CMSC 426/626: Secure Coding Krishna M. Sivalingam Sources: From Secure Coding, Mark and van Wyk, O’Reilly, 2003 www.cert.org/secure-coding Where can errors occur? During entire software lifecycle Security Architecture/Design stage Man-in-the-middle attack Race condition attack Replay attack Implementation Stage Buffer overflow attack Parsing error attack Back door attacks (aka Trapdoors) Code Maintenance Stage Flaw Classifications Landwehr’s Scheme Bishop’s Scheme Aslam’s Scheme Du/Mathur’s classification Flaws are Intentional and Inadvertent Inadvertent Flaw Classifications Validation Error Domain Error Serialization and Aliasing Inadequate Authentication and Identification Boundary Condition Violation Other exploitable logic error Study of Buffer Overflow Attack Cowan, Crispin, Perry Wagle, Calton Pu, Steve Beattie, and Jonathan Walpole. "Buffer Overflows: Attacks and Defenses for the Vulnerability of the Decade." Proceedings of DARPA Information Survivability Conference and Expo (DISCEX), 1999 http://insecure.org/stf/mudge_buffer_overflow_tutorial.html Buffer Overflows Inject attack code by overflowing the buffer Usually involves adding code based on target machines’ CPU opcodes Execute code with all the privileges of the vulnerable program Thus, if program is running as root, attacker can run at will any code as root Typically, manage to invoke execve /bin/sh or similar to get a root shell Program Segments An executing program consists of: Code Initialized Data Global variables Stack Heap (for dynamic allocation) Remember that local variables, return address, etc. are stored in the stack when a function is invoked When a local variable is over-run, it can alter return address, etc. Where to Inject Code On the stack (automatic variables) On the heap (malloc or calloc variables) In static data areas Executable code need not be restricted to the overflowing buffer – code can be injected elsewhere One can also use existing code For example, if exec(arg) exists in program, modify running code by making arg point to “/bin/sh” Jump to Attacker’s Code Activation Record Overflow into return address on the stack and make it point at the code. Function pointers Overflow into “void (*foo())()” and it point at the code Setjmp and longjmp commands, that are used for checkpointing and recovery Alter address given to longjmp to point to attacker’s code Buffer Overflow Details Look at Mudge’s sample buffer overflow attack Buffer Overflow Defenses Writing Correct Code Vulnerable programs continue to emerge on a regular basis C has many error-prone idioms and a culture that favors performance over correctness. Static Analysis Tools Fortify – looks for vulnerable constructs Too many false positives From Crispin Cowan’s SANS 2000 Talk on Web Buffer Overflow Defenses Non-executable buffers Non executable data segments Optimizing compiles emit code into program data segments Non executable stack segments Highly effective against code injection on the stack but not against code injections on the heap or static variables. Buffer Overflow Defenses Array Bound Checking Can run 12x-30x slower a[3] is checked but *(a+3) is not Type safe languages: Java or ML There are millions of lines of C code in operating systems and security system applications Attack the Java Virtual Machine which is a C program StackGuard program: Adds a “canary” value, which is a 32-bit random # or a known string terminator (CR, LF, ‘\0’, etc.) Compiler adds canary and system can check for this value at runtime Entire RedHat system has been recompiled with this and shown to be less vulnerable Race Conditions http://seclab.cs.ucdavis.edu/projects/vulnerabilities/s criv/ucd-ecs-95-08.pdf http://citeseer.ist.psu.edu/bishop96checking.html http://www.mirrors.wiretapped.net/security/develo pment/secure-programming/bishop-dilger-1996checking-for-race-conditions-in-file-accesses.pdf Race condition: What is it? Consider a setuid program, owned by root UserA is presently executing the program, hence is running it as root Assume that the program wants to write to a file. The system must check whether UserA has the right privileges on this file, checked as follows: if (access(filename, W_OK) == 0){ if ((fd = open(filename, O_WRONLY)) == NULL){ perror(filename); return(0); } /* now write to the file */ Race condition: What is it? In the time between verifying access and opening the file, if the file referred to changes, then its access will not have been checked Called TOCTTOU (Time-of-check-To-Time-ofUse) binding flaw For example, if access is originally checked on /tmp/X AND before execution of write statement: /tmp/X is deleted AND Hard link from /etc/passwd is created to /tmp/X Then, process will write to /etc/passwd! Present in xterm program, while logging sessions Source: Bishop and Dilger’s 1996 paper in Computing Systems Race conditions, contd. Similar attack possible on binmail program Binmail appends mail to an existing mail spool file E.g. /usr/spool/mail/jkl Binmail verifies if file exists (and is not a symbolic link) Before binmail writes to file, jkl is deleted AND made a hard link to /etc/passwd Now, binmail appends data to /etc/passwd Attacker can create a new account with no password and root privileges Note that binding flaws do not arise when file descriptors are used! Good Practices in Implementation Inform Yourself Follow Vulnerability Discussions and Alerts (eg. www.cert.org) Read books and papers on secure coding practices, analyses of software flaws, etc. Explore open source software Examples of how to and how not to write code Good Practices in Implementation Handle Data with Caution Cleanse data: Examine input data for malicious intent (altering character sets, using dis-allowed characters) Perform bounds checking Check array indices Check configuration files Can be modified by attacker Check command-line parameters Don’t trust web URLs and parameters within Be careful of web content (variables hidden in HTML fields) Good Practices in Implementation Check web cookies Check environment variables Set valid initial values for data Understand filename references and use them correctly Check for indirect file references (e.g. Shortcuts, symbolic links) Be careful of how program and data files are located (as in searching using PATH variable) Reuse “Good” Code whenever Practical Good Practices in Implementation Sound Review Processes Perform Peer review of Code Perform Independent Validation and Verification Use automated security tools Static Code checkers RATS - Rough Auditing Tool for Security SPLINT – Source code scanner http://splint.org/ Uno: http://spinroot.com/uno/ Runtime checkers Libsafe: http://directory.fsf.org/libsafe.html PurifyPlus: http://www-306.ibm.com/software/awdtools/purifyplus/ Immunix Tools: Good Practices in Implementation Profiling Tools Papillon for Solaris: http://www.roqe.org/papillon/ Gprof from GNU Janus – policy enforcement and profiling; http://www.cs.berkeley.edu/~daw/janus/ Black-box Testing for Fault-Injection Tools Appscan: http://www.watchfire.com/securityzone/default.aspx Whisker: wiretrip.net ISS Database Scanner: http://www.iss.net/ Perform network-based vulnerability scans Nmap: http://insecure.org/nmap/ Nessus: http://www.nessus.org/ ISS Internet Scanner Good Practices in Implementation Make Generous Use of Checklists Security checklists must be created and checked against. For example: Application requires password for access All user ID logins are unique Uses role-based access control Encryption is used Code should be Maintainable Practice standards of in-line documentation Remove obsolete code Test all code changes Implementation, Don’ts Don’t write code that uses relative filenames Fully qualified filenames should be used Don’t refer to a file twice in the same program by its name Always use file descriptors after initial open Prevents “race condition attack” that exploit time between access check and file execution Don’t invoke untrusted programs from within trusted ones Avoid using setuid or similar mechanisms whenever possible Don’t assume that users are not malicious Implementation, Don’ts Don’t dump core – code must fail gracefully Coredump can be used to extract valuable data stored in memory during execution Don’t assume that a system call (or any function call) is always successful – always check for return values and error variable values Computer-based random number generators are “pseudo-random” and can have repitition Don’t invoke shell or command line from within a program Don’t use world writable storage, even for temporary files Implementation, Don’ts Don’t trust user-writable storage not to be tampered with Don’t keep sensitive data in a database without password protection Don’t code usernames/passwords into an application Don’t echo passwords! Don’t rely on host-level file protection mechanisms Don’t make access decisions based on environment variables or command-line arguments Don’t issue passwords via email To be Continued