--- title: Phrack Prophile on Halvar Flake author: Phrack Staff --- |=-----------------------------------------------------------------------=| |=-----------------=[ PHRACK PROPHILE ON Halvar Flake ]=-----------------=| |=-----------------------------------------------------------------------=| |=--------------------------=[ Phrack Staff ]=---------------------------=| |=-----------------------------------------------------------------------=| |=---=[ Specs Name: Thomas Dullien Handle: Halvar Flake / HalVar on IRC Handle origin: Children's cartoon about a smart viking child AKA: elwood_blues@viking.ruhr.com in the 90s Affiliations: TESO Active Since: I think 1996 or 1997? Country: US, Germany, Switzerland, France Website: thomasdullien.github.io GitHub: github.com/thomasdullien Projects/Code: Uh. Yeah. |=---=[ Background I'll steal Gera's approach and have a short version and a lengthy historians version: When I was 8 or 9 I asked my brother "how do you crack copy protections", and he said "it's easy, you just invert a jump". I didn't know what a jump was, but remembered the advice. I learnt assembly at 16, then started cracking and virus writing, then moved into reverse engineering and hacking/vuln-dev, started training government folks on doing that to finance my math degree, accidentally started a company, got acquired by Google when my health was bad, recovered for a few years, got tired of security, started another company in performance optimization, sold that, then spent a few years caring for family. Now interested in LLM and AI stuff, because everything is in motion and new and interesting. ---------- Version for historians and really bored readers: -------------- My dad struggled professionally in his 40s while I was on the way. Money was tight. He answered a newspaper ad and became a developer. My 6 year older brother bought himself a self-assembly computer at age 8 or 9 and taught himself game programming and cracking; he sold two commercial games at age 14. He was the smart nerd in the family. I wasn't. Dad taught me BBx BASIC when I was 8, but didn't do much with it. I typed in magazine listings to get games, but couldn't debug them to get them to run. I asked my brother how to crack copy protections, and he said: "It is easy, you just invert a jump. You need to know assembly." - and that's all I remembered. I became the artsy kid instead, drawing, and as a teen trying my hand at editing Quake 3D models, using a software called "QuakeMe" that had some limitation in the shareware (non-paid) version. I went to the US for a high school exchange year. At Barnes&Noble I saw "Master Class Assembly Language" by Wrox Press and bought it. It taught me what a JMP is, how to put the CPU into protected mode, how a DPMI extender and paging works, and all the things I needed. I installed an IRC client and found the cracking scene, joining #cracking4newbies (you'd get booted from #cracking immediately if you weren't obviously good). I started learning and cracked QuakeME with a *terrible* butcher job, but it worked. I was hooked. I read ed!son's cracking tutorial, then stumbled into what would become Fravia's reverse engineering website. I had found a practically infinite fountain of puzzles and peers to puzzle with. The EFNet cracking scene was a very vibrant place. An entire generation of curious kids, spanning 10-15 years age difference, congregated in shared spaces. There was a lot of 1990s weirdness (from today's perspective toxic or even sociopathic), but many weird kids found their communities. I met many people that are now important in the security world there - too many to name. Rolf Rolles, scut, acp, random_, Neel Mehta, Stephan Esser, Anders Vogh, Solar Designer, vizions, the_grugq, the_owl, G-R0M, iceman.de, Nishad Herath, etc. etc. etc. The cracking scene ran on a piece of software called W32dasm, but I was an early adopter of IDA Pro when it first became "available". Slow and deliberate suited me, and IDA was a tool for thorough work. I used IDA almost daily for the next 20 years (the last 10 years years it's been perhaps twice a year). I've been an IDA user for 30 years now. I hacked my first computer in 1996 by abusing the PHF vulnerability, a machine at a Mexican university. I didn't know what to do from here, as I didn't know Unix at all. I read VLAD and 29A magazine. The Win32.Cabanas source code and article in 29A were an absolute eye-opener for me, and I meditated on Win32.Cabanas for a long while. Solar Designer's Crackme that included custom virtual machines and everything involved with executable packing (PECrypt, Procdump32). I read Phrack, too, but until Smashing the Stack, a lot of hacking articles seemed alien to me. The cracking scene was full of assembly, interrupts, some crypto even, and many hacking articles were "hey we social engineered our way in" and "here is a manual for some obscure telco system". Fascinating, but far away from what I knew. "Smashing the stack" was an eye opener to me. And writing shellcode was easy if you were used to writing keygens in assembly. But as a cracking scene kid I knew nothing about UNIX, or using a shell other than the terrible CMD.EXE. I returned to Germany a changed person, deeply into the cracking scene, and interested in the social status within that scene and all the puzzles. I had two years of German high school to go. I earned some money cutting sheets of metal reactants with a scissor at a firm nearby. I saw an ad that a copy protection dongle company was looking for people to test the security of the dongle. I applied, and then learnt during the interview process that my future boss was iceman.de. The job was great and allowed me to work from a cafe or a bar with a laptop, which was helpful because the relationship between my parents had deteriorated, and with it my ability to find peace at home. At the time, there was a conference/party called "UC3D" - "underground conference 3D" - which was one of the last places where the cracking scene, the demo scene, and the hacking scene still had touchpoints (the scenes had largely split and diverged). I took a train to Bingen in Germany and met random and lorian, who I had known online for years. The party had live-on-site-demo-coding competitions, but also a "hardware compression" competition (where you had to crush a piece of hardware to the smallest possible volume). I met some other youths that were dabbling in hacking and that were talking about having compromised