The root causes of many security vulnerabilities include a pernicious combination of two problems, often regarded as inescapable aspects of computing. First, the protection mechanisms provided by the mainstream processor architecture and C/C++ language abstractions, dating back to the 1970s and before, provide only coarse-grain virtual-memory-based protection. Second, mainstream system engineering relies almost exclusively on test-and-debug methods, with (at best) prose specifications. These methods have historically sufficed commercially for much of the computer industry, but they fail to prevent large numbers of exploitable bugs, and the security problems that this causes are becoming ever more acute.
In this paper we show how more rigorous engineering methods can be applied to the development of a new security-enhanced processor architecture, with its accompanying hardware implementation and software stack. We use formal models of the complete instruction-set architecture (ISA) at the heart of the design and engineering process, both in lightweight ways that support and improve normal engineering practice -- as documentation, in emulators used as a test oracle for hardware and for running software, and for test generation -- and for formal verification. We formalise key intended security properties of the design, and establish that these hold with mechanised proof. This is for the same complete ISA models (complete enough to boot operating systems), without idealisation.
We do this for CHERI, an architecture with hardware capabilities that supports fine-grained memory protection and scalable secure compartmentalisation, while offering a smooth adoption path for existing software. CHERI is a maturing research architecture, developed since 2010, with work now underway to explore its possible adoption in mass-market commercial processors. The rigorous engineering work described here has been an integral part of its development to date, enabling more rapid and confident experimentation, and boosting confidence in the design.
C remains central to our infrastructure, making verification of C code an essential and much-researched topic, but the semantics of C is remarkably complex, and important aspects of it are still unsettled, leaving programmers and verification tool builders on shaky ground.
This paper describes a tool, Cerberus-BMC, that for the first time provides a principled reference semantics that simultaneously supports (1) a choice of concurrency memory model (including substantial fragments of the C11, RC11, and Linux kernel memory models), (2) a modern memory object model, and (3) a well-validated thread-local semantics for a large fragment of the language. The tool should be useful for C programmers, compiler writers, verification tool builders, and members of the C/C++ standards committees.
This technical report describes CHERI ISAv7, the seventh version of the Capability Hardware Enhanced RISC Instructions (CHERI) Instruction-Set Architecture (ISA) being developed by SRI International and the University of Cambridge. This design captures nine years of research, development, experimentation, refinement, formal analysis, and validation through hardware and software implementation. CHERI ISAv7 is a substantial enhancement to prior ISA versions. We differentiate an architecture-neutral protection model vs. architecture-specific instantiations in 64-bit MIPS, 64-bit RISC-V, and x86-64. We have defined a new CHERI Concentrate compression model. CHERI-RISC-V is more substantially elaborated. A new compartment-ID register assists in resisting microarchitectural side-channel attacks. Experimental features include linear capabilities, capability coloring, temporal memory safety, and 64-bit capabilities for 32-bit architectures.
CHERI is a hybrid capability-system architecture that adds new capability-system primitives to commodity 64-bit RISC ISAs, enabling software to efficiently implement fine-grained memory protection and scalable software compartmentalization. Design goals include incremental adoptability within current ISAs and software stacks, low performance overhead for memory protection, significant performance improvements for software compartmentalization, formal grounding, and programmer-friendly underpinnings. We have focused on providing strong, non-probabilistic, efficient architectural foundations for the principles of least privilege and intentional use in the execution of software at multiple levels of abstraction, preventing and mitigating vulnerabilities.
The CHERI system architecture purposefully addresses known performance and robustness gaps in commodity ISAs that hinder the adoption of more secure programming models centered around the principle of least privilege. To this end, CHERI blends traditional paged virtual memory with an in-address-space capability model that includes capability registers, capability instructions, and tagged memory. CHERI builds on the C-language fat-pointer literature: its capabilities can describe fine-grained regions of memory, and can be substituted for data or code pointers in generated code, protecting data and also improving control-flow robustness. Strong capability integrity and monotonicity properties allow the CHERI model to express a variety of protection properties, from enforcing valid C-language pointer provenance and bounds checking to implementing the isolation and controlled communication structures required for software compartmentalization.
