Backwards-Compatible Bounds Checking for Arrays and Pointers In

Total Page:16

File Type:pdf, Size:1020Kb

Backwards-Compatible Bounds Checking for Arrays and Pointers In Backwardscompatible b ounds checking for arrays and p ointers in C programs Richard W M Jones and Paul H J Kelly Department of Computing Imp erial College of Science Technology and Medicine Queens Gate London SW BZ Abstract within which the resulting p ointer should p oint A p ointer in C can b e used in a context divorced This pap er presents a new approach to enforcing array from the name of the storage region for which it is valid b ounds and p ointer checking in the C language Check its intended referent and this has prevented a fully ing is rigorous in the sense that the result of p ointer satisfactory b ounds checking mechanism from b eing de arithmetic must refer to the same ob ject as the orig veloped There is overwhelming evidence that b ounds inal p ointer this ob ject is sometimes called the in checking is desirable and a number of schemes have tended referent The novel asp ect of this work is b een presented The main dierence b etween our work that checked co de can interoperate without restriction and Kendalls bcc and Steens rtcc is that in with unchecked co de without interface problems with our scheme the representation of p ointers is unchanged some eective checking and without false alarms This This is crucial since it means that interoperation with backwards compatibility prop erty allows the overheads nonchecked mo dules and libraries still works and much of checking to b e conned to susp ect mo dules and also checking is still p ossible Compared with interpretative facilitates the use of libraries for which source co de is schemes like SabreC we oer the p otential for much not available The pap er describ es the scheme its pro higher p erformance Patil and Fischer present a totype implementation as an extension to the GNU C sophisticated technique with very low overheads using compiler presents exp erimental results to evaluate its a second CPU to p erform checking in parallel Unfor eectiveness and discusses p erformance issues and the tunately their scheme requires function interfaces to b e eectiveness of some simple optimisations changed to carry information ab out p ointers so also has the interoperation problem Another approach is exemplied by the commercially Introduction and related work available checking package Purify Purify pro cesses the binary representation of the software so can handle C is unusual among programming languages in provid binaryonly co de Each memory access instruction is ing the programmer with the full p ower of p ointers mo died to maintain a bit map of valid storage regions Languages in the PascalAlgol family have arrays and and whether each byte has b een initialised Accesses p ointers with the restriction that arithmetic on p oint to unallo cated or uninitialised lo cations are rep orted ers is disallowed Languages like BCPL allow arbitrary as errors Purify catches many imp ortant bugs and is op erations on p ointers but lack types and so require fairly ecient However Purify do es not catch abuse clumsy scaling by ob ject sizes of p ointer arithmetic which yields a p ointer to a valid An advantage of the PascalAlgol approach is that region which is not the intended referent Fischer and array references can b e checked at runtime fairly e Patil provide evidence for the imp ortance of this ciently in fact so eciently that there is a go o d case renement for b oundschecking in pro duction co de Bounds check Our goals in this pap er are to describ e a metho d of ing is easy for arrays b ecause the array subscript syn b ounds checking C programs that fullls the following tax sp ecies b oth the address calculation and the array criteria Backwards compatibility the ability to mix checked co de and unchecked libraries for which the source Presented at AADEBUG LinkopingSweden may b e proprietary or otherwise unavailable Works with all common C programming styles Rigorously rejects violations of the ANSI C stan Bounds checking is not blo cked or weakened by the dard use of a cast ie type co ercion Casts can prop erly b e used to change the type of the ob ject to which a Checks static and stack ob jects as well as ob jects p ointer refers but cannot b e used to turn a p ointer dynamically allo cated with malloc to one ob ject into a p ointer to another A corollary is that b ounds checking is not type checking it do es Understands scop e of automatic variables not prevent storage from b eing declared with one data Performance including the ability to b e able to structure and used with another distribute programs with checks compiled in More subtly note that for this reason b ounds check ing in C cannot easily validate use of arrays of structs There remain some circumstances in which checking is which contain arrays in turn incomplete as we describ e later these are fairly un Casts and unions can b e used to create a p ointer common in practice The main shortcoming of the im from an ob ject of any other type in a machinedependent plementation describ ed in this pap er is that the p erfor way This cannot b e checked using our technique nor mance is currently p o or However the approach has by earlier approaches to b ounds checking since there is fundamental p erformance advantages over previously no ob ject for the p ointer to b e derived from published work Because checked co de interoperates easily with unchecked co de the p erformance p enalty is conned to those mo dules where it is needed Fur The technique and its advantages thermore there is substantial scop e for optimisation of lo opinvariant p ointers and p ointers which are induc In