and networked hundreds of machines, and there was a lot of talk about using those machines in DDoS networks. I was disappointed by some of the "hackers" I talked to, because they seemed uninterested in writing their own exploits (which is really all I cared about). The party was a bit somber because the french sceners didn't come after the ETOILENOIRE busts the week before. Shortly thereafter a "mass exodus" from the cracking scene into the hacking scene happened. Scut, an early TESO members also had home in cracking, invited a bunch of crackers into TESO. It worked out great. The cracking scene folks merged with the existing highly technical hacking folks, and for a brief moment TESO was among the most respected "hacking groups" for sheer technical output. ADM was quite possibly deemed cooler, but TESO was up there, and for a brief moment we had an extremely collaborative culture. We wrote up everything we learned in "TESO informationals", text files that described what we were up to - many of which would've justified a conference talk. We hung out on IRC, we exchanged literature recommendations, and helped each other solve problems. Near high school graduation I saw a post on Bugtraq about a stack overflow in Netscape Enterprise Server, an important commercial webserver at the time (it ran CNN.com, which was an important news website, culturally about as relevant as Fox today). The people that had posted the overflow seemed to struggle with exploitation, and I emailed them telling them I could do it. They emailed back that if I succeeded they'd send me $1k. I sent the exploit the next day. They offered me a job in Bangkok, and given that I was about to graduate from high school and had a few months of downtime before starting my replacement service for the mandatory military service, I agreed. I was happy, because in the past year my home life had gotten very difficult due to things outside of my control, so a path offering me a quick escape and some money was great. I graduated from German high school and took a flight to Bangkok the next day. It was there that I met a few old-school hackers that had started a security consultancy. I also met Fyodor Yarochkin. We worked together, and I consider him a close friend and a great human being to this day. It was fun times -- software was so full of bugs, and people were not used to buying a pentest and getting hacked by zero days in core infra services. It was also fun to be 18 and in a foreign land with good food and interesting work. I remember Fyodor and me playing around with restricted x86 shellcode at the time, and managing to do alphanumeric shellcode with '_' and '-' added (don't ask me why I needed those characters now), and scripting IDA trying to do static analysis using IDC. It was fun. Fyodor also got me interested in all the strange other architectures at the time, from PA-RISC (where the stack grows upwards) to SPARC (to this day the most beautiful assembly). I returned to Germany to start the military replacement. It was in a home for troubled youth in a very rough part of town. The job gave me about 2 hours each morning to hack on my own stuff before the kids arrived. While in Thailand I had agreed with a fellow windows hacker, Joey__, that we should meet. He was one of the few people at the time that wrote Windows exploits (I can really just remember a handful of nicknames: darkspyrit (Barnaby Jack), Joey__, a Japanese person that called himself UNYON / dark penguin (?) that later worked at EEye ... and a bit later Greg Hoglund (?). Everybody kinda knew (of) each other. Joey__ had hired two cracking sceners that I valued extremely highly, and they had set up shop in Sri Lanka, and there they had done magical things to extend SoftICE, the kernel debugger every windows exploit person lived in. SoftICE is a hard thing to explain to people nowadays, but it was a software kernel debugger that largely inserted itself *under* the OS, with limited understanding of the actual operating system, and which would freeze the entire system. You'd hit CTRL-D, a black window would pop up, and the entire OS, including the clock etc. would stop. It was great for providing extremely fine-grained control and the ability to single-step etc., but this was before virtualization on x86 was a thing, so you'd go into SoftICE and your music playback would hang on the last sample, and your machine clock would stop, and literally your entire system stays put until you single step again. And if you were deep in some tricky self-modifying code, you didn't really dare go out of it. Anyhow, these friends of mine had reverse engineered the (closed source) SoftICE to an enormous extent and then written third-party extensions to it, from the practical (using SoftICE as a tracer to unpack packed PE files) to the whimsical (adding a below-OS-level MP3 player so you could continue listening to music while you're debugging). I had always admired Joey__ and we agreed submit to the same Blackhat conference so we could meet. I submitted, and got accepted, and somehow Joey__ didn't submit -- I'm hazy on the details -- but through this I ended up presenting at Blackhat Amsterdam 2000 on reverse engineering. Afterwards I was asked if I could talk more about this. I was 19, fresh out of high-school and needed a way to finance myself (I was not going to allow myself to be dependent on my parents, given the difficult situation at home). I started doing reverse engineering and exploit development trainings at Blackhat in order to finance my studies. Here I got into contact with all the usual three-letter suspects. In my first class, an experienced "DoD" reverse engineer (with like 13 or 15 years experience doing reverse engineering, an unimaginable timespan for a 19 year old) sat in the class with some junior folks and I was confused why they'd send someone with that pedigree to a class like mine, taught by a kid that taught himself to reverse just a few years ago. The person was extremely nice, and explained that he's mainly there to check out how I'm doing things, but that he doesn't have time to train all the people that need to learn this stuff. Contact with the US intelligence sector was a bit unsettling at first -- I had grown up in a *deeply* pacifist core family (both my parents saw WW2 as children, and both had deep emotional scars from it - which led them to embrace pacifism to an extreme extent, in spite of a very prussian and quite martial family history). So