Is Windows inherently more vulnerable to malware attacks than OS X?

[the only trolling here is being perpetrated by those sending people here to be outraged]

It took an attack on a Windows production server, not devotion to Apple, to put that provocative title on this entry.

On August 13 at 3:04 AM, a Windows server that I’ve been running for all of two weeks–it just replaced an Xserve G5–was attacked by a new strain of malware. This worm/trojan/backdoor/proxy/IRCbot/DDOS agent shared some characteristics with a known exploit, but it went well beyond what was described. I believed at the time of the infection, and even more strongly now, that this exploit’s latent damage potential has been underestimated. I view the terse and vague update on the CERT site regarding the less tenacious strain of this beast with a sense of foreboding.

The attack I encountered occasioned a re-examination of a common question: Is Windows more vulnerable to malware than OS X? I’ve encountered no clearer or more definitive proof point than this attack. To set the stage, I’ll describe the malware’s methods. The only victim requirement is that a Windows system–client or server from 2000 and XP on up, 32 and 64-bit–be on an Internet-accessible IP address and listening for socket requests to the Windows Server service. The attacker connects to the Windows Server service, overflows a fixed-length buffer and tricks the service into executing code contained in a portion of the buffer. The attack edits the Registry to turn off the Windows firewall and packet filter, disables notifications that you’re running with reduced security, and opens your system to anonymous access. It then uses the Registry to insert plant a pair of Windows services that run with SYSTEM privileges. Processes owned by that pseudo-user can literally do anything, unchecked, to the local machine. The malware services launch and announce your exploited system’s presence via IRC and IM. After that, an IRC bot or (sub)human driver can make your system do whatever it wants, including making it a nest for more malware. In my case, it was so eager to scan the Internet for other systems to infect that it locked my server’s CPUs at 100 percent and gave itself away.

To nail itself in place, two services watch for and regenerate each other even if their files are deleted. The malware adds an entry to Administrator’s login script, and it watches for a privileged invocation of Windows Explorer (like Finder) and attaches a malicious thread to that.

I’ve been giving it great deal of thought, and I came up with a reasons pointing to the likelihood that Windows is at greater risk of catastrophic attacks. It’s not easy reading, but it was either this dense packing or a book-length blog post.

• All Windows background processes/daemons are spawned from a single hyper-privileged process and referred to as services.
• By default, Windows launches all services with SYSTEM-level privileges.
• SYSTEM is a pseudo-user (LocalSystem) that trumps Administrator (like UNIX’s root) in privileges. SYSTEM cannot be used to log in, but it also has no password, no login script, no shell and no environment, therefore
• The activity of SYSTEM is next to impossible to control or log.
• Most of the code running on any Windows system at a given time is related to services, most or all of which run with SYSTEM privileges, therefore
• Successful infection of running Windows software carries a good chance of access to SYSTEM privileges.
• Windows buries most privileged software, service executables and configuration files in a single, unstructured massive directory (SYSTEM32) that is frequently used by third parties. Windows will notify you on an attempt to overwrite one of its own system files stored here, but does not try to protect privileged software.
• Microsoft does not sign or document the name and purpose of the files it places in SYSTEM32.
• Windows has no equivalent to OS X’s bill of materials, so it cannot validate permissions, dates and checksums of system and third-party software.
• Windows requires that users log in with administrative privileges to install software, which causes many to use privileged accounts for day-to-day usage.
• Windows requires extraordinary effort to extract the path to, and the files and TCP/UDP ports opened by, running services, and to certify that they are valid.
• Microsoft made it easy for commercial applications to refuse a debugger’s attempt to attach to a process or thread. Attackers use this same mechanism to cloak malware. A privileged user must never be denied access to a debugger on any system. My right to track down malware on my computers trumps vendors’ interests in preventing piracy or reverse-engineering. Maintaining that right is one of the reasons that open source commercial OS kernels are so vital.
• Access to the massive, arcane, nearly unstructured, non-human-readable Windows Registry, which was to be obsolete by now, remains the only resource a Windows attacker needs to analyze and control a Windows system.
• Another trick that attackers learned from Microsoft is that Registry entries can be made read-only even to the Administrator, so you can find an exploit and be blocked from disarming it.
• Malicious code or data can be concealed in NTFS files’ secondary streams. These are similar to HFS forks, but so few would think to look at these.
• One of the strongest tools that Microsoft has to protect users from malware is Access Control Lists (ACLs), but standard tools make ACLs difficult to employ, so most opt for NTFS’s inadequate standard access rights.

