WEBVTT - Automatic transcript; recognition may contain errors. Reviewed for technical wording and speaker labels. Times match the published episode. 00:00:07.250 --> 00:00:09.910 Live from Utrecht, this is the Van Wirdum Sjorsnado. 00:00:10.190 --> 00:00:10.390 Hello. 00:00:11.380 --> 00:00:16.540 Sjors, we're recording this on Wednesday evening, Dutch time. Bitcoin just broke 00:00:16.720 --> 00:00:19.180 20k dollars. Are you ecstatic? 00:00:19.540 --> 00:00:19.900 No, no. 00:00:20.360 --> 00:00:23.300 Okay. This episode, we're going to discuss open source? 00:00:23.820 --> 00:00:24.300 Yes. 00:00:24.480 --> 00:00:28.140 I'm just going to skip over the whole price thing. We're going to discuss open 00:00:28.140 --> 00:00:32.800 source and why it's useful. Or free software and why it's useful. Are you on the 00:00:32.800 --> 00:00:34.800 free software train or on the open source train? 00:00:35.360 --> 00:00:38.600 I'm on every train. I like trains, but tell me. 00:00:38.740 --> 00:00:42.200 I can tell you the difference because a lot of people don't know this. There 00:00:42.200 --> 00:00:46.640 isn't a lot of difference except for like a philosophical difference. So the idea 00:00:46.640 --> 00:00:51.920 is that Richard Stallman, he founded the free software movement. And the idea 00:00:51.920 --> 00:00:57.280 there was if software is closed source, then there is sort of a power 00:00:57.280 --> 00:01:02.970 relationship between developers and users. Because users don't 00:01:02.970 --> 00:01:08.540 know what software they are running. And we'll get to this in a bit I guess, well 00:01:08.540 --> 00:01:12.660 I can explain this very briefly right now the reason is… because you know this 00:01:12.660 --> 00:01:16.820 but I'm explaining it to our listeners, yeah, the reason is that the actual 00:01:16.820 --> 00:01:21.140 software you're running on your computer are binaries they're ones and zeros 00:01:21.140 --> 00:01:27.140 that's the stuff computers can read while humans when they write software they 00:01:27.140 --> 00:01:32.810 write computer code and the two aren't the same thing. So when you're running 00:01:32.810 --> 00:01:38.450 closed software, you're just running the binaries and you're not exactly sure 00:01:38.450 --> 00:01:43.610 what your computer is actually doing. So what your computer could be doing, for 00:01:43.610 --> 00:01:48.350 example, is spy on you. If the developer put some malware into the closed 00:01:48.350 --> 00:01:54.270 software, then your computer could spy on you or could do all sorts of stuff that 00:01:54.270 --> 00:01:58.790 you don't actually want the software to do. So you have to trust the developer. 00:01:59.210 --> 00:02:02.770 In that sense, you have to trust the developer that he didn't include malware 00:02:02.770 --> 00:02:08.250 into your binaries, into your software. So Richard Stallman, he didn't like this 00:02:08.250 --> 00:02:12.230 idea. That wasn't his vision for the future. So he started a free software 00:02:12.230 --> 00:02:18.190 movement where the source code had to be available so people could actually check 00:02:18.190 --> 00:02:21.770 what they were running on their computer. So in that sense, there wasn't a power 00:02:21.770 --> 00:02:26.610 relationship anymore. They didn't need to trust their developer. So free in that 00:02:26.610 --> 00:02:31.280 context means freedom. It doesn't mean free as in free beer. 00:02:31.540 --> 00:02:36.620 Yeah, so in Dutch we have the word vrij, which means freedom, and gratis, which 00:02:36.620 --> 00:02:40.820 means free beer. So we can intuitively understand this. 00:02:40.960 --> 00:02:43.440 Yeah, we actually have two different words for that. Yeah, exactly. 00:02:44.280 --> 00:02:46.700 And I'm sure German has 27 words for it. 00:02:47.250 --> 00:02:47.770 Probably. 00:02:49.290 --> 00:02:54.610 Where was I? So yeah, that was Stallman's vision. Then I think in the early 00:02:54.610 --> 00:02:59.270 90s, there was a different, I can't remember the guy who wrote it, but there was 00:02:59.270 --> 00:03:04.090 this paper about the cathedral and the bazaar, the bazaar and the cathedral, 00:03:04.270 --> 00:03:09.090 something like that. It was like a Linux contributor. And he explained the 00:03:09.090 --> 00:03:13.790 benefits of free software, as it was just called until then, from a kind of 00:03:13.790 --> 00:03:19.360 different perspective, where he explained how… free software could 00:03:19.360 --> 00:03:24.080 actually provide high quality code because there's a lot of people checking the 00:03:24.080 --> 00:03:29.380 code. Enough eyeballs make all bugs shallow. That's the saying. So he came up 00:03:29.380 --> 00:03:33.740 with a more sort of pragmatic