thread is considered bad practice, however, thefinally
block is still executed. Assuming consistency of object state is maintained there, this may not be an issue.Dhruv Mohindra
The NCE uses a deprecated method
Thread.stop()
. Not sure if this qualifies and how I can suggest better solutions without this NCE. Perhaps this whole guideline belongs in MET15-J. Do not use deprecated or obsolete methods.Dhruv Mohindra
The blocking I/O NCEs are less appealing because they share a socket between multiple threads. This is essentially coarse grained locking and violates LCK09-J. Do not perform operations that may block while holding a lock. A better idea is to leave the socket local to the method in the NCEs so that each thread has its own copy. But then we cannot provide ths shutdown method in the CS. Consequently, the CS can solve the problem by using a static thread local and providing a way to shut down the socket.
David Svoboda
In the 2nd NCCE/CS code samples, the socket object is private to the class; consequently the socket is only accessible to the thread running the
run()
method. In the CS closure of the socket is encapsulated in theshutdown()
method; hence no socket-sharing is going on. So the only way you can relate this rule to CON20 is the implicit locking that goes on when a socket blocks on input, which is what this rule is all about.David Svoboda
I've added an
isRunning
flag to prevent multiple threads from blocking on the socket.Dhruv Mohindra
In the new title "Ensure that threads are properly terminated", what does properly constitute? Is thread cancellation proper termination (cancellation != forced thread shutdown != Thread.stop())? This guideline only talks about cleanly stopping threads as opposed to cleanly canceling threads. So the title was - "Ensure that threads are stopped cleanly".
Ulrich Scholz
I wonder if Thread.stop() would be appropriate in the following case: We build a tool that consists of a complex parser and an outer control loop. Although we are careful (and check for interruption etc.), there's always a chance that the parser will enter an infinite loop. If so, the control loop kills the parser with Thread.stop(). Because there is no connection between the two parts of our tool (locking, shared data structures, ...), I see no problem in doing so. The only other option I see is starting the parser in its own JVM, which is then killed in case.
David Svoboda
I suppose with sufficient work, you can argue that Thread.stop() is appropriate in cancelling a specific thread. In order to do that you must know the following:
1. The thread cannot be stopped while in a section of code that must not throw an exception...such as in a finally clause.
2. The details of how Thread.stop() does its work. This includes details of your JVM and OS...and hardware most likely.
I suspect that having a sufficient understanding of these factors will be very difficult...harder than understanding why your control loop might freeze and preventing that.