Why this can’t happen under OS X:

• OS X has no user account with privileges exceeding root.
• Maximum privilege is extended only to descendants of process ID 1 (init or Darwin’s launchd), a role that is rarely used and closely scrutinized.
• Unlike services.exe, launchd executes daemons and scheduled commands in a shell that’s subject to login scripts, environment variables, resource limits, auditing and all security features of Darwin/OS X.
• Apple’s daemons have man pages, and third parties are duty-bound to provide the same. Admins also expect to be able to run daemons, with verbose reporting, in a shell for testing.
• OS X Man pages document daemons’ file dependencies, so administrators can easily rework file permissions to match daemons’ reduced privileges.
• Launchd can tripwire directories so that if they’re altered unexpectedly, launchd triggers a response.
• If an attacker takes over a local or remote console, any effort to install software or alter significant system settings cannot proceed without entering the administrator’s user name and password, even if the console is already logged in as a privileged user. In other words, even having privileges doesn’t ensure that even an inside hacker can arrange to keep them.
• OS X has a single console and a single system log, both in plain text.
• OS X’s nearest equivalent to the Registry is Netinfo, but this requires authentication for modification. In later releases of OS X, it is fairly sparse.
• Applications have their own per-user and system-wide properties files, private Registries if you like, stored in human-readable files in standard locations.
• Every installed file is traceable to a bill of materials that can verify that the file is meant to exist, and that it and all of its dependencies match their original checksums. Mac users, back up and protect your Receipts folder!
• The directories used to hold OS X’s privileged system executables are sacred. Anything new that pops up there is immediately suspect.
• OS X does not require that a user be logged in as an administrator to install software. The user or someone aiding the install needs to know the name and password of a local administrative user to complete the install. On a network, most software is installed using Remote Desktop, an inexpensive Systems Management Server-like console.
• The UNIX/POSIX API, standard command-line tools and open source tools leave malware unable to hide from a competent OS X administrator. It takes a new UNIX programmer longer to choose an editor than it does to write a console app that walks the process tree listing privileged processes. Finding the owners of open TCP/UDP ports or open files is similarly trivial. The "system" is not opaque.
• Basic OS X features can be put to use to make life miserable for malware. For example, Windows’ hackable restore points are done better by OS X’s ability to create encrypted, read-only disk images. They’re simpler than archives, and you can mount them as volumes anywhere in your file hierarchy.
• Likewise, OS X Server will image any Mac client or server’s local drives and maintain safe copies that can be used not only for restoration, but which can be booted from to guarantee that there’s no trace of infection.
• When erase-and-reinstall is the only way to be sure, OS X Server automates it. It can safely capture the affected Mac’s active drives before having that Mac boot from the fresh install image.

So, after all this, do I have enough to judge Windows inherently more vulnerable to severe malware than OS X? I do.

I’ve been writing about these shortcomings for years, and it always traces back to Microsoft’s untenable policy of maintaining gaps in Windows security to avoid competing with 3rd party vendors and certified partners. Apple’s taking a different approach: What users need is in the box: Anti-virus, anti-spam, encryption, image backup and restore, offsite safe storage through .Mac, and launchd. Pretty soon any debate with Microsoft over security can be ended in one round when Apple stands up, says "launchd," and sits back down.