Provided by: rbldnsd_0.995_i386
rbldnsd - DNS daemon suitable for running DNS-based blocklists
rbldnsd options zone:dataset...
rbldnsd is a small DNS-protocol daemon which is designed to handle
queries to DNS-based IP-listing or NAME-listing services. Such
services are a simple way to share/publish a list of IP addresses or
(domain) names which are "listed" for for some reason, for example in
order to be able to refuse a service to a client which is "listed" in
rbldnsd is not a general-purpose nameserver. It will answer to A and
TXT (and SOA and NS if such RRs are specified) queries, and has limited
ability to answer to some other types of queries.
rbldnsd tries to handle data from two different perspectives: given a
set (or several) of "listed entries" (e.g. IP address ranges or domain
names), it builds and serves a DNS zone. Note the two are not the
same: list of spammer’s IPs is NOT a DNS zone, but may be represented
and used as such, provided that some additional information necessary
to build complete DNS zone (e.g. NS and SOA records, maybe A records
necessary for http to work) is available. In this context, rbldnsd is
very different from other general-purpose nameservers such as BIND or
NSD: rbldnsd operates with datasets (sets of entries - IP addresses or
domain names, logically grouped together), while other general-purpose
nameservers operates with zones. The way how rbldnsd operates may be
somewhat confusing to BIND experts.
For rbldnsd, a building block is a dataset: e.g., set of
insecure/abuseable hosts (IP addresses), set of network ranges that
belongs to various spam operations (IP ranges), domain names that
belong to spammers (RHSBL) and so on. Usually, different kind of
information is placed into separate file, for easy maintenance. From a
number of such datasets, rbldnsd constructs a number of DNS zones as
specified on command line. A single dataset may be used for several
zones, and a single zone may be constructed from several datasets.
rbldnsd will answer queries to DNS zones specified on the command line
as a set of zone specifications. Each zone specification consists of
zone basename, dataset type and a comma-separated list of files that
forms a given dataset: zone:type:file,file,...
Several zones may be specified in command line, so that rbldnsd will
answer queries to any of them. Also, a single zone may be specified
several times with different datasets, so it is possible to form a zone
from a combination of several different dataset. The same dataset may
be reused for several zones too (and in this case, it will be read into
memory only once).
There are several dataset formats available, each is suitable and
optimized (in terms of memory, speed and ease of use) for a specific
task. Available dataset types may be grouped into the following
lists of IP addresses. When a query is done to a zone with such
data, query is interpreted as an IP address in a reverse form
(similar to in-addr.arpa zone). If the address is found in
dataset data, rbldnsd will return A and TXT records specified in
data for that IP. This is a classical IP-based blocklist.
lists of domain names. Similar to list of IP addresses, but
with generic domain names instead of IPs (wildcards allowed).
This type of data may be used to form a blocklist of e.g.
sender domain names.
generic list of various types of records, as an auxilary data to
form a complete nameserver. This format is similar to
bind-style datafiles, but very simplified. One may specify A,
TXT, NS and MX records here.
combined set, different datasets from the list above combined in
the single (set of) source files, for easy maintenance.
acl, or Access Control List. This is a pseudo dataset, that
works by overweriting query results based on the requestor
(peer) IP address.
The following options may be specified:
rbldnsd will change its userid to the specified user, which
defaults to rbldns, and group, which by default is the primary
group of a user. rbldnsd will refuse to run as the root user,
since this is insecure.
rbldnsd will chroot to rootdir if specified. Data files should
be available inside rootdir.
rbldnsd will change its working directory to workdir (after
chrooting to rootdir if -r option is also specified). May be
used to shorten filename paths.
This option is required. rbldnsd will bind to specified address
and port (port defaults to port 53, domain). Either numeric IP
address or a hostname may be specified, and either port number
or service name is accepted. It is possible to specify several
addresses to listen on this way, by repeating -b option.
Additionally, if there are several addresses listed for a
hostname, rbldnsd will listen on all of them. Note that rbldnsd
will work slightly faster if only one listening address is
specified. Note the delimiter between host and port is a slash
(/), not a colon, to be able to correctly handle IPv6 addresses.
