netfilter: flowtable bridge and vlan enhancements
Subsystems (5)
Pablo Neira Ayuso
maintainerv3 cover netfilter: flowtable bridge and vlan enhancements
Hi,
The following patchset augments the Netfilter flowtable fastpath [1] to support for network topologies that combine IP forwarding, bridge and VLAN devices.
A typical scenario that can benefit from this infrastructure is composed of several VMs connected to bridge ports where the bridge master device 'br0' has an IP address. A DHCP server is also assumed to be running to provide connectivity to the VMs. The VMs reach the Internet through 'br0' as default gateway, which makes the packet enter the IP forwarding path. Then, netfilter is used to NAT the packets before they leave through the wan device.
Something like this:
fast path
.------------------------.
/ \
| IP forwarding |
| / \ .
| br0 eth0
. / \
-- veth1 veth2
.
.
.
eth0
ab:cd:ef:ab:cd:ef
VMThe idea is to accelerate forwarding by building a fast path that takes packets from the ingress path of the bridge port and place them in the egress path of the wan device (and vice versa). Hence, skipping the classic bridge and IP stack paths.
This patchset is composed of:
Patch #1 adds a placeholder for the hash calculation, instead of using
the dir field. Changes in v3: this patch is new.
Patch #2 adds the transmit path type field to the flow tuple. Two transmit
paths are supported so far: the neighbour and the xfrm transmit paths. This patch comes in preparation to add a new direct ethernet transmit path (see patch #7).
Changes in v3: I have simplified xmit path type logic.
Patch #3 adds dev_fill_forward_path() and .ndo_fill_forward_path() to
netdev_ops. This new function describes the list of netdevice hops
to reach a given destination MAC address in the local network topology,
e.g.
IP forwarding
/ \
br0 eth0
/ \
veth1 veth2
.
.
.
eth0
ab:cd:ef:ab:cd:ef
where veth1 and veth2 are bridge ports and eth0 provides Internet
connectivity. eth0 is the interface in the VM which is connected to
the veth1 bridge port. Then, for packets going to br0 whose
destination MAC address is ab:cd:ef:ab:cd:ef, dev_fill_forward_path()
provides the following path: br0 -> veth1.Patch #4 adds .ndo_fill_forward_path for VLAN devices, which provides the next
device hop via vlan->real_dev. This annotates the VLAN id and protocol. This is useful to know what VLAN headers are expected from the ingress device. This also provides information regarding the VLAN headers to be pushed in the egress path.
Patch #5 adds .ndo_fill_forward_path for bridge devices, which allows to make
lookups to the FDB to locate the next device hop (bridge port) in the forwarding path.
Changes in v3: use READ_ONCE() as requested by Nikolay Aleksandrov.
Patch #6 updates the flowtable to use the dev_fill_forward_path()
infrastructure to obtain the ingress device in the fastpath.
Changes in v3: If user specifies:
devices = { br0, veth1 }
^^^
then, this still works and the fastpath is only created between br0 and veth1 using the neighbour path. Thus, classic bridge path is exercised. If user specifies:
devices = { veth0, veth1 }
^^^^^
then, dev_forward_path() provides an earlier ingress fastpath ie. flowtable lookups for incoming packets happen in veth0 instead of br0).
There is fallback to use the neighbour path if:
- dev_fill_forward_path() finds no real ingress device.
- the ingress device that is obtained is not part of the flowtable devices.
- this route has a xfrm policy.
Patch #7 updates the flowtable to use dev_fill_forward_path() to obtain the
egress device in the forwarding path. This also adds the direct ethernet transmit path, which pushes the ethernet header to the packet and send it through dev_queue_xmit(). This patch adds support for the bridge, so bridge ports use this direct xmit path.
Changes in v3:
- add the direct xmit path and use it for bridge devices type, formely in v2 this came as two separated patches.
- use union so it's either the direct xmit info or the IP dst that is stored in the flow tuple to reduce memory footprint.
- use dev_hard_header() in the direct xmit path to simplify VLAN support coming patch #8.
Patch #8 adds ingress VLAN support (up to 2 VLAN tags, QinQ). The VLAN
information is also provided by dev_fill_forward_path(). Store the VLAN id and protocol in the flow tuple for hash lookups. The VLAN support in the xmit path is achieved by annotating the first vlan device found in the xmit path and by calling dev_hard_header() (previous patch #7) before dev_queue_xmit().
Changes in v3:
- Use VLAN device xmit path instead of manually pushing the VLAN headers from the flowtable datapath to reuse the existing device xmit infrastructure.
Patch #9 extends nft_flowtable.sh selftest: This is adding a test to
cover bridge and vlan support coming in this patchset.
Changes in v3: this patch is new.
= Performance numbers
My testbed environment consists of three containers:
192.168.20.2 .20.1 .10.1 10.141.10.2
veth0 veth0 veth1 veth0
ns1 <---------> nsr1 <--------> ns2
SNAT
iperf -c iperf -s
where nsr1 is used for forwarding. There is a bridge device br0 in nsr1, veth0 is a port of br0. SNAT is performed on the veth1 device of nsr1.
- ns2 runs iperf -s
- ns1 runs iperf -c 10.141.10.2 -n 100G
My results are:
- Baseline (no flowtable, classic forwarding path + netfilter): ~16 Gbit/s
- Fastpath (with flowtable, this patchset): ~25 Gbit/s
This is an improvement of ~50% compared to baseline. My testbed is not optimized, there are a few of instrumentation toggles turned on in my kernel for debugging and this equipment is aging a bit.
Please apply, thanks.
Pablo Neira Ayuso (9):
netfilter: flowtable: add hash offset field to tuple
netfilter: flowtable: add xmit path types
net: resolve forwarding path from virtual netdevice and HW destination address
net: 8021q: resolve forwarding path for vlan devices
bridge: resolve forwarding path for bridge devices
netfilter: flowtable: use dev_fill_forward_path() to obtain ingress device
netfilter: flowtable: use dev_fill_forward_path() to obtain egress device
netfilter: flowtable: add vlan support
selftests: netfilter: flowtable bridge and VLAN support
include/linux/netdevice.h | 35 +++ include/net/netfilter/nf_flow_table.h | 43 +++- net/8021q/vlan_dev.c | 15 ++ net/bridge/br_device.c | 24 +++ net/core/dev.c | 31 +++ net/netfilter/nf_flow_table_core.c | 51 +++-- net/netfilter/nf_flow_table_ip.c | 200 ++++++++++++++---- net/netfilter/nft_flow_offload.c | 159 +++++++++++++- .../selftests/netfilter/nft_flowtable.sh | 82 +++++++ 9 files changed, 580 insertions(+), 60 deletions(-)
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 1/9 netfilter: flowtable: add hash offset field to tuple
Add a placeholder field to calculate hash tuple offset. Similar to 2c407aca6497 ("netfilter: conntrack: avoid gcc-10 zero-length-bounds warning").
--- include/net/netfilter/nf_flow_table.h | 4 ++++ net/netfilter/nf_flow_table_core.c | 6 +++--- 2 files changed, 7 insertions(+), 3 deletions(-)
include/net/netfilter/nf_flow_table.h +4 -0
@@ -107,6 +107,10 @@ struct flow_offload_tuple {u8 l3proto;u8 l4proto; + +/* All members above are keys for lookups, see flow_offload_hash(). */ +struct { } __hash; +u8 dir;u16 mtu;
net/netfilter/nf_flow_table_core.c +3 -3
@@ -191,14 +191,14 @@ static u32 flow_offload_hash(const void *data, u32 len, u32 seed) {const struct flow_offload_tuple *tuple = data; -return jhash (tuple, offsetof(struct flow_offload_tuple , dir), seed); +return jhash (tuple, offsetof(struct flow_offload_tuple , __hash), seed); }static u32 flow_offload_hash_obj (const void *data,u32 len,u32 seed) {const struct flow_offload_tuple_rhash *tuplehash = data; -return jhash (&tuplehash->tuple, offsetof(struct flow_offload_tuple , dir), seed); +return jhash (&tuplehash->tuple, offsetof(struct flow_offload_tuple , __hash), seed); } static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
@@ -207,7 +207,7 @@ static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,const struct flow_offload_tuple *tuple = arg->key;const struct flow_offload_tuple_rhash *x = ptr; -if (memcmp (&x->tuple, tuple, offsetof(struct flow_offload_tuple , dir))) +if (memcmp (&x->tuple, tuple, offsetof(struct flow_offload_tuple , __hash)))return 1 ;return 0 ;
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 2/9 netfilter: flowtable: add xmit path types
In reply to Pablo Neira AyusoAdd the xmit_type field that defines the two supported xmit paths in the flowtable data plane, which are the neighbour and the xfrm xmit paths. This patch prepares for new flowtable xmit path types to come.