everything to do with the military, with espionage, and with state- to-state conflict was deeply suspicious to me, and more so to my father. I was introduced to a few folks that had just left the agencies -- I met Dave Aitel, and some of his former colleagues, and I realized that a lot of these folks were intelligent, resourceful, and often quite introspective about the moral and ethical dimensions of their work. This soothed my instinctive antipathy. TESO suffered from leaks, and people that got their hands on the exploits created a bit of havoc. This led to incidents, like FBI folks asking me and another TESO member to have a chat over coffee in Vancouver while we were attending CanSecWest. I still have the pin they gave me somewhere. "We don't mind you two, but you have some really antisocial friends" is a quote I remember vividly, and having to explain to them that Burneye (the ELF cryptor) wasn't so much designed to annoy forensic investigators but rather a desperate attempt at controlling exploit leaks better. TESO's previously free-wheeling internal research culture nosedived, and everybody tried locking stuff down more. The leaks did not stop (perhaps some of the infra was compromised), and this got so bad that TESO members would often only learn of the research of another TESO researcher *after* that research had gotten leaked and published. [There is a lesson I learnt from this: In many organizations with extremely locked-down information flow, the only person that has a holistic view into what the organisation is doing is the successful attacker, and that there is a crossover point where information compartmentation kneecaps the defenders. I have seen that lesson play out repeatedly throughout my career - in a most egregious case, a friend of mine worked at a company defensively and wanted access to some internal documentation. He couldn't get it, but some other friend of mine that had not worked at that company got that documentation by monitoring a (likely Chinese) APT exfil box.] I had to decide on a subject to study at university. I tried law (family tradition), but a centralized body assigned university seats for subjects in high demand, and they assigned me to a city I didn't want to move to; I'd have to wait two semesters to study law in the city where my then-gf lived. First day of the semester I need to pick a subject. I went to the only subject that didn't have people queuing for it: Mathematics, with the intent to do two semesters to teach myself how to think clearly. I had also read Schneier's "Applied Cryptography" and thought some math would be useful. For the first time in my life a subject was genuinely hard for me, and I was middling, if not in the lower half of students. The first lecture opened with the Prof saying "look at your neighbor to your left and to your right. If you're here by Christmans, they won't be.". I was too proud to quit, so I slogged on, and eventually had the realization that if I ever wanted to be just an epsilon above useless, I'd have to decide to love it. I did, and I still love it. [In my first Analysis lecture, I met my friend Soeren, who was (and is) the best C++ developer I have ever known, and who I had the pleasure of hiring for my company later, and worked with at Google. I loved conversations with experienced C++ developers, because their gripes about the sharp corners of their language would directly inform my code auditing and vulnerabilty hunting strategies. That said, I love chatting with Soeren about most topics until this day, even if (or perhaps because?) we often disagree.] 9/11 happened a month or two into my first year at uni. Everything went crazy and complicated. The GWoT started, and my view of US actions had to become more nuanced very quickly, in particular in the run-up to the second Iraq war. I still earned my money doing trainings. I was studying mathematics and struggling with it, only to fly to Las Vegas or other places at regular intervals to teach a class full of people much older than me (and often in government roles) on reverse engineering and vulnerability development. It was very schizophrenic, and a tiny bit isolating. [I also had the strangest conversations with my professors when I had to tell them "hey, I cannot take the final exam at the prescribed data because my livelihood depends on me being gone *that week*, can I take the re-exam for people that are sick for the first one?". I remember feeling encouraged about completing my math degree when one of my professors told me "young man, you might make an OK mathematician if you'd just cut out all this computer stuff".] In 2003 Microsoft published a bunch of patches for the Windows DCOM system after the hacking group LSD had reported a bunch of Windows remotes. The In order to mitigate the risks MS wanted to have the patches available but provided no information about the vulnerabilities. I got annoyed by having the patched executables but no way to understand what the bugs were that we patched. Microsofts release executables were shipped with something that is called PGO nowadays (basic block reordering based on profiling data -- Microsoft even had a patent on this stuff), and their patches weren't -- so any byte-level comparison was out because everything had gotten shuffled. The fact that all the bytes were shuffled but the graph structure had remained stable for almost all functions led me to build the first prototype of BinDiff. Near the end of my Vordiplom (roughly equiv to undergrad, but not a stand-alone degree) I took a cryptography seminar. The topic I was assigned was hash functions, and the book was Stinson's "Cryptography: Theory and Practice". I found the chapter on hash functions inscrutable, but I had read MD5 *implementations* during my code auditing and of course been exposed to hash functions when cracking. So I decided to switch the literature from Stinson to Dobbertin's MD4 collision paper, not knowing that that paper had a reputation of being hard to read. I presented my part of the seminar, and it was the only time in my undergrad that I did well. The Prof took me aside after the seminar, and told me: "Listen, I see your grades and how you're doing here. And I see how you did on this seminar. My recommendation to you is: Go do something that *really* interests you for a change." I had heard that Dobbertin had just become a Professor in Bochum, and I decided to switch universities with the goal of writing my master thesis on MD5. I