[xdp]refactor offload=pktsw,csum and optimize code

Makefile

@@ -14,7 +14,7 @@ endif
 ifeq ($(OS),Linux)
 LINUX_ARCH = $(ARCH)
 else
-LINUX_ARCH = $(shell docker run --rm vproxyio/compile:latest uname -m)
+LINUX_ARCH := $(shell docker run --rm vproxyio/compile:latest uname -m)
 endif
 #
 ifeq ($(LINUX_ARCH),arm64)
@@ -86,7 +86,7 @@ _add_linux_so_to_zip:
 	mkdir -p ./io/vproxy/
 	cp ./base/src/main/c/libvfdposix.so ./io/vproxy/libvfdposix-$(LINUX_ARCH).so
 	cp ./submodules/vpxdp/libvpxdp.so ./io/vproxy/libvpxdp-$(LINUX_ARCH).so
-	cp ./submodules/xdp-tools/lib/libxdp/libxdp.1.4.0 ./io/vproxy/libxdp-$(LINUX_ARCH).so
+	cp ./submodules/vpxdp/submodules/xdp-tools/lib/libxdp/libxdp.so.1.4.0 ./io/vproxy/libxdp-$(LINUX_ARCH).so
 	cp ./libmsquic.so ./io/vproxy/libmsquic-$(LINUX_ARCH).so
 	cp ./base/src/main/c/libmsquic-java.so ./io/vproxy/libmsquic-java-$(LINUX_ARCH).so
 	cp ./submodules/fubuki/target/release/libfubukil.so ./io/vproxy/libfubuki-$(LINUX_ARCH).so
@@ -118,6 +118,8 @@ endif
 .PHONY: jar-with-lib
 ifeq ($(OS),Linux)
 jar-with-lib: clean jar native _add_linux_so_to_zip
+.PHONY: jar-with-lib-skip-native
+jar-with-lib-skip-native: clean-jar jar _add_linux_so_to_zip
 else
 jar-with-lib: clean jar native _add_linux_so_to_zip jar-with-lib-skip-native
 .PHONY: jar-with-lib-no-docker

app/src/main/java/io/vproxy/app/app/cmd/Flag.java

@@ -16,7 +16,6 @@ public enum Flag {
     disable("disable"),
 
     rxgencsum("rx-gen-csum"),
-    offload("offload"),
     ;
     public final String fullname;
 

app/src/main/java/io/vproxy/app/app/cmd/HelpCommand.java

@@ -1,6 +1,7 @@
 package io.vproxy.app.app.cmd;
 
