We covered the basic concepts of blockchain, cryptocurrencies, and coin mining in our previous blog.
The beginning of in-browser mining
On the other hand, for malicious actors, the web-based approach also provides extra flexibility: there is no need to store everything on the same page. Functionality and components can be split across multiple domains, and previously rented or hacked servers can be (re)used for a new purpose.
Coinhive et al.
Fingers soon started to point at the creators, primarily as a result of sites not always being willing to provide an opt-out from running the scripts on their visitors’ PCs. Many didn’t even bother to inform their visitors about their computers being used in this way.
In response, Coinhive created a new version which would become enabled if permission was explicitly given. Unfortunately this offered no guarantee of Coinhive users adopting the new build, especially ones with clear malicious intent in mind.
Naturally, the strong uptake of Coinhive resulted in the development of competing services in late 2017 and early 2018 along with several easy to use mining plugins for popular web platforms such as WordPress.
The first ‘Minimum Viable Product’ (MVP) release of WebAssembly was in March 2017 and the standard is currently supported by all major browsers including Firefox, Chrome, WebKit /Safari and Microsoft Edge.
While it has numerous use-cases, many of which are listed within the WebAssembly design documents, it’s easy to see how the technology would appeal to coin miner developers: in-browser miners have to live with a significant performance drop compared to ‘native’ code, and WebAssembly’s focus on speed assists in closing this gap. For reference, the developers of Coinhive provide the following performance guidance:
Is it malware?
Whether or not in-browser mining is inherently malicious is a matter of contention, with even the security industry harbouring mixed views.
Running a hashing algorithm on someone else's PC without their knowledge is a malicious action: ultimately, it is appropriating someone else’s resources without their permission. While it shouldn’t do any harm to the data or integrity of the machine, it will result in a slower-responding PC and an increased electricity bill, especially if goes unnoticed for a long period of time.
On the other hand, performing hashing with a user’s consent is not a malicious action – at least some users may be willing to run these scripts if it means no adverts on a website. In early 2018 the Salon website experimented with this approach.
From here onwards, please note that when we use the term Coinhive we are exclusively referring to abuse of the Coinhive code and service for the purposes of mining without user consent.
When good dogs do bad things
Malicious users quickly came up with multiple schemes to get miners into people’s browsers. Chrome extensions featuring Coinhive code were quickly developed, providing a much-improved return over simple web pages as the miner would run whenever the browser was open. In parallel, a number of Coinhive blocker extensions were developed to try and prevent unauthorised mining on websites.
Google's Web Store policy was initially permissive with regards to mining extensions: as long as the extension's sole purpose was mining and the user was adequately informed, it would not be banned. Still, many failed to comply with these policies and, as a result, Google recently moved to tighten the rules effectively banning any mining extensions from the Chrome Web Store.
This left the bad guys with the tried-and-true approach of website code injection.
How does all this work in the wild?
As the popularity of browser-based miners is high amongst cybercriminals, a vulnerable web site can potentially be compromised by more than one actor: as an example, the web site shown below was compromised by three separate cybercriminals - all using slightly different browser miner tools. Note the lack of visible evidence of the miners on the site itself.
The image below shows the full infection chain visible in the packet capture when visiting this site:
By the time we analysed the site the first of the three miners had been removed, however the other two were still working. The injected code for this inactive miner is shown below.
Active Infection #1: ‘Basic’ Miner
The first active miner is a widely available browser miner tool (‘Basic Miner’ in the diagrams).
It is less sophisticated and easier to block with a web security gateway than the second active miner as it uses minimal obfuscation and static domains for relaying mining traffic to mining pools through WebSocket-based intermediary proxy servers. The image below shows the injection for the ‘basic miner’.
Upon loading, it first downloads the CryptoNight WASM binary and then immediately begins mining, routing its traffic through publicly known coinhive.com WebSocket proxies (see below).
