DragonFlyBSD

mh I did some more testing

tcpdump -ttt -vvv -i fxp0 shows the usual things like AC-Name,
AC-Cookie, etc.

but tcpdump -ttt -vvv -i tun0 shows nothing

so what's wrong? netgraph?

On 5/24/06, Stefan Krüger <skrueger@meinberlikomm.de> wrote:

hi folks,

I can't connect to the Internet using ppp and pppoe (-current, sources
updated around May 21)

I did a full buildworld/buildkernel, I did not change any
configuration files (ppp.conf etc) and it's a mystery to me why ppp
can't establish a connection (there isn't any interessting in
/var/log/ppp.log)

Can you post a `boot -v' dmesg?

Best Regards,
sephe

sure,

http://tecneeq.dyndns.org/~cosmicdj/dmesg.txt

I've also logged the "killed my pppoe connection"-cvsup run, it's
available at:

http://tecneeq.dyndns.org/~cosmicdj/src-20060522

On Sun, May 28, 2006 at 07:00:51PM +0100, Stefan Kr?ger wrote:

I've also logged the "killed my pppoe connection"-cvsup run, it's
available at:

http://tecneeq.dyndns.org/~cosmicdj/src-20060522

So if you checkout the source with date=2006.05.20.06.32.00
and reinstall the kernel, does pppoe connection come back?

YONETANI Tomokazu wrote:

I've also logged the "killed my pppoe connection"-cvsup run, it's
available at:

http://tecneeq.dyndns.org/~cosmicdj/src-20060522

So if you checkout the source with date=2006.05.20.06.32.00
and reinstall the kernel, does pppoe connection come back?

yes, it's back now (note: I also rebuilt userland)

related (IMHO) commits made between 20. May and 22. May:

http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00108.html
the big #include cleanup, maybe unrelated

http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00113.html
netmsg embeded in the mbuf, hhhmmm... no idea

http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00116.html
"Fix a build issue with libnetgraph" well pppoe uses netgraph...

http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00120.html
another big #include cleanup

unfortunately most of them are quite "big", touching many files etc; so
any help debugging this further appreciated

:YONETANI Tomokazu wrote:
:>> I've also logged the "killed my pppoe connection"-cvsup run, it's
:>> available at:
:>>
:>> http://tecneeq.dyndns.org/~cosmicdj/src-20060522
:>
:> So if you checkout the source with date=2006.05.20.06.32.00
:> and reinstall the kernel, does pppoe connection come back?
:
:yes, it's back now (note: I also rebuilt userland)
:
:related (IMHO) commits made between 20. May and 22. May:
:....
:
:http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00113.html
:netmsg embeded in the mbuf, hhhmmm... no idea
:
:unfortunately most of them are quite "big", touching many files etc; so
:any help debugging this further appreciated

It's probably this one, I'll take a look at it as soon as I get the
    scheduler bug fixed.

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>

On Mon, Jun 05, 2006 at 01:31:04PM -0700, Matthew Dillon wrote:

:http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00113.html
:netmsg embeded in the mbuf, hhhmmm... no idea
:
:unfortunately most of them are quite "big", touching many files etc; so
:any help debugging this further appreciated

It's probably this one, I'll take a look at it as soon as I get the
scheduler bug fixed.

Ok, here's a bandaid until Matt makes his mind on how to fix this issue
(other "do not reply!" comments may need similar bandaids depending
on your network configuration):

Index: ng_base.c ===================================================================
RCS file: /home/source/dragonfly/cvs/src/sys/netgraph/netgraph/ng_base.c,v
retrieving revision 1.20
diff u -r1.20 ng_base.c
-- ng_base.c 1 Jun 2006 06:10:54 0000 1.20
++ ng_base.c 6 Jun 2006 11:11:36 -0000
@ -2087,7 +2087,9 @
}
}
out:
/* pmsg was embedded in the mbuf, do not reply! /
/ netgraph code may receive non embedded-in-mbuf netmsg's */
+ if (pmsg != &m->mh_netmsg.nm_lmsg)
+ lwkt_replymsg(pmsg->nm_lmsg, 0);
return(EASYNC);
}

