bridge: Return an error when enabling STP in netns.
Subsystems (3)
PATCH 1/1 bridge: Return an error when enabling STP in netns.
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n
# ip link add br0 type bridge
# ip link add veth0 type veth peer name veth1
# ip link set veth0 master br0 up
# ip link set veth1 master br0 up
# ip link set br0 type bridge stp_state 1
# ip link set br0 up
# sleep 30
# ip -s link show br0
2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000
link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff
RX: bytes packets errors dropped missed mcast
956553768 12861249 0 0 0 12861249 <-. Keep
TX: bytes packets errors dropped carrier collsns | increasing
1027834 11951 0 0 0 0 <-' rapidlyThis is because llc_rcv() drops all packets in non-root netns and BPDU is dropped.
Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.
Note this commit will be reverted later when we namespacify the whole LLC infra.
--- net/bridge/br_stp_if.c | 5 +++++ 1 file changed, 5 insertions(+)
net/bridge/br_stp_if.c +5 -0
@@ -201,6 +201,11 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); +if (!net_eq (dev_net (br->dev), &init_net)) { +NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); +return -EINVAL ; + } +if (br_mrp_enabled (br)) {NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled if MRP is already enabled" );
Signature · 1 line
2.30.2
Quoted 46 lines · Kuniyuki Iwashima
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n # ip link add br0 type bridge # ip link add veth0 type veth peer name veth1 # ip link set veth0 master br0 up # ip link set veth1 master br0 up # ip link set br0 type bridge stp_state 1 # ip link set br0 up # sleep 30 # ip -s link show br0 2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000 link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 956553768 12861249 0 0 0 12861249 <-. Keep TX: bytes packets errors dropped carrier collsns | increasing 1027834 11951 0 0 0 0 <-' rapidlyThis is because
llc_rcv()drops all packets in non-root netns and BPDU is dropped.Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.Note this commit will be reverted later when we namespacify the whole LLC infra.
Fixes e730c15519d0 ("[NET]: Make packet reception network namespace safe")Suggested-by Harry Coin <[email protected]>Signed-off-by Kuniyuki Iwashima <[email protected]>--- net/bridge/br_stp_if.c | 5 +++++ 1 file changed, 5 insertions(+)net/bridge/br_stp_if.c +0
-0@@ -201,6 +201,11 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {
net/bridge/br_stp_if.c int br_stp_set_enabled(struct net_bridge *br, unsigned long val, · line 211
Quoted 10 lines
net/bridge/br_stp_if.c +5
-0ASSERT_RTNL (); +if (!net_eq (dev_net (br->dev), &init_net)) { +NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); +return -EINVAL ; + } +if (br_mrp_enabled (br)) {NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled if MRP is already enabled" );
Oh well, I guess this should do for now. Hopefully we don't break scripts that rely on configuring the bridge without any errors. Thanks for the patch,
Quoted 29 lines · Kuniyuki Iwashima
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n # ip link add br0 type bridge # ip link add veth0 type veth peer name veth1 # ip link set veth0 master br0 up # ip link set veth1 master br0 up # ip link set br0 type bridge stp_state 1 # ip link set br0 up # sleep 30 # ip -s link show br0 2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000 link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 956553768 12861249 0 0 0 12861249 <-. Keep TX: bytes packets errors dropped carrier collsns | increasing 1027834 11951 0 0 0 0 <-' rapidlyThis is because
llc_rcv()drops all packets in non-root netns and BPDU is dropped.Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.
Quoted 5 lines
Note this commit will be reverted later when we namespacify the whole LLC infra.
Fixes e730c15519d0 ("[NET]: Make packet reception network namespace safe")Suggested-by Harry Coin <[email protected]>
I'm not sure that's accurate. I read his response in the link below and he says "I'd rather be warned than blocked" and "But better warned and awaiting a fix than blocked", which I agree with. The patch has the potential to cause a lot of regressions, but without actually fixing the problem.