 import io.vproxy.app.app.Application;
+import io.vproxy.app.app.cmd.handle.param.OffloadHandle;
 import io.vproxy.app.app.cmd.handle.resource.SwitchHandle;
 import io.vproxy.base.Config;
 import io.vproxy.base.util.coll.Tuple;
@@ -447,6 +448,7 @@ public enum ParamMan {
         framesize("frame-size", null, "size of a frame"),
         busypoll("busy-poll", null, "a number indicating whether to enable busy poll, " +
             "and may set the SO_BUSY_POLL_BUDGET as well"),
+        offload("offload", null, "specify offload operations"),
         ip("ip", null, "ip address or ip/mask"),
         iface("iface", null, "connected interface in switch"),
         ;
@@ -470,7 +472,6 @@ public enum FlagMan {
         force("force", null, "forcibly to do something"),
         zerocopy("zerocopy", null, "indicate to perform zerocopy operations"),
         rxgencsum("rx-gen-csum", null, "generate checksum before receiving the packet into vswitch"),
-        offload("offload", null, "offload operations from java"),
         enable("enable", null, "enable the resource"),
         disable("disable", null, "disable the resource"),
         ;
@@ -1372,15 +1373,17 @@ public enum ResMan {
                 new ResActMan(ActMan.addto, "add xdp socket into the switch", Arrays.asList(
                     new ResActParamMan(ParamMan.umem, "umem for the xdp socket to use. See `umem` for more info"),
                     new ResActParamMan(ParamMan.queue, "the queue index to bind to"),
-                    new ResActParamMan(ParamMan.rxringsize, "rx ring size", "" + SwitchUtils.RX_TX_CHUNKS),
-                    new ResActParamMan(ParamMan.txringsize, "tx ring size", "" + SwitchUtils.RX_TX_CHUNKS),
+                    new ResActParamMan(ParamMan.rxringsize, "rx ring size", "" + SwitchUtils.DEFAULT_RX_TX_CHUNKS),
+                    new ResActParamMan(ParamMan.txringsize, "tx ring size", "" + SwitchUtils.DEFAULT_RX_TX_CHUNKS),
                     new ResActParamMan(ParamMan.mode, "mode of the xsk, enum: {SKB, DRIVER}, see doc for more info", "" + BPFMode.SKB),
                     new ResActParamMan(ParamMan.busypoll, "whether to enable busy poll, and set SO_BUSY_POLL_BUDGET. Set this option to 0 to disable busy poll", "0"),
-                    new ResActParamMan(ParamMan.vrf, "vrf which the iface is assigned to")
+                    new ResActParamMan(ParamMan.vrf, "vrf which the iface is assigned to"),
+                    new ResActParamMan(ParamMan.offload, "support " + OffloadHandle.PACKET_SWITCHING + "," + OffloadHandle.CHECKSUM + ". " +
+                                                         "Note that checksum offloading requires kernel version >= 6.8 and not all drivers support this feature",
+                        "(none)")
                 ), Arrays.asList(
                     new ResActFlagMan(FlagMan.zerocopy, "allow kernel to use zerocopy machanism", false),
-                    new ResActFlagMan(FlagMan.rxgencsum, "generate checksum in native code before receiving the packet in java", false),
-                    new ResActFlagMan(FlagMan.offload, "offload mac switching to xdp program", false)
+                    new ResActFlagMan(FlagMan.rxgencsum, "generate checksum in native code before receiving the packet in java", false)
                 ), Arrays.asList(
                     new Tuple<>(
                         "add xdp xdptut-4667 to switch sw0 umem umem0 queue 0 rx-ring-size 2048 tx-ring-size 2048 mode SKB vrf 1 zerocopy",
@@ -1487,9 +1490,9 @@ public enum ResMan {
         umem("umem", null, "umem for xdp sockets to use",
             Arrays.asList(
                 new ResActMan(ActMan.addto, "add a umem to a switch", Arrays.asList(
-                    new ResActParamMan(ParamMan.chunks, "how many chunks are there in this umem", "" + (SwitchUtils.RX_TX_CHUNKS * 2)),
-                    new ResActParamMan(ParamMan.fillringsize, "size of the fill ring", "" + SwitchUtils.RX_TX_CHUNKS),
-                    new ResActParamMan(ParamMan.compringsize, "size of the comp ring", "" + SwitchUtils.RX_TX_CHUNKS),
+                    new ResActParamMan(ParamMan.chunks, "how many chunks are there in this umem", "" + (SwitchUtils.DEFAULT_UMEM_CHUNKS)),
+                    new ResActParamMan(ParamMan.fillringsize, "size of the fill ring", "" + SwitchUtils.DEFAULT_RX_TX_CHUNKS),
+                    new ResActParamMan(ParamMan.compringsize, "size of the comp ring", "" + SwitchUtils.DEFAULT_RX_TX_CHUNKS),
                     new ResActParamMan(ParamMan.framesize, "size of the frame, must be 2048 or 4096", "" + SwitchUtils.TOTAL_RCV_BUF_LEN / 2)
                 ), Arrays.asList(
                     new Tuple<>(

app/src/main/java/io/vproxy/app/app/cmd/ModuleCommands.kt

@@ -919,11 +919,11 @@ class ModuleCommands private constructor() : Commands() {
           it + ResActParam(Param.mode) { BPFModeHandle.check(it) }
           it + ResActParam(Param.busypoll) { BusyPollHandle.check(it) }
           it + ResActParam(Param.vrf, required) { VrfParamHandle.check(it) }
+          it + ResActParam(Param.offload) { OffloadHandle.check(it) }
         },
         flags = {
           it + ResActFlag(Flag.zerocopy)
           it + ResActFlag(Flag.rxgencsum)
-          it + ResActFlag(Flag.offload)
         },
         exec = execUpdate { XDPHandle.add(it) }
       )

app/src/main/java/io/vproxy/app/app/cmd/Param.java

@@ -60,6 +60,7 @@ public enum Param {
     compringsize("comp-ring-size"),
     framesize("frame-size"),
     busypoll("busy-poll"),
+    offload("offload"),
 