reason why free software was a good idea. This 00:03:33.740 --> 00:03:39.600 convinced the Netscape people to turn Netscape, the internet browser, into 00:03:39.600 --> 00:03:44.000 an open source project. I'm calling it open source now. And that was Firefox. And 00:03:44.000 --> 00:03:48.420 I'm calling it open source now because this group of people, they sort of 00:03:48.420 --> 00:03:53.900 rebranded free software to open source to sort of more accentuate 00:03:53.900 --> 00:03:57.500 these different benefits. So they weren't necessarily 00:03:58.900 --> 00:04:03.440 proposing or they weren't in favor of open sourcing software for this 00:04:03.440 --> 00:04:09.400 philosophical freedom reasons that Stallman was advertising, that he was 00:04:09.400 --> 00:04:13.340 promoting, but more this pragmatic attitude. So that's that's where sort of the 00:04:13.340 --> 00:04:16.960 difference between free software and open source stems from. 00:04:16.960 --> 00:04:20.180 Was there also a difference between freeware because of course that was a term 00:04:20.180 --> 00:04:22.860 that was going around too but freeware could still be closed source? 00:04:22.860 --> 00:04:28.450 I don't know. I do know that basically every free software project is also an 00:04:28.450 --> 00:04:33.050 open source project. There are very subtle differences for some of the licenses, 00:04:33.070 --> 00:04:37.590 but it's basically the same thing, just sort of explained with different 00:04:37.590 --> 00:04:38.430 philosophies. 00:04:38.770 --> 00:04:41.730 All right. Well, in our case, we're going to be quite pragmatic. So the 00:04:41.730 --> 00:04:43.110 terminology is less important. 00:04:43.350 --> 00:04:45.450 You want to just go with open source? 00:04:46.130 --> 00:04:46.610 Yes. 00:04:46.930 --> 00:04:50.750 Okay. So we're just going to discuss open source. Okay. So Bitcoin is an open 00:04:50.750 --> 00:04:51.290 source project? 00:04:51.740 --> 00:04:52.260 Yes. 00:04:52.460 --> 00:04:55.760 Why is that very important in the context of Bitcoin, Sjors? 00:04:56.160 --> 00:05:01.620 Well, imagine you're trying to use Bitcoin and you install a computer program and 00:05:01.620 --> 00:05:06.120 it gives you an address. And then it turns out there's some code in there that 00:05:06.120 --> 00:05:09.100 just steals your Bitcoin. That would be bad. 00:05:09.560 --> 00:05:10.360 Yeah, that would be bad. 00:05:10.540 --> 00:05:16.120 So at minimum, you want to know what code you're actually about to send your 00:05:16.120 --> 00:05:17.540 pension to. 00:05:17.900 --> 00:05:23.230 Yep. Well, this is actually… this is exactly… I just gave you the spying example 00:05:23.230 --> 00:05:26.570 of Stallman, but this would obviously be another great example where, you know, 00:05:26.570 --> 00:05:31.590 you don't want… you don't want to trust the developers to steal… to not steal 00:05:31.590 --> 00:05:32.270 your coins. 00:05:32.690 --> 00:05:37.510 It's a very extreme example. It makes it very clear why you really need the 00:05:37.510 --> 00:05:40.630 maximum transparency of what the hell is running on your machine. 00:05:40.830 --> 00:05:41.270 Exactly. 00:05:42.160 --> 00:05:44.120 And we're going to go down that rabbit hole a little bit. 00:05:44.300 --> 00:05:45.940 Yeah, because it's not that easy. 00:05:46.220 --> 00:05:46.980 No, it's not. 00:05:47.140 --> 00:05:51.500 It would be nice if it was that easy, but it's actually a lot harder than it 00:05:51.500 --> 00:05:56.120 sounds to make sure that the code on your computer is actually doing what you 00:05:56.120 --> 00:05:56.820 want it to do. 00:05:57.020 --> 00:06:01.780 Yeah, because one thing is you want whatever Bitcoin code is running to be open 00:06:01.780 --> 00:06:05.580 source so that you can see what it is. But most computer programs, as we talked 00:06:05.580 --> 00:06:09.560 about in the first episode, use libraries or dependencies. Use some other piece 00:06:09.560 --> 00:06:13.330 of software. That in turn uses some other piece of software that in turn uses 00:06:13.330 --> 00:06:14.330 some other piece of software… 00:06:14.330 --> 00:06:20.090 I want to… I want to take this one step at a time. First step. So the the code of 00:06:20.090 --> 00:06:24.290 Bitcoin it's open source, it's on the… it's hosted on the… it's GitHub 00:06:24.290 --> 00:06:24.950 repository. 00:06:24.950 --> 00:06:28.810 Yeah it's a Git repository which is also hosted on GitHub. 