4. NAME SERVERS
Name servers are the repositories of information that make up the domain
database. The database is divided up into sections called zones, which
are distributed among the name servers. While name servers can have
several optional functions and sources of data, the essential task of a
name server is to answer queries using data in its zones. By design,
name servers can answer queries in a simple manner; the response can
always be generated using only local data, and either contains the
answer to the question or a referral to other name servers "closer" to
the desired information.
A given zone will be available from several name servers to insure its
availability in spite of host or communication link failure. By
administrative fiat, we require every zone to be available on at least
two servers, and many zones have more redundancy than that.
A given name server will typically support one or more zones, but this
gives it authoritative information about only a small section of the
domain tree. It may also have some cached non-authoritative data about
other parts of the tree. The name server marks its responses to queries
so that the requester can tell whether the response comes from
authoritative data or not.
4.2. How the database is divided into zones
The domain database is partitioned in two ways: by class, and by "cuts"
made in the name space between nodes.
The class partition is simple. The database for any class is organized,
delegated, and maintained separately from all other classes. Since, by
convention, the name spaces are the same for all classes, the separate
classes can be thought of as an array of parallel namespace trees. Note
that the data attached to nodes will be different for these different
parallel classes. The most common reasons for creating a new class are
the necessity for a new data format for existing types or a desire for a
separately managed version of the existing name space.
Within a class, "cuts" in the name space can be made between any two
adjacent nodes. After all cuts are made, each group of connected name
space is a separate zone. The zone is said to be authoritative for all
names in the connected region. Note that the "cuts" in the name space
may be in different places for different classes, the name servers may
be different, etc.
These rules mean that every zone has at least one node, and hence domain
name, for which it is authoritative, and all of the nodes in a
particular zone are connected. Given, the tree structure, every zone
has a highest node which is closer to the root than any other node in
the zone. The name of this node is often used to identify the zone.
It would be possible, though not particularly useful, to partition the
name space so that each domain name was in a separate zone or so that
all nodes were in a single zone. Instead, the database is partitioned
at points where a particular organization wants to take over control of
a subtree. Once an organization controls its own zone it can
unilaterally change the data in the zone, grow new tree sections
connected to the zone, delete existing nodes, or delegate new subzones
under its zone.
If the organization has substructure, it may want to make further
internal partitions to achieve nested delegations of name space control.
In some cases, such divisions are made purely to make database
maintenance more convenient.
4.2.1. Technical considerations
The data that describes a zone has four major parts:
- Authoritative data for all nodes within the zone.
- Data that defines the top node of the zone (can be thought of
as part of the authoritative data).
- Data that describes delegated subzones, i.e., cuts around the
bottom of the zone.
- Data that allows access to name servers for subzones
(sometimes called "glue" data).
All of this data is expressed in the form of RRs, so a zone can be
completely described in terms of a set of RRs. Whole zones can be
transferred between name servers by transferring the RRs, either carried
in a series of messages or by FTPing a master file which is a textual
The authoritative data for a zone is simply all of the RRs attached to
all of the nodes from the top node of the zone down to leaf nodes or
nodes above cuts around the bottom edge of the zone.
Though logically part of the authoritative data, the RRs that describe
the top node of the zone are especially important to the zone's
management. These RRs are of two types: name server RRs that list, one
per RR, all of the servers for the zone, and a single SOA RR that
describes zone management parameters.
The RRs that describe cuts around the bottom of the zone are NS RRs that
name the servers for the subzones. Since the cuts are between nodes,
these RRs are NOT part of the authoritative data of the zone, and should
be exactly the same as the corresponding RRs in the top node of the
subzone. Since name servers are always associated with zone boundaries,
NS RRs are only found at nodes which are the top node of some zone. In
the data that makes up a zone, NS RRs are found at the top node of the
zone (and are authoritative) and at cuts around the bottom of the zone
(where they are not authoritative), but never in between.
One of the goals of the zone structure is that any zone have all the
data required to set up communications with the name servers for any
subzones. That is, parent zones have all the information needed to
access servers for their children zones. The NS RRs that name the
servers for subzones are often not enough for this task since they name
the servers, but do not give their addresses. In particular, if the
name of the name server is itself in the subzone, we could be faced with
the situation where the NS RRs tell us that in order to learn a name
server's address, we should contact the server using the address we wish
to learn. To fix this problem, a zone contains "glue" RRs which are not
part of the authoritative data, and are address RRs for the servers.
