In what is sure to be my last blog (?!) born out of reading “This Is How They Tell Me The World Ends” by Nicole Perlroth, this article is basically a
quick dive into a single paragraph that contains one sentence with an alleged fact pertaining to vulnerability history. As a self-described Vulnerability Historian, this is of course of interest especially if the statement is questionable. From page 63 of the book, here is the paragraph for full context and the relevant bits in bold:
But that’s not how he would be remembered. One month after unleashing his white paper on the world, Charlie [Miller] made an even bigger name for himself with the first remote hack of an iPhone. The conventional wisdom had always been that the iPhone – with its sleek design and closely held code – was more secure than the alternatives. But Charlie blew a hole right through that theory. He demonstrated before an audience of hundreds how easily he could remotely control anyone’s iPhone simply by steering their browser to a malicious website he created.
With that, we’ll examine three components of this claim:
- Was the vulnerability remote?
- Was the vulnerability in an iPhone?
- Was Miller the first?
Before jumping to conclusions on those answers, there’s a twist or two! If you’re already grumbling over me being “wordy” you can scroll down to the end for the cliff notes TL;DR and miss the rabbit hole adventure. And also thank me for not posting each section separately, teasing it out and making you wait two weeks for an answer.
Was it Remote?
Perlroth cites the quoted section above from a 2007 article by John Schwartz titled “iPhone Flaw Lets Hackers Take Over, Security Firm Says“. To make sure we understand the context from this article, with an important bit highlighted:
Once he was there, the site injected a bit of code into the iPhone that then took over the phone. The phone promptly followed instructions to transmit a set of files to the attacking computer that included recent text messages — including one that had been sent to the reporter’s cellphone moments before — as well as telephone contacts and email addresses. “We can get any file we want,” he said. Potentially, he added, the attack could be used to program the phone to make calls, running up large bills or even turning it into a portable bugging device.
For clarity, and to show this was widely reported, we see from Farhad Manjoo of Salon in his article “Security researchers find a dangerous iPhone flaw” that the attack vector is stated more clearly:
The hack — first reported by John Schwartz in Monday’s New York Times — can be activated through a malicious Web site, a Wi-Fi access point or a link sent to the phone through e-mail or a text message. After it’s activated, an attacker can make the phone transmit files or passwords, run up wireless services or even record audio and relay it back to the attacker.
The reason the attack vector is so important is that it speaks to the first part of the claim in which Perlroth says it was the “first remote hack”. In the context of vulnerabilities, remote means that a vulnerability can be exploited remotely without user interaction from the victim. If the exploit requires the victim to perform an action of any kind, including clicking a link, it is a user-assisted or context-dependent attack. While that is a serious attack, since we know the bar for clicking a link is low, it is still important to make this distinction. Why? Let’s start with risk scoring and refer to Remote Code Execution (RCE) and Arbitrary Code Execution (ACE) for reference.
Using the Common Vulnerability Scoring System (CVSS), an industry standard for better or worse, we get four sets of scores to look at. First, understand that many organizations use a three-tier “stoplight” system for general risk severity (i.e. low, medium, high) or a five-tier system that adds an ‘informational’ and ‘critical’ rating. The five-tier system breaks down as 0.0 (informational), 0.1 – 3.9 (low), 4.0 – 6.9 (medium), 7.0 – 8.9 (high), 9.0 – 10.0 (critical). For organizations that prioritize at this higher level first, focusing on critical before moving to high-risk, the difference between an 8.9 and 9.0 may mean a lot. So let’s compare a RCE versus an ACE looking at both CVSS version 2 and 3, which are the same in spirit but different in implementation:
As we see, arbitrary code execution under CVSSv3 is scored as 8.8 which is only “high” risk while under CVSSv2 it is “critical”. Compare that to remote code execution which is “critical” under both scoring systems. So the distinction between remote and user-assisted is important in both severity and factual accuracy. Jumping back to specifics of the attack:
The hack — first reported by John Schwartz in Monday’s New York Times — can be activated through a malicious Web site, a Wi-Fi access point or a link sent to the phone through e-mail or a text message.
