| Message ID | 1309518770-8547-1-git-send-email-equinox@diac24.net |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
On Fri, 1 Jul 2011 13:12:50 +0200 David Lamparter <equinox@diac24.net> wrote: > this first and foremost fixes handling of bonding frames, which were > incorrectly forwarded until now. they need to never cross a bridge. > > it also introduces a new switch to control handling of the other > not-that-special groups; if you want them forwarded despite having > STP running, there's a sysfs knob for that. you can implement your > local policy with ebtables then. > > in the end, we now match hardware switch behaviour rather closely, but > still additionally allow playing tricks on things like 802.1X. > > Signed-off-by: David Lamparter <equinox@diac24.net> > Cc: Stephen Hemminger <shemminger@linux-foundation.org> > Cc: Nick Carter <ncarter100@gmail.com> Forwarding pause frames is wrong. I wonder if the best solution for this crap is to just write a userland program to do the forwarding. -- 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
On Fri, Jul 01, 2011 at 09:26:12AM -0700, Stephen Hemminger wrote: > On Fri, 1 Jul 2011 13:12:50 +0200 > David Lamparter <equinox@diac24.net> wrote: > > > this first and foremost fixes handling of bonding frames, which were > > incorrectly forwarded until now. they need to never cross a bridge. > > > > it also introduces a new switch to control handling of the other > > not-that-special groups; if you want them forwarded despite having > > STP running, there's a sysfs knob for that. you can implement your > > local policy with ebtables then. > > > > in the end, we now match hardware switch behaviour rather closely, but > > still additionally allow playing tricks on things like 802.1X. > > > > Signed-off-by: David Lamparter <equinox@diac24.net> > > Cc: Stephen Hemminger <shemminger@linux-foundation.org> > > Cc: Nick Carter <ncarter100@gmail.com> > > Forwarding pause frames is wrong. None of the patches discussed forwards pause frames. > I wonder if the best solution for this crap is to just write > a userland program to do the forwarding. You can't do that without moving the remaining STP bits to userspace, since if you want to keep STP in-kernel, you still need some policy. Also, there is a fundamental conflict between a working bridge and the desire to work as fully transparent L2 tap. As long as we forward 802.3ad/bonding frames, we are a broken bridge. Yet we still want that for the tap case. Plus, we don't need the userspace daemon if we can set the policy with ebtables - which we can do if and only if we allow stripping down the built-in restrictions. I think the variant that I suggested to Michał, with a 3-value knob "drop it if STP" / "forward except pause/bond" / "forward all" is the best way to go. It leaves the default usable but allows controlling everything through ebtables. -David -- 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/net/bridge/br_device.c b/net/bridge/br_device.c index c188c80..4a01a3e 100644 --- a/net/bridge/br_device.c +++ b/net/bridge/br_device.c @@ -359,6 +359,7 @@ void br_dev_setup(struct net_device *dev) memcpy(br->group_addr, br_group_address, ETH_ALEN); br->stp_enabled = BR_NO_STP; + br->stp_forward_802local = 0; br->designated_root = br->bridge_id; br->bridge_max_age = br->max_age = 20 * HZ; br->bridge_hello_time = br->hello_time = 2 * HZ; diff --git a/net/bridge/br_input.c b/net/bridge/br_input.c index f3ac1e8..0060b10 100644 --- a/net/bridge/br_input.c +++ b/net/bridge/br_input.c @@ -160,13 +160,24 @@ rx_handler_result_t br_handle_frame(struct sk_buff **pskb) p = br_port_get_rcu(skb->dev); if (unlikely(is_link_local(dest))) { - /* Pause frames shouldn't be passed up by driver anyway */ - if (skb->protocol == htons(ETH_P_PAUSE)) - goto drop; - - /* If STP is turned off, then forward */ - if (p->br->stp_enabled == BR_NO_STP && dest[5] == 0) - goto forward; + switch (dest[5]) { + /* Pause/3x frames shouldn't be passed up by driver anyway + * LACP/3ad can never be allowed to cross even a dumb hub + */ + case 0x01: /* pause */ + case 0x02: /* bonding */ + break; + + case 0x00: /* STP */ + if (p->br->stp_enabled == BR_NO_STP) + goto forward; + break; + + default: + if (p->br->stp_enabled == BR_NO_STP + || p->br->stp_forward_802local) + goto forward; + } if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN, skb, skb->dev, NULL, br_handle_local_finish)) { diff --git a/net/bridge/br_private.h b/net/bridge/br_private.h index 54578f2..909814f 100644 --- a/net/bridge/br_private.h +++ b/net/bridge/br_private.h @@ -213,6 +213,8 @@ struct net_bridge BR_USER_STP, /* new RSTP in userspace */ } stp_enabled; + bool stp_forward_802local; + unsigned char topology_change; unsigned char topology_change_detected; diff --git a/net/bridge/br_sysfs_br.c b/net/bridge/br_sysfs_br.c index 68b893e..8004bf7 100644 --- a/net/bridge/br_sysfs_br.c +++ b/net/bridge/br_sysfs_br.c @@ -149,6 +149,29 @@ static ssize_t store_stp_state(struct device *d, static DEVICE_ATTR(stp_state, S_IRUGO | S_IWUSR, show_stp_state, store_stp_state); +static ssize_t show_stp_forward_802local(struct device *d, + struct device_attribute *attr, + char *buf) +{ + struct net_bridge *br = to_bridge(d); + return sprintf(buf, "%d\n", br->stp_forward_802local); +} + +static int set_stp_forward_802local(struct net_bridge *br, unsigned long val) +{ + br->stp_forward_802local = !!val; + return 0; +} + +static ssize_t store_stp_forward_802local(struct device *d, + struct device_attribute *attr, + const char *buf, size_t len) +{ + return store_bridge_parm(d, buf, len, set_stp_forward_802local); +} +static DEVICE_ATTR(stp_forward_802local, S_IRUGO | S_IWUSR, + show_stp_forward_802local, store_stp_forward_802local); + static ssize_t show_priority(struct device *d, struct device_attribute *attr, char *buf) { @@ -652,6 +675,7 @@ static struct attribute *bridge_attrs[] = { &dev_attr_max_age.attr, &dev_attr_ageing_time.attr, &dev_attr_stp_state.attr, + &dev_attr_stp_forward_802local.attr, &dev_attr_priority.attr, &dev_attr_bridge_id.attr, &dev_attr_root_id.attr,
this first and foremost fixes handling of bonding frames, which were incorrectly forwarded until now. they need to never cross a bridge. it also introduces a new switch to control handling of the other not-that-special groups; if you want them forwarded despite having STP running, there's a sysfs knob for that. you can implement your local policy with ebtables then. in the end, we now match hardware switch behaviour rather closely, but still additionally allow playing tricks on things like 802.1X. Signed-off-by: David Lamparter <equinox@diac24.net> Cc: Stephen Hemminger <shemminger@linux-foundation.org> Cc: Nick Carter <ncarter100@gmail.com> --- net/bridge/br_device.c | 1 + net/bridge/br_input.c | 25 ++++++++++++++++++------- net/bridge/br_private.h | 2 ++ net/bridge/br_sysfs_br.c | 24 ++++++++++++++++++++++++ 4 files changed, 45 insertions(+), 7 deletions(-)