-4 Use IPv4 listening socket/transport, do not attempt to use IPv6
(ignored if rbldnsd was built without IPv6 support).
-6 Use IPv6 listening socket/transport, do not attempt to use IPv4
(this option will be reported as error if IPv6 support was not
Set default reply time-to-live (TTL) value to be defttl, and set
constraints for TTL to minttl and maxttl. Default applies when
there’s no TTL defined in a given scope (in data file), and
constraints are applied when such value provided in data. Any
of the values may be omitted, including trailing colon (:)
characters, e.g. "-t 30" set default TTL to be 30 secound, and
"-t ::120" or "-t ::2m" sets maximum allowed TTL to 2 minutes.
All 3 values are in time units, with optional suffix: s
(secounds, default), m (minutes), h (hours), d (days) or w
(weeks). Zero minttl or maxttl means no corresponding
constraint will be enforced. Default defttl is 35m.
Set interval between checking for zone file changes to be check,
default is 1m (one minute). rbldnsd will check zone file’s last
modification time every so often, and if it detects a change,
zone will be automatically reloaded. Setting this value to 0
disables automatic zone change detection. This procedure may
also be triggered by sending a SIGHUP signal to rbldnsd (see
SIGNALS section below).
-e Allow non-network addresses to be used in CIDR ranges.
Normally, rbldnsd rejects addresses such as 127.2.3.4/24, where
prefix is not within the network mask derived from bit number
(in this case, correct form is 127.2.3.0/24, note the trailing
zero in prefix). This is done in order to catch possible typos
in zones (rbldnsd will warn about a problem and will ignore such
entry). This option disables checking whether the CIDR prefix
fits within the network mask.
-q Quick and quiet start. Normally, rbldnsd does socket
initialization and zone load in foreground, writing progress and
statistic to standard output, and aborts in case of any errors.
With this flag, nothing will be printed and first zone load will
be done in background (unless -n option is also given), and zone
load errors will be non-fatal.
Write rbldnsd’s pid to the specified pidfile, so it will be
easily findable. This file gets written before entering a
chroot jail (if specified) and before changing userid, so it’s
ok to specify e.g. /var/run/rbldnsd.pid here.
Specifies a file to which log all requests made. This file is
created after entering a chroot jail and becoming a user.
Logfiles may be quite large, esp. on busy sites (rbldnsd will
log every recognized request if this option is specified). This
option is mainly intended for debugging purposes. Upon
receiption of SIGHUP signal, rbldnsd reopens its logfile. If
logfile prefixed with a plus sign (+), logging will not be
buffered (i.e. each line will be flushed to disk); by default,
logging is buffered to reduce system load. Specify a single
hyphen (-) as a filename to log to standard output
(filedescriptor 1), either buffered by default, or line-buffered
if specified as ‘+-’ (standard output will not be "reopened"
upon receiving SIGHUP signal, but will be flushed in case
logging is buffered).
Specifies a file where rbldnsd will write a line with short
statistic summary of queries made per zone, every check (-c)
interval. Format of each line is:
timestamp zone:qtot:qok:qnxd:bin:bout zone:...
where timestamp is unix time (secounds since epoch), zone is the
name of the base zone, qtot is the total number of queries
received, qok - number of positive replies, qnxd - number of
NXDOMAIN replies, bin is the total number of bytes read from
network (excluding IP/UDP overhead and dropped packets), bout is
the total number of bytes written to network. Ther are as many
such tuples as there are zones, and one extra, total typle at
the end, with zone being "*", like:
1234 bl1.ex:10:5:4:311:432 bl2.ex:24:13:7:248:375 *:98:35:12:820:987
Note the total values may be larger than the sum of per-zone
values, due to queries made against unlisted zones, or
Rbldnsd will write bare timestamp to statsfile when it is
starting up, shutting down or when statistic counters are being
reset after receiving SIGUSR2 signal (see below), to indicate
the points where the counters are starting back from zero.
By default, rbldnsd writes absolute counter values into
statsfile (number of packets (bytes) since startup or last
reset). statsfile may be prefixed with plus sign (+), in which
case rbldnsd will write delta values, that is, number of packets
or bytes since last write, or number of packets (bytes) per unit
of time ("incremental" mode, hence the "+" sign).