CHERI's hybrid capability-system approach, inspired by the Capsicum security model, allows incremental adoption of capability-oriented design: software implementations that are more robust and resilient can be deployed where they are most needed, while leaving less critical software largely unmodified, but nevertheless suitably constrained to be incapable of having adverse effects. Potential deployment scenarios include low-level software Trusted Computing Bases (TCBs) such as separation kernels, hypervisors, and operating-system kernels, as well as userspace TCBs such as language runtimes and web browsers. We also see potential early-use scenarios around particularly high-risk software libraries (such as data compression, protocol parsing, and image processing), which are concentrations of both complex and historically vulnerability-prone code exposed to untrustworthy data sources, while leaving containing applications unchanged.
The CHERI architecture allows pointers to be implemented as capabilities (rather than integer virtual addresses) in a manner that is compatible with, and strengthens, the semantics of the C language. In addition to the spatial protections offered by conventional fat pointers, CHERI capabilities offer strong integrity, enforced provenance validity, and access monotonicity. The stronger guarantees of these architectural capabilities must be reconciled with the real-world behavior of operating systems, run-time environments, and applications. When the process model, user-kernel interactions, dynamic linking, and memory management are all considered, we observe that simple derivation of architectural capabilities is insufficient to describe appropriate access to memory. We bridge this conceptual gap with a notional abstract capability that describes the accesses that should be allowed at a given point in execution, whether in the kernel or userspace. To investigate this notion at scale, we describe the first adaptation of a full C-language operating system (FreeBSD) with an enterprise database (PostgreSQL) for complete spatial and referential memory safety. We show that awareness of abstract capabilities, coupled with CHERI architectural capabilities, can provide more complete protection, strong compatibility, and acceptable performance overhead compared with the pre-CHERI baseline and software-only approaches. Our observations also have potentially significant implications for other mitigation techniques.
We are building accurate full-scale mathematical models of some of the key computational abstractions (processor architectures, programming languages, concurrent OS interfaces, and network protocols), studying how this can best be done, and investigating how such models can be used for new verification research and in new systems and programming language research. For many of these abstractions, our work has exposed and clarified fundamental questions about what the abstractions are, and provided tools to let them be explored. Supporting all this, we are also developing new specification tools. Most of our models and tools are publicly available under permissive open-source licences.
Architecture specifications notionally define the fundamental interface between hardware and software: the envelope of allowed behaviour for processor implementations, and the basic assumptions for software development and verification. But in practice, they are typically prose and pseudocode documents, not rigorous or executable artifacts, leaving software and verification on shaky ground.
In this paper, we present rigorous semantic models for the sequential behaviour of large parts of the mainstream ARMv8-A, RISC-V, and MIPS architectures, and the research CHERI-MIPS architecture, that are complete enough to boot operating systems, variously Linux, FreeBSD, or seL4. Our ARMv8-A models are automatically translated from authoritative ARM-internal definitions, and (in one variant) tested against the ARM Architecture Validation Suite.
We do this using a custom language for ISA semantics, Sail, with a lightweight dependent type system, that supports automatic generation of emulator code in C and OCaml, and automatic generation of proof-assistant definitions for Isabelle, HOL4, and (currently only for MIPS) Coq. We use the former for validation, and to assess specification coverage. To demonstrate the usability of the latter, we prove (in Isabelle) correctness of a purely functional characterisation of ARMv8-A address translation. We moreover integrate the RISC-V model into the RMEM tool for (user-mode) relaxed-memory concurrency exploration. We prove (on paper) the soundness of the core Sail type system.
We thereby take a big step towards making the architectural abstraction actually well-defined, establishing foundations for verification and reasoning.
The semantics of pointers and memory objects in C has been a vexed question for many years. C values cannot be treated as either purely abstract or purely concrete entities: the language exposes their representations, but compiler optimisations rely on analyses that reason about provenance and initialisation status, not just runtime representations. The ISO WG14 standard leaves much of this unclear, and in some respects differs with de facto standard usage --- which itself is difficult to investigate.
