WEBVTT 00:00:00.149 --> 00:00:14.099 prerol music 00:00:14.099 --> 00:00:23.599 Herald: So a very warm welcome to Thomas Roth. He is a security researcher and his 00:00:23.599 --> 00:00:28.980 specialty is exploiting techniques and reverse engineering and industrial 00:00:28.980 --> 00:00:37.590 security. And the talk today will be about out SCADA the gateway to shell. 00:00:37.590 --> 00:00:45.364 applause 00:00:45.364 --> 00:00:50.390 And just one little notice: this talk will be in English and will be translated 00:00:50.390 --> 00:00:53.580 in German as well. Thomas Roth: Thank you. 00:00:53.580 --> 00:00:55.090 Herald: Yes. 00:00:55.090 --> 00:00:59.290 Thomas Roth: Awesome, thank you. OK, yeah. Welcome to my talk gateway to shell. Who 00:00:59.290 --> 00:01:03.850 am I? He already introduced me, but still my name is Thomas Roth. I'm a security 00:01:03.850 --> 00:01:08.850 researcher. I do a lot of low level security, so a lot of ARM reverse 00:01:08.850 --> 00:01:13.321 engineering, Coldfire and so on. And yeah, you can find me on Twitter or if you 00:01:13.321 --> 00:01:20.730 want to write me an email. Feel free to send me one to thomas@stacksmashing.net. 00:01:20.730 --> 00:01:25.720 Before we start a short introduction to the background of this talk, so, this year 00:01:25.720 --> 00:01:30.830 I did some SCADA penetration tests and I found that while the PLC sensors 00:01:30.830 --> 00:01:35.210 are pretty well covered in the security research area, I found that all the small 00:01:35.210 --> 00:01:39.810 devices that surround SCADA environments are not really well covered. So basically 00:01:39.810 --> 00:01:44.060 we have the big Siemens PLCs and so on, and there's a lot of research going on 00:01:44.060 --> 00:01:48.760 about them. But there are also a ton of other small Ethernet devices involved in 00:01:48.760 --> 00:01:56.700 industrial networks that are not really researched very well yet. And all devices 00:01:56.700 --> 00:02:00.570 that we're going to talk about are running their latest respective firmware. 00:02:00.570 --> 00:02:07.310 Unfortunately, there will be zero days and these are not theoretical attacks. Like if 00:02:07.310 --> 00:02:12.489 you go to Shodan or similar search engine, you can find tens of thousands of these 00:02:12.489 --> 00:02:18.359 devices vulnerable and open in the Internet. So let me give you a quick 00:02:18.359 --> 00:02:24.779 introduction into the terminology in SCADA, because I know in the title I say 00:02:24.779 --> 00:02:29.079 SCADA, but actually it should be ICS, which stands for industrial control 00:02:29.079 --> 00:02:36.900 systems, because basically ICS describes the whole system from your supervision, 00:02:36.900 --> 00:02:42.069 the big room with all the big screens up to your PLCs the sensors, the actors and 00:02:42.069 --> 00:02:46.930 so on that you will find in your installation. And the term SCADA just 00:02:46.930 --> 00:02:50.959 describes the supervision and control centers. So the big screens that you might 00:02:50.959 --> 00:02:55.040 know from movies and so on, where when the bad guy comes, suddenly all the lights 00:02:55.040 --> 00:03:02.400 turn red. Then there's something called a PLC, which is programable logic 00:03:02.400 --> 00:03:06.889 controller. It's basically like an Arduino, just for industrial applications 00:03:06.889 --> 00:03:11.909 and they are really easy to program and you can get them from Siemens or Schneider 00:03:11.909 --> 00:03:17.610 and so on and so forth. Then there is something called an RTU, a remote terminal 00:03:17.610 --> 00:03:22.279 unit, which is a small device that generally are, well, back in the day, was 00:03:22.279 --> 00:03:27.029 only used for monitoring. But today you can actually program a lot of RTUs. So 00:03:27.029 --> 00:03:33.280 it's kind of a mix between a PLC and an RTU. So it's basically a PLC in a remote 00:03:33.280 --> 00:03:41.269 location. Alrighty, to the actual topic, industrial control gateways. So when you 00:03:41.269 --> 00:03:45.870 look at industrial control network, you'll find that there are a lot of different 00:03:45.870 --> 00:03:50.281 sensors and actors and a lot of them speak different protocols. So, for example, some 00:03:50.281 --> 00:03:56.470 might be serial, some might be IP, some might be Modbus and so on. And so you can 00:03:56.470 --> 00:04:01.459 buy these small gateways that connect all these different protocols to an IP 00:04:01.459 --> 00:04:06.539 network. So, for example, via Ethernet or even via GPRS or Wi-Fi and so on. And I've 00:04:06.539 --> 00:04:11.829 seen them in almost any industrial installation that I've seen. So, for 00:04:11.829 --> 00:04:16.440 example, they're used in power plants. They are used in water dam control 00:04:16.440 --> 00:04:22.880 systems. They are used to control the power grid and so on. And the security 00:04:22.880 --> 00:04:27.080 concept is, "Hey, but these devices are airgapped!", so it doesn't matter really 00:04:27.080 --> 00:04:31.599 if they are vulnerable or not fully up to date and so on, but that's not really true 00:04:31.599 --> 00:04:34.919 because a lot of these devices, while they might be airgapped, they also have 00:04:34.919 --> 00:04:42.650 antennas and they are interconnected by a ton of different wireless protocols such 00:04:42.650 --> 00:04:50.970 as Wi-Fi, LoRa or GSM or even proprietary radio links. So, yeah, and even the 00:04:50.970 --> 00:04:54.940 case studies show that basically in this case, you would have a monitoring network 00:04:54.940 --> 00:04:59.650 that's connected via the cellular network to control the water mains and so on and 00:04:59.650 --> 00:05:04.889 check the pressure. Or even worse, they even recommend that you connect the actors 00:05:04.889 --> 00:05:10.430 like valves and water level gotchas and so on over GPS, which we know is not a secure 00:05:10.430 --> 00:05:17.830 protocol to do anything that could be critical. Or you have stuff like 