00:06:28.810 --> 00:06:34.590 Right. Okay sorry thanks for that correction. So it's on GitHub so anyone with 00:06:34.590 --> 00:06:39.840 the skills can look at this source code and you know check that it does what it's 00:06:39.840 --> 00:06:40.500 supposed to do, so… 00:06:40.500 --> 00:06:45.400 In addition, everybody who has that skill can compile it themselves rather than 00:06:45.400 --> 00:06:46.140 downloading it. 00:06:46.140 --> 00:06:51.400 Right. First question is yours, because I actually cannot read this at all. How 00:06:51.400 --> 00:06:53.480 many people do you think can actually read this? 00:06:53.480 --> 00:06:58.940 Well that depends on what you mean by actually read? I mean, how many people are 00:06:58.940 --> 00:07:03.620 computer literate in general? Probably many, many tens of millions in the world. 00:07:04.140 --> 00:07:09.560 How many can roughly read what a C program is doing? I guess, again, several 00:07:09.560 --> 00:07:13.300 million, tens of millions probably. But the number of people who can actually 00:07:13.300 --> 00:07:18.340 understand what the Bitcoin software is doing is probably a lot smaller. And the 00:07:18.340 --> 00:07:21.900 number of people who actually do, in addition to being able to, is in the dozens. 00:07:22.320 --> 00:07:28.080 And even then, it's hyper-specialized. So somebody might know everything 00:07:28.080 --> 00:07:33.120 about peer-to-peer networking code and have never looked at some other part of the code. 00:07:33.280 --> 00:07:38.720 So this sounds like a small number. Why is it so small and can it be bigger? How 00:07:38.720 --> 00:07:41.260 could we make this bigger in the future? 00:07:42.140 --> 00:07:47.880 Well, one thing is you can make source code more clean, more readable. So then 00:07:47.880 --> 00:07:50.600 there are just more people who can read it because it's just better. 00:07:50.820 --> 00:07:53.140 So why isn't that the case right now, for example? 00:07:53.940 --> 00:07:59.320 Well, it's better than it was. But when Satoshi wrote it, everything was one file 00:07:59.320 --> 00:08:05.060 with God knows how many lines of code in it. And that's very, very, very hard to 00:08:05.060 --> 00:08:05.540 reason about. 00:08:06.970 --> 00:08:10.830 For the non-programmers, that means, reason about means… 00:08:11.850 --> 00:08:16.710 You're looking at the code and you see, okay, there's a function called make a 00:08:16.710 --> 00:08:20.230 private key. Okay, what does that function do? Oh, calling this other function. 00:08:20.370 --> 00:08:24.050 Where's that other function? Oh, it's 20,000 lines up in the same file. Let me 00:08:24.050 --> 00:08:28.330 scroll 20,000 lines up, have a look at that code, and it's calling something 00:08:28.330 --> 00:08:32.690 else. Oh, but it's not calling something, it's referring to a variable. Oh, but 00:08:32.690 --> 00:08:37.000 this variable… can be accessed in like 15 different places at the same time 00:08:37.000 --> 00:08:38.440 somewhere in this file. 00:08:39.220 --> 00:08:42.040 Reason about means understand what the hell is going on. 00:08:42.240 --> 00:08:42.420 Yes. 00:08:42.740 --> 00:08:47.560 Right. Okay, so there's not that many, but hopefully it's improving or it's 00:08:47.560 --> 00:08:51.320 getting easier, but that's a work in progress and it's still pretty hard. 00:08:51.320 --> 00:08:54.080 Yeah, we're getting a little bit of help from all these altcoins, which are 00:08:54.080 --> 00:08:58.260 cloning the Bitcoin code. Not all altcoins are, but many are. And they're cloning 00:08:58.260 --> 00:09:01.760 it and they're working on it and they might occasionally find bugs too. Or at 00:09:01.760 --> 00:09:02.560 least they're looking at it. 00:09:02.780 --> 00:09:08.580 Right. Okay, so let's say I trust that this process where you know, 00:09:08.740 --> 00:09:11.920 a bunch of people can look at it and I kind of trust that they're not all 00:09:11.920 --> 00:09:17.080 cheating and this process is sort of working. At that point, I can 00:09:17.080 --> 00:09:23.360 download the binaries from BitcoinCore.org and I should be totally fine, right? 00:09:23.660 --> 00:09:23.940 No. 00:09:24.200 --> 00:09:26.140 Oh, what's the next problem? 00:09:26.900 --> 00:09:32.080 Well, I mean, there's a lot of problems. First of all, who says BitcoinCore.org 00:09:32.080 --> 00:09:34.500 is run by the same people you just mentioned? 00:09:36.250 --> 00:09:41.150 It might not be. Well, okay, maybe you can still prove that. But then maybe the 00:09:41.150 --> 00:09:45.230 site is hacked or the site isn't hacked, but the DNS is hacked. There's lots of 00:09:45.230 --> 00:09:47.930 reasons why the thing you download is not the thing you think you're downloading. 00:09:48.150 --> 00:09:53.310 It's called malware. So one thing that open source projects almost always do 00:09:53.310 --> 00:09:59.070 is publish a checksum, which is basically saying when you download this thing 00:09:59.070 --> 00:10:03.050 and you run this little script on it, it should have the following checksum. 