changes

早期freebsd实现

Kenneth Almquist (no relation to Phil, as far as I know) posted a patch for
res_send() that lets it keep track of servers that are responding "SERVFAIL"
or some other fatal condition; these servers are NOT used for retries of the
current query.  This information is not persistent between calls to res_send()
since future calls will probably be for different {name,type} queries, which
will not neccessarily fail in the same way.  This change is trivial and makes
a measurable difference in the amount of DNS traffic on my local net.

4.9-BETA ------------------- April 17, 1993 -- Paul Vixie -- DECWRL

"Peter Koch" <[e-mail]>'s previous patch caused core
dumps on some systems.  I fixed part of it and Peter sent me a fix for
the rest of it.  All is now well.

The Bind Operations Guide in doc/BOG has been updated to 4.9.  Also, the
man page in man/named.8 has had some patches applied.  The copyrights are
all fixed now.  Let's get this thing OUT of here!

4.9-ALPHA ------------------- March 15, 1993 -- Paul Vixie -- DECWRL

There was a really bad bug affecting wildcards.  I received a patch
from "Peter Koch" <[e-mail]> which fixes some of
it, but I can't quite motivate myself to fix the rest of it since I
know that what's _really_ wrong is going to require chainsaws and
dynamite to fix and that'll add another year to the release.  I think
that this patch will hold us for a while.

There are a LOT of portability changes that I'm holding onto, especially
including 64-bit fixes.  Do not submit any more portability changes
until 4.9.1 opens.  Go ahead and make them, but be prepared to remake
them later.  Let me know what you are doing but don't send me any diffs
for portability until I ask for them.  4.9 has been stuck in the barrel
for way too long already -- patches that don't fix RFC-noncompliance or
core dumps will just go into my "todo" folder (which is presently a
black hole of great mass).

4.9-ALPHA ---------------- Febrtuary 2, 1992 -- Paul Vixie -- DECWRL

Mostly portability fixes.  The nslookup "lex" problem is BSDI-specific
and I'm not going to hold up release because of it.  This will be the
last alpha release before the public beta.  It is, as usual, running
the DEC.COM primary name service and has done so for more than a week
without any problems.

4.9-ALPHA ---------------- January 10, 1993 -- Paul Vixie -- DECWRL

Once I get the known bug in nslookup (see below) fixed, this version is going
to go into public beta.  I would appreciate it if everyone would try it out.

KNOWN BUG IN THIS RELEASE: something wild is going on inside of the yylex()
routing on BSD/386 systems.  It only affects nslookup.  I'm still trying to
figure out how I'm going to debug this; lex experts, please see what's going
on.  None of the changes since the 930105 release should have been capable
of producing this change, but something is sure doing it.

I finally fixed the {GET,PUT}{SHORT,LONG} macros to stop issuing warnings
on HP-UX systems.  They are also warning-free on Ultrix(SPIM,VAX), BSDI(386),
and SunOS(SPARC) systems.  I took the plunge and changed the internal functions
in res/res_comp.c to depend on these macros instead of duplicating the code,
and everything still works.

Tom Limoncelli found three ancient memory leaks.  I fixed two of them
but the last one looks too much like a "cannot happen" for me to be
willing to experiment with it.  Besides which, it's "very" minor.

Uses setsid() on POSIX systems.  PID file is now optional.  (arc@sgi)

Comments (";" or "#") are now allowed in resolv.conf (arc@sgi).

Documentation and copyright changes in README.

Known to compile on NeXT machines.

Some portability changes for AIX, whose CC is very picky.

I forgot to mention in the 921227 release that T_RP is supported (arc@sgi).

I included a number of changes that Alan Barrett has been trying to get
in since the 921221 version.  Most are portability-related, and the few
things that are functional are changes to my own previous additions :-),
so I'm fairly sure that they are doing the right thing.  Alan's changes
include:

    include/arpa/nameser.h
	improved error diagnosis in the BYTE_ORDER configuration.

	changed hp9000 test to hp9000s300.  As far as I know, there is
	no hp9000 preprocessor symbol.  Should probably add other
	hp9000s<whatever> tests, but have not done so.

    named/ns.h
	Moved the XFER-related stuff from the end of the file to near
	the top, where it is grouped with similar stuff.

    Makefiles:
	Add SYSLIBS variable, so folk can compile with -lBSD easily.

	Changed install targets to make them easier to customise.

	make links wasn't handling named.{reload,restart}*

	Add ${CDEBUG} flag to link step.  Some debuggers don't work
	right if the program isn't linked with the -g flag.

    struct timeval members are declared as unsigned long on some systems.
	Add casts to (long) in several if statements that appear to assume
	that tv_sec is signed.

    PID_FIX in ns_main.c controlled more than just whether or not the
    pid file gets fixed.
	Changed it to control only that one feature.

    For debugging, it is useful for a nameserver to listen to non-standard
    port, but to forward requests to a standard port.
	Add "-p remote/local" option to named/ns_main.c.
	Also needed some other changes elsewhere.

    Don't forward back to the host that asked us a question, unless they
    asked from some port other than their nameserver port.  This allows a
    dig or nslookup user on a host to ask us questions with
    recursion-desired, where we are willing to recursively ask the
    nameserver on their host.  However, if a nameserver asks us something
    we will not recurse back to them.
	nslookup() in named/ns_forw.c checks for this and returns -1.
	ns_forw() and sysquery() notice this and return SERVFAIL.

	Moved the nsContainsUs functionality from a separate routine
	into nslookup().  No need to do the same tree walk several times.