In this paper we explore the possible source-language semantics for memory objects and pointers, in ISO C and in C as it is used and implemented in practice, focussing especially on pointer provenance. We aim to, as far as possible, reconcile the ISO C standard, mainstream compiler behaviour, and the semantics relied on by the corpus of existing C code. We present two coherent proposals, tracking provenance via integers and not; both address many design questions. We highlight some pros and cons and open questions, and illustrate the discussion with a library of test cases. We make our semantics executable as a test oracle, integrating it with the Cerberus semantics for much of the rest of C, which we have made substantially more complete and robust, and equipped with a web-interface GUI. This allows us to experimentally assess our proposals on those test cases. To assess their viability with respect to larger bodies of C code, we analyse the changes required and the resulting behaviour for a port of FreeBSD to CHERI, a research architecture supporting hardware capabilities, which (roughly speaking) traps on the memory safety violations which our proposals deem undefined behaviour. We also develop a new runtime instrumentation tool to detect possible provenance violations in normal C code, and apply it to some of the SPEC benchmarks. We compare our proposal with a source-language variant of the twin-allocation LLVM semantics proposal of Lee et al. Finally, we describe ongoing interactions with WG14, exploring how our proposals could be incorporated into the ISO standard.
Conventional computer engineering relies on test-and-debug development processes, with the behaviour of common interfaces described (at best) with prose specification documents. But prose specifications cannot be used in test-and-debug development in any automated way, and prose is a poor medium for expressing complex (and loose) specifications.
The TCP/IP protocols and Sockets API are a good example of this: they play a vital role in modern communication and computation, and interoperability between implementations is essential. But what exactly they are is surprisingly obscure: their original development focussed on “rough consensus and running code”, augmented by prose RFC specifications that do not precisely define what it means for an implementation to be correct. Ultimately, the actual standard is the de facto one of the common implementations, including, for example, the 15000--20000 lines of the BSD implementation --- optimised and multithreaded C code, time-dependent, with asynchronous event handlers, intertwined with the operating system, and security-critical.
This paper reports on work done in the Netsem project to develop lightweight mathematically rigorous techniques that can be applied to such systems: to specify their behaviour precisely (but loosely enough to permit the required implementation variation) and to test whether these specifications and the implementations correspond, with specifications that are executable as test oracles. We developed post-hoc specifications of TCP, UDP, and the Sockets API, both of the service that they provide to applications (in terms of TCP bidirectional stream connections), and of the internal operation of the protocol (in terms of TCP segments and UDP datagrams), together with a testable abstraction function relating the two. These specifications are rigorous, detailed, readable, with broad coverage, and are rather accurate. Working within a general-purpose proof assistant (HOL4), we developed language idioms (within higher-order logic) in which to write the specifications: operational semantics with nondeterminism, time, system calls, monadic relational programming, etc. We followed an experimental semantics approach, validating the specifications against several thousand traces captured from three implementations (FreeBSD, Linux, and WinXP). Many differences between these were identified, and a number of bugs. Validation was done using a special-purpose symbolic model checker programmed above HOL4.
Having demonstrated that our logic-based engineering techniques suffice for handling real-world protocols, we argue that similar techniques could be applied to future critical software infrastructure at design time, leading to cleaner designs and (via specification-based testing) more robust and predictable implementations. In cases where specification looseness can be controlled, this should be possible with lightweight techniques, without the need for a general-purpose proof assistant, at relatively little cost.
ARM has a relaxed memory model, previously specified in informal prose for ARMv7 and ARMv8. Over time, and partly due to work building formal semantics for ARM concurrency, it has become clear that some of the complexity of the model is not justified by the potential benefits. In particular, the model was originally non-multicopy-atomic: writes could become visible to some other threads before becoming visible to all --- but this has not been exploited in production implementations, the corresponding potential hardware optimisations are thought to have insufficient benefits in the ARM context, and it gives rise to subtle complications when combined with other ARMv8 features. The ARMv8 architecture has therefore been revised: it now has a multicopy-atomic model. It has also been simplified in other respects, including more straightforward notions of dependency, and the architecture now includes a formal concurrency model.
In this paper we detail these changes and discuss their motivation. We define two formal concurrency models: an operational one, simplifying the Flowing model of Flur et al., and the axiomatic model of the revised ARMv8 specification. The models were developed by an academic group and by ARM staff, respectively, and this extended collaboration partly motivated the above changes. We prove the equivalence of the two models. The operational model is integrated into an executable exploration tool with new web interface, demonstrated by exhaustively checking the possible behaviours of a loop-unrolled version of a Linux kernel lock implementation, a previously known bug due to unprevented speculation, and a fixed version.
