cover image for post 'The DGA of QSnatch'

The DGA of QSnatch

Table of Contents
Disclaimer

These are just unpolished notes. The content likely lacks clarity and structure; and the results might not be adequately verified and/or incomplete.

Changes
  • 2019-11-13 21:11:03: Added two more QSnatch samples with the simpler DGA.
  • 2019-11-14 12:08:08: Version timestamps from version A samples added.
DGArchive

The DGA in this blog post has been implemented by the DGArchive project.

Malpedia

For more information about the malware in this blog post see the Malpedia entry on QSnatch.

QSnatch is a malware that infects QNAP NAS devices. It collects and exfiltrates user credentials from vulnerable devices, and can also load malicious code from its command and control (C2) servers. These C2 servers are resolved by algorithmically generated domains.

The National Cyber Security Centre of Finland (NCSC-FI) published an article about QSnatch in late October 2019 and made the threat known, but samples on Virustotal date back to at least June 2019.

I found seven different QSnatch samples with two different DGAs. Both versions are very similar, with one being a simpler version of the other. I called the more complicated version Version A and the simpler one Version B. QSnatch is implemented as shell scripts so it is trivial to reimplement the DGA in Python.

Version A

These are two samples with the more complicted version of the DGA from Virustotal. This version of QSnatch comes with a timestamp as version information:

MD5
372140d7c2c68dc2c8dc137d1a471e9f
SHA1
986f38a04937ede2000e8f25e59ea438ee265e24
SHA256
3c38e7bb004b000bd90ad94446437096f46140292a138bfc9f7e44dc136bac8d
Version Timestamp
2019-03-20 5:00 UTC
Size
41KB, 41655 Bytes
Uploaded to Virustotal
2019-11-04 13:39:51 UTC

and this sample (which is also the one currently delivered as of 2019-11-14 11:00 UTC):

MD5
60567a1d2b2e02e93ffc162e6a70d60c
SHA1
1f1bf0bd2df89029d5267130f014ab5aa133c3ae
SHA256
9526ccdeb9bf7cfd9b34d290bdb49ab6a6acefc17bff0e85d9ebb46cca8b9dc2
Version Timestamp
2019-05-17 5:00 UTC
Size
41KB, 41104 Bytes
Uploaded to Virustotal
2019-06-09 22:56:07 UTC

Both samples have a timestamp set to exactly 05:00 UTC, which could mean that the timestamp is in fact a simple date without time information, generated in a timezone UTC+5.

The DGA generates domains like the following:

t2q2r.cf
gc9nz.tk
07tvvc.com
7ubqo.ml
53bcm.de
6zltf.rocks
hv7uv.mx
nypno.biz
qkzccy.net
rassb.cn

Bash

The following screenshots shows the shell script lines responsible for generating the domains:

The DGA

In the following I will briefly discuss the main components of the DGA.

TLDs

The complicated version of the DGA of QSnatch uses a whopping 145 top level domain, including many gTLDS like rocks, mobi, today; and a few sld/tld-tuples where domain registrations are only possible on the third level, like .com.ua, .com.bn, .com.by. The domains are listed in a space separated string as tld/number-pairs divided by colons:

domainexts='cf:0 tk:0 com:1 ml:0 de:0 rocks:0 mx:0 biz:0 ...``

The number next to the domain specifies the number of extra characters in the hostname. In the analysed sample, .com, .net and .org have 1 extra character specified, meaning their hostname has one extra character compared the remaining tlds, which all have 0 extra characters set.

The script has a bug for the domain .com.bn which is listed with a leading dot, resulting in invalid domains with doubled points, e.g., r4rb..com.bn.

Domain Timespans

The generated domains have different validities of 15 days (1296000 seconds) down to 1 hour (3600 seconds). The DGA generates domains for all specified intervals starting with the largest timespan.

for interval in '1296000' '432000' '86400' '28800' '7200' '3600'; do

Hostname Lengths

The hostnames have varying lengths, which are calculated by adding a global length value and the number next to the list of top level domains. The global lengths are 5, 3, then 4. As a result, .cf hostnames have a length of 3 to 5, while .com hostnames have a length of 4 to 6.

for length in 5 3 4; do

Iterating over all TLDs

The third loop iterates over all top level domains.

n=0; while [ "$n" -lt $domainextcnt ]; do

Time permitting, the DGA generates 6 (number of timespans) * 3 (number of hostname lengths) * 145 (number of tlds) = 2610 domains.

