Re: Who does this really affect, its hard to tell....
Trump followers: 32m on Twitter plus his POTUS account of 18m and 22M on Facebook
I think that we can afford to lose 60m users.
A multi-year effort to update the internet's overall security has been put on hold just days before it was due to be introduced, over fears that as many as 60 million people could be forced offline. DNS overseer ICANN announced on Thursday it had postponed the rollout of a new root zone "key signing key" (KSK) used to secure …
I suspect it only affects DNS resolvers running on ISP servers, and individual users won't be affected at all, even when you're running an ancient version of windows...
In order to state the query as 'authoritative' you'll need that key stuff mentioned in RFC 8145 is for the conversation between the 'resolver' (running on the client) and the server, using DNSSEC. I don't believe that DNSSEC is actually _REQUIRED_ though, and older servers should still work.
I would expect older clients NOT using DNSSEC to work just fine, also.
If you're trying to resolve the queries yourself, and NOT use an ISP server for DNS, then maybe this will affect you. Or not.
If you're using a forwarding server or cacheing server from your ISP (or 188.8.131.52 for google's DNS server) then I'd expect it to work just fine and not break anything.
but worst case you could temporarily turn off DNSSEC [though I doubt it would be necessary]
The question here is MOST likely what the ISP cacheing servers will be doing, and whether those would all need to be updated. And yeah, it could cause a BIG problem if they can't resolve DNS queries any more...
> If you're using a forwarding server or cacheing server from your ISP (or 184.108.40.206 for google's DNS server) then I'd expect it to work just fine and not break anything.
Thats not right. If your ISP enabled DNSsec resolvers, and their system doesn't follow the automatic KSK addition mechanism that is required for the 2017 KSK key roll, *all* their lookups will fail when the old 2011 KSK stops signing the responses from '.'
The client doesn't request DNSsec (well, it could and check itself), but all the resolvers upstream need to be able to follow the KSK addition into their keystores via the proper method if they do DNSsec resolving themselves. (which most ISP servers do, unless your ISP is a small podunk one that doesn't follow current standards). All client lookups will fail of the ISP resolver is broken.
Since Government users typically demand that due to their standards they have to follow, most large ISPs have followed suit.
The organization is planning to publish a full list of resolvers that listed having only the 2010 KSK key, and then ask the internet community to help identify where they are and figure out what the problem is, and how to update them.
Did anyone else read this and think that these resolvers are likely to be running out of date software with known issues? I expect certain sections of the internet community will be very quick in identifying the servers and updating them, but I'm not sure their actions will be considered 'helpful'.
I say just do it.
Then all the companies running DNS servers and not maintaining them will get an in-flood of angry customers some will leave and the company will learn it's lesson to not leave stale infrastructure out there.
Heck doing it is likely to also halt a number of DDoS attack vectors due to loads of DNS servers suddenly being updated.
"DNS is hierarchical in nature, and we don't know how far up the tree the rot has settled."
And you'd hope they'd quickly identify and fix it. Sometimes it takes things to break completely before someone does something about creaky infrastructure, in fact that's how a lot of companies work with their IT budget!
If the servers are 'unidentified' then the only real way to find them is to implement the changes. If they pussyfoot around this then they'll never be in a position to make the switch. The broken link will soon be discovered and whoever owns/provides the DNS server can then explain to their customers why service was lost,.
"Unless they're servers for entire regions or even companies, meaning nowhere to defect. "
That kind of defection is accompanied by a P45 or a monopolies investigation.
Back in the days when open mail relays were a major problem, getting japanese admins to fix their boxes was quite hard, with most either claiming they were required to keep the things open, were fully standards compliant or simply blocked complaints from people who'd been hit, making open japanese relays an seemingly intractable problem...
...Until some bright sparks in Tokyo hit on the idea of notifying japanese media about the problems and the TV channels delighted in naming and shaming companies which were assisting hackers/spammers by not securing their computer networks(*) - which wasn't so much loss of face as being publicly kicked in the ballsack as far as management was concerned. As a result it usually took less than a day from the time that reporters started asking questions to the time that the servers either went offline permanently or were fixed.
(*) The public had been sensitised to the problem due to massive spam campaigns targetting mobiles.
Management facing adverse publicity and/or investigations in western countries has about the same reaction. Embarrassment is a fantastic teacher/persuader when dealing with refuseniks (either the management who refuse to let changes be made or the admins who refuse to do it)
I meant to say entire COUNTRIES. Plus you can't tell if it's YOUR country that would be affected or not since the servers could be upstream of you. Sounds fun to say, "let 'em suffer" until you discover YOU'RE suffering...AND can't change ISPs because ALL of them were affected at the same time.