this section we review earlier approaches and explain tion variables Because the p ointer representation is the basis for the new approach unchanged there is no residual overhead once checking co de is eliminated We return to this issue in Section Earlier approaches to carrying b ounds information Overview of this pap er The next section reviews the problem of b ounds check Storage object ing for C and the limitations the language places on the checking that can b e done In the following section the new approach is introduced and we explain how unlike earlier schemes our b ounds checking scheme al base: lows interoperation with unchecked co de Then we give pointer: some details of our implementation and discuss some their eectiveness Finally we dis optimisations and limit: the eectiveness of the scheme in the light of our cuss Enhanced pointer exp erience with some large and wellknown C programs Figure Mo died p ointer representation p ointer Ob jects b ounds checking in C and its baseaddressextent triple limitations In earlier work in this area b ounds information is carried with each p ointer at runtime A ANSI C conveniently allows us to dene an object as simple approach is to represent each p ointer as a triple the fundamental unit of memory allo cation Ob jects the p ointer together with the storage regions base ad are created by declarations or allo cations such as those dress and limit or extent Checking is then straight shown in Table which may b e static automatic ie forward The larger size of p ointers requires changes in stackallocated or dynamically allo cated storage allo cation and the co de generator must b e mo d Ob jects are stored sequentially in memory and can ied to copy p ointers correctly The change in p ointer not overlap Op erations are p ermitted which manipu size can b e avoided by replacing each p ointer with an late p ointers within ob jects but p ointer op erations are index into a table which contains the p ointerbaselimit not p ermitted to cross b etween two ob jects There is triple no ordering dened b etween ob jects and the program The net eect of b oth metho ds is the same When mer should never b e allowed to make assumptions ab out the program at runtime comes to use a p ointer it how ob jects are arranged in memory must rst verify that the op eration that is ab out to b e int a A simple variable int a An array struct f g a A single record struct f g a An array of records malloc A single unit of memory allo cated with malloc Table Typical ob jects p erformed is correct It uses the information ab out the Every valid p ointervalued expression in C derives base and size of the array or structure b eing p ointed to its result from exactly one original storage ob ject If to decide if a particular index is legal the result of the p ointer calculation refers to a dierent ob ject it is invalid Although it sometimes useful to know where an in Unchanged p ointer representation valid p ointer has b een calculated rep orting every in stance can yield many false alarms We therefore re The problem with b oth these schemes is that the mo d place such incorrectlyderived p ointers with a p ointer ied p ointer representation is not interpreted correctly value which is always invalid called ILLEGAL Dened by co de compiled without b ounds checking enabled as void in our implementation This ensures This is a problem wherever a p ointer is passed to or that a b ounds error is rep orted when the p ointer is ac from an unchecked pro cedure whether as a parameter tually used a result or in a global variable It is of course often p ossible to translate p ointers where necessary called encapsulation in bcc and rtcc but this is incon Example p ointers to ob jects venient and dicult to do reliably eg where a func tion p ointer may refer either to checked or an unchecked Because of these diculties in rtcc only op routine Dead space between objects erating system calls are encapsulated all libraries
Recommended publications
  • Bounds Checking on GPU
    Noname manuscript No. (will be inserted by the editor) Bounds Checking on GPU Troels Henriksen Received: date / Accepted: date Abstract We present a simple compilation strategy for safety-checking array indexing in high-level languages on GPUs. Our technique does not depend on hardware support for abnormal termination, and is designed to be efficient in the non-failing case. We rely on certain properties of array languages, namely the absence of arbitrary cross-thread communication, to ensure well-defined execution in the presence of failures. We have implemented our technique in the compiler for the functional array language Futhark, and an empirical eval- uation on 19 benchmarks shows that the geometric mean overhead of checking array indexes is respectively 4% and 6% on two different GPUs. Keywords GPU · functional programming · compilers 1 Introduction Programming languages can be divided roughly into two categories: unsafe languages, where programming errors can lead to unpredictable results at run- time; and safe languages, where all risky operations are guarded by run-time checks. Consider array indexing, where an invalid index will lead an unsafe lan- guage to read from an invalid memory address. At best, the operating system will stop the program, but at worst, the program will silently produce invalid results. A safe language will perform bounds checking to verify that the array index is within the bounds of the array, and if not, signal that something is amiss. Some languages perform an abnormal termination of the program and print an error message pointing to the offending program statement. Other languages throw an exception, allowing the problem to be handled by the pro- gram itself.
