NetBSD Problem Report #9413
Received: (qmail 13820 invoked from network); 13 Feb 2000 23:01:44 -0000
Message-Id: <20000213230129.AFCC5D48@WintelKiller.HEH.Uni-Oldenburg.DE>
Date: Mon, 14 Feb 2000 00:01:29 +0100 (MET)
From: Thilo.Manske@HEH.Uni-Oldenburg.DE
Reply-To: Thilo.Manske@HEH.Uni-Oldenburg.DE
To: gnats-bugs@gnats.netbsd.org
Subject: resolver weirdnesses
X-Send-Pr-Version: 3.95
>Number: 9413
>Category: lib
>Synopsis: strange problems with NetBSD's resolver
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: itojun
>State: closed
>Class: sw-bug
>Submitter-Id: net
>Arrival-Date: Sun Feb 13 15:03:00 +0000 2000
>Closed-Date: Sat Feb 26 03:29:25 +0000 2000
>Last-Modified: Sat Feb 26 03:30:27 +0000 2000
>Originator: Thilo Manske
>Release: 4th and 12th February
>Organization:
Dies ist Thilos Unix Signature! Viel Spass damit.
>Environment:
System: NetBSD WintelKiller 1.4S NetBSD 1.4S (WintelKiller) #194: Sat Feb 12 15:16:01 MET 2000 thilo@WintelKiller:/usr/src/sys/arch/i386/compile/WintelKiller i386
No IPv6 enabled at all.
>Description:
I've discovered strange actions by NetBSD's resolver:
When I issue rlogin seti-100 NetBSD does this:
WintelKiller-100.65372 > seti-100.domain: 10782+ AAAA? seti-100.HEH.Uni-Oldenburg.DE. (47)
seti-100.domain > WintelKiller-100.65372: 10782 NXDomain* 0/1/0 (115)
WintelKiller-100.65371 > seti-100.domain: 10783+ AAAA? seti-100. (26)
seti-100.domain > WintelKiller-100.65371: 10783 NXDomain 0/1/0 (99)
HEH.Uni-Oldenburg.de is my defaultdomain, set in /etc/resolv.conf (search
option))
What's strange is, that
A) seti-100 is in /etc/hosts (and nsswitch.conf set properly)
B) I don't have IPv6 enabled, so I'm not intereseted in "AAAA" records
C) Some tools do this (rlogin, telnet, ftp, rsh, tcpdump...) some not
(i.e. work correctly; ping and ntpdate e.g. - and host of course (doesn't
use NetBSD'S resolver, right?))
A dummy program using gethostbyname(3) works correctly as well.
When our site and thus my cacheing nameserver (seti-100) has lost its
connection to the rest of the 'net this saturday name resolving needed
ages. (And that was the reason I've started investigating this :-)
Here's a lookup of an nonexistent host (not present in /etc/host):
WintelKiller-100.65376 > seti-100.domain: 19293+ AAAA? doesnotexist.HEH.Uni-Oldenburg.DE. (51)
seti-100.domain > WintelKiller-100.65376: 19293 NXDomain* 0/1/0 (119)
WintelKiller-100.65375 > seti-100.domain: 19294+ AAAA? doesnotexist. (30)
seti-100.domain > WintelKiller-100.65375: 19294 NXDomain 0/1/0 (103)
WintelKiller-100.65374 > seti-100.domain: 19295+ A? doesnotexist.HEH.Uni-Oldenburg.DE. (51)
seti-100.domain > WintelKiller-100.65374: 19295 NXDomain* 0/1/0 (119)
WintelKiller-100.65373 > seti-100.domain: 19296+ A? doesnotexist. (30)
seti-100.domain > WintelKiller-100.65373: 19296 NXDomain 0/1/0 (103)
Looks like the resolver does this in some cases:
1. IPv6 adress requests (AAAA),
2. following the rules in /etc/nsswitch.conf
I've browsed the resolver sources in src/lib/libc/net quickly to find
a maybe undocumented option of nsswitch.conf but didn't succeed.