--- include/net/netfilter/nf_flow_table.h | 11 +++++++-- net/netfilter/nf_flow_table_core.c | 1 + net/netfilter/nf_flow_table_ip.c | 32 ++++++++++++++++++--------- net/netfilter/nft_flow_offload.c | 20 +++++++++++++++-- 4 files changed, 50 insertions(+), 14 deletions(-)
include/net/netfilter/nf_flow_table.h +9 -2
@@ -89,6 +89,11 @@ enum flow_offload_tuple_dir {FLOW_OFFLOAD_DIR_MAX =IP_CT_DIR_MAX }; +enum flow_offload_xmit_type { +FLOW_OFFLOAD_XMIT_NEIGH =0 , +FLOW_OFFLOAD_XMIT_XFRM , +}; + struct flow_offload_tuple { union { struct in_addr src_v4;
@@ -111,7 +116,8 @@ struct flow_offload_tuple {/* All members above are keys for lookups, see flow_offload_hash(). */ struct { } __hash; -u8 dir; + u8 dir:6, + xmit_type:2;u16 mtu;
@@ -158,7 +164,8 @@ static inline __s32 nf_flow_timeout_delta(unsigned int timeout)struct nf_flow_route {struct { -struct dst_entry *dst; +struct dst_entry *dst; +enum flow_offload_xmit_type xmit_type; } tuple[FLOW_OFFLOAD_DIR_MAX ]; };
net/netfilter/nf_flow_table_core.c +1 -0
@@ -95,6 +95,7 @@ static int flow_offload_fill_route(struct flow_offload *flow, } flow_tuple->iifidx = other_dst->dev->ifindex; + flow_tuple->xmit_type = route->tuple[dir].xmit_type; flow_tuple->dst_cache = dst;return 0 ;
net/netfilter/nf_flow_table_ip.c +22 -10
@@ -220,10 +220,20 @@ static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)return true; } -static int nf_flow_offload_dst_check (struct dst_entry *dst) +static inline struct dst_entry * +nft_flow_dst (struct flow_offload_tuple_rhash *tuplehash) { -if (unlikely (dst_xfrm (dst))) +return tuplehash->tuple.dst_cache; +} + +static int nf_flow_offload_dst_check (struct flow_offload_tuple_rhash *tuplehash) +{ +struct dst_entry *dst; + +if (unlikely (tuplehash->tuple.xmit_type ==FLOW_OFFLOAD_XMIT_XFRM )) { + dst =nft_flow_dst (tuplehash);return dst_check (dst,0 ) ?0 :-1 ; + }return 0 ; }
@@ -265,8 +275,6 @@ nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb, dir = tuplehash->tuple.dir; flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); - rt = (struct rtable *)flow->tuplehash[dir].tuple.dst_cache; - outdev = rt->dst.dev;if (unlikely (nf_flow_exceeds_mtu (skb, flow->tuplehash[dir].tuple.mtu)))return NF_ACCEPT ;
@@ -280,7 +288,7 @@ nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,flow_offload_refresh (flow_table, flow); -if (nf_flow_offload_dst_check (&rt->dst)) { +if (nf_flow_offload_dst_check (tuplehash)) {flow_offload_teardown (flow);return NF_ACCEPT ; }
@@ -295,13 +303,16 @@ nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,if (flow_table->flags &NF_FLOWTABLE_COUNTER )nf_ct_acct_update (flow->ct, tuplehash->tuple.dir, skb->len); -if (unlikely (dst_xfrm (&rt->dst))) { + rt = (struct rtable *)tuplehash->tuple.dst_cache; + +if (unlikely (tuplehash->tuple.xmit_type ==FLOW_OFFLOAD_XMIT_XFRM )) {memset (skb->cb,0 ,sizeof (struct inet_skb_parm ));IPCB (skb)->iif = skb->dev->ifindex;IPCB (skb)->flags =IPSKB_FORWARDED ;return nf_flow_xmit_xfrm (skb, state, &rt->dst); } + outdev = rt->dst.dev; skb->dev = outdev; nexthop =rt_nexthop (rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);skb_dst_set_noref (skb, &rt->dst);
@@ -506,8 +517,6 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb, dir = tuplehash->tuple.dir; flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]); - rt = (struct rt6_info *)flow->tuplehash[dir].tuple.dst_cache; - outdev = rt->dst.dev;if (unlikely (nf_flow_exceeds_mtu (skb, flow->tuplehash[dir].tuple.mtu)))return NF_ACCEPT ;
@@ -518,7 +527,7 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,flow_offload_refresh (flow_table, flow); -if (nf_flow_offload_dst_check (&rt->dst)) { +if (nf_flow_offload_dst_check (tuplehash)) {flow_offload_teardown (flow);return NF_ACCEPT ; }
@@ -536,13 +545,16 @@ nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,if (flow_table->flags &NF_FLOWTABLE_COUNTER )nf_ct_acct_update (flow->ct, tuplehash->tuple.dir, skb->len); -if (unlikely (dst_xfrm (&rt->dst))) { + rt = (struct rt6_info *)tuplehash->tuple.dst_cache; + +if (unlikely (tuplehash->tuple.xmit_type ==FLOW_OFFLOAD_XMIT_NEIGH )) {memset (skb->cb,0 ,sizeof (struct inet6_skb_parm ));IP6CB (skb)->iif = skb->dev->ifindex;IP6CB (skb)->flags =IP6SKB_FORWARDED ;return nf_flow_xmit_xfrm (skb, state, &rt->dst); } + outdev = rt->dst.dev; skb->dev = outdev; nexthop =rt6_nexthop (rt, &flow->tuplehash[!dir].tuple.src_v6);skb_dst_set_noref (skb, &rt->dst);
net/netfilter/nft_flow_offload.c +18 -2
@@ -19,6 +19,22 @@ struct nft_flow_offload {struct nft_flowtable *flowtable; }; +static enum flow_offload_xmit_type nft_xmit_type (struct dst_entry *dst) +{ +if (dst_xfrm (dst)) +return FLOW_OFFLOAD_XMIT_XFRM ; + +return FLOW_OFFLOAD_XMIT_NEIGH ; +} + +static void nft_default_forward_path (struct nf_flow_route *route, +struct dst_entry *dst_cache, +enum ip_conntrack_dir dir) +{ + route->tuple[dir].dst = dst_cache; + route->tuple[dir].xmit_type =nft_xmit_type (dst_cache); +} + static int nft_flow_route(const struct nft_pktinfo *pkt, const struct nf_conn *ct, struct nf_flow_route *route,
@@ -44,8 +60,8 @@ static int nft_flow_route(const struct nft_pktinfo *pkt,if (!other_dst)return -ENOENT ; - route->tuple[dir].dst = this_dst; - route->tuple[!dir].dst = other_dst; +nft_default_forward_path (route, this_dst, dir); +nft_default_forward_path (route, other_dst, !dir);return 0 ; }
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 3/9 net: resolve forwarding path from virtual netdevice and HW destination address
In reply to Pablo Neira AyusoThis patch adds dev_fill_forward_path() which resolves the path to reach the real netdevice from the IP forwarding side. This function takes as input the netdevice and the destination hardware address and it walks down the devices calling .ndo_fill_forward_path() for each device until the real device is found.
For instance, assuming the following topology:
IP forwarding
/ \
br0 eth0
/ \
eth1 eth2
.
.
.
ethXab:cd:ef:ab:cd:ef
where eth1 and eth2 are bridge ports and eth0 provides WAN connectivity. ethX is the interface in another box which is connected to the eth1 bridge port.
For packets going through IP forwarding to br0 whose destination MAC address is ab:cd:ef:ab:cd:ef, dev_fill_forward_path() provides the following path:
br0 -> eth1
.ndo_fill_forward_path for br0 looks up at the FDB for the bridge port from the destination MAC address to get the bridge port eth1.
This information allows to create a fast path that bypasses the classic bridge and IP forwarding paths, so packets go directly from the bridge port eth1 to eth0 (wan interface) and vice versa.
fast path
.------------------------.
/ \
| IP forwarding |
| / \ \/
| br0 eth0
. / \
-> eth1 eth2
.
.