Active Infection #2: ‘Advanced’ Miner
Many publicly available ‘grey/black’ miners also have private versions for subscribers which are generally intended to be much harder to detect. The second active sample (‘Advanced Miner’ in the diagrams) is potentially one of these. Its injection code is shown below.
It uses a URL shortening service to retrieve the main mining script and uses a slightly more advanced obfuscation technique than the ‘basic miner’ – although it should be noted that this is still only simple base64 obfuscation, something which would be considered very basic amongst more advanced threats such as exploit kits.
Interestingly, this miner uses at least one dynamic-DNS proxy server (see code snippet below) for relaying mining data through WebSocket connections making detection and blocking much more difficult.
Comparing the Two Active Miners
It is worth noting that the underlying code in the ‘basic’ and ‘advanced’ miners is very similar – this broadly appears to be the case with a large number of in-the-wild WebAssembly-based miner, many of which appear to be minor variants of codebases available on GitHub. These ‘copycats’ are mostly tweaked only slightly in order to avoid detection, usually by altering function names in a ‘creative way’ (batmanrobin, darkshadow, krypt0n1ght) or by completely removing references to the original C source.
The table below compares the functionality of the two miners.
|Easily detectable, contains in-browser mining related keywords.
|More innocuous looking, the main miner script is hidden behind a URL shortener.
|Miner Script Obfuscation
|Almost non-existent, only the most critical keywords are obfuscated using hex values.
|Base64 encoding for the whole script, critical keywords are hex encoded, plus some dynamic obfuscation.
|WebSocket Proxy Infrastructure
|Publicly known and static, uses coinhive.com WebSocket servers.
|Private and dynamically updated, making takedown much more difficult.
A peculiarity of in-browser miners is their reliance on WebSockets for communication. As processes running in a browser sandbox are not permitted to open system sockets, WebSockets were designed to allow full-duplex, asynchronous communications between code running on a webpage and servers – for example, chat applications such as Slack make heavy use of WebSockets.
As the standard for WebSockets calls for sessions to be opened as HTTP and then ‘Upgraded’ to a WebSocket (see the example below), this does not allow for direct communication with the majority of coin mining ‘pools’ which generally use a protocol called Stratum.
As a result, the operators of in-browser mining operations need to set up WebSocket servers to listen for connections from their miners and either process this data themselves if they also operate their own mining pool or ‘unwrap’ the traffic and forward it to a public pool if they don’t.
While a coin mining script may appear on dozens or even hundreds of websites, there are likely to be significantly fewer of these WebSocket servers involved in any deployment, thus dramatically simplifying blocking using web proxies: no ability to communicate with the mining pool generally means no mining activity.
Note: The following actions are undertaken at your own risk.
Something of a nuclear option, only Firefox and Chrome support disabling Wasm as of April 2018. To do this, follow the instructions below. Before doing this, consider that this approach is likely to prevent the normal operation of an increasing number of sites as the technology is further adopted.
In Chrome: navigate to chrome://flags/#enable-webassembly and change it to disabled
Conclusions & protection statement
With the ever-increasing popularity of cryptocurrencies (despite the unstable value of Bitcoin in 2018) the ongoing interest in coin mining is no surprise. The question for the security industry is the one posed above: is it malware?
Reduced to its simplest, the mining process is nothing more than arithmetic instructions executed in order to solve a pre-defined piece of a puzzle. This can hardly be considered malicious on its own.
The key question is whether this activity occurs with the explicit understanding and approval of the user. As such, blanket blocking all mining scripts without examining their context is not necessarily the best approach.
Forcepoint take a combinatory approach to detection and blocking of coin miners associated with compromised websites, blocking the instances of the scripts which we identify but – more critically – blocking the WebSocket command/relay servers which entire campaigns depend on.
Thus, Forcepoint customers are protected against this threat at the following stages of attack:
Stage 5 (Dropper File) - Malicious files are prevented from being downloaded.
Stage 6 (Call Home) - Attempts to contact the C2 server are blocked.
Indicators of Compromise
Note: The following list of IOCs is non-exhaustive and focuses primarily on the Wasm samples and campaigns discussed in this article.