These RRs are only necessary if the name server's name is "below" the
cut, and are only used as part of a referral response.
4.2.2. Administrative considerations
When some organization wants to control its own domain, the first step
is to identify the proper parent zone, and get the parent zone's owners
to agree to the delegation of control. While there are no particular
technical constraints dealing with where in the tree this can be done,
there are some administrative groupings discussed in [RFC-1032] which
deal with top level organization, and middle level zones are free to
create their own rules. For example, one university might choose to use
a single zone, while another might choose to organize by subzones
dedicated to individual departments or schools. [RFC-1033] catalogs
available DNS software an discusses administration procedures.
Once the proper name for the new subzone is selected, the new owners
should be required to demonstrate redundant name server support. Note
that there is no requirement that the servers for a zone reside in a
host which has a name in that domain. In many cases, a zone will be
more accessible to the internet at large if its servers are widely
distributed rather than being within the physical facilities controlled
by the same organization that manages the zone. For example, in the
current DNS, one of the name servers for the United Kingdom, or UK
domain, is found in the US. This allows US hosts to get UK data without
using limited transatlantic bandwidth.
As the last installation step, the delegation NS RRs and glue RRs
necessary to make the delegation effective should be added to the parent
zone. The administrators of both zones should insure that the NS and
glue RRs which mark both sides of the cut are consistent and remain so.
4.3. Name server internals
4.3.1. Queries and responses
The principal activity of name servers is to answer standard queries.
Both the query and its response are carried in a standard message format
which is described in [RFC-1035]. The query contains a QTYPE, QCLASS,
and QNAME, which describe the types and classes of desired information
and the name of interest.
The way that the name server answers the query depends upon whether it
is operating in recursive mode or not:
- The simplest mode for the server is non-recursive, since it
can answer queries using only local information: the response
contains an error, the answer, or a referral to some other
server "closer" to the answer. All name servers must
implement non-recursive queries.
- The simplest mode for the client is recursive, since in this
mode the name server acts in the role of a resolver and
returns either an error or the answer, but never referrals.
This service is optional in a name server, and the name server
may also choose to restrict the clients which can use
Recursive service is helpful in several situations:
- a relatively simple requester that lacks the ability to use
anything other than a direct answer to the question.
- a request that needs to cross protocol or other boundaries and
can be sent to a server which can act as intermediary.
- a network where we want to concentrate the cache rather than
having a separate cache for each client.
Non-recursive service is appropriate if the requester is capable of
pursuing referrals and interested in information which will aid future
The use of recursive mode is limited to cases where both the client and
the name server agree to its use. The agreement is negotiated through
the use of two bits in query and response messages:
- The recursion available, or RA bit, is set or cleared by a
name server in all responses. The bit is true if the name
server is willing to provide recursive service for the client,
regardless of whether the client requested recursive service.
That is, RA signals availability rather than use.
- Queries contain a bit called recursion desired or RD. This
bit specifies specifies whether the requester wants recursive
service for this query. Clients may request recursive service
from any name server, though they should depend upon receiving
it only from servers which have previously sent an RA, or
servers which have agreed to provide service through private
agreement or some other means outside of the DNS protocol.
The recursive mode occurs when a query with RD set arrives at a server
which is willing to provide recursive service; the client can verify
that recursive mode was used by checking that both RA and RD are set in
the reply. Note that the name server should never perform recursive
service unless asked via RD, since this interferes with trouble shooting
of name servers and their databases.
If recursive service is requested and available, the recursive response
to a query will be one of the following:
- The answer to the query, possibly preface by one or more CNAME
RRs that specify aliases encountered on the way to an answer.
- A name error indicating that the name does not exist. This
may include CNAME RRs that indicate that the original query
name was an alias for a name which does not exist.
- A temporary error indication.
If recursive service is not requested or is not available, the non-
recursive response will be one of the following:
- An authoritative name error indicating that the name does not
- A temporary error indication.
- Some combination of:
RRs that answer the question, together with an indication
whether the data comes from a zone or is cached.
