Network Working Group Robert Elz Internet Draft University of Melbourne Expiration Date: December 1996 Randy Bush RGnet, Inc. Scott Bradner Harvard University June 1996 Selection and Operation of Secondary DNS Servers draft-ietf-dnsind-2ndry-02.txt Status of this Memo This document is an Internet-Draft. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." To learn the current status of any Internet-Draft, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe), munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast). 1. Abstract This draft discusses the selection of secondary servers for DNS zones, and related issues. kre & randy & sob [Page 1] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 2. Introduction Poor choice of secondary servers for DNS zones seems to currently be an endemic problem. This draft discusses some of the issues, and attempts to give some guidance in the matter of the selection of the required secondary DNS server(s). 3. Definitions For the purposes of this document, and only this document, the following definitions apply: DNS The Domain Name System [RFC1034, RFC1035]. Zone A part of the DNS tree, that is treated as a unit. Forward Zone A zone containing data used to map host names and domains to addresses, mail exchange targets, etc. Contrasts with Reverse Zone, used to map addresses back to names. Server An implementation of the DNS protocols able to provide answers to queries. Answers may be from information known by the server, or information obtained from another server. Authoritative Server A server that knows the content of a DNS zone from local knowledge, and thus can answer queries about that zone without needing to query other servers. Listed Server An Authoritative Server for which there is an "NS" resource record (RR) in the zone. Primary Server An authoritative server for which the zone information is locally configured. Sometimes known as a Master server. Secondary Server An authoritative server that obtains information about a zone from a Primary Server via a zone transfer mechanism. Sometimes known as a Slave Server. Stealth Server An authoritative server, usually secondary, which is not a Listed Server. kre & randy & sob [Page 2] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 4. Secondary Servers A prime purpose of secondary servers is to allow information from the Domain Name System to be available widely and reliably to clients throughout the Internet (that is, throughout the world), even when the primary server is unavailable or unreachable. They can also spread the name resolution load, that purpose is not treated further here. When selecting secondary servers, attention should be given to the various likely failure modes, and servers should be placed so that it is likely that at least one server will be available to all significant parts of the Internet, for any likely failure. Consequently, placing all servers at the local site, while easy to arrange, and easy to manage, is not a good policy, as they are susceptible, together, to being disconnected from the Internet due to a single link failure, or a site (or sometimes, building or even room) wide power failure. Secondary servers should be placed at both topologically and geographically dispersed locations on the Internet, to minimise the likelihood of a single failure disabling all of them. That is, secondary servers should be at geographically distant locations, so it is unlikely that events like power loss, etc, will disrupt all of them simultaneously. The should also be connected to the net via quite diverse paths, so the failure of any one link, or of routing within some segment of the network (such as an ISP) will not make all of the servers unreachable. While it is unfortunately quite common, servers for a zone should certainly not all be placed on the same LAN segment in the same room of the same building - or any of those. Such a configuration almost defeats the requirement, and utility, of having multiple servers. The only redundancy provided in that configuration is for the case when one server is down, whereas there are many other possible failure modes. 5. Unreachable servers Similarly, listing as a server, or as the address of a server, a name or value that cannot be reached from much of the network is undesirable. Not only does such a server add no reliability at all, it actually causes the network as a whole (the Internet) problems, as the lack of reachability cannot be ascertained other than by attempted use and the subsequent lack of response (or occasionally ICMP error response). Further, even that is generally kre & randy & sob [Page 3] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 indistinguishable from a simple packet loss, so the sequence must be repeated, several times, to give any real evidence of an unreachable server. Further, the whole thing needs to be repeated from time to time to distinguish a permanently unreachable server from a temporarily unreachable one. Worst of all, all this may potentially be done by every other DNS server on the Internet. To avoid this problem, NS records for a zone returned in any response should list only servers that the client requesting the information, and others to which it may forward the reply, are likely to be able to reach. Additionally, addresses of the servers returned must all be reachable. As the addresses of each server form a Resource Record Set [KRE1996b], all must be returned (or none), thus it is not acceptable to elide addresses of servers that are unreachable, or to return them with a low TTL (while returning others with a higher TTL). In particular, when some servers are behind a firewall, which disallows DNS queries or responses, their names, or addresses, should not be returned to clients outside the firewall. Similarly, servers outside the firewall should not be made known to clients inside it, if the clients would be unable to query those servers. Implementing this usually requires dual DNS setups, one for internal use, the other for external use. Such a setup often solves other problems with environments like this. When a server is at a firewall boundary, reachable from both sides, but using different addresses, that server should be given two names, each name associated with appropriate A records, such that each appears to be reachable only on the appropriate side of the firewall. This should then be treated just like two servers, one on each side of the firewall. Special care will need to be taken to allow such a server to return the correct responses to clients on each side. A similar problem occurs with DNS servers located in parts of the net that are often disconnected from the Internet as a whole, for example, those which connect via an intermittent connection that is often down. A similar solution may need to be adopted, though here much of the zone information can often be replicated, with only NS records being adjusted. Servers in this environment often need special provision to give them access to the root servers. Often this is accomplished via "fake root" configurations. In such a case the servers should be kept well isolated from the rest of the DNS, lest their unusual configuration pollute others. kre & randy & sob [Page 4] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 6. How many secondaries? The DNS specification requires at least two servers for every zone. That is, usually, the primary and one secondary. While two, carefully placed, are usually sufficient, occasions where two are insufficient are frequent enough that we advise the use of more than two listed servers. Various problems can cause a server to be unavailable for extended periods - during such a period, a zone with only two listed servers is actually running with just one. Since any server may occasionally be unavailable, for all kinds of reasons, this zone is likely, at times, to have no functional servers at all. On the other hand, having large numbers of servers adds little benefit, while adding costs. At the simplest, more servers cause packets to be larger, so requiring more bandwidth. This may seem, and realistically is, trivial, however there is a limit to the size of a DNS packet, and causing that limit to be reached has more serious performance implications. It is wise to stay well clear of it. More servers also increase both the likelihood that one server will be misconfigured, or malfunction, without being detected. It is recommended that three servers be provided for most organisation level zones, with at least one well removed from the others. For zones where even higher reliability is required, four, or even five, servers may be desirable. Two, or occasionally three of five, would be at the local site, with the others not geographically or topologically close to the site, or each other. Reverse zones, that is, sub-domains of .IN-ADDR.ARPA, tend to be less crucial, and less servers, less distributed, will often suffice. 6.1. Stealth Servers Servers which are authoritative for the zone, but not listed in NS records (also known as "stealth" servers) are not included in the count of servers. It can often be useful for all servers at a site to be authoritative (secondary), but only one or two be listed servers, the rest being unlisted for all local zones, that is, to be stealth servers. This allows those servers to provide answers to local queries directly, without needing to consult another server. If it were necessary to consult another server, it would usually be necessary for the root servers to be consulted, in order to follow the delegation tree - that the zone is local would not be known. This would mean that some local queries may not be able to be answered if external communications were disrupted. kre & randy & sob [Page 5] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 Listing all such servers in NS records, if more than one or two, would cause the rest of the Internet to spend unnecessary effort attempting to contact all servers at the site when the whole site is inaccessible due to link or routing failures. 7. Serial Number Maintenance Secondary servers use the serial number in the SOA record of the zone to determine when it is necessary to update their local copy of the zone. Serial numbers are basically just 32 bit unsigned integers that wrap around from the biggest possible value to zero again. See [KRE1996a] for a more complete definition of the serial number. The serial number must be incremented every time a change, or group of changes, is made to the zone on the primary server, in order to cause secondary servers to update their copies of the zone. Note that it is not possible to decrement a serial number, increments are the only defined modification. Occasionally due to editing errors, or other factors, it may be necessary to cause a serial number to become smaller. Do not simply decrease the serial number, secondary servers will ignore that change, and further, will ignore any further increments until the earlier large value is exceeded. Instead, given that serial numbers wrap from large to small, in absolute terms, increment the serial number, several