1#!/bin/bash 2# SPDX-License-Identifier: GPL-2.0 3 4source lib.sh 5 6BPF_FILE="lib/xdp_dummy.bpf.o" 7readonly BASE="ns-$(mktemp -u XXXXXX)" 8readonly SRC=2 9readonly DST=1 10readonly DST_NAT=100 11readonly NS_SRC=$BASE$SRC 12readonly NS_DST=$BASE$DST 13 14# "baremetal" network used for raw UDP traffic 15readonly BM_NET_V4=192.168.1. 16readonly BM_NET_V6=2001:db8:: 17 18# "overlay" network used for UDP over UDP tunnel traffic 19readonly OL_NET_V4=172.16.1. 20readonly OL_NET_V6=2001:db8:1:: 21readonly NPROCS=`nproc` 22 23cleanup() { 24 local ns 25 local -r jobs="$(jobs -p)" 26 [ -n "${jobs}" ] && kill -1 ${jobs} 2>/dev/null 27 28 for ns in $NS_SRC $NS_DST; do 29 ip netns del $ns 2>/dev/null 30 done 31} 32 33trap cleanup EXIT 34 35create_ns() { 36 local net 37 local ns 38 39 for ns in $NS_SRC $NS_DST; do 40 ip netns add $ns 41 ip -n $ns link set dev lo up 42 43 # disable route solicitations to decrease 'noise' traffic 44 ip netns exec $ns sysctl -qw net.ipv6.conf.default.router_solicitations=0 45 ip netns exec $ns sysctl -qw net.ipv6.conf.all.router_solicitations=0 46 done 47 48 ip link add name veth$SRC type veth peer name veth$DST 49 50 for ns in $SRC $DST; do 51 ip link set dev veth$ns netns $BASE$ns 52 ip -n $BASE$ns link set dev veth$ns up 53 ip -n $BASE$ns addr add dev veth$ns $BM_NET_V4$ns/24 54 ip -n $BASE$ns addr add dev veth$ns $BM_NET_V6$ns/64 nodad 55 done 56 ip -n $NS_DST link set veth$DST xdp object ${BPF_FILE} section xdp 2>/dev/null 57} 58 59create_vxlan_endpoint() { 60 local -r netns=$1 61 local -r bm_dev=$2 62 local -r bm_rem_addr=$3 63 local -r vxlan_dev=$4 64 local -r vxlan_id=$5 65 local -r vxlan_port=4789 66 67 ip -n $netns link set dev $bm_dev up 68 ip -n $netns link add dev $vxlan_dev type vxlan id $vxlan_id \ 69 dstport $vxlan_port remote $bm_rem_addr 70 ip -n $netns link set dev $vxlan_dev up 71} 72 73create_vxlan_pair() { 74 local ns 75 76 create_ns 77 78 for ns in $SRC $DST; do 79 # note that 3 - $SRC == $DST and 3 - $DST == $SRC 80 create_vxlan_endpoint $BASE$ns veth$ns $BM_NET_V4$((3 - $ns)) vxlan$ns 4 81 ip -n $BASE$ns addr add dev vxlan$ns $OL_NET_V4$ns/24 82 done 83 for ns in $SRC $DST; do 84 create_vxlan_endpoint $BASE$ns veth$ns $BM_NET_V6$((3 - $ns)) vxlan6$ns 6 85 ip -n $BASE$ns addr add dev vxlan6$ns $OL_NET_V6$ns/24 nodad 86 done 87 88 # preload neighbur cache, do avoid some noisy traffic 89 local addr_dst=$(ip -j -n $BASE$DST link show dev vxlan6$DST |jq -r '.[]["address"]') 90 local addr_src=$(ip -j -n $BASE$SRC link show dev vxlan6$SRC |jq -r '.[]["address"]') 91 ip -n $BASE$DST neigh add dev vxlan6$DST lladdr $addr_src $OL_NET_V6$SRC 92 ip -n $BASE$SRC neigh add dev vxlan6$SRC lladdr $addr_dst $OL_NET_V6$DST 93} 94 95is_ipv6() { 96 if [[ $1 =~ .*:.* ]]; then 97 return 0 98 fi 99 return 1 100} 101 102run_test() { 103 local -r msg=$1 104 local -r dst=$2 105 local -r pkts=$3 106 local -r vxpkts=$4 107 local bind=$5 108 local rx_args="" 109 local rx_family="-4" 110 local family=-4 111 local filter=IpInReceives 112 local ipt=iptables 113 114 printf "%-40s" "$msg" 115 116 if is_ipv6 $dst; then 117 # rx program does not support '-6' and implies ipv6 usage by default 118 rx_family="" 119 family=-6 120 filter=Ip6InReceives 121 ipt=ip6tables 122 fi 123 124 rx_args="$rx_family" 125 [ -n "$bind" ] && rx_args="$rx_args -b $bind" 126 127 # send a single GSO packet, segmented in 10 UDP frames. 