00:05:17.830 --> 00:05:24.160 water storage tanks that are controlled via Wi-Fi and so on or even public in the 00:05:24.160 --> 00:05:33.349 Internet. So, yeah, these devices are airgapped? Nope. So attacking in the field 00:05:33.349 --> 00:05:37.889 I already mentioned, if you go to Shodan, you will find a ton of different 00:05:37.889 --> 00:05:42.970 devices reachable via the Internet and even via GPS. So if you live 00:05:42.970 --> 00:05:49.090 close to, for example, a dam or something, it's kind of interesting to look at an SDR 00:05:49.090 --> 00:05:52.290 or similar radio equipment to see what's going over the airwaves, because you will 00:05:52.290 --> 00:05:59.090 find a ton of interesting stuff and sometimes, you can even very trivially get 00:05:59.090 --> 00:06:03.930 a physical access to the in field devices because they might just be in a white box 00:06:03.930 --> 00:06:07.569 somewhere hidden. And if you break into it, you can pull out the SIM card and it 00:06:07.569 --> 00:06:12.419 will put you directly into the SCADA network, if you're lucky. Don't do that, 00:06:12.419 --> 00:06:13.760 by the way. 00:06:13.760 --> 00:06:17.169 laughter 00:06:17.169 --> 00:06:24.539 So, yeah, let's let's hack some gateways. So the equipment you will need to and 00:06:24.539 --> 00:06:29.039 everything in this talk was done on this desk, just using these devices here, you 00:06:29.039 --> 00:06:33.000 really just need a laptop, you need an oscilloscope or similar measurement 00:06:33.000 --> 00:06:37.320 equipment just to ensure that you don't burn out your logic analyzer. You need a 00:06:37.320 --> 00:06:43.230 logic analyzer, a soldering iron, a multimeter and a power supply. And that's 00:06:43.230 --> 00:06:48.360 really basically it, because you can hack almost any embedded device that's using 00:06:48.360 --> 00:06:56.530 these devices and to find potential targets. I have this kind of map where 00:06:56.530 --> 00:07:02.139 first try to understand, can I get the firmware of the device or do I have to 00:07:02.139 --> 00:07:07.251 somehow, for example, use J-Tech to get it out of the device? Can I actually buy the 00:07:07.251 --> 00:07:12.460 devices at a sensible price? Because some of these devices cost like 600 € or so, 00:07:12.460 --> 00:07:18.360 and if you buy ten of them, that gets expensive very quickly. And so, uh, I need 00:07:18.360 --> 00:07:24.360 to check eBay and see what devices can I actually buy. And they should be half what 00:07:24.360 --> 00:07:29.449 current, because if you look at all the devices, like 10 years old or so, they are 00:07:29.449 --> 00:07:33.940 completely broken. You don't even have to look to start to look at their security. 00:07:33.940 --> 00:07:40.819 So, yeah, the first device that I that I choose to really look at was the moxa 00:07:40.819 --> 00:07:50.689 W2150A, which is this small device, which is also mounted on the board right here, 00:07:50.689 --> 00:07:54.319 mainly because I found the phone was available and it looked like an 00:07:54.319 --> 00:07:58.669 interesting device because it has Wi-Fi and so if I managed to break into it, I 00:07:58.669 --> 00:08:07.930 can jump an airgap potentially. And the W2150A is just a simple device server. So 00:08:07.930 --> 00:08:14.939 you can connect any serial device, any RS485 device simply to it and it will be 00:08:14.939 --> 00:08:20.669 exposed via Ethernet or even via Wi-Fi. And you can download the firmware publicly 00:08:20.669 --> 00:08:29.270 and it's available on eBay relatively cheap. So like 150 bucks or something. So 00:08:29.270 --> 00:08:33.290 I downloaded the firmware and I looked at the entropy of the firmware and 00:08:33.290 --> 00:08:37.090 I immediately saw that the entropy is very high, which means either it's very 00:08:37.090 --> 00:08:41.590 compressed or it's encrypted, unfortunately, using a tool called 00:08:41.590 --> 00:08:46.510 binwalk, which is really useful for looking into firmwares I saw that there's 00:08:46.510 --> 00:08:51.510 no compression detected. And so it was very likely that this firmware image is 00:08:51.510 --> 00:08:59.940 encrypted. But I noticed on the Web page that before you upgrade to version 2.0 or 00:08:59.940 --> 00:09:08.650 2.1 of the firmware, you must upgrade to the firmware version 1.11. And I thought, 00:09:08.650 --> 00:09:13.540 that's interesting. Let's look at the release notes for version 1.11. And it 00:09:13.540 --> 00:09:22.330 turns out that 1.11 adds the support for the encrypted firmware. So I downloaded 00:09:22.330 --> 00:09:28.093 the one point eleven firmware and sure enough, it's unencrypted. And if you've 00:09:28.093 --> 00:09:33.710 ever done anything with ARM before, if you just look into a firmware hex dump, you 00:09:33.710 --> 00:09:39.930 can immediately recognize whether it's ARM or not, because the first four bits of each 00:09:39.930 --> 00:09:45.580 instructions are the conditional bits and those are almost always E. So if 00:09:45.580 --> 00:09:50.320 you see a Hexdump and roughly every fourth byte is an E, you know, this is an ARM 00:09:50.320 --> 00:09:57.550 firmware and it's not encrypted or anything else. And so, yeah, sure enough, 00:09:57.550 --> 00:10:02.640 I ran binwalk on this image. This time we see there is a huge drop in entropy, which 00:10:02.640 --> 00:10:08.570 is the bootloader and so on, and then a high entropy, which is basically the all 00:10:08.570 --> 00:10:15.280 the compressed filesystems and so on. And binwalk was able to detect the SquashFS 00:10:15.280 --> 00:10:22.060 filesystem and extract it for me very, very easy. And so my goal was to extract 00:10:22.060 --> 00:10:27.250 the firmware, find the firmware upgrade code and somehow try to decipher the new 00:10:27.250 --> 00:10:34.250 firmware. And so I was browsing through the files and sure enough, found the file 00:10:34.250 --> 00:10:40.810 that was helpfully called libupgradeFirmware.so and if we look into 00:10:40.810 --> 00:10:45.010 the symbols, which they luckily didn't remove or anything, there is a beautiful 00:10:45.010 --> 00:10:48.066 symbol called firmware decrypt. 