"I say just do it.
Then all the companies running DNS servers and not maintaining them will get an in-flood of angry customers some will leave and the company will learn it's lesson to not leave stale infrastructure out there."
Maybe a half-way house. Roll it out for a short period then roll back (they did have a roll back facility in their plans, didn't they? Then announce "Did you have a problem? If so you've got a month to fix it, then it's permanent."
"Yeah, just fuck over millions of customers and businesses just to make a point to a few ISPs. Great plan."
I can only assume the 10 thumbs down I got for this post mean there really are 10 people too stupid to spot blindingly obvious sarcasm, or there are 10 people who really are such bastards that they think entire communities should suffer just to make a point to some DNS admins at an ISP. Either way, you're morons.
actually to quote Matt Larson :
" Historically there has been no way to determine which trust anchors DNSSEC validators have configured, making it difficult to assess the potential impact of the root KSK rollover. "Signaling Trust Anchor Knowledge in DNS Security Extensions (DNSSEC)" (defined in RFC 8145) is a recent protocol extension that allows a validator to report which trust anchors it has configured for a zone to that zone's name servers. The protocol was only finalized in April, 2017, and only the most recent versions of BIND (9.10.5b1 and 9.11.0b3 and later) and Unbound (1.6.4 and later) support it. This protocol was not expected to have sufficient deployment to provide useful information for the first root KSK rollover."
updating the DNS infrastructure has a side effect that finally DNSSEC with ECC specifically Ed25519
ed25519 is a public-key signature system invented by Bernstein et al. that is standardized for use in internet protocols as RFC 8032. In RFC 8080, ed25519 (and ed448) were standardized for use in DNSSEC in February 2017.
DNS software supporting ed25519
An incomplete list of DNS tools and servers that support or will support ed25519:
BIND 9 (unreleased) has support based on the upcoming OpenSSL 1.1.1
Knot DNS (unreleased) will support ed25519 in 2.6 using GnuTLS 3.6.0
Unbound (since 1.6.4) supports ed25519 via the unreleased OpenSSL 1.1.1 and since version 1.6.6 via libnettle
PowerDNS Authoritative Server (since 4.0.0) has support for signing with ed25519 using libsodium
PowerDNS Recursor (since 4.0.5) has support for validating ed25519 using libsodium
ed25519 signature is 64 bytes long, compared to 256 bytes for an RSA 2048 signature. These smaller signatures ensure that DNS amplification attacks are less severe than before, without sacrificing the security of DNSSEC.
I do not quite understand. It does not appear that the new KSK is somehow linked to the new signing algorithm ed25519 , or at least the article is silent on this. From what I read, it is a new DNSSEC signing key for the root servers, so as long as your (new or not so-new) DNS server has the new DNSSEC keys installed, it should just work. The RFC 8145 seem to be about enabling the root (and not so root) DNS servers to build up a knowledge about who has the new key and who has not, and it does not need to be installed everywhere. Neither seem related to the new key signing algo.
Unless the requirement of the new key is the use of the new signing algo, which would be a pretty important point to make in the article (and yet it is no there)?
Umm BIND is currently releasing 9.11.2, 9.10.6, 9.9.11
dnssec looks like it was added about 9.9 - see matrix
Additional info - addition of ED25519 alg is in proto 9.12 - in beta release:
So is this something that only ISPs have to know about - as they are the ones running DNS servers mainly?
Could an affected end user with a crappy ISP just tell his computer to use google's 220.127.116.11?
It would be fun to email my ISP for confirmation thy are aware of this upcoming change.
Are there more layers of DNS ( like managers) . some group of "master servers"?
EE might say to me - Yeah we're ready but unable to get comfirmation from our DNS chain wholsaler that they are prepared.
"Are there more layers of DNS ( like managers) . some group of "master servers"?"
Yes. DNS is hierarchical in nature with varying levels. The "master servers" are the 13 root servers at the top of the hierarchy. It's THEIR keys that are being changed, and each layer down needs to copy or everything below them can go dark.
All of those, and more, are solved at other levels of the protocol stack (TLS, SSH, etc). No need to get DNS involved. Or benefit from doing so.
If you haven't protected the actual data stream, you need to do so regardless since there are many ways to direct it into the arms of an attacker apart from the few venues DNSSEC protects against. And once you have done so, you no longer need DNSSEC with its downsides.