     url("url"),
     cls("class"),

app/src/main/java/io/vproxy/app/app/cmd/handle/param/OffloadHandle.java

@@ -0,0 +1,53 @@
+package io.vproxy.app.app.cmd.handle.param;
+
+import io.vproxy.app.app.cmd.Command;
+import io.vproxy.app.app.cmd.Param;
+
+import java.util.Arrays;
+
+public class OffloadHandle {
+    private OffloadHandle() {
+    }
+
+    public static final String PACKET_SWITCHING = "pktsw";
+    public static final String CHECKSUM = "csum";
+
+    public static boolean check(Command cmd, String... available) {
+        if (!cmd.args.containsKey(Param.offload)) {
+            return true;
+        }
+        var lsAvailable = Arrays.asList(available);
+        var offload = cmd.args.get(Param.offload);
+        var split = offload.split(",");
+        for (var s : split) {
+            s = s.trim();
+            if (!lsAvailable.contains(s)) {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    public static boolean isPacketSwitchingOffloaded(Command cmd) {
+        return isOffloaded(cmd, PACKET_SWITCHING);
+    }
+
+    public static boolean isChecksumOffloaded(Command cmd) {
+        return isOffloaded(cmd, CHECKSUM);
+    }
+
+    private static boolean isOffloaded(Command cmd, String match) {
+        var offload = cmd.args.get(Param.offload);
+        if (offload == null) {
+            return false;
+        }
+        var split = offload.split(",");
+        for (var s : split) {
+            s = s.trim();
+            if (s.equals(match)) {
+                return true;
+            }
+        }
+        return false;
+    }
+}

app/src/main/java/io/vproxy/app/app/cmd/handle/resource/ArpHandle.java

@@ -83,18 +83,18 @@ public static List<ArpEntry> list(Resource parent) throws Exception {
             macInArpEntries.add(a.mac);
             var macEntry = macEntriesMap.get(a.mac);
             if (macEntry == null) {
-                result.add(new ArpEntry(a.mac, a.ip, null, a.getTTL(), -1, false));
+                result.add(new ArpEntry(a.mac, a.ip, null, a.getTTL(), -1, false, 0));
             } else {
                 var iface = macEntry.iface;
                 var ttl = macEntry.getTTL();
-                result.add(new ArpEntry(a.mac, a.ip, iface, a.getTTL(), ttl, macEntry.isOffloaded()));
+                result.add(new ArpEntry(a.mac, a.ip, iface, a.getTTL(), ttl, macEntry.isOffloaded(), macEntry.getOffloadedCount()));
             }
         }
         for (var m : macEntries) {
             if (macInArpEntries.contains(m.mac)) {
                 continue;
             }
-            result.add(new ArpEntry(m.mac, null, m.iface, -1, m.getTTL(), m.isOffloaded()));
+            result.add(new ArpEntry(m.mac, null, m.iface, -1, m.getTTL(), m.isOffloaded(), m.getOffloadedCount()));
         }
         result.sort((a, b) -> {
             // sort by mac
@@ -144,23 +144,9 @@ public static void remove(Command cmd) throws Exception {
         net.arpTable.remove(mac);
     }
 
-    public static class ArpEntry {
-        public final MacAddress mac;
-        public final IP ip;
-        public final Iface iface;
-        public final long arpTTL;
-        public final long macTTL;
-        public final boolean offloaded;
-
-        public ArpEntry(MacAddress mac, IP ip, Iface iface, long arpTTL, long macTTL, boolean offloaded) {
-            this.mac = mac;
-            this.ip = ip;
-            this.iface = iface;
-            this.arpTTL = arpTTL;
-            this.macTTL = macTTL;
-            this.offloaded = offloaded;
-        }
-
+    public record ArpEntry(
+        MacAddress mac, IP ip, Iface iface, long arpTTL, long macTTL, boolean offloaded, int offloadedCount
+    ) {
         private String strForIp(ArpEntry e) {
             if (e.ip == null) return "";
             return e.ip.formatToIPString();
@@ -236,7 +222,7 @@ public String toString(List<ArpEntry> all) {
             String strMacTTL = strForMacTTL(this);
             sb.append(split).append("MAC-TTL:").append(strMacTTL);
             if (offloaded) {
-                sb.append("/offload");
+                sb.append("/offload(").append(offloadedCount).append(")");
             }
             return sb.toString();
         }

app/src/main/java/io/vproxy/app/app/cmd/handle/resource/UMemHandle.java

@@ -31,9 +31,9 @@ public static UMem get(String alias, String swAlias) throws NotFoundException {
     }
 