00:10:48.066 --> 00:10:51.150 laughter 00:10:51.150 --> 00:10:56.430 So we load the whole thing into disassembler and we see that 00:10:56.430 --> 00:11:03.870 there's some fancy XORing going on in the bottom left corner. And I'm 00:11:03.870 --> 00:11:08.000 going to walk you through what's, what exactly is happening in this code. 00:11:08.000 --> 00:11:13.310 So basically, first, there's a variable called password loaded into the registar 00:11:13.310 --> 00:11:21.790 R2 and then a second count variable is basically set and it starts looping and 00:11:21.790 --> 00:11:35.540 increasing always by four and goes through this whole xor shebang and it turns out 00:11:35.540 --> 00:11:41.200 that this is the obfuscation method for the AES Key. So, in password, in memory, 00:11:41.200 --> 00:11:45.950 we have an obfuscated key and we can be obfusciated by just implementing the code 00:11:45.950 --> 00:11:53.890 we see here in C or in the emulator. And sure enough, eventually this 00:11:53.890 --> 00:12:03.330 will be used as the key into the ECB 128 AES decryption. And so I implemented the 00:12:03.330 --> 00:12:08.760 whole thing in C, it was almost a copy paste from the decompiler, so you can in 00:12:08.760 --> 00:12:14.500 IAD Pro, you just hit F5, copy the C code at the bit, fix the memory offsets and so 00:12:14.500 --> 00:12:20.020 on. And you have the whole key obfuscation method basically reverse engineered almost 00:12:20.020 --> 00:12:25.630 automatically. And so I compile it. And sure enough, Moxa key extration, it turns 00:12:25.630 --> 00:12:31.200 out that the key is two eight eight seven Conn seven five six four. I build a short 00:12:31.200 --> 00:12:39.260 script to decrypt the 2.1 firmware and this time Binwalk finds all the files and 00:12:39.260 --> 00:12:41.740 we can start reverse engineering the actual firmware. 00:12:41.740 --> 00:12:48.519 applause 00:12:48.519 --> 00:12:54.180 The scripts for this are available on my github. I'll push the actual decrypts stuff 00:12:54.180 --> 00:12:59.810 after the talk because this is the first time this has been released. And so after 00:12:59.810 --> 00:13:03.470 I was at this point, I knew that the firmware is.. I can decrypted the firmware 00:13:03.470 --> 00:13:07.930 I can look into it. By the way, it's not signed or anything. The only verification 00:13:07.930 --> 00:13:14.480 method is CRC32. And so at this point I knew, OK, I can buy this device and 00:13:14.480 --> 00:13:19.980 start playing with it. And so I went to eBay, I bought one. I got it. I screwed it 00:13:19.980 --> 00:13:24.140 open. And sure enough, there's an ARM processor in there. It's an Freescale 00:13:24.140 --> 00:13:28.640 i.MX25, which is just a regular ARM processor. It's like 400 MHz or something, 00:13:28.640 --> 00:13:34.880 I don't know. And I started probing all the all the small pins inside of the 00:13:34.880 --> 00:13:43.040 device to try to find JTAG or serial or anything. And so I actually hooked up 00:13:43.040 --> 00:13:47.320 my power supply to foot pedal so that I can probe and just press with my foot to 00:13:47.320 --> 00:13:54.250 reset the device. And sure enough, I found that there's a full serial console 00:13:54.250 --> 00:14:00.660 available inside of the device on these pins. And if you boot the device, it even 00:14:00.660 --> 00:14:05.160 tells you, please press enter to activate this console, and so you do that and you 00:14:05.160 --> 00:14:07.463 are root on the device. 00:14:07.463 --> 00:14:14.820 applause 00:14:14.820 --> 00:14:18.660 So that's kind of cool, but that means that you require physical access, so 00:14:18.660 --> 00:14:23.530 that's not really a vulnerability, but it's very nice to have when doing security 00:14:23.530 --> 00:14:29.420 research because it means you can suddenly debug all the code on there. And so if you 00:14:29.420 --> 00:14:35.050 write an exploit, you can just touch GDB to the binary and start very, very simply, 00:14:35.050 --> 00:14:40.420 writing the exploit. So at this point, I was trying to look at the available 00:14:40.420 --> 00:14:46.010 services on the device. So for example, there is a web interface, there's a 00:14:46.010 --> 00:14:52.530 proprietary configuration protocol, there's telnet, there's snmp, there is a 00:14:52.530 --> 00:14:58.910 serial driver protocol and so on. And I started looking at the web interface and 00:14:58.910 --> 00:15:03.740 there was cross site scripting that was Cross site request forgery, there was 00:15:03.740 --> 00:15:07.440 insecure authentication where they basically hash on the client. So they have 00:15:07.440 --> 00:15:12.520 some JavaScript that hashes your password and then locks you in. Then there's a 00:15:12.520 --> 00:15:17.720 command injection which lets you execute code as root, there are stack overflows. 00:15:17.720 --> 00:15:23.675 And just a week ago there was a zero day released for the web server. So yeah, demo 00:15:23.675 --> 00:15:36.820 time. So just let me open up the Moxa Pitch right here. And so this one is 00:15:36.820 --> 00:15:41.240 authenticated, so I'll just enter the default password, which, by the way, in 00:15:41.240 --> 00:15:46.050 the field will 90 percent of the time these devices will be configured with 00:15:46.050 --> 00:15:54.950 default credentials. But still, so, if we just start browsing through this thing and 00:15:54.950 --> 00:16:00.140 go to the basic settings, we can start with a simple cross site scripting just in 00:16:00.140 --> 00:16:08.810 the device name. One sec, so just for example we just paste in some JavaScript. 00:16:08.810 --> 00:16:15.017 Submit the whole thing, and hello 34c3. 