Stelios Sidiroglou-Douskos

Managing performance vs. accuracy trade-offs with loop perforation

Many modern computations (such as video and audio encoders, Monte Carlo simulations, and machine learning algorithms) are designed to trade off accuracy in return for increased performance. To date, such computations typically use ad-hoc, domain-specific techniques developed specifically for the computation at hand. Loop perforation provides a general technique to trade accuracy for performance by transforming loops to execute a subset of their iterations. A criticality testing phase filters out critical loops (whose perforation produces unacceptable behavior) to identify tunable loops (whose perforation produces more efficient and still acceptably accurate computations). A perforation space exploration algorithm perforates combinations of tunable loops to find Pareto-optimal perforation policies. Our results indicate that, for a range of applications, this approach typically delivers performance increases of over a factor of two (and up to a factor of seven) while changing the result that the application produces by less than 10%.

Control Jujutsu: On the Weaknesses of Fine-Grained Control Flow Integrity

Control flow integrity (CFI) has been proposed as an approach to defend against control-hijacking memory corruption attacks. CFI works by assigning tags to indirect branch targets statically and checking them at runtime. Coarse-grained enforcements of CFI that use a small number of tags to improve the performance overhead have been shown to be ineffective. As a result, a number of recent efforts have focused on fine-grained enforcement of CFI as it was originally proposed. In this work, we show that even a fine-grained form of CFI with unlimited number of tags and a shadow stack (to check calls and returns) is ineffective in protecting against malicious attacks. We show that many popular code bases such as Apache and Nginx use coding practices that create flexibility in their intended control flow graph (CFG) even when a strong static analyzer is used to construct the CFG. These flexibilities allow an attacker to gain control of the execution while strictly adhering to a fine-grained CFI. We then construct two proof-of-concept exploits that attack an unlimited tag CFI system with a shadow stack. We also evaluate the difficulties of generating a precise CFG using scalable static analysis for real-world applications. Finally, we perform an analysis on a number of popular applications that highlights the availability of such attacks.

Automatic runtime error repair and containment via recovery shepherding

We present a system, RCV, for enabling software applications to survive divide-by-zero and null-dereference errors. RCV operates directly on off-the-shelf, production, stripped x86 binary executables. RCV implements recovery shepherding, which attaches to the application process when an error occurs, repairs the execution, tracks the repair effects as the execution continues, contains the repair effects within the application process, and detaches from the process after all repair effects are flushed from the process state. RCV therefore incurs negligible overhead during the normal execution of the application. We evaluate RCV on all divide-by-zero and null-dereference errors available in the CVE database [2] from January 2011 to March 2013 that 1) provide publicly-available inputs that trigger the error which 2) we were able to use to trigger the reported error in our experimental environment. We collected a total of 18 errors in seven real world applications, Wireshark, the FreeType library, Claws Mail, LibreOffice, GIMP, the PHP interpreter, and Chromium. For 17 of the 18 errors, RCV enables the application to continue to execute to provide acceptable output and service to its users on the error-triggering inputs. For 13 of the 18 errors, the continued RCV execution eventually flushes all of the repair effects and RCV detaches to restore the application to full clean functionality. We perform a manual analysis of the source code relevant to our benchmark errors, which indicates that for 11 of the 18 errors the RCV and later patched versions produce identical or equivalent results on all inputs.

Automatic error elimination by horizontal code transfer across multiple applications

We present Code Phage (CP), a system for automatically transferring correct code from donor applications into recipient applications that process the same inputs to successfully eliminate errors in the recipient. Experimental results using seven donor applications to eliminate ten errors in seven recipient applications highlight the ability of CP to transfer code across applications to eliminate out of bounds access, integer overflow, and divide by zero errors. Because CP works with binary donors with no need for source code or symbolic information, it supports a wide range of use cases. To the best of our knowledge, CP is the first system to automatically transfer code across multiple applications.

Missing the Point(er): On the Effectiveness of Code Pointer Integrity

Memory corruption attacks continue to be a major vector of attack for compromising modern systems. Numerous defenses have been proposed against memory corruption attacks, but they all have their limitations and weaknesses. Stronger defenses such as complete memory safety for legacy languages (C/C++) incur a large overhead, while weaker ones such as practical control flow integrity have been shown to be ineffective. A recent technique called code pointer integrity (CPI) promises to balance security and performance by focusing memory safety on code pointers thus preventing most control-hijacking attacks while maintaining low overhead. CPI protects access to code pointers by storing them in a safe region that is protected by instruction level isolation. On x86-32, this isolation is enforced by hardware, on x86-64 and ARM, isolation is enforced by information hiding. We show that, for architectures that do not support segmentation in which CPI relies on information hiding, CPIs safe region can be leaked and then maliciously modified by using data pointer overwrites. We implement a proof-of-concept exploit against Nginx and successfully bypass CPI implementations that rely on information hiding in 6 seconds with 13 observed crashes. We also present an attack that generates no crashes and is able to bypass CPI in 98 hours. Our attack demonstrates the importance of adequately protecting secrets in security mechanisms and the dangers of relying on difficulty of guessing without guaranteeing the absence of memory leaks.