00:10:03.430 --> 00:10:07.370 That's one thing you can do. But then can you trust the checksum that you 00:10:07.370 --> 00:10:10.930 downloaded? I don't know, because whoever hacked the site might have also hacked 00:10:10.930 --> 00:10:16.810 the checksum. So then what you do is you sign the checksum. So, for example, 00:10:16.870 --> 00:10:22.030 a well-known person, in this case, Wladimir van der Laan, he signs the checksum 00:10:22.030 --> 00:10:27.070 with a signature, with a key, with a PGP key that's publicly known. It's been the 00:10:27.070 --> 00:10:32.300 same for like 10 years. So then at least, you know, you kind of have something to check. 00:10:32.480 --> 00:10:37.560 Okay, so how does Wladimir know that the binaries he got 00:10:37.560 --> 00:10:41.220 actually reflect the open source code from the Git? 00:10:41.440 --> 00:10:45.780 Well, he knows because he did it. So he took the source code, he ran a command, 00:10:45.840 --> 00:10:46.860 and he got the binary. 00:10:47.300 --> 00:10:51.820 Yeah, and by he ran a command, you mean he put it through some other piece of 00:10:51.820 --> 00:10:56.320 software that… produces binaries from the open source software. 00:10:56.520 --> 00:10:58.360 Yeah, a compiler and a bunch of other tools. 00:10:58.520 --> 00:10:58.640 Yeah. 00:10:59.500 --> 00:11:02.280 So that's great. But then the question is, how do you know? 00:11:02.440 --> 00:11:02.660 Right. 00:11:03.300 --> 00:11:08.820 And here it gets a little bit more complicated. Ideally, what you do is you run 00:11:08.820 --> 00:11:13.800 the same command and you also compile it, and then hopefully you get the same result. 00:11:14.060 --> 00:11:14.460 Right. 00:11:14.940 --> 00:11:19.180 And sometimes that works with some projects, but as projects get really 00:11:19.180 --> 00:11:24.620 complicated, it often doesn't work because it can depend on some very specific 00:11:24.620 --> 00:11:29.160 details on your computer system what the exact binary file is going to be. So for 00:11:29.160 --> 00:11:33.420 example, the software uses libraries. And those libraries are living on your 00:11:33.420 --> 00:11:36.840 system. So we talked about that in one of the first episodes about libraries in 00:11:36.840 --> 00:11:40.320 general. These libraries might live in your system and these libraries get 00:11:40.320 --> 00:11:44.500 updated all the time. And maybe you updated two months ago and Wladimir is like 00:11:44.500 --> 00:11:50.320 very accurate and he updated yesterday. And so the final product contains a 00:11:50.320 --> 00:11:53.820 different version of a library. And if you only change one letter in a computer 00:11:53.820 --> 00:11:57.850 program, then boom, your checksum doesn't work anymore. So that's one of the 00:11:57.850 --> 00:11:58.590 things that can go wrong. 00:11:58.690 --> 00:12:03.490 Hang on. One step at a time. Why do I even need to care that my checksum matches 00:12:03.490 --> 00:12:06.610 whatever Wladimir signed if I compiled it myself? 00:12:06.970 --> 00:12:09.870 If all you want is to compile it yourself, you don't care. 00:12:10.250 --> 00:12:10.370 Right. 00:12:10.570 --> 00:12:16.370 But basically what this whole mechanism relies on is that some people check. And 00:12:16.370 --> 00:12:20.900 that if some people find a problem, they're going to sound the alarm bell. And so 00:12:20.900 --> 00:12:25.840 your security model kind of depends on hoping that somebody will do this checking 00:12:25.840 --> 00:12:27.700 for you because you didn't compile it yourself. 00:12:28.080 --> 00:12:33.760 Right. Everyone who compiles the software should compile into the same checksum 00:12:33.760 --> 00:12:37.000 because that's how you know everyone's running the same software and no one's 00:12:37.000 --> 00:12:38.300 being [unclear]… 00:12:40.070 --> 00:12:42.430 Malicious software, for example. Yeah, so you, as somebody who wants to make sure 00:12:42.430 --> 00:12:45.350 there are no shenanigans going on, you go to BitcoinCore.org, you download the 00:12:45.350 --> 00:12:49.050 binary, you just put it in a nice place, and then you compile it yourself and you 00:12:49.050 --> 00:12:53.110 see, hey, is this the same? If not, you go on Twitter and on the news media and 00:12:53.110 --> 00:12:54.750 you say, hey, there's malware on this website. 00:12:55.010 --> 00:12:55.210 Right. 00:12:55.610 --> 00:12:59.670 However, that is not trivial because of, for example, these libraries that might 00:12:59.670 --> 00:13:02.830 be slightly different. So you get a different checksum, even though there are no 00:13:02.830 --> 00:13:05.150 shenanigans going on. It's just your computer is different. 