128 # Always expect 10 UDP frames on RX side as rx socket does 129 # not enable GRO 130 ip netns exec $NS_DST $ipt -A INPUT -p udp --dport 4789 131 ip netns exec $NS_DST $ipt -A INPUT -p udp --dport 8000 132 ip netns exec $NS_DST ./udpgso_bench_rx -C 2000 -R 100 -n 10 -l 1300 $rx_args & 133 local spid=$! 134 wait_local_port_listen "$NS_DST" 8000 udp 135 ip netns exec $NS_SRC ./udpgso_bench_tx $family -M 1 -s 13000 -S 1300 -D $dst 136 local retc=$? 137 wait $spid 138 local rets=$? 139 if [ ${rets} -ne 0 ] || [ ${retc} -ne 0 ]; then 140 echo " fail client exit code $retc, server $rets" 141 ret=1 142 return 143 fi 144 145 local rcv=`ip netns exec $NS_DST $ipt"-save" -c | grep 'dport 8000' | \ 146 sed -e 's/\[//' -e 's/:.*//'` 147 if [ $rcv != $pkts ]; then 148 echo " fail - received $rcv packets, expected $pkts" 149 ret=1 150 return 151 fi 152 153 local vxrcv=`ip netns exec $NS_DST $ipt"-save" -c | grep 'dport 4789' | \ 154 sed -e 's/\[//' -e 's/:.*//'` 155 156 # upper net can generate a little noise, allow some tolerance 157 if [ $vxrcv -lt $vxpkts -o $vxrcv -gt $((vxpkts + 3)) ]; then 158 echo " fail - received $vxrcv vxlan packets, expected $vxpkts" 159 ret=1 160 return 161 fi 162 echo " ok" 163} 164 165run_test_csum() { 166 local -r msg="$1" 167 local -r dst="$2" 168 local csum_error_filter=UdpInCsumErrors 169 local csum_errors 170 171 printf "%-40s" "$msg" 172 173 is_ipv6 "$dst" && csum_error_filter=Udp6InCsumErrors 174 175 ip netns exec "$NS_DST" iperf3 -s -1 >/dev/null & 176 wait_local_port_listen "$NS_DST" 5201 tcp 177 local spid="$!" 178 ip netns exec "$NS_SRC" iperf3 -c "$dst" -t 2 >/dev/null 179 local retc="$?" 180 wait "$spid" 181 local rets="$?" 182 if [ "$rets" -ne 0 ] || [ "$retc" -ne 0 ]; then 183 echo " fail client exit code $retc, server $rets" 184 ret=1 185 return 186 fi 187 188 csum_errors=$(ip netns exec "$NS_DST" nstat -as "$csum_error_filter" | 189 grep "$csum_error_filter" | awk '{print $2}') 190 if [ -n "$csum_errors" ] && [ "$csum_errors" -gt 0 ]; then 191 echo " fail - csum error on receive $csum_errors, expected 0" 192 ret=1 193 return 194 fi 195 echo " ok" 196} 197 198run_bench() { 199 local -r msg=$1 200 local -r dst=$2 201 local family=-4 202 203 printf "%-40s" "$msg" 204 if [ $NPROCS -lt 2 ]; then 205 echo " skip - needed 2 CPUs found $NPROCS" 206 return 207 fi 208 209 is_ipv6 $dst && family=-6 210 211 # bind the sender and the receiver to different CPUs to try 212 # get reproducible results 213 ip netns exec $NS_DST bash -c "echo 2 > /sys/class/net/veth$DST/queues/rx-0/rps_cpus" 214 ip netns exec $NS_DST taskset 0x2 ./udpgso_bench_rx -C 2000 -R 100 & 215 local spid=$! 216 wait_local_port_listen "$NS_DST" 8000 udp 217 ip netns exec $NS_SRC taskset 0x1 ./udpgso_bench_tx $family -l 3 -S 1300 -D $dst 218 local retc=$? 219 wait $spid 220 local rets=$? 