--- 1/draft-ietf-dnsext-wcard-clarify-00.txt 2006-02-04 23:12:57.000000000 +0100 +++ 2/draft-ietf-dnsext-wcard-clarify-01.txt 2006-02-04 23:12:57.000000000 +0100 @@ -1,20 +1,20 @@ Internet Engineering Task Force B. Halley Internet-Draft Nominum E. Lewis ARIN -June 17, 2003 Expires: December 17, 2003 +August 10, 2003 Expires: February 10, 2004 Clarifying the Role of Wild Card Domains in the Domain Name System - + Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. 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. @@ -22,51 +22,56 @@ 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". The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. + Comments on this document can be sent to the editors or the mailing + list for the DNSEXT WG, namedroppers@ops.ietf.org. + Abstract The definition of wild cards is recast from the original in RFC 1034, in words that are more specific and in line with RFC 2119. This document is meant to supplement the definition in RFC 1034 and to alter neither the spirit nor intent of that definition. 1 Introduction The first section of this document will give a crisp overview of what -is begin defined, as well as the motivation for what amounts to a -simple rewording of an original document. Examples are included to -help orient the reader. +is begin defined, as well as the motivation rewording of an original document +and making a change to bring the specification in line with implementations. +Examples are included to help orient the reader. Wild card domain names are defined in Section 4.3.3. of RFC 1034 as "instructions for synthesizing RRs." [RFC1034] The meaning of this is that a specific, special domain name is used to construct responses in instances in which the query name is not otherwise represented in a zone. A wild card domain name has a specific range of influence on query names (QNAMEs) within a given class, which is rooted at the domain name containing the wild card label, and is limited by explicit entries, zone cuts and empty non-terminal domains (see section 1.3 of this document). Note that a wild card domain name has no special impact on the search for a query type (QTYPE). If a domain name is found that matches the QNAME (exact or a wild card) but the QTYPE is not found at that point, the proper response is that there is no data available. The search does not continue on to seek other wild cards that might match the QTYPE. To illustrate, a wild card owning an MX RR does not 'cover' other names -in the zone that own an A RR. +in the zone that own an A RR. There are certain special case RR types +that will be singled out for discussion, the SOA RR, NS RR, CNAME RR, +and DNAME RR. Why is this document needed? Empirical evidence suggests that the words in RFC 1034 are not clear enough. There exist a number of implementations that have strayed (each differently) from that definition. There also exists a misconception of operators that the wild card can be used to add a specific RR type to all names, such as the MX RR example cited above. This document is also needed as input to efforts to extend DNS, such as the DNS Security Extensions [RFC 2535]. Lack of a clear base specification has proven to result in extension documents that have unpredictable consequences. (This is true in general, not just @@ -76,41 +81,48 @@ regarding authenticated denial of existence, a service introduced in the DNS Security Extensions [RFC 2535]. Prior to the work leading up to this document, it had been feared that a large number of proof records (NXTs) might be needed in each reply because of the unknown number of potential wild card domains that were thought to be applicable. One outcome of this fear is a now discontinued document solving a problem that is now known not to exist. I.e., this clarification has the impact of defending against unwarranted protocol surgery. It is not "yet another" effort to just rewrite the early specifications for the sake of purity. +Although the effort to define the DNS Security Extensions has prompted this +document, the clarifications herein relate to basic DNS only. No DNS +Security Extensions considerations are mentioned in the document. + 1.1 Document Limits -This document limits itself to reinforcing the concepts in RFC 1034. -Any deviation from this should be brought to the attention of the editors. +This document limits itself to reinforcing the concepts in RFC 1034. In +the effort to do this, a few issues have been discussed that change +parts of what is in RFC 1034. The