WebSocket Relay Servers
cfcnet[.]top cfcnet[.]gdn cfcdist[.]gdn cfcdist[.]loan cpufan[.]club coinhive[.]com
Hashes (WebAssembly Files)
978c69b6b93a66e47236abc0f36c79096429111d 5117b6d9fd649e5946be0d3cbe4f285d14f64ca2 99e5c6e9bd5ffebf7afe070fb53ee08f10cc707f 96d82c07ef7e15465073bff9e07c30e54f867f50 8a31ff73da067ac87e098effcce909390b568965 16340175dad11597028d434a34365ee46fb815d9 f43a873add72fa6e6b5465d5cdd7ec275dadfaa5 f3353f1a3da825bcc4a552805cf345531a1e48dd eba038b18f32227a9dd1e761fb69029dda6bdbdd 730efe35879cd09c2dc1e18c21f79896fe611120 72616671f4fcebf2b357de851c3598b33108c9a1 957b140c2b28508c78e9b526769f2e9ab6e664da 54612d3dc9ac23c1b49ac3243db12eb5c003118e a3da43bb12093017143c2888c8fddd7e12479bd6 b6c6e98ddca1030b4d9261972506088115fcd04f 58776b6dccf4c0d7a329e64fdeb4f6d13dc3cd2e cbf527baabc7e344cb8f9ec37f0ee490ce800dc1 9392401eb40c0bdac3e22d5430a79af16f3424d5 bcf77c7424c4707b9ea57c82abf1a0859a456784 ba38dc9fb45946228081bd6932e533461e4b3a98 ef58f8c31ab2bc380afc13f691f7b9415aa2dac2 07506bbb3873180d97834622cd55e6cd890a92d2 19c0abdfcf37fdf0bd102b90f6c05b081ea772d0 7bc4adbb43f26f0d954946ae3c4a5e3b91882e9c 04e7b1e1251fb0c18753d0b413f69d5c34c17acf f6a08de1f9c9f79d39455ab31d91c1b44f859116 2a35af592d672d45848548fa17c67b7cc952f05d bc387716f4de26ae78a8c10c9d41b3f40cf517ab d0a66bc0e408e04776e0bec1fc18864974df0198 da6269118b9b8aec68ba7c22de689a7f8e3e1d42 ad765a663c40457cf96c8023500d24f02711f30d b3338cae76b9eacf477af92b4bcfd8cba8f2a516 59d33942585d970b75d2c2e9993d18335edb07bc 8ca54a1a926e5cb0b3a33baa3c58ff3acdbfc515 432fac1f706388be3f7257dd71388021e53038ad 971960066074c62f986215cdf0d85f36b0b33f0f 032c4d6c50c11b3dc247f1941bf77700720ac7e0 29122c16de434eccb4c3abe0c899ae47f9829dde 8c964de7bb13fa9683eed76c419256462393c55c 957cc99377fadb192483a9810660a2fc1f5451fb 608a65e5d08e4cb8362ffbb773a99e78b89d1b5c a0afabb912d593dd2ca9332a1367c1a7614dc3e3 f234378e7b3a93e1dfe06f3efebcee204becaa49 6b7423ab78e38cb8a4263d12c1c2158536fd02cd 57d9f3e422fca7dddbaa381230a6e20921d4a2ac 11c91a43f77f0a7b1da10f52160d52478464c86a b1ce36407b3a456e8bd0fb33b777d84f884db072 c8e27ae2938ea4be173e796f83c843009d21d883 198370faa7488ad8bb95d170880bd52045576eff 4387f76d9c49232f62a41c55c6a48eb38dc97eee 786df01387091a23b15b11385e61e3d441b50bca b291f9545faea0cc4ff8ab44e7ea766f85d4993c 5abf24ed6a7b85b0fae6560fe67bcdc9f8406a11 edd53e16aee9344d580bc3e57b1b2e30d11f5d14 c147b98d461210f8aa835bb9636db2dbf358cd0a