6.resource exhaustion

Course 1: Overview of Secure Programming, Section 6Pascal Meunier, Ph.D., M.Sc., CISSPMay 2004; updated August 12, 2004Developed thanks to support and contributions from Symantec Corporation, support from the NSF SFS Capacity Building Program (Award Number 0113725) and the Purdue e-Enterprise CenterCopyright (2004) Purdue Research Foundation. All rights reserved.

Course 1 Learning Plan

Security overview and patching Public vulnerability databases and resources Secure software engineering Security assessment and testing Shell and environment Resource management Trust management

Learning objectives

Be able to identify resources at risk Understand how resources become at risk from

denial-of-service attacks Be able to decide which resources need to be

exposed Understand how to mitigate resource exhaustion

risks

Resource Management: Outline

Motivation Resource identification Resource exhaustion

– CPU exhaustion – Network applications and protocols vulnerabilities– Generous Protocols and Algorithms– Other asymmetric attacks

Memory Management

How Important is Availability?

How important is it to have resources available at some specific times or all the time?

Market for 99.99% availability systems or even "5 nines" (99.999%)– Worth a lot of money for some businesses– Redundant hardware, fault-resistant software

Problem: Infrastructure, hardware, software usually designed for functionality and performance in normal situations, not robustness vs. worst-case scenarios– Malicious people can engineer worst-case scenarios to

come true

Availability Through Software

Monitoring software that relaunches an application whenever it crashes or quits

Redundant software installations on different partitions

Disk images (e.g., Ghost) Virtual machines that reload running images

(VMWare) The above are just mitigating setups that don't fix

the original problem and can still cause interruptions– Perfect software is not possible, but better software is

Denial of Service

The unavailability of a needed resource, most often due to a malicious entity.

Sometimes not the primary goal– Side effect of another attack– Failure to achieve worse results

Perhaps used to evade traceability, law enforcement or accountability– disable logging mechanisms– disable detection and alert systems

submerge human with messages, human may disable the attacked defense mechanism!

Resource Identification

Shared resources are exposed to resource exhaustion attacks– Memory– Hard Drive space– Bandwidth– CPU– Entropy (for random number generation)– Database engines– Servers– Analysts– Wireless Mice, Keyboards– Wireless NICs

Question

Which one of these resources is susceptible to a resource exhaustion attack?

a) Electric power b) Chair c) Trackpad

Resource Exhaustion

May happen whenever there are:– A finite number of resources– A finite rate (e.g., processing)

Hard to defend against some variants– Sometimes a balancing act– Analogy: By staying home, the risk of meeting unpleasant

people is removed, but the cure may be worse than the risk.

How Resource Exhaustion Happens

Spend time processing requests from illegitimate (but possibly legitimate) users

Allow legitimate users to hog resources Improperly free resources no longer needed Sometimes design errors, sometimes

implementation

Resource Exhaustion Enablers

Expensive Tasks– Algorithms– Encryption, Compression and Encoding

e.g., DVDs are expensive to compress

Generous Protocols and Algorithms– Anonymous or unauthenticated allocation of computer

resources– Amplifiers (broadcasts, subscriptions, distributed systems)

Coding errors turned into vulnerabilities– Memory Leaks and other memory management errors

Design errors– Absence of policies, restrictions, access control, partitions

or compartments, backups, failover or redundant systems

Example: Disk

Risk: Disk or partition is unavailable because it is completely filled

Threat: one user can rob all others of disk space (including the use of a partition)

Resource managed by the operating system Enabling factors

– Missing or no quotas specified by OS for users and processes /tmp directory

– fill up the disk (or partition) with temp files

/var directory– use up the disk (or partition) with logs

Question

Identify the correct resource exhaustion enablers:

a) Memory failures

b) Generous protocols and algorithms

c) Expensive hardware

Question

Identify the correct resource exhaustion enablers:

a) Memory failures

b) Generous protocols and algorithms

c) Expensive hardware

CPU Exhaustion Attacks

Uninterruptible tasks Unwise operational order

– Perform a series of complex operations first, before checking the request's validity

Asymmetric attacks– Cost for attacker is much smaller than for defender– Algorithmic complexity attacks

Uninterruptible Tasks

CAN-1999-1285Linux 2.1.132 and earlier allows local users to cause a denial of service (resource exhaustion) by reading a large buffer from a random device (e.g. /dev/urandom), which cannot be interrupted until the read has completed.

CPU not available until random numbers have all been calculated

Unwise Operational Order

A firewall’s job is to block trafficDon’t perform expensive operations on traffic you’re blocking anyway!

CAN-2002-1203 IBM SecureWay Firewall before 4.2.2 performs extra processing before determining that a packet is invalid and dropping it, which allows remote attackers to cause a denial of service (resource exhaustion) via a flood of malformed TCP packets without any flags set.

Asymmetric CPU Attacks

Cryptographic algorithms are typically expensive– Initiate communications so server generates keys, etc...

