1. INTRODUCTION l4fwd is a sample application to demonstrate and test TLDK TCP/UDP functionalities. Depending on configuration it can do simple send, recv or both over opened TCP/UDP streams. Also it implements ability to do TCP/UDP packet forwarding between different streams, so it is possible to use the l4fwd application as a TCP/UDP proxy. The l4fwd application logically is divided into two parts, Back End (BE) and Front End (FE). 1.1 Back End (BE) BE is responsible for: - RX over DPDK ports and feed them into TCP/UDP TLDK context(s) (via tle_*_rx_bulk). - retrieve packets ready to be send out from TCP/UDP TLDK context(s) and TX them over destined DPDK port. - Multiple RX/TX queues per port are supported by RSS. Right now the number of TX is same as the number of RX queue. - Each BE lcore can serve multiple DPDK ports, TLDK TCP/UDP contexts. BE configuration record format: port=,addr=,masklen=,mac= port - DPDK port id to be used to send packets to the destination. It is an mandatory option. addr - destination network address. It is an mandatory option. masklen - destination network prefix length. It is an mandatory option. mac - destination Ethernet address. It is an mandatory option. mtu - MTU to be used on that port (= application data size + L2/L3/L4 headers sizes, default=1514). It is an optional option. Below are some example of BE entries port=0,masklen=16,addr=192.168.0.0,mac=01:de:ad:be:ef:01 port=0,addr=2001:4860:b002::,masklen=64,mac=01:de:ad:be:ef:01 These examples are also available in be.cfg file. 1.2 Front End (FE) FE is responsible for: - to open configured TCP/UDP streams and perform send/recv over them. These streams can belong to different TCP/UDP contexts. Each lcore can act as BE and/or FE. In UDP mode the application can reassemble input fragmented IP packets, and fragment outgoing IP packets (if destination MTU is less then packet size). FE configuration record format: lcore=,op=<"rx|tx|echo|fwd|rxtx">,\ laddr=,lport=,raddr=,rport=,\ [txlen=,fwladdr=,fwlport=,fwraddr=,fwrport=,\ belcore=,rxlen=] lcore - EAL lcore to manage that stream(s) in the FE. It is an mandatory option. belcore - EAL lcore to manage that stream(s) in the BE. It is an optional option. lcore and belcore can specify the same cpu core. op - operation to perform on that stream: "rx" - do receive only on that stream. "tx" - do send only on that stream. "echo" - mimic recvfrom(..., &addr);sendto(..., &addr); on that stream. "fwd" - forward packets between streams. "rxtx" - Receive/reply transactions on stream. It is an mandatory option. laddr - local address for the stream to open. It is an mandatory option. lport - local port for the stream to open. It is an mandatory option. raddr - remote address for the stream to open. It is an mandatory option. rport - remote port for the stream to open. It is an mandatory option. txlen - data length sending in each packet. (mandatory for "tx" & "rxtx" modes only). rxlen - Expected response length (mandatory for "rxtx" mode only). fwladdr - local address for the forwarding stream(s) to open (mandatory for "fwd" mode only). fwlport - local port for the forwarding stream(s) to open (mandatory for "fwd" mode only). fwraddr - remote address for the forwarding stream(s) to open (mandatory for "fwd" mode only). fwrport - remote port for the forwarding stream(s) to open (mandatory for "fwd" mode only). Below are some example of FE entries lcore=3,op=echo,laddr=192.168.1.233,lport=0x8000,raddr=0.0.0.0,rport=0 lcore=3,op=tx,laddr=192.168.1.233,lport=0x8001,raddr=192.168.1.56,\ rport=0x200,txlen=72 lcore=3,op=rx,laddr=::,lport=0x200,raddr=::,rport=0,txlen=72 lcore=3,op=fwd,laddr=0.0.0.0,lport=11211,raddr=0.0.0.0,rport=0,\ fwladdr=::,fwlport=0,fwraddr=2001:4860:b002::56,fwrport=11211 These examples are also available in fe.cfg file with some more explanation. 1.3 Configuration files format - each record on a separate line. - lines started with '#' are treated as comments. - empty lines (containing whitespace chars only) are ignored. - kvargs style format for each record. - each FE record correspond to at least one stream to be opened (could be multiple streams in case of op="fwd"). - each BE record define a ipv4/ipv6 destination. 