DNS just about barely works as it is. The average quality of zones and delegations is simply appalling. Adding a PKI on top of it certainly won't help it work better. And considering adding said PKI has no benefits whatsoever *, why even consider doing it, much less do it on a global scale?
* At most it would stop something along the lines of Web sites getting defaced because of DNS hijacks, like after compromising hosts in the same subnets as the DNS servers. Not exactly an issue worth remodeling fundamental parts of the Internet for... If you can have much more fun than that, the issue is with the actual service (read: cleartext authentication, no HTTPS, etc), not with the DNS infrastructure.
PS. The proper "upgrading of tools for DNSSEC" in a small organization consists of making sure it's disabled everywhere.
Having just completed a training course which covered DNSSEC, the Root Key Signing Keys are using algorithm 8 which RSA/SHA256 and is compatible with Bind 9.10.3 which is available for MintLinux and Ubuntu Mate for Raspberry Pi. The new KSK works just fine. After five years of giving training on DNSSEC, 99% of students who were familiar with DNS had never heard of DNSSEC (despite it being 16 years old).
For those unfamilar with DNSSEC at present there is a double Key Rollover, both the KSK and Zone Signing Key (ZSK) are rolling over to a new key.
dig @18.104.22.168 . dnskey +multiline
; <<>> DiG 9.10.6 <<>> @22.214.171.124 . dnskey +multiline
; (1 server found)
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 5983
;; flags: qr rd ra ad; QUERY: 1, ANSWER: 4, AUTHORITY: 0, ADDITIONAL: 1
;; OPT PSEUDOSECTION:
; EDNS: version: 0, flags:; udp: 512
;; QUESTION SECTION:
;. IN DNSKEY
;; ANSWER SECTION:
. 166290 IN DNSKEY 257 3 8 (
) ; KSK; alg = RSASHA256 ; key id = 19036
. 166290 IN DNSKEY 256 3 8 (
) ; ZSK; alg = RSASHA256 ; key id = 15768
. 166290 IN DNSKEY 256 3 8 (
) ; ZSK; alg = RSASHA256 ; key id = 46809
. 166290 IN DNSKEY 257 3 8 (
) ; KSK; alg = RSASHA256 ; key id = 20326
;; Query time: 9 msec
;; SERVER: 126.96.36.199#53(188.8.131.52)
;; WHEN: Fri Sep 29 19:51:13 BST 2017
;; MSG SIZE rcvd: 1128
DNSSEC is used to digitally sign DNS records by adding a RRSIG record which is cryptographically signed hash of a set of records within a Zone. So the NS records inside a Zone would have a single RRSIG to allow authentication and validation of all the NS records as if you query the NS records all will be returned.
In terms of handling the new KSK, if the DNS Server supports RFC5011 the KSK rollover is handled automatically BIND has support this for a couple of years. RFC7344 allows the automatic update of the Child Zone Delegated Signer Record using the CDS Record and CDNSKEY record both supported in Bind 9.11
Unfortunately many commercial DNS providers do not support RFC5011, nor do they support RFC7344.
The Root KSK should have been replaced in June 2015 under the contract ICANN operated under here we are seven years later and the its going to be delayed again.
Personally the root key change should not be delayed... A competent DNS Architect/Administrator should have planned for this two years ago.
> Just look at IPv6
I'm looking at IPv6. Mobile really made it a slam dunk use-case.
56-60% of all my email users come in over IPv6.
I'm not a large web content provider, so I can't show the same stats there, but I'd bet that Facebook is showing numbers equally impressive.
Look at the ISPs or companies like Facebook that are 100% IPv6 internal with only IPv4 gateways now.
Look at the IOC 2017 IPv6 report for more evidence of ISPs considering dropping IPv4 native in the next *handful* of years.
The one case where everybody is dragging their feet?
Enterprise fears IPv6, buried their heads in the sand, even though they probably have significant IPv6 traffic internally traversing their network. They need to figure out that those OSs running internally are all doing IPv6 native now, and learn how to properly secure it (a single external breach could setup a IPv6 RA and proxy, and funnel all the Enterprise traffic out beyond the firewall in a heartbeat) and embrace it. IPv4 is going away, Enterprise needs to learn that.
Well, enterprise happens to be most likely place to find IPv4-ONLY equipment: acquired before IPv6 was a thing yet too indispensable and/or too expensive to replace. You can't roll out IPv6 selectively because the old stuff will lose touch with complicated bodges, becoming a case of "If it ain't broke..." best not to rock the boat internally and address external IPV6 needs via specialized structures: dedicated subnets, gateways, proxies, etc.
Biting the hand that feeds IT © 1998–2019