thought at the time that the construct looked shaky enough to have a go at it. I'd have a semester off beforehand. I decided to go to Argentina, where many friends were working for Gera's company, CORE. Those 2-3 months were some of the best in my life. Not particularly healthy - I'd take Spanish classes in the morning, then sit in the CORE office working mostly on my own stuff and chatting with friends that worked at CORE, then go home to hack on stuff with my flatmate, sleep on a matress in an otherwise bare apartment (that initially did not even have a chair or table; we got one when our backs started hurting), and repeat. I drank way too much Mate and slept way too little, not to mention lack of exercise. But I will forever cherish the mornings in Belgrano, near the Subte station of Congreso di Tucuman, where we'd order orange juice and tostados mixtos for breakfast after a nuit blanche spent hacking on stuff. BinDiff really became useful, and I hacked up a website for the company around it. I flew to LA from Argentina to teach a training class on patch analysis. There was demand. [My flatmate made me understand the extreme importance of exploit reliability. He was the personality type that'd spend 3 months on stabilizing a LPE (an unthinkable time investment at the time) to make sure it'd always always work. I love him as a dear friend to this day.] This Microsoft PCT SSL security patch hit while I was in Argentina, and it was BinDiff's first "big moment" -- the tool pinpointed the fix, and within 24h of patch release there was a working and reliable exploit in CORE Impact, allowing remote access to any unpatched IIS installation with SSL enabled. [I was never a fan of "regular" shellcodes, so I had adapted a friends all-assembly PE loader into a shellcode (you'd give it a PE file and an anonymous blob of +x memory, and it would load the PE file there and call the entry point). This allowed me to write all my post-exploitation stuff in C/C++ and just load DLLs into the remote memory space. It had the benefit that you'd never touch disk with any of your tools, and forensic memory acquisition wasn't very common yet (this was 2004 after all). The SSL PCT exploit was so reliable that it allowed loading DLLs over and over again; I remember abusing it and eventually managing to lock up the remote process after having loaded 30+ DLLs into the other process (I never really implemented unloading).] Good times. It was here in Argentina that I first learned about Gera's heap visualization work that he had done for the dtlogin exploit, and it was an epiphany for me. [Over the decades I have written and re-written large-scale heap visualizers 3 or 4 times, and I had to resort to GPU acceleration to deal with all those rectangles as early as 2006 or so. I still cannot imagine doing any sophisticated exploitation without a good heap visualizer. Perhaps it's time to write a Phrack article with an implementation about one?] I returned to Germany and moved to Bochum, into a new flat, in early summer 2004. In August 2004, Wang et. al. published their MD5 collision, and then visited Bochum. I was invited into the discussions with Dobbertin and his grad students and Mrs Wang. This was extremely interesting, validating but also disappointing: My hunch about MD5 fragility had been correct, but now I'd need to work on something else. Unfortunately, Prof Dobbertin fell ill shortly after the semester started, mostly disappeared from University life and eventually died of cancer. I was stranded in Bochum, the the reason why I was there gone, and I had to wait for the Cryptography department char being re-appointed. At the same time people wanted to buy BinDiff, and I set up a German GmbH to sell it (people had asked for all sorts of source code escrow etc. to buy from me as an individual, once I had a legal entity all these questions went away). (Modest) revenue started coming in, but enough to reduce my travel schedule for trainings. And a short while after, Rolf Rolles contacted me - he had manually decompiled the BinDiff plugin binary to re-compilable source code, and wanted to see if he could work with me on the tool. [I had known Rolf from the #c4n channel, and held his work in high esteem - but the fact that he had decompiled BinDiff blew my mind. We came to an agreement on how to collaborate, and Rolf made many important contributions to the core BinDiff algorithms. While him and me parted ways in an unfriendly and ugly manner a few years later, I still hold him in the highest regard - technically, there are few people that can match his depth of expertise, his persistence, and his vision.] I also was introduced to Ero Carrera, who was working at F-Secure at the time and had also looked at using the graph structure of malware to perform malware clustering. He also joined the "company", and we ended up renting an office. We were initially called SABRE Security ("security analysis by binary reverse engineering"), and the website had grand plans: BinDiff, BinNavi, and BinAudit, three products that we had planned. We planned to finance the development for each step from the money from the previous step. The roadmap was: Use BinDiff revenue to build a tool that allows input crafting (one of my motivations for studying MD5 was that I wanted to build better solvers for the type of equation systems that come up in code analysis, mixing boolean arithmetic with mod 2^32 or mod 2^64 arithmetic) to reach a given point in a program (BinNavi), and to use the money from that to build BinAudit, a hypothetical bug detection program that would identify security issues in binaries. Over the 7 year history of the company, we never started actual real development on BinAudit, but we did build BinDiff, and BinNavi, including the REIL IR that was supposed to be used to feed into the solvers. SABRE (later zynamics) was, in retrospect, an insanely hubristic project. [We had planned to do BinDiff, all of z3/KLEE, and then a hypothetical strong static analysis tool on binaries (like Codesurfer/x86?), all without external funding and bootstrapped on training revenues. In 2008 or so I thought about raising funding and realized that our business plan was essentially "we will solve a bunch of fundamentally hard CS problems and then sell the results to 1000 people at $800 a pop". I have to smile with a sort of benign grandfatherly smile at my naive ambition and overconfidence of my youth.] We had rename to zynamics because we got