In my quest to collapse an array of machines into Solaris zones, I have appplication zones connected to a firewall zone. The path between host0 and host1 is host0-zone/host0-vnic/etherstub0/fw0-vnic/firewall-zone <=> firewall-zone/fw1-vnic/etherstub1/host1-vnic/host1-zone SunOS hydra2 5.11 net-virt_xb_40_snv_101_103108 sun4u sparc sun4u stub0 etherstub 9000 unknown -- stub1 etherstub 9000 unknown -- host0 vnic 9000 up stub0 host1 vnic 9000 up stub1 gw0 vnic 9000 up stub0 gw1 vnic 9000 up stub1 My crude test for 1500 would be 1472/3, for 9000 ... ping -I 1 fred 8658 1 id OK, but 8659 fails. When I try to reflect this in the etherstub''s mtu, set-linkprop -p mtu=1500 (or any other number) fails If the path from the firewall-zone is via a real interface, all is well since mtu=1500. I''ll have a look at the code, but I''d guess there''s an offset or length arithmetic error. BTW I did remove IPF from the mix when testing. Now to update IPF ! pjc, enjoying the fruits of the XB team''s efforts. -- This message posted from opensolaris.org
Re: Etherstub MTU seems to be >9000 ^h^h^h^h^h < 9000 mea culpa, pjc -- This message posted from opensolaris.org
Peter J. Cherny
2008-Nov-12 01:27 UTC
[crossbow-discuss] Etherstub MTU seems to be < 9000 Q&D fix
ifconfig xxx mtu 8986 on the vnic endpoints fixes the problem ... 14 bytes ??? -- This message posted from opensolaris.org
Boyd Adamson
2008-Nov-12 01:54 UTC
[crossbow-discuss] Etherstub MTU seems to be < 9000 Q&D fix
"Peter J. Cherny" <peterc at luddite.com.au> writes:> ifconfig xxx mtu 8986 on the vnic endpoints > fixes the problem ... 14 bytes ???Target MAC address (6 bytes) + Source MAC address (6 bytes) + Type (2 bytes) ? Certainly these fields shouldn''t be counted in the MTU, but perhaps it''s where they come from. Boyd
Hi Peter, You''re hitting a known issue (6766660). The fix was integrated in our project gate last week and will be in the next set of bits. Nicolas. On Nov 11, 2008, at 4:42 PM, Peter J. Cherny wrote:> In my quest to collapse an array of machines into Solaris zones, > I have appplication zones connected to a firewall zone. > The path between host0 and host1 is > host0-zone/host0-vnic/etherstub0/fw0-vnic/firewall-zone <=> > firewall-zone/fw1-vnic/etherstub1/host1-vnic/host1-zone > > SunOS hydra2 5.11 net-virt_xb_40_snv_101_103108 sun4u sparc sun4u > stub0 etherstub 9000 unknown -- > stub1 etherstub 9000 unknown -- > host0 vnic 9000 up stub0 > host1 vnic 9000 up stub1 > gw0 vnic 9000 up stub0 > gw1 vnic 9000 up stub1 > > My crude test for 1500 would be 1472/3, for 9000 ... > ping -I 1 fred 8658 1 id OK, but 8659 fails. > > When I try to reflect this in the etherstub''s mtu, > set-linkprop -p mtu=1500 (or any other number) fails > > If the path from the firewall-zone is via a real interface, > all is well since mtu=1500. > > I''ll have a look at the code, but I''d guess there''s an offset > or length arithmetic error. > BTW I did remove IPF from the mix when testing. > > Now to update IPF ! > > pjc, enjoying the fruits of the XB team''s efforts. > -- > This message posted from opensolaris.org > _______________________________________________ > crossbow-discuss mailing list > crossbow-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss-- Nicolas Droux - Solaris Kernel Networking - Sun Microsystems, Inc. nicolas.droux at sun.com - http://blogs.sun.com/droux