-n Do not become a daemon. Normally, rbldnsd will fork and go to
the background after successful initialization. This option
disables this behaviour.
-f Request rbldnsd to continue processing requests during data
reloads. Rbldnsd forks a child process to handle requests while
parent reloads the data. This ensures smooth operations, but
requires more memory, since two copies of data is keept in
memory during reload process.
-d Dump all zones to stdout in BIND format and exit. This may be
suitable to convert easily editable rbldnsd-style data into BIND
zone. rbldnsd dumps all zones as one stream, so one may want to
specify only one zone with -d. Zone file will have appropriate
$ORIGIN tags. Note that data generated may be really huge (as
BIND format isn’t appropriate for this sort of things), and some
entries may not be really the same in BIND as in rbldnsd (e.g.,
IP netblocks of large size will be represented as wildcard
entries - 10.0.0.0/8 will become *.10; excluded entries will be
represented by a CNAME to ‘excluded’ name, so such name should
not be present in a data set). In this mode, rbldnsd ignores -r
(root directory) option.
-v Do not show exact version information in response to
version.bind CH TXT queries (by default rbldnsd responds to such
queries since version 0.98). With single -v, rbldnsd will only
return "rbldnsd" to the caller, without the version number.
Second -v disables providing any information in response to such
requests, i.e. rbldnsd will return REFUSE code.
DATASET TYPES AND FORMATS
Dataset files are text files which are interpreted depending on type
specified in command line. Empty lines and lines starting with hash
character (#) or semicolon (;) are ignored, except for a special case
outlined below in section titled "Special Entries".
A (comma-separated) list of files in dataset specification (in
type:file,file,...) is interpreted as if all files where logically
combined into one single file.
rbldnsd is designed to service a DNSBL, where each entry have single A
record and optional TXT record assotiated with it. rbldnsd allows to
specify A value and TXT template either for each entry individually, or
to use default A value and TXT template pair for a group of entries.
See section "Resulting A values and TXT templates" below for a way to
If a line starts with a dollar sign ($), hash character and a dollar
sign (#$), semicolon and dollar sign (;#) or colon and a dollar sign
(:$), it is interpreted in a special way, regardless of dataset type
(this is one exception where a line starting with hash character is not
ignored - to be able to use zone files for both rbldnsd and for DJB’s
rbldns). The following keywords, following a dollar sign, are
$SOA ttl origindn persondn serial refresh retry expire minttl
Specifies SOA (Start Of Authority) record for all zones using
this dataset. Only first SOA record is interpreted. This is
the only way to specify SOA - by default, rbldnsd will not add
any SOA record into answers, and will REFUSE to answer to
certain queries (notably, SOA query to zone’s base domain name).
It is recommended, but not mandatory to specify SOA record for
every zone. If no SOA is given, negative replies will not be
cacheable by caching nameservers. Only one, first $SOA line is
recognized in every dataset (all subsequent $SOA lines
encountered in the same dataset are silently ignored). When
constructing a zone, SOA will be taken from first dataset where
$SOA line is found, in an order as specified in command line,
subsequent $SOA lines, if any, are ignored. This way, one may
overwrite $SOA found in 3rd party data by prepending small local
file to the dataset in question, listing it before any other
If serial value specified is zero, timestamp of most recent
modified file will be substituted as serial.
If ttl field is zero, default ttl (-t option or last $TTL value,
see below) will be used.
All time fields (ttl, refresh, retry, expire, minttl) may be
specified in time units. See -t option for details.
$NS ttl nameserverdn nameserverdn...
Specifies NS (Name Server) records for all zones using this
dataset. Only first $NS line in a dataset is recognized, all
subsequent lines are silently ignored. When constructing a zone
from several datasets, rbldnsd uses nameservers from $NS line in
only first dataset where $NS line is given, in command-line
order, just like for $SOA record. Only first 20 namservers are
recognized. Individual nameserver(s) may be prefixed with a
minus sign (-), which means this single nameserver will be
ignored by rbldnsd. This is useful to temporary comment out one
nameserver entry without removing it from the list. If ttl is
zero, default ttl will be used. The list of NS records, just
like $SOA value, are taken from the first data file in a dataset
where the $NS line is found, subsequent $NS lines, if any, are
Specifies TTL (time-to-live) value for all records in current
dataset. See also -t option. $TTL special overrides -t value
on a per-dataset basis.
