Skip to phishing-filter section
Recently I switched Firefox’s Android app from Preview Nightly to Nightly after the former has been deprecated. The switch entailed migrating the configurations; a config I need to migrate over is DNS-over-HTTPS (DoH). I verified the Quad9’s DoH address through its instruction page (tips: you can use “https://126.96.36.199/dns-query" instead of “https://dns.quad9.net/dns-query" so that a browser doesn’t need to query the IP behind dns.quad9.net). I also checked out its recent article that mentioned it blocks more malicious and phishing websites (via DNS-blocking) compared to other well-known DNS service–like Cloudflare and OpenDNS–according to an independent test. The test was measured based on DShield.org Suspicious Domain List.
(Edit) DShield.org dataset previously did not sourced from PhishTank and URLhaus datasets and were not actually used in that DNS test. DShield later switched after it found out the previous dataset was outdated. In light of this discovery, I conduct a DNS filtering test using datasets sourced from URLhaus and PhishTank, refer to this post for result.
I was intrigued by the DShield list as I created a blocklist (urlhaus-filter) that is also based on URLhaus. I then checked out its another source, the PhishTank list. PhishTank operates similarly to URLhaus, the links are user-submitted. User can vote on submitted links (of other users’) are indeed phishing websites. The database is available in various formats including CSV. This seemed ideal to be processed into a blocklist, just like what I did in urlhaus-filter. To avoid duplicate effort, I did a search on FilterLists and there is a domain-based blocklist (“Phishing Bad Sites“) that is based on PhishTank.
Domain-based blocklist is created by stripping out the path of the original links, leaving the domain only (e.g.
). This blocks the whole website, instead of specific webpages; it also significantly reduces the file size, not just from the path stripping, but also de-duplication of domains. However, one thing I learned from urlhaus-filter is that many malicious links are also hosted on popular domains, like Google Docs and Dropbox; such is the fate of file-hosting service, it will inadvertently be abused to host malicious content. To avoid blocking those popular services, I utilise Umbrella Popularity List and Tranco List to remove popular domains from urlhaus-filter and minimise false positive. Since uBlock Origin (uBO) supports blocking webpages via static filter (e.g.
||example.com/foo-page$all), malicious webpages (of popular domains) are still blocked in the uBO-specific filter.
I ran a quick check on “Phishing Bad Sites” filter:
$ grep -F 'google' 2020-07-07-phishing.bad.sites.conf
The search result included Google Drive and Google Play. Hence, I find it necessary to create a new PhishTank-based filter, with minimal false positives and available in various formats (other than uBO). Another reason is that despite PhishTank being operated by OpenDNS, it does not block all of the verified phishing websites.
I presents phishing-filter, a blocklist to restrict >14K phishing websites. uBlock Origin (uBO) users can import phishing-filter.txt to install the filter. Other formats includes domain-based, hosts-based, dnsmasq, bind and unbound, refer to the repository for installation guide. The blocklist utilises similar approach as urlhaus-filter to exclude popular domains. Phishing links found in popular domains are still included in the “phishing-filter.txt”, hence I recommend to use uBO for best result.
The workflow is largely similar to what I did in urlhaus-filter, so I don’t have to reinvent the wheel here. I did take the opportunity to improve the repository’s folder structure, which I find a bit messy in urlhaus-filter. urlhaus-filter’s folder structure is still retained as is because changing it would induce breaking change. In phishing-filter, all generated filters are put in
src/ folder and
utils folder contains csvquote binaries.
csvquote is a workaround for the use of optional quote in PhishTank database. A URL is quoted only when there is a comma.
cut having incorrect result, the comma in the link is trimmed off,
$ cat phishtank.csv | cut -f 2 -d "," http://example-phishing.com/lorem "https://example.phishing.net/ipsum
csvquote works by escaping the comma(s) in the column and then un-escape back.
$ cat phishtank.csv | csvquote | cut -f 2 -d "," | csvquote -u http://example-phishing.com/lorem "https://example.phishing.net/ipsum,ipsum"
I then remove the quotes with
sed 's/"//g'. It will be more convenient if all URLs are quoted–just like URLhaus.csv–I can simply use
cut -f 2 -d '"'. I know I’m not supposed to use
cut to process csv, but there is no csv-processing command line tools available in Alpine Linux official packages.
Before I stumbled upon csvquote, I also considered xsv and csvtools. xsv is written in Rust and can be installed via Cargo package manager, which is available in Alpine Linux. I couldn’t compile csvtools under Alpine because it’s not compatible with musl. Ultimately, I chose csvquote due to simplicity, minimal size and being compatible with both glibc and musl. Speaking of glibc and musl, the binaries are incompatible with each other; I couldn’t run glibc-compiled csvquote on Alpine (musl-based), neither did musl-compiled binary on Ubuntu. I utilise a musl detection script that can adjust the binary choice automatically.
"$(ldd /bin/ls | grep 'musl' || [ $? = 1 ])" if [ -z "$LIBC" ]; then rm -f "/tmp/musl.log" # Not Musl CSVQUOTE="../utils/csvquote-bin-glibc" else # Musl CSVQUOTE="../utils/csvquote-bin-musl" fi cat "phishtank.csv" | \ "./$CSVQUOTE" | \ cut -f 2 -d "," | \ "./$CSVQUOTE" -u | \ ...LIBC=
OpenDNS and PhishTank are either trademarks or registered trademarks of OpenDNS, LLC.