     public static void add(Command cmd) throws Exception {
-        int chunksSize = RingSizeHandle.get(cmd, Param.chunks, SwitchUtils.RX_TX_CHUNKS * 2);
-        int fillRingSize = RingSizeHandle.get(cmd, Param.fillringsize, SwitchUtils.RX_TX_CHUNKS);
-        int compRingSize = RingSizeHandle.get(cmd, Param.compringsize, SwitchUtils.RX_TX_CHUNKS);
+        int chunksSize = RingSizeHandle.get(cmd, Param.chunks, SwitchUtils.DEFAULT_UMEM_CHUNKS);
+        int fillRingSize = RingSizeHandle.get(cmd, Param.fillringsize, SwitchUtils.DEFAULT_RX_TX_CHUNKS);
+        int compRingSize = RingSizeHandle.get(cmd, Param.compringsize, SwitchUtils.DEFAULT_RX_TX_CHUNKS);
         int frameSize = FrameSizeHandle.get(cmd, SwitchUtils.TOTAL_RCV_BUF_LEN / 2);
 
         Switch sw = Application.get().switchHolder.get(cmd.prepositionResource.alias);

app/src/main/java/io/vproxy/app/app/cmd/handle/resource/XDPHandle.java

@@ -21,16 +21,17 @@ public static void add(Command cmd) throws Exception {
         UMem umem = UMemHandle.get(cmd.args.get(Param.umem), sw.alias);
 
         int queueId = QueueHandle.get(cmd);
-        int rxRingSize = RingSizeHandle.get(cmd, Param.rxringsize, SwitchUtils.RX_TX_CHUNKS);
-        int txRingSize = RingSizeHandle.get(cmd, Param.txringsize, SwitchUtils.RX_TX_CHUNKS);
+        int rxRingSize = RingSizeHandle.get(cmd, Param.rxringsize, SwitchUtils.DEFAULT_RX_TX_CHUNKS);
+        int txRingSize = RingSizeHandle.get(cmd, Param.txringsize, SwitchUtils.DEFAULT_RX_TX_CHUNKS);
         boolean zeroCopy = cmd.flags.contains(Flag.zerocopy);
         int busyPollBudget = 0;
         if (cmd.args.containsKey(Param.busypoll)) {
             busyPollBudget = BusyPollHandle.get(cmd);
         }
         boolean rxGenChecksum = cmd.flags.contains(Flag.rxgencsum);
         int vrf = VrfParamHandle.get(cmd);
-        boolean offload = cmd.flags.contains(Flag.offload);
+        boolean pktswOffloaded = OffloadHandle.isPacketSwitchingOffloaded(cmd);
+        boolean csumOffloaded = OffloadHandle.isChecksumOffloaded(cmd);
 
         var nic = cmd.resource.alias;
         var mode = BPFModeHandle.get(cmd, BPFMode.SKB);
@@ -41,12 +42,26 @@ public static void add(Command cmd) throws Exception {
 
         try {
             var xskMap = bpfobj.getMap(Prebuilt.DEFAULT_XSKS_MAP_NAME);
-            var macMap = createResult.map();
+            var mac2portMap = createResult.mac2portMap();
+            var port2devMap = createResult.port2devMap();
             var srcmac2countMap = bpfobj.getMap(Prebuilt.DEFAULT_SRC_MAC_TO_COUNT_MAP_NAME);
+            var passmacMap = bpfobj.getMap(Prebuilt.DEFAULT_PASS_MAC_MAP_NAME);
 
-            sw.addXDP(nic, vrf, umem, new XDPIface.XDPParams(
-                queueId, rxRingSize, txRingSize, mode, zeroCopy, busyPollBudget, rxGenChecksum, offload,
-                new XDPIface.BPFInfo(bpfobj, xskMap, macMap, srcmac2countMap, createResult.groupName())));
+            var bpfInfo = new XDPIface.BPFInfo(bpfobj, xskMap, mac2portMap, port2devMap, srcmac2countMap, passmacMap, createResult.groupName());
+
+            sw.addXDP(nic, vrf, umem, new XDPIface.XDPParamsBuilder()
+                .setQueueId(queueId)
+                .setBPFInfo(bpfInfo)
+                .setRxRingSize(rxRingSize)
+                .setTxRingSize(txRingSize)
+                .setMode(mode)
+                .setZeroCopy(zeroCopy)
+                .setBusyPollBudget(busyPollBudget)
+                .setRxGenChecksum(rxGenChecksum)
+                .setPktswOffloaded(pktswOffloaded)
+                .setCsumOffloaded(csumOffloaded)
+                .build()
+            );
         } catch (Exception e) {
             Logger.error(LogType.SYS_ERROR, "failed to create xdp interface " + nic, e);
             createResult.release(true);

app/src/main/java/io/vproxy/app/controller/DockerNetworkDriverImpl.java

@@ -386,31 +386,41 @@ private XDPIface createXDPIface(Switch sw, UMem umem, VirtualNetwork net, String
             bpfobj = BPFObject.loadAndAttachToNic(nicname, BPFMode.SKB, true);
         }
 