Specifies maximum size of IPv4 range allowed for IPv4-based
datasets. If an entry covers more IP addresses than range-size,
it will be ignored (and a warning will be logged). range-size
may be specified as a number of hosts, like 256, or as network
prefix lenght, like /24 (the two are the same):
This constraint is active for a dataset it is specified in, and
can be owerwritten (by subsequent $MAXRANGE statement) by a
smaller value, but can not be increased.
(n is a single digit). Specifies a substitution variable for
use as $n placeholders (the $n entries are ignored in generic
daaset). See section "Resulting A values and TXT templates"
below for description and usage examples.
A set of IP addresses or CIDR address ranges, together with A and TXT
resulting values. IP addresses are specified one per line, by an IP
address prefix (initial octets), complete IP address, CIDR range, or IP
prefix range (two IP prefixes or complete addresses delimited by a
dash, inclusive). Examples, to specify 127.0.0.0/24:
to specify 127.16.0.0-127.31.255.255:
Note that in prefix range, last boundary is completed with all-ones
(255), not all-zeros line with first boundary and a prefix alone. In
prefix ranges, if last boundary is only one octet (127.16-31), it is
treated as "suffix", as value of last specified octet of the first
boundary prefix (127.16.0-31 is treated as 127.16.0.0-127.16.31.255,
After an IP address range, A and TXT values for a given entry may be
specified. If none given, default values in current scope (see below)
applies. If a value starts with a colon, it is interpreted as a pair
of A record and TXT template, delimited by colon (:127.0.0.2:This entry
is listed). If a value does not start with a colon, it is interpreted
as TXT template only, with A record defaulting to the default A value
in current scope.
IP address range may be followed by a comment char (either hash
character (#) or semicolon (;)), e.g.:
127/8 ; loopback network
In this case all characters up to the end of line are ignored, and
default A and TXT values will be used for this IP range.
Every IP address that fits within any of specified ranges is "listed",
and rbldnsd will respond to reverse queries against it within specified
zone with positive results. In contrast, if an entry starts with an
exclamation sign (!), this is an exclusion entry, i.e. corresponding
address range is excluded from being listed (and any value for this
record is ignored). This may be used to specify large range except
some individual addresses, in a compact form.
If a line starts with a colon (:), this line specifies the default A
value and TXT template to return (see below) for all subsequent entries
up to end of current file. If no default entry specified, and no value
specified for a given record, rbldnsd will return 127.0.0.2 for
matching A queries and no record for matching TXT queries. If TXT
record template is specified and contains occurences of of dollar sign
($), every such occurence is replaced with an IP address in question,
so singe TXT template may be used to e.g. refer to a webpage for an
additional information for a specific IP address.
Set of IP4 CIDR ranges with corresponding (A, TXT) values. Similar to
ip4set, but uses different internal representation (implemented as a
patricia trie), accepts CIDR ranges only (not a.b.c.d-e.f.g.h), allows
to specify only one value per CIDR range, and returns only one, most
close matching, entry on queries. Exclusions are supported too. This
dataset is not memory-efficient to store many single IP addresses, but
it is ok to use it to store many possible wide CIDR ranges.
"trivial" ip4set: a set of single IP addresses (one per line), with the
same A+TXT template. This dataset type is more efficient than ip4set
(in both memory usage and access times), but have obvious limitation.
It is intended for DNSBLs like DSBL.org, ORDB.org and similar, where
each entry uses the same default A+TXT template. This dataset uses
only half a memory for the same list of IP addresses compared to
Set of (possible wildcarded) domain names with associated A and TXT
values. Similar to ip4set, but instead of IP addresses, data consists
of domain names (not in reverse form). One domain name per line,
possible starting with wildcard (either with star-dot (*.) or just a
dot). Entry starting with exclamation sign is exclusion. Default
value for all subsequent lines may be specified by a line starting with
Wildcards are interpreted as follows:
only example.com domain is listed, not subdomains thereof. Not
a wildcard entry.
all subdomains of example.com are listed, but not example.com
all subdomains of example.com and example.com itself are listed.