221 if [ ${rets} -ne 0 ] || [ ${retc} -ne 0 ]; then 222 echo " fail client exit code $retc, server $rets" 223 ret=1 224 return 225 fi 226} 227 228for family in 4 6; do 229 BM_NET=$BM_NET_V4 230 OL_NET=$OL_NET_V4 231 IPT=iptables 232 SUFFIX=24 233 VXDEV=vxlan 234 PING=ping 235 236 if [ $family = 6 ]; then 237 BM_NET=$BM_NET_V6 238 OL_NET=$OL_NET_V6 239 SUFFIX="64 nodad" 240 VXDEV=vxlan6 241 IPT=ip6tables 242 # Use ping6 on systems where ping doesn't handle IPv6 243 ping -w 1 -c 1 ::1 > /dev/null 2>&1 || PING="ping6" 244 fi 245 246 echo "IPv$family" 247 248 create_ns 249 run_test "No GRO" $BM_NET$DST 10 0 250 cleanup 251 252 create_ns 253 ip netns exec $NS_DST ethtool -K veth$DST generic-receive-offload on 254 ip netns exec $NS_DST ethtool -K veth$DST rx-gro-list on 255 run_test "GRO frag list" $BM_NET$DST 1 0 256 cleanup 257 258 # UDP GRO fwd skips aggregation when find an udp socket with the GRO option 259 # if there is an UDP tunnel in the running system, such lookup happen 260 # take place. 261 # use NAT to circumvent GRO FWD check 262 create_ns 263 ip -n $NS_DST addr add dev veth$DST $BM_NET$DST_NAT/$SUFFIX 264 ip netns exec $NS_DST ethtool -K veth$DST generic-receive-offload on 265 ip netns exec $NS_DST ethtool -K veth$DST rx-udp-gro-forwarding on 266 ip netns exec $NS_DST $IPT -t nat -I PREROUTING -d $BM_NET$DST_NAT \ 267 -j DNAT --to-destination $BM_NET$DST 268 run_test "GRO fwd" $BM_NET$DST_NAT 1 0 $BM_NET$DST 269 cleanup 270 271 create_ns 272 run_bench "UDP fwd perf" $BM_NET$DST 273 ip netns exec $NS_DST ethtool -K veth$DST rx-udp-gro-forwarding on 274 run_bench "UDP GRO fwd perf" $BM_NET$DST 275 cleanup 276 277 create_vxlan_pair 278 ip netns exec $NS_DST ethtool -K veth$DST generic-receive-offload on 279 ip netns exec $NS_DST ethtool -K veth$DST rx-gro-list on 280 run_test "GRO frag list over UDP tunnel" $OL_NET$DST 10 10 281 cleanup 282 283 # use NAT to circumvent GRO FWD check 284 create_vxlan_pair 285 ip -n $NS_DST addr add dev $VXDEV$DST $OL_NET$DST_NAT/$SUFFIX 286 ip netns exec $NS_DST ethtool -K veth$DST generic-receive-offload on 287 ip netns exec $NS_DST ethtool -K veth$DST rx-udp-gro-forwarding on 288 ip netns exec $NS_DST $IPT -t nat -I PREROUTING -d $OL_NET$DST_NAT \ 289 -j DNAT --to-destination $OL_NET$DST 290 291 # load arp cache before running the test to reduce the amount of 292 # stray traffic on top of the UDP tunnel 293 ip netns exec $NS_SRC $PING -q -c 1 $OL_NET$DST_NAT >/dev/null 294 run_test "GRO fwd over UDP tunnel" $OL_NET$DST_NAT 10 10 $OL_NET$DST 295 cleanup 296 297 # force segmentation and re-aggregation 298 create_vxlan_pair 299 ip netns exec "$NS_DST" ethtool -K veth"$DST" generic-receive-offload on 300 ip netns exec "$NS_SRC" ethtool -K veth"$SRC" tso off 301 ip -n "$NS_SRC" link set dev veth"$SRC" mtu 1430 302 303 # forward to a 2nd veth pair 304 ip -n "$NS_DST" link add br0 type bridge 305 ip -n "$NS_DST" link set dev veth"$DST" master br0 306 307 # segment the aggregated TSO packet, without csum offload 308 ip -n "$NS_DST" link add veth_segment type veth peer veth_rx 309 for FEATURE in tso tx-udp-segmentation tx-checksumming; do 310 ip netns exec "$NS_DST" ethtool -K veth_segment "$FEATURE" off 311 done 312 ip -n "$NS_DST" link set dev veth_segment master br0 up 313 ip -n "$NS_DST" link set dev br0 up 314 ip -n "$NS_DST" link set dev veth_rx up 315 316 # move the lower layer IP in the last added veth 317 for ADDR in "$BM_NET_V4$DST/24" "$BM_NET_V6$DST/64"; do 318 # the dad argument will let iproute emit a unharmful warning 319 # with ipv4 addresses 320 ip -n "$NS_DST" addr del dev veth"$DST" "$ADDR" 321 ip -n "$NS_DST" addr add dev veth_rx "$ADDR" \ 322 nodad 2>/dev/null 323 done 324 325 run_test_csum "GSO after GRO" "$OL_NET$DST" 326 cleanup 327done 328 329exit $ret 330