I've discovered the problem with world build around 4th February but it
still exists with yesterday's sources.
Here are parts of a ktrace "rsh seti-100 echo":
(I hope the interesting ones):
1190 rsh NAMI "/etc/nsswitch.conf"
1190 rsh RET __stat13 0
1190 rsh CALL open(0x480cf882,0,0x1b6)
1190 rsh NAMI "/etc/hosts"
1190 rsh RET open 3
1190 rsh CALL __fstat13(0x3,0xefbfd53c)
1190 rsh RET __fstat13 0
1190 rsh CALL read(0x3,0x8050000,0x2000)
1190 rsh GIO fd 3 read 64 bytes
"127.0.0.1 localhost
10.2.0.1 seti-100
10.2.0.2 WintelKiller-100
"
1190 rsh RET read 64/0x40
1190 rsh CALL read(0x3,0x8050000,0x2000)
1190 rsh GIO fd 3 read 0 bytes
""
1190 rsh RET read 0
1190 rsh CALL close(0x3)
1190 rsh RET close 0
1190 rsh CALL madvise(0x8050000,0x2000,0x6)
1190 rsh RET madvise 0
1190 rsh CALL gettimeofday(0xefbfcd98,0)
1190 rsh RET gettimeofday 0
1190 rsh CALL getpid
1190 rsh RET getpid 1190/0x4a6
1190 rsh CALL open(0x480d1237,0,0x1b6)
1190 rsh NAMI "/etc/resolv.conf"
1190 rsh RET open 3
1190 rsh CALL __fstat13(0x3,0xefbfccd8)
1190 rsh RET __fstat13 0
1190 rsh CALL read(0x3,0x8050000,0x2000)
1190 rsh GIO fd 3 read 48 bytes
"search HEH.Uni-Oldenburg.DE
nameserver 10.2.0.1
"
1190 rsh RET read 48/0x30
1190 rsh CALL read(0x3,0x8050000,0x2000)
1190 rsh GIO fd 3 read 0 bytes
""
1190 rsh RET read 0
1190 rsh CALL close(0x3)
1190 rsh RET close 0
1190 rsh CALL madvise(0x8050000,0x2000,0x6)
1190 rsh RET madvise 0
1190 rsh CALL socket(0x2,0x2,0)
1190 rsh RET socket 3
1190 rsh CALL connect(0x3,0x480e18a0,0x10)
1190 rsh RET connect 0
1190 rsh CALL sendto(0x3,0xefbfc9c0,0x2f,0,0,0)
1190 rsh GIO fd 3 wrote 47 bytes
"gL\^A\0\0\^A\0\0\0\0\0\0\bseti-100\^CHEH\rUni-Oldenburg\^BDE\0\0\^\\0\
\^A"
1190 rsh RET sendto 47/0x2f
1190 rsh CALL poll(0xefbfc778,0x1,0x1388)
1190 rsh RET poll 1
1190 rsh CALL recvfrom(0x3,0xefbfd25c,0x400,0,0xefbfc78c,0xefbfc774)
1190 rsh GIO fd 3 read 115 bytes
"gL\M^E\M^C\0\^A\0\0\0\^A\0\0\bseti-100\^CHEH\rUni-Oldenburg\^BDE\0\0\
\^\\0\^A\^CHEH\rUni-Oldenburg\^BDE\0\0\^F\0\^A\0\^AQ\M^@\0$\^FServer\
\M-@/\^Droot\M-@Ow5\M-b\M-U\0\08@\0\0\^N\^P\0 :\M^@\0\^AQ\M^@"
1190 rsh RET recvfrom 115/0x73
1190 rsh CALL close(0x3)
1190 rsh RET close 0
1190 rsh CALL socket(0x2,0x2,0)
1190 rsh RET socket 3
1190 rsh CALL connect(0x3,0x480e18a0,0x10)
1190 rsh RET connect 0
1190 rsh CALL sendto(0x3,0xefbfc9c0,0x1a,0,0,0)
1190 rsh GIO fd 3 wrote 26 bytes
"gM\^A\0\0\^A\0\0\0\0\0\0\bseti-100\0\0\^\\0\^A"
1190 rsh RET sendto 26/0x1a
1190 rsh CALL poll(0xefbfc778,0x1,0x1388)
1190 rsh RET poll 1
1190 rsh CALL recvfrom(0x3,0xefbfd25c,0x400,0,0xefbfc78c,0xefbfc774)
1190 rsh GIO fd 3 read 99 bytes
"gM\M^A\M^C\0\^A\0\0\0\^A\0\0\bseti-100\0\0\^\\0\^A\0\0\^F\0\^A\0\0"N\0\
>\^AA\fROOT-SERVERS\^CNET\0
hostmaster\binternic\M-@4w5\M-g4\0\0\a\b\0\0\^C\M^D\0 :\M^@\0\^AQ\
\M^@"
1190 rsh RET recvfrom 99/0x63
1190 rsh CALL close(0x3)
1190 rsh RET close 0
I hope this is enough information :-)
>How-To-Repeat:
"tcpdump -vv -i <interface> port domain" in one shell
and "rlogin <somehost>" etc in another one.
