Concurrent Execution using Shared Resource with Improper Synchronization ('Race Condition')

The program contains a code sequence that can run concurrently with other code, and the code sequence requires temporary, exclusive access to a shared resource, but a timing window exists in which the shared resource can be modified by another code sequence that is operating concurrently.


Description

This can have security implications when the expected synchronization is in security-critical code, such as recording whether a user is authenticated or modifying important state information that should not be influenced by an outsider.

A race condition occurs within concurrent environments, and is effectively a property of a code sequence. Depending on the context, a code sequence may be in the form of a function call, a small number of instructions, a series of program invocations, etc.

A race condition violates these properties, which are closely related:

Exclusivity - the code sequence is given exclusive access to the shared resource, i.e., no other code sequence can modify properties of the shared resource before the original sequence has completed execution.

Atomicity - the code sequence is behaviorally atomic, i.e., no other thread or process can concurrently execute the same sequence of instructions (or a subset) against the same resource.

A race condition exists when an "interfering code sequence" can still access the shared resource, violating exclusivity. Programmers may assume that certain code sequences execute too quickly to be affected by an interfering code sequence; when they are not, this violates atomicity. For example, the single "x++" statement may appear atomic at the code layer, but it is actually non-atomic at the instruction layer, since it involves a read (the original value of x), followed by a computation (x+1), followed by a write (save the result to x).

The interfering code sequence could be "trusted" or "untrusted." A trusted interfering code sequence occurs within the program; it cannot be modified by the attacker, and it can only be invoked indirectly. An untrusted interfering code sequence can be authored directly by the attacker, and typically it is external to the vulnerable program.

Demonstrations

The following examples help to illustrate the nature of this weakness and describe methods or techniques which can be used to mitigate the risk.

Note that the examples here are by no means exhaustive and any given weakness may have many subtle varieties, each of which may require different detection methods or runtime controls.

Example One

This code could be used in an e-commerce application that supports transfers between accounts. It takes the total amount of the transfer, sends it to the new account, and deducts the amount from the original account.

$transfer_amount = GetTransferAmount();
$balance = GetBalanceFromDatabase();

if ($transfer_amount < 0) {
  FatalError("Bad Transfer Amount");
}
$newbalance = $balance - $transfer_amount;
if (($balance - $transfer_amount) < 0) {
  FatalError("Insufficient Funds");
}
SendNewBalanceToDatabase($newbalance);
NotifyUser("Transfer of $transfer_amount succeeded.");
NotifyUser("New balance: $newbalance");

A race condition could occur between the calls to GetBalanceFromDatabase() and SendNewBalanceToDatabase().

Suppose the balance is initially 100.00. An attack could be constructed as follows:

In the following pseudocode, the attacker makes two simultaneous calls of the program, CALLER-1 and CALLER-2. Both callers are for the same user account.
CALLER-1 (the attacker) is associated with PROGRAM-1 (the instance that handles CALLER-1). CALLER-2 is associated with PROGRAM-2.
CALLER-1 makes a transfer request of 80.00.
PROGRAM-1 calls GetBalanceFromDatabase and sets $balance to 100.00
PROGRAM-1 calculates $newbalance as 20.00, then calls SendNewBalanceToDatabase().
Due to high server load, the PROGRAM-1 call to SendNewBalanceToDatabase() encounters a delay.
CALLER-2 makes a transfer request of 1.00.
PROGRAM-2 calls GetBalanceFromDatabase() and sets $balance to 100.00. This happens because the previous PROGRAM-1 request was not processed yet.
PROGRAM-2 determines the new balance as 99.00.
After the initial delay, PROGRAM-1 commits its balance to the database, setting it to 20.00.
PROGRAM-2 sends a request to update the database, setting the balance to 99.00

At this stage, the attacker should have a balance of 19.00 (due to 81.00 worth of transfers), but the balance is 99.00, as recorded in the database.

To prevent this weakness, the programmer has several options, including using a lock to prevent multiple simultaneous requests to the web application, or using a synchronization mechanism that includes all the code between GetBalanceFromDatabase() and SendNewBalanceToDatabase().

Example Two

The following function attempts to acquire a lock in order to perform operations on a shared resource.

void f(pthread_mutex_t *mutex) {

  pthread_mutex_lock(mutex);

  /* access shared resource */


  pthread_mutex_unlock(mutex);

}

However, the code does not check the value returned by pthread_mutex_lock() for errors. If pthread_mutex_lock() cannot acquire the mutex for any reason, the function may introduce a race condition into the program and result in undefined behavior.

In order to avoid data races, correctly written programs must check the result of thread synchronization functions and appropriately handle all errors, either by attempting to recover from them or reporting it to higher levels.

int f(pthread_mutex_t *mutex) {

  int result;

  result = pthread_mutex_lock(mutex);
  if (0 != result)
    return result;


  /* access shared resource */


  return pthread_mutex_unlock(mutex);

}

Example Three

Suppose a processor's Memory Management Unit (MMU) has 5 other shadow MMUs to distribute its workload for its various cores. Each MMU has the start address and end address of "accessible" memory. Any time this accessible range changes (as per the processor's boot status), the main MMU sends an update message to all the shadow MMUs.

Suppose the interconnect fabric does not prioritize such "update" packets over other general traffic packets. This introduces a race condition. If an attacker can flood the target with enough messages so that some of those attack packets reach the target before the new access ranges gets updated, then the attacker can leverage this scenario.

See Also

SEI CERT Oracle Secure Coding Standard for Java - Guidelines 08. Visibility and Atomicity (VNA)

Weaknesses in this category are related to the rules and recommendations in the Visibility and Atomicity (VNA) section of the SEI CERT Oracle Secure Coding Standard fo...

SFP Secondary Cluster: Race Condition Window

This category identifies Software Fault Patterns (SFPs) within the Race Condition Window cluster (SFP20).

CERT C++ Secure Coding Section 14 - Concurrency (CON)

Weaknesses in this category are related to rules in the Concurrency (CON) section of the CERT C++ Secure Coding Standard. Since not all rules map to specific weaknesse...

Comprehensive CWE Dictionary

This view (slice) covers all the elements in CWE.

Weaknesses for Simplified Mapping of Published Vulnerabilities

CWE entries in this view (graph) may be used to categorize potential weaknesses within sources that handle public, third-party vulnerability information, such as the N...

Weaknesses without Software Fault Patterns

CWE identifiers in this view are weaknesses that do not have associated Software Fault Patterns (SFPs), as covered by the CWE-888 view. As such, they represent gaps in...


Common Weakness Enumeration content on this website is copyright of The MITRE Corporation unless otherwise specified. Use of the Common Weakness Enumeration and the associated references on this website are subject to the Terms of Use as specified by The MITRE Corporation.