    [Show full text]
  • Cmsc330 Cybersecurity
    cmsc330 Cybersecurity Cybersecurity Breaches Major security breaches of computer systems are a fact of life. They affect companies, governments, and individuals. Focusing on breaches of individuals' information, consider just a few examples: Equifax (2017) - 145 million consumers’ records Adobe (2013) - 150 million records, 38 million users eBay (2014) - 145 million records Anthem (2014) - Records of 80 million customers Target (2013) - 110 million records Heartland (2008) - 160 million records Vulnerabilities: Security-relevant Defects The causes of security breaches are varied but many of them, including those given above, owe to a defect (or bug) or design flaw in a targeted computer system's software. The software problem can be exploited by an attacker. An exploit is a particular, cleverly crafted input, or a series of (usually unintuitive) interactions with the system, which trigger the bug or flaw in a way that helps the attacker. Kinds of Vulnerability One obvious sort of vulnerability is a bug in security policy enforcement code. For example, suppose you are implementing an operating system and you have code to enforce access control policies on files. This is the code that makes sure that if Alice's policy says that only she is allowed to edit certain files, then user Bob will not be allowed to modify them. Perhaps your enforcement code failed to consider a corner case, and as a result Bob is able to write those files even though Alice's policy says he shouldn't. This is a vulnerability. A more surprising sort of vulnerability is a bug in code that seems to have nothing to do with enforcing security all.
    [Show full text]
  • Safe Arrays Via Regions and Dependent Types
    Safe Arrays via Regions and Dependent Types Christian Grothoff1 Jens Palsberg1 Vijay Saraswat2 1 UCLA Computer Science Department University of California, Los Angeles [email protected], [email protected] 2 IBM T.J. Watson Research Center P.O. Box 704, Yorktown Heights, NY 10598, USA [email protected] Abstract. Arrays over regions of points were introduced in ZPL in the late 1990s and later adopted in Titanium and X10 as a means of simpli- fying the programming of high-performance software. A region is a set of points, rather than an interval or a product of intervals, and enables the programmer to write a loop that iterates over a region. While conve- nient, regions do not eliminate the risk of array bounds violations. Until now, language implementations have resorted to checking array accesses dynamically or to warning the programmer that bounds violations lead to undefined behavior. In this paper we show that a type system for a language with arrays over regions can guarantee that array bounds vi- olations cannot occur. We have developed a core language and a type system, proved type soundness, settled the complexity of the key deci- sion problems, implemented an X10 version which embodies the ideas of our core language, written a type checker for our X10 version, and ex- perimented with a variety of benchmark programs. Our type system uses dependent types and enables safety without dynamic bounds checks. 1 Introduction Type-safe languages allow programmers to be more productive by eliminating such difficult-to-find errors as memory corruption. However, type safety comes at a cost in runtime performance due to the need for dynamic safety checks.