discussions have been held within the +DNS Extensions Working Group. -Two changes to the text of RFC 1034 that fall within the realm of -clarifying the wild card definition have been suggested. (Changes aren't -really clarifications.) The two suggestions are barring the ownership -by a wild card domain of an CNAME resource record set and barring the -ownership by a wild card domain of a NS resource record set. Both -of these have some merit, but do not belong in a document that has not -yet been reviewed by the working group. +Briefly, the issues raised include: + - The lack of clarity in the definition of domain name existence + - Implications of a wild card domain name owning any of the following + resource record sets: DNAME [RFC 2672], CNAME, NS, and SOA + - Whether RFC 1034 meant to allow special processing of CNAME RR's + owned by wild card domain names 1.2 Existence The notion that a domain name 'exists' will arise numerous times in this discussion. RFC 1034 raises the issue of existence in a number of places, usually in reference to non-existence and often in reference to processing -involving wild card domain names. RFC 1034 does contain algorithms that +involving wild card domain names. RFC 1034 contains algorithms that describe how domain names impact the preparation of an answer and does -define wild cards as a means of synthesizing answers. +define wild cards as a means of synthesizing answers. Because of this +a discussion on wild card domain names has to start with the issue of +existence. To help clarify the topic of wild cards, a positive definition of existence is needed. Complicating matters, though, is the realization that existence is relative. To an authoritative server, a domain name exists if the domain name plays a role following the algorithms of preparing a response. To a resolver, a domain name exists if there is any data available corresponding to the name. The difference between the two is the synthesis of records according to a wild card. For the purposes of this document, the point of view of an authoritative @@ -118,21 +130,21 @@ the execution of the algorithms in RFC 1034. 1.3 An Example For example, consider this wild card domain name: *.example. Any query name under example. is a candidate to be matched (answered) by this wild card, i.e., to have an response returned that is synthesized from the wild card's RR sets. Although any name is a candidate, not all queries will match. -To further illustrate this, consider this example: +To further illustrate this, consider this zone: $ORIGIN example. @ IN SOA NS NS * TXT "this is a wild card" MX 10 mailhost.example. host1 A 10.0.0.1 _ssh._tcp.host1 SRV _ssh._tcp.host2 SRV @@ -199,30 +211,33 @@ interior node, owning or not owning any resource set, that has a leaf node owning a resource set as a subdomain," is the proper interpretation of the middle sentence. As an aside, an "authoritative name error" has been called NXDOMAIN in some RFCs, such as RFC 2136 [RFC 2136]. NXDOMAIN is the mnemonic assigned to such an error by at least one implementation of DNS. As this mnemonic is specific to implementations, it is avoided in the remainder of this document. -1.3 Terminology +1.5 Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in the document entitled "Key words for use in RFCs to Indicate Requirement Levels." [RFC2119] Requirements are denoted by paragraphs that begin with with the following convention: 'R'.. +Quotations of RFC 1034 (as has already been done once above) are denoted by +a '#' in the leftmost column. + 2 Defining the Wild Card Domain Name A wild card domain name is defined by having the initial label be: 0000 0001 0010 1010 (binary) = 0x01 0x2a (hexadecimal) This defines domain names that may play a role in being a wild card, that is, being a source for synthesized answers. Domain names conforming to this definition that appear in queries and RDATA sections do not have any special role. These cases will be described in more detail in @@ -276,21 +291,21 @@ A note on terminology. A domain transcends zones, i.e., all DNS data is in the root domain but segmented into zones of control. In this document, there are references to a "domain name" in the context of existing "in a zone." In this usage, a domain name is the root of a domain, not the entire domain. The domain's root point is said to "exist in a zone" if the zone is authoritative for the name. RR sets existing in a domain need not be owned by the domain's root domain name, but are owned