Sound input filter generation for integer overflow errors

We present a system, SIFT, for generating input filters that nullify integer overflow errors associated with critical program sites such as memory allocation or block copy sites. SIFT uses a static pro- gram analysis to generate filters that discard inputs that may trigger integer overflow errors in the computations of the sizes of allocated memory blocks or the number of copied bytes in block copy operations. Unlike all previous techniques of which we are aware, SIFT is sound -- if an input passes the filter, it will not trigger an integer overflow error at any analyzed site. Our results show that SIFT successfully analyzes (and therefore generates sound input filters for) 56 out of 58 memory allocation and block memory copy sites in analyzed input processing modules from five applications (VLC, Dillo, Swfdec, Swftools, and GIMP). These nullified errors include six known integer overflow vulnerabilities. Our results also show that applying these filters to 62895 real-world inputs produces no false positives. The analysis and filter generation times are all less than a second.

Targeted Automatic Integer Overflow Discovery Using Goal-Directed Conditional Branch Enforcement

We present a new technique and system, DIODE, for auto- matically generating inputs that trigger overflows at memory allocation sites. DIODE is designed to identify relevant sanity checks that inputs must satisfy to trigger overflows at target memory allocation sites, then generate inputs that satisfy these sanity checks to successfully trigger the overflow. DIODE works with off-the-shelf, production x86 binaries. Our results show that, for our benchmark set of applications, and for every target memory allocation site exercised by our seed inputs (which the applications process correctly with no overflows), either 1) DIODE is able to generate an input that triggers an overflow at that site or 2) there is no input that would trigger an overflow for the observed target expression at that site.

Program fracture and recombination for efficient automatic code reuse

We present a new code transfer technique, program fracture and recombination, for automatically replacing, deleting, and/or combining code from multiple applications. Benefits include automatic generation of new applications incorporating the best or most desirable functionality developed anywhere, the automatic elimination of errors and security vulnerabilities, effective software rejuvenation, the automatic elimination of obsolete or undesirable functionality, and improved performance, energy efficiency, simplicity, analyzability, and clarity. The technique may be particularly appropriate for high performance computing. The field has devoted years of effort to developing efficient (but complex) implementations of standard linear algebra operations with good numerical properties. At the same time these operations also have very simple but inefficient implementations, often with poor numerical properties. Program fracture and recombination allows developers to work with the simple implementation during development and testing, then use program fracture and recombination to automatically find and deploy the most appropriate implementation for the hardware platform at hand. The benefits include reduced implementation effort, increased code clarity, and the ability to automatically search for and find efficient implementations with good numerical properties.

AutoRand: Automatic Keyword Randomization to Prevent Injection Attacks

AutoRand automatically transforms Java applications to use SQL keyword randomization to defend against SQL injection vulnerabilities. AutoRand is completely automatic. Unlike previous approaches it requires no manual modifications to existing code and does not require source it works directly on Java bytecode. It can thus easily be applied to the large numbers of existing potentially insecure applications without developer assistance. Our key technical innovation is augmented strings. Augmented strings allow extra information such as random keys to be embedded within a string. AutoRand transforms string operations so that the extra information is transparent to the program, but is always propagated with each string operation. AutoRand checks each keyword at SQL statements for the random key. Experimental results on large, production Java applications and malicious inputs provided by an independent evaluation team hired by an agency of the United States government showed that AutoRand successfully blocked all SQL injection attacks and preserved transparent execution for benign inputs, all with low overhead.

An adversarial evaluation of network signaling and control mechanisms

Network signaling and control mechanisms are critical to coordinate such diverse defense capabilities as honeypots and honeynets, host-based defenses, and online patching systems, any one of which might issue an actionable alert and provide security-critical data. Despite considerable work in exploring the trust requirements of such defenses and in addressing the distribution speed of alerts, little work has gone into identifying how the underlying transport systems behave under adversarial scenarios. In this paper, we evaluate the reliability and performance trade-offs for a variety of control channel mechanisms that are suitable for coordinating large-scale collaborative defenses when under attack. Our results show that the performance and reliability characteristics change drastically when one evaluates the systems under attack by a sophisticated and targeted adversary. Based on our evaluation, we explore available design choices to reinforce the reliability of the control channel mechanisms. To that end, we propose ways to construct a control scheme to improve network coverage without imposing additional overhead.