    [Show full text]
  • Runtime Defenses 1 Baggy Bounds Checking
    Runtime Defenses Nicholas Carlini 7 September, 2011 1 Baggy Bounds Checking 1.1 Motivation There are large amounts of legacy C programs which can not be rewritten entirely. Since memory must be manually managed in C, and bounds are not checked automatically on pointer dereferences, C often contains many memory-saftey bugs. One common type of bug is buffer overflows, where the length of an array is not checked before copying data in to the array: data is written past the end of the array and on top of whatever happens to be at that memory location. 1.2 Baggy Bounds Checking Baggy bounds checking was proposed as a possible defense which would detect when pointers wrote to a memory location out of bounds. It is a runtime defense, which uses source code instrumentation. At a very high level, baggy bounds checking keeps track of bounds information on allocated objects. When a pointer which indexes in to allocated memory is dereferenced, bounds checking code ensures that the pointer still is indexing the same allocated memory location. If the pointer points out of bounds of that object, the program is halted. Baggy bounds checking uses a buddy allocator to return objects of size 2n, and fills the remaining unused bytes with 0s. A bounds table is created containing the log of the sizes of every allocated object. Using this table allows for single-memory-access lookups, which is much more efficient than other methods which require multiple memory accesses. 1.3 Baggy Bounds Checking Failures Baggy bounds checking does not work in all situations, however.
    [Show full text]
  • Spring 2015 CS 161 Computer Security Optional Notes Memory
    CS 161 Optional Notes Spring 2015 Computer Security 1 Memory safety | Attacks and Defenses In the first few lectures we will be looking at software security|problems associated with the software implementation. You may have a perfect design, a perfect specification, perfect algorithms, but still have implementation vulnerabilities. In fact, after configuration errors, implementation errors are probably the largest single class of security errors exploited in practice. We will start by looking at a particularly prevalent class of software flaws, those that concern memory safety. Memory safety refers to ensuring the integrity of a program's data structures: preventing attackers from reading or writing to memory locations other than those intended by the programmer. Because many security-critical applications have been written in C, and because C has peculiar pitfalls of its own, many of these examples will be C-specific. Implementation flaws can in fact occur at all levels: in improper use of the programming language, the libraries, the operating system, or in the application logic. We will look at some of these others later in the course. 1 Buffer overflow vulnerabilities We'll start with one of the most common types of error|buffer overflow (also called buffer overrun) vulnerabilities. Buffer overflow vulnerabilities are a particular risk in C. Since it is an especially widely used systems programming language, you might not be surprised to hear that buffer overflows are one of the most pervasive kind of implementation flaws around. As a low-level language, we can think of C as a portable assembly language. The programmer is exposed to the bare machine (which is one reason that C is such a popular systems language).
    [Show full text]
  • Implicit Array Bounds Checking on 64-Bit Architectures
    Implicit Array Bounds Checking on 64-bit Architectures CHRIS BENTLEY, SCOTT A. WATTERSON, DAVID K. LOWENTHAL, and BARRY ROUNTREE The University of Georgia Several programming languages guarantee that array subscripts are checked to ensure they are within the bounds of the array. While this guarantee improves the correctness and security of array- based code, it adds overhead to array references. This has been an obstacle to using higher-level languages, such as Java, for high-performance parallel computing, where the language specification requires that all array accesses must be checked to ensure they are within bounds. This is because, in practice, array-bounds checking in scientific applications may increase execution time by more than a factor of 2. Previous research has explored optimizations to statically eliminate bounds checks, but the dynamic nature of many scientific codes makes this difficult or impossible. Our ap- proach is, instead, to create a compiler and operating system infrastructure that does not generate explicit bounds checks. It instead places arrays inside of Index Confinement Regions (ICRs), which are large, isolated, mostly unmapped virtual memory regions. Any array reference outside of its bounds will cause a protection violation; this provides implicit bounds checking. Our results show that when applying this infrastructure to high-performance computing programs written in Java, the overhead of bounds checking relative to a program with no bounds checks is reduced from an average of 63% to an average of 9%. Categories and Subject Descriptors: D.3.4 [Programming Languages]: Processors— Optimization General Terms: Measurement, Performance Additional Key Words and Phrases: Array-bounds checking, virtual memory, 64-bit architectures 1.