.
ethXab:cd:ef:ab:cd:ef
--- include/linux/netdevice.h | 27 +++++++++++++++++++++++++++ net/core/dev.c | 31 +++++++++++++++++++++++++++++++ 2 files changed, 58 insertions(+)
include/linux/netdevice.h +27 -0
@@ -833,6 +833,27 @@ typedef u16 (*select_queue_fallback_t)(struct net_device *dev,struct sk_buff *skb, struct net_device *sb_dev); +enum net_device_path_type { +DEV_PATH_ETHERNET =0 , +}; + +struct net_device_path { +enum net_device_path_type type; +const struct net_device *dev; +}; + +#define NET_DEVICE_PATH_STACK_MAX 5 + +struct net_device_path_stack { +int num_paths; +struct net_device_path path[NET_DEVICE_PATH_STACK_MAX ]; +}; + +struct net_device_path_ctx { +const struct net_device *dev; +const u8 *daddr; +}; +enum tc_setup_type {TC_SETUP_QDISC_MQPRIO ,TC_SETUP_CLSU32 ,
@@ -1279,6 +1300,8 @@ struct netdev_net_notifier {* struct net_device *(*ndo_get_peer_dev)(struct net_device *dev); * If a device is paired with a peer device, return the peer instance. * The caller must be under RCU read context. +* int (*ndo_fill_forward_path)(struct net_device_path_ctx *ctx, struct net_device_path *path); +* Get the forwarding path to reach the real device from the HW destination address */ struct net_device_ops {int (*ndo_init)(struct net_device *dev);
@@ -1487,6 +1510,8 @@ struct net_device_ops {int (*ndo_tunnel_ctl)(struct net_device *dev,struct ip_tunnel_parm *p,int cmd);struct net_device * (*ndo_get_peer_dev)(struct net_device *dev); +int (*ndo_fill_forward_path)(struct net_device_path_ctx *ctx, +struct net_device_path *path); };/**
@@ -2824,6 +2849,8 @@ void dev_remove_offload(struct packet_offload *po);int dev_get_iflink (const struct net_device *dev);int dev_fill_metadata_dst (struct net_device *dev,struct sk_buff *skb); +int dev_fill_forward_path (const struct net_device *dev,const u8 *daddr, +struct net_device_path_stack *stack);struct net_device *__dev_get_by_flags (struct net *net,unsigned short flags,unsigned short mask);struct net_device *dev_get_by_name (struct net *net,const char *name);
net/core/dev.c +31 -0
@@ -846,6 +846,37 @@ int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) }EXPORT_SYMBOL_GPL (dev_fill_metadata_dst); +int dev_fill_forward_path (const struct net_device *dev,const u8 *daddr, +struct net_device_path_stack *stack) +{ +const struct net_device *last_dev; +struct net_device_path_ctx ctx; +struct net_device_path *path; +int ret =0 ; + +memset (&ctx,0 ,sizeof (ctx)); + ctx.dev = dev; + ctx.daddr = daddr; + +while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { + last_dev = ctx.dev; + + path = &stack->path[stack->num_paths++]; + ret = ctx.dev->netdev_ops->ndo_fill_forward_path (&ctx, path); +if (ret <0 ) +return -1 ; + +if (WARN_ON_ONCE (last_dev == ctx.dev)) +return -1 ; + } + path = &stack->path[stack->num_paths++]; + path->type =DEV_PATH_ETHERNET ; + path->dev = ctx.dev; + +return ret; +} +EXPORT_SYMBOL_GPL (dev_fill_forward_path); +/** * __dev_get_by_name - find a device by its name * @net: the applicable net namespace
Signature · 1 line
2.20.1
Jakub Kicinski
maintainerRe: [PATCH net-next,v3 3/9] net: resolve forwarding path from virtual netdevice and HW destination address
Quoted 6 lines · On Wed, 11 Nov 2020 20:37:31 +0100 Pablo Neira Ayuso wrote:
net/core/dev.c +6
-0+int dev_fill_forward_path (const struct net_device *dev,const u8 *daddr, +struct net_device_path_stack *stack) +{ +const struct net_device *last_dev; +struct net_device_path_ctx ctx; +struct net_device_path *path;
net/core/dev.c int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) · line 864
Quoted 10 lines
net/core/dev.c +10
-0+int ret =0 ; + +memset (&ctx,0 ,sizeof (ctx)); + ctx.dev = dev; + ctx.daddr = daddr; + +while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { + last_dev = ctx.dev; + + path = &stack->path[stack->num_paths++];
I don't see you checking that this stack doesn't overflow.
What am I missing?
Quoted 14 lines
net/core/dev.c +14
-0+ ret = ctx.dev->netdev_ops->ndo_fill_forward_path (&ctx, path); +if (ret <0 ) +return -1 ; + +if (WARN_ON_ONCE (last_dev == ctx.dev)) +return -1 ; + } + path = &stack->path[stack->num_paths++]; + path->type =DEV_PATH_ETHERNET ; + path->dev = ctx.dev; + +return ret; +} +EXPORT_SYMBOL_GPL (dev_fill_forward_path);
Pablo Neira Ayuso
maintainerRe: [PATCH net-next,v3 3/9] net: resolve forwarding path from virtual netdevice and HW destination address
Quoted 20 lines · Jakub Kicinski
Quoted 6 lines · On Wed, 11 Nov 2020 20:37:31 +0100 Pablo Neira Ayuso wrote:
net/core/dev.c +6
-0+int dev_fill_forward_path (const struct net_device *dev,const u8 *daddr, +struct net_device_path_stack *stack) +{ +const struct net_device *last_dev; +struct net_device_path_ctx ctx; +struct net_device_path *path;net/core/dev.c int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) · line 864
Quoted 10 lines
net/core/dev.c +10
-0+int ret =0 ; + +memset (&ctx,0 ,sizeof (ctx)); + ctx.dev = dev; + ctx.daddr = daddr; + +while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { + last_dev = ctx.dev; + + path = &stack->path[stack->num_paths++];I don't see you checking that this stack doesn't overflow.
Quoted 1 line
What am I missing?
Yes, this check in missing indeed.
I'll fix it and will include it in the next patchset version including Nikolay's request.
Thank you.
Pablo Neira Ayuso
maintainerPATCH v3 4/9 net: 8021q: resolve forwarding path for vlan devices
In reply to Pablo Neira AyusoAdd .ndo_fill_forward_path for vlan devices.
For instance, assuming the following topology:
IP forwarding
/ \
eth0.100 eth0
|
eth0
.
.
.
ethX
ab:cd:ef:ab:cd:efFor packets going through IP forwarding to eth0.100 whose destination MAC address is ab:cd:ef:ab:cd:ef, dev_fill_forward_path() provides the following path:
eth0.100 -> eth0
--- include/linux/netdevice.h | 7 +++++++ net/8021q/vlan_dev.c | 15 +++++++++++++++ 2 files changed, 22 insertions(+)
include/linux/netdevice.h +7 -0
@@ -835,11 +835,18 @@ typedef u16 (*select_queue_fallback_t)(struct net_device *dev,enum net_device_path_type {DEV_PATH_ETHERNET =0 , +DEV_PATH_VLAN , };struct net_device_path {enum net_device_path_type type;const struct net_device *dev; +union { +struct { +u16 id; +__be16 proto; + } vlan; + }; };#define NET_DEVICE_PATH_STACK_MAX 5
net/8021q/vlan_dev.c +15 -0
@@ -767,6 +767,20 @@ static int vlan_dev_get_iflink(const struct net_device *dev)return real_dev->ifindex; } +static int vlan_dev_fill_forward_path (struct net_device_path_ctx *ctx, +struct net_device_path *path) +{ +struct vlan_dev_priv *vlan =vlan_dev_priv (ctx->dev); + + path->type =DEV_PATH_VLAN ; + path->vlan.id = vlan->vlan_id; + path->vlan.proto = vlan->vlan_proto; + path->dev = ctx->dev; + ctx->dev = vlan->real_dev; + +return 0 ; +} +static const struct ethtool_ops vlan_ethtool_ops = { .get_link_ksettings = vlan_ethtool_get_link_ksettings, .get_drvinfo = vlan_ethtool_get_drvinfo,
@@ -805,6 +819,7 @@ static const struct net_device_ops vlan_netdev_ops = {#endif .ndo_fix_features = vlan_dev_fix_features, .ndo_get_iflink = vlan_dev_get_iflink, + .ndo_fill_forward_path = vlan_dev_fill_forward_path, };static void vlan_dev_free (struct net_device *dev)
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 5/9 bridge: resolve forwarding path for bridge devices
In reply to Pablo Neira AyusoAdd .ndo_fill_forward_path for bridge devices.