Don’t know if message is good until decrypted– Send random messages

IPSEC design vulnerability

Algorithmic Complexity Attacks

Exploit worst-case scenario of algorithms– Hash algorithms (Crosby and Wallach 2003)

Data structure pollution Bro IDS (Intrusion Detection System) dropping 70% packets Normally O(N), becomes O(N2) with malicious input

– if N is 1000, cost is 1000 times higher than expected!

– Quicksort: O(N2) instead of O(NlogN)– Python regular expression engine

Exponential blowout with malicious input

– Fix: use algorithms that are not vulnerable "universal hash algorithms" designed to avoid the

vulnerability Please see http://www.cs.rice.edu/~scrosby/hash/

Question

Algorithmic complexity attacks work because:

a) they attack complex algorithms

b) they exploit the worst-case behavior of algorithms

c) there were errors in the implementation of the algorithms

Question

Algorithmic complexity attacks work because:

a) they attack complex algorithms

b) they exploit the worst-case behavior of algorithms

c) there were errors in the implementation of the algorithms

Discussion

How would you prevent or defend against:– Uninterruptible tasks– Unwise operational order– Asymmetric attacks

Discussion Sample Answers

How would you prevent or defend against:– Uninterruptible tasks

Limit CPU slices per user– move part of algorithm out of kernel space

– Unwise operational order Do not invest in something that may be worthless until you

know for sure you have to (may not be initially obvious)

– Asymmetric attacks Limit the rate of the expensive events/origin

Network Application and Protocol Vulnerabilities

Can produce:– Memory exhaustion– Numbered resource (e.g., ports) exhaustion– Bandwidth exhaustion...

Ports and Thread Exhaustion

In TCP/IP, an application uses a ”port”, a positive number less than 65536. Example: port 80 for web servers.

Passive FTP:FTP server reserves a random port (above 1024) for use by a client and waits for the client to connect there.

What if client doesn’t connect ever?

Ports example

CAN-2002-0221

Etype Eserv 2.97 allows remote attackers to cause a denial of service (resource exhaustion) via a large number of PASV commands that consume ports 1024 through 5000, which prevents the server from accepting valid PASV.

Threads Example

Microsoft NT architecture: FTP and Web services on the same computer share a common thread poolExhausting the FTP thread pool will cause failed connection requests for the Web service.

CVE-1999-1148IIS processes passive FTP connection requests by assigning a thread to each port waiting for a client to connect

Sockets

Socket: Data structure to record which application talks to what

Internet sockets (AF_INET):

– Which application reserved which port

– One IP address, one port = one socket you can listen to

– Incoming connections are recorded with additional sockets

– Number of sockets -1 = number of clients

Sockets example

CVE-2001-08306tunnel 0.08 and earlier does not properly close sockets that were initiated by a client, which allows remote attackers to cause a denial of service (resource exhaustion) by repeatedly connecting to and disconnecting from the server.

Generous Protocols and Algorithms

A Protocol or Algorithm that allocates resources based on (perhaps initially) anonymous or unauthenticated requests

Can you name one?

TCP/IP Generosity

The TCP/IP protocol allocates memory at the beginning stage of a communication, upon reception of a packet with the “SYN” flag, to keep track of communications (e.g., socket).

Early TCP/IP implementations kept the memory allocated for a very long time...

SYN flood attack:The sending of numerous SYN packets until all the memory available for keeping track of new connections has been consumed.

Generosity in Stateful Protocols

Protocols that maintain state information are necessarily more vulnerable to DoS attacks.– Above a certain treshold, quality of service breaks down

Connectionless protocols show progressive degradation with load

Conversion of stateful into stateless protocols– Not easy in all cases, but can solve SYN flood– Idea: encrypt the state data, and return it to client

No memory usage Increased CPU and bandwidth usage trade-off

– Reference: Aura and Nikander 1997

Amplification

Form of generosity Example: ICMP ping

– Request-response protocol– Unauthenticated– Can send request to a broadcast address

All computers respond!– To who? A spoofed IP address == Smurf attack

Bandwidth can be completely consumed by the response– Overwhelm victim destination computer– Other victims

» hosts on affected networks» networks in between broadcast and destination

Question

Can you name another amplification mechanism used by attackers?

a) Challenge-response mechanism

b) Encryption

c) Distributed Denial-of-Service attacks

Question

Can you name another amplification mechanism used by attackers?

a) Challenge-response mechanism

b) Encryption

c) Distributed Denial-of-Service attacks

In DDoS attacks, amplification is provided by numerous "zombie" (compromised) computers obeying remote commands.

Work-Around for Generosity

Quickly expire transactions (connections, etc...) that block while waiting on input– especially anonymous users

Exercise

Name the vulnerability in this pseudo-code, and explain why it is vulnerable:

1 Wait for client connection2 Validate input3 Create user object4 Match user against potential dates5 Prepare report6 Verify that user paid subscription; if so send back report, if not send bill7 Repeat (i.e. go to line 1)

Exercise

Name the vulnerability in this pseudo-code, and explain why it is vulnerable:

That is an unwise operational ordering. It can result in a resource exhaustion because expensive operations are performed before a request validity check.