This is a shortcut: to list a domain name itself and all it’s
subdomains, one may either specify two lines (example.com and
*.example.com), or one line (.example.com).
This dataset type may be used instead of ip4set, provided all CIDR
ranges are expanded and reversed (but in this case, TXT template will
be expanded differently).
Generic type, simplified bind-style format. Every record should be on
one line (line continuations are not supported), and should be
specified completely (i.e. all domain names in values should be
fully-qualified, entry name may not be omitted). No wildcards are
accepted. Only A, TXT, and MX records are recognized. TTL value may
be specified before record type. Examples:
# specify some values for current zone
$NS 0 ns1.ex.com ns2.ex.com
# record with TTL
www 3000 A 127.0.0.1
about TXT "ex.com combined blocklist"
This is a special dataset that stores no data by itself but acts like a
container for several other datasets of any type except of combined
type itself. The data file contains an optional common section, where
various specials are recognized like $NS, $SOA, $TTL (see above), and a
series of sections, each of which defines one (nested) dataset and
several subzones of the base zone, for which this dataset should be
consulted. New (nested) dataset starts with a line
$DATASET type[:name] subzone subzone...
and all subsequent lines up to the end of current file or to next
$DATASET line are interpreted as a part of dataset of type type, with
optional name (name is used for logging purposes only, and the whole
":name" (without quotes or square brackets) part is optional). Note
that combined datasets cannot be nested. Every subzone will always be
relative to the base zone name specified on command line. If subzone
specified as single character "@", dataset will be connected to the
base zone itself.
This dataset type aims to simplify subzone maintenance, in order to be
able to include several subzones in one file for easy data transfer,
atomic operations and to be able to modify list of subzones on remote
Example of a complete dataset that contains subzone ‘proxies’ with a
list of open proxies, subzone ‘relays’ with a list of open relays,
subzone ‘multihop’ with output IPs of multihop open relays, and the
base zone itself includes proxies and relays but not multihops:
# common section
$NS 1w ns1.ex.com ns2.ex.com
$SOA 1w ns1.ex.com admin.ex.com 0 2h 2h 1w 1h
# list of open proxies,
# in ‘proxies’ subzone and in base zone
$DATASET ip4set:proxy proxies @
:2:Open proxy, see http://bl.ex.com/proxy/$
# list of open relays,
# in ‘relays’ subzone and in base zone
$DATASET ip4set:relay relays @
:3:Open relay, see http://bl.ex.com/relay/$
# list of optputs of multistage relays,
# in ‘multihop’ subzone only
$DATASET ip4set:multihop-relay multihop
:4:Multihop open relay, see http://bl.ex.com/relay/$
# for the base zone and all subzones,
# include several additional records
$DATASET generic:common proxies relays multihop @
@ A 127.0.0.8
www A 127.0.0.8
@ MX 10 mx.ex.com
# the above results in having the following records
# (provided that the base zone specified is bl.ex.com):
# proxies.bl.ex.com A 127.0.0.8
# www.proxies.bl.ex.com 127.0.0.8
# relays.bl.ex.com A 127.0.0.8
# www.relays.bl.ex.com 127.0.0.8
# multihop.bl.ex.com A 127.0.0.8
# www.multihop.bl.ex.com 127.0.0.8
# bl.ex.com A 127.0.0.8
# www.bl.ex.com 127.0.0.8
Note that $NS and $SOA values applies to the base zone only, regardless
of the placement in the file. Unlike the $TTL values and $n
substitutions, which may be both global and local for a given
Resulting A values and TXT templates
In all zone file types except generic, A values and TXT templates are
specified as following:
If a line starts with a colon, it specifies default A and TXT for all
subsequent entries in this dataset. Similar format is used to specify
values for individual records, with the A value (enclosed by colons)
127.0.0.2 :127.0.0.2:Blacklisted: http://example.com/bl?$
or, without specific A value:
127.0.0.2 Blacklisted: http://example.com/bl?$
Two parts of a line, delimited by second colon, specifies A and TXT
record values. Both are optional. By default (either if no default
line specified, or no IP address within that line), rbldnsd will return
127.0.0.2 as A record. 127.0.0 prefix for A value may be omitted, so
the above example may be simplified to:
There is no default TXT value, so rbldnsd will not return anything for
TXT queries it TXT isn’t specified.