--- include/linux/netdevice.h | 1 + net/bridge/br_device.c | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+)
include/linux/netdevice.h +1 -0
@@ -836,6 +836,7 @@ typedef u16 (*select_queue_fallback_t)(struct net_device *dev,enum net_device_path_type {DEV_PATH_ETHERNET =0 ,DEV_PATH_VLAN , +DEV_PATH_BRIDGE , };struct net_device_path {
net/bridge/br_device.c +24 -0
@@ -391,6 +391,29 @@ static int br_del_slave(struct net_device *dev, struct net_device *slave_dev)return br_del_if (br, slave_dev); } +static int br_fill_forward_path (struct net_device_path_ctx *ctx, +struct net_device_path *path) +{ +struct net_bridge_fdb_entry *f; +struct net_bridge_port *dst; +struct net_bridge *br; + +if (netif_is_bridge_port (ctx->dev)) +return -1 ; + + br =netdev_priv (ctx->dev); + f =br_fdb_find_rcu (br, ctx->daddr,0 ); +if (!f || !f->dst) +return -1 ; + + dst =READ_ONCE (f->dst); + path->type =DEV_PATH_BRIDGE ; + path->dev = dst->br->dev; + ctx->dev = dst->dev; + +return 0 ; +} +static const struct ethtool_ops br_ethtool_ops = { .get_drvinfo = br_getinfo, .get_link = ethtool_op_get_link,
@@ -425,6 +448,7 @@ static const struct net_device_ops br_netdev_ops = { .ndo_bridge_setlink = br_setlink, .ndo_bridge_dellink = br_dellink, .ndo_features_check = passthru_features_check, + .ndo_fill_forward_path = br_fill_forward_path, };static struct device_type br_type = {
Signature · 1 line
2.20.1
Re: [PATCH net-next,v3 5/9] bridge: resolve forwarding path for bridge devices
Quoted 34 lines · Pablo Neira Ayuso
Add .ndo_fill_forward_path for bridge devices.
Signed-off-by Pablo Neira Ayuso <[email protected]>--- include/linux/netdevice.h | 1 + net/bridge/br_device.c | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+)include/linux/netdevice.h +1
-0@@ -836,6 +836,7 @@ typedef u16 (*select_queue_fallback_t)(struct net_device *dev,enum net_device_path_type {DEV_PATH_ETHERNET =0 ,DEV_PATH_VLAN , +DEV_PATH_BRIDGE , };struct net_device_path {net/bridge/br_device.c +6
-0@@ -391,6 +391,29 @@ static int br_del_slave(struct net_device *dev, struct net_device *slave_dev)return br_del_if (br, slave_dev); } +static int br_fill_forward_path (struct net_device_path_ctx *ctx, +struct net_device_path *path) +{ +struct net_bridge_fdb_entry *f; +struct net_bridge_port *dst; +struct net_bridge *br;
net/bridge/br_device.c static int br_del_slave(struct net_device *dev, struct net_device *slave_dev) · line 409
Quoted 10 lines
net/bridge/br_device.c +10
-0+ +if (netif_is_bridge_port (ctx->dev)) +return -1 ; + + br =netdev_priv (ctx->dev); + f =br_fdb_find_rcu (br, ctx->daddr,0 ); +if (!f || !f->dst) +return -1 ; + + dst =READ_ONCE (f->dst);
While this is ok, I meant that you have to test the ptr after. In theory between the !f->dst test above and now it could've become null, so to make it future-proof please test the null dst after deref, after the READ_ONCE(). I realize currently there are still problems after the change but we'll fix them.
Quoted 19 lines
net/bridge/br_device.c +8
-0+ path->type =DEV_PATH_BRIDGE ; + path->dev = dst->br->dev; + ctx->dev = dst->dev; + +return 0 ; +} +static const struct ethtool_ops br_ethtool_ops = { .get_drvinfo = br_getinfo, .get_link = ethtool_op_get_link,@@ -425,6 +448,7 @@ static const struct net_device_ops br_netdev_ops = { .ndo_bridge_setlink = br_setlink, .ndo_bridge_dellink = br_dellink, .ndo_features_check = passthru_features_check, + .ndo_fill_forward_path = br_fill_forward_path, };static struct device_type br_type = {
Re: [PATCH net-next,v3 5/9] bridge: resolve forwarding path for bridge devices
Quoted 42 lines · Nikolay Aleksandrov
Quoted 41 lines · Pablo Neira Ayuso
Add .ndo_fill_forward_path for bridge devices.
Signed-off-by Pablo Neira Ayuso <[email protected]>--- include/linux/netdevice.h | 1 + net/bridge/br_device.c | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+)include/linux/netdevice.h +1
-0@@ -836,6 +836,7 @@ typedef u16 (*select_queue_fallback_t)(struct net_device *dev,enum net_device_path_type {DEV_PATH_ETHERNET =0 ,DEV_PATH_VLAN , +DEV_PATH_BRIDGE , };struct net_device_path {net/bridge/br_device.c +13
-0@@ -391,6 +391,29 @@ static int br_del_slave(struct net_device *dev, struct net_device *slave_dev)return br_del_if (br, slave_dev); } +static int br_fill_forward_path (struct net_device_path_ctx *ctx, +struct net_device_path *path) +{ +struct net_bridge_fdb_entry *f; +struct net_bridge_port *dst; +struct net_bridge *br; + +if (netif_is_bridge_port (ctx->dev)) +return -1 ; + + br =netdev_priv (ctx->dev); + f =br_fdb_find_rcu (br, ctx->daddr,0 ); +if (!f || !f->dst)
Quoted 9 lines
net/bridge/br_device.c static int br_del_slave(struct net_device *dev, struct net_device *slave_dev) · line 409
Quoted 3 lines
net/bridge/br_device.c +3
-0+return -1 ; + + dst =READ_ONCE (f->dst);While this is ok, I meant that you have to test the ptr after. In theory between the !f->dst test above and now it could've become null, so to make it future-proof please test the null dst after deref, after the
READ_ONCE(). I realize currently there are still problems after the change but we'll fix them.
On a second read this sounds too vague. :) I meant that there are still similar issues in the bridge code today which I'll take care of. For this patch what I suggested should be enough.
Thanks,
Nik
Quoted 20 lines
Quoted 19 lines
net/bridge/br_device.c +8
-0+ path->type =DEV_PATH_BRIDGE ; + path->dev = dst->br->dev; + ctx->dev = dst->dev; + +return 0 ; +} +static const struct ethtool_ops br_ethtool_ops = { .get_drvinfo = br_getinfo, .get_link = ethtool_op_get_link,@@ -425,6 +448,7 @@ static const struct net_device_ops br_netdev_ops = { .ndo_bridge_setlink = br_setlink, .ndo_bridge_dellink = br_dellink, .ndo_features_check = passthru_features_check, + .ndo_fill_forward_path = br_fill_forward_path, };static struct device_type br_type = {
Pablo Neira Ayuso
maintainerPATCH v3 6/9 netfilter: flowtable: use dev_fill_forward_path() to obtain ingress device
In reply to Pablo Neira AyusoObtain the ingress device in the tuple from the route in the reply direction. Use dev_fill_forward_path() instead to get the real ingress device for this flow.
Fall back to use the ingress device that the IP forwarding route provides if:
- dev_fill_forward_path() finds no real ingress device.
- the ingress device that is obtained is not part of the flowtable devices.
- this route has a xfrm policy.
--- include/net/netfilter/nf_flow_table.h | 3 + net/netfilter/nf_flow_table_core.c | 3 +- net/netfilter/nft_flow_offload.c | 101 +++++++++++++++++++++++++- 3 files changed, 102 insertions(+), 5 deletions(-)
include/net/netfilter/nf_flow_table.h +3 -0
@@ -165,6 +165,9 @@ static inline __s32 nf_flow_timeout_delta(unsigned int timeout)struct nf_flow_route {struct {struct dst_entry *dst; +struct { +u32 ifindex; + } in;enum flow_offload_xmit_type xmit_type; } tuple[FLOW_OFFLOAD_DIR_MAX ]; };
net/netfilter/nf_flow_table_core.c +1 -2
@@ -79,7 +79,6 @@ static int flow_offload_fill_route(struct flow_offload *flow, enum flow_offload_tuple_dir dir) {struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple; -struct dst_entry *other_dst = route->tuple[!dir].dst;struct dst_entry *dst = route->tuple[dir].dst;if (!dst_hold_safe (route->tuple[dir].dst))
@@ -94,7 +93,7 @@ static int flow_offload_fill_route(struct flow_offload *flow,break ; } - flow_tuple->iifidx = other_dst->dev->ifindex; + flow_tuple->iifidx = route->tuple[dir].in.ifindex; flow_tuple->xmit_type = route->tuple[dir].xmit_type; flow_tuple->dst_cache = dst;
net/netfilter/nft_flow_offload.c +98 -3
133 lines not on this page · raw message
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 7/9 netfilter: flowtable: use dev_fill_forward_path() to obtain egress device
In reply to Pablo Neira AyusoThe egress device in the tuple is obtained from route. Use dev_fill_forward_path() instead to provide the real egress device for this flow whenever this is available.