Memory Management Problems

Memory leaks (very common)– Memory that is never freed, for every request– CAN-2003-0032

Memory leak in libmcrypt before 2.5.5 allows attackers to cause a denial of service (memory exhaustion)

Double free– CVE-2002-0059

The decompression algorithm in zlib 1.1.3 and earlier, as used in many different utilities and packages, causes inflateEnd to release certain memory more than once (a "double free"), which may allow local and remote attackers to execute arbitrary code via a block of malformed compression data.

Memory Management Problems (cont.)

Use of freed memory– CAN-2002-1490

NetBSD 1.4 through 1.6 beta allows local users to cause a denial of service (kernel panic) via a series of calls to the TIOCSCTTY ioctl, which causes an integer overflow in a structure counter and sets the counter to zero, which frees memory that is still in use by other processes.

Freeing wrong memory– CAN-2003-0525

The getCanonicalPath function in Windows NT 4.0 may free memory that it does not own and cause heap corruption, which allows attackers to cause a denial of service (crash) via requests that cause a long file name to be passed to getCanonicalPath, as demonstrated on the IBM JVM...

Memory Management Problems (cont.)

Information leakage– CAN-2003-0048

PuTTY 0.53b and earlier did not clear logon credentials from memory, including plaintext passwords, which could allow attackers with access to memory to steal the SSH credentials.

– CAN-2003-0047 SSH2 clients for VanDyke (1) SecureCRT 4.0.2 and 3.4.7, (2) SecureFX 2.1.2 and 2.0.4, and (3) Entunnel 1.0.2 and earlier, do not clear logon credentials from memory, including plaintext passwords...

– CAN-2003-0001 Multiple ethernet Network Interface Card (NIC) device drivers do not pad frames with null bytes...

Notes About Information Leakage

Overwrite sensitive memory to prevent leakage Compilers may remove calls to bzero and memset

during optimization– Use SecureZeroMemory in Windows– Use spc_memset from Secure Programming Cookbook

Use memory locking to prevent passwords and keys from being saved to disk (virtual memory, swap space)– mlock– AllocateUserPhysicalPages and VirtualLock

Disable crash dumps (core files)– setrlimit(RLIMIT_CORE, …)

Memory Management Problems (cont.)

Invalid memory references– CAN-2002-1294

The Microsoft Java implementation, as used in Internet Explorer, can provide HTML object references to applets via Javascript, which allows remote attackers to cause a denial of service (crash due to illegal memory accesses) ...

– CAN-2002-1289 The Microsoft Java implementation, as used in Internet Explorer, allows remote attackers to read restricted process memory, cause a denial of service (crash), and possibly execute arbitrary code via the getNativeServices function, which creates an instance of the com.ms.awt.peer.INativeServices (INativeServices) class, whose methods do not verify the memory addresses that are passed as parameters.

Memory Management Problems (cont.)

Memory exposures– CAN-2002-1125

FreeBSD port programs that use libkvm for FreeBSD 4.6.2-RELEASE and earlier, including (1) asmon, (2) ascpu, (3) bubblemon, (4) wmmon, and (5) wmnet2, leave open file descriptors for /dev/mem and /dev/kmem, which allows local users to read kernel memory.

– CAN-2002-0973 Integer signedness error in several system calls for FreeBSD 4.6.1 RELEASE-p10 and earlier may allow attackers to access sensitive kernel memory via large negative values to the (1) accept, (2) getsockname, and (3) getpeername system calls, and the (4) vesa FBIO_GETPALETTE ioctl.

Exhausting Memory for Data Structures

Process and other tables Buffer pools File descriptors Sockets Etc...

Human Resource Exhaustion

Typical street scenario:– Some people distract the person guarding assets while

others steal things or otherwise violate policies

Information security: – Create many alerts and warnings so that an analyst or

user is overwhelmed and can't identify the dangerous ones. IDS flooding tools available

– "Stick" (Giovanni 2001)

– Attack against online support and services Chat bots (Gabriolovich and Gontmahker 2003)

– Defense: Reverse Turing Tests» Recognizing distorted letters in an image» Riddles

Discussion

Discuss the similarities between SPAM and human resource exhaustion attacks.

True or False?

Denial of service attacks are all caused by resource exhaustion

All shared resources risk being exhausted

Lower resource cost to the attacker than the defender is indicative of a resource exhaustion vulnerability

Questions?

About These Slides

You are free to copy, distribute, display, and perform the work; and to

make derivative works, under the following conditions.

– You must give the original author and other contributors credit

– The work will be used for personal or non-commercial educational uses

only, and not for commercial activities and purposes

– For any reuse or distribution, you must make clear to others the terms of

use for this work

– Derivative works must retain and be subject to the same conditions, and

contain a note identifying the new contributor(s) and date of modification

– For other uses please contact the Purdue Office of Technology

Commercialization.

Developed thanks to the support of Symantec Corporation

Pascal Meunierpmeunier@purdue.eduContributors:Jared Robinson, Alan Krassowski, Craig Ozancin, Tim Brown, Wes Higaki, Melissa Dark, Chris Clifton, Gustavo Rodriguez-Rivera