rlogin host
>Fix:
>Release-Note:
>Audit-Trail:
From: itojun@iijlab.net
To: Thilo.Manske@HEH.Uni-Oldenburg.DE
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: lib/9413: resolver weirdnesses
Date: Mon, 14 Feb 2000 11:39:48 +0900
>>Synopsis: strange problems with NetBSD's resolver
>I've discovered strange actions by NetBSD's resolver:
>When I issue rlogin seti-100 NetBSD does this:
>HEH.Uni-Oldenburg.de is my defaultdomain, set in /etc/resolv.conf (search
>option))
>What's strange is, that
>A) seti-100 is in /etc/hosts (and nsswitch.conf set properly)
>B) I don't have IPv6 enabled, so I'm not intereseted in "AAAA" records
Sorry, this is (one of) known problem.
There are multiple reasons on it.
Dual stack ready userland programs (like rsh/rlogin as you've
experiencded) uses get{addr,name}info, not gethostby*, to query names.
get{addr,name}info calls gethostbyname2 twice, once for AAAA and
once for A (since there's no other good primitive there).
This result in:
AAAA query on /etc/hosts
AAAA query on DNS
A query on /etc/hosts
A query on DNS
Also, (due to binary distribution and other reasons) current userland
programs cannot/do not change benavior based on IPv4/v6 kernel support.
One way to address this problem is modify resolver internal, and
provide primitive function that does
AAAA query on /etc/hosts
A query on /etc/hosts
AAAA query on DNS
A query on DNS
when you specify "hosts: files dns" in nsswitch.conf.
Current internal primitives do not do this for me.
Another way (which can coexist with the former) is to implement
part of post-RFC2553 getaddrinfo specification (AI_ADDRCONFIG flag),
and use it from userland programs. This allows get{addr,name}info
to behave differently based on kernel supported address families.
I'll try to explore both sooner.
>C) Some tools do this (rlogin, telnet, ftp, rsh, tcpdump...) some not
>(i.e. work correctly; ping and ntpdate e.g. - and host of course (doesn't
>use NetBSD'S resolver, right?))
programs which work "correctly" are not dual stack ready (IPv4-only),
and uses gethostby*.
itojun
Responsible-Changed-From-To: lib-bug-people->itojun
Responsible-Changed-By: itojun
Responsible-Changed-When: Sun Feb 13 23:27:30 PST 2000
Responsible-Changed-Why:
mine
From: Juergen Hannken-Illjes <hannken@eis.cs.tu-bs.de>
To: Thilo.Manske@heh.uni-oldenburg.de
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: lib/9413: resolver weirdnesses
Date: Mon, 14 Feb 2000 10:01:54 +0100 (MET)
The cause is: rcmd -> orcmd_af() -> getaddrinfo().
Our libc is always INET6 enabled.
I use ::ffff:10.x.x.x records to bypass these lookups.
