Message ID | m1ocwfsgw7.fsf@fess.ebiederm.org |
---|---|
State | Accepted, archived |
Delegated to: | David Miller |
Headers | show |
Eric W. Biederman wrote: > When running in a network namespace whose only link to > the outside world is a macvlan device, not being > able to create another macvlan is a real pain. > > So modify macvlan creation to allow automatically forward > a creation of a macvlan on a macvlan to become a creation > of a macvlan on the underlying network device. I'm not sure I understand the constallation, what is the underlying device in this case? A device outside the namespace? -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Patrick McHardy <kaber@trash.net> writes: > Eric W. Biederman wrote: >> When running in a network namespace whose only link to >> the outside world is a macvlan device, not being >> able to create another macvlan is a real pain. >> >> So modify macvlan creation to allow automatically forward >> a creation of a macvlan on a macvlan to become a creation >> of a macvlan on the underlying network device. > > I'm not sure I understand the constallation, what is the underlying > device in this case? A device outside the namespace? Yes. Typical usage would be: eth0 in the initial namespace. A macvlan off of eth0 in each child namespace. Which works fine until I do things like create a network namespace when I am already inside of a network namespace. A child of a child. In which case I have to start rigging up something like a pair of veths an bridging or routing to get outside connectivity. Or roughly: ip link add mv0 link eth0 type macvlan. ip link add mv1 link eth0 type macvlan. ip link set mv0 netns 1234 ip link set mv1 netns 6789 Then later I would find it very handy to do: echo $$ -> 1234 ip link add mv3 link mv0 type macvlan ip link set mv3 netns 101112 The fact that we only bridge traffic on the ingress from the external world is also a pain, but that isn't trivial to fix, and fixing it might possibly break valid macvlan users. Eric -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Eric W. Biederman wrote: > Patrick McHardy <kaber@trash.net> writes: > >>> So modify macvlan creation to allow automatically forward >>> a creation of a macvlan on a macvlan to become a creation >>> of a macvlan on the underlying network device. >> I'm not sure I understand the constallation, what is the underlying >> device in this case? A device outside the namespace? > > Yes. > > Typical usage would be: > > eth0 in the initial namespace. > A macvlan off of eth0 in each child namespace. > > Which works fine until I do things like create a network namespace > when I am already inside of a network namespace. A child of a child. > In which case I have to start rigging up something like a pair of > veths an bridging or routing to get outside connectivity. > > Or roughly: > ip link add mv0 link eth0 type macvlan. > ip link add mv1 link eth0 type macvlan. > ip link set mv0 netns 1234 > ip link set mv1 netns 6789 > > Then later I would find it very handy to do: > echo $$ -> 1234 > ip link add mv3 link mv0 type macvlan > ip link set mv3 netns 101112 That makes sense of course. I'm mainly wondering whether a namespace should be able to directly affect the real device like this. This might move it to promiscous mode, or affect other performce-relevant settings. Its also looks like you can steal the MAC address of a different macvlan device this way and have the packets directed to you (new devices are added to the beginning of the hash chains, so they are found first on lookups). -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Patrick McHardy <kaber@trash.net> writes: > That makes sense of course. I'm mainly wondering whether a namespace > should be able to directly affect the real device like this. This > might move it to promiscous mode, or affect other performce-relevant > settings. Its also looks like you can steal the MAC address of a > different macvlan device this way and have the packets directed to you > (new devices are added to the beginning of the hash chains, so they > are found first on lookups). To a large extent those are things that we already can do, simply by having multiple mcavlans in different network namespaces. I could push it into promiscous mode by adding more multicast listeners, and I could steal the mac address of another macvlan by changing my mac address if I happen to come first in the hash chain. Hmm. Actually that appears to be a macvlan bug. It looks like if I change the macaddress on a macvlan we don't update the hash chains. So unless we have the same low byte we will be on the wrong hash chain and not receive the packets for the mac we specified. Ouch! It is also trivial to spoof a different macvlan device by using PF_PACKET and sending packets with the source mac address of another macvlan. Also this still requires CAP_NET_ADMIN, as much as I would like to remove that restriction. Your concerns don't appear to be new to allowing the creation of a macvlan from a macvlan or fundamental to creating a macvlan from a macvlan. You still must have access to at least a macvlan in your namespace to create a new one. So I don't think those