A referral to name servers which have zones which are closer
ancestors to the name than the server sending the reply.
- RRs that the name server thinks will prove useful to the
The actual algorithm used by the name server will depend on the local OS
and data structures used to store RRs. The following algorithm assumes
that the RRs are organized in several tree structures, one for each
zone, and another for the cache:
1. Set or clear the value of recursion available in the response
depending on whether the name server is willing to provide
recursive service. If recursive service is available and
requested via the RD bit in the query, go to step 5,
otherwise step 2.
2. Search the available zones for the zone which is the nearest
ancestor to QNAME. If such a zone is found, go to step 3,
otherwise step 4.
3. Start matching down, label by label, in the zone. The
matching process can terminate several ways:
a. If the whole of QNAME is matched, we have found the
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, change QNAME to the canonical name in
the CNAME RR, and go back to step 1.
Otherwise, copy all RRs which match QTYPE into the
answer section and go to step 6.
b. If a match would take us out of the authoritative data,
we have a referral. This happens when we encounter a
node with NS RRs marking cuts along the bottom of a
Copy the NS RRs for the subzone into the authority
section of the reply. Put whatever addresses are
available into the additional section, using glue RRs
if the addresses are not available from authoritative
data or the cache. Go to step 4.
c. If at some label, a match is impossible (i.e., the
corresponding label does not exist), look to see if a
the "*" label exists.
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.
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.
4. Start matching down in the cache. If QNAME is found in the
cache, copy all RRs attached to it that match QTYPE into the
answer section. If there was no delegation from
authoritative data, look for the best one from the cache, and
put it in the authority section. Go to step 6.
5. Using the local resolver or a copy of its algorithm (see
resolver section of this memo) to answer the query. Store
the results, including any intermediate CNAMEs, in the answer
section of the response.
6. Using local data only, attempt to add other RRs which may be
useful to the additional section of the query. Exit.
In the previous algorithm, special treatment was given to RRs with owner
names starting with the label "*". Such RRs are called wildcards.
Wildcard RRs can be thought of as instructions for synthesizing RRs.
When the appropriate conditions are met, the name server creates RRs
with an owner name equal to the query name and contents taken from the
This facility is most often used to create a zone which will be used to
forward mail from the Internet to some other mail system. The general
idea is that any name in that zone which is presented to server in a
query will be assumed to exist, with certain properties, unless explicit
evidence exists to the contrary. Note that the use of the term zone
here, instead of domain, is intentional; such defaults do not propagate
across zone boundaries, although a subzone may choose to achieve that
appearance by setting up similar defaults.
The contents of the wildcard RRs follows the usual rules and formats for
RRs. The wildcards in the zone have an owner name that controls the
query names they will match. The owner name of the wildcard RRs is of
the form "*.<anydomain>", where <anydomain> is any domain name.
<anydomain> should not contain other * labels, and should be in the
authoritative data of the zone. The wildcards potentially apply to
descendants of <anydomain>, but not to <anydomain> itself. Another way
to look at this is that the "*" label always matches at least one whole
label and sometimes more, but always whole labels.
Wildcard RRs do not apply:
- When the query is in another zone. That is, delegation cancels
the wildcard defaults.
- When the query name or a name between the wildcard domain and
the query name is know to exist. For example, if a wildcard
RR has an owner name of "*.X", and the zone also contains RRs
attached to B.X, the wildcards would apply to queries for name
Z.X (presuming there is no explicit information for Z.X), but
not to B.X, A.B.X, or X.
A * label appearing in a query name has no special effect, but can be
used to test for wildcards in an authoritative zone; such a query is the
only way to get a response containing RRs with an owner name with * in
it. The result of such a query should not be cached.
Note that the contents of the wildcard RRs are not modified when used to
To illustrate the use of wildcard RRs, suppose a large company with a
large, non-IP/TCP, network wanted to create a mail gateway. If the
company was called X.COM, and IP/TCP capable gateway machine was called
A.X.COM, the following RRs might be entered into the COM zone:
X.COM MX 10 A.X.COM
*.X.COM MX 10 A.X.COM
A.X.COM A 18.104.22.168
A.X.COM MX 10 A.X.COM
*.A.X.COM MX 10 A.X.COM
This would cause any MX query for any domain name ending in X.COM to
return an MX RR pointing at A.X.COM. Two wildcard RRs are required
since the effect of the wildcard at *.X.COM is inhibited in the A.X.COM
subtree by the explicit data for A.X.COM. Note also that the explicit
MX data at X.COM and A.X.COM is required, and that none of the RRs above
would match a query name of XX.COM.