    While trying to track down various problems, added detection
    and logging of errors in several syscalls in ns_main.c.

    Avoid integer overflow in roundtrip time calc in ns_resp.
    This needs a definition for INT_MAX.

    Fixed root zone transfer bug.  Also corrected some slightly misleading
    comments in the doaxfr() code, and added some more comments.

4.9-ALPHA ---------------- January 5, 1993 -- Paul Vixie -- DECWRL

This one was built and tested on Ultrix 4.2 (SPIM, MIPS CC and GCC),
BSD/386 (Gamma.4), Sun SPARC (4.0.3, sorry, that's the latest I have),
4.3BSD Reno (VAX, PCC), and Ultrix 3.0 (VAX PCC).

Moved res/defs.h to conf/portability.h; named/options.h to conf/options.h.

Portability changes for O_NDELAY.  SUNOS is really strange about this.

Removed some unneccessary goto's added to ns_main.c on 1jan.  Oops.

Art Harkin of HP sent in a number of small (read: obviously correct)
improvements, some related to portability, some to functionality.

4.9-ALPHA ---------------- January 1, 1993 -- Paul Vixie -- DECWRL

Changed all O_NONBLOCK to O_NDELAY.  Changed all {r}index to str{r}chr.

Added some SysV support in the form of bcopy->memcpy, bzero->memset.

Added C_HS support to named-xfer ([e-mail]).

Fixed a line-number problem in asp's "include" logic ([e-mail]).

streamq's were being used after free().  bug report from [e-mail]
and [e-mail].  bug fix by vixie.

In the resolver, we now default to address 127.0.0.1 rather than 0.0.0.0.
There's a comment in the code that explains why.

In the resolver, [e-mail] changed it to use inet_aton()
and included that function for those not running 4.4bsd.

[e-mail] also provided lots of portability fixes and
general cleanups, in particular to nslookup which he maintains for CSRG.

4.9-ALPHA ---------------- December 27, 1992 -- Paul Vixie -- DECWRL

Added strtoul() to libresolv.a since it's yet another neccessary function
that older systems don't have.  If we can stomach strcasecmp() we can sure
handle this.

Moved res/named/gethostnamadr.c to res/gethnamaddr.c (note basename change)
and res/named/sethostent.c to res/sethostent.c.  Since the host table stuff
isn't in separate files any more I saw no reason to retain the subdirectory.

Updated all the copyrights and applied the small lint changes that bring
the baseline of this version from "4.8.3 as seen on ucbarpa" up to "4.8.3
as released with net-2".  Thanks to the alpha testers for pointing this out
to me and for sending in the diffs.

With much howling and screaming, I ported this to UMIPS (MIPS System V).
There are a lot of really bad things going on in their libc.a, and now
they're going on in BIND as well.

I added a "res/defs.h" file and then proceeded to include it from all kinds
of files that aren't in res/.  I'm thinking of moving it but I'm also trying
to figure out where -- include/ is the wrong place.  res/defs.h has in it all
the ugly ifdef's needed to figure out whether this is a late-model BSD system,
a POSIX system, or just old.

All the "#endif" and "#else" cpp directives now have comments around their
annotations.  It turns out that System V CPP complains about "#endif DEBUG"
but has no problem with "#endif /*DEBUG*/".  In many cases where the #ifdef
was obviously visible and unambiguous, I simply removed the annotation.

The "l" is now a ";".  Thanks to all who replied :-).

There was a very bad bug in the named-xfer interface.  'nuff said.

AIX needs a 32-bit field for PID's.  I can't imagine.  But it's fixed.

The "domain" directive in named.boot is now an option, defaulting to off.

There was a benign bug in sqrm().

doaxfr() is now shorter and clearer.

There is an "include" directive in the named.boot file now.  Its syntax is
simple: "include somefile".  No quotes, no "#", no <brackets>.  This feature
was in 4.9-ALPHA as well, courtesy of Andrew Partan.  I forgot to document it.

4.9-ALPHA ---------------- December 21, 1992 -- Paul Vixie -- DECWRL

This release incorporates fixes from a lot of people, including many from
DECWRL.  Some fixes are just lint; some are to avoid dumping core on non-VAX
computers; many are to fix promiscuity, corruption, and rudeness.

Various internal DEC programmers have ported the old 4.8.3 code to various
not-entirely-BSD-like platforms and turned up some interesting lint.  All
of this has been fixed.  Also, we fixed a bad bug in the handling of timeouts
and SERVFAIL's when forwarders and slave are both used.

I have made major changes to the code inside the ALLOW_UPDATES ifdef's, but
I don't use it and have never compiled with that option turned on so I don't
know if it still works.  Given that SNMP has come and there is an IETF WG for
SNMP management of the DNS, I am thinking very seriously of purging all of the
ALLOW_UPDATES code in 4.9.1.  I suspect that Mike Schwartz will let me know if
this is ok..

(interrim "KJB" notes) ------------------- March, 1992 -- Paul Vixie -- DECWRL

If we are about to forward a query for some zone for which we are one of the
servers, we send back a SERVFAIL instead.  If we don't have it, chances are
good that the other name servers won't have it either.  This is the major
cause of "network meltdown" when the root servers declare you as a name server
for some zone you don't know about and havn't configured yourself for.

Fixed a memory leak such that if db_update() fails to update the database
from a response packet, a databuf will no longer be orphaned.  Also fixed
what looks like a similar leak in the ALLOW_UPDATES code but I don't use it
that hasn't been tested.

Fixed a memory sponge such that if we forward a query to someone who is not
ever going to answer it, we will eventually expire it from our query queue.