Thomas Hühn | 1 Jan 18:41 2004
Picon

Re: Problem with history - cant fgets after article

Also sprach Russ Allbery <rra <at> stanford.edu>:

>> I haven't found much about that "cant fgets after article" in Google
>> Groups and don't know what's wrong.
>
> This sounds really familiar.  I'm sure that I've seen this before in
> news.software.nntp or on this mailing list.  I *think* this means innd has
> died; is that what you're seeing?  Is there anything in errlog?

innd dies, but usually accepts a few articles (this time one, I'm not
sure whether it sometimes accepts more) before doing so.

>> In the meantime I get even more errors:
>
>> | Dec 28 23:41:48 daphne rnews: offered
>> | <200312282208.hBSM8Nn8004028 <at> jupiter.gwalter.demon.co.uk> uucp.gnuu
>> | Dec 28 23:41:48 daphne innd: dbz not open for this history file
>> | /self/inn/db/history
>> | Dec 28 23:41:48 daphne innd: dbz not open for this history file
>> | /self/inn/db/history
>> | Dec 28 23:41:48 daphne innd: tradindexed: index inode mismatch for
>> | F7961EF57C01E9591BA1834459B07AC9
>
>> Do you have a clue what might have gong wrong?
>
> Well, that would seem to indicate that your history file is hosed.  I'm
> not sure about the tradindexed error; that particular message is sometimes
> a harmless warning.  tdx-util -A will audit your overview database and see
> if it can find any other problems.

(Continue reading)

Katsuhiro Kondou | 1 Jan 18:57 2004

Re: syslog to localhost with controlchan

In article <87u13ic39f.fsf <at> windlord.stanford.edu>,
	Russ Allbery <rra <at> stanford.edu> wrote;

} I wonder if calling:
} 
}     setlogsock [ 'unix', 'stream', 'tcp', 'udp', 'console' ];
} 
} would do what we want, and if that's portable back to the oldest version
} of Perl that we care about....  No, it looks like the array form was added
} in 5.8.0.  Hm.  We could try using that only with recent versions of Perl,
} since OS X seems to have a pretty current one.

Ah, this reminds me effective patch only for OS X, and
this works fine.

Index: control/controlchan.in
===================================================================
RCS file: /Users/kondou/news/inn/repository/inn/control/controlchan.in,v
retrieving revision 1.7
diff -u -r1.7 controlchan.in
--- controlchan.in	2002/11/21 00:03:15	1.7
+++ controlchan.in	2004/01/01 17:52:19
 <at>  <at>  -50,7 +50,7  <at>  <at> 

 if ($use_syslog) {
     eval "sub Sys::Syslog::_PATH_LOG { '/dev/log' }" if $^O eq 'dec_osf';
-    Sys::Syslog::setlogsock('unix') if $^O =~ /linux|dec_osf/;
+    Sys::Syslog::setlogsock('unix') if $^O =~ /linux|dec_osf|darwin/;
     openlog('controlchan', 'pid', $inn::syslog_facility);
 }
(Continue reading)

Katsuhiro Kondou | 1 Jan 18:57 2004

Re: syslog to localhost with controlchan

In article <87y8suc3ja.fsf <at> windlord.stanford.edu>,
	Russ Allbery <rra <at> stanford.edu> wrote;

} Hm.  For any other Unix box, I'd consider that a configuration error bad
} enough to not be worth working around in software (not the DNS part, which
} shouldn't matter, but the /etc/hosts part), but I have no idea what the
} Mac OS X defaults look like.  And maybe my feeling is wrong, given how
} much dynamic IP assignment happens these days.