4.3.4. Negative response caching (Optional)
The DNS provides an optional service which allows name servers to
distribute, and resolvers to cache, negative results with TTLs. For
example, a name server can distribute a TTL along with a name error
indication, and a resolver receiving such information is allowed to
assume that the name does not exist during the TTL period without
consulting authoritative data. Similarly, a resolver can make a query
with a QTYPE which matches multiple types, and cache the fact that some
of the types are not present.
This feature can be particularly important in a system which implements
naming shorthands that use search lists beacuse a popular shorthand,
which happens to require a suffix toward the end of the search list,
will generate multiple name errors whenever it is used.
The method is that a name server may add an SOA RR to the additional
section of a response when that response is authoritative. The SOA must
be that of the zone which was the source of the authoritative data in
the answer section, or name error if applicable. The MINIMUM field of
the SOA controls the length of time that the negative result may be
Note that in some circumstances, the answer section may contain multiple
owner names. In this case, the SOA mechanism should only be used for
the data which matches QNAME, which is the only authoritative data in
Name servers and resolvers should never attempt to add SOAs to the
additional section of a non-authoritative response, or attempt to infer
results which are not directly stated in an authoritative response.
There are several reasons for this, including: cached information isn't
usually enough to match up RRs and their zone names, SOA RRs may be
cached due to direct SOA queries, and name servers are not required to
output the SOAs in the authority section.
This feature is optional, although a refined version is expected to
become part of the standard protocol in the future. Name servers are
not required to add the SOA RRs in all authoritative responses, nor are
resolvers required to cache negative results. Both are recommended.
All resolvers and recursive name servers are required to at least be
able to ignore the SOA RR when it is present in a response.
Some experiments have also been proposed which will use this feature.
The idea is that if cached data is known to come from a particular zone,
and if an authoritative copy of the zone's SOA is obtained, and if the
zone's SERIAL has not changed since the data was cached, then the TTL of
the cached data can be reset to the zone MINIMUM value if it is smaller.
This usage is mentioned for planning purposes only, and is not
recommended as yet.
4.3.5. Zone maintenance and transfers
Part of the job of a zone administrator is to maintain the zones at all
of the name servers which are authoritative for the zone. When the
inevitable changes are made, they must be distributed to all of the name
servers. While this distribution can be accomplished using FTP or some
other ad hoc procedure, the preferred method is the zone transfer part
of the DNS protocol.
The general model of automatic zone transfer or refreshing is that one
of the name servers is the master or primary for the zone. Changes are
coordinated at the primary, typically by editing a master file for the
zone. After editing, the administrator signals the master server to
load the new zone. The other non-master or secondary servers for the
zone periodically check for changes (at a selectable interval) and
obtain new zone copies when changes have been made.
To detect changes, secondaries just check the SERIAL field of the SOA
for the zone. In addition to whatever other changes are made, the
SERIAL field in the SOA of the zone is always advanced whenever any
change is made to the zone. The advancing can be a simple increment, or
could be based on the write date and time of the master file, etc. The
purpose is to make it possible to determine which of two copies of a
zone is more recent by comparing serial numbers. Serial number advances
and comparisons use sequence space arithmetic, so there is a theoretic
limit on how fast a zone can be updated, basically that old copies must
die out before the serial number covers half of its 32 bit range. In
practice, the only concern is that the compare operation deals properly
with comparisons around the boundary between the most positive and most
negative 32 bit numbers.
The periodic polling of the secondary servers is controlled by
parameters in the SOA RR for the zone, which set the minimum acceptable
polling intervals. The parameters are called REFRESH, RETRY, and
EXPIRE. Whenever a new zone is loaded in a secondary, the secondary
waits REFRESH seconds before checking with the primary for a new serial.