This is clearly an arbitrary code execution situation as it requires a victim to visit a malicious web page in some manner. That distinction is one that Charlie Miller has made himself many times over the years. This is not a remote vulnerability. At this point it would be more accurate to say “first user-assisted code execution vulnerability in an iPhone“. That’s a bit different huh?
Was the vulnerability in an iPhone?
The simple answer is of course, “yes”. But it’s more complicated than that which we’ll see, as well as why that is important. When attributing a vulnerability to a given device like an iPhone we should note if the vulnerable code is in Apple’s own iPhone code or a third-party library used by the device. This distinction starts us down a rabbit hole.
Available for: iPhone v1.0
Impact: Viewing a maliciously crafted web page may lead to arbitrary code execution
How do we know it was this vulnerability and not a subsequent one? Perloth says it came one month after Miller’s paper, “The Legitimate Vulnerability Market” from May, 2007. Miller and Honoroff’s vulnerability was shared with Apple on July 17 and publicly disclosed on July 19. Close enough to a month and the next iPhone update was 1.1.1 in September which did not credit Miller. You can also notice that while Perlroth credits Charlie Miller, it was additionally credited to a second individual, Jake Honoroff.
We can see that the first two advisories attribute the vulnerability to the code in Safari, while the Safari advisory attributes the vulnerability to WebKit, an open-source web browser engine used by Google Chrome, Apple Safari, Mozilla Firefox, Microsoft IE (recent versions) and other browsers. But the advisory tells us the issue is actually in Perl Compatible Regular Expressions (PCRE), which is a library used within a library (WebKit) used within Safari used within the iPhone. At this point it would be more accurate to say “first user-assisted code execution vulnerability in a transient dependency used by the iPhone“. That’s quite different huh?
We need to go further down the rabbit hole though. Since the vulnerability is in WebKit, which existed before the iPhone and the first security patch, we need to consider if any prior WebKit vulnerabilities might have impacted the iPhone and simply weren’t reported as such. We know iPhone development began in 2004 and the first release was June 29, 2007. We don’t know what that development was like, specifically how often they pulled in upstream code in WebKit. In theory that gives us a 3.5 year window but I think it is safe to say the developers would pull in code more often. There are at least two WebKit exploits from 2006, only one later in the year disclosed on November 14 that is ACE. I’d suspect that was patched well before the iPhone release since it was patched in macOS at that time.
Next we need to consider if other Safari vulnerabilities might have impacted the iPhone. One vulnerability jumps out quickly, an ACE in Safari patched on June 12, but it only impacts installs on Windows. Next we have a vague disclosure on June 11, 2007 about “ten flaws” in the SVG parsing engine that weren’t reported to Apple (CVE-2007-3718). These very well could represent vulnerabilities that impacted the iPhone, we simply don’t know. There were two more ACE vulnerabilities reported in Safari with no indication they were fixed, just reported (CVE-2007-3187). These could very well affect the iPhone as well.
Finally, we have to consider if vulnerabilities in other third-party libraries used in the iPhone affect it. Apple doesn’t publish a list of those libraries but based on prior disclosures that affect macOS, which could also affect the iPhone, those include expat, troff, Libxslt, ICU / ICU4C, libXfont, libxml2, glibc, and some FreeBSD BDF font handling code. That’s a lot of code we don’t know about that is certainly a concern.
Did Miller’s vulnerability in question affect the iPhone? Yes, but, at this point it would be more accurate to say “first publicly disclosed user-assisted code execution vulnerability in a third-party library used by the iPhone after commercial sales began“. That’s even more specific huh?
Was Miller the First?