threatened with a trademark lawsuit. We had rather interesting customers that were often unable to talk to us, and fielded support requests such as the following: Them: "How does BinDiff work on architectures you don't support?" Us: "Can you tell us the architecture, we'll support it?" Them: "No." Us: "Can you give us a list of architectures, one of which may contain yours, and we support all of them?" Them: "No." Us: "Uhhhh." We also had to deal with many suspicious customers that wanted to drop by our office and pay in cash, and misc. other madness. We ended up having the company secretary do basic background checking of our customers. The stories I could tell. [A surprising customer segment we didn't know about were car tuners, and that amused us to no end. We also got to do contract reverse engineering for a few government-run attacks involving worms and rootkits that would only get publicly discussed a few years later. We had front-row seats for Stuxnet, front row enough to notice the body odours of the actors involved.] zynamics was fun, we grew to 10-12 people, but it was also deeply unhealthy for me. I ran myself ragged: Every time our money in the bank would dip below my safety margin I'd go and sell extra trainings to make up for the shortfall, then sales would recover and I'd still have to do the trainings, while trying to finish my degree(s) (first my MSc, then I embarked on a PhD). So developed stress-induced immune failure -- I got shingles, whooping cough, and a bursitis in short sequence, with the bursitis spiking inflammation markers beyond all reason. My doctor eventually told me: "Young man, we are seeing symptoms in you that we normally see in 80 year old men with immune failure. Whatever you are doing to yourself, if you want to see the age of 35, you need to stop doing that." I took a vacation and started surfing, and decided I need a way out of the treadmill. This was around the time of the Wikileaks madness. Julien Assange spent a good amount of time next to our zynamics tent at the dutch hacking camp eating from our barbecue and discussing, and during my "vacation" I had my car broken into, all my papers stolen, and when I drove back through Paris my drink got spiked -- but somehow in those crazy years all of that seemed par for the course. Ero and me started scheming about potentially finding a buyer, initially in the antivirus industry; we got very lucky that the Chinese Gov got caught hacking Google at just the right time. So after a year of ridiculous regotiation between us and mighty Google (11k people, growing rapidly) we sold zynamics on the 28th of Feb 2011. Everybody got a good outcome (although pocket change by today's standards), and the entire zynamics team transitioned into Google where we got to scale up our compute (we had struggled with computational cost for some of our malware work prior to that). I joined Google - it was an exciting place in 2011 - and Google made me a much better computer scientist but a much worse reverse engineer (just because the nature of my work shifted a lot). [It was good, though, my boss at Google was one of the best human beings I know, and fundamentally committed to viewing security problems through a system architecture lens (rather than a bolt-on product).] My involvement in vuln-dev shifted more theoretical -- I had had the important "insights" into the proper way to define what Sergey Bratus called a weird machine in the late 2000s, and at Google I then began thinking about scale ("if I can fill DRAM with Javascript bytecode, how many ads do I have to buy until a random cosmic ray bitflip will corrupt *some* laptop out there and give me control").] I was very happy to be able to collaborate with Mark Seaborn when Flippy the Laptop with terrible RAM was found and sparked Mark's successful run to his Rowhammer exploit. My contribution to this was largely figuring out that you can do 2-sided hammering by staring at some data visualisation, and also the idea of filling memory with PTEs (although iirc someone else had suggested that shortly before me, so I will only claim partial credit here). Mark did all the real work. I started getting tired of security, took a sabbatical, finished the weird machine paper (which was a bit of a slog, but important) and returned into an IC role at Project Zero. P0 was full of awesome human beings, and being in a research group like that felt somewhat liberating. [When you're managing the build-out of a piece of software, the work is very interlocking, like an orchestra where everybody has to be very precise. P0-style research is more like a jazz jam, there's more room for improvisation.] I got more and more frustrated with security. I also have always admired people that had multiple different careers, from my favorite applied mathematicians to people like Gera. So after sitting next to Jann Horn and learning about eBPF by watching him, I decided it is time to do something in performance land, and quit Google to start a performance optimization company. I got together with a vuln-dev friend of mine, Sean Heelan, with whom I had discussed a lot during his PhD work. We started the company, and ran it for a short 2.5 years during COVID. We raised funding this time, from a16z, built the world's first multi-runtime eBPF based continuous production profiler, and got acquired by Elastic. It was a short run, but fun - somehow a fantastic team assembled itself, we did great stuff, and then everything was over before we could really blink twice. Unfortunately, my mother fell into a coma the day the deal closed, and then died after 9 weeks of ICU later, and I had small kids etc, so the aftermath was rather difficult for me. On the positive side, Elastic open-sourced our code and donated it to OTel, so chances are that if you are running a continuous profiler these days that you are running something based on our code. So optimyze didn't end up the success I had hoped for, but it had an impact, even if it might be just enough to let me die with a net-zero CO2 budget at the end of my life. And ... well, it's 5 years later now. LLMs are now a thing and are reshaping everything, including security, and I find myself back looking at some security stuff, because things are changing rapidly enough for everything to be *interesting* again. As Hunter S. Thompson said: When the going gets weird, the weird turn pro, and I'm here for the ride, at least momentarily. |=---=[ Inspiration For some unknown reason, I