...
If however, one follows a fine-grained locking approach by using member locks, deadlocks can still arise unless it is ensured that each and every thread always requests locks in the same order.
Noncompliant Code Example
This noncompliant code example shows a more subtle deadlock issue caused due to incorrect implementation of synchronization. Assume that an attacker has two bank accounts and is capable of requesting two amount transfer operations in succession. This corresponds to two threads as shown in main()
. Objects a
and b
are constructed where the amount will be transferred from a
to b
by first depositing the amount from a
to b
and then zeroing out a
(and vice-versa for the second call to withdraw).
An insidious deadlock condition can sometimes arise. Both the threads invoke the synchronized deposit
method on objects a
and b
respectively and there is no interleaving of locks. However, the deposit
method is designed to request a lock on object a
for the first thread and object b
for the second. If both threads have acquired separate locks together and the call to withdraw()
requests an already held lock, there is a deadlock situation. In case Thread 1 finishes executing before Thread 2, there would not be any issues. Sequences where the threads alternate, such as, T1, T2, T1, T2 can result in a deadlock (Tx = Thread number).
Code Block | ||
---|---|---|
| ||
class BadLocks {
private int balance; // total amount
private BadLocks(int balance) {
this.balance = balance;
}
private synchronized void withdraw(BadLocks bl) {
System.out.println("Withdrawing amount...");
this.balance = 0;
}
private synchronized void deposit(BadLocks bl) {
System.out.println("Depositing amount...");
bl.balance += this.balance;
bl.withdraw(bl);
}
public static void main(String[] args) throws Exception {
final BadLocks a = new BadLocks(5000);
final BadLocks b = new BadLocks(6000);
// These two threads correspond to two malicious requests triggered by the attacker
Thread t1 = new Thread(new Runnable() {
public void run() {
a.deposit(b);
}
});
Thread t2 = new Thread(new Runnable() {
public void run() {
b.deposit(a);
}
});
t1.start();
t2.start();
}
}
|
Compliant Solution
To be compliant, do not request a lock when it is already held by another object where that object is waiting for the requesting object to release its own lock. This can be achieved by invoking the withdraw
method without qualifying it with the bl
parameter.
Code Block | ||
---|---|---|
| ||
withdraw(bl);
|
Noncompliant Code Example
In this noncompliant code example, a deadlock occurs if the transfer
method acquires the monitors in decreasing numeric order, while the sumHelper
method uses an increasing numeric order to acquire its locks.
Code Block | ||
---|---|---|
| ||
class Stocks implements FundConstants { static int[] balances = new int[noOfStocks]; static Object[] locks = new Object[noOfStocks]; static { for (int n=0; n<noOfStocks; n++) { balances[n] = 10000; locks[n] = new Object(); } } static void transfer(Transfer t) { int lo, hi; if (t.fundFrom > t.fundTo) {// acquires the monitors in decreasing numeric order lo = t.fundFrom; hi = t.fundTo; } else { lo = t.fundTo; hi = t.fundFrom; } synchronized (locks[lo]) { synchronized (locks[hi]) { balances[t.fundFrom] -= t.amount; balances[t.fundTo] += t.amount; } } } static int sumHelper (int next) { synchronized (locks[next]) { if (next == (noOfStocks-1)) { return balances[next]; } else { return balances[next] + sumHelper(next+1); } } } static void checkSystem() { int actual = 0; actual = sumHelper(0); System.out.println("Actual balance is" + actual); } } |
Compliant Solution
To implement a fine-grained locking strategy, request a separate lock for each position in the balances
array. Since one cannot lock on primitive types, a direct lock on the items in the balances
array cannot be obtained. Instead, one has to create an array of Object (locks
).
...
Code Block | ||
---|---|---|
| ||
class Stocks implements FundConstants { static int[] balances = new int[noOfStocks]; static Object[] locks = new Object[noOfStocks]; static { for (int n=0; n<noOfStocks; n++) { balances[n] = 10000; locks[n] = new Object(); } } static void transfer(Transfer t) { int lo, hi; if (t.fundFrom < t.fundTo) { lo = t.fundFrom; hi = t.fundTo; } else { lo = t.fundTo; hi = t.fundFrom; } synchronized (locks[lo]) { synchronized (locks[hi]) { balances[t.fundFrom] -= t.amount; balances[t.fundTo] += t.amount; } } } static int sumHelper (int next) { synchronized (locks[next]) { if (next == (noOfStocks-1)) { return balances[next]; } else { return balances[next] + sumHelper(next+1); } } } static void checkSystem() { int actual = 0; actual = sumHelper(0); System.out.println("Actual balance is " + actual); } } |
Risk Assessment
Fine-grained locking may result in deadlocks if some thread does not always request locks in the same order as others.
Rule | Severity | Likelihood | Remediation Cost | Priority | Level |
---|---|---|---|---|---|
CON34- J | low | likely | high | P3 | L3 |
Automated Detection
TODO
Related Vulnerabilities
Search for vulnerabilities resulting from the violation of this rule on the CERT website.
References
Wiki Markup |
---|
\[[JLS 05|AA. Java References#JLS 05]\] [Chapter 17, Threads and Locks|http://java.sun.com/docs/books/jls/third_edition/html/memory.html] \[[SDN 08|AA. Java References#SDN 08]\] [Sun Developer Network Tech tips|http://java.sun.com/developer/TechTips/2000/tt0328.html] \[[MITRE 09|AA. Java References#MITRE 09]\] [CWE ID 412|http://cwe.mitre.org/data/definitions/412.html] "Unrestricted Lock on Critical Resource" |
...