by other domain names in the domain. -4 Impact of a Wild Card Domain In a Query Message +4 Impact of a Wild Card Domain In a Query Message or in an RDATA field When a wild card domain name appears in a question, e.g., the query name is "*.example.", the response in no way differs from any other query. In other words, the wild card label in a QNAME has no special meaning, and query processing will proceed using '*' as a literal query name. R4.1 A wild card domain name acting as a QNAME MUST be treated as any other QNAME, there MUST be no special processing accorded it. If a wild card domain name appears in the RDATA of a CNAME RR or any @@ -305,23 +320,36 @@ processing accorded it. 5 Impact of a Wild Card Domain On a Response The description of how wild cards impact response generation is in RFC 1034, section 4.3.2. That passage contains the algorithm followed by a server in constructing a response. Within that algorithm, step 3, part 'c' defines the behavior of the wild card. The algorithm is directly quoted in lines that begin with a '#' sign. Commentary is interleaved. -[Note that are no requirements specifically listed in this section. The -text here is explanatory and interpretative. There is no change to -the algorithm specified in RFC 1034.] +There is a documentation issue deserving some explanation. The algorithm +in RFC 1034, section 4.3.2. is not intended to be pseudo code, i.e., it's +steps are not intended to be followed in strict order. The "algorithm" +is a suggestion. As such, in step 3, parts a, b, and c, do not have to +be implemented in that order. + +Another issue needing explanation is that RFC 1034 is a full standard. +There is another RFC, RFC 2672, which makes, or proposes an adjustment +to RFC 1034's section 4.3.2 for the sake of the DNAME RR. RFC 2672 is +a proposed standard. The dilemma in writing these clarifications is +knowing which document is the one being clarified. Fortunately, the +difference between RFC 1034 and RFC 2672 is not significant with respect +to wild card synthesis, so this document will continue to state that +it is clarifying RFC 1034. If RFC 2672 progresses along the standards +track, it will need to refer to modifying RFC 1034's algorithm as +amended here. The context of part 'c' is that the search is progressing label by label through the QNAME. (Note that the data being searched is the authoritative data in the server, the cache is searched in step 4.) Step 3's part 'a' covers the case that the QNAME has been matched in full, regardless of the presence of a CNAME RR. Step 'b' covers crossing a cut point, resulting in a referral. All that is left is to look for the wild card. Step 3 of the algorithm also assumes that the search is looking in the zone closest to the answer, i.e., in the same class as QCLASS and as @@ -357,21 +385,21 @@ For the sake of brevity, the closest enclosing node can be referred to as the "closest encloser." The closest encloser is the most important concept in this clarification. Describing the closest encloser is a bit tricky, but it is an easy concept. To find the closest encloser, you have to first locate the zone that is the authority for the query name. This eliminates the need to be concerned that the closest encloser is a cut point. In addition, we can assume too that the query name does not exist, hence the closest encloser is not equal to the query name. We can assume away these two cases because they are -handled in steps a and b of section 4.3.2.'s algorithm. +handled in steps 2, 3a and 3b of section 4.3.2.'s algorithm. What is left is to identify the existing domain name that would have been up the tree (closer to the root) from the query name. Knowing that an exact match is impossible, if there is a "*" label descending from the unique closest encloser, this is the one and only wild card from which an answer can be synthesized for the query. To illustrate, using the example in section 1.2 of this document, the following chart shows QNAMEs and the closest enclosers. In Appendix A there is another chart showing unusual cases. @@ -382,23 +410,21 @@ _telnet._tcp.host2.example. host2.example. no wild card _telnet._tcp.host3.example. example. *.example. _chat._udp.host3.example. example. *.example. Note that host1.subdel.example. is in a subzone, so the search for it ends in a referral in part 'b', thus does not enter into finding a closest encloser. The fact that a closest encloser will be the only superdomain that can have a candidate wild card will