Message ID | 20090315161217.7fa2c3a7@nehalam |
---|---|
State | RFC, archived |
Delegated to: | David Miller |
Headers | show |
Mon, Mar 16, 2009 at 12:12:17AM CET, shemminger@linux-foundation.org wrote: >On Sat, 14 Mar 2009 10:49:11 +0100 >Jiri Pirko <jpirko@redhat.com> wrote: > >> Sat, Mar 14, 2009 at 06:39:32AM CET, shemminger@linux-foundation.org wrote: >> >On Fri, 13 Mar 2009 19:33:04 +0100 >> >Jiri Pirko <jpirko@redhat.com> wrote: >> > >> >> Hi all. >> >> >> >> This is only a draft of patch to consult. I'm aware that it should be divided >> >> into multiple patches. I want to know opinion from you folks. >> >> >> >> The problem is described in following bugzilla: >> >> https://bugzilla.redhat.com/show_bug.cgi?id=487763 >> >> >> >> Basically here's what's going on. In every mode, bonding interface uses the same >> >> mac address for all enslaved devices. Except for mode balance-alb. When you put >> >> this kind of bond device into a bridge it will only add one of mac adresses into >> >> a hash list of mac addresses, say X. This mac address is marked as local. But >> >> this bonding interface also has mac address Y. Now then packet arrives with >> >> destination address Y, this address is not marked as local and the packed looks >> >> like it needs to be forwarded. This packet is then lost which is wrong. >> >> >> >> Notice that interfaces can be added and removed from bond while it is in bridge. >> >> Therefore I introduce another function pointer in struct net_device_ops - >> >> ndo_check_mac_address. This function when it's implemented should check passed >> >> mac address against the one set in device. I'm using this in bonding driver when >> >> the bond is in mode balance-alb to walk thru all slaves and checking if any of >> >> them equals passed address. >> >> >> >> Then in bridge function br_handle_frame_finish() I'm using ndo_check_mac_address >> >> to recognize the destination mac address as local. >> >> >> >> Please look at this and tell me what you think about it. >> >> >> >> Thanks >> >> >> >> Jirka >> >> >> > >> >A better and more general way to do this have the dev_set_mac_address >> >function check the return of the notifier and unwind. Then any protocol >> >can easily prevent address from changing. >> >> Can you please describe this thougth a bit more? I can't understand it now... >> >> Thanks >> >> Jirka > >Something like this: > >--- a/net/core/dev.c 2009-03-15 15:55:02.098126056 -0700 >+++ b/net/core/dev.c 2009-03-15 16:02:43.999251305 -0700 >@@ -3830,6 +3830,7 @@ int dev_set_mac_address(struct net_devic > { > const struct net_device_ops *ops = dev->netdev_ops; > int err; >+ char save_addr[MAX_ADDR_LEN]; > > if (!ops->ndo_set_mac_address) > return -EOPNOTSUPP; >@@ -3837,9 +3838,17 @@ int dev_set_mac_address(struct net_devic > return -EINVAL; > if (!netif_device_present(dev)) > return -ENODEV; >+ >+ memcpy(save_addr, dev->dev_addr, dev->addr_len); > err = ops->ndo_set_mac_address(dev, sa); >- if (!err) >- call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); >+ if (err) >+ return err; >+ >+ err = call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); >+ if (err) { >+ memcpy(sa->sa_data, save_addr, dev->addr_len); >+ ops->ndo_set_mac_address(dev, sa); >+ } > return err; > } > > >And something like this: > >--- a/drivers/net/bonding/bond_main.c 2009-03-15 16:03:53.909000973 -0700 >+++ b/drivers/net/bonding/bond_main.c 2009-03-15 16:11:43.227127031 -0700 >@@ -3534,6 +3534,7 @@ static int bond_slave_netdev_event(unsig > { > struct net_device *bond_dev = slave_dev->master; > struct bonding *bond = netdev_priv(bond_dev); >+ int err; > > switch (event) { > case NETDEV_UNREGISTER: >@@ -3570,6 +3571,15 @@ static int bond_slave_netdev_event(unsig > * servitude. > */ > break; >+ case NETDEV_CHANGEADDR: >+ if (bond->params.mode == BOND_MODE_ALB) >+ err = bond_alb_check_mac_address(bond); >+ else if (compare_ether_addr(bond_dev->dev_addr, addr) != 0) >+ err = -EINVAL; >+ >+ if (err) >+ return notifier_from_errno(err); >+ break; > case NETDEV_CHANGENAME: > /* > * TODO: handle changing the primary's name > Yes, I think the changing mac address of slaves should be also handled by bonding driver. But my patch fixes a different issue. See, unlike