The new FLOW_OFFLOAD_XMIT_DIRECT type uses dev_queue_xmit() to transmit ethernet frames. Cache the source and destination hardware address to use dev_queue_xmit() to transfer packets.
The FLOW_OFFLOAD_XMIT_DIRECT replaces FLOW_OFFLOAD_XMIT_NEIGH if dev_fill_forward_path() finds a direct transmit path.
--- include/net/netfilter/nf_flow_table.h | 16 +++++- net/netfilter/nf_flow_table_core.c | 35 ++++++++++--- net/netfilter/nf_flow_table_ip.c | 72 +++++++++++++++++++++------ net/netfilter/nft_flow_offload.c | 25 ++++++++-- 4 files changed, 118 insertions(+), 30 deletions(-)
include/net/netfilter/nf_flow_table.h +14 -2
@@ -92,6 +92,7 @@ enum flow_offload_tuple_dir {enum flow_offload_xmit_type {FLOW_OFFLOAD_XMIT_NEIGH =0 ,FLOW_OFFLOAD_XMIT_XFRM , +FLOW_OFFLOAD_XMIT_DIRECT , };struct flow_offload_tuple {
@@ -120,8 +121,14 @@ struct flow_offload_tuple { xmit_type:2 ;u16 mtu; - -struct dst_entry *dst_cache; +union { +struct dst_entry *dst_cache; +struct { +u32 ifidx; +u8 h_source[ETH_ALEN ]; +u8 h_dest[ETH_ALEN ]; + } out; + }; };struct flow_offload_tuple_rhash {
@@ -168,6 +175,11 @@ struct nf_flow_route {struct {u32 ifindex; } in; +struct { +u32 ifindex; +u8 h_source[ETH_ALEN ]; +u8 h_dest[ETH_ALEN ]; + } out;enum flow_offload_xmit_type xmit_type; } tuple[FLOW_OFFLOAD_DIR_MAX ]; };
net/netfilter/nf_flow_table_core.c +28 -7
@@ -81,9 +81,6 @@ static int flow_offload_fill_route(struct flow_offload *flow,struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;struct dst_entry *dst = route->tuple[dir].dst; -if (!dst_hold_safe (route->tuple[dir].dst)) -return -1 ; -switch (flow_tuple->l3proto) {case NFPROTO_IPV4 : flow_tuple->mtu =ip_dst_mtu_maybe_forward (dst, true);
@@ -94,12 +91,36 @@ static int flow_offload_fill_route(struct flow_offload *flow, } flow_tuple->iifidx = route->tuple[dir].in.ifindex; + +switch (route->tuple[dir].xmit_type) { +case FLOW_OFFLOAD_XMIT_DIRECT : +memcpy (flow_tuple->out.h_dest, route->tuple[dir].out.h_dest, +ETH_ALEN ); +memcpy (flow_tuple->out.h_source, route->tuple[dir].out.h_source, +ETH_ALEN ); + flow_tuple->out.ifidx = route->tuple[dir].out.ifindex; +break ; +case FLOW_OFFLOAD_XMIT_XFRM : +case FLOW_OFFLOAD_XMIT_NEIGH : +if (!dst_hold_safe (route->tuple[dir].dst)) +return -1 ; + + flow_tuple->dst_cache = dst; +break ; + } flow_tuple->xmit_type = route->tuple[dir].xmit_type; - flow_tuple->dst_cache = dst;return 0 ; } +static void nft_flow_dst_release (struct flow_offload *flow, +enum flow_offload_tuple_dir dir) +{ +if (flow->tuplehash[dir].tuple.xmit_type ==FLOW_OFFLOAD_XMIT_NEIGH || + flow->tuplehash[dir].tuple.xmit_type ==FLOW_OFFLOAD_XMIT_XFRM ) +dst_release (flow->tuplehash[dir].tuple.dst_cache); +} +int flow_offload_route_init (struct flow_offload *flow,const struct nf_flow_route *route) {
@@ -118,7 +139,7 @@ int flow_offload_route_init(struct flow_offload *flow,return 0 ; err_route_reply: -dst_release (route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL ].dst); +nft_flow_dst_release (flow,FLOW_OFFLOAD_DIR_ORIGINAL );return err; }
@@ -169,8 +190,8 @@ static void flow_offload_fixup_ct(struct nf_conn *ct)static void flow_offload_route_release (struct flow_offload *flow) { -dst_release (flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL ].tuple.dst_cache); -dst_release (flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY ].tuple.dst_cache); +nft_flow_dst_release (flow,FLOW_OFFLOAD_DIR_ORIGINAL ); +nft_flow_dst_release (flow,FLOW_OFFLOAD_DIR_REPLY ); }void flow_offload_free (struct flow_offload *flow)
net/netfilter/nf_flow_table_ip.c +56 -16
117 lines not on this page · raw message
net/netfilter/nft_flow_offload.c +20 -5
@@ -39,12 +39,11 @@ static void nft_default_forward_path(struct nf_flow_route *route,static int nft_dev_fill_forward_path (const struct nf_flow_route *route,const struct dst_entry *dst_cache,const struct nf_conn *ct, -enum ip_conntrack_dir dir, +enum ip_conntrack_dir dir,u8 *ha,struct net_device_path_stack *stack) {const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;struct net_device *dev = dst_cache->dev; -unsigned char ha[ETH_ALEN ];struct neighbour *n;u8 nud_state;
@@ -66,10 +65,14 @@ static int nft_dev_fill_forward_path(const struct nf_flow_route *route,struct nft_forward_info {const struct net_device *indev; +u8 h_source[ETH_ALEN ]; +u8 h_dest[ETH_ALEN ]; +enum flow_offload_xmit_type xmit_type; };static void nft_dev_path_info (const struct net_device_path_stack *stack, -struct nft_forward_info *info) +struct nft_forward_info *info, +unsigned char *ha) {const struct net_device_path *path;int i;
@@ -79,10 +82,14 @@ static void nft_dev_path_info(const struct net_device_path_stack *stack,switch (path->type) {case DEV_PATH_ETHERNET : info->indev = path->dev; +memcpy (info->h_dest, path->dev->dev_addr,ETH_ALEN ); +memcpy (info->h_source, ha,ETH_ALEN );break ;case DEV_PATH_VLAN :break ;case DEV_PATH_BRIDGE : +memcpy (info->h_dest, path->dev->dev_addr,ETH_ALEN ); + info->xmit_type =FLOW_OFFLOAD_XMIT_DIRECT ;break ; } }
@@ -113,14 +120,22 @@ static void nft_dev_forward_path(struct nf_flow_route *route,const struct dst_entry *dst = route->tuple[dir].dst;struct net_device_path_stack stack = {};struct nft_forward_info info = {}; +unsigned char ha[ETH_ALEN ]; -if (nft_dev_fill_forward_path (route, dst, ct, dir, &stack) >=0 ) -nft_dev_path_info (&stack, &info); +if (nft_dev_fill_forward_path (route, dst, ct, dir, ha, &stack) >=0 ) +nft_dev_path_info (&stack, &info, ha);if (!info.indev || !nft_flowtable_find_dev (info.indev, ft))return ; route->tuple[!dir].in.ifindex = info.indev->ifindex; + +if (info.xmit_type ==FLOW_OFFLOAD_XMIT_DIRECT ) { +memcpy (route->tuple[dir].out.h_dest, info.h_source,ETH_ALEN ); +memcpy (route->tuple[dir].out.h_source, info.h_dest,ETH_ALEN ); + route->tuple[dir].out.ifindex = info.indev->ifindex; + route->tuple[dir].xmit_type = info.xmit_type; + } } static int nft_flow_route(const struct nft_pktinfo *pkt,
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 8/9 netfilter: flowtable: add vlan support
In reply to Pablo Neira AyusoAdd the vlan id and protocol to the flow tuple to uniquely identify flows from the receive path. For the transmit path, dev_hard_header() on the vlan device push the headers. This patch includes support for two VLAN headers (QinQ) from the ingress path.