Hi Nicolas, Is it possible to see the issue description? I suspect I am hitting the same problem. I tried searching OpenSolaris bug database with the issue ID of 6766660 but nothing showed up. Thank you, Vasiliy Nicolas Droux wrote:> Hi Peter, > > You''re hitting a known issue (6766660). The fix was integrated in our > project gate last week and will be in the next set of bits. > > Nicolas. > > On Nov 11, 2008, at 4:42 PM, Peter J. Cherny wrote: > > >> In my quest to collapse an array of machines into Solaris zones, >> I have appplication zones connected to a firewall zone. >> The path between host0 and host1 is >> host0-zone/host0-vnic/etherstub0/fw0-vnic/firewall-zone <=> >> firewall-zone/fw1-vnic/etherstub1/host1-vnic/host1-zone >> >> SunOS hydra2 5.11 net-virt_xb_40_snv_101_103108 sun4u sparc sun4u >> stub0 etherstub 9000 unknown -- >> stub1 etherstub 9000 unknown -- >> host0 vnic 9000 up stub0 >> host1 vnic 9000 up stub1 >> gw0 vnic 9000 up stub0 >> gw1 vnic 9000 up stub1 >> >> My crude test for 1500 would be 1472/3, for 9000 ... >> ping -I 1 fred 8658 1 id OK, but 8659 fails. >> >> When I try to reflect this in the etherstub''s mtu, >> set-linkprop -p mtu=1500 (or any other number) fails >> >> If the path from the firewall-zone is via a real interface, >> all is well since mtu=1500. >> >> I''ll have a look at the code, but I''d guess there''s an offset >> or length arithmetic error. >> BTW I did remove IPF from the mix when testing. >> >> Now to update IPF ! >> >> pjc, enjoying the fruits of the XB team''s efforts. >> -- >> This message posted from opensolaris.org >> _______________________________________________ >> crossbow-discuss mailing list >> crossbow-discuss at opensolaris.org >> http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss >> > >
On Nov 13, 2008, at 1:08 AM, Vasiliy Baranov wrote:> Hi Nicolas, > > Is it possible to see the issue description? I suspect I am hitting > the > same problem. I tried searching OpenSolaris bug database with the > issue > ID of 6766660 but nothing showed up.Hi Vasily, If you''re seeing problems with thelarge packets between VNICs defined on top of the same NIC this is probably caused by the same issue. The problem can be seen with large ICMP packets as described by Peter, or even TCP traffic. The regression was introduced in our build 42 but the fix unfortunately didn''t make it in that build. We''re hoping to send out new bits with that fix in the few next days. Nicolas.> > > Thank you, > Vasiliy > > Nicolas Droux wrote: >> Hi Peter, >> >> You''re hitting a known issue (6766660). The fix was integrated in our >> project gate last week and will be in the next set of bits. >> >> Nicolas. >> >> On Nov 11, 2008, at 4:42 PM, Peter J. Cherny wrote: >> >> >>> In my quest to collapse an array of machines into Solaris zones, >>> I have appplication zones connected to a firewall zone. >>> The path between host0 and host1 is >>> host0-zone/host0-vnic/etherstub0/fw0-vnic/firewall-zone <=> >>> firewall-zone/fw1-vnic/etherstub1/host1-vnic/host1-zone >>> >>> SunOS hydra2 5.11 net-virt_xb_40_snv_101_103108 sun4u sparc sun4u >>> stub0 etherstub 9000 unknown -- >>> stub1 etherstub 9000 unknown -- >>> host0 vnic 9000 up stub0 >>> host1 vnic 9000 up stub1 >>> gw0 vnic 9000 up stub0 >>> gw1 vnic 9000 up stub1 >>> >>> My crude test for 1500 would be 1472/3, for 9000 ... >>> ping -I 1 fred 8658 1 id OK, but 8659 fails. >>> >>> When I try to reflect this in the etherstub''s mtu, >>> set-linkprop -p mtu=1500 (or any other number) fails >>> >>> If the path from the firewall-zone is via a real interface, >>> all is well since mtu=1500. >>> >>> I''ll have a look at the code, but I''d guess there''s an offset >>> or length arithmetic error. >>> BTW I did remove IPF from the mix when testing. >>> >>> Now to update IPF ! >>> >>> pjc, enjoying the fruits of the XB team''s efforts. >>> -- >>> This message posted from opensolaris.org >>> _______________________________________________ >>> crossbow-discuss mailing list >>> crossbow-discuss at opensolaris.org >>> http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss >>> >> >> > > _______________________________________________ > crossbow-discuss mailing list > crossbow-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss-- Nicolas Droux - Solaris Kernel Networking - Sun Microsystems, Inc. nicolas.droux at sun.com - http://blogs.sun.com/droux
Nicolas Droux wrote:> On Nov 13, 2008, at 1:08 AM, Vasiliy Baranov wrote: > > >> Hi Nicolas, >> >> Is it possible to see the issue description? I suspect I am hitting >> the >> same problem. I tried searching OpenSolaris bug database with the >> issue >> ID of 6766660 but nothing showed up. >> > > Hi Vasily, > > If you''re seeing problems with thelarge packets between VNICs defined > on top of the same NIC this is probably caused by the same issue. The > problem can be seen with large ICMP packets as described by Peter, or > even TCP traffic. The regression was introduced in our build 42 but > the fix unfortunately didn''t make it in that build. > > We''re hoping to send out new bits with that fix in the few next days. > > Nicolas. >Nicolas, Thank you. Yes, I am running a number of guests while booted in xVM and if I try transmitting anything large between a guest and dom0 it just hangs, although short pings go fine in both directions and I can transmit anything between a guest and other hosts without any problems. ifconfig eth0 mtu 1486 within the guest fixes the problem for me. Thanks, Vasiliy> >> Thank you, >> Vasiliy >> >> Nicolas Droux wrote: >> >>> Hi Peter, >>> >>> You''re hitting a known issue (6766660). The fix was integrated in our >>> project gate last week and will be in the next set of bits. >>> >>> Nicolas. >>> >>> On Nov 11, 2008, at 4:42 PM, Peter J. Cherny wrote: >>> >>> >>> >>>> In my quest to collapse an array of machines into Solaris zones, >>>> I have appplication zones connected to a firewall zone. >>>> The path between host0 and host1 is >>>> host0-zone/host0-vnic/etherstub0/fw0-vnic/firewall-zone <=> >>>> firewall-zone/fw1-vnic/etherstub1/host1-vnic/host1-zone >>>> >>>> SunOS hydra2 5.11 net-virt_xb_40_snv_101_103108 sun4u sparc sun4u >>>> stub0 etherstub 9000 unknown -- >>>> stub1 etherstub 9000 unknown -- >>>> host0 vnic 9000 up stub0 >>>> host1 vnic 9000 up stub1 >>>> gw0 vnic 9000 up stub0 >>>> gw1 vnic 9000 up stub1 >>>> >>>> My crude test for 1500 would be 1472/3, for 9000 ... >>>> ping -I 1 fred 8658 1 id OK, but 8659 fails. >>>> >>>> When I try to reflect this in the etherstub''s mtu, >>>> set-linkprop -p mtu=1500 (or any other number) fails >>>> >>>> If the path from the firewall-zone is via a real interface, >>>> all is well since mtu=1500. >>>> >>>> I''ll have a look at the code, but I''d guess there''s an offset >>>> or length arithmetic error. >>>> BTW I did remove IPF from the mix when testing. >>>> >>>> Now to update IPF ! >>>> >>>> pjc, enjoying the fruits of the XB team''s efforts. >>>> -- >>>> This message posted from opensolaris.org >>>> _______________________________________________ >>>> crossbow-discuss mailing list >>>> crossbow-discuss at opensolaris.org >>>> http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss >>>> >>>> >>> >> _______________________________________________ >> crossbow-discuss mailing list >> crossbow-discuss at opensolaris.org >> http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss >> > >