options

早期freebsd实现

OPTIONS
    Original: Paul Vixie, 28Mar92
    Revised: $Id: OPTIONS,v 4.9.1.8 1993/11/03 12:25:50 vixie Exp $

Options available in this version of BIND are controlled by conf/options.h,
rather than by $(DEFS) in the Makefile.  The options are:

DEBUG (origin: U C Berkeley)
	enables the -d command line option, and allows SIGUSR1 to increment
and SIGUSR2 to decrement the internal variable "debug", which in turn controls
hundreds of fprintf()'s out to /usr/tmp/named.run.
	you probably want this.  it makes the binary bigger but not slower (or
at least not much slower), but SIGUSR[12] are the only way you'll track down
misconfigured name servers that hose you down with billions of bogus requests.

ALLOW_T_UNSPEC (origin: MIT Project Athena)
	enables the "unspec" RR type for ancient athena software that does not
know about TXT RR's.
	you probably do not care about this.

ALLOW_UPDATES (origin: Mike Schwartz, University of Washington)
	enables "dynamic updates", described in "doc/DynamicUpdate".  this lets
you update named's in-memory database on the fly if you have the right client.
there is absolutely no security around this; if you enable it, anyone who can
reach your server can update your database.
	you probably do not want to enable this.

INVQ (origin: U C Berkeley, with #ifdef's by Paul Vixie)
	enables "inverse queries", which in all of the internet only one client
ever uses: ancient nslookup.  if you build named with INVQ defined as "1",
you get the time-honored behaviour of supporting this whole class of
queries for no real purpose other than to waste a few hundred kilobytes of
your memory and about 3% of named's total CPU time.  if you build with
INVQ set to "0", old nslookups will not be able to reach your server in their
startup phase, and you will have to use the "server" command after it fails
over to some other server.
	you probably want this.

DSTORAGE (origin: U C Berkeley, with #ifdef's by Paul Vixie)
	enables a malloc-debugger that checks for overruns on both ends of
each allocated block of memory.  used when debugging since C has no bounds
or type checking.
	you probably do not want this.

DMALLOC (origin: Paul Vixie of Digital)
	enables a malloc-debugger that traces all allocated blocks of memory
such that SIGIOT's output (see STATS, above) includes a list of all malloc's
in the program, how many times each has been called, how many blocks of memory
allocated by that malloc are not yet free, and how many bytes they use up.
under each one will be a list of each free/realloc that has deallocated a block
of that malloc's memory, and how many times it has done so.
	this is extremely helpful for finding memory leaks.  as such, you
probably do not want this unless you are debugging named.

CRED (origin: Paul Vixie of Digital)
	enables a system of "credibility checking" on all data in the memory-
resident database.  every RR that comes in will be tagged with a credibility
index with zone files being highest, followed by authoritative answers, then
non-authoritative answers, then finally by additional data.  when any RR is
being added to a node ("name") in the database, all RR's of that type with a
lower credibility index will be flushed.  this tends to do away with additional
data, which is one of the greatest sources of database pollution in the DNS.
data that comes in with lower credibility than what we already have is ignored.
	with CRED enabled, additional data is deprecated such that every
time an additional-data RR is used, its Time To Live (TTL) is multiplied by
0.95, effectively lowering it by 5% of its current value.  this causes
additional data to be timed out rather quickly, and as soon as it times
out, a sysquery() will be sent to some authoritative server, which in turn
results in a real live answer which tends to lock out future additional
data on that <name,type> tuple.
	due to source dependencies, CRED also controls a bug fix that keeps
all sysquery() responses from being entered into the "root cache".  you can
see the effect of this by dumping your database to disk with SIGINT and
looking at the bottom of the file.  try it with and without CRED, letting a
few million queries through first.  without CRED, you'll see a bunch of
non-root junk in the section of the dump that is reserved for the "hints".
	you probably want this.