00:16:15.017 --> 00:16:19.560 applause 00:16:19.560 --> 00:16:23.770 I know what you're thinking, like cross site scripting, come on, that's not a 00:16:23.770 --> 00:16:28.530 vulnerability, that's just nothing. So let's look at the ping test that's 00:16:28.530 --> 00:16:33.910 integrated into this device. And funilly, a different device from Moxa that runs an 00:16:33.910 --> 00:16:39.570 entirely different firmware had the same vulnerability in the past. But if I just 00:16:39.570 --> 00:16:46.390 paste in my ping, so my IP address, a semicolon and then, for example, I cut 00:16:46.390 --> 00:16:51.970 /etc/passwd and activate enter. Here we go. 00:16:51.970 --> 00:17:00.060 applause 00:17:00.060 --> 00:17:08.199 Kind of funny, but, yes, for sure not intended. All righty, but I know what 00:17:08.199 --> 00:17:12.740 you're thinking, right, these are authenticated bugs in the web interface, 00:17:12.740 --> 00:17:17.460 so we need something unauthenticated. We want something that's like cool and a real 00:17:17.460 --> 00:17:23.420 exploit. Right? And so I decided to look at the.. this custom TCP protocol, which 00:17:23.420 --> 00:17:29.430 runs on Port 4900. And my goal was to reverse engineer the whole protocol and 00:17:29.430 --> 00:17:34.030 build a fuzzer for it, to find vulnerabilities, that turned out not to be 00:17:34.030 --> 00:17:40.990 necessary. So during some testing, I just sent a lot of bytes onto this thing and 00:17:40.990 --> 00:17:49.140 enabled crash debugging via the serial console. And sure enough, it crashed and 00:17:49.140 --> 00:17:58.740 put my program countdown right to 0x41414140. Wonderful. Thank you, Moxa. 00:17:58.740 --> 00:18:04.370 applause 00:18:04.370 --> 00:18:21.550 So, Demo time. So let's increase the size of this a bit. So I built a small script. 00:18:21.550 --> 00:18:34.490 Just called moxa_pown and I'll just supply the IP address to it. Let's see. Opening a 00:18:34.490 --> 00:18:43.620 second shell to connect to it via netcat. Here we go, we have a root shell on the 00:18:43.620 --> 00:18:44.600 device. 00:18:44.600 --> 00:18:54.263 applause 00:18:54.263 --> 00:19:01.540 So, yeah, that was the Moxa w21508, basically rolls of the tongue. And so the 00:19:01.540 --> 00:19:08.690 next device I decided to look at was the Advantech EKI-1522 which you can find 00:19:08.690 --> 00:19:17.460 right here. And it's, again, just a simple serial device server this time without 00:19:17.460 --> 00:19:21.410 Wi-Fi, even though they are available with Wi-Fi. It comes with two Ethernet ports 00:19:21.410 --> 00:19:26.060 two serial ports and so on. And I basically followed the same steps again. 00:19:26.060 --> 00:19:31.170 So I looked at the.. I downloaded the firmware. I looked at the edit using 00:19:31.170 --> 00:19:35.800 binwalk. And this time we see almost no entropy. So there is.. this guy is 00:19:35.800 --> 00:19:40.280 basically completely unencrypted. And again, we saw some ARM 32 bit it runs a 00:19:40.280 --> 00:19:51.010 Linux kernel, 2.6.31 and a BOA Web server where the last update was in 2005. And the 00:19:51.010 --> 00:19:56.770 firmware, I think, is from 2017. So these are kind of outdated. And I found 00:19:56.770 --> 00:20:01.230 during the initial analysis just of the firmware that the main binary to look at 00:20:01.230 --> 00:20:07.180 will be this edgserver binary. And so I loaded it into IDA pro and looked at the 00:20:07.180 --> 00:20:12.780 different things that calls. And there are a lot of calls to functions like 00:20:12.780 --> 00:20:18.340 string copy, to system, to sprintf and so on that are generally kind of considered 00:20:18.340 --> 00:20:25.660 unsecure. And sure enough, I am doing static analysis. I found that there's some 00:20:25.660 --> 00:20:33.630 code for sending an email as an alert, for example, when the system reboots. And 00:20:33.630 --> 00:20:39.250 the full command invocation is mailx -s blah blah blah, and we have control over 00:20:39.250 --> 00:20:46.160 some parts in the string because we can configure the two address in the UI. And if 00:20:46.160 --> 00:20:51.040 we look at what's happening here, it basically just sets up this 00:20:51.040 --> 00:20:56.500 format string. Then it goes to include the subject and then it gets some arguments 00:20:56.500 --> 00:21:04.260 from the stack and basically calls into system. And so there's no filtering 00:21:04.260 --> 00:21:09.930 going on at all. So we have an unfiltered part of the system, invocation, code 00:21:09.930 --> 00:21:15.380 execution. And this was before I had the device in my hand. And this is kind of a 00:21:15.380 --> 00:21:19.470 funny story because I first bought because it was just 40 bucks, I bought this 00:21:19.470 --> 00:21:24.770 device, which in the firmware has the same bug, but the mail functionality is broken, 00:21:24.770 --> 00:21:33.780 so I couldn't test it. So I had to go to eBay again, buy another one and buy the 00:21:33.780 --> 00:21:38.950 bigger one. And so I ordered the bigger one on eBay. Looks like this. It comes 00:21:38.950 --> 00:21:45.660 with a Cavium CNS C.P.U. It has JTAG exposed on the bottom there and serial 00:21:45.660 --> 00:21:50.940 console is available again without any authentication. So beautiful. You just 00:21:50.940 --> 00:21:57.809 connect your bus pirate or your UART adapter to it and you have full serial 00:21:57.809 --> 00:22:06.740 console. So, again, we had to look at finding vulnerabilities for this device 00:22:06.740 --> 00:22:11.559 and there, again, a ton of different services, there's like a Web interface 00:22:11.559 --> 00:22:15.670 available. There is a proprietary configuration protocol that's based on 00:22:15.670 --> 00:22:22.760 UDP. There is Telnet, there's snmp, there's a serial driver protocol and so 00:22:22.760 --> 00:22:28.380 on. And again, looked at the website and again, cross site scripting cross side 00:22:28.380 --> 00:22:33.280 request forgery, command injection, broken authentication, which basically if you log 00:22:33.280 --> 00:22:38.710 in from one computer, it uses, I think http digest authentication, you can 00:22:38.710 --> 00:22:42.690 connect from a completely different computer and it doesn't ask for a 00:22:42.690 --> 00:22:49.700 password. I don't know why that is, but.. Yeah. So I was thinking I was doing 00:22:49.700 --> 00:22:52.130 something wrong, but it turned out it was just broken. 