told other kids when I was 8 or 9 that I'd become a hacker and steal $1bn from the defense budget. I don't know where that came from. I had a nanny as a young child, and her husband was my hero. The closest thing to a grandfather I had. He was a bona fide safe-cracker; banks would hire him to open the vaults they had accidentally locked. He taught me to play poker, and sadly passed from a pulmonary embolism shortly after I turned 6 - during a routine hospital stay he didn't receive blood thinners. I have always loved heist movies. "The Thieves" is one of my favorites. My father would go on walks when I was a kid, and to entertain me, he'd tell me "the history of bank robbery"; from barter exchange to the appearance of money and banks to the various ways banks were secured. I loved those stories. There was a German criminal called "Dagobert", Arno Funke, who fascinated me as a child, he would build all sorts of improbable technical contraptions (like radio-controlled electromagnetic backpacks to be attached to the outside of a high speed train). I always loved the idea of complex improvised contraptions doing improbable things. |=---=[ Early Influences My brother. Then probably Fravia, and the rest of the cracking scene. I never knew ed!son, but his cracking tutorial was huge for me. |=---=[ Memorable Experiences Writing shellcode on my AMD 486 knock-off, and being too clever by half: To avoid NUL bytes, I'd push the bytes for "pop ebp, call ebp" onto the stack, and then issue "call esp". It calculated the delta offset just fine on my machine, but I didn't know Intel CPUs had an erratum in the call esp instruction, so most machines on the internet would crash. Standing on a train platform in summer 2000, eating from a bag of cherries and slowly realizing the full impact of pre-malloc integer overflows. Reading OpenSSH again and again, and mistakenly assuming fatal() just "prints and exits", not realizing it does a bunch of heap cleanup, hence missing my best shot at having my own OpenSSH remote. Meeting FX in pre-Kathrina New Orleans and ending up on a stage singing Karaoke to a song with him that he didn't know. Getting double-sided row-hammering to work and buying a pack of cigarettes to smoke a single victory cigarette (after many years of not smoking). Meeting members of the cracking scene for the first time. Trading a bottle of Jaegermeister against a bottle of Vodka with Dmitri Skylarov two nights before his arrest. Getting to know Saumil Shah, and becoming friends, and the many times we have met and enjoyed each other's company since. His wedding, and travelling through Rajastan and Jaisalmer multiple times with him. Partying with Caddis and Duke and Joey__ in Las Vegas. Developing an escalating and unhealthy taste for really shitty and unpalatable bugs, like a Mozilla bug that would accidentally multiply a large area of your heap with a random large integer. Getting code execution on a Cisco PIX with dvorak by exploiting an unitialized stack variable in their ISAKMP implementation. Oh, and carrying a printed bound book of the IPSec RFCs (including IKE) around with me, reading it and annotating it. Trying to convince fuzz-testing people to give me all the bugs they deem unexploitable "for science". Showing a slide "Friends don't let friends work for ISS" at the end of a Blackhat talk where my 4 co-presenters were all ISS employees. Partying in a Sikh bar in 2000 in Kuala Lumpur. Building my first fuzzer, hating myself for it, and then being surprised when the person I gave it to found dozens of bugs with it. |=---=[ What is the achievement you're most proud of? I think there's a few: - I am proud that BinDiff is still in use, 15 years after we stopped doing serious development on it. I would've *never* imagined that that could happen. - I am proud that the techniques we invented at optimyze for profiling are now the standard way continuous profiling is done. - I am proud of the Weird Machine paper. That thing was hard to write, because translating something I understood deeply on an intuitive level into formal writing was hard and painful. It was gratifying though - several people that do work I admire have told me the paper spoke to them directly and articulated what they did and felt. - I am proud of the work we did at zynamics; we did a lot of very innovative stuff that really moved the needle for RE tooling. - I am proud of having had a tiny role in the chain of events that let to Blackhoodie, even if it was almost immeasurably small. - I am proud of the great and high-impact work that some of the folks that attended my vuln-dev and reverse engineering classes have done. I like to kid myself into believing I had a part in that. Perhaps the proudest thing is: I think it is difficult (although in all likelihood not impossible) to find people that would say I treated them badly, unfairly, or unjustly. You can't be friends with everybody, and you can't always make everybody happy, but you can try to be fair and square with everybody. |=---=[ What is something you are not proud of? There are things I stated with great certainty when I was younger that were kinda sloppy science. I'd rightfully criticize my younger self for those things. I wasn't always as kind as I could've been when I was younger, and even today, I can get impatient and then accidentally hurtful. A friend told me that in technical discussions with me they get panicked flashbacks to math exams in school, and I don't like doing that. I have a set of values, and the belief that I need to apply these with great rigor to myself and with lenience toward others. This leads to me constantly falling short of my expectations for myself, and I am not proud of each of these shortfalls. |=---=[ What would you like to see published in Phrack? Ohhh shiny. So I am aware that career incentives and legalities etc. make it difficult to publish the sort of work I'd love to read in Phrack. But I love everything that involves taking things apart: Tooling for systems analysis (heap introspection? Better fuzzing? Code analysis? Anything else?), or for the analysis of complex systems (all that AI mechanistic interpretability should have a hacker equivalent), stuff about new computing platforms and paradigms (CUDA? Low level TPU stuff?), things about physical attacks (EMFI?), etc. etc. In essence "cool stuff that pushes the boundary of what is possible, ideally in a way that is reproducible with small budget at home, and that provides true