> >Number: 9413
> >Category: lib
> >Synopsis: strange problems with NetBSD's resolver
> ...
--
Juergen Hannken-Illjes - hannken@eis.cs.tu-bs.de - TU Braunschweig (Germany)
From: Thilo Manske <Thilo.Manske@HEH.Uni-Oldenburg.DE>
To: itojun@iijlab.net
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: lib/9413: resolver weirdnesses
Date: Mon, 14 Feb 2000 13:05:04 +0100
On Mon, Feb 14, 2000 at 11:39:48AM +0900, itojun@iijlab.net wrote:
> Dual stack ready userland programs (like rsh/rlogin as you've
> experiencded) uses get{addr,name}info, not gethostby*, to query names.
> get{addr,name}info calls gethostbyname2 twice, once for AAAA and
> once for A (since there's no other good primitive there).
> This result in:
> AAAA query on /etc/hosts
> AAAA query on DNS
> A query on /etc/hosts
> A query on DNS
Oh, yes. Now I remember a post (IIRC from you) about this. Maybe this
should be documented in the BUGS section of nsswitch.conf?
And thanks to Juergen for the workaround!
--
Dies ist Thilos Unix Signature! Viel Spass damit.
From: itojun@iijlab.net
To: Thilo Manske <Thilo.Manske@HEH.Uni-Oldenburg.DE>
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: lib/9413: resolver weirdnesses
Date: Mon, 14 Feb 2000 21:07:48 +0900
>> Dual stack ready userland programs (like rsh/rlogin as you've
>> experiencded) uses get{addr,name}info, not gethostby*, to query names.
>> get{addr,name}info calls gethostbyname2 twice, once for AAAA and
>> once for A (since there's no other good primitive there).
>> This result in:
>> AAAA query on /etc/hosts
>> AAAA query on DNS
>> A query on /etc/hosts
>> A query on DNS
>Oh, yes. Now I remember a post (IIRC from you) about this. Maybe this
>should be documented in the BUGS section of nsswitch.conf?
>And thanks to Juergen for the workaround!
This is not just about nsswitch.conf. This is about all part of
resolver code... a big one.
anyway I should put some notes somewhere, thanks for suggestion.
itojun
From: Thilo Manske <Thilo.Manske@HEH.Uni-Oldenburg.DE>
To: itojun@iijlab.net
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: lib/9413: resolver weirdnesses
Date: Mon, 14 Feb 2000 13:58:07 +0100
On Mon, Feb 14, 2000 at 09:07:48PM +0900, itojun@iijlab.net wrote:
> >Oh, yes. Now I remember a post (IIRC from you) about this. Maybe this
> >should be documented in the BUGS section of nsswitch.conf?
> >And thanks to Juergen for the workaround!
>
> This is not just about nsswitch.conf. This is about all part of
> resolver code... a big one.
Yes, of course. But the manpage of nsswitch.conf is the first you read
if you want know how to control hostname lookups.
And if you read there:
EXAMPLES
To lookup hosts in /etc/hosts and then from the DNS, and lookup user in-
formation from NIS then files, use:
hosts: files dns
[...]
You expect it to work that way.
So I think the manpage of nsswitch.conf would be the best place.
--
Dies ist Thilos Unix Signature! Viel Spass damit.
State-Changed-From-To: open->feedback
State-Changed-By: itojun
State-Changed-When: Thu Feb 24 21:24:40 PST 2000
State-Changed-Why:
it should be okay by now, by getaddrinfo change i made couple of days before.
State-Changed-From-To: feedback->closed
State-Changed-By: itojun
State-Changed-When: Fri Feb 25 19:29:25 PST 2000
State-Changed-Why:
confirmed by submitter
>Unformatted:
(Contact us)
$NetBSD: query-full-pr,v 1.51 2026/08/10 02:28:17 riastradh Exp $
$NetBSD: gnats_config.sh,v 1.10 2026/05/13 22:00:09 riastradh Exp $
Copyright © 1994-2026
The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.