If this check cannot be completed, new checks are started every RETRY
seconds. The check is a simple query to the primary for the SOA RR of
the zone. If the serial field in the secondary's zone copy is equal to
the serial returned by the primary, then no changes have occurred, and
the REFRESH interval wait is restarted. If the secondary finds it
impossible to perform a serial check for the EXPIRE interval, it must
assume that its copy of the zone is obsolete an discard it.
When the poll shows that the zone has changed, then the secondary server
must request a zone transfer via an AXFR request for the zone. The AXFR
may cause an error, such as refused, but normally is answered by a
sequence of response messages. The first and last messages must contain
the data for the top authoritative node of the zone. Intermediate
messages carry all of the other RRs from the zone, including both
authoritative and non-authoritative RRs. The stream of messages allows
the secondary to construct a copy of the zone. Because accuracy is
essential, TCP or some other reliable protocol must be used for AXFR
Each secondary server is required to perform the following operations
against the master, but may also optionally perform these operations
against other secondary servers. This strategy can improve the transfer
process when the primary is unavailable due to host downtime or network
problems, or when a secondary server has better network access to an
"intermediate" secondary than to the primary.
Resolvers are programs that interface user programs to domain name
servers. In the simplest case, a resolver receives a request from a
user program (e.g., mail programs, TELNET, FTP) in the form of a
subroutine call, system call etc., and returns the desired information
in a form compatible with the local host's data formats.
The resolver is located on the same machine as the program that requests
the resolver's services, but it may need to consult name servers on
other hosts. Because a resolver may need to consult several name
servers, or may have the requested information in a local cache, the
amount of time that a resolver will take to complete can vary quite a
bit, from milliseconds to several seconds.
A very important goal of the resolver is to eliminate network delay and
name server load from most requests by answering them from its cache of
prior results. It follows that caches which are shared by multiple
processes, users, machines, etc., are more efficient than non-shared
5.2. Client-resolver interface
5.2.1. Typical functions
The client interface to the resolver is influenced by the local host's
conventions, but the typical resolver-client interface has three
1. Host name to host address translation.
This function is often defined to mimic a previous HOSTS.TXT
based function. Given a character string, the caller wants
one or more 32 bit IP addresses. Under the DNS, it
translates into a request for type A RRs. Since the DNS does
not preserve the order of RRs, this function may choose to
sort the returned addresses or select the "best" address if
the service returns only one choice to the client. Note that
a multiple address return is recommended, but a single
address may be the only way to emulate prior HOSTS.TXT
2. Host address to host name translation
This function will often follow the form of previous
functions. Given a 32 bit IP address, the caller wants a
character string. The octets of the IP address are reversed,
used as name components, and suffixed with "IN-ADDR.ARPA". A
type PTR query is used to get the RR with the primary name of
the host. For example, a request for the host name
corresponding to IP address 22.214.171.124 looks for PTR RRs for
domain name "126.96.36.199.IN-ADDR.ARPA".
3. General lookup function
This function retrieves arbitrary information from the DNS,
and has no counterpart in previous systems. The caller
supplies a QNAME, QTYPE, and QCLASS, and wants all of the
matching RRs. This function will often use the DNS format
for all RR data instead of the local host's, and returns all
RR content (e.g., TTL) instead of a processed form with local
When the resolver performs the indicated function, it usually has one of
the following results to pass back to the client:
- One or more RRs giving the requested data.
In this case the resolver returns the answer in the
- A name error (NE).
This happens when the referenced name does not exist. For
example, a user may have mistyped a host name.
- A data not found error.
This happens when the referenced name exists, but data of the
appropriate type does not. For example, a host address
function applied to a mailbox name would return this error
since the name exists, but no address RR is present.
It is important to note that the functions for translating between host
names and addresses may combine the "name error" and "data not found"
error conditions into a single type of error return, but the general
function should not. One reason for this is that applications may ask
first for one type of information about a name followed by a second
request to the same name for some other type of information; if the two
errors are combined, then useless queries may slow the application.
While attempting to resolve a particular request, the resolver may find
that the name in question is an alias. For example, the resolver might
find that the name given for host name to address translation is an
alias when it finds the CNAME RR. If possible, the alias condition
should be signalled back from the resolver to the client.