As a default for Mac OS X, computer name is asked to be
input when started up first time, while IP address is
not.  That name is also set as hostname, but installation
of inn assumes, I believe, requiring a little bit higher
experience of unix.  I agree with your opinion that system
misconfiguration should not be helped by application, but
there is a case that someone like me uses inn as stand
alone server w/o dns service.  So, I think we should take
care of it.
--

-- 
Katsuhiro Kondou

Russ Allbery | 1 Jan 19:30 2004
Picon

Re: Problem with history - cant fgets after article

Thomas H=FChn <inn <at> daphne.gnuu.de> writes:
> Also sprach Russ Allbery <rra <at> stanford.edu>:

>> This sounds really familiar.  I'm sure that I've seen this before in
>> news.software.nntp or on this mailing list.  I *think* this means innd
>> has died; is that what you're seeing?  Is there anything in errlog?

> innd dies, but usually accepts a few articles (this time one, I'm not
> sure whether it sometimes accepts more) before doing so.

Try upgrading to a current STABLE snapshot from INN 2.4.0.  There was a
bug in INN that caused it to die when processing certain malformed
messages, and I'm wondering if you're running into that.

Upgrading to a STABLE snapshot should be pretty painless; just configure,
make, make update, and restart INN.

> I even get those errors if I try to "clone" the old INN on the other
> computer as much as possible. For now, I don't know what more I could
> try... perhaps just playing a bit more with the permissions of rnews?
> Could something like that trigger this error? :-(

No, if innd is dying, there's a software bug somewhere (well, it might be
hardware, but I think that's fairly unlikely).

--=20
Russ Allbery (rra <at> stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.
(Continue reading)

Russ Allbery | 1 Jan 19:32 2004
Picon

Re: syslog to localhost with controlchan

Katsuhiro Kondou <Katsuhiro_Kondou <at> isc.org> writes:

> Ah, this reminds me effective patch only for OS X, and
> this works fine.

I'll get this applied; it's a good workaround for now at least.

> Index: control/controlchan.in
> ===================================================================
> RCS file: /Users/kondou/news/inn/repository/inn/control/controlchan.in,v
> retrieving revision 1.7
> diff -u -r1.7 controlchan.in
> --- controlchan.in	2002/11/21 00:03:15	1.7
> +++ controlchan.in	2004/01/01 17:52:19
>  <at>  <at>  -50,7 +50,7  <at>  <at> 

>  if ($use_syslog) {
>      eval "sub Sys::Syslog::_PATH_LOG { '/dev/log' }" if $^O eq 'dec_osf';
> -    Sys::Syslog::setlogsock('unix') if $^O =~ /linux|dec_osf/;
> +    Sys::Syslog::setlogsock('unix') if $^O =~ /linux|dec_osf|darwin/;
>      openlog('controlchan', 'pid', $inn::syslog_facility);
>  }
>  logmsg('starting');

> But, I wonder if any unix system whose syslog does not utilize unix
> domain socket.  In another words, cannot any unix system use
> "Sys::Syslog::setlogsock('unix')"?

Solaris may not be able to; I'm not sure.  I'll experiment.  (Solaris uses
a stream device rather than a Unix domain socket.)
(Continue reading)

Katsuhiro Kondou | 2 Jan 10:12 2004

Re: syslog to localhost with controlchan

In article <874qvfldty.fsf <at> windlord.stanford.edu>,
	Russ Allbery <rra <at> stanford.edu> wrote;

} I'll get this applied; it's a good workaround for now at least.

Could you also apply to cnfsstat.in?

} Solaris may not be able to; I'm not sure.  I'll experiment.  (Solaris uses
} a stream device rather than a Unix domain socket.)

IIRC(when I administrated biglobe's server over two
years ago), it somewhat used inet domain.
--

-- 
Katsuhiro Kondou

Steve Youngs | 2 Jan 15:40 2004
X-Face
Picon

Re: Problem with history - cant fgets after article

|--==> "RA" == Russ Allbery <rra <at> stanford.edu> writes:
  RA> Try upgrading to a current STABLE snapshot from INN 2.4.0.
  RA> There was a bug in INN that caused it to die when processing
  RA> certain malformed messages, and I'm wondering if you're running
  RA> into that.

Hi Russ!

