Document: WG14 N1194
Date: 2006/18/10
Author: Douglas Gwyn


Defect Report #XXX

Previous Defect Report < - > Next Defect Report


Submitter: Gwyn (US)
Submission Date: 2006-10-17
Source: Douglas A. Gwyn <gwyn@arl.army.mil>
Version: 1.0
Date: 2006-10-17
Subject: gets is generally unsafe

Summary

The gets function's notorious vulnerability to buffer overrun ought to be addressed.

Rationale

The gets function draws much criticism due to its vulnerability to buffer overrun, which is inherent in its legacy interface specification. Its very presence in the Standard has been taken by many as evidence of WG14's ineptitude or lack of concern for software reliability, despite arguments to the contrary. The Committee may be more favorably regarded within the programming community if it takes reasonable steps to address this issue.

The recent publication of TR 24731, which specifies an alternative function gets_s that could be used instead of gets, does not satisfy the critics who claim that the continuing existence of the gets specification in the C standard amounts to an endorsement of its unsafe use in new programs.

Discussion

Consider this representative usage of gets:

#include <stdio.h>
static char line[BUFSIZ];     /* BUFSIZ is bigger than any normal text line */
extern void process(char *);
int main(void) {
    while (gets(line))
        process(line);        /* may invoke puts(line), etc. */
    return 0;
}
   
This shows how convenient the gets interface is. The well-known problem with this interface occurs when the standard input stream contains a text line longer than the allocated size of the buffer; because gets has no way to know that size, it blindly continues to store data beyond the end of the array, with potentially devastating impact on program operation. The infamous 1988 Morris Internet worm was merely the first of many attacks that exploit this behavior to breach security in network applications.

What might be done to improve the specification for gets so that the safety of this exceptionally convenient interface can be assured? (I do not recommend removing it altogether!) It seems evident that the only feasible change would be to impose a limit on the amount of data transferred. Requiring the programmer to establish the limit through some additional interface would sacrifice the convenience. The alternative is to impose some constant limit, in which case the remaining question is what would be a suitable constant. That can be answered by examining existing uses of gets to determine typical buffer sizes. It appears that two usage patterns predominate: Using the BUFSIZ macro which happens to be conveniently at hand as a consequence of #include <stdio.h>, or using some assumed text-line length such as 80.

Restricting the amount of data transferred to only 80 characters may be too severe for many applications, and in any case it would necessitate the introduction of a new limit macro such as LINE_MAX to provide a convenient way for programmers to declare suitable buffer arrays. Therefore I recommend instead that the existing usage of BUFSIZ be legitimatized, as follows.

Suggested Technical Corrigendum

Add the following sentence to the Description in subclause 7.19.7.7 (The gets function)], between the two existing sentences:
At most BUFSIZ-1 characters are copied to the array; excessive characters are discarded.
(The portion after the semicolon isn't strictly necessary, but it adds clarity.)

Impact

The proposed change to the gets specification would have the effect of preventing buffer overruns in many existing applications. Overly long input lines would be silently truncated (which is better than the alternative of treating them as multiple lines).

Existing applications using small buffers would not be automatically rescued by this change; however, there would be a simple source-code fix (change the buffer size). New applications would obtain safe behavior by using the known limit for buffer allocation, exactly as in the above example.

Adoption of such a change would demonstrate the committee's willingness to improve specifications compatibly with the existing standard, without resorting to unnecessarily drastic measures.


Committee Discussion (for history only)

???

Proposed Technical Corrigendum

???


Previous Defect Report < - > Next Defect Report