00:13:05.560 --> 00:13:11.470 Right, so if I compile the Bitcoin Core software on my… and you 00:13:11.470 --> 00:13:16.350 compile it on your MacBook, they could still compile into different binaries? 00:13:16.510 --> 00:13:19.390 Yes, because there might be some subtle differences. And it's not just libraries. 00:13:19.390 --> 00:13:25.250 It can even be the time of your computer. So because as you're copying stuff, 00:13:25.570 --> 00:13:30.370 there's some random output that contains a timestamp, maybe a log. And if the log 00:13:30.370 --> 00:13:33.490 is included in the final product, then there's a different timestamp in your 00:13:33.490 --> 00:13:36.450 version than in my version because we didn't compile it exactly at the same time. 00:13:38.270 --> 00:13:39.530 So that's a problem. 00:13:39.710 --> 00:13:43.870 Right, so somehow we need to make sure that the same source code compiles into 00:13:43.870 --> 00:13:45.350 the exact same binaries. 00:13:45.570 --> 00:13:49.410 Yeah, at least if your goal is to verify that nothing went wrong, right? Because 00:13:49.410 --> 00:13:53.050 normally if you're just using it yourself, you don't care about that. So this 00:13:53.050 --> 00:13:57.570 phenomena is called deterministic builds. So deterministic really just implies 00:13:57.570 --> 00:14:01.350 given a source, you're going to get the same binary, and if you change one letter 00:14:01.350 --> 00:14:04.390 in the source, you're going to get a different binary, but everybody will get the 00:14:04.390 --> 00:14:06.750 same, basically, if they make the same change. 00:14:06.890 --> 00:14:07.750 So how is this done? 00:14:08.520 --> 00:14:13.840 So this is difficult. And the current way that Bitcoin Core is doing this is 00:14:13.840 --> 00:14:19.100 called Gitian. And just to sum that up, it's basically, you take an Ubuntu 00:14:19.100 --> 00:14:24.500 machine, could be a virtual machine, or could be a real machine. And you just 00:14:24.500 --> 00:14:28.460 take a very specific Ubuntu version, I think, that you download. And many people 00:14:28.460 --> 00:14:32.800 in the world have seen that version, so you kind of trust it. And then inside 00:14:32.800 --> 00:14:37.880 that machine, you build another virtual machine. And inside… that virtual 00:14:37.880 --> 00:14:40.880 machine, there are all sorts of little changes made to make sure that that 00:14:40.880 --> 00:14:45.500 machine is identical for everyone who builds this thing. So I think it uses a 00:14:45.500 --> 00:14:49.020 fake time and all the files are in the same place and all the libraries are the 00:14:49.020 --> 00:14:54.860 exact same versions, etc. And then you build Bitcoin Core. And then you look at 00:14:54.860 --> 00:14:56.580 the checksums inside that virtual machine. 00:14:57.810 --> 00:15:03.490 Right. Yeah. So it's kind of like running a computer within your actual computer. 00:15:03.710 --> 00:15:04.670 Within a computer. 00:15:04.950 --> 00:15:08.690 And everyone's sort of running the same computer within the computer, in their 00:15:08.690 --> 00:15:13.410 actual computer. And therefore the software they're compiling into binaries in 00:15:13.410 --> 00:15:17.310 that computer, and the computer is resulting in the exact same binaries. 00:15:17.530 --> 00:15:17.930 That's right. 00:15:18.130 --> 00:15:23.620 This is turned into a checksum. And if the checksums match, then the 00:15:23.620 --> 00:15:29.600 developers sign because they can all verify that, yep, it all worked out 00:15:29.600 --> 00:15:33.860 and this is the correct checksum and you can trust this because we're not all 00:15:33.860 --> 00:15:36.680 going to cheat on you. Unless they are, but at least… 00:15:36.680 --> 00:15:37.340 But then you can catch them. 00:15:37.580 --> 00:15:37.740 Yeah. 00:15:38.040 --> 00:15:40.920 Anybody has the opportunity to see that there's a shenanigan going on. 00:15:41.020 --> 00:15:43.940 Anyone can follow this process and catch that something… 00:15:44.120 --> 00:15:47.790 Yeah, in theory. In practice, it's a pain. It's a huge pain to get the system 00:15:47.790 --> 00:15:51.330 working. There's not many open source projects that use this. As far as I know, 00:15:51.410 --> 00:15:54.490 Bitcoin Core and Tor do. Maybe a few others, but not a lot. 00:15:54.830 --> 00:15:56.490 Maybe some other cryptocurrencies. 00:15:56.870 --> 00:16:00.250 Some, but a lot of them, even if they've cloned Bitcoin Core, they stopped doing 00:16:00.250 --> 00:16:01.790 this process because it's too much work. 00:16:02.090 --> 00:16:02.210 Right. 00:16:03.010 --> 00:16:04.150 But, okay. 00:16:04.930 --> 00:16:05.930 So far, so good. 00:16:06.110 --> 00:16:10.190 So it's great, but there is another problem because the rabbit hole is deeper. 