00:22:52.130 --> 00:22:54.855 laughter 00:22:54.855 --> 00:23:03.170 So, yeah, and there's, again, a stack overflow in another protocol. So I guess, 00:23:03.170 --> 00:23:13.980 again, demo time. Let's first look at the device itself, so, you know the 00:23:13.980 --> 00:23:22.830 password, firstly, we have a nice device description here. This is just a basic web 00:23:22.830 --> 00:23:29.320 interface. Right. And we can, again, just copy in some basic JavaScript 00:23:29.320 --> 00:23:38.620 hit the save button. Reload and there we go, cross site scripting yet again, OK, 00:23:38.620 --> 00:23:49.129 again, not really interesting. Right. So, um, let's look at the stack overflow. 00:23:49.129 --> 00:24:04.070 Again, I have a small script advantech_pown. For the IP there. And we have netcat 00:24:04.070 --> 00:24:12.090 running on there. Sure enough, there we go, that's root on the Advantech device 00:24:12.090 --> 00:24:13.810 again, via stack overflow. 00:24:13.810 --> 00:24:25.516 applause 00:24:25.516 --> 00:24:31.500 Yeah, so two of three devices have basically broken already. Let's look 00:24:31.500 --> 00:24:38.150 at the next one. This one is a Lantronix EDS2100. And this one is kind of 00:24:38.150 --> 00:24:43.770 interesting because it's not ARM. I normally I almost exclusively do ARMs. So 00:24:43.770 --> 00:24:48.500 this one was kind of interesting. And this device, which is mounted somewhere right 00:24:48.500 --> 00:24:57.390 here. Yeah. This device comes with a serial to ethernet secure device server. 00:24:57.390 --> 00:25:01.929 It has two serial ports. It has Ethernet and you can buy it in two 00:25:01.929 --> 00:25:07.830 variants. One comes with Linux and one is Evolution OS, which is I guess, a 00:25:07.830 --> 00:25:14.880 proprietary operating system from Lantronics. And I'm using the EvolutionOS 00:25:14.880 --> 00:25:22.120 variant in this talk. Looking at the firmware it turns out it's unencrypted and 00:25:22.120 --> 00:25:28.240 it's coldfire architecture, which I've never done really anything with before, 00:25:28.240 --> 00:25:32.630 and there are no obvious external software components. So if you go through this, 00:25:32.630 --> 00:25:37.440 through the firmware, you'll find there's an SSH implementation, there's an SSL 00:25:37.440 --> 00:25:42.810 implementation, but it's not openSSL and it's not anything very well known. And the 00:25:42.810 --> 00:25:47.490 same is true for the web server and so on. It's not really anything that's well 00:25:47.490 --> 00:25:56.500 known. And this time, while probing the device, I did not really find anything 00:25:56.500 --> 00:26:01.580 interesting in terms of serial consoles or so, but it just found a potential debugger 00:26:01.580 --> 00:26:05.730 port, but it didn't have a fitting debugger unfortunately. The CPU is from 00:26:05.730 --> 00:26:14.760 NXP runs at 160MHz or something. Yeah. This time we actually have a web 00:26:14.760 --> 00:26:21.660 interface, we have Telnet SSL and it even has a file system, so you have like FTP 00:26:21.660 --> 00:26:26.210 and TFTP which allows you to download the configuration, upload the configuration 00:26:26.210 --> 00:26:30.980 and so on. And it's kind of hard to secure it correctly because there are so many 00:26:30.980 --> 00:26:37.000 protocols and it's not really clear what's set up by default. But yeah, you get 00:26:37.000 --> 00:26:44.350 the idea. And this time the web interface was surprisingly secure. So there was no 00:26:44.350 --> 00:26:50.230 cross site scripting. There was no command injection, because there's also not really 00:26:50.230 --> 00:26:55.440 a shell that you could execute commands into. But I still found some stuff. 00:26:55.440 --> 00:27:01.540 One is the configuration injection, which allows you basically to change the format 00:27:01.540 --> 00:27:06.630 of the configuration using a different field. And I found an authentication 00:27:06.630 --> 00:27:11.970 bypass, so I was able to write a small piece of code that takes a while and then 00:27:11.970 --> 00:27:23.650 completely removes the password from the device. Demo time. So if we connect to the 00:27:23.650 --> 00:27:29.750 Lantronics device, it will currently ask for a password, which in theory we don't 00:27:29.750 --> 00:27:44.580 have. Let's clean up here a bit. I know it's just. And let's run Lantronix_pown, 00:27:44.580 --> 00:27:51.300 oh, that was fast. That worked. Yeah, sure enough, the password is gone. 00:27:51.300 --> 00:27:59.980 applause 00:27:59.980 --> 00:28:07.230 Awesome. To be honest, I didn't expect the demos to go so smoothly, so I put in an 00:28:07.230 --> 00:28:13.710 hour for the talk for this went very well so far, so that's good. So before we 00:28:13.710 --> 00:28:22.170 finish already, some other devices are even worse. So, for example, as I 00:28:22.170 --> 00:28:26.549 mentioned, I bought some other devices, for example, this Advantaech device and 00:28:26.549 --> 00:28:31.610 this Moxa device and this Lantronix device, which are basically the 00:28:31.610 --> 00:28:38.940 predecessors of the other devices. And those guys are really interesting to look 00:28:38.940 --> 00:28:45.850 at, one could say. So, some of those are running eCos, which is an embedded Linux 00:28:45.850 --> 00:28:52.390 platform, which was last released in 2009, and some devices run a Linux kernel with 00:28:52.390 --> 00:28:57.570 the 2.4 version and you see Linux without any memory protection whatsoever. So even 00:28:57.570 --> 00:29:03.640 if they, so even a small stack overflow in one of the userspace applications gives 00:29:03.640 --> 00:29:08.640 you full root access to the device because you can directly