have an impact when it comes to -designing authenticated denial of existence proofs. (This concept -is not introduced until DNS Security Extensions are considered in -upcoming sections.) +designing authenticated denial of existence proofs. # If the "*" label does not exist, check whether the name # we are looking for is the original QNAME in the query # or a name we have followed due to a CNAME. If the name # is original, set an authoritative name error in the # response and exit. Otherwise just exit. The above passage says that if there is not even a wild card domain name to match at this point (failing to find an explicit answer elsewhere), we are to return an authoritative name error at this point. If we were @@ -406,398 +432,147 @@ a no error return code is appropriate, with just the CNAME RR (or sequence of CNAME RRs) in the answer section. # If the "*" label does exist, match RRs at that node # against QTYPE. If any match, copy them into the answer # section, but set the owner of the RR to be QNAME, and # not the node with the "*" label. Go to step 6. This final paragraph covers the role of the QTYPE in the process. Note that if no resource record set matches the QTYPE the result is that no data -is copied, but the search still ceases ("Go to step 6."). - -6 Authenticated Denial and Wild Cards - -In unsecured DNS, the only concern when there is no data to return to -a query is whether the domain name from which the answer comes exists or -not, whether or not a name error is indicated in the return code. In -either case the answer section is empty or contained just a sequence of -CNAME RR sets. - -In securing DNS, authenticated denial of existence is a service that is -provided. The chosen solution to provide this service is to generate -resource records indicating what is protected in a zone and to digitally -sign these. - -The resource records that do this, as defined in RFC 2535, are NXT RRs. - -There are three points to consider when clarifying the topic of wild card -domain names. One is the construction of the records. The second is -the inclusion of records in responses. The third is the interpretation -of the records in a response by the resolver. - -In short, authenticated denial has to be sure to prove that the closest -encloser does not equal the query name, whether there is a wild card -name directly under the closest encloser. - -6.1 Preparing Wild Card Domain Name Owned Non-existence Proofs - -During the creation of the authenticated denial records, the wild card -domain name plays no special role, in the same manner as the wild card -domain name playing no special role in a query. - -There are two considerations with regards to preparing non-existence -proofs. - -R6.1 Any mechanism used to provide authenticated denial MUST reveal the - closest enclosing existing domain name for the query. If this is not - provided, the resolver will not be able to ascertain the identity - of an appropriate wild card domain name. - -R6.2 If a zone is signed in such a way that offers authenticated denial - of existence, wild card domain name owned RR sets MUST be signed. - Otherwise the determination of the "closest encloser" is not possible. - -6.2 Role of Wild Cards in Answers - -There are three cases to address. The first is synthesizing from wild card -domain name with data, the second is negatively synthesizing from an -existing wild card, and the third is denying that neither an exact match, -referral, nor wild card exist to answer the query. - -6.2.1 Synthesizing From a Wild Card - -When preparing an answer from a wild card domain name, the answer needs -to include proof that the exact match of the QNAME and QCLASS does not -exist. This is needed because synthesis of the answer replaces the "*" -label with the QNAME without securing the result. The resolver will -realize that the answer was derived from a wild card, but cannot -detect whether an exact match was maliciously omitted. - -R6.3 When synthesizing a positive answer from a wild card domain name, the - answer MUST include proof that the exact match for the QNAME and - QCLASS does not exist. - -Note that a proof that the QTYPE does not exist at the QNAME and QCLASS is -not sufficient to justify synthesis from a wild card. - -6.2.2. Synthesizing Authoritative No Error, No Data From a Wild Card - -When synthesizing a negative answer that is derived from a wild card, -meaning that a wild card matched the QNAME (no exact match happened for -QNAME) but that there is no match