XFRNETS (origin: Paul Vixie of Digital)
	enables the "xfrnets" command in named.boot.  this has the same
syntax as "forwarders" and "sortlist" -- that is, a list of dotted quads.
each one is a network (16.0.0.0 and 130.180.0.0 are examples) or a host.
if you put any xfrnets commands into your named.boot, then zone transfers
will only be honored if they come from inside one of the specified
networks.  this is very useful if you want to keep people outside from
being able to trivially map your entire network, but it doesn't stop them
from iterating so it's more annoying than secure.
	this feature was once called "tcplist" out of ignorance on my part,
but with advice from phil almquist i decided to rename it "xfrnets" and make
it only control zone transfers -- previously it controlled all TCP connections
which made certain TCP-only resolvers unable to use our servers.  the "tcplist"
syntax still works; it is a synonym for "xfrnets".
	you probably do not care about this.

PID_FIX (origin: Don Lewis of Harris)
	tells named that if it starts up but can't keep going because another
nameserver is already running (and sitting on the server port), it should
put the /etc/named.pid (/var/run/named.pid) file back the way it found it.
	you probably want this.

FWD_LOOP (origin: Don Lewis of Harris)
	tells named that if you list any of your own IP addresses in a
"forwarders" command in your named.boot file, you should be scolded.
	you probably want this.

NO_GLUE (origin: Don Lewis of Harris, and Andrew Partan of UUNET)
	tells named-xfer that incoming zone transfers should be checked
for "glue" that comes from a zone outside the zone being transferred, and
comment this garbage out in the zone file so that when named reads in the
zone file after named-xfer exits, the garbage will not be entered into the
memory-resident database.
	also tells named that when it is performing an outgoing zone
transfer, it should not send any of these "glue" records.
	you definitely want this.

BOGUSNS (origin: Piet Beertema of EUNet)
	enables the "bogusns" command in named.boot.  this has the same
syntax as forwarders and sortlist.  any NS RR's that come in whose addresses
are on the list of "bogusns" addresses will be ignored.  this is the last
resort when someone is bogusly advertising themselves as a root server.
	you probably want this, just in case, though you won't use it often.

QRYLOG (origin: Bryan Beecher of UMich)
	enables "query logging", such that SIGWINCH toggles tracing of all
incoming queries.  the trace is sent to syslog, and is huge, but when you
need this you will need it bad and it does not slow named down or make it
larger.
	If you define QRYLOG you may also start up named in query logging
mode by using the -q flag.  If you do so you will probably want to analyze
the logs produced, the dnsstats and lamers scrips (in the contrib/umich
directory) will do it for you.
	you probably want this.

LOGFAC (origin: various people)
	If you start up named with the -q flag you will be logging
large amounts of data, and probably will not want them logged to the
default logging facility, which is LOG_DAEMON.  You will want to
redefine LOGFAC, presumably to LOC_LOCALn (0 <= n <= 7).  Remember to
modify /etc/syslog.conf appropriately.
	This only works on a system with a modern syslogd.

YPKLUDGE (origin: Piet Beertema of EUNet)
	certain versions of NIS/YP are capable of using the DNS for names
that cannot be found in the YP servers.  of these, certain versions can't
tell the difference between a dotted quad and a domain name, and they send
queries to the DNS for dotted quads as if they were domain names.  if your
named does not do anything special with these queries, they will end up
getting forwarded to other servers, effectively hosing all of you down with
endless useless network traffic.  YPKLUDGE enables some checking in named
that lets it catch these bogus queries and send back immediate errors.
	If you run "ypserv -i" you definitely want this, as a malconfigured
NIS server can cause DNS "flood" queries otherwise. Trust me.

TRACEROOT (origin: [e-mail] and Bryan Beecher of UMich)
	enables some checking in named for bogus root nameservers.  This
code has been in use at U-M for years, so it is pretty well tested, plus we
have never been burned by the "bogus root NS scares" that have plagued the
DNS off and on.
	this is a feature that people will very much want to use.

LOCALDOM (origin: Berkeley)
	if set, the "domain" directive is recognized in the named.boot file.
this causes us to retry queries with the specified domain appended to the
name if the first lookup fails.  this is a very bad idea since a given name
server will often be used by clients in more than one domain -- a name server
should _not_ make any presumptions as to the "home domain" of a requestor.