How about simply removing the error [1]? Since iproute2 commit 844c37b42373 ("libnetlink: Handle extack messages for non-error case"), it can print extack warnings and not only errors. With the diff below:
# unshare -n # ip link add name br0 type bridge # ip link set dev br0 type bridge stp_state 1 Warning: bridge: STP can't be enabled in non-root netns. # echo $? 0
[1]
net/bridge/br_stp_if.c +1 -3
@@ -201,10 +201,8 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); -if (!net_eq (dev_net (br->dev), &init_net)) { +if (!net_eq (dev_net (br->dev), &init_net))NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); -return -EINVAL ; - }if (br_mrp_enabled (br)) { NL_SET_ERR_MSG_MOD(extack,
Quoted 26 lines
Signed-off-by Kuniyuki Iwashima <[email protected]>--- net/bridge/br_stp_if.c | 5 +++++ 1 file changed, 5 insertions(+)net/bridge/br_stp_if.c +5
-0@@ -201,6 +201,11 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); +if (!net_eq (dev_net (br->dev), &init_net)) { +NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); +return -EINVAL ; + } +if (br_mrp_enabled (br)) {NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled if MRP is already enabled" );Signature · 1 line
2.30.2
Nikolay Aleksandrov
maintainerQuoted 61 lines · Ido Schimmel
Quoted 29 lines · Kuniyuki Iwashima
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n # ip link add br0 type bridge # ip link add veth0 type veth peer name veth1 # ip link set veth0 master br0 up # ip link set veth1 master br0 up # ip link set br0 type bridge stp_state 1 # ip link set br0 up # sleep 30 # ip -s link show br0 2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000 link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 956553768 12861249 0 0 0 12861249 <-. Keep TX: bytes packets errors dropped carrier collsns | increasing 1027834 11951 0 0 0 0 <-' rapidlyThis is because
llc_rcv()drops all packets in non-root netns and BPDU is dropped.Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.Quoted 5 lines
Note this commit will be reverted later when we namespacify the whole LLC infra.
Fixes e730c15519d0 ("[NET]: Make packet reception network namespace safe")Suggested-by Harry Coin <[email protected]>I'm not sure that's accurate. I read his response in the link below and he says "I'd rather be warned than blocked" and "But better warned and awaiting a fix than blocked", which I agree with. The patch has the potential to cause a lot of regressions, but without actually fixing the problem.
How about simply removing the error [1]? Since iproute2 commit
844c37b42373("libnetlink: Handle extack messages for non-error case"), it can print extack warnings and not only errors. With the diff below:# unshare -n # ip link add name br0 type bridge # ip link set dev br0 type bridge stp_state 1 Warning: bridge: STP can't be enabled in non-root netns. # echo $? 0[1]
net/bridge/br_stp_if.c +0
-0@@ -201,10 +201,8 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {
net/bridge/br_stp_if.c int br_stp_set_enabled(struct net_bridge *br, unsigned long val, · line 208
Quoted 10 lines
net/bridge/br_stp_if.c +1
-3ASSERT_RTNL (); -if (!net_eq (dev_net (br->dev), &init_net)) { +if (!net_eq (dev_net (br->dev), &init_net))NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); -return -EINVAL ; - }if (br_mrp_enabled (br)) { NL_SET_ERR_MSG_MOD(extack,
I'd prefer this approach to changing user-visible behaviour and potential regressions. Just change the warning message.
Thanks,
Nik
Quoted 2 lines · Nikolay Aleksandrov
I'd prefer this approach to changing user-visible behaviour and potential regressions. Just change the warning message.
Yea, I noticed after sending that the message no longer fits :)
Quoted 68 lines · Nikolay Aleksandrov
Quoted 67 lines · Ido Schimmel
Quoted 29 lines · Kuniyuki Iwashima
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n # ip link add br0 type bridge # ip link add veth0 type veth peer name veth1 # ip link set veth0 master br0 up # ip link set veth1 master br0 up # ip link set br0 type bridge stp_state 1 # ip link set br0 up # sleep 30 # ip -s link show br0 2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000 link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 956553768 12861249 0 0 0 12861249 <-. Keep TX: bytes packets errors dropped carrier collsns | increasing 1027834 11951 0 0 0 0 <-' rapidlyThis is because
llc_rcv()drops all packets in non-root netns and BPDU is dropped.Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.Quoted 5 lines
Note this commit will be reverted later when we namespacify the whole LLC infra.
Fixes e730c15519d0 ("[NET]: Make packet reception network namespace safe")Suggested-by Harry Coin <[email protected]>I'm not sure that's accurate. I read his response in the link below and he says "I'd rather be warned than blocked" and "But better warned and awaiting a fix than blocked", which I agree with. The patch has the potential to cause a lot of regressions, but without actually fixing the problem.