When A value is specified for a given entry, but TXT template is
omitted, there may be two cases interpreted differently, namely,
whenever there’s a second semicolon (:) after the A value. If there’s
no second semicolon, default TXT value for this scope will be used. In
contrast, when second semicolon is present, no TXT template will be
generated at all. All possible cases are outlined in the following
# default A value and TXT template
:127.0.0.2:IP address $ is listed
# 127.0.0.4 will use default A and TXT
# 127.0.0.5 will use specific A and default TXT
# 127.0.0.6 will use specific a and no TXT
# 127.0.0.7 will use default A and specific TXT
127.0.0.7 IP address $ running an open relay
In a TXT template, some references to substitution variables are
replaced with values of that variables. In particular, single dollar
sign ($) is replaced by a listed entry (an IP address in question for
IP-based datasets and the domain name for domain-based datasets).
$n-style constructs, where n is a single digit, are replaced by a
substitution variable $n defined for this dataset in current scope (see
section "Special Entries" above). To specify a dollar sign as-is, use
For example, the following lines:
$1 See http://www.example.com/bl
$2 for details
127.0.0.2 $1/spammer/$ $2
127.0.0.3 $1/relay/$ $2
127.0.0.4 This spammer wants some $$$$. $1/$
will result in the following text to be generated:
See http://www.example.com/bl/spammer/127.0.0.2 for details
See http://www.example.com/bl/relay/127.0.0.3 for details
This spammer wants some $$. See http://www.example.com/bl/127.0.0.4
This is not a real dataset, while the syntax and usage is the same as
with other datasets. Instead of defining which records exists in a
given zone and which does not, acl dataset specifies which client hosts
(peers) are allowed to query the given zone. The dataset specifies a
set of IPv4 address ranges (currently, IPv6 addresses are not
supported) in a form of CIDRs (with the syntax exactly the same as
understood by ip4trie dataset), together with action specifiers. When
a query is made from an IP address listed (not for the IP address),
specified action changes rules used to construct the reply. Possible
actions and their meanings are:
ignore all queries from this IP address altogether. rbldnsd
acts like there was no query received at all. This is the
refuse all queries from the IP in question. rbldnsd will always
return REFUSED DNS response code.
:empty pretend there’s no data in all other datasets for the given
zone. This means that all the clients in question will always
receive reply from rbldnsd telling that the requested IP address
or domain name is not listed in a given DNSBL. rbldnsd still
replies to metadata queries (SOA and NS records, and to all
queries satisfied by generic dataset if specified for the given
zone) as usual.
usual A+TXT template as used by other datasets. This means that
rbldnsd will reply to any valid DNSBL query with "it is listed"
answer, so that the client in question will see every IP address
or domain name is listed in a given DNSBL. TXT record used in
the reply is taken from the acl dataset instead of real
datasets. Again, just like with empty case, rbldnsd will
continue replying to metadata queries (including generic
datasets if any) as usual.
Only one ACL dataset can be specified for a given zone, and each zone
must have at least one non-acl dataset. It is also possible to specify
one global ACL dataset, by specifying empty zone name (which is not
allowed for other dataset types), like
rbldnsd ... :acl:filename...
In this case the ACL defined in filename applies to all zones. If
there are both global ACL and local zone-specific ACL specified, both
will be consulted and actions taken in the order specified above, ie,
if either ACL returns ignore for this IP, the request will be ignored,
else if either ACL returns refuse, the query will be refused, and so
on. If both ACLs specifies "always listed" A+TXT template, the reply
will contain A+TXT from global ACL.
For this dataset type, only a few $-style specials are recognized. In
particular, $SOA and $NS keywords are not allowed. When rbldnsd
performs $ substitution in the TXT template returned from ACL dataset,
it will use client IP address to substitute for a single $ character,
instead of the IP address or domain name found in the original query.