exploit the kernel and 00:29:08.640 --> 00:29:12.840 there are unfixed public vulnerabilities. So in the first penetration test that I 00:29:12.840 --> 00:29:19.290 did, that included actually this device and Moxa and part of a small one. I found 00:29:19.290 --> 00:29:25.170 that using SNMPWwalk, it gives you back the administration password via SNMP. 00:29:25.170 --> 00:29:26.780 laughing 00:29:26.780 --> 00:29:31.500 And so I went online. I tried to report it. And it turns out it's well known 00:29:31.500 --> 00:29:34.160 there's a metasploit module for this 00:29:34.160 --> 00:29:36.830 laughing 00:29:36.830 --> 00:29:41.690 and it's unfixed, OK? And these devices are still in support. So I don't know why 00:29:41.690 --> 00:29:50.950 the vendor is not patching this. Yeah. So the summary with trivial vulnerabilities 00:29:50.950 --> 00:29:56.520 in most devices, or at least all that I've looked at, there are no security 00:29:56.520 --> 00:30:00.571 mitigations whatsoever. So they don't even enable like the compiler flags that you 00:30:00.571 --> 00:30:05.850 just set and then you have at least some kind of stack protection and some like 00:30:05.850 --> 00:30:11.070 stack cookies and whatnot. And some vendors are really bad at responding to 00:30:11.070 --> 00:30:18.429 vulnerability reports. So, yeah, I'm not going to name the vendor, but not even, on 00:30:18.429 --> 00:30:22.180 Twitter I asked them to please give me a security contact and they responded, 00:30:22.180 --> 00:30:26.840 please use our contact form. I said I did, three times. I send you emails, you're not 00:30:26.840 --> 00:30:30.600 responding to me. And so they stopped responding to me on Twitter too. 00:30:30.600 --> 00:30:40.809 laughing applause 00:30:40.809 --> 00:30:47.200 So how to mitigate? Well, the only way that I would see to mitigate against this, 00:30:47.200 --> 00:30:53.380 and I'm more on the deconstructive side of the story, is defense in depth. So never 00:30:53.380 --> 00:30:56.850 directly expose any of these devices to the Internet, even if they say they 00:30:56.850 --> 00:31:02.490 support VPN, even if they say they are a secure device of whatever, just don't do 00:31:02.490 --> 00:31:08.780 it. Get a real VPN gateway and make sure that you never rely on a single level of, 00:31:08.780 --> 00:31:16.169 for example, encryption. So, for example, WPA2 was broken by the crack attack and 00:31:16.169 --> 00:31:20.799 they actually released a patch for it after two months. And these are these are 00:31:20.799 --> 00:31:26.370 still two months where you are exposed to vulnerability on your potentially mission- 00:31:26.370 --> 00:31:33.760 critical system. Also never use GPRS for these devices without VPN because it just, 00:31:33.760 --> 00:31:41.480 it will go wrong. Okay. Yeah, thank you. I guess now we have time for Q&A. Thank you 00:31:41.480 --> 00:31:43.282 all for coming. 00:31:43.282 --> 00:31:49.170 applause 00:31:49.170 --> 00:31:57.990 Herald: Thank you very much for the talk. So we have very much time for Q&A. So 00:31:57.990 --> 00:32:03.980 please line up to the microphones and we have someone at microphone 4 already. 00:32:03.980 --> 00:32:09.220 Mic 4: Yes, hello. Hello. Thanks for your talk. This is.. obviously this is a 00:32:09.220 --> 00:32:14.792 problem. This is a part of the bigger problem of security in IT. Right. In 00:32:14.792 --> 00:32:18.950 anything related to any kind of technology. And this is only going to go 00:32:18.950 --> 00:32:25.230 worse with time, right. Internet of shit, internet of things and so and so on, so 00:32:25.230 --> 00:32:31.740 forth. So my question is, you gave some ideas how to mitigate this in this very 00:32:31.740 --> 00:32:36.540 specific area that use VPN, et cetera, et cetera. But my question is, so hacker 00:32:36.540 --> 00:32:42.462 community is not very, let's say, interested in regulation. Right? And when 00:32:42.462 --> 00:32:46.610 we see, when we see a government trying to do something with technology that usually 00:32:46.610 --> 00:32:51.580 goes bad, we have this idea in our head that, OK, this can only go like this can 00:32:51.580 --> 00:32:56.741 only go bad. Right. But so my question is: do you think that perhaps there is some 00:32:56.741 --> 00:33:00.810 space for regulation here? T: There's definitely space for 00:33:00.810 --> 00:33:07.450 regulation, but I think regulation does not solve the underlying technical issues. 00:33:07.450 --> 00:33:13.611 So these devices, it's 2017 and these devices are using C-code. I think that's 00:33:13.611 --> 00:33:18.580 just asking for trouble, basically. And so we really need to see this shift, even in 00:33:18.580 --> 00:33:22.690 the embedded world, to switch to memory safe languages, for example Rust or 00:33:22.690 --> 00:33:28.129 something similar, and really to stop using C in this kind of context. I don't 00:33:28.129 --> 00:33:35.729 think there's anyone who can .. Thank you. applause 00:33:35.729 --> 00:33:39.072 T: But there's definitely space for regulation. 00:33:39.072 --> 00:33:43.173 Herald: Since there was a question from the Internet. 00:33:43.173 --> 00:33:47.530 Signal Angel: OK, yeah, the Internet wants to know why you are not naming the bad 00:33:47.530 --> 00:33:51.980 vendor, because it looks like it's the only option basically if they don't 00:33:51.980 --> 00:33:57.990 respond to you. Let's say I asked them on Twitter and my Twitter is right there. And 00:33:57.990 --> 00:34:02.640 if you click on Tweets and Replies.. laughter 00:34:02.640 --> 00:34:05.850 Signal Angel: Yeah, somebody just posted the link on IRC. 00:34:05.850 --> 00:34:10.869 laughter T: I did not name them, just for the 00:34:10.869 --> 00:34:13.309 record. laughter 00:34:13.309 --> 00:34:17.029 applause Herald: So we have a question from 00:34:17.029 --> 00:34:23.369 microphone number 2. Mic 2: So you shown an exploit for the 00:34:23.369 --> 00:34:29.559 last device that disabled authentication. What did you use to achieve that? 00:34:29.559 --> 00:34:35.529 T: So this one is unpatched and not yet fixed, so I would rather not disclose the 00:34:35.529 --> 00:34:38.720 details yet. Mic 2: OK. 00:34:38.720 --> 00:34:42.919 Herald: Microphone number 1, please. Mic 1: I wonder if you've also been 00:34:42.919 --> 00:34:47.729 looking at a building automation system, control systems, or just industrial 00:34:47.729 --> 00:34:53.510 automation control systems? T: So you can use these devices basically 00:34:53.510 --> 00:35:00.609 wherever you want. And I think some of the Moxa ones are used in home automation. But 00:35:00.609 --> 00:35:05.920 I've looked at I guess Crestron, it's called? But not in a lot of detail. So I'm 00:35:05.920 --> 00:35:09.509 more on the industrial side at the moment. Mic 1: Thanks. 