Previous work on the semantics of relaxed shared-memory concurrency has only considered the case in which each load reads the data of exactly one store. In practice, however, multiprocessors support mixed-size accesses, and these are used by systems software and (to some degree) exposed at the C/C++ language level. A semantic foundation for software, therefore, has to address them.
We investigate the mixed-size behaviour of ARMv8 and IBM POWER architectures and implementations: by experiment, by developing semantic models, by testing the correspondence between these, and by discussion with ARM and IBM staff. This turns out to be surprisingly subtle, and on the way we have to revisit the fundamental concepts of coherence and sequential consistency, which change in this setting. In particular, we show that adding a memory barrier between each instruction does not restore sequential consistency. We go on to extend the C/C++11 model to support non-atomic mixed-size memory accesses.
This is a necessary step towards semantics for real-world shared-memory concurrent code, beyond litmus tests.
The C/C++11 concurrency model balances two goals: it is relaxed enough to be efficiently implementable and (leaving aside the “thin-air” problem) it is strong enough to give useful guarantees to programmers. It is mathematically precise and has been used in verification research and compiler testing. However, the model is expressed in an axiomatic style, as predicates on complete candidate executions. This suffices for computing the set of allowed executions of a small litmus test, but it does not directly support the incremental construction of executions of larger programs. It is also at odds with conventional operational semantics, as used implicitly in the rest of the C/C++ standards.
Our main contribution is the development of an operational model for C/C++11 concurrency. This covers all the features of the previous formalised axiomatic model, and we have a mechanised proof that the two are equivalent, in Isabelle/HOL. We also integrate this semantics with an operational semantics for sequential C (described elsewhere); the combined semantics can incrementally execute programs in a small fragment of C.
Doing this uncovered several new aspects of the C/C++11 model: we show that one cannot build an equivalent operational model that simply follows program order, sequential consistent order, or the synchronises-with order. The first negative result is forced by hardware-observable behaviour, but the latter two are not, and so might be ameliorated by changing C/C++11. More generally, we hope that this work, with its focus on incremental construction of executions, will inform the future design of new concurrency models.
Beneath the surface, software usually depends on complex linker behaviour to work as intended. Even linking hello_world.c is surprisingly involved, and systems software such as libc and operating system kernels rely on a host of linker features. But linking is poorly understood by working programmers and has largely been neglected by language researchers.
In this paper we survey the many use-cases that linkers support and the poorly specified linker speak by which they are controlled: metadata in object files, command-line options, and linker-script language. We provide the first validated formalisation of a realistic executable and linkable format (ELF), and capture aspects of the Application Binary Interfaces for four mainstream platforms (AArch64, AMD64, Power64, and IA32). Using these, we develop an executable specification of static linking, covering (among other things) enough to link small C programs (we use the example of bzip2) into a correctly running executable. We provide our specification in Lem and Isabelle/HOL forms. This is the first formal specification of mainstream linking. We have used the Isabelle/HOL version to prove a sample correctness property for one case of AMD64 ABI relocation, demonstrating that the specification supports formal proof, and as a first step towards the much more ambitious goal of verified linking. Our work should enable several novel strands of research, including linker-aware verified compilation and program analysis, and better languages for controlling linking.
C remains central to our computing infrastructure. It is notionally defined by ISO standards, but in reality the properties of C assumed by systems code and those implemented by compilers have diverged, both from the ISO standards and from each other, and none of these are clearly understood.
We make two contributions to help improve this error-prone situation. First, we describe an in-depth analysis of the design space for the semantics of pointers and memory in C as it is used in practice. We articulate many specific questions, build a suite of semantic test cases, gather experimental data from multiple implementations, and survey what C experts believe about the de facto standards. We identify questions where there is a consensus (either following ISO or differing) and where there are conflicts. We apply all this to an experimental C implemented above capability hardware. Second, we describe a formal model, Cerberus, for large parts of C. Cerberus is parameterised on its memory model; it is linkable either with a candidate de facto memory object model, under construction, or with an operational C11 concurrency model; it is defined by elaboration to a much simpler Core language for accessibility, and it is executable as a test oracle on small examples.
This should provide a solid basis for discussion of what mainstream C is now: what programmers and analysis tools can assume and what compilers aim to implement. Ultimately we hope it will be a step towards clear, consistent, and accepted semantics for the various use-cases of C.