    [Show full text]
  • Ocaml Inside: a Drop-In Replacement for Libtls
    OCaml inside: a drop-in replacement for libtls Enguerrand Decorne (speaker), Jeremy Yallop, David Kaloper-Meršinjak University of Cambridge Computer Laboratory Introduction: openssl to libtls to libnqsb-tls case, these are declarations of the functions in the libtls inter- The C programming language pervades systems software. An op- face, which our library exports. The source files contain definitions erating system in the Unix tradition consists of a kernel, written of those functions, which mediate between the libtls interface in C, and a collection of libraries and executables, also written in and the ocaml-tls implementation, and which can be compiled C, which communicate in large part via APIs defined as C types and linked together with ocaml-tls, the OCaml runtime, and the and functions. Systems built in C typically suffer from a number of OCaml code that implements libnqsb-tls to build a shared li- problems, ranging from buffer overflows and other violations that brary. follow inevitably from unrestricted access to memory, to awkward let tls_server ()= APIs that result from an inexpressive type system and a lack of let tls_server= automatic memory management. { error= None; config= None; fd= None; The openssl library, which implements the cryptographic pro- state=‘NotConfigured; linger= None} in Root.create tls_server |> from_voidp tls tocols TLS and SSL, suffers from both these problems. The lack of bounds checking in C led to the notorious Heartbleed bug in let()=I.internal "tls_server" 2014; a study two years earlier found that almost no clients of (void @-> returning(ptr tls)) tls_server openssl use the library correctly, apparently due to its unhelpful interface (Georgiev et al.
    [Show full text]
  • A Comprehensive Approach to Array Bounds Check Elimination for Java
    A Comprehensive Approach to Array Bounds Check Elimination for Java Feng Qian, Laurie Hendren, and Clark Verbrugge School of Computer Science, McGill University {fqian,hendren,clump}@cs.mcgill.ca Abstract. This paper reports on a comprehensive approach to elimi- nating array bounds checks in Java. Our approach is based upon three analyses. The first analysis is a flow-sensitive intraprocedural analysis called variable constraint analysis (VCA). This analysis builds a small constraint graph for each important point in a method, and then uses the information encoded in the graph to infer the relationship between array index expressions and the bounds of the array. Using VCA as the base analysis, we also show how two further analyses can improve the results of VCA. Array field analysis is applied on each class and pro- vides information about some arrays stored in fields, while rectangular array analysis is an interprocedural analysis to approximate the shape of arrays, and is useful for finding rectangular (non-ragged) arrays. We have implemented all three analyses using the Soot bytecode opti- mization/annotation framework and we transmit the results of the anal- ysis to virtual machines using class file attributes. We have modified the Kaffe JIT, and IBM’s High Performance Compiler for Java (HPCJ) to make use of these attributes, and we demonstrate significant speedups. 1 Introduction The Java programming language is becoming increasingly popular for the im- plementation of a wide variety of application programs, including loop-intensive programs that use arrays. Java compilers translate high-level programs to Java bytecode and this bytecode is either executed by a Java virtual machine (usu- ally including a JIT compiler), or it is compiled by an ahead-of-time compiler to native code.
    [Show full text]
  • Backwards-Compatible Array Bounds Checking for C with Very Low Overhead∗
    Backwards-Compatible Array Bounds Checking for C with Very Low Overhead∗ Dinakar Dhurjati and Vikram Adve Department of Computer Science University of Illinois at Urbana-Champaign 201 N. Goodwin Ave., Urbana, IL 61801 {dhurjati,vadve}@cs.uiuc.edu ABSTRACT work on detecting array bounds violations or buffer over- The problem of enforcing correct usage of array and pointer runs, because the best existing solutions to date are either references in C and C++ programs remains unsolved. The far too expensive for use in deployed production code or approach proposed by Jones and Kelly (extended by Ruwase raise serious practical difficulties for use in real-world devel- and Lam) is the only one we know of that does not require opment situations. significant manual changes to programs, but it has extremely The fundamental difficulty of bounds checking in C and high overheads of 5x-6x and 11x–12x in the two versions. In C++ is the need to track, at run-time, the intended tar- this paper, we describe a collection of techniques that dra- get object of each pointer value (called the intended referent matically reduce the overhead of this approach, by exploit- by Jones and Kelly [10]). Unlike safe languages like Java, ing a fine-grain partitioning of memory called Automatic pointer arithmetic in C and C++ allows a pointer to be com- Pool Allocation. Together, these techniques bring the aver- puted into the middle of an array or string object and used age overhead checks down to only 12% for a set of bench- later to further index into the object.