for QTYPE there, at most two negative -answers are needed, possibly one. As in 6.2.1, a proof that the exact -match failed is needed. A second proof is needed to show that the wild -card domain name does not have the QTYPE. Depending on the method of -authenticated denial, these this could be possible with one statement. - -R6.4 When synthesizing a negative answer from a wild card domain name, the - answer MUST include proof that the exact match of the QNAME and - QCLASS does not exist and that the QTYPE matches no RR set at the - wild card. If this answer can be optimized, an implementation - SHOULD reduce the number of records included in the response. - -6.2.3. Answering With an Authoritative Name Error - -When answering with a result code of a name error, the answer needs to -provide proof that neither the exact match for QNAME and QCLASS exists -nor that a wild card domain name exists as a subdomain of the closest -enclosing domain name. - -R6.5 When preparing a reply with an authoritative name error, the answer - MUST include proof that the exact match for the QNAME and QCLASS - does not exist and that no wild card is available to provide a match. - -6.2.4. The Remaining Case (Authoritative No Error, No Data at QNAME) - -When answering negatively because there is a match for QNAME and QCLASS -but no match for the QTYPE, only a proof for that is needed. Just as -the search does not proceed onto a search for the wild card in this -case, neither does the construction of the negative answer proof. - -R6.6 When preparing a reply in which there is an exact match of the - QNAME and QCLASS, but there is no RR set matching the QTYPE, - the reply SHOULD NOT contain any proof regarding the wild card - domain name. - -6.3 Interpreting Negative Answers Involving Wild Cards - -There are three requirements for resolvers when it comes to handling -negative answers generated as described in section 6.2. - -R6.7 A resolver MUST confirm that the negative data relates to the - query submitted. - -It is incumbent upon the resolver to interpret the answer correctly. - -R6.8 A resolver MUST confirm that an answer synthesized from a wild - card domain name is done so only in an authoritative absence of - a domain name with the query name and query class. - -In the case of a wild card synthesized answer, the resolver has to -see that the query name and class has no node, proving that a synthesized -answer would be appropriate (subject to validation of it). - -R6.9 A resolver MUST confirm that an authoritative name error is - valid if there is proof that both domain name matching the query - name and class and if there is proof that the closest encloser - does not have a wild card domain name as an immediate descendent. - -Before concluding that an authoritative name error is justified, a -resolver has to determine that neither an exact match for the query -name and class exists nor an appropriate wild card domain name. - -6.4 Authenticated Denial, Wild Card Domain Names, and Opt-In - -When considering the Opt-In proposal [WIP], it is wise to not combine -a zone that adheres to both opt-in and that has a wild card domain -name. The reason is rooted in that the synthesis of an answer is done -by substituting the QNAME for the wild card domain name in the answer. -Because this is unsecured, and the is ambiguity regarding whether a -negative proof can be provided for the exact match (when it is outside -the opt-in secured area), a definitive proof of authenticated denial -is not possible. - -For a more complete discussion of this topic, please refer to the document -describing the Opt-In proposal, referenced above. - -7 Analytical Proof That NXT Names the Closest Encloser - -How does one know, and (more importantly) *prove* using NXT records, what -the closest encloser of a given QNAME is? This section answers that -question with a rigorous proof, because security is the topic. - -7.1 Background to the Proof - -We'd like to have empty non-terminals provably exist in secure zones. -In other words, if someone has: - - a.b.c 3600 IN A 10.0.0.1 - -in their zone, but does not have any records with owner names "c" or -"b.c", we'd like to be able to say (with proof) that "nodes 'c' and -'b.c' exist and yet have no RRs." - -We want this because it is the behavior mandated by the nameserver -algorithm in section 4.3.2 of RFC 1034, and because it is regarded by -most as a better, more "natural" behavior than the alternative of -treating such empty non-terminals as being