diff options
author | Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk> | 2020-05-29 16:29:53 +0100 |
---|---|---|
committer | Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk> | 2020-06-05 21:03:26 +0100 |
commit | 7b4877c2040a63332cd1f10023c51698fec2ed98 (patch) | |
tree | 80e72dcf92dc71419fe672622b4cc66ec9605a25 /target/linux/generic | |
parent | 898969636d873c999fc905efa3c5a9973be736d8 (diff) | |
download | upstream-7b4877c2040a63332cd1f10023c51698fec2ed98.tar.gz upstream-7b4877c2040a63332cd1f10023c51698fec2ed98.tar.bz2 upstream-7b4877c2040a63332cd1f10023c51698fec2ed98.zip |
kernel: sch_cake: use skb hash improve wireguard compatibility
While the other fq-based qdiscs take advantage of skb->hash and doesn't
recompute it if it is already set, sch_cake does not.
This was a deliberate choice because sch_cake hashes various parts of the
packet header to support its advanced flow isolation modes. However,
foregoing the use of skb->hash entirely loses a few important benefits:
- When skb->hash is set by hardware, a few CPU cycles can be saved by not
hashing again in software.
- Tunnel encapsulations will generally preserve the value of skb->hash from
before the encapsulation, which allows flow-based qdiscs to distinguish
between flows even though the outer packet header no longer has flow
information.
It turns out that we can preserve these desirable properties in many cases,
while still supporting the advanced flow isolation properties of sch_cake.
This patch does so by reusing the skb->hash value as the flow_hash part of
the hashing procedure in cake_hash() only in the following conditions:
- If the skb->hash is marked as covering the flow headers (skb->l4_hash is
set)
AND
- NAT header rewriting is either disabled, or did not change any values
used for hashing. The latter is important to match local-origin packets
such as those of a tunnel endpoint.
The immediate motivation for fixing this was the recent patch to WireGuard
to preserve the skb->hash on encapsulation. As such, this is also what I
tested against; with this patch, added latency under load for competing
flows drops from ~8 ms to sub-1ms on an RRUL test over a WireGuard tunnel
going through a virtual link shaped to 1Gbps using sch_cake. This matches
the results we saw with a similar setup using sch_fq_codel when testing the
WireGuard patch.
Fixes: 046f6fd5daef ("sched: Add Common Applications Kept Enhanced (cake) qdisc")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Diffstat (limited to 'target/linux/generic')
-rw-r--r-- | target/linux/generic/backport-5.4/398-v5.9-net-sch_cake-Take-advantage-of-skb-hash-where-appropriate.patch | 170 |
1 files changed, 170 insertions, 0 deletions
diff --git a/target/linux/generic/backport-5.4/398-v5.9-net-sch_cake-Take-advantage-of-skb-hash-where-appropriate.patch b/target/linux/generic/backport-5.4/398-v5.9-net-sch_cake-Take-advantage-of-skb-hash-where-appropriate.patch new file mode 100644 index 0000000000..7b3396c6c3 --- /dev/null +++ b/target/linux/generic/backport-5.4/398-v5.9-net-sch_cake-Take-advantage-of-skb-hash-where-appropriate.patch @@ -0,0 +1,170 @@ +From b0c19ed6088ab41dd2a727b60594b7297c15d6ce Mon Sep 17 00:00:00 2001 +From: =?UTF-8?q?Toke=20H=C3=B8iland-J=C3=B8rgensen?= <toke@redhat.com> +Date: Fri, 29 May 2020 14:43:44 +0200 +Subject: [PATCH] sch_cake: Take advantage of skb->hash where appropriate +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +While the other fq-based qdiscs take advantage of skb->hash and doesn't +recompute it if it is already set, sch_cake does not. + +This was a deliberate choice because sch_cake hashes various parts of the +packet header to support its advanced flow isolation modes. However, +foregoing the use of skb->hash entirely loses a few important benefits: + +- When skb->hash is set by hardware, a few CPU cycles can be saved by not + hashing again in software. + +- Tunnel encapsulations will generally preserve the value of skb->hash from + before the encapsulation, which allows flow-based qdiscs to distinguish + between flows even though the outer packet header no longer has flow + information. + +It turns out that we can preserve these desirable properties in many cases, +while still supporting the advanced flow isolation properties of sch_cake. +This patch does so by reusing the skb->hash value as the flow_hash part of +the hashing procedure in cake_hash() only in the following conditions: + +- If the skb->hash is marked as covering the flow headers (skb->l4_hash is + set) + +AND + +- NAT header rewriting is either disabled, or did not change any values + used for hashing. The latter is important to match local-origin packets + such as those of