insight". How about a tiny LLM that you can run at home that can't do anything but reverse engineering? Or tools to steal model weights from a running CUDA process? Timing attacks against sandboxes? Soooo much stuff. Perhaps we need some code to allow better diagramming from ASCII... |=---=[ How did Phrack influence you and helped shape who you are? Without Barnaby Jack's Phrack article, I don't think I would've continued with Windows hacking. But even before that, Phrack was an important part of my "teenage education" -- I read voraciously, from stuff like the "Phone Losers of America" via "Confidence remains high" to 29A and VLAD. Textfiles were very formative for me. I was a teen while route ran Phrack, and I admired everybody that was involved. And to some extent the values touted by some old textfiles are values that seeped into all of us? |=---=[ What is your favourite bug/exploit? Each one has their own little story. They end up like pets. One of my favorites was a RealServer bug where RealServer would take a request URL and break it up into parts, splitting the string at '/', and putting a sequence of char* into an array of char* on the stack - so you'd overflow the stack buffer with char* to your own data, and you'd overwrite EIP with that pointer, so it was really "send the right string and watch it get executed". I remember Dave Aitel being confused about accidentally executing code while fuzzing *without doing* anything. Gorgeous bug. Cisco PIX had a funny bug in IKE where you could just pretend you had already passed the key exchange phase when you hadn't, and the implementation would start HMAC'ing your data, meaning copying it into the buffer for MD5Update - except that the pointers in that buffer had never been initialized. So if you managed to pollute the stack properly, you'd get a "write your data anywhere" primitive. When the /GS mitigations came in, the compilers would add stack canaries to many functions. But there is a funny (obvious, but not really documented) interaction between modern C++ programming and stack canaries: The stack canary is only active at the *end* of a function, but most functions receive pointers to stack-allocated C++ objects further up on the stack. The net result of this is that if you have an unbounded stack overflow of the type GS tries to mitigate, chances are you can overflow long enough to hit a vtable higer up on the stack. Adobe Reader had such a bug, and you got control on the last call before the GS check. Tremendously gratifying. Anything ever where you get to corrupt some sort of loopy bytecode. |=---=[ Will mitigations eventually make exploitation impossible? I think properly implemented fine-grained memory tagging can make the vast majority of memory corruption bugs impossible or at least fiendishly hard to exploit. That said - incentives are usually aligned against security, and it is not unlikely that we will simply never get ubiquitous memory tagging in all devices; things like "how do you do memory tagging in a heterogeneous compute environment where many different devices hack around in the same shared memory" are not always clear-cut. I don't think we have a much better mitigation on the cost/benefit tradeoff currently. Time will tell. To be perfectly honest, I have oscillated on this. I was doubtful about the future of buffer overflows as early as the 30th October 2000, and then as time passed I got very optimistic about their future. There will be other bugs, less powerful perhaps, that will still be worth exploiting. |=---=[ Would you recommend newcomers to contribute to open source projects? I recommend everybody contributes to open source. Spreading source code makes us all richer. I also understand that economics can dictate not sharing the source code -- so there is no moral judgement if you don't -- but I generally think that sharing code when you can is a fundamentally altruistic move that makes everybody better off. |=---=[ Your opinion in the infosec scene now vs then? We have all gotten much older ;) and we can now talk about health issues as much as bugs. Middle-aged mutant ninja turtles indeed. We need to recall that the 90s were a weird time, geopolitically, as some serious scholars seriously considered "the end of history" and that we'd all just trade and be happy, a form of capitalism-meets-hippie-ideals. 9/11 put an end to that, and the aftermath gave everybody stark reminders what power and steel and torture means. I share some of Gera's concerns about "mercenaries everywhere", I don't really like the militarization of everything. Then again, this might just be the way the world works: Any sufficiently useful technology will be co-opted by power and used in war. The early days of manned flight are a good example. We went from the Wright Brothers to WW1 dogfights in 16 years, and to the massacre at Guernica in 35 years, carpetbombing Germany's industrial centers in 40 years. Of course the structure and personalities of pilots changed through that. And those that were around when "flight" meant "running down a beautiful grassy you flap your arms to perhaps get 300m of lift", seeing jet fighters bomb civilians will always (and in some sense rightfully) seem like a terrible betrayal of the craft. Everyone is much more specialized now - the early 2010s still had people that could operate accross the full spectrum, from vuln-dev to persistence to doing actual hacking. This is getting rare, people are very much in their niche. Perhaps AI models provide some ability to become "well-rounded" again, as they can help skill up in foreign areas. I also think the scene is under harsher economic pressures. When there is no money, the weirdoes that everybody says have no "common sense" congregate and find joy. Then money arrives, and it changes both the dynamics but also the composition of the group. So everything is more professional, more structured, more organized, more employer-driven, people worry about career trajectories, and they expect to have a professional life in security. That's a very different vibe than a bunch of kids that were thrilled that anyone would pay them for opening IDA, and that could still avoid thinking about the consequences of everything, and that had no expectations any of this would last more than a few years. To finish: I used to joke that there are less than a dozen reasons why working in security is fun, and I know most of them personally. Then over time I had to increase that number, and