non-existent. - -There are two ways to achieve this. One way is to instantiate all -the implied empty non-terminals, and then add NXT and SIG(NXT) to them. -This works, but is a burden to the server in storage and computation -resources. It especially complicates updates, since any deletion of -the last record at a node necessitates a computation to determine -which empty non-terminals are no longer relevant and thus must also be -deleted. - -The second way is to infer the existence of the empty non-terminals -from the names of the nodes with real data (i.e. the names in the NXT -chain). - -Using this technique, the "deepest existing ancestor" a.k.a. the "most -enclosing name" of any query name Q can be easily found, and proved to -exist. This allows great efficiency in the wild card matching -algorithm as well, since only one wild card possibility exists and must -subsequently be either proven to exist or proven not to exist. This -is a big improvement on the "empty non-terminals do not exist" -approach, which has many more possible candidate wild card names which -must be proven not to exist. - -7.2 Definitions and Preliminaries - -When we say "subdomain" anywhere below, we mean "is contained within the -domain (in the sense that RFC 1034 describes), or is equal to the domain". -I.e., we're treating it like "subset" in mathematics. - -X is a "superdomain" of Y iff. Y is a subdomain of X. - -A name is an "owner name in zone Z" if it is an owner name, is a subdomain -of the origin of zone Z, and is not glue (or otherwise beneath a zone cut -of zone Z). - -A name N is "directly in zone Z" iff. there is some owner name in Z equal -to N. - -A name N is "inferred to be in zone Z", if it is not directly in zone Z, -but is a superdomain of some direct name of Z and is still a subdomain of -Z. I.e., it is an "empty non-terminal" required to make the path from the -zone origin to some name directly in Z. - -A name is "in zone Z" if it is directly in zone Z, or is inferred to be in -zone Z. - -Let "<" denote the DNSSEC name order relation. - -The "greatest common superdomain" of names A and B, denoted GCS(A,B), is -the greatest (according to the DNSSEC ordering) name X such that X is a -superdomain of both A and B. I.e. it is the "deepest common ancestor" of -A and B. GCS(A,B) always exists, because the root name is a superdomain -of all names. - -Let Q be a name which is a subdomain of the origin of zone Z. - -7.3 Bounds of Q in Z - -There is always a name directly in Z, call it "GLB(Q,Z)", which is the -greatest lower bound of Q. I.e. GLB(Q,Z) <= Q, and for all N in Z where -N <= Q, N <= GLB(Q,Z). - -There may or may not be a name directly in Z, call it "LUB(Q,Z)", which is -the least upper bound of Q. If there is no N directly in Z such that -N >= Q, then there is no LUB(Q,Z). If there is some N directly in Z where -N >= Q, then there is an LUB(Q,Z) >= Q such that if N >= Q, then -LUB(Q,Z) <= N. - -So, GLB(Q,Z) <= Q < LUB(Q,Z), if the least upper bound exists. - -GLB(Q,Z) will have a NXT record which: - - If GLB(Q,Z) = Q, proves that Q is directly in Z - - If GLB(Q,Z) != Q, proves that Q is not directly in Z - -The "next domain name" field of this NXT record is the LUB, unless it is -the zone origin (the DNSSEC "end of chain" marker) and Q != the origin of -Z, in which case there is no LUB. - -THEOREM 1: Let A, B, and Q be subdomains of Z. Let A <= B and B <= Q. Then - - GCS(Q, A) <= GCS(Q, B) - -Proof: - -Assume GCS(Q, A) > GCS(Q, B). Then A must have more labels in common with -Q than B, but since A and B are less than Q, that means that A > B by the -DNSSEC ordering, which is a contradiction since A <= B. - -THEOREM 2: Let A, B, and Q be subdomains of Z. Let A >= B and B >= Q. Then - - GCS(Q, A) <= GCS(Q, B) - -Proof: - -Assume GCS(Q, A) > GCS(Q, B). Then A must have more labels in common with -Q than B, but since A and B are greater than Q, that means that A < B by -the DNSSEC ordering, which is a contradiction since A >= B. - -7.4 Greatest Ancestor of Q in Z - -The "greatest ancestor of Q in Z", denoted GA(Q,Z), is the greatest N in Z, -directly or inferred, such that Q is a subdomain of N. GA(Q,Z) is also -called the "most enclosing name of Q in Z" or the "deepest ancestor of -Q in Z". - -GA(Q,Z) always exists. Since Q is a subdomain of the origin of Z, and the -origin of Z is "directly in zone Z", so there's always at least one N in Z -such that Q is a subdomain of N. - -THEOREM 3: Let Q be a subdomain of the origin of zone Z. If