Aborting Domain Generation

Before looping over the top level domains, the DGA checks if more than 10 minutes (600 seconds) passed since starting a timespan block. If ten minutes elapsed, then generating domains for that particular timespan is aborted, unless it is the last timespan:

test "$(( $timenow - $timestart ))" -gt 600 && test "$interval" != "3600" && break

Hostname String Generation

The following command generates the string which is later trimmed into a hostname:

hostname=$(echo \
  "$(( $(date +%s) / $interval ))IbjGOEgnuD${ext}" | \
  openssl dgst -sha1 -binary | \
  openssl base64 | \
  sed 'y/ABCDEFGHIJKLMNOPQRSTUVWXYZ-+\//abcdefghijklmnopqrstuvwxyzabc/;s/=//g')
  1. The command ($(date +%s) / $interval ) divides the current unix timestamp by the timespan length.
  2. The seed IbjGOEgnuD and the top level domain ${ext} are appended to the string from Step 1.
  3. openssl dgst -sha1 -binary generates the SHA1-hash of the string (including a newline character \n from the echo command).
  4. openssl base64 converts the hash into base64.
  5. sed 'y/.../;s/=//g') converts the base64 string into lowercase, replaces -, + and \\ with a, b and c respectively, and removes =-padding.

Trimming to desired length

Next, the hostname is trimmed to the desired length. The length can not be smaller than 3, which is never the case for the configured lengths:

trycnt=0
while [ ${#host} -gt "$l" ] && [ $trycnt -lt 3 ]; do
  trycnt=$(( $trycnt + 1 ))
  host=${host%?}
done

Concatenating the Hostname and SLD

Finally, the hostname and tld are joined to produce the DGA domain:

curl --connect-timeout "$curlconntimeout" -m 30 -k -o "$outfile" "https://${host}.${ext}/qnap_firmware.xml?t=$(date +%s)"

Python

This is a reimplementation of Version A of the DGA in Python 3:

import time
import hashlib
import base64
import argparse
from datetime import datetime


TLDS = {
    "cf": 0, "tk": 0, "com": 1, "ml": 0, "de": 0, "rocks": 0, "mx": 0,
    "biz": 0, "net": 1, "cn": 0, "ga": 0, "gq": 0, "org": 1, "top": 0, "nl": 0,
    "men": 0, "ws": 0, "se": 0, "info": 0, "xyz": 0, "today": 0, "ru": 0,
    "ec": 0, "co": 0, "ee": 0, "rs": 0, "com.sv": 0, "com.cy": 0, "co.zw": 0,
    "kg": 0, "com.ge": 0, "tl": 0, "name": 0, "tw": 0, "lv": 0, "bs": 0,
    "li": 0, "ng": 0, "ae": 0, "bt": 0, "tv": 0, "pe": 0, "uz": 0, "me": 0,
    "gy": 0, "am": 0, "kr": 0, "by": 0, "fr": 0, "com.uy": 0, "com.lb": 0,
    "com.br": 0, "vu": 0, "hk": 0, "in": 0, "re": 0, "ch": 0, "af": 0,
    "com.ps": 0, "ug": 0, "dz": 0, "pro": 0, "co.th": 0, "sg": 0, "cd": 0,
    "so": 0, "mo": 0, "co.id": 0, "co.il": 0, "com.do": 0, "ke": 0, "cx": 0,
    "ro": 0, "id": 0, "pm": 0, "hm": 0, "vg": 0, "az": 0, "com.eg": 0, "bz": 0,
    "su": 0, "com.ar": 0, "gg": 0, "com.lr": 0, "pa": 0, "com.ve": 0, "al": 0,
    "fm": 0, "to": 0, "mu": 0, "co.ck": 0, "pk": 0, "co.rs": 0, "cw": 0,
    "nr": 0, "gd": 0, "gl": 0, "ac": 0, "lk": 0, "md": 0, "fi": 0, "sx": 0,
    "lc": 0, "es": 0, "cc": 0, "cm": 0, "la": 0, "co.za": 0, "je": 0, "cz": 0,
    "jp": 0, "ai": 0, "pw": 0, "bg": 0, "nu": 0, "ag": 0, "bm": 0, "eu": 0,
    "com.my": 0, "sc": 0, "ax": 0, "wf": 0, "ly": 0, "qa": 0, "vn": 0, "aq": 0,
    "mobi": 0, "com.tr": 0, "com.ua": 0, "com.py": 0, "hk.org": 0,
    "south.am": 0, "com.kh": 0, "co.zm": 0, "ru.net": 0, "com.km": 0, "tt": 0,
    "kn": 0, "co.ls": 0, "co.fk": 0, "uy.com": 0, "com.gu": 0, ".com.bn": 0,
    "com.pf": 0, "com.fj": 0
}
SEED = "IbjGOEgnuD"


def dga(date):
    def unix(date):
        unix = int(time.mktime(date.timetuple()))
        return unix

    HOUR = 3600
    DAY = 24*HOUR
    for interval in [15*DAY, 5*DAY, 1*DAY, 8*HOUR, 2*HOUR, 1*HOUR]:
        for length in [5, 3, 4]:
            for tld, l in TLDS.items():
                min_length = l + length
                key = f"{unix(date)//interval}{SEED}{tld}\n".encode('ascii')
                key_hash = hashlib.sha1(key).digest()
                key_hash_b64 = base64.b64encode(key_hash).decode('ascii')
                key_hash_b64_noeq_lc = key_hash_b64.rstrip("=").lower()
                trantab = str.maketrans("-+/", "abc")
                hostname_src = key_hash_b64_noeq_lc.translate(trantab)
                hostname_len = max(min_length, 3)
                hostname = hostname_src[:hostname_len]
                domain = f"{hostname}.{tld}"
                yield domain