    [Show full text]
  • Liquid Types ∗
    Liquid Types ∗ Patrick M. Rondon Ming Kawaguchi Ranjit Jhala University of California, San Diego {prondon,mwookawa,jhala}@cs.ucsd.edu Abstract The utility of these type systems stems from their ability to pre- We present Logically Qualified Data Types, abbreviated to Liquid dict, at compile-time, invariants about the run-time values com- Types, a system that combines Hindley-Milner type inference with puted by the program. Unfortunately, classical type systems only Predicate Abstraction to automatically infer dependent types pre- capture relatively coarse invariants. For example, the system can cise enough to prove a variety of safety properties. Liquid types express the fact that a variable i is of the type int, meaning that allow programmers to reap many of the benefits of dependent it is always an integer, but not that it is always an integer within a types, namely static verification of critical properties and the elim- certain range, say between 1 and 99. Thus, the type system is un- ination of expensive run-time checks, without the heavy price of able to statically ensure the safety of critical operations, such as a manual annotation. We have implemented liquid type inference in division by i, or the accessing of an array a of size 100 at an index i. Instead, the language can only provide a weaker dynamic safety DSOLVE, which takes as input an OCAML program and a set of log- ical qualifiers and infers dependent types for the expressions in the guarantee at the additional cost of high performance overhead. In an exciting development, several authors have proposed the OCAML program.
    [Show full text]
  • Coverity Support for SEI CERT C, C++, and Java Coding Standards
    Coverity Support for SEI CERT C, C++, and Java Coding Standards Ensure the safety, The SEI CERT C, C++, and Oracle Java Coding Standards are lists of rules and reliability, and security recommendations for writing secure code in the C, C++, and Java programming languages They represent an important milestone in introducing best practices for of software written in C, ensuring the safety, reliability, security, and integrity of software written in C/C++ and C++, and Java Java Notably, the standards are designed to be enforceable by software code analyzers using static analysis techniques This greatly reduces the cost of compliance by way of automation Adhering to coding standards is a crucial step in establishing best coding practices Standards adherence is particularly important in safety-critical, high-impact industries, such as automotive, medical, and networking Software defects in products coming from these industries manifest themselves physically and tangibly—often with life- threatening consequences Synopsys provides a comprehensive solution for the SEI CERT C/C++ Coding Standards rules, along with high-impact SEI CERT Oracle Java Coding Standards (online version) rules and SEI CERT C Coding Standard recommendations (online version) Coverity static analysis implements the Rules category within the CERT C/ C++ standards, high-impact CERT Java L1 rules, and methods for managing violations and reporting on them Coverity also supports some of the best practices from the Recommendations category for the CERT C standard Acknowledgement
    [Show full text]
  • Checking Array Bounds by Abstract Interpretation and Symbolic Expressionsétienne Fausto Spoto, Etienne Payet
    Checking Array Bounds by Abstract Interpretation and Symbolic ExpressionsÉtienne Fausto Spoto, Etienne Payet To cite this version: Fausto Spoto, Etienne Payet. Checking Array Bounds by Abstract Interpretation and Symbolic Ex- pressionsÉtienne. 9th International Joint Conference on Automated Reasoning (IJCAR’18), Jul 2018, Oxford, United Kingdom. pp.706-722, 10.1007/978-3-319-94205-6_46. hal-01921666 HAL Id: hal-01921666 https://hal.univ-reunion.fr/hal-01921666 Submitted on 14 Nov 2018 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Checking Array Bounds by Abstract Interpretation and Symbolic Expressions Etienne´ Payet1 and Fausto Spoto2 1 Laboratoire d'Informatique et de Math´ematiques, Universit´ede la R´eunion,France 2 Dipartimento di Informatica, Universit`adi Verona, Italy Abstract. Array access out of bounds is a typical programming er- ror. From the '70s, static analysis has been used to identify where such errors actually occur at runtime, through abstract interpretation into linear constraints. However, feasibility and scalability to modern object- oriented code has not been established yet. This article builds on previ- ous work on linear constraints and shows that the result does not scale, when polyhedra implement the linear constraints, while the more ab- stract zones scale to the analysis of medium-size applications.
    [Show full text]