--- include/net/netfilter/nf_flow_table.h | 15 +++- net/netfilter/nf_flow_table_core.c | 6 ++ net/netfilter/nf_flow_table_ip.c | 108 +++++++++++++++++++++----- net/netfilter/nft_flow_offload.c | 25 +++++- 4 files changed, 131 insertions(+), 23 deletions(-)
include/net/netfilter/nf_flow_table.h +12 -3
@@ -95,6 +95,8 @@ enum flow_offload_xmit_type { FLOW_OFFLOAD_XMIT_DIRECT, }; +#define NF_FLOW_TABLE_VLAN_MAX 2 + struct flow_offload_tuple { union { struct in_addr src_v4;
@@ -113,13 +115,17 @@ struct flow_offload_tuple {u8 l3proto;u8 l4proto; +struct { +u16 id; +__be16 proto; + } in_vlan[NF_FLOW_TABLE_VLAN_MAX ];/* All members above are keys for lookups, see flow_offload_hash(). */ struct { } __hash; - u8 dir:6, - xmit_type:2; - + u8 dir:4, + xmit_type:2, + in_vlan_num:2;u16 mtu;union {struct dst_entry *dst_cache;
@@ -174,6 +180,9 @@ struct nf_flow_route {struct dst_entry *dst;struct {u32 ifindex; +u16 vid[NF_FLOW_TABLE_VLAN_MAX ]; +__be16 vproto[NF_FLOW_TABLE_VLAN_MAX ]; +u8 num_vlans; } in;struct {u32 ifindex;
net/netfilter/nf_flow_table_core.c +6 -0
@@ -80,6 +80,7 @@ static int flow_offload_fill_route(struct flow_offload *flow, {struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;struct dst_entry *dst = route->tuple[dir].dst; +int i;switch (flow_tuple->l3proto) {case NFPROTO_IPV4 :
@@ -91,6 +92,11 @@ static int flow_offload_fill_route(struct flow_offload *flow, } flow_tuple->iifidx = route->tuple[dir].in.ifindex; +for (i =0 ; i < route->tuple[dir].in.num_vlans; i++) { + flow_tuple->in_vlan[i].id = route->tuple[dir].in.vid[i]; + flow_tuple->in_vlan[i].proto = route->tuple[dir].in.vproto[i]; + } + flow_tuple->in_vlan_num = route->tuple[dir].in.num_vlans;switch (route->tuple[dir].xmit_type) {case FLOW_OFFLOAD_XMIT_DIRECT :
net/netfilter/nf_flow_table_ip.c +89 -19
209 lines not on this page · raw message
net/netfilter/nft_flow_offload.c +24 -1
@@ -65,6 +65,10 @@ static int nft_dev_fill_forward_path(const struct nf_flow_route *route,struct nft_forward_info {const struct net_device *indev; +const struct net_device *vlandev; +__u16 vid[NF_FLOW_TABLE_VLAN_MAX ]; +__be16 vproto[NF_FLOW_TABLE_VLAN_MAX ]; +u8 num_vlans;u8 h_source[ETH_ALEN ];u8 h_dest[ETH_ALEN ];enum flow_offload_xmit_type xmit_type;
@@ -86,6 +90,16 @@ static void nft_dev_path_info(const struct net_device_path_stack *stack,memcpy (info->h_source, ha,ETH_ALEN );break ;case DEV_PATH_VLAN : +if (info->num_vlans >=NF_FLOW_TABLE_VLAN_MAX ) { + info->indev =NULL ; +break ; + } +if (!info->vlandev) + info->vlandev = path->dev; + + info->vid[info->num_vlans] = path->vlan.id; + info->vproto[info->num_vlans] = path->vlan.proto; + info->num_vlans++;break ;case DEV_PATH_BRIDGE :memcpy (info->h_dest, path->dev->dev_addr,ETH_ALEN );
@@ -121,6 +135,7 @@ static void nft_dev_forward_path(struct nf_flow_route *route,struct net_device_path_stack stack = {};struct nft_forward_info info = {};unsigned char ha[ETH_ALEN ]; +int i;if (nft_dev_fill_forward_path (route, dst, ct, dir, ha, &stack) >=0 )nft_dev_path_info (&stack, &info, ha);
@@ -129,11 +144,19 @@ static void nft_dev_forward_path(struct nf_flow_route *route,return ; route->tuple[!dir].in.ifindex = info.indev->ifindex; +for (i =0 ; i < info.num_vlans; i++) { + route->tuple[!dir].in.vid[i] = info.vid[i]; + route->tuple[!dir].in.vproto[i] = info.vproto[i]; + } + route->tuple[!dir].in.num_vlans = info.num_vlans;if (info.xmit_type ==FLOW_OFFLOAD_XMIT_DIRECT ) {memcpy (route->tuple[dir].out.h_dest, info.h_source,ETH_ALEN );memcpy (route->tuple[dir].out.h_source, info.h_dest,ETH_ALEN ); - route->tuple[dir].out.ifindex = info.indev->ifindex; +if (info.vlandev) + route->tuple[dir].out.ifindex = info.vlandev->ifindex; +else + route->tuple[dir].out.ifindex = info.indev->ifindex; route->tuple[dir].xmit_type = info.xmit_type; } }
Signature · 1 line
2.20.1
Pablo Neira Ayuso
maintainerPATCH v3 9/9 selftests: netfilter: flowtable bridge and VLAN support
In reply to Pablo Neira AyusoThis patch adds two new tests to cover bridge and VLAN support:
- Add a bridge device to the Router1 (nsr1) container and attach the veth0 device to the bridge. Set the IP address to the bridge device to exercise the bridge forwarding path.
- Add VLAN encapsulation between to the bridge device in the Router1 and one of the sender containers (ns1).
--- .../selftests/netfilter/nft_flowtable.sh | 82 +++++++++++++++++++ 1 file changed, 82 insertions(+)
tools/testing/selftests/netfilter/nft_flowtable.sh +82 -0
@@ -371,6 +371,88 @@ else ip netns exec nsr1 nft list ruleset fi +# Another test: +# Add bridge interface br0 to Router1, with NAT enabled. +ip -net nsr1 link add name br0 type bridge +ip -net nsr1 addr flush dev veth0 +ip -net nsr1 link set up dev veth0 +ip -net nsr1 link set veth0 master br0 +ip -net nsr1 addr add 10.0.1.1/24 dev br0 +ip -net nsr1 addr add dead:1::1/64 dev br0 +ip -net nsr1 link set up dev br0 + +ip netns exec nsr1 sysctl net.ipv4.conf.br0.forwarding=1 > /dev/null + +# br0 with NAT enabled. +ip netns exec nsr1 nft -f - <<EOF +flush table ip nat +table ip nat { + chain prerouting { + type nat hook prerouting priority 0; policy accept; + meta iif "br0" ip daddr 10.6.6.6 tcp dport 1666 counter dnat ip to 10.0.2.99:12345 + } + + chain postrouting { + type nat hook postrouting priority 0; policy accept; + meta oifname "veth1" counter masquerade + } +} +EOF + +if test_tcp_forwarding_nat ns1 ns2; then + echo "PASS: flow offloaded for ns1/ns2 with bridge NAT" +else + echo "FAIL: flow offload for ns1/ns2 with bridge NAT" 1>&2 + ip netns exec nsr1 nft list ruleset + ret=1 +fi + +# Another test: +# Add bridge interface br0 to Router1, with NAT and VLAN. +ip -net nsr1 link set veth0 nomaster +ip -net nsr1 link set down dev veth0 +ip -net nsr1 link add link veth0 name veth0.10 type vlan id 10 +ip -net nsr1 link set up dev veth0 +ip -net nsr1 link set up dev veth0.10 +ip -net nsr1 link set veth0.10 master br0 + +ip -net ns1 addr flush dev eth0 +ip -net ns1 link add link eth0 name eth0.10 type vlan id 10 +ip -net ns1 link set eth0 up +ip -net ns1 link set eth0.10 up +ip -net ns1 addr add 10.0.1.99/24 dev eth0.10 +ip -net ns1 route add default via 10.0.1.1 +ip -net ns1 addr add dead:1::99/64 dev eth0.10 + +if test_tcp_forwarding_nat ns1 ns2; then + echo "PASS: flow offloaded for ns1/ns2 with bridge NAT and VLAN" +else + echo "FAIL: flow offload for ns1/ns2 with bridge NAT and VLAN" 1>&2 + ip netns exec nsr1 nft list ruleset + ret=1 +fi + +# restore test topology (remove bridge and VLAN) +ip -net nsr1 link set veth0 nomaster +ip -net nsr1 link set veth0 down +ip -net nsr1 link set veth0.10 down +ip -net nsr1 link delete veth0.10 type vlan +ip -net nsr1 link delete br0 type bridge +ip -net ns1 addr flush dev eth0.10 +ip -net ns1 link set eth0.10 down +ip -net ns1 link set eth0 down +ip -net ns1 link delete eth0.10 type vlan + +# restore address in ns1 and nsr1 +ip -net ns1 link set eth0 up +ip -net ns1 addr add 10.0.1.99/24 dev eth0 +ip -net ns1 route add default via 10.0.1.1 +ip -net ns1 addr add dead:1::99/64 dev eth0 +ip -net ns1 route add default via dead:1::1 +ip -net nsr1 addr add 10.0.1.1/24 dev veth0 +ip -net nsr1 addr add dead:1::1/64 dev veth0 +ip -net nsr1 link set up dev veth0 + KEY_SHA="0x"$(ps -xaf | sha1sum | cut -d " " -f 1) KEY_AES="0x"$(ps -xaf | md5sum | cut -d " " -f 1) SPI1=$RANDOM
Signature · 1 line
2.20.1
Jakub Kicinski
maintainerQuoted 29 lines · On Wed, 11 Nov 2020 20:37:28 +0100 Pablo Neira Ayuso wrote:
The following patchset augments the Netfilter flowtable fastpath [1] to support for network topologies that combine IP forwarding, bridge and VLAN devices.