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="QSnatch dga")
    parser.add_argument(
        "-d", "--datetime",
        help="date time for which to generate domains, e.g., "
             "2019-11-11 18:00:00")
    args = parser.parse_args()
    if args.datetime:
        d = datetime.strptime(args.datetime, "%Y-%m-%d %H:%M:%S")
    else:
        d = datetime.now()
    for domain in dga(d):
        print(domain)

Version B

Version B of the DGA is simpler than the previously described version. I found these five samples on Virustotal that use the simpler version of the DGA:

MD5SHA1SHA256SizeUploaded to Virustotal
8cee2a187198648c199c1d135c918a3aa9f39f3b832344a79d32d92ac56c50cdaff0b93c09ab3031796bea1b8b79fcfd2b86dac8f38b1f95f0fce6bd2590361f6dcd676419KB, 19120 Bytes2019-09-19 16:02:14 UTC
c49ac8cfe022ff6acb8eb0036e2fc1a1e30ce38ff0ce46d8256d06fb3d5e13bf3abb10125cb5dce0a1e03fc4d3ffc831e4a356bce80e928423b374fc80ee997e7c62d3f819KB, 19038 Bytes2019-11-04 21:08:21 UTC
4affa116b27f2d977a756e353f77b8f5e8bb081056542504b5a69bd5f202cf77fac0a64f8fd16e639f99cdaa7a2b730fc9af34a203c41fb353eaa250a536a09caf78253b52KB, 53113 Bytes2019-09-19 17:14:38 UTC
421f006756f72cabc1ffb796c6cdb5c05ca92d6f02019519de593758583d7ca5a4bf9f235130282cdb4e371b5b9257e6c992fb7c11243b2511a6d4185eafc0faa0e0a3a628KB, 28671 Bytes2019-10-07 22:53:17 UTC
421240952a097e904df778590caa966858523de660632c6b84ffbd243cc75f4fb576980a15892206207fdef1a60af17684ea18bcaa5434a1c7bdca55f460bb69abec0bdc22KB, 21938 Bytes2019-09-18 02:37:51 UTC

Examples of domains of this DGA are:

t2q2rs.cf
t2q2rsa.cf
t2q2rsaz.cf
t2q2rsazo.cf
t2q2rsazo1.cf
gc9nzf.tk
gc9nzfb.tk
gc9nzfbt.tk
gc9nzfbt3.tk
gc9nzfbt3i.tk

Bash

These are the lines of QSnatch that generate the domains:

The DGA

The DGA uses fewer tlds, without tld-specific hostname length specifiers. The algorithm only uses one timespan (15 days). It uses more hostname lengths (6, 7, 8, 9 and 10) which are generated one after another for a specific hostname pattern.

The generation of the string for the hostname is the same as for the more complicated DGA version, including the seed IbjGOEgnuD. Because Version A mostly uses hostname lengths of 5 and shorter, while Version B uses lengths of 6 up to 10, only some domains from Version A with an extra character overlap with Version B (.com, .net, .org). For example, 07tvvc.com is a valid domain for both DGAs.

Python

import time
import hashlib
import base64
import argparse
from datetime import datetime


TLDS =[
    'cf', 'tk', 'ml', 'ga', 'gq', 'com', 'biz', 'org', 'de', 'rocks', 
    'mx', 'cn', 'top', 'nl', 'men', 'ws', 'se', 'info', 'xyz', 'net', 'today', 
    'ru', 'fi', 'name', 'to', 'in', 'com.ua', 'vg', 'vn', 'cd'
]
SEED = "IbjGOEgnuD"


def dga(date):
    def unix(date):
        unix = int(time.mktime(date.timetuple()))
        return unix

    HOUR = 3600
    DAY = 24*HOUR
    INTERVAL = 15*DAY
    for tld in TLDS:
        key = f"{unix(date)//INTERVAL}{SEED}{tld}\n".encode('ascii')
        key_hash = hashlib.sha1(key).digest()
        key_hash_b64 = base64.b64encode(key_hash).decode('ascii')
        key_hash_b64_noeq_lc = key_hash_b64.rstrip("=").lower()
        trantab = str.maketrans("-+/", "abc")
        hostname_src = key_hash_b64_noeq_lc.translate(trantab)
        for hostname_len in range(6, 11):
            hostname = hostname_src[:hostname_len]
            domain = f"{hostname}.{tld}"
            yield domain


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="QSnatch DGA Version 1")
    parser.add_argument(
        "-d", "--datetime",
        help="date time for which to generate domains, e.g., "
             "2019-11-11 18:00:00")
    args = parser.parse_args()
    if args.datetime:
        d = datetime.strptime(args.datetime, "%Y-%m-%d %H:%M:%S")
    else:
        d = datetime.now()
    for domain in dga(d):
        print(domain)

You also find both versions of the QSnatch DGA in my collection of domain generation algorithms on GitHub.