LUB(Q,Z) -exists, then: - - GA(Q,Z) = the greater of GCS(Q, GLB(Q,Z)) and GCS(Q, LUB(Q,Z)) +is copied, but the search still ceases ("Go to step 6."). In the following +section, a suggested change is made to this, under the heading "CNAME RRs +at a Wild Card Domain Name." -otherwise +6 Considerations with Special Types - GA(Q,Z) = GCS(Q, GLB(Q,Z)) +For the purposes of this section, "special" means that a record induces +processing at the server beyond simple lookup. The special types in this +section are SOA, NS, CNAME, and DNAME. SOA is special because it is used +as a zone marker and has an impact on step 2 of the algorithm in 4.3.2. +NS denotes a cut point and has an impact on step 3b. CNAME redirects the +query and is mentioned in steps 3a and 3b. DNAME is a "CNAME generator." -Proof: +6.1 SOA RR's at a Wild Card Domain Name -We can eliminate the trivial case where Q is directly in Z, since in that -case GA(Q,Z) is obviously Q. +If the owner of an SOA record conforms to the basic rules of owning an +SOA RR (meaning it is the apex of a zone) the impact on the search algorithm +is not in section 3c (where records are synthesized) as would be expected. +The impact is really in step 2 of the algorithm, the choice of zone. -For notational convenience, let +We are no longer talking about whether or not an SOA RR can be synthesized +in a response because we are shifting attention to step 2. We are now talking +about what it means for a name server to synthesize a zone for a response. +To date, no implementation has done this. Thinking ahead though, anyone +choosing to pursue this would have to be aware that a server would have to +be able to distinguish between queries for data it will have to synthesize and +queries that ought to be treated as if they were prompted by a lame delegation. - L = GCS(Q, GLB(Q,Z)) - U = GCS(Q, LUB(Q,Z)) +It is not a protocol error to have an SOA RR owned by a wild card domain name, +just as it is not an error to have zone name be syntactically equivalent to a +domain name. However, this situation requires careful consideration of how +a server chooses the appropriate zone for an answer. And an SOA RR is +not able to be synthesized as in step 3c. -Assume L and U both exist. Assume there is an M in Z that is greater than -both L and U, and is a superdomain of Q. +6.2 NS RR's at a Wild Card Domain Name -If M is directly in Z, then M > GLB(Q,Z). This is because if M were -<= GLB(Q,Z), then GCS(Q,M) would be <= L by Theorem 1. If M is directly -in Z, it cannot be >= Q since it is a superdomain of Q and M != Q. So, -we have GLB(Q,Z) < M < Q, which implies that GLB(Q,Z) is not the greatest -lower bound, which is a contradiction. +Complimentary to the issue of an SOA RR owned by a wild card domain name is +the issue of NS RR's owned by a wild card domain name. In this instance, +each machine being referred to in the RDATA of the NS RR has to be able to +understand the impact of this on step 2, the choosing of the authoritative +zone. -If M is inferred to be in Z, then there is some N directly in Z and M is a -superdomain of N. Either N < Q or N > Q (since Q is not directly in Z). +Referring to the same machine in such a NS RR will probably not work well. +This is because the server may become confused as to whether the query name +ought to be answered by the zone owning the NS RR in question or a synthesized +zone. (It isn't known in advance that the query name will invoke the wild +card synthesis.) -If N < Q, then N > GLB(Q,Z). If N were <= GLB(Q,Z), then the GCS(Q,N) -would be <= L by Theorem 1, but GCS(Q,N) = M, and M > L. We thus have a -contradiction, since this implies that GLB(Q,Z) is not the greatest lower -bound. +The status of other RR's owned by a wild card domain name is the same as +if the owner name was not a wild card domain name. I.e., when there is a +NS RR at a wild card domain name, other records are treated as being below +the zone cut. -If N > Q, then N < LUB(Q,Z). If N were >= LUB(Q,Z), then the GCS(Q,N) -would be <= U by Theorem 2, but GCS(Q,N) = M, and M > U. We thus have a -contradiction, since this implies that LUB(Q,Z) is not the least upper bound. +Is it not a protocol error to have a NS RR owned by a wild card domian name, +complimentary to the case of a SOA RR. However, for this to work, an +implementation has to know how to synthesize a zone. -Now we deal with the case where U doesn't exist. Again, assume M in Z that -is greater than L, and is a superdomain of Q. +6.3 CNAME RR's at a Wild