issues bear on my patch. That said I am not opposed conceptually to something that is much harder to abuse, and works better for the network namespace case. Eric -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Eric W. Biederman wrote: > Patrick McHardy <kaber@trash.net> writes: > >> That makes sense of course. I'm mainly wondering whether a namespace >> should be able to directly affect the real device like this. This >> might move it to promiscous mode, or affect other performce-relevant >> settings. Its also looks like you can steal the MAC address of a >> different macvlan device this way and have the packets directed to you >> (new devices are added to the beginning of the hash chains, so they >> are found first on lookups). > > To a large extent those are things that we already can do, simply by > having multiple mcavlans in different network namespaces. I could > push it into promiscous mode by adding more multicast listeners, > and I could steal the mac address of another macvlan by changing > my mac address if I happen to come first in the hash chain. > > Hmm. Actually that appears to be a macvlan bug. It looks like if I > change the macaddress on a macvlan we don't update the hash chains. > So unless we have the same low byte we will be on the wrong hash chain > and not receive the packets for the mac we specified. Ouch! The address can only be changed while the device is down and unhashed. > It is also trivial to spoof a different macvlan device by using > PF_PACKET and sending packets with the source mac address of > another macvlan. Yes, but that doesn't allow one namespace to deny service to a different one. > Also this still requires CAP_NET_ADMIN, as much as I would like > to remove that restriction. > > Your concerns don't appear to be new to allowing the creation > of a macvlan from a macvlan or fundamental to creating > a macvlan from a macvlan. You still must have access to at > least a macvlan in your namespace to create a new one. So > I don't think those issues bear on my patch. No, they're not, but it seemed worth pointing out. Your patch looks perfectly fine. Acked-by: Patrick McHardy <kaber@trash.net> -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Patrick McHardy <kaber@trash.net> writes: > Eric W. Biederman wrote: >> Patrick McHardy <kaber@trash.net> writes: >> >>> That makes sense of course. I'm mainly wondering whether a namespace >>> should be able to directly affect the real device like this. This >>> might move it to promiscous mode, or affect other performce-relevant >>> settings. Its also looks like you can steal the MAC address of a >>> different macvlan device this way and have the packets directed to you >>> (new devices are added to the beginning of the hash chains, so they >>> are found first on lookups). >> >> To a large extent those are things that we already can do, simply by >> having multiple mcavlans in different network namespaces. I could >> push it into promiscous mode by adding more multicast listeners, >> and I could steal the mac address of another macvlan by changing >> my mac address if I happen to come first in the hash chain. >> >> Hmm. Actually that appears to be a macvlan bug. It looks like if I >> change the macaddress on a macvlan we don't update the hash chains. >> So unless we have the same low byte we will be on the wrong hash chain >> and not receive the packets for the mac we specified. Ouch! > > The address can only be changed while the device is down and unhashed. Point. The dev_unicast/dev_unicast_delete in macvlan_set_mac_address appears to be completely unnecessary then. >> It is also trivial to spoof a different macvlan device by using >> PF_PACKET and sending packets with the source mac address of >> another macvlan. > > Yes, but that doesn't allow one namespace to deny service to > a different one. No for that I guess it seems I just need to down the interface change the mac and up the interface again. >> Also this still requires CAP_NET_ADMIN, as much as I would like >> to remove that restriction. >> >> Your concerns don't appear to be new to allowing the creation >> of a macvlan from a macvlan or fundamental to creating >> a macvlan from a macvlan. You still must have access to at >> least a macvlan in your namespace to create a new one. So >> I don't think those issues bear on my patch. > > No, they're not, but it seemed worth pointing out. Your patch > looks perfectly fine. Thanks. Would you be opposed to changes that made macvlan more robust. Such as refusing to come up if the macaddress is already in use. And perhaps denying the sending of packets with the wrong source mac? > Acked-by: Patrick McHardy <kaber@trash.net> -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Eric W. Biederman wrote: > Patrick McHardy <kaber@trash.net> writes: > >>> Hmm. Actually that appears to be a macvlan bug. It looks like if I >>> change the macaddress on a macvlan we don't update the hash chains. >>> So unless we have the same low byte we will be on the wrong hash chain >>> and not receive the packets for the mac we specified. Ouch! >> The address can only be changed while the device is down and unhashed. > > Point. The dev_unicast/dev_unicast_delete in macvlan_set_mac_address > appears to be completely unnecessary then. I think thats correct. >> No, they're not, but it seemed worth pointing out. Your patch >> looks perfectly fine. > > Thanks. > > Would you be opposed to changes that made macvlan more robust. > Such as refusing to come up if the macaddress is already in use. > And perhaps denying the sending of packets with the wrong source > mac? Refusing duplicate MACs (on one underlying device) makes sense, the results are undefined currently. About the filtering - I don't like the idea too much given that we already have multiple possiblities to do that. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Patrick McHardy <kaber@trash.net> writes: > Eric W. Biederman wrote: >> Patrick McHardy <kaber@trash.net> writes: >> >>>> Hmm. Actually that appears to be a macvlan bug. It looks like if I >>>> change the macaddress on a macvlan we don't update the hash chains. >>>> So unless we have the same low byte we will be on the wrong hash chain >>>> and not receive the packets for the mac we specified. Ouch! >>> The address can only be changed while the device is down and unhashed. >> >> Point. The dev_unicast/dev_unicast_delete in macvlan_set_mac_address >> appears to be completely unnecessary then. > > I think thats correct. Actually it is really weird. We can change the mac address while the devices is running but the code is broken because it does not update the hash table. > Refusing duplicate MACs (on one underlying device) makes sense, the > results are undefined currently. Then I will do that just for consistency. > About the filtering - I don't like > the idea too much given that we already have multiple possiblities > to do that. Agreed, and even more I can think of some reasons why would not want that. The observation I have is that if really want separation you need separate vlans to the switch. As the similar hardware based features also don't perform egress filtering. Eric -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Eric W. Biederman wrote: > Patrick McHardy <kaber@trash.net> writes: > >> Eric W. Biederman wrote: >>> Patrick McHardy <kaber@trash.net> writes: >>> >>>>> Hmm. Actually that appears to be a macvlan bug. It looks like if I >>>>> change the macaddress on a macvlan we don't update the hash chains. >>>>> So unless we have the same low byte we will be on the wrong hash chain >>>>> and not receive the packets for the mac we specified. Ouch! >>>> The address can only be changed while the device is down and unhashed. >>> Point. The dev_unicast/dev_unicast_delete in macvlan_set_mac_address >>> appears to be completely unnecessary then. >> I think thats correct. > > Actually it is really weird. We can change the mac address while > the devices is running but the code is broken because it does > not update the hash table. Thats strange. I know the assumption was that this can't be done. But I can't find anything preventing it, not even in older versions. >> Refusing duplicate MACs (on one underlying device) makes sense, the >> results are undefined currently. > > Then I will do that just for consistency. Thanks. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Eric W. Biederman wrote: > Thanks. > Would you be opposed to changes that made macvlan more robust. > Such as refusing to come up if the macaddress is already in use. > And perhaps denying the sending of packets with the wrong source > mac? > I wouldn't deny sending with wrong source mac..ethernet interfaces can do this, and mac-vlan should look as much like ethernet is possible. I'm all for making it harder to mis-configure things (like dup macs on single interface, etc). Thanks, Ben
From: Patrick McHardy <kaber@trash.net> Date: Fri, 06 Mar 2009 16:08:45 +0100 > Acked-by: Patrick McHardy <kaber@trash.net> Applied to net-next-2.6 -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/net/macvlan.c b/drivers/net/macvlan.c index 7e24b50..b5241fc 100644 --- a/drivers/net/macvlan.c +++ b/drivers/net/macvlan.c @@ -461,12 +461,13 @@ static int macvlan_newlink(struct net_device *dev, if (lowerdev == NULL) return -ENODEV; - /* Don't allow macvlans on top of other macvlans - its not really - * wrong, but lockdep can't handle it and its not useful for anything - * you couldn't do directly on top of the real device. + /* When creating macvlans on top of other macvlans - use + * the real device as the lowerdev. */ - if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) - return -ENODEV; + if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) { + struct macvlan_dev *lowervlan = netdev_priv(lowerdev); + lowerdev = lowervlan->lowerdev; + } if (!tb[IFLA_MTU]) dev->mtu = lowerdev->mtu;
When running in a network namespace whose only link to the outside world is a macvlan device, not being able to create another macvlan is a real pain. So modify macvlan creation to allow automatically forward a creation of a macvlan on a macvlan to become a creation of a macvlan on the underlying network device. Signed-off-by: Eric Biederman <ebiederm@aristanetworks.com> --- drivers/net/macvlan.c | 11 ++++++----- 1 files changed, 6 insertions(+), 5 deletions(-)