A typical scenario that can benefit from this infrastructure is composed of several VMs connected to bridge ports where the bridge master device 'br0' has an IP address. A DHCP server is also assumed to be running to provide connectivity to the VMs. The VMs reach the Internet through 'br0' as default gateway, which makes the packet enter the IP forwarding path. Then, netfilter is used to NAT the packets before they leave through the wan device.
Something like this:
fast path .------------------------. / \ | IP forwarding | | / \ . | br0 eth0 . / \ -- veth1 veth2 . . . eth0 ab:cd:ef:ab:cd:ef VM
Quoted 4 lines
The idea is to accelerate forwarding by building a fast path that takes packets from the ingress path of the bridge port and place them in the egress path of the wan device (and vice versa). Hence, skipping the classic bridge and IP stack paths.
The problem that immediately comes to mind is that if there is any dynamic forwarding state the cache you're creating would need to be flushed when FDB changes. Are you expecting users would plug into the flowtable devices where they know things are fairly static?
Pablo Neira Ayuso
maintainerQuoted 35 lines · Jakub Kicinski
Quoted 29 lines · On Wed, 11 Nov 2020 20:37:28 +0100 Pablo Neira Ayuso wrote:
The following patchset augments the Netfilter flowtable fastpath [1] to support for network topologies that combine IP forwarding, bridge and VLAN devices.
A typical scenario that can benefit from this infrastructure is composed of several VMs connected to bridge ports where the bridge master device 'br0' has an IP address. A DHCP server is also assumed to be running to provide connectivity to the VMs. The VMs reach the Internet through 'br0' as default gateway, which makes the packet enter the IP forwarding path. Then, netfilter is used to NAT the packets before they leave through the wan device.
Something like this:
fast path .------------------------. / \ | IP forwarding | | / \ . | br0 eth0 . / \ -- veth1 veth2 . . . eth0 ab:cd:ef:ab:cd:ef VMQuoted 4 lines
The idea is to accelerate forwarding by building a fast path that takes packets from the ingress path of the bridge port and place them in the egress path of the wan device (and vice versa). Hence, skipping the classic bridge and IP stack paths.
Quoted 4 lines
The problem that immediately comes to mind is that if there is any dynamic forwarding state the cache you're creating would need to be flushed when FDB changes. Are you expecting users would plug into the flowtable devices where they know things are fairly static?
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
Thank you.
Quoted 8 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Quoted 8 lines
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
This should work in the case when the bridge ports are normal NICs or switchdev ports, right?
In that case, relying on link state is brittle as you can easily have a switch or a media converter between the bridge and the end-station:
br0 br0
/ \ / \
eth0 eth1 eth0 eth1
/ \ => / \ [sw0] [sw1] [sw0] [sw1]
/ \ / \
A A
In a scenario like this, A has clearly moved. But neither eth0 nor eth1 has seen any changes in link state.
This particular example is a bit contrived. But this is essentially what happens in redundant topologies when reconfigurations occur (e.g. STP).
These protocols will typically signal reconfigurations to all bridges though, so as long as the affected flows are flushed at the same time as the FDB it should work.
Interesting stuff!
Jakub Kicinski
maintainerQuoted 9 lines · On Sat, 14 Nov 2020 15:00:03 +0100 Tobias Waldekranz wrote:
Quoted 8 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Quoted 8 lines
Quoted 8 lines
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
Indeed, 2 VMs A and B, talking to each other, A is _outside_ the system (reachable via eth0), B is inside (veth1). When A moves inside and gets its veth. Neither B's veth1 not eth0 will change state, so cache wouldn't get flushed, right?
Quoted 17 lines
This should work in the case when the bridge ports are normal NICs or switchdev ports, right?
In that case, relying on link state is brittle as you can easily have a switch or a media converter between the bridge and the end-station:
br0 br0 / \ / \ eth0 eth1 eth0 eth1 / \ => / \[sw0] [sw1] [sw0] [sw1]
/ \ / \
A AIn a scenario like this, A has clearly moved. But neither eth0 nor eth1 has seen any changes in link state.
Quoted 8 lines
This particular example is a bit contrived. But this is essentially what happens in redundant topologies when reconfigurations occur (e.g. STP).
These protocols will typically signal reconfigurations to all bridges though, so as long as the affected flows are flushed at the same time as the FDB it should work.
Interesting stuff!
Agreed, could be interesting for all NAT/conntrack setups, not just VMs.
Pablo Neira Ayuso
maintainerQuoted 16 lines · Jakub Kicinski
Quoted 15 lines · On Sat, 14 Nov 2020 15:00:03 +0100 Tobias Waldekranz wrote:
Quoted 14 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
Quoted 7 lines
Quoted 2 lines
Quoted 2 lines
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
Indeed, 2 VMs A and B, talking to each other, A is _outside_ the system (reachable via eth0), B is inside (veth1). When A moves inside and gets its veth. Neither B's veth1 not eth0 will change state, so cache wouldn't get flushed, right?
The flow tuple includes the input interface as part of the hash key, so packets will not match the existing entries in the flowtable after the topology update. The stale flow entries are removed after 30 seconds if no matching packets are seen. New flow entries will be created for the new topology, a few packets have to go through the classic forwarding path so the new flow entries that represent the flow in the new topology are created.
Jakub Kicinski
maintainerQuoted 20 lines · On Mon, 16 Nov 2020 23:18:15 +0100 Pablo Neira Ayuso wrote:
Quoted 19 lines · Jakub Kicinski
Quoted 17 lines · On Sat, 14 Nov 2020 15:00:03 +0100 Tobias Waldekranz wrote:
Quoted 16 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
Quoted 8 lines
Quoted 4 lines
Indeed, 2 VMs A and B, talking to each other, A is _outside_ the system (reachable via eth0), B is inside (veth1). When A moves inside and gets its veth. Neither B's veth1 not eth0 will change state, so cache wouldn't get flushed, right?
The flow tuple includes the input interface as part of the hash key, so packets will not match the existing entries in the flowtable after the topology update.
To be clear - the input interface for B -> A traffic remains B. So if B was talking to A before A moved it will keep hitting the cached entry.
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Quoted 5 lines
The stale flow entries are removed after 30 seconds
if no matching packets are seen. New flow entries will be created for the new topology, a few packets have to go through the classic forwarding path so the new flow entries that represent the flow in the new topology are created.
Pablo Neira Ayuso
maintainerQuoted 26 lines · Jakub Kicinski
Quoted 25 lines · On Mon, 16 Nov 2020 23:18:15 +0100 Pablo Neira Ayuso wrote:
Quoted 23 lines · Jakub Kicinski
Quoted 9 lines · On Sat, 14 Nov 2020 15:00:03 +0100 Tobias Waldekranz wrote:
Quoted 8 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Quoted 8 lines
Quoted 8 lines
Existing flows still follow the old path. The same happens with FIB currently.
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
Indeed, 2 VMs A and B, talking to each other, A is _outside_ the system (reachable via eth0), B is inside (veth1). When A moves inside and gets its veth. Neither B's veth1 not eth0 will change state, so cache wouldn't get flushed, right?
Quoted 7 lines
Quoted 3 lines
The flow tuple includes the input interface as part of the hash key, so packets will not match the existing entries in the flowtable after the topology update.
To be clear - the input interface for B -> A traffic remains B. So if B was talking to A before A moved it will keep hitting the cached entry.
Yes, Traffic for B -> A still hits the cached entry.
Quoted 2 lines
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Yes, Traffic for A -> B won't match the flowtable entry, this will update the cache.
Jakub Kicinski
maintainerQuoted 5 lines · On Mon, 16 Nov 2020 23:36:15 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Yes, Traffic for A -> B won't match the flowtable entry, this will update the cache.
That's assuming there will be A -> B traffic without B sending a request which reaches A, first.
Pablo Neira Ayuso
maintainerQuoted 9 lines · Jakub Kicinski
Quoted 5 lines · On Mon, 16 Nov 2020 23:36:15 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Yes, Traffic for A -> B won't match the flowtable entry, this will update the cache.
That's assuming there will be A -> B traffic without B sending a request which reaches A, first.