00:16:11.320 --> 00:16:14.360 And there's actually two different problems, but they're kind of the same. So 00:16:14.360 --> 00:16:17.720 let's start with the first thing. Let's say you have read every single line of 00:16:17.720 --> 00:16:22.120 code in Bitcoin Core. And you can say, okay, I've read every single line in 00:16:22.120 --> 00:16:24.920 there. I understand every single line of it. It's just like, you know, when you 00:16:24.920 --> 00:16:28.580 read the Facebook terms and conditions. But then it turns out the Facebook terms 00:16:28.580 --> 00:16:32.660 and conditions point to some other document. Like, for example, I don't know, the 00:16:32.660 --> 00:16:34.720 United States law. All of it. 00:16:36.590 --> 00:16:41.750 So, you know, with phrases like as defined in law. Now you have a problem because 00:16:41.750 --> 00:16:45.830 Bitcoin Core uses all sorts of other things. And so you have to inspect those 00:16:45.830 --> 00:16:46.370 things too. 00:16:46.850 --> 00:16:49.010 Yeah, dependencies, these are called. 00:16:49.210 --> 00:16:54.510 Because the dependencies could also be stealing your coins. So they should be 00:16:54.510 --> 00:16:55.150 open source too. 00:16:56.480 --> 00:16:59.440 Bitcoin Core doesn't use that many dependencies anymore, right? 00:16:59.720 --> 00:17:03.460 Exactly. So one of the constraints when working on Bitcoin Core is to try and 00:17:03.460 --> 00:17:07.860 keep the number of dependencies as small as possible and also not update them all 00:17:07.860 --> 00:17:11.100 the time. Because of course, the people who maintain those dependencies know that 00:17:11.100 --> 00:17:15.300 Bitcoin Core is using it, right? So you need to be somewhat on your toes to make 00:17:15.300 --> 00:17:17.740 sure that those projects are being scrutinized too. 00:17:17.920 --> 00:17:23.720 So let's say some dependency is corrupted. What could that mean for Bitcoin? 00:17:24.300 --> 00:17:26.220 Could they… Go on. 00:17:26.740 --> 00:17:29.460 Yeah, if some dependency is corrupted, it could steal your coins. 00:17:29.700 --> 00:17:30.740 Right. That bad. 00:17:31.280 --> 00:17:32.580 Yeah, that's your worst case. 00:17:32.760 --> 00:17:32.960 Okay. 00:17:33.240 --> 00:17:38.810 And this actually happened, at least in another project, called Copay, which is 00:17:38.810 --> 00:17:43.550 a… I think it's a library for wallets in general, used by BitPay, but by other 00:17:43.550 --> 00:17:44.150 companies too. 00:17:44.410 --> 00:17:44.670 Right. 00:17:45.050 --> 00:17:48.850 And it's written in a different programming language, but the general idea is the 00:17:48.850 --> 00:17:52.330 same. You know, they have a piece of software that's open source. Everybody can 00:17:52.330 --> 00:17:56.890 review it, but it uses dependencies. And those dependencies use dependencies and 00:17:56.890 --> 00:18:00.350 those and those and those and those. And in this case, they were using npm, the 00:18:00.350 --> 00:18:05.620 node package manager. And Node.js and that package manager is basically a very 00:18:05.620 --> 00:18:10.060 large open source community. And they've very much focused on making very modular 00:18:10.060 --> 00:18:15.480 packages. So I think there's an individual package for just addition or 00:18:15.480 --> 00:18:21.300 multiplication or fairly trivial packages. And every single package links 00:18:21.300 --> 00:18:25.380 to a place on GitHub. So it's all open source. And every package could have its 00:18:25.380 --> 00:18:29.560 own maintainer who can release updates whenever they want. And so now you have a 00:18:29.560 --> 00:18:33.760 problem because you might be pulling in 10,000 dependencies without even 00:18:33.760 --> 00:18:37.700 realizing it because you only pull in maybe five dependencies, but those each 00:18:37.700 --> 00:18:42.300 pull in 50 dependencies and those each pull in another 50 dependencies. 00:18:42.420 --> 00:18:48.110 And if any of these is corrupted, it could… At least theoretically, include 00:18:48.110 --> 00:18:49.490 coin-stealing malware. 00:18:50.050 --> 00:18:54.190 Yeah, and so it depends. I mean, there are some ways in theory that you can try 00:18:54.190 --> 00:18:58.550 to avoid that by encapsulating, by saying, okay, this piece of code, I'm going to 00:18:58.550 --> 00:19:01.510 run that code. I don't trust it, but I'm going to put it in some place where it 00:19:01.510 --> 00:19:06.130 cannot do anything other than that I want it to do. But with JavaScript, which is 00:19:06.130 --> 00:19:10.450 what they're using, at least at the time, this is two years ago, that was very 00:19:10.450 --> 00:19:16.290 difficult to do. So any JavaScript… that is run can do anything in the 00:19:16.290 --> 00:19:21.110 entire browser. So in this case with the Copay wallet there would be private keys 00:19:21.110 --> 00:19:25.890 somewhere inside the browser and a piece of malware could just say 00:19:25.890 --> 00:19:29.590 window.blahblahblah.stealcoins basically. 