Rbldnsd handles the following signals:
SIGHUP recheck zone files and reload any outdated ones. This is done
automatically if enabled, see -c option. Additionally, rbldnsd
will reopen logfile upon receiving SIGHUP, if specified (-l
Log current statistic counters into syslog. Rbldnsd collects
how many packets it handled, how many bytes was received, sent,
how many OK requests/replies (and how many answer records) was
received/sent, how many NXDOMAIN answers was sent, and how many
errors/refusals/etc was sent, in a period of time.
The same as SIGUSR1, but reset all counters and start new sample
Some unsorted usage notes follows.
Generating and transferring data files
When creating a data file for rbldnsd (and for anything else, it is a
general advise), it is a good idea to create the data in temporary file
and rename the temp file when all is done. Never try to write to the
main file directly, it is possible that at the same time, rbldnsd will
try to read it and will get incomplete data as the result. The same
applies to copying data using cp(1) utility and similar (including
scp(1)), that performs copying over existing data. Even if you’re sure
noone is reading the data while you’re copying or generating it,
imagine what will happen if you will not be able to complete the
process for whatever reason (interrupt, filesystem full, endless number
of other reasons...). In most cases is better to keep older but
correct data instead of leaving incomplete/corrupt data in place.
scp remote:data target.tmp && mv target.tmp target
scp remote:data target
./generate.pl > target.tmp && mv target.tmp target
./generate.pl > target
From this point of view, rsync(1) command seems to be safe, as it
always creates temporary file and renames it to the destination only
when all is ok (but note the --partial option, which is good for
downloading something but may be wrong to transfer data files --
usually you don’t want partial files to be loaded). In contrast,
scp(1) command is not safe, as it performs direct copying. You may
still use scp(1) in a safe manner, as shown in the example above.
Also try to eliminate a case when two (or more) processes performs data
copying/generation at the same time to the same destination. When your
data is generated by a cron job, use file locking (create separate lock
file (which should never be removed) and flock/fcntl it in exclusive
mode without waiting, exiting if lock fails) before attempting to do
other file manipulation.
Absolute vs relative domain names
All keys specified in dataset files are always relative to the zone
base DN. In contrast, all the values (NS and SOA records, MX records
in generic dataset) are absolute. This is different from BIND
behaviour, where trailing dot indicates whenever this is an absolute or
relative DN. Trailing dots in domain names are ignored by rbldnsd.
Several zones may be served by rbldnsd, every zone may consist of
several datasets. There are numerous ways to combine several data
files into several zones. For example, suppose you have a list of
dialup ranges in file named ‘dialups’, and a list of spammer’s ip
addresses in file named ‘spammers’, and want to serve 3 zones with
rbldnsd: dialups.bl.ex.com, spam.bl.ex.com and bl.ex.com which is a
combination of the two. There are two ways to do this:
rbldnsd options... \
rbldnsd options... \
(note you should specify combined bl.ex.com zone after all its subzones
in a command line, or else subzones will not be consulted at all).
In the first form, there will be 3 independent data sets, and every
record will be stored 2 times in memory, but only one search in
internal data structures will be needed to resolve queries for
aggregate bl.ex.com. In second form, there will be only 2 data sets,
every record will be stored only once (both datasets will be reused),
but 2 searches will be performed by rbldnsd to answer queries against
aggregate zone (but difference in speed is almost unnoticeable). Note
that when aggregating several data files into one dataset, an exclusion
entry in one file becomes exclusion entry in the whole dataset (which
may be a problem when aggregating dialups, where exclusions are common,
with open relays/proxies, where exclusions are rare if at all used).
Similar effect may be achieved by using combined dataset type,
sometimes more easily. combined dataset results in every nested
dataset to be used independantly, like in second form above.
combined dataset requires rbldnsd to be the authoritative nameserver
for the whole base zone. Most important, one may specify SOA and NS
records for the base zone only. So, some DNSBLs which does not use a
common subzone for the data, cannot use this dataset. An example being
DSBL.org DNSBL, where each of list.dsbl.org, multihop.dsbl.org and
unconfirmed.dsbl.org zones are separate, independant zones with
different set of nameservers. But for DSBL.org, where each dataset is
really independant and used only once (there’s no (sub)zone that is as
a combinations of other zones), combined dataset isn’t necessary. In
contrast, SORBS.net zones, where several subzones used and main zone is
a combination of several subzones, combined dataset is a way to go.