B might send packets to A but this will not get anywhere. Assuming TCP, this will trigger retransmissions so B -> A will kick in to refresh the entry.
Is this scenario that you describe a showstopper?
Thank you.
Pablo Neira Ayuso
maintainerQuoted 8 lines · Pablo Neira Ayuso
Quoted 7 lines · Jakub Kicinski
Quoted 5 lines · On Mon, 16 Nov 2020 23:36:15 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Yes, Traffic for A -> B won't match the flowtable entry, this will update the cache.
Quoted 8 lines
Quoted 2 lines
That's assuming there will be A -> B traffic without B sending a request which reaches A, first.
B might send packets to A but this will not get anywhere. Assuming TCP, this will trigger retransmissions so B -> A will kick in to refresh the entry.
Is this scenario that you describe a showstopper?
I have been discussing the topology update by tracking fdb updates with the bridge maintainer, I'll be exploring extensions to the existing fdb_notify() infrastructure to deal with this scenario you describe. On my side this topology update scenario is not a priority to be supported in this patchset, but it's feasible to support it later on.
Jakub Kicinski
maintainerQuoted 9 lines · On Sat, 21 Nov 2020 13:31:38 +0100 Pablo Neira Ayuso wrote:
Quoted 8 lines · Pablo Neira Ayuso
Quoted 7 lines · Jakub Kicinski
Quoted 5 lines · On Mon, 16 Nov 2020 23:36:15 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Are you saying A -> B traffic won't match so it will update the cache, since conntrack flows are bi-directional?
Yes, Traffic for A -> B won't match the flowtable entry, this will update the cache.
Quoted 8 lines
Quoted 8 lines
Quoted 2 lines
That's assuming there will be A -> B traffic without B sending a request which reaches A, first.
B might send packets to A but this will not get anywhere. Assuming TCP, this will trigger retransmissions so B -> A will kick in to refresh the entry.
Is this scenario that you describe a showstopper?
Sorry I got distracted.
Quoted 6 lines
I have been discussing the topology update by tracking fdb updates with the bridge maintainer, I'll be exploring extensions to the existing
fdb_notify()infrastructure to deal with this scenario you describe. On my side this topology update scenario is not a priority to be supported in this patchset, but it's feasible to support it later on.
My concern is that invalidation is _the_ hard part of creating caches. And I feel like merging this as is would be setting our standards pretty low.
Please gather some review tags from senior netdev developers. I don't feel confident enough to apply this as 100% my own decision.
Pablo Neira Ayuso
maintainerQuoted 4 lines · Jakub Kicinski
Quoted 3 lines · On Sat, 21 Nov 2020 13:31:38 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines · Pablo Neira Ayuso
Quoted 1 line · Jakub Kicinski
On Mon, 16 Nov 2020 23:36:15 +0100 Pablo Neira Ayuso wrote:
[...]
Quoted 10 lines
Quoted 6 lines
I have been discussing the topology update by tracking fdb updates with the bridge maintainer, I'll be exploring extensions to the existing
fdb_notify()infrastructure to deal with this scenario you describe. On my side this topology update scenario is not a priority to be supported in this patchset, but it's feasible to support it later on.My concern is that invalidation is _the_ hard part of creating caches. And I feel like merging this as is would be setting our standards pretty low.
Interesting, let's summarize a bit to make sure we're on the same page:
- This "cache" is optional, you enable it on demand through ruleset.
- This "cache" is configurable, you can specify through ruleset policy what policies get into the cache and _when_ they are placed in the cache.
- This is not affecting any existing default configuration, neither Linux networking not even classic path Netfilter configurations, it's a rather new thing.
- This is showing performance improvement of ~50% with a very simple testbed. With pktgen, back few years ago I was reaching x2.5 performance boost in software in a pktgen testbed.
- This is adding minimal changes to netdev_ops, just a single callback.
For the live VM migration you describe, connections might time out, but there are many use-cases where this is still valid, some of them has been described already here.
Quoted 2 lines
Please gather some review tags from senior netdev developers. I don't feel confident enough to apply this as 100% my own decision.
Fair enough.
This requirement for very specific Netfilter infrastructure which does not affect any other Networking subsystem sounds new to me.
What senior developers specifically you would like I should poke to get an acknowledgement on this to get this accepted of your preference?
Thank you.
Jakub Kicinski
maintainerQuoted 7 lines · On Sat, 21 Nov 2020 19:56:21 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Please gather some review tags from senior netdev developers. I don't feel confident enough to apply this as 100% my own decision.
Fair enough.
This requirement for very specific Netfilter infrastructure which does not affect any other Networking subsystem sounds new to me.
You mean me asking for reviews from other senior folks when I don't feel good about some code? I've asked others the same thing in the past, e.g. Paolo for his RPS thing.
Quoted 3 lines
What senior developers specifically you would like I should poke to get an acknowledgement on this to get this accepted of your preference?
I don't want to make a list. Maybe netconf attendees are a safe bet?
Pablo Neira Ayuso
maintainerHi Jakub,
Quoted 4 lines · Jakub Kicinski
Quoted 3 lines · On Sat, 21 Nov 2020 19:56:21 +0100 Pablo Neira Ayuso wrote:
Quoted 2 lines
Please gather some review tags from senior netdev developers. I don't feel confident enough to apply this as 100% my own decision.
Quoted 8 lines
Quoted 4 lines
Fair enough.
This requirement for very specific Netfilter infrastructure which does not affect any other Networking subsystem sounds new to me.
You mean me asking for reviews from other senior folks when I don't feel good about some code? I've asked others the same thing in the past, e.g. Paolo for his RPS thing.
No, I'm perfectly fine with peer review.
Note that I am sending this to net-next as a patchset (not as a PR) _only_ because this is adding a new .ndo_fill_forward_path to netdev_ops.
That's the only thing that is relevant to the Netdev core infrastructure IMO, and this new .ndo that is private, not exposed to userspace.
Let's have a look at the diffstats again:
include/linux/netdevice.h | 35 +++ include/net/netfilter/nf_flow_table.h | 43 +++- net/8021q/vlan_dev.c | 15 ++ net/bridge/br_device.c | 27 +++ net/core/dev.c | 46 ++++ net/netfilter/nf_flow_table_core.c | 51 +++-- net/netfilter/nf_flow_table_ip.c | 200 ++++++++++++++---- net/netfilter/nft_flow_offload.c | 159 +++++++++++++- .../selftests/netfilter/nft_flowtable.sh | 82 +++++++ 9 files changed, 598 insertions(+), 60 deletions(-)
So this is adding _minimal_ changes to the NetDev infrastructure. Most of the code is an extension to the flowtable Netfilter infrastructure. And the flowtable is a cache since its conception.
I am adding the .ndo indirection to avoid the dependencies with Netfilter modules, e.g. Netfilter could use direct reference to bridge function, but that would pull in bridge modules.
Quoted 5 lines
Quoted 3 lines
What senior developers specifically you would like I should poke to get an acknowledgement on this to get this accepted of your preference?
I don't want to make a list. Maybe netconf attendees are a safe bet?
I have no idea who to ask to, traditionally it's the NetDev maintainer (AFAIK it's only you at this stage) that have the last word on something to get this merged.
I consider all developers that have reviewed this patchset to be senior developers.
Pablo Neira Ayuso
maintainerQuoted 12 lines · Tobias Waldekranz
Quoted 11 lines · Pablo Neira Ayuso
If any of the flowtable device goes down / removed, the entries are removed from the flowtable. This means packets of existing flows are pushed up back to classic bridge / forwarding path to re-evaluate the fast path.
For each new flow, the fast path that is selected freshly, so they use the up-to-date FDB to select a new bridge port.
Existing flows still follow the old path. The same happens with FIB currently.
Quoted 8 lines
Quoted 5 lines
It should be possible to explore purging entries in the flowtable that are stale due to changes in the topology (either in FDB or FIB).
What scenario do you have specifically in mind? Something like VM migrates from one bridge port to another?
This should work in the case when the bridge ports are normal NICs or switchdev ports, right?
Yes.
Quoted 14 lines
In that case, relying on link state is brittle as you can easily have a switch or a media converter between the bridge and the end-station:
br0 br0 / \ / \ eth0 eth1 eth0 eth1 / \ => / \[sw0] [sw1] [sw0] [sw1]
/ \ / \
A AIn a scenario like this, A has clearly moved. But neither eth0 nor eth1 has seen any changes in link state.
Quoted 6 lines
This particular example is a bit contrived. But this is essentially what happens in redundant topologies when reconfigurations occur (e.g. STP).
These protocols will typically signal reconfigurations to all bridges though, so as long as the affected flows are flushed at the same time as the FDB it should work.
Yes, watching the FDB should allow to clean up stale flows immediately.