00:19:30.610 --> 00:19:32.350 And did this actually happen? 00:19:32.280 --> 00:19:36.530 Or, well, somebody wrote that malware. I don't think it was exploited in the 00:19:36.530 --> 00:19:39.500 wild. I think it was detected… on… someone… someone… 00:19:39.010 --> 00:19:44.390 Because I vaguely remember this, but someone actually got this kind of malware 00:19:44.390 --> 00:19:49.530 into the Copay wallet, basically, into the Copay library. This was actually done. 00:19:49.550 --> 00:19:52.670 In a dependency of a dependency of a dependency. So what they did is they found 00:19:52.670 --> 00:19:57.570 some random faraway dependency down, deep down in the tree right, that's actually 00:19:57.570 --> 00:19:59.350 used by millions of projects. 00:20:01.060 --> 00:20:05.440 And that dependency was no longer maintained. So somebody wrote it, everybody 00:20:05.440 --> 00:20:11.200 uses it, and then the guy or girl no longer maintains it. And so the attacker 00:20:11.200 --> 00:20:15.000 sent an email to that previous maintainer saying, hey, I really love your 00:20:15.000 --> 00:20:19.780 project, I care about this, maybe I can take over from you. And so he got the 00:20:19.780 --> 00:20:24.840 keys to the kingdom, and he was able to publish updates. And so then he published 00:20:24.840 --> 00:20:29.080 an update that contained some malware. Some coin stealing code that was 00:20:29.080 --> 00:20:33.780 specifically designed to attack Copay wallets or the general library. 00:20:35.200 --> 00:20:39.580 And what it did was, well, first of all, he hid it. So, you know, it's open 00:20:39.580 --> 00:20:43.880 source, but if you release an update, you can do different versions. So you can 00:20:43.880 --> 00:20:46.600 do a minor update saying, oh, this is just a small change and you can do major 00:20:46.600 --> 00:20:52.590 updates. And most software will constrain these updates. So it will 00:20:52.590 --> 00:20:56.150 automatically update for you, but only for minor updates. It won't automatically 00:20:56.150 --> 00:21:01.990 upgrade a major upgrade. So what the attacker did is he made a minor update, and 00:21:01.990 --> 00:21:06.990 then immediately afterwards made a major update, which undid the attack. So if 00:21:06.990 --> 00:21:10.670 you looked at the most recent version of the code, there would not be any attack 00:21:10.670 --> 00:21:13.790 code in there. So anybody inspecting the open source would say, okay, this is 00:21:13.790 --> 00:21:19.130 fine. But if you looked at the specific minor version that was being used, then 00:21:19.130 --> 00:21:24.230 it was there. So you don't only have to review the most recent source codes, you 00:21:24.230 --> 00:21:29.070 have to review the source code specific version that you're using for all of your 00:21:29.070 --> 00:21:30.830 dependencies. It's completely impossible. 00:21:31.130 --> 00:21:36.510 Right. So Copay was open source, but because of these dependencies and the 00:21:36.510 --> 00:21:40.610 dependencies on the dependencies, it's still not going to solve your problems. 00:21:40.870 --> 00:21:45.470 No, and it was very, very subtle, right? I guess it was found by somebody very, 00:21:45.470 --> 00:21:48.990 very carefully looking for this sort of stuff because it's very hard to stumble 00:21:48.990 --> 00:21:54.830 into it. And it was even made that it wouldn't reveal itself early because what 00:21:54.830 --> 00:21:59.070 you kind of want to do as an attacker is you want to look for a very big bounty 00:21:59.070 --> 00:22:03.130 and then take that. Because as soon as you start stealing coins, if you only 00:22:03.130 --> 00:22:07.030 steal one Satoshi and the person losing that one Satoshi notices it, they're 00:22:07.030 --> 00:22:09.250 going to sound the alarm bell and then people are going to start looking where 00:22:09.250 --> 00:22:13.410 the malware is, knowing that it exists and they're going to find it. So it was 00:22:13.410 --> 00:22:16.510 basically it had a condition in there that says it has to be at least a couple of 00:22:16.510 --> 00:22:20.830 Bitcoin in there. And only then am I going to attack. And I guess that never 00:22:20.830 --> 00:22:24.550 happened, fortunately, because they [unclear] it on time. But this is a risk. 00:22:24.750 --> 00:22:28.150 So what's the solution? Just not depend on dependencies? 