Up till very recently, through the samples we had learned that the Mirai DGA seeds are all fixed to 0, as detailed in blog Now Mirai Has DGA Feature Built in, and were able to predict all corresponding DGA domains.
Surprisingly, although we have not see any related samples, just few days ago, our PassiveDNS based anomaly detect module captured some new domains that matches the characteristics of the mirai DGA algorithm but not belong to any seed 0 series.
The L2 domains conform to mirai DGA: 12 characters and a-y only. But these domains do not belong to DGA seed 0 series. This leads us to speculate that a new mirai variant and new DGA seed is emerging.
DGA Domains Generated With Non-zero DGA Seed
The new suspicious domain are listed below:
And the domains's query patterns are as follows:
A detailed study on these domains shows more findings:
- The L2 domains conform to mirai DGA, 12 characters, a-y only. But these domains does not belong to DGA seed 0 series.
- All TLDs for these domains are fixed to
.online, differing from previous TLDs of
support. Maybe the author want just keep "online", but no more "tech support".
- Of all the domains, the first seen starts almost strictly at 00:00:00 of the day, with the last seen almost always strictly ends at 00:00:00 on the next. Those time window overlaps between the two consequential domains are very short, in minutes. Considering such a strict time window control, we think one possible explanation is that those DNS queries were eventually launched by some mirai bots.
New DGA Seed 0x91 Brute-forced
Under the guidance of the precise mirai algorithm, we are pretty sure we are able to brute-force the new DGA seeds, no matter where it hides in the int32 4G space. And the brute force did not take long before we cracked the new seed:
The new DGA seed is used to predict all the 2016-12 domains, as follows. Note that almost all these domains are registered by mirai author, with the exception 18/19/27/29/30 on which the algorithm itself can not generated a functional DGA domain.
- 2016/12/08 pcrpxewicouh.online
- 2016/12/09 rwoywonuobcr.online
- 2016/12/10 liusqxocbedg.online
- 2016/12/11 ndoiabgxgmew.online
- 2016/12/12 hwjqtwkectlo.online
- 2016/12/13 mtoyrjnvlqdx.online
- 2016/12/14 ddfqdttkmoyv.online
- 2016/12/15 iaxjxyqjckqi.online
- 2016/12/16 nuuxndqlhiwb.online
- 2016/12/17 pxvmqwpemiif.online
- 2016/12/18 dfftxpajygxy
- 2016/12/19 fiotbgopgnxv
- 2016/12/20 shjwhbdgjyba.online
- 2016/12/21 xmjvlucdsegk.online
- 2016/12/22 irkbpugkwsir.online
- 2016/12/23 nwnrbhnesmtk.online
- 2016/12/24 wafunxdngsfc.online
- 2016/12/25 ydcvedrxcmym.online
- 2016/12/26 sggiyqadywsv.online
- 2016/12/27 ujmnvkyeltfv
- 2016/12/28 cmhewcvopvno.online
- 2016/12/29 hrjlyymassqx
- 2016/12/30 xsvftelyclfh
- 2016/12/31 dxukryyyqnhl.online
Although we can explain most of what we saw in our data, there are still some open questions. We list them here with a hope to see feedback from the security community.
- On December 12, there is no 0x91 DGA domain and corresponding DNS data flow. As shown in the previous table, the 2016-12-12 0x91 domain is hwjqtwkectlo.online. This domain has not been registered yet, and we are pretty sure there is no DNS query for this domain, otherwise we could tell from the DNS error data stream. Why this domain is not registered? Maybe the author forget to do that. But why there is not even a single DNS request? Especially given the continuous DNS requests before and after the day?
- For all the bot behind the DNS query, are they from the same group, or totally different ones in each day? For the former, why do they say good-bye to the old C2 and embrace the new one at about exactly 00:00:00? For the latter, why the traffic waves look so similar for each days?
- The bot traffic keeps falling in recent days, why? Are these bots completely get out of the mirai botnet? Or just captured by another mirai C2?