Signal Handler Race Condition

The product uses a signal handler that introduces a race condition.


Description

Race conditions frequently occur in signal handlers, since signal handlers support asynchronous actions. These race conditions have a variety of root causes and symptoms. Attackers may be able to exploit a signal handler race condition to cause the product state to be corrupted, possibly leading to a denial of service or even code execution.

These issues occur when non-reentrant functions, or state-sensitive actions occur in the signal handler, where they may be called at any time. These behaviors can violate assumptions being made by the "regular" code that is interrupted, or by other signal handlers that may also be invoked. If these functions are called at an inopportune moment - such as while a non-reentrant function is already running - memory corruption could occur that may be exploitable for code execution. Another signal race condition commonly found occurs when free is called within a signal handler, resulting in a double free and therefore a write-what-where condition. Even if a given pointer is set to NULL after it has been freed, a race condition still exists between the time the memory was freed and the pointer was set to NULL. This is especially problematic if the same signal handler has been set for more than one signal -- since it means that the signal handler itself may be reentered.

There are several known behaviors related to signal handlers that have received the label of "signal handler race condition":

Shared state (e.g. global data or static variables) that are accessible to both a signal handler and "regular" code

Shared state between a signal handler and other signal handlers

Use of non-reentrant functionality within a signal handler - which generally implies that shared state is being used. For example, malloc() and free() are non-reentrant because they may use global or static data structures for managing memory, and they are indirectly used by innocent-seeming functions such as syslog(); these functions could be exploited for memory corruption and, possibly, code execution.

Association of the same signal handler function with multiple signals - which might imply shared state, since the same code and resources are accessed. For example, this can be a source of double-free and use-after-free weaknesses.

Use of setjmp and longjmp, or other mechanisms that prevent a signal handler from returning control back to the original functionality

While not technically a race condition, some signal handlers are designed to be called at most once, and being called more than once can introduce security problems, even when there are not any concurrent calls to the signal handler. This can be a source of double-free and use-after-free weaknesses.

Signal handler vulnerabilities are often classified based on the absence of a specific protection mechanism, although this style of classification is discouraged in CWE because programmers often have a choice of several different mechanisms for addressing the weakness. Such protection mechanisms may preserve exclusivity of access to the shared resource, and behavioral atomicity for the relevant code:

Avoiding shared state

Using synchronization in the signal handler

Using synchronization in the regular code

Disabling or masking other signals, which provides atomicity (which effectively ensures exclusivity)

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 registers the same signal handler function with two different signals (CWE-831). If those signals are sent to the process, the handler creates a log message (specified in the first argument to the program) and exits.

char *logMessage;

void handler (int sigNum) {

  syslog(LOG_NOTICE, "%s\n", logMessage);
  free(logMessage);
  /* artificially increase the size of the timing window to make demonstration of this weakness easier. */

  sleep(10);
  exit(0);

}

int main (int argc, char* argv[]) {

  logMessage = strdup(argv[1]);
  /* Register signal handlers. */

  signal(SIGHUP, handler);
  signal(SIGTERM, handler);
  /* artificially increase the size of the timing window to make demonstration of this weakness easier. */

  sleep(10);

}

The handler function uses global state (globalVar and logMessage), and it can be called by both the SIGHUP and SIGTERM signals. An attack scenario might follow these lines:

The program begins execution, initializes logMessage, and registers the signal handlers for SIGHUP and SIGTERM.

The program begins its "normal" functionality, which is simplified as sleep(), but could be any functionality that consumes some time.

The attacker sends SIGHUP, which invokes handler (call this "SIGHUP-handler").

SIGHUP-handler begins to execute, calling syslog().

syslog() calls malloc(), which is non-reentrant. malloc() begins to modify metadata to manage the heap.

The attacker then sends SIGTERM.

SIGHUP-handler is interrupted, but syslog's malloc call is still executing and has not finished modifying its metadata.

The SIGTERM handler is invoked.

SIGTERM-handler records the log message using syslog(), then frees the logMessage variable.

At this point, the state of the heap is uncertain, because malloc is still modifying the metadata for the heap; the metadata might be in an inconsistent state. The SIGTERM-handler call to free() is assuming that the metadata is inconsistent, possibly causing it to write data to the wrong location while managing the heap. The result is memory corruption, which could lead to a crash or even code execution, depending on the circumstances under which the code is running.

Note that this is an adaptation of a classic example as originally presented by Michal Zalewski [REF-360]; the original example was shown to be exploitable for code execution.

Also note that the strdup(argv[1]) call contains a potential buffer over-read (CWE-126) if the program is called without any arguments, because argc would be 0, and argv[1] would point outside the bounds of the array.

Example Two

The following code registers a signal handler with multiple signals in order to log when a specific event occurs and to free associated memory before exiting.

#include <signal.h>
#include <syslog.h>
#include <string.h>
#include <stdlib.h>

void *global1, *global2;
char *what;
void sh (int dummy) {

  syslog(LOG_NOTICE,"%s\n",what);
  free(global2);
  free(global1);
  /* Sleep statements added to expand timing window for race condition */

  sleep(10);
  exit(0);

}

int main (int argc,char* argv[]) {

  what=argv[1];
  global1=strdup(argv[2]);
  global2=malloc(340);
  signal(SIGHUP,sh);
  signal(SIGTERM,sh);
  /* Sleep statements added to expand timing window for race condition */

  sleep(10);
  exit(0);

}

However, the following sequence of events may result in a double-free (CWE-415):

a SIGHUP is delivered to the process

sh() is invoked to process the SIGHUP

This first invocation of sh() reaches the point where global1 is freed

At this point, a SIGTERM is sent to the process

the second invocation of sh() might do another free of global1

this results in a double-free (CWE-415)

This is just one possible exploitation of the above code. As another example, the syslog call may use malloc calls which are not async-signal safe. This could cause corruption of the heap management structures. For more details, consult the example within "Delivering Signals for Fun and Profit" [REF-360].

See Also

Comprehensive Categorization: Concurrency

Weaknesses in this category are related to concurrency.

SFP Secondary Cluster: Missing Lock

This category identifies Software Fault Patterns (SFPs) within the Missing Lock cluster (SFP19).

Concurrency Issues

Weaknesses in this category are related to concurrent use of shared resources.

Comprehensive CWE Dictionary

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

CWE Cross-section

This view contains a selection of weaknesses that represent the variety of weaknesses that are captured in CWE, at a level of abstraction that is likely to be useful t...

Weaknesses Introduced During Implementation

This view (slice) lists weaknesses that can be introduced during implementation.


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.