00:22:28.650 --> 00:22:32.130 Pretty much. I mean, I think there are now, you know, over the last couple of 00:22:32.130 --> 00:22:36.650 years, there's some companies that will sort of screen as a service that might 00:22:36.650 --> 00:22:39.930 actually go through all these dependencies. But what you really want to do is you 00:22:39.930 --> 00:22:43.910 want to have very few dependencies. And especially you want to stay away from 00:22:43.910 --> 00:22:47.850 things that have nested dependencies. So in the case of Bitcoin Core, it's not 00:22:47.850 --> 00:22:52.570 too bad. It has, I think, about 10 dependencies that do not have a bunch of 00:22:52.570 --> 00:22:53.390 nested dependencies. 00:22:54.150 --> 00:22:54.550 Right. 00:22:54.550 --> 00:22:59.130 It's not a big tree. It's relatively shallow. So you'd have to go after those 00:22:59.130 --> 00:23:02.800 dependencies directly to attack. So that's good news. 00:23:03.440 --> 00:23:09.230 Okay, sure. I think that part is clear. Now, I have another question for you. I 00:23:09.230 --> 00:23:13.450 think you have a vague idea where I'm going because we already discussed exactly 00:23:13.450 --> 00:23:17.690 where we're going. We just discussed how you have deterministic builds and how 00:23:17.690 --> 00:23:23.330 different developers are all using this Gitian building thing to get the exact 00:23:23.330 --> 00:23:29.260 same binaries and sign all of that. Now, here's my next question. What if the 00:23:29.260 --> 00:23:34.120 Gitian building process itself is corrupted somehow? Is that possible? 00:23:34.790 --> 00:23:40.650 Yeah, or specifically Gitian uses Ubuntu. And what if somebody says, hey, this 00:23:40.650 --> 00:23:43.850 Bitcoin project is pretty cool. This Ubuntu project is pretty cool. Let me 00:23:43.850 --> 00:23:49.450 contribute some source to Ubuntu. And now when everybody runs their Gitian 00:23:49.450 --> 00:23:55.240 builder, which includes Ubuntu, there is a compiler on Ubuntu. And maybe that 00:23:55.240 --> 00:23:59.920 compiler is modified to, if it compiles Bitcoin Core, it actually adds some code 00:23:59.920 --> 00:24:00.720 to steal coins. 00:24:02.250 --> 00:24:02.610 Right. 00:24:02.730 --> 00:24:05.390 It would be very, very scary. Because you still have deterministic builds, 00:24:05.450 --> 00:24:07.990 because everybody would be using the same malware to build it. 00:24:08.470 --> 00:24:12.450 Yeah, I guess that's a dependency in itself then, right? That's like a dependency 00:24:12.450 --> 00:24:14.810 for Ubuntu, or am I saying that right? 00:24:14.850 --> 00:24:18.130 I guess there's two kinds of dependencies. One is the dependency you're actively 00:24:18.130 --> 00:24:21.330 running, you know, that's inside the binary that you're shipping to your 00:24:21.330 --> 00:24:24.850 customers. But the other dependency, and that's a real can of worms, are all the 00:24:24.850 --> 00:24:26.830 tools that you're using to produce the binary. 00:24:26.990 --> 00:24:27.330 Right. 00:24:27.590 --> 00:24:29.390 And even to download the binary, but yeah. 00:24:29.730 --> 00:24:35.570 Yeah, so if the tools you use to build Bitcoin Core… is corrupted then you still 00:24:35.570 --> 00:24:39.030 have a problem because all of the developers are getting the self the same 00:24:39.030 --> 00:24:43.170 binaries from their Gitian process but if that's corrupted anyways I think our 00:24:43.170 --> 00:24:44.010 listeners get it. 00:24:44.010 --> 00:24:47.110 So what's… what you're hoping is that the people who are maintaining all these 00:24:47.110 --> 00:24:50.650 compilers and all the other things know what they're doing and would never let 00:24:50.650 --> 00:24:54.210 any such a backdoor through. But, you know, that would be boring. So how do we 00:24:54.210 --> 00:24:54.890 get more paranoid? 00:24:54.890 --> 00:24:57.970 How do we get more paranoid? How do we solve this problem? 00:24:57.970 --> 00:25:03.630 Well, the key there is to make everything open source, and 00:25:03.630 --> 00:25:08.670 everything a deterministic build. So not just Bitcoin is a deterministic build, 00:25:08.830 --> 00:25:12.770 but every dependency of Bitcoin is a deterministic build. And every tool that is 00:25:12.770 --> 00:25:17.410 used to build Bitcoin is a deterministic build, including the compiler. And this 00:25:17.410 --> 00:25:23.020 is where we introduce Guix. This is a project Carl Dong has been working on and 00:25:23.020 --> 00:25:25.820 has given several talks on that we'll probably link to in the show notes. 00:25:26.060 --> 00:25:29.020 Yeah, Carl Dong from, he's with Chaincode Labs, right? 00:25:29.220 --> 00:25:29.340 Yeah. 00:25:29.860 --> 00:25:35.360 Yeah, so the trick then, it's a difficult trick… Well, it sounds very difficult 00:25:35.360 --> 00:25:41.160