I, too, have had the "can't fgets after article" problem for ages
now.  Today I built/installed inn-STABLE-20040101, and yes, it _still_
has the problem.

I can trigger it every time without fail with `rnews -U'.

Here's a backtrace from rnews:

GNU gdb 5.2
Copyright 2002 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB.  Type "show warranty" for details.
This GDB was configured as "i386-slackware-linux"...
(gdb) run -U
Starting program: /usr/local/news/bin/rnews -U

Program received signal SIGPIPE, Broken pipe.
0x40101708 in write () from /lib/libc.so.6
(gdb) bt
#0  0x40101708 in write () from /lib/libc.so.6
(Continue reading)

Roxik | 2 Jan 16:19 2004

Problem with PGP


 Hello

 I installed inn 2.4.0+20030912
 It works great, but I have problems with PGP signed control message:

 At last I see this:
 Jan  1 06:21:00 news pgpverify[2680]: gpg exited status 2 processing
 <checkgroups-pl-2004.Jan.01-062010 <at> usenet.pl>
 Jan  1 06:21:00 news controlchan[10967]: skipping checkgroups
 newgroup <at> usenet.pl(pgpverify failed) in
 <checkgroups-pl-2004.Jan.01-062010 <at> usenet.pl>

 It's a same situation with newgroup and rmgroup.
 I looked up into this message and it will be ok.

 I use gpg:
 gpg (GnuPG) 1.2.3
 Supported algorithms:
 Pubkey: RSA, RSA-E, RSA-S, ELG-E, DSA, ELG
 Cipher: 3DES, CAST5, BLOWFISH, AES, AES192, AES256, TWOFISH
 Hash: MD5, SHA1, RIPEMD160, SHA256
 Compression: Uncompressed, ZIP, ZLIB

 p.s.1
 This message:
 ftp://ftp.isc.org/usenet/control/pl/pl.sci.historia.gz
 give me a similar error message. I changed mode of this group manually, but
 why "is pgp failed"?
 I have'nt any other problems with normal groups
(Continue reading)

Kai Henningsen | 2 Jan 17:43 2004
Picon

Re: Problem with history - cant fgets after article

sryoungs <at> bigpond.net.au (Steve Youngs)  wrote on 03.01.04 in <microsoft-free.87oetml8hf.fsf <at> eicq.dnsalias.org>:

(...)
> #22 0x4033257e in analyze_tree () from /lib/libc.so.6
> #23 0x4033257e in analyze_tree () from /lib/libc.so.6
> #24 0x4033257e in analyze_tree () from /lib/libc.so.6
> #25 0x4033257e in analyze_tree () from /lib/libc.so.6
> #26 0x4033256c in analyze_tree () from /lib/libc.so.6
> #27 0x4033256c in analyze_tree () from /lib/libc.so.6
> #28 0x4033257e in analyze_tree () from /lib/libc.so.6
> #29 0x4033257e in analyze_tree () from /lib/libc.so.6
> #30 0x403324e8 in analyze () from /lib/libc.so.6
> #31 0x40331f19 in re_compile_internal () from /lib/libc.so.6
> #32 0x4033196c in regcomp () from /lib/libc.so.6
(...)

> My suspicion is that glibc is the culprit.  I've got this horrible
> nagging feeling that I first saw this after a libc upgrade (I'm
> currently running glibc 2.3.2).

That seems very likely: this is dying deep in the guts of regcomp().

Have you tried to rebuild with libc headers matching the .so you're using  
now?

MfG Kai

Thomas Hühn | 2 Jan 19:24 2004
Picon

Re: Problem with history - cant fgets after article

Also sprach Steve Youngs <sryoungs <at> bigpond.net.au>:

> My suspicion is that glibc is the culprit.  I've got this horrible
> nagging feeling that I first saw this after a libc upgrade (I'm
> currently running glibc 2.3.2).

I have 2.2.5 from Debian stable, FWIW.

Thomas


Gmane