00:35:09.509 --> 00:35:15.079 Herald: Microphone number 3. Mic 3: Any field experience or even just 00:35:15.079 --> 00:35:21.259 opinions on using industrial strength Raspberry Pi hardware with community 00:35:21.259 --> 00:35:25.559 supported Linux distributions or something like OpenBC whatever on them. 00:35:25.559 --> 00:35:30.869 T: Yeah. So I guess the big trouble there is support, right? There are some, some 00:35:30.869 --> 00:35:34.579 German companies and so on that provide support for industrial Raspberry Pis and 00:35:34.579 --> 00:35:40.789 even like nice casing and so on. But I'm not sure if really Raspberry Pi is the way 00:35:40.789 --> 00:35:45.240 to go here. I think there are boards that are.. the problem is not the 00:35:45.240 --> 00:35:49.720 underlying stack, right? It's not the hardware. Really, that's the issue. It's 00:35:49.720 --> 00:35:55.950 the software. And you will have the same issues on on the Raspberry Pi. So, yeah, I 00:35:55.950 --> 00:36:00.880 guess you could buy these devices, which are like industrial grade shockproof and 00:36:00.880 --> 00:36:07.460 whatnot, and put some Linux on it and do it better. But I don't think that 00:36:07.460 --> 00:36:11.650 the hardware or platform will change anything at the moment. 00:36:11.650 --> 00:36:16.319 Herald: There is another question from microphone number 4. 00:36:16.319 --> 00:36:21.749 Mic 4: Hi, more a social question, did you get in contact with any development team, 00:36:21.749 --> 00:36:25.849 software development team in any of these companies, or might it be that there is no 00:36:25.849 --> 00:36:33.080 one behind the emails and everything? T: So I guess some of these companies are 00:36:33.080 --> 00:36:37.349 really so big, that they don't reply to you if you don't have a support contract 00:36:37.349 --> 00:36:45.049 with them. But, for example, the support of the ones that are not on my Twitter is 00:36:45.049 --> 00:36:49.730 kind of decent when it comes to two security reports. And so my next steps 00:36:49.730 --> 00:36:57.220 will be to go via the ICS Cert, but, you know, to report them. So, yes, there are 00:36:57.220 --> 00:37:03.737 development teams that will get in contact with you, just not from all vendors. 00:37:03.737 --> 00:37:06.670 Herald: Thank you. We have another question from the Internet. 00:37:06.670 --> 00:37:13.960 Signal Angel: Hello? OK. The Internet wants to know what to do about, because 00:37:13.960 --> 00:37:18.259 there are a lot of old devices in the field, how do you propose a vendor should 00:37:18.259 --> 00:37:24.200 deal with legacy devices and updates? T: Yeah, so keeping legacy devices 00:37:24.200 --> 00:37:29.680 supported is very expensive because, for example, if you buy a Qualcomm chip, they 00:37:29.680 --> 00:37:35.089 will eventually drop support for the Linux kernel for it and so on. But if you buy 00:37:35.089 --> 00:37:39.619 like a Freescale automotive chip, they guarantee you a certain time of support. 00:37:39.619 --> 00:37:43.490 But then you actually have to invest the money to regularly provide the updates and 00:37:43.490 --> 00:37:48.859 ensure that your devices are secure. The problem is that the lifetime of industrial 00:37:48.859 --> 00:37:55.470 installations currently is much larger than the lifetime of this processors' supports 00:37:55.470 --> 00:38:00.819 and so on. So I guess we'll have to get used to upgrading our hardware regularly 00:38:00.819 --> 00:38:07.400 or switch to, or figure out a different way of deploying secure software onto 00:38:07.400 --> 00:38:11.259 them. But I really think the underlying problem is, that we are still using 00:38:11.259 --> 00:38:16.229 memory unsafe languages. And I guess the fact that there's cross site scripting 00:38:16.229 --> 00:38:20.150 just shows that there's no security awareness really at those vendors 00:38:20.150 --> 00:38:29.395 whatsoever. At some of the vendors. Herald: So, microphone number 2, please. 00:38:29.395 --> 00:38:34.349 Mic 2: I was wondering, you mentioned that some of these facilities use GPRS. 00:38:34.349 --> 00:38:36.390 T: Yeah. Mic 2: Do you know if they have mostly 00:38:36.390 --> 00:38:40.749 their own closed infrastructure, or if they're using general consumer telecom 00:38:40.749 --> 00:38:44.849 stuff? T: So they will use commercial 00:38:44.849 --> 00:38:50.479 networks mostly, and then they have custom EPNs which have an IPSec tunnel or 00:38:50.479 --> 00:38:55.700 something similar to their premises. But there's also there's also a company that 00:38:55.700 --> 00:39:02.589 sells industrial control SIM cards which give you a public IP and you don't 00:39:02.589 --> 00:39:08.100 want to search on Shodan for that vendor. Mic 2: Yeah. Thank you. 00:39:08.100 --> 00:39:11.050 Herald: There is a question from microphone number 3. 