In most cases a resolver simply restarts the query at the new name when
it encounters a CNAME. However, when performing the general function,
the resolver should not pursue aliases when the CNAME RR matches the
query type. This allows queries which ask whether an alias is present.
For example, if the query type is CNAME, the user is interested in the
CNAME RR itself, and not the RRs at the name it points to.
Several special conditions can occur with aliases. Multiple levels of
aliases should be avoided due to their lack of efficiency, but should
not be signalled as an error. Alias loops and aliases which point to
non-existent names should be caught and an error condition passed back
to the client.
5.2.3. Temporary failures
In a less than perfect world, all resolvers will occasionally be unable
to resolve a particular request. This condition can be caused by a
resolver which becomes separated from the rest of the network due to a
link failure or gateway problem, or less often by coincident failure or
unavailability of all servers for a particular domain.
It is essential that this sort of condition should not be signalled as a
name or data not present error to applications. This sort of behavior
is annoying to humans, and can wreak havoc when mail systems use the
While in some cases it is possible to deal with such a temporary problem
by blocking the request indefinitely, this is usually not a good choice,
particularly when the client is a server process that could move on to
other tasks. The recommended solution is to always have temporary
failure as one of the possible results of a resolver function, even
though this may make emulation of existing HOSTS.TXT functions more
5.3. Resolver internals
Every resolver implementation uses slightly different algorithms, and
typically spends much more logic dealing with errors of various sorts
than typical occurances. This section outlines a recommended basic
strategy for resolver operation, but leaves details to [RFC-1035].
5.3.1. Stub resolvers
One option for implementing a resolver is to move the resolution
function out of the local machine and into a name server which supports
recursive queries. This can provide an easy method of providing domain
service in a PC which lacks the resources to perform the resolver
function, or can centralize the cache for a whole local network or
All that the remaining stub needs is a list of name server addresses
that will perform the recursive requests. This type of resolver
presumably needs the information in a configuration file, since it
probably lacks the sophistication to locate it in the domain database.
The user also needs to verify that the listed servers will perform the
recursive service; a name server is free to refuse to perform recursive
services for any or all clients. The user should consult the local
system administrator to find name servers willing to perform the
This type of service suffers from some drawbacks. Since the recursive
requests may take an arbitrary amount of time to perform, the stub may
have difficulty optimizing retransmission intervals to deal with both
lost UDP packets and dead servers; the name server can be easily
overloaded by too zealous a stub if it interprets retransmissions as new
requests. Use of TCP may be an answer, but TCP may well place burdens
on the host's capabilities which are similar to those of a real
In addition to its own resources, the resolver may also have shared
access to zones maintained by a local name server. This gives the
resolver the advantage of more rapid access, but the resolver must be
careful to never let cached information override zone data. In this
discussion the term "local information" is meant to mean the union of
the cache and such shared zones, with the understanding that
authoritative data is always used in preference to cached data when both
The following resolver algorithm assumes that all functions have been
converted to a general lookup function, and uses the following data
structures to represent the state of a request in progress in the
SNAME the domain name we are searching for.
STYPE the QTYPE of the search request.
SCLASS the QCLASS of the search request.
SLIST a structure which describes the name servers and the
zone which the resolver is currently trying to query.
This structure keeps track of the resolver's current
best guess about which name servers hold the desired
information; it is updated when arriving information
changes the guess. This structure includes the
equivalent of a zone name, the known name servers for
the zone, the known addresses for the name servers, and
history information which can be used to suggest which
server is likely to be the best one to try next. The
zone name equivalent is a match count of the number of
labels from the root down which SNAME has in common with
the zone being queried; this is used as a measure of how
"close" the resolver is to SNAME.
SBELT a "safety belt" structure of the same form as SLIST,
which is initialized from a configuration file, and
lists servers which should be used when the resolver
doesn't have any local information to guide name server
selection. The match count will be -1 to indicate that
no labels are known to match.
CACHE A structure which stores the results from previous
responses. Since resolvers are responsible for
discarding old RRs whose TTL has expired, most
implementations convert the interval specified in
arriving RRs to some sort of absolute time when the RR
is stored in the cache. Instead of counting the TTLs
down individually, the resolver just ignores or discards
old RRs when it runs across them in the course of a
search, or discards them during periodic sweeps to
reclaim the memory consumed by old RRs.