How about simply removing the error [1]? Since iproute2 commit
844c37b42373("libnetlink: Handle extack messages for non-error case"), it can print extack warnings and not only errors. With the diff below:# unshare -n # ip link add name br0 type bridge # ip link set dev br0 type bridge stp_state 1 Warning: bridge: STP can't be enabled in non-root netns. # echo $? 0[1]
net/bridge/br_stp_if.c +1
-3@@ -201,10 +201,8 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); -if (!net_eq (dev_net (br->dev), &init_net)) { +if (!net_eq (dev_net (br->dev), &init_net))NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); -return -EINVAL ; - }
Quoted 8 lines
net/bridge/br_stp_if.c int br_stp_set_enabled(struct net_bridge *br, unsigned long val, · line 208
Quoted 3 lines
net/bridge/br_stp_if.c +0
-0if (br_mrp_enabled (br)) { NL_SET_ERR_MSG_MOD(extack,I'd prefer this approach to changing user-visible behaviour and potential regressions. Just change the warning message.
Thanks,
Nik
Remember, the only way it's honest to 'warn but not block' STP in netns is trust in the Kuniyuki's assertion that the llc will be 'namespacified' in a near term frame. As STP is not only non-functional in a netns, but will in fact bring down every connected system in a packet storm should a L2 loop exist the situation is much worse than a merely inaccessible extra feature. This as the only reason STP exists is to avoid crashing sites owing to packet storms arising from L2 loops. I think as this bug is a potential 'site killer' (which in fact happened to me!) The Linux dev community has an obligation to either hard block this or commit to a fix time frame and merely warn.
Thanks
Harry Coin
From: Ido Schimmel <[email protected]>
Date: Wed, 12 Jul 2023 17:48:20 +0300
Quoted 42 lines · Kuniyuki Iwashima
Quoted 29 lines
When we create an L2 loop on a bridge in netns, we will see packets storm even if STP is enabled.
# unshare -n # ip link add br0 type bridge # ip link add veth0 type veth peer name veth1 # ip link set veth0 master br0 up # ip link set veth1 master br0 up # ip link set br0 type bridge stp_state 1 # ip link set br0 up # sleep 30 # ip -s link show br0 2: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default qlen 1000 link/ether b6:61:98:1c:1c:b5 brd ff:ff:ff:ff:ff:ff RX: bytes packets errors dropped missed mcast 956553768 12861249 0 0 0 12861249 <-. Keep TX: bytes packets errors dropped carrier collsns | increasing 1027834 11951 0 0 0 0 <-' rapidlyThis is because
llc_rcv()drops all packets in non-root netns and BPDU is dropped.Let's show an error when enabling STP in netns.
# unshare -n # ip link add br0 type bridge # ip link set br0 type bridge stp_state 1 Error: bridge: STP can't be enabled in non-root netns.Quoted 5 lines
Note this commit will be reverted later when we namespacify the whole LLC infra.
Fixes e730c15519d0 ("[NET]: Make packet reception network namespace safe")Suggested-by Harry Coin <[email protected]>I'm not sure that's accurate. I read his response in the link below and he says "I'd rather be warned than blocked" and "But better warned and awaiting a fix than blocked", which I agree with. The patch has the potential to cause a lot of regressions, but without actually fixing the problem.
Quoted 3 lines
How about simply removing the error [1]? Since iproute2 commit
844c37b42373("libnetlink: Handle extack messages for non-error case"), it can print extack warnings and not only errors. With the diff below:
This is good to know and I also prefer this approach! I'll post v2.
Thanks!
Quoted 50 lines
# unshare -n # ip link add name br0 type bridge # ip link set dev br0 type bridge stp_state 1 Warning: bridge: STP can't be enabled in non-root netns. # echo $? 0[1]
net/bridge/br_stp_if.c +1
-3@@ -201,10 +201,8 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); -if (!net_eq (dev_net (br->dev), &init_net)) { +if (!net_eq (dev_net (br->dev), &init_net))NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); -return -EINVAL ; - }if (br_mrp_enabled (br)) { NL_SET_ERR_MSG_MOD(extack,Quoted 25 lines
Signed-off-by Kuniyuki Iwashima <[email protected]>--- net/bridge/br_stp_if.c | 5 +++++ 1 file changed, 5 insertions(+)net/bridge/br_stp_if.c +5
-0@@ -201,6 +201,11 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val, {ASSERT_RTNL (); +if (!net_eq (dev_net (br->dev), &init_net)) { +NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled in non-root netns" ); +return -EINVAL ; + } +if (br_mrp_enabled (br)) {NL_SET_ERR_MSG_MOD (extack,"STP can't be enabled if MRP is already enabled" );Signature · 1 line
2.30.2