00:39:11.050 --> 00:39:14.999 Mic 3: Hi there, isn't economics meant to solve some of these problems? We're not 00:39:14.999 --> 00:39:20.359 talking about dirt cheap devices. How surely at 300 bucks you should better have 00:39:20.359 --> 00:39:24.539 someone who's read security one and one. How long before a large organization gets 00:39:24.539 --> 00:39:28.200 the result of their security audit and goes to the aforementioned vendors and 00:39:28.200 --> 00:39:32.960 says, provide us something that's not trivially hackable, otherwise we stop 00:39:32.960 --> 00:39:37.839 buying your rubbish? T: Well, I mean, it's the same in all of 00:39:37.839 --> 00:39:45.329 IT, right? So everything has vulnerabilities. And yes, there should be 00:39:45.329 --> 00:39:50.400 market pressure. But that's why I'm trying to raise awareness for the issues that 00:39:50.400 --> 00:39:53.270 these devices have. Mic 3: Thanks. 00:39:53.270 --> 00:39:55.729 Herald: There's another question from the Internet. 00:39:55.729 --> 00:40:01.339 Signal Angel: Yep. The Internet wants to know how and if it's a good idea to raise 00:40:01.339 --> 00:40:06.549 the level of awareness in public, because they think it's a good approach to make 00:40:06.549 --> 00:40:11.869 people, the public know that, well, infrastructure in the cities is at risk. 00:40:11.869 --> 00:40:16.000 T: Uh, sorry. Could you repeat the first part of the question? 00:40:16.000 --> 00:40:21.339 Signal Angel: Yeah. They want to know how to raise awareness for this in the public? 00:40:21.339 --> 00:40:27.789 T: Good question. I guess we need some news articles or something about this in 00:40:27.789 --> 00:40:32.800 regular paper, but I personally think it's just an accident waiting to happen. So 00:40:32.800 --> 00:40:37.999 eventually someone will turn off the lights in a city or wherever, will open a 00:40:37.999 --> 00:40:44.773 flood valve or something. And that's when the awareness will start. 00:40:44.773 --> 00:40:47.813 Herald: There's another question from microphone number 4. 00:40:47.813 --> 00:40:51.680 Mic 4: OK, for what kind of industrial processes are these devices you just 00:40:51.680 --> 00:40:57.108 demoed used? T: So I've seen them in power utility. I 00:40:57.108 --> 00:41:02.350 know they're used in water dam control systems. They are used and in 00:41:02.350 --> 00:41:07.039 serial connecting a CNC machine to the network, they are used in connecting all 00:41:07.039 --> 00:41:10.690 kinds of stuff. Because if you have a big plant, you have a ton of different 00:41:10.690 --> 00:41:15.719 sensors. So you might, you might need the water level sensor. And for whatever 00:41:15.719 --> 00:41:20.680 reason, you only can get it with a modbus and then you need to convert the modbus to 00:41:20.680 --> 00:41:25.119 TCP and then you need one of these gateways. And so, I've seen in one 00:41:25.119 --> 00:41:28.529 cabinet, 20 of them. So they're really used a lot I guess. 00:41:28.529 --> 00:41:31.869 Mic 4: OK, thank you. I just retweeted your tweet to Star Alliance. 00:41:31.869 --> 00:41:37.979 T: Huh. laughs Thank you. laughs Herald: So there's another question from 00:41:37.979 --> 00:41:41.260 the Internet. Signal Angel: Yeah, the Internet wants to 00:41:41.260 --> 00:41:50.749 know if you did any research on MQTT for example from like Beckhoff uses? 00:41:50.749 --> 00:41:54.489 T: I actually talked to someone who recommended me to look at Beckhoff 00:41:54.489 --> 00:41:58.249 yesterday, but I've not looked at them whatsoever yet. 00:41:58.249 --> 00:42:01.900 Herald: And there's another question from microphone 3. 00:42:01.900 --> 00:42:07.450 Mic 3: OK, could you show the Moxa web panel, because I would like to double 00:42:07.450 --> 00:42:16.619 check, which proves that they and they would like you to see their Web page. And 00:42:16.619 --> 00:42:24.050 I think this browser isn't very secure. T: OK, let's take a look. 00:42:24.050 --> 00:42:29.160 Mic 3: Yeah, and under gohead the webserver small print. 00:42:29.160 --> 00:42:41.527 laughter Herald: Nice finding. 00:42:41.527 --> 00:42:47.859 T: That's probably the issue here. laughs 00:42:47.859 --> 00:42:55.658 Herald: Are there any more questions? Any questions from the Internet? 00:42:55.658 --> 00:43:02.009 Signal Angel: The internet wants to know how a memory safe language would prevent 00:43:02.009 --> 00:43:08.750 the authentication bypasses you showed? T: Not one would not be protected against 00:43:08.750 --> 00:43:13.130 but it protects against a ton of other stuff. It's just one example of where the 00:43:13.130 --> 00:43:18.420 industry needs to change. We need to stop using memory unsafe languages. We need to 00:43:18.420 --> 00:43:23.910 start really thinking about security design from the start, and we must not in 00:43:23.910 --> 00:43:28.319 2017, there's no excuse for having cross site scripting or anything on the web 00:43:28.319 --> 00:43:35.720 page. That's also if we in the Lantronics website, if you click logout, 00:43:35.720 --> 00:43:39.479 it tells you logout is not supported in your browser. 00:43:39.479 --> 00:43:43.290 laughter T: Probably because I'm not using Internet 00:43:43.290 --> 00:43:48.130 Explorer five. Herald: So there's another question from 00:43:48.130 --> 00:43:53.239 microphone number 3. Mic 3: Any remote part of the exploit 00:43:53.239 --> 00:43:57.750 where you did a buffer overflow - I think. 00:43:57.750 --> 00:44:01.490 T: Yeah? Mic 3: What I'm wondering is, are 00:44:01.490 --> 00:44:07.180 there.. isn't it like very standard to have ALSR on these devices? 00:44:07.180 --> 00:44:10.239 T: No! laughts It should be, but it isn't. 00:44:10.239 --> 00:44:16.199 Mic 3: Okay. Thank you though. That was pretty much my question. 00:44:16.199 --> 00:44:23.428 Herald: Is there another question from the Internet? It doesn't seem like it? 00:44:23.428 --> 00:44:36.052 Signal Angel: So, one just came in, OK, if you want to hear it. Ok, nope. 00:44:36.052 --> 00:44:41.329 laughter Herald: So, all right, give a very warm 00:44:41.329 --> 00:44:43.329 applause to Thomas Roth again! 00:44:43.329 --> 00:44:46.779 applause 00:44:46.779 --> 00:44:59.882 postroll music 00:44:59.882 --> 00:45:08.000 Subtitles created by c3subtitles.de in the year 2021. Join, and help us!