The top level algorithm has four steps:
1. See if the answer is in local information, and if so return
it to the client.
2. Find the best servers to ask.
3. Send them queries until one returns a response.
4. Analyze the response, either:
a. if the response answers the question or contains a name
error, cache the data as well as returning it back to
b. if the response contains a better delegation to other
servers, cache the delegation information, and go to
c. if the response shows a CNAME and that is not the
answer itself, cache the CNAME, change the SNAME to the
canonical name in the CNAME RR and go to step 1.
d. if the response shows a servers failure or other
bizarre contents, delete the server from the SLIST and
go back to step 3.
Step 1 searches the cache for the desired data. If the data is in the
cache, it is assumed to be good enough for normal use. Some resolvers
have an option at the user interface which will force the resolver to
ignore the cached data and consult with an authoritative server. This
is not recommended as the default. If the resolver has direct access to
a name server's zones, it should check to see if the desired data is
present in authoritative form, and if so, use the authoritative data in
preference to cached data.
Step 2 looks for a name server to ask for the required data. The
general strategy is to look for locally-available name server RRs,
starting at SNAME, then the parent domain name of SNAME, the
grandparent, and so on toward the root. Thus if SNAME were
Mockapetris.ISI.EDU, this step would look for NS RRs for
Mockapetris.ISI.EDU, then ISI.EDU, then EDU, and then . (the root).
These NS RRs list the names of hosts for a zone at or above SNAME. Copy
the names into SLIST. Set up their addresses using local data. It may
be the case that the addresses are not available. The resolver has many
choices here; the best is to start parallel resolver processes looking
for the addresses while continuing onward with the addresses which are
available. Obviously, the design choices and options are complicated
and a function of the local host's capabilities. The recommended
priorities for the resolver designer are:
1. Bound the amount of work (packets sent, parallel processes
started) so that a request can't get into an infinite loop or
start off a chain reaction of requests or queries with other
implementations EVEN IF SOMEONE HAS INCORRECTLY CONFIGURED
2. Get back an answer if at all possible.
3. Avoid unnecessary transmissions.
4. Get the answer as quickly as possible.
If the search for NS RRs fails, then the resolver initializes SLIST from
the safety belt SBELT. The basic idea is that when the resolver has no
idea what servers to ask, it should use information from a configuration
file that lists several servers which are expected to be helpful.
Although there are special situations, the usual choice is two of the
root servers and two of the servers for the host's domain. The reason
for two of each is for redundancy. The root servers will provide
eventual access to all of the domain space. The two local servers will
allow the resolver to continue to resolve local names if the local
network becomes isolated from the internet due to gateway or link
In addition to the names and addresses of the servers, the SLIST data
structure can be sorted to use the best servers first, and to insure
that all addresses of all servers are used in a round-robin manner. The
sorting can be a simple function of preferring addresses on the local
network over others, or may involve statistics from past events, such as
previous response times and batting averages.
Step 3 sends out queries until a response is received. The strategy is
to cycle around all of the addresses for all of the servers with a
timeout between each transmission. In practice it is important to use
all addresses of a multihomed host, and too aggressive a retransmission
policy actually slows response when used by multiple resolvers
contending for the same name server and even occasionally for a single
resolver. SLIST typically contains data values to control the timeouts
and keep track of previous transmissions.
Step 4 involves analyzing responses. The resolver should be highly
paranoid in its parsing of responses. It should also check that the
response matches the query it sent using the ID field in the response.
The ideal answer is one from a server authoritative for the query which
either gives the required data or a name error. The data is passed back
to the user and entered in the cache for future use if its TTL is
greater than zero.
If the response shows a delegation, the resolver should check to see
that the delegation is "closer" to the answer than the servers in SLIST
are. This can be done by comparing the match count in SLIST with that
computed from SNAME and the NS RRs in the delegation. If not, the reply
is bogus and should be ignored. If the delegation is valid the NS
delegation RRs and any address RRs for the servers should be cached.
The name servers are entered in the SLIST, and the search is restarted.
If the response contains a CNAME, the search is restarted at the CNAME
unless the response has the data for the canonical name or if the CNAME
is the answer itself.
Details and implementation hints can be found in [RFC-1035].