Card Domain Name -The cases where M is directly in Z, or where M is inferred and N < Q are as -above. Now we deal with the case where N > Q. First we note that since < -is a well-ordering of the names in Z, if there are any upper bounds to Q in -Z, then there must be a least upper bound. Now, if N existed, it would be -an upper bound of Q in Z, and hence a least upper bound would have to exist, -but there is no least upper bound of Q in Z by assumption, so we again have -a contradiction. +The issue of CNAME RR's owned by wild card domain names has prompted a +suggested change to the last paragraph of step 3c of the algorithm in 4.3.2. +The changed text is this: -Q.E.D. + If the "*" label does exist and if the data at the node is + a CNAME and QTYPE doesn't match CNAME, copy the CNAME RR + into the answer section of the response, set the owner + of the CNAME RR to be QNAME, and then change QNAME to the + canonical name in the CNAME RR, and go back to step 1. -7.5 Conclusion of the Proof + If the "*" label does exist and either QTYPE is CNAME or the + data at the node is not a CNAME, then match RRs at that node + against QTYPE. If any match, copy them into the answer + section, but set the owner of the RR to be QNAME, and + not the node with the "*" label. Go to step 6. -We've shown how to find the "closest encloser" of any given QNAME by looking -at the QNAME along with the owner name and "next domain name" field of the -NXT record which proves the QNAME doesn't exist. The technique works even -when the closest encloser is an inferred name. +Apologies if the above isn't clear, but an attempt was made to stitch +together the passage using just the phrases in section 3a and 3c of the +algorithm so as to preserve the original flavor. -Knowing the closest encloser lets us do very simple wild card checking in -secure zones, since the only possible matching wild card is +In case the passage as suggested isn't clear enough, the intent is to make +"landing" at a wild card name and finding a CNAME the same as if this happened +as a result of a direct match. I.e., Finding a CNAME at the name matched +in step 3c is supposed to have the same impact as finding the CNAME in step +3a. - *. +6.4 DNAME RR's at a Wild Card Domain Name -We simply lookup that name, and if found, proceed accordingly. If not, we -add the NXT record which proves it doesn't exist to the authority section. +The specification of the DNAME RR, which is at the proposed level of +standardization, is not as mature as the full standard in RFC 1034. Because +of this, or the reason for this is, there appears to be a host of issues +with that definition and it's rewrite of the algorithm in 4.3.2. For the +time being, when it comes to wild card processing issues, a DNAME can +be considered to be a CNAME synthesizer. A DNAME at a wild card domain name +is effectively the same as a CNAME at a wild card domain name. -8 Security Considerations +7 Security Considerations This document is refining the specifications to make it more likely that security can be added to DNS. No functional additions are being made, just refining what is considered proper to allow the DNS, security of the DNS, and extending the DNS to be more predictable. -9 References +8 References Normative References [RFC 20] ASCII Format for Network Interchange, V.G. Cerf, Oct-16-1969 [RFC 1034] Domain Names - Concepts and Facilities, P.V. Mockapetris, Nov-01-1987 [RFC 1035] Domain Names - Implementation and Specification, P.V Mockapetris, Nov-01-1987 [RFC 2119] Key Words for Use in RFCs to Indicate Requirement Levels, S Bradner, March 1997 -Non-normative References +Informative References [RFC 2136] Dynamic Updates in the Domain Name System (DNS UPDATE), P. Vixie, Ed., S. Thomson, Y. Rekhter, J. Bound, April 1997 [RFC 2535] Domain Name System Security Extensions, D. Eastlake, March 1999 -[WIP] DNSSEC Opt-In, Internet Draft, R. Arends, M. Kosters, D. Blacka, 2002 +[RFC 2672] Non-Terminal DNS Name Redirection, M. Crawford, August 1999 -10 Others Contributing to This Document +9 Others Contributing to This Document Others who have directly caused text to appear in the document: Paul Vixie and Olaf Kolkman. Many others have indirect influences on the content. -11 Editors +10 Editors Name: Bob Halley Affiliation: Nominum, Inc. Address: 2385 Bay Road, Redwood City, CA 94063 USA Phone: +1-650-381-6016 EMail: Bob.Halley@nominum.com Name: Edward Lewis Affiliation: ARIN Address: 3635 Concorde Pkwy, Suite 200, Chantilly, VA 20151 USA