logo

Unintended Reentrant Invocation of Non-reentrant Code Via Nested Calls

Draft Base
Structure: Simple
Description

This vulnerability occurs when a non-reentrant function is called, and during its execution, another call is triggered that unexpectedly re-enters the same non-reentrant code path, corrupting its internal state.

Extended Description

In complex software, a single function call can branch into many unexpected execution paths, especially when processing external inputs. Attackers can often manipulate these inputs—like scripts in a web browser or embedded code in a PDF—to force the program into a state where a nested call re-enters code that wasn't designed to handle re-invocation. This is dangerous because the non-reentrant code typically relies on global or static data that gets corrupted when called again before the first invocation finishes. For developers, the core issue is that a function assumes its own internal state or global resources remain stable for the duration of its execution. When a nested call—perhaps via a callback, event handler, or virtual method—breaks that assumption, it leads to race conditions, data corruption, or crashes. To prevent this, you must audit code paths triggered during non-reentrant operations and isolate or protect any state that shouldn't be accessed concurrently, even from within the same thread.
Common Consequences 1
Scope: Integrity

Impact: Unexpected State

Exploitation of this weakness can leave the application in an unexpected state and cause variables to be reassigned before the first invocation has completed. This may eventually result in memory corruption or unexpected code execution.
Potential Mitigations 2
Phase: Architecture and Design
When architecting a system that will execute untrusted code in response to events, consider executing the untrusted event handlers asynchronously (asynchronous message passing) as opposed to executing them synchronously at the time each event fires. The untrusted code should execute at the start of the next iteration of the thread's message loop. In this way, calls into non-reentrant code are strictly serialized, so that each operation completes fully before the next operation begins. Special attention must be paid to all places where type coercion may result in script execution. Performing all needed coercions at the very beginning of an operation can help reduce the chance of operations executing at unexpected junctures.

Effectiveness: High

Phase: Implementation
Make sure the code (e.g., function or class) in question is reentrant by not leveraging non-local data, not modifying its own code, and not calling other non-reentrant code.

Effectiveness: High
Demonstrative Examples 2
The implementation of the Widget class in the following C++ code is an example of code that is not designed to be reentrant. If an invocation of a method of Widget inadvertently produces a second nested invocation of a method of Widget, then data member backgroundImage may unexpectedly change during execution of the outer call.

Code Example:

Bad
C++
c++
Looking closer at this example, Widget::click() calls backgroundImage->click(), which in turn calls scriptEngine->fireOnImageClick(). The code within fireOnImageClick() invokes the appropriate script handler routine as defined by the document being rendered. In this scenario this script routine is supplied by an adversary and this malicious script makes a call to Widget::changeBackgroundImage(), deleting the Image object pointed to by backgroundImage. When control returns to Image::click, the function's backgroundImage "this" pointer (which is the former value of backgroundImage) is a dangling pointer. The root of this weakness is that while one operation on Widget (click) is in the midst of executing, a second operation on the Widget object may be invoked (in this case, the second invocation is a call to different method, namely changeBackgroundImage) that modifies the non-local variable.
This is another example of C++ code that is not designed to be reentrant.

Code Example:

Bad
C++
c++
The expected order of operations is a call to Request::setup(), followed by a call to Request::send(). Request::send() calls scriptEngine->coerceToString(_data) to coerce a script-provided parameter into a string. This operation may produce script execution. For example, if the script language is ECMAScript, arbitrary script execution may result if _data is an adversary-supplied ECMAScript object having a custom toString method. If the adversary's script makes a new call to Request::setup, then when control returns to Request::send, the field uri and the local variable credentials will no longer be consistent with one another. As a result, credentials for one resource will be shared improperly with a different resource. The root of this weakness is that while one operation on Request (send) is in the midst of executing, a second operation may be invoked (setup).
Observed Examples 2
CVE-2014-1772In this vulnerability, by registering a malicious onerror handler, an adversary can produce unexpected re-entrance of a CDOMRange object. [REF-1098]
CVE-2018-8174This CVE covers several vulnerable scenarios enabled by abuse of the Class_Terminate feature in Microsoft VBScript. In one scenario, Class_Terminate is used to produce an undesirable re-entrance of ScriptingDictionary during execution of that object's destructor. In another scenario, a vulnerable condition results from a recursive entrance of a property setter method. This recursive invocation produces a second, spurious call to the Release method of a reference-counted object, causing a UAF when that object is freed prematurely. This vulnerability pattern has been popularized as "Double Kill". [REF-1099]
References 2
Root Cause Analysis of CVE-2014-1772 - An Internet Explorer Use After Free Vulnerability
Jack Tang
05-11-2014
https://www.trendmicro.com/en_us/research.html(2023-04-07)
ID: REF-1098
It's Time To Terminate The Terminator
Simon Zuckerbraun
15-05-2018
https://www.zerodayinitiative.com/blog/2018/5/15/its-time-to-terminate-the-terminator
ID: REF-1099
Applicable Platforms
Languages:
Not Language-Specific : Undetermined
Related Attack Patterns
Related Weaknesses
ChildOf: Insufficient Control Flow Management (CWE-691)
PeerOf: Use of a Non-reentrant Function in a Concurrent Context (CWE-663)
CanPrecede: Use After Free (CWE-416)