a tunnel endpoint. + +The immediate motivation for fixing this was the recent patch to WireGuard +to preserve the skb->hash on encapsulation. As such, this is also what I +tested against; with this patch, added latency under load for competing +flows drops from ~8 ms to sub-1ms on an RRUL test over a WireGuard tunnel +going through a virtual link shaped to 1Gbps using sch_cake. This matches +the results we saw with a similar setup using sch_fq_codel when testing the +WireGuard patch. + +Fixes: 046f6fd5daef ("sched: Add Common Applications Kept Enhanced (cake) qdisc") +Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com> +Signed-off-by: David S. Miller <davem@davemloft.net> +Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk> +--- + net/sched/sch_cake.c | 65 ++++++++++++++++++++++++++++++++++---------- + 1 file changed, 51 insertions(+), 14 deletions(-) + +--- a/net/sched/sch_cake.c ++++ b/net/sched/sch_cake.c +@@ -585,26 +585,48 @@ static bool cobalt_should_drop(struct co + return drop; + } + +-static void cake_update_flowkeys(struct flow_keys *keys, ++static bool cake_update_flowkeys(struct flow_keys *keys, + const struct sk_buff *skb) + { + #if IS_ENABLED(CONFIG_NF_CONNTRACK) + struct nf_conntrack_tuple tuple = {}; +- bool rev = !skb->_nfct; ++ bool rev = !skb->_nfct, upd = false; ++ __be32 ip; + + if (tc_skb_protocol(skb) != htons(ETH_P_IP)) +- return; ++ return false; + + if (!nf_ct_get_tuple_skb(&tuple, skb)) +- return; ++ return false; + +- keys->addrs.v4addrs.src = rev ? tuple.dst.u3.ip : tuple.src.u3.ip; +- keys->addrs.v4addrs.dst = rev ? tuple.src.u3.ip : tuple.dst.u3.ip; ++ ip = rev ? tuple.dst.u3.ip : tuple.src.u3.ip; ++ if (ip != keys->addrs.v4addrs.src) { ++ keys->addrs.v4addrs.src = ip; ++ upd = true; ++ } ++ ip = rev ? tuple.src.u3.ip : tuple.dst.u3.ip; ++ if (ip != keys->addrs.v4addrs.dst) { ++ keys->addrs.v4addrs.dst = ip; ++ upd = true; ++ } + + if (keys->ports.ports) { +- keys->ports.src = rev ? tuple.dst.u.all : tuple.src.u.all; +- keys->ports.dst = rev ? tuple.src.u.all : tuple.dst.u.all; ++ __be16 port; ++ ++ port = rev ? tuple.dst.u.all : tuple.src.u.all; ++ if (port != keys->ports.src) { ++ keys->ports.src = port; ++ upd = true; ++ } ++ port = rev ? tuple.src.u.all : tuple.dst.u.all; ++ if (port != keys->ports.dst) { ++ port = keys->ports.dst; ++ upd = true; ++ } + } ++ return upd; ++#else ++ return false; + #endif + } + +@@ -625,23 +647,36 @@ static bool cake_ddst(int flow_mode) + static u32 cake_hash(struct cake_tin_data *q, const struct sk_buff *skb, + int flow_mode, u16 flow_override, u16 host_override) + { ++ bool hash_flows = (!flow_override && !!(flow_mode & CAKE_FLOW_FLOWS)); ++ bool hash_hosts = (!host_override && !!(flow_mode & CAKE_FLOW_HOSTS)); ++ bool nat_enabled = !!(flow_mode & CAKE_FLOW_NAT_FLAG); + u32 flow_hash = 0, srchost_hash = 0, dsthost_hash = 0; + u16 reduced_hash, srchost_idx, dsthost_idx; + struct flow_keys keys, host_keys; ++ bool use_skbhash = skb->l4_hash; + + if (unlikely(flow_mode == CAKE_FLOW_NONE)) + return 0; + +- /* If both overrides are set we can skip packet dissection entirely */ +- if ((flow_override || !(flow_mode & CAKE_FLOW_FLOWS)) && +- (host_override || !(flow_mode & CAKE_FLOW_HOSTS))) ++ /* If both overrides are set, or we can use the SKB hash and nat mode is ++ * disabled, we can skip packet dissection entirely. If nat mode is ++ * enabled there's another check below after doing the conntrack lookup. ++ */ ++ if ((!hash_flows || (use_skbhash && !nat_enabled)) && !hash_hosts) + goto skip_hash; + + skb_flow_dissect_flow_keys(skb, &keys, + FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL); + +- if (flow_mode & CAKE_FLOW_NAT_FLAG) +- cake_update_flowkeys(&keys, skb); ++ /* Don't use the SKB hash if we change the lookup keys from conntrack */ ++ if (nat_enabled && cake_update_flowkeys(&keys, skb)) ++ use_skbhash = false; ++ ++ /* If we can still use the SKB hash and don't need the host hash, we can ++ * skip the rest of the hashing procedure ++ */ ++ if (use_skbhash && !hash_hosts) ++ goto skip_hash; + + /* flow_hash_from_keys() sorts the addresses by value, so we have + * to preserve their order in a separate data structure to treat +@@ -680,12 +715,14 @@ static u32 cake_hash(struct cake_tin_dat + /* This *must* be after the above switch, since as a + * side-effect it sorts the src and dst addresses. + */ +- if (flow_mode & CAKE_FLOW_FLOWS) ++ if (hash_flows && !use_skbhash) + flow_hash = flow_hash_from_keys(&keys); + + skip_hash: + if (flow_override) + flow_hash = flow_override - 1; ++ else if (use_skbhash) ++ flow_hash = skb->hash; + if (host_override) { + dsthost_hash = host_override - 1; + srchost_hash = host_override - 1; |