Since the iPhone advisory above covers the first security update for the device, that advisory represents the first batch of vulnerabilities patched after public release. The next thing we need to look at are the other vulnerabilities patched; are any of them ACE or RCE? Yes, one of the four other vulnerabilities is an ACE as well (CVE-2007-2399). It is described as:
Impact: Visiting a maliciously crafted website may lead to an unexpected application termination or arbitrary code execution
Description: An invalid type conversion when rendering frame sets could lead to memory corruption. Visiting a maliciously crafted web page may lead to an unexpected application termination or arbitrary code execution. Credit to Rhys Kidd of Westnet for reporting this issue.
So there are two ACE vulnerabilities fixed in the same advisory. How did Schwartz at the New York Times know that Miller and Honoroff’s vulnerability was first? Because Miller likely told him so. In the article Schwartz quotes Lynn Fox from Apple so they talked but I suspect that Schwartz did not verify that information and Fox did not volunteer it. From the NYT article:
The researchers, working for Independent Security Evaluators, a company that tests its clients’ computer security by hacking it, said that they could take control of iPhones through a WiFi connection or by tricking users into going to a Web site that contains malicious code. The hack, the first reported, allowed them to tap the wealth of personal information the phones contain.
A spokeswoman for Apple, Lynn Fox, said, “Apple takes security very seriously and has a great track record of addressing potential vulnerabilities before they can affect users.”
Per that article and other sources, we know that Independent Security Evaluators (ISE) reported the vulnerability to Apple on July 17. Looking in VulnDB I can see that Kidd reported his find to Apple on June 13, over a month before ISE did, and it is in the third-party library WebKit rather than a transient dependency of WebKit. So that settles it, right? Not quite.
We know that between these two vulnerabilities, Miller was not first. But we also know that neither were remote code execution either. Moving past the iPhone 1.0.1 update, we have to go through each subsequent release to figure out if any of the fixed vulnerabilities qualify. Fortunately, we only have to go one more version to 1.1.1 before we have our first candidate. On September 27, 2007, the update fixed vulnerability in Bluetooth functionality that can be exploited remotely:
Impact: An attacker within Bluetooth range may be able to cause an unexpected application termination or arbitrary code execution
Description: An input validation issue exists in the iPhone’s Bluetooth server. By sending maliciously-crafted Service Discovery Protocol (SDP) packets to an iPhone with Bluetooth enabled, an attacker may trigger the issue, which may lead to unexpected application termination or arbitrary code execution. This update addresses the issue by performing additional validation of SDP packets. Credit to Kevin Mahaffey and John Hering of Flexilis Mobile Security for reporting this issue.
This technically qualifies as the first remote vulnerability in the iPhone! However, notice that it has to be exploited from within Bluetooth range which severely limits exploitation. In such cases CVSS would be scored as AV:A, meaning adjacent network, dropping the score a good bit. While this does fit the bill, meaning Kevin and John deserve serious kudos, it isn’t remote in the context most associate the term with. So let’s keep going to see the first fully remote vulnerability in an iPhone. We pass the releases for 1.1.2, 1.1.3, 2.0, and 2.1 to find the next of interest in 2.2 on November 20, 2008:
Impact: Viewing a maliciously crafted TIFF image may lead to an unexpected application termination or arbitrary code execution
Description: Multiple uninitialized memory access issues exist in libTIFF’s handling of LZW-encoded TIFF images. Viewing a maliciously crafted TIFF image may lead to an unexpected application termination or arbitrary code execution. This update addresses the issue through proper memory initialization and additional validation of TIFF Images.
Impact: Viewing a maliciously crafted TIFF image may lead to an unexpected device reset
Description: A memory exhaustion issue exists in the handling of TIFF images. Viewing a maliciously crafted TIFF image may lead to an unexpected device reset. This update addresses the issue by limiting the amount of memory allocated to open a TIFF image. Credit to Sergio ‘shadown’ Alvarez of n.runs AG for reporting this issue.