nowadays I could probably name 3 dozen or more people whom I have gotten to know through security that I hold in high esteem. So while perhaps the percentage of people I like has gone down, the overall number has gone up, and since I am good at avoiding people I don't like, my life is a bit richer with the scene today than it was with the scene back then, by sheer numbers. |=---=[ Your opinion on conferences? There are how many conferences per day now? I like some of them. The best ones are still meetings in someone's attic with a bunch of people discussing stuff that isn't meant for broader consumption. They do exist. Then there are the offensive and research-heavy conferences, and finally the big "industry" conferences that are also sales events. They have their place, and I enjoy attending one or perhaps two per year. It's good to meet people in person, and it's motivating to be exposed to what everybody is doing. I am also happy whenever I see younger people than me do good work (by now almost everyone is younger ;). |=---=[ Recommendations Technical Books: "A New Era in Computing" - 1993, MIT Press. Best book. I steal ideas from it all the time. Beyond that: Too many to list or too few to list. I have shelves full of books I will never touch again, but can't bring myself to throw away. Other Books: Heinlein's "The Moon is a harsh mistress" Lem's "Memoirs found in a bathtub" Mathematical biographies or autobiographies: Ulam, Rota, Thorpe, Erdos, Friedman Anything by Giancarlo Rota Essays by Vladimir Arnold Anything by Serhij Zhadan Descartes Error A History of Computating in the 20th century Young researcher that impress(ed) me: Jann Horn is deeply impressive, and very unique. Many of my P0 colleagues were special and very unique, each in their own ways. Gal Beniamini was fast and deep, a force of nature. Many young folks I meet at offensive conferences are doing fascinating work. I watch the talks, and I feel clumsy and old. |=---=[ Reflections Hacker Spirit: Boundless curiosity, a desire to understand even if people say "no need to look at that", and a true desire for experimentation and learning by absorbing a topic and not just reading about it. As long as we can kindle curiosity and generous sharing of what we learn, the hacker spirit is alive. It's science, in the most basic, child-play, curiosity, joyful way. Experiment, learn, and explain. Exploit Industry: You don't argue with the existence of rain. You decide what level of exposure you can tolerate, and you act according to that. The only person that can judge whether you can still look into the mirror tomorrow is you. Make your 8 year old self proud, and make your 80 year old self proud. Early Programming: See above. Career and Burnout: In a career, you need to balance exploitation (earning money) with exploration (doing things that fill your heart with joy). I saw people burn out because they focused on exploitation too much. I always swung back and forth between tool building and vuln-dev, and I would not have been happy with only one of them. And even with that I eventuall did burn out. So I went and did something else for a while. The world is full of marvels, and if you're willing to go deep and understand and be helpful and kind, there is life outside of security if you ever need it. There's a helpful analogy: Your life is a long car ride, and each of the aspects of "you" is a passenger. You need to find a way to keep them all happy, and that begins by listening to them, and not locking one of them into the trunk for years. |=---=[ Insights Important principle: As a child I understood the "canary" principle: Enforcement of rules is often by making an example of the most egregious offender. If you're breaking the rules, don't be in the first line. And if the first line gets hit, hit the ground and stay there for longer than reasonable. Hacking Milestones: I recommend "not getting arrested, sued, or seriously harmed" is pretty high on the list to achieve before you're 25. "Make decisions to make your 8 year old self and your 80 year old self proud" is the milestone for 40. After that, be aware that you're sitting in failing machinery that needs serious maintenance. Nontraditional Hacking: Chiptuning of cars was entertaining. The demoscene was entertaining. The game cheating scene had some very strong technical folks. Do something new. Build something with your hands that nobody else has. The "Art" of Hacking: Finding unexpected uses of something, either for beauty, for fun, or actual utility / gain. |=---=[ Favorites Programming Languages: I actually like C++ these days. Stockholm syndrome. Pwnie Award: Yes. I got one, Lifetime Achievement, and it feels it was a lifetime ago. Best Hack: Detecting a read from the IDT and protecting my debug registers from being overwritten by setting the GD bit, then setting a breakpoint on memory read, trapping that, and feeding a fake value back to a piece of software that tried to verify the presence of an untampered device driver on Win9X. What a mess. Dongle copyprotection madness. Software: I loved SoftICE. I loved IDA. Nowadays, I just want a compiler and vim and the freedom to play. Websites: Gatherings: The dutch hacker camps. HAL, HIP, WTF etc. -- each time I go, they refresh my spirit and make me want to be around a bit longer. Locations: Bay of Biscay. Northern Spain. East Taiwan. Rocha in Uruguay. Portugal. Tanegashima. |=---=[ Personal Other Interests: I like lots of things. Surfing is a fascination, but I suck at it. Physical simulation is fascinating, but my Physics is poor. I love working with wood, but I suck at it. I like drawing, I suck a bit less. I like playing the guitar, I suck at it. In some sense, I enthusiastically suck at a lot of things, but I love the process of learning so much. Philosophy: "Mathematicians need pen, paper, and a wastebasket to do their work. Philosophies only need pen and paper." - and nowadays LLMs do both. Zines: Phrack. tmp.out. CRH, Phone Losers of America, zf0, el8, VLAD, 29A Zines are the best thing, and I could hug the younger generation for bringing zine culture back and nurturing it better than my generation ever did. |=---=[ Quotes "Theo, there are no more bugs in UNIX" -- Duke in conversation with Theo de Raadt, around 2001, when Theo tried to quiz him for bugs. |=---=[ Closing Thoughts This is pushing 1000 lines. I really tend to ramble when y'all let me. |=[ EOF ]=---------------------------------------------------------------=|