On Tue, Jun 06, 2006 at 08:26:06PM +0900, YONETANI Tomokazu wrote:

On Mon, Jun 05, 2006 at 01:31:04PM -0700, Matthew Dillon wrote:

:http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00113.html
:netmsg embeded in the mbuf, hhhmmm... no idea
:
:unfortunately most of them are quite "big", touching many files etc; so
:any help debugging this further appreciated

It's probably this one, I'll take a look at it as soon as I get the
scheduler bug fixed.

Ok, here's a bandaid until Matt makes his mind on how to fix this issue
(other "do not reply!" comments may need similar bandaids depending
on your network configuration):

Index: ng_base.c

Oops, copy&paste error...

--- sys/netgraph/netgraph/ng_base.c Sat Jun 03 08:06:31 2006 0000
++ sys/netgraph/netgraph/ng_base.c Tue Jun 06 22:55:38 2006 0900
@ -2087,7 +2087,9 @ ngintr(struct netmsg pmsg)
}
}
out:
- / pmsg was embedded in the mbuf, do not reply! /
/ netgraph code may receive non embedded-in-mbuf netmsg's */
+ if (&pmsg->nm_lmsg != &m->m_hdr.mh_netmsg.nm_lmsg)
+ lwkt_replymsg(&pmsg->nm_lmsg, 0);
return(EASYNC);
}

YONETANI Tomokazu wrote:

On Tue, Jun 06, 2006 at 08:26:06PM +0900, YONETANI Tomokazu wrote:

On Mon, Jun 05, 2006 at 01:31:04PM -0700, Matthew Dillon wrote:

:http://leaf.dragonflybsd.org/mailarchive/commits/2006-05/msg00113.html
:netmsg embeded in the mbuf, hhhmmm... no idea
:
:unfortunately most of them are quite "big", touching many files etc; so
:any help debugging this further appreciated

It's probably this one, I'll take a look at it as soon as I get the
scheduler bug fixed.

Ok, here's a bandaid until Matt makes his mind on how to fix this issue
(other "do not reply!" comments may need similar bandaids depending
on your network configuration):

Index: ng_base.c

Oops, copy&paste error...

--- sys/netgraph/netgraph/ng_base.c Sat Jun 03 08:06:31 2006 0000
++ sys/netgraph/netgraph/ng_base.c Tue Jun 06 22:55:38 2006 0900
@ -2087,7 +2087,9 @ ngintr(struct netmsg pmsg)
}
}
out:
- / pmsg was embedded in the mbuf, do not reply! /
/ netgraph code may receive non embedded-in-mbuf netmsg's */
+ if (&pmsg->nm_lmsg != &m->m_hdr.mh_netmsg.nm_lmsg)
+ lwkt_replymsg(&pmsg->nm_lmsg, 0);
return(EASYNC);
}

this one works, thanks :)

:> out:
:> - /* pmsg was embedded in the mbuf, do not reply! /
:> + / netgraph code may receive non embedded-in-mbuf netmsg's */
:> + if (&pmsg->nm_lmsg != &m->m_hdr.mh_netmsg.nm_lmsg)
:> + lwkt_replymsg(&pmsg->nm_lmsg, 0);
:> return(EASYNC);
:> }
:
:this one works, thanks :)

Hmm.  It looks like I didn't take into account calls made to
    schednetisr(), whos messages do need to be replied.

I'll commit a more involved fix.  It is going to be a little messier
    then I would like because messages embedded in a mbuf can't be checked
    after the mbuf has been freed.

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>

:> Ok, here's a bandaid until Matt makes his mind on how to fix this issue
:> (other "do not reply!" comments may need similar bandaids depending
:> on your network configuration):
:>
:> Index: ng_base.c

It turned out not be as messy as I though it would be.  I'll commit
    a fix soon.

Both PPP and NETGRAPH needed the fix.

-Matt