These two vulnerabilities are interesting because there is a potential for a remote attack here, but the advisory doesn’t make it clear in wording and they don’t provide CVSS scores. Since an image can be delivered a wide variety of ways, including via SMS, the fact that these occur in the ImageIO subsystem is of note. The Apple Developer documentation backs up this thought:
The Image I/O programming interface framework allows applications to read and write most image file formats. This framework offers high efficiency, color management, and access to image metadata.
A bit light on details but this suggests that if e.g. an SMS messaging app, or any other that remotely receives content and processes it, could be an avenue for remote code execution. Based on a chat with a colleague, it would require the victim opening the SMS app at the very least which is a low bar for exploitation, but he does not think the iPhone SMS app renders the images as a preview without clicking into a specific message. Low bar, but still requires some user interaction. We see the exact same thing for CVE-2008-3623 and CVE-2009-0040 in the iPhone 3.0 update on June 17, 2009. It is interesting to note that we’re now two years after the iPhone’s release and full remote vulnerability with no limitations or caveats.
Impact: Viewing a maliciously crafted image may lead to an unexpected application termination or arbitrary code execution
Impact: Processing a maliciously crafted PNG image may lead to an unexpected application termination or arbitrary code execution
Impact: Receiving a maliciously crafted SMS message may lead to an unexpected service interruption or arbitrary code execution
Description: A memory corruption issue exists in the decoding of SMS messages. Receiving a maliciously crafted SMS message may lead to an unexpected service interruption or arbitrary code execution. This update addresses the issue through improved error handling. Credit to Charlie Miller of Independent Security Evaluators, and Collin Mulliner of Fraunhofer SIT for reporting this issue.
This has all the makings of what we’re after. While the advisory says “arbitrary code execution” that is a qualifier to “decoding of SMS messages”. Receiving the message triggers it as the payload is processed regardless of loading the message specifically. But notice that the same issue was also found by Collin Mulliner. So who found or reported it to Apple first? That is what ultimately settles this question. Since it lists two people with two different affiliations, that typically means mutual discovery or a “vulnerability collision”.
I reached out to a contact at Apple and asked if they could figure out which of the two sent the email first to settle this debate. Low and behold, I was told that it was a single mail sent June 18, 2009 and both were listed as creditees! That is backed up by a presentation at Black Hat USA 2009 titled “Fuzzing the Phone in your Phone” given by both individuals.
We began the blog with a quote from Nicole Perlroth’s book, “This Is How They Tell Me The World Ends”, in which she says “One month after unleashing his white paper on the world, Charlie [Miller] made an even bigger name for himself with the first remote hack of an iPhone.” The question is if that quote is accurate, understanding she is citing CVE-2007-3944. The answer is, it’s complicated. Here’s the facts as best I can tell:
- Was it remote? No, the cited vulnerability is a user-assisted issue and cannot be carried out remotely w/o that victim clicking something.
- Was Miller the first? It’s complicated.
- If we go strictly by CVE-2007-3944, then no Miller was not the first. Rhys Kidd disclosed a user-assisted vulnerability in WebKit, the rendering engine in Safari, over one month earlier. Further, Jake Honoroff co-disclosed the vulnerability Miller found.
- If we go by remote without interaction but limited in attacker location, then no, Kevin Mahaffey and John Hering are the first with CVE-2007-3753 that must be exploited over Bluetooth.
- If we go by the actual first remote vulnerability, CVE-2009-2204 around two years later, then yes but Miller co-discovered it with Collin Mulliner and both earned that distinction.
In short: no, kind of, no, no, yes but. So Perlroth is incorrect in her assertion and very likely included it after talking to Miller for her book. The problem is that in the context of the 2007 vulnerability, Miller was wrong and Perlroth et al did not properly fact check that detail, instead relying on a co-worker’s article as gospel. We don’t know if Miller mentioned Honoroff in his discussions with Perlroth or if her text was written outside the scope of her discussion with Miller, but that detail was trivial to find if the claim was fact checked beyond her colleague’s article that also omitted it.
Regardless, I believe we have a definitive answer as to that one line in the book. It took eight pages to get to this answer and I apologize for that (not really).