in any other bonding modes, in balance-alb mode incoming packets have multiple MAC adresses (of any of enslaved devices). This causes problem because bridge only recognize one of them (the mac of master which is the mac on one of the slaves) as local - the other MAC's are not recognized as they are a part of port and therefore handled as general MAC adresses. This is the problem. I can see two solutions. Either like my patch or somehow allow bridge to know more MAC addressses per port (maybe netdev can be changed to know more then one MAC address). Any thoughts? Thanks Jirka > > -- 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
From: Jiri Pirko <jpirko@redhat.com> Date: Mon, 16 Mar 2009 12:11:28 +0100 > I can see two solutions. Either like my patch or somehow allow bridge to know > more MAC addressses per port (maybe netdev can be changed to know more then > one MAC address). > > Any thoughts? The netdev struct already supports having a list of multiple unicast MAC addresses, it can probably be used and inspected for this. I'll hold off on your patch until we make some more progress on this discussion. -- 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
Thu, Mar 19, 2009 at 07:20:03AM CET, davem@davemloft.net wrote: >From: Jiri Pirko <jpirko@redhat.com> >Date: Mon, 16 Mar 2009 12:11:28 +0100 > >> I can see two solutions. Either like my patch or somehow allow bridge to know >> more MAC addressses per port (maybe netdev can be changed to know more then >> one MAC address). >> >> Any thoughts? > >The netdev struct already supports having a list of multiple unicast >MAC addresses, it can probably be used and inspected for this. Yes I was looking at this thing yesterday (uc_list). But this list serves to different purpose. Do you think that it will be correct to use it for this? I would maybe like to make a new list similar to this for our purpose (say addr_list). I think it would be more correct. Eventually in the furute we would use this list as a primary place to store device address instead of dev_addr value and make it more general (as device generally may have more adresses). Just a thought... > >I'll hold off on your patch until we make some more progress on >this discussion. -- 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
David Miller wrote: > From: Jiri Pirko <jpirko@redhat.com> > Date: Mon, 16 Mar 2009 12:11:28 +0100 > >> I can see two solutions. Either like my patch or somehow allow bridge to know >> more MAC addressses per port (maybe netdev can be changed to know more then >> one MAC address). >> >> Any thoughts? > > The netdev struct already supports having a list of multiple unicast > MAC addresses, it can probably be used and inspected for this. > > I'll hold off on your patch until we make some more progress on > this discussion. From reading the balance-alb description, I get the impression that this mode is simply not meant to be used with bridging: Adaptive load balancing: includes balance-tlb plus receive load balancing (rlb) for IPV4 traffic, and does not require any special switch support. The receive load balancing is achieved by ARP negotiation. The bonding driver intercepts the ARP Replies sent by the local system on their way out and overwrites the source hardware address with the unique hardware address of one of the slaves in the bond such that different peers use different hardware addresses for the server. In any case I'd tend to say that if bond-alb mode mangles outgoing MAC addresses, it should restore the original one for received packets and keep the hacks local to bonding. -- 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
From: Jiri Pirko <jpirko@redhat.com> Date: Thu, 19 Mar 2009 09:44:45 +0100 > Yes I was looking at this thing yesterday (uc_list). But this list serves > to different purpose. Do you think that it will be correct to use it for this? I > would maybe like to make a new list similar to this for our purpose > (say addr_list). I think it would be more correct. Whatever you do with that list privately inside of the bonding driver should be fine. It might upset something in the generic code if you don't clean it up before deregistration of the bonding device, so just be tidy. -- 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