All authoritative nameservers should be set up similarily
When you have several nameservers for your zone, set them all in a
similar way. Namely, if one is set up using combined dataset, all the
rest should be too, or else DNS meta-data will be broken. This is
because metadata (SOA and NS) records returned by nameservers using
combined and other datasets will have different origin. With combined
dataset, rbldnsd return NS and SOA records for the base zone, not for
any subzone defined inside the dataset. Given the above example with
dialups.bl.ex.com, spammers.bl.ex.com and aggregate bl.ex.com zones,
and two nameservers, first is set up in any ways described above (using
individual datasets for every of the 3 zones), and second is set up for
the whole bl.ex.com zone using combined dataset. In this case, for
queries against dialups.bl.ex.com, first nameserver will return NS
dialups.bl.ex.com. IN NS a.ns.ex.com.
while second will always use base zone, and NS records will look like
bl.ex.com. IN NS a.ns.ex.com.
All authoritative nameservers for a zone must have consistent metadata
records. The only way to achieve this is to use similar configuration
(combined or not) on all nameservers. Have this in mind when using
other software for a nameserver.
Generic dataset usage
generic dataset type is very rudimentary. It’s purpose is to
complement all the other type to form complete nameserver that may
answer to A, TXT and MX queries. This is useful mostly to define A
records for HTTP access (relays.bl.example.com A, www.bl.example.com A
just in case), and maybe descriptive texts as a TXT record.
Since rbldnsd only searches one, most closely matching (sub)zone for
every request, one cannot specify a single e.g. generic dataset in
proxies TXT list of open proxies
www.proxies A 127.0.0.8
relays TXT list of open relays
www.relays A 127.0.0.9
for several (sub)zones, each of which are represented as a zone too
(either in command line or as combined dataset). Instead, several
generic datasets should be specified, separate one for every (sub)zone.
If the data for every subzone is the same, the same, single dataset may
be used, but it should be specified for every zone it should apply to
(see combined dataset usage example above).
Most of the bugs outlined in this section aren’t really bugs, but
present due to non-standartized and thus unknown expected behaviour of
a nameserver that serves a DNSBL zone. rbldnsd matches BIND runtime
behaviour where appropriate, but not always.
rbldnsd lowercases some domain names (the ones that are lookup keys,
e.g. in ‘generic’ and ‘dnset’ datasets) when loading, to speed up
lookup operations. This isn’t a problem in most cases.
There is no TCP mode. If a resource record does not fit in UDP packet
(512 bytes), it will be silently ignored. For most usages, this isn’t
a problem, because there should be only a few RRs in an answer, and
because one record is usually sufficient to decide whenever a given
entry is "listed" or not. rbldnsd isn’t a full-featured nameserver,
rbldnsd will not always return a list of nameserver records in the
AUTHORITY section of every positive answer: NS records will be provided
(if given) only if there’s a room for them in single UDP packet. If
records does not fit, AUTHORITY section will be empty.
rbldnsd does not allow AXFR operations. For DNSBLs, AXFR is the
stupidiest yet common thing to do - use rsync for zone transfers
instead. This isn’t a bug in rbldnsd itself, but in common practice of
using AXFR and the like to transfer huge zones in a format which isn’t
suitable for such a task. Perhaps in the future, if there will be some
real demand, I’ll implement AXFR "server" support (so that rbldnsd will
be able to act as master for BIND nameservers, but not as secondary),
but the note remains: use rsync.
rbldnsd truncates all TXT records to be at most 255 bytes. DNS specs
allows longer TXTs, but long TXTs is something that should be avoided
as much as possible - TXT record is used as SMTP rejection string.
Note that DNS UDP packet is limited to 512 bytes. rbldnsd will log a
warning when such truncation occurs.
This manpage corresponds to rbldnsd version 0.995fR.
The rbldnsd daemon written by Michael Tokarev <firstname.lastname@example.org>, based on
ideas by Dan Bernstein and his djbdns package.
Jul 2004 rbldnsd(8)