-        BPFMap mac2portMap;
+        BPFMap mac2portMap = null;
+        BPFMap port2devMap = null;
+        BPFMap srcmac2countMap = null;
         try {
-            mac2portMap = bpfobj.getMap(Prebuilt.DEFAULT_MAC_TO_PORT_MAP_NAME);
-        } catch (Exception e) {
-            Logger.error(LogType.SYS_ERROR, "failed to retrieve mac2port map", e);
             if (isPodNic) {
-                reusedResult.release(true);
-            } else {
-                bpfobj.release(true);
+                mac2portMap = bpfobj.getMap(Prebuilt.DEFAULT_MAC_TO_PORT_MAP_NAME);
+                port2devMap = bpfobj.getMap(Prebuilt.DEFAULT_PORT_TO_DEV_MAP_NAME);
+                srcmac2countMap = bpfobj.getMap(Prebuilt.DEFAULT_SRC_MAC_TO_COUNT_MAP_NAME);
             }
+        } catch (Exception e) {
+            Logger.error(LogType.SYS_ERROR, "failed to retrieve mac2port map", e);
+            reusedResult.release(true);
             throw e;
         }
 
+        var bpfInfo = new XDPIface.BPFInfo(
+            bpfobj,
+            bpfobj.getMap(Prebuilt.DEFAULT_XSKS_MAP_NAME),
+            mac2portMap,
+            port2devMap,
+            srcmac2countMap,
+            null,
+            isPodNic ? SHARED_BPF_MAP_GROUP_NAME : null
+        );
+
         XDPIface iface;
         try {
             iface = sw.addXDP(nicname, net != null ? net.vrf : NETWORK_ENTRY_VRF, umem,
-                new XDPIface.XDPParams(0, 32, 32, BPFMode.SKB,
-                    false, 0, false, isPodNic,
-                    new XDPIface.BPFInfo(
-                        bpfobj,
-                        bpfobj.getMap(Prebuilt.DEFAULT_XSKS_MAP_NAME),
-                        isPodNic ? mac2portMap : null,
-                        isPodNic ? bpfobj.getMap(Prebuilt.DEFAULT_SRC_MAC_TO_COUNT_MAP_NAME) : null,
-                        isPodNic ? SHARED_BPF_MAP_GROUP_NAME : null
-                    )));
+                new XDPIface.XDPParamsBuilder()
+                    .setQueueId(0)
+                    .setBPFInfo(bpfInfo)
+                    .setRxGenChecksum(!NativeXDP.supportTxMetadata)
+                    .setPktswOffloaded(isPodNic)
+                    .setCsumOffloaded(true)
+                    .build());
         } catch (Exception e) {
             Logger.error(LogType.SYS_ERROR, "failed to create xdp interface: " + nicname, e);
             if (isPodNic) {

app/src/main/java/io/vproxy/app/process/Shutdown.java

@@ -1,6 +1,7 @@
 package io.vproxy.app.process;
 
 import io.vproxy.app.app.*;
+import io.vproxy.app.app.cmd.handle.param.OffloadHandle;
 import io.vproxy.app.app.util.SignalHook;
 import io.vproxy.app.controller.HttpController;
 import io.vproxy.app.controller.RESPController;
@@ -758,8 +759,14 @@ public static String currentConfig() {
                         + " mode " + xdp.params.mode().name()
                         + " busy-poll " + xdp.params.busyPollBudget()
                         + " vrf " + xdp.vrf;
-                    if (xdp.params.offload()) {
-                        cmd += " offload";
+                    if (xdp.params.pktswOffloaded() || xdp.params.csumOffloaded()) {
+                        cmd += " offload ";
+                        var ls = new ArrayList<String>();
+                        if (xdp.params.pktswOffloaded())
+                            ls.add(OffloadHandle.PACKET_SWITCHING);
+                        if (xdp.params.csumOffloaded())
+                            ls.add(OffloadHandle.CHECKSUM);
+                        cmd += String.join(",", ls);
                     }
                     if (xdp.params.zeroCopy()) {
                         cmd += " zerocopy";