times, until it has reached the value desired. At each step, wait until all secondary servers have updated to the new value before proceeding. For example, assume that the serial number of a zone was 10, but has accidentally been set to 1000, and it is desired to set it back to 11. Do not simply change the value from 1000 to 11, a secondary server that has seen the 1000 value (and in practice, there is always at least one) will ignore this change, and continue to use the version of the zone with serial number 1000, until the primary server's serial number exceeds that value. This may be a long time - in fact, the secondary often expires its copy of the zone before the zone is ever updated again. Instead, for this example, set the primary's serial number to 2147484647, and wait for the secondary servers to update to that zone. The value 2147484647 is calculated by taking the current serial number (1000), adding 2^31 (2147483648) and subtracting 1, and taking the result modulo 4294967296 (2^32). That is by adding 2^31 - 1 (2147483647), which is the largest defined serial number kre & randy & sob [Page 6] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 increment [KRE1996a]. This value is a serial number that is larger than 1000, and further, is the biggest serial number that is larger than 1000. This latter attribute is not essential to the procedure, a slightly smaller number could be chosen. Next, after all servers needing updating have the zone with that serial number, the serial number can be set to 11, as 2147484647 + 2147483647 is 4294968294, and 4294968294 modulo 4294967296 is 998, which is bigger than 11. Hence the change from 2147484647 to 11 is an increment. Note that secondary servers that had not updated to the 1000 version of the zone will ignore the first step in this procedure and remain with their serial number set at 10, as 2147484647 is more than 2147483647 bigger than 10, and hence appears as an older version of the zone (that is, in serial number arithmetic, 2147484647 < 10, whereas 2147484647 > 1000). Clearly it is neither important, nor possible, to wait for those servers to follow this procedure. However, the second increment, to 11, will be seen as an increment by all servers, which will all update their copies of the zone, and hence all be consistent again. When following this procedure, it is essential to verify that all relevant servers have been updated at each step, never assume anything. Failing to do this can result in a worse mess than existed before the attempted correction. Also beware that it is the relationship between the values of the various serial numbers that is important, not the absolute values. The values used above are correct for that one example only. Also, note that not all nameserver implementations correctly implement serial number operations. With such servers as secondaries there is typically no way to cause the serial number to become smaller, other than contacting the administrator of the server and requesting that all existing data for the zone be purged. Then that the secondary be loaded again from the primary, as if for the first time. It remains safe to carry out the above procedure, as the malfunctioning servers will need manual attention in any case. After the sequence of serial number changes described above, conforming secondary servers will have been reset. Then when the primary server has the correct (desired) serial number, contact the remaining secondary servers and request their understanding of the correct serial number be manually corrected. Perhaps also suggest that they upgrade their software to a standards conforming implementation. kre & randy & sob [Page 7] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 A server which does not implement this algorithm is defective, and may be detected as follows. At some stage, usually when the absolute integral value of the serial number becomes smaller, a server with this particular defect will ignore the change. Servers with this type of defect can be detected by sending a query for the SOA, after waiting at least the time specified in the SOA refresh field, as it is reported from the secondary. Servers with this defect will still have the old serial number. We are not aware of other means to detect this defect. 8. Security Considerations This document does not consider security. The mention of firewalls in section 5 is purely because they are a fact of life (and an impediment to orderly communications). It is not intended to imply that a firewall is in any way useful for security purposes. It is not believed that anything in this document adds to any security issues that may exist with the DNS, nor does it do anything to lessen them. 9. References [RFC1034] Domain Names - Concepts and Facilities, P. Mockapetris, ISI, November 1987. [RFC1035] Domain Names - Implementation and Specification, P. Mockapetris, ISI, November 1987 [KRE1996a] Serial Number Arithmetic, R. Elz, R. Bush, Work in Progress (RFC pending), April 1996. [KRE1996b] Clarifications to the DNS specification, R. Elz, R. Bush, Work In Progress (internet-draft), May 1996. kre & randy & sob [Page 8] Internet Draft draft-ietf-dnsind-2ndry-02.txt June 1996 10. Authors' Addresses Robert Elz Computer Science University of Melbourne Parkville, Vic, 3052 Australia. EMail: kre@munnari.OZ.AU Randy Bush RGnet, Inc. 9501 SW Westhaven Portland, Oregon, 97225 United States. EMail: randy@psg.com Scott Bradner Harvard University 1350 Mass Ave Cambridge, MA, 02138 United States. EMail: sob@harvard.edu kre & randy & sob [Page 9]