Thu, Mar 19, 2009 at 11:21:43AM CET, davem@davemloft.net wrote: >From: Jiri Pirko <jpirko@redhat.com> >Date: Thu, 19 Mar 2009 09:44:45 +0100 > >> Yes I was looking at this thing yesterday (uc_list). But this list serves >> to different purpose. Do you think that it will be correct to use it for this? I >> would maybe like to make a new list similar to this for our purpose >> (say addr_list). I think it would be more correct. > >Whatever you do with that list privately inside of the bonding >driver should be fine. Well I do not need it only inside the bonding driver. I want bridge to use this list when adding a device in it and get mac addresses from there into its hashlist (to recognize these addresses as local). > >It might upset something in the generic code if you don't clean >it up before deregistration of the bonding device, so just be >tidy. -- 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
Thu, Mar 19, 2009 at 09:50:03AM CET, kaber@trash.net wrote: > David Miller wrote: >> From: Jiri Pirko <jpirko@redhat.com> >> Date: Mon, 16 Mar 2009 12:11:28 +0100 >> >>> I can see two solutions. Either like my patch or somehow allow bridge to know >>> more MAC addressses per port (maybe netdev can be changed to know more then >>> one MAC address). >>> >>> Any thoughts? >> >> The netdev struct already supports having a list of multiple unicast >> MAC addresses, it can probably be used and inspected for this. >> >> I'll hold off on your patch until we make some more progress on >> this discussion. > > From reading the balance-alb description, I get the impression that this > mode is simply not meant to be used with bridging: > > Adaptive load balancing: includes balance-tlb plus > receive load balancing (rlb) for IPV4 traffic, and > does not require any special switch support. The > receive load balancing is achieved by ARP negotiation. > The bonding driver intercepts the ARP Replies sent by > the local system on their way out and overwrites the > source hardware address with the unique hardware > address of one of the slaves in the bond such that > different peers use different hardware addresses for > the server. > > In any case I'd tend to say that if bond-alb mode mangles outgoing MAC > addresses, it should restore the original one for received packets > and keep the hacks local to bonding. To let bonding driver to resolve this I think there will be needed some kind of hook in netif_receive_skb() as for example bridge has. I would rather do this more general and transparent. -- 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
--- a/net/core/dev.c 2009-03-15 15:55:02.098126056 -0700 +++ b/net/core/dev.c 2009-03-15 16:02:43.999251305 -0700 @@ -3830,6 +3830,7 @@ int dev_set_mac_address(struct net_devic { const struct net_device_ops *ops = dev->netdev_ops; int err; + char save_addr[MAX_ADDR_LEN]; if (!ops->ndo_set_mac_address) return -EOPNOTSUPP; @@ -3837,9 +3838,17 @@ int dev_set_mac_address(struct net_devic return -EINVAL; if (!netif_device_present(dev)) return -ENODEV; + + memcpy(save_addr, dev->dev_addr, dev->addr_len); err = ops->ndo_set_mac_address(dev, sa); - if (!err) - call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); + if (err) + return err; + + err = call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); + if (err) { + memcpy(sa->sa_data, save_addr, dev->addr_len); + ops->ndo_set_mac_address(dev, sa); + } return err; } And something like this: --- a/drivers/net/bonding/bond_main.c 2009-03-15 16:03:53.909000973 -0700 +++ b/drivers/net/bonding/bond_main.c 2009-03-15 16:11:43.227127031 -0700 @@ -3534,6 +3534,7 @@ static int bond_slave_netdev_event(unsig { struct net_device *bond_dev = slave_dev->master; struct bonding *bond = netdev_priv(bond_dev); + int err; switch (event) { case NETDEV_UNREGISTER: @@ -3570,6 +3571,15 @@ static int bond_slave_netdev_event(unsig * servitude. */ break; + case NETDEV_CHANGEADDR: + if (bond->params.mode == BOND_MODE_ALB) + err = bond_alb_check_mac_address(bond); + else if (compare_ether_addr(bond_dev->dev_addr, addr) != 0) + err = -EINVAL; + + if (err) + return notifier_from_errno(err); + break; case NETDEV_CHANGENAME: /* * TODO: handle changing the primary's name