2. REQUIREMENTS DPDK libraries (16.11 or higher) TLDK libraries (1.0) Back-End (BE) configuration file Front-End(FE) configuration file 3. USAGE l4fwd -- \ -P | --promisc /* promiscuous mode enabled. */ \ -R | --rbufs /* max recv buffers per stream. */ \ -S | --sbufs /* max send buffers per stream. */ \ -s | --streams /* streams to open per context. */ \ -b | --becfg /* backend configuration file. */ \ -f | --fecfg /* frontend configuration file. */ \ -c | --txcnt /* file with TX payload (used by rxtx mode). */ \ -U | --udp /* run the app to handle UDP streams only. */ \ -T | --tcp /* run the app to handle TCP streams only. */ \ -L | --listen /* open TCP streams in server mode (listen). */ \ -a | --enable-arp /* enable arp responses (request not supported) */ \ -v | --verbose /* different level of verbose mode */ \ -H | --hash /* hash algorithm i.e. siphash or jhash to be */ \ /* used to generate the sequence number. */ \ -K | --seckey /* 16 character long secret key used by */ \ /* hash algorithms to generate the */ \ /* sequence number. */ \ -M | --mbuf-num /* other than default number of mbufs per pool. */ \ ... Note that: options -U and -T cannot be used together. Option -L can be used only with option -T. portX_params: port=,lcore=[-],[lcore=[-],]\ [rx_offload=,tx_offload=,mtu=,ipv4=,ipv6=] portX_params are used to configure the particular DPDK device (rte_ethdev port), and specify BE lcore that will handle RX/TX from/to the device and manage BE part of corresponding TCP/UDP context. Multiple BE lcore can be specified. port - DPDK port id (RSS are supported when multiple lcores are specified for a port). It is an mandatory option. lcore - EAL lcore id to handle IO over that port (rx_burst/tx_burst). several ports can be managed by the same lcore, and same port can be managed by more than one lcore. It is an mandatory option. At least one lcore option has to be specified. lcore range can be specified in one lcore option. e.g. lcore=2-3,lcore=6 will enable lcores 2, 3, and 6 to handle BE. rx_offload - RX HW offload capabilities to enable/use on this port. (bitmask of DEV_RX_OFFLOAD_* values). It is an optional option. tx_offload - TX HW offload capabilities to enable/use on this port. (bitmask of DEV_TX_OFFLOAD_* values). mtu - MTU to be used on that port (= application data size + L2/L3/L4 headers sizes, default=1514). ipv4 - ipv4 address to assign to that port. ipv6 - ipv6 address to assign to that port. At least one of ipv4/ipv6 values have to be specified for each port. 3.1 RSS If multiple lcore is specified per DPDK port, the following RSS hash will be enabled on that port: ETH_RSS_UDP, or ETH_RSS_TCP The RSS queue qid will handle the stream according to the TCP/UDP source ports of the stream. The qid can be calculated as below qid = (src_port % power_of_2(n)) % n where n is number of lcore used to mane the DPDK port. 4. EXAMPLES 4.1 Sample testbed +----------------------------+ +-------------------------------+ | TLDK Box | | Linux Box | | | | | | port 0 +----------------+ port 0 | | 192.168.1.1 | | 192.168.1.2 | | 2001:4860:b002::1 | | 2001:4860:b002::2 | | AA:BB:CC:DD:EE:F1 | | AA:BB:CC:DD:EE:F2 | +----------------------------+ +-------------------------------+ 4.2 UDP, "rx" mode, IPv4-only, Single core This example shows receiving data from a IPv4 stream. The TLDK UDP server runs on single core where both BE and FE run on cpu core 3. be.cfg file contains: port=0,masklen=24,addr=192.168.1.0,mac=AA:BB:CC:DD:EE:F2 fe.cfg file contains (udp server listening to port 6000): lcore=3,op=rx,laddr=192.168.1.1,lport=6000,raddr=0.0.0.0,rport=0 run the l4fwd application as below (DPDK port 0 (pci 01:00.0)): l4fwd --lcores='3' -w 01:00.0 -- \ --promisc --rbufs 0x100 --sbufs 0x100 --streams 0x100 --fecfg fe.cfg \ --becfg be.cfg -U port=0,lcore=3,ipv4=192.168.1.1 This will create TLDK UDP context on lcore=3 (BE lcore) to manage DPDK port 0. The port 0 will have IPv4 address 192.168.1.1. All the streams will be in server mode and also managed by lcore 3. 4.3 UDP, "echo" mode, IPv6-only, Multicore This example shows receiving data from a IPv6 stream and sending the data back through the same IPv6 stream. The TLDK UDP server runs on multicore where BE runs on cpu core 2 and FE runs on cpu core 3. be.cfg file contains: port=0,masklen=64,addr=2001:4860:b002::,mac=AA:BB:CC:DD:EE:F2 fe.cfg file contains (udp server listening to port 6000): lcore=3,op=rx,laddr=2001:4860:b002::1,lport=6000,raddr=::,rport=0 run the l4fwd application as below (DPDK port 0 (pci 01:00.0)): l4fwd --lcores='2,3' -w 01:00.0 -- \ --promisc --rbufs 0x100 --sbufs 0x100 --streams 0x100 --fecfg fe.cfg \ --becfg be.cfg -U port=0,lcore=2,ipv6=2001:4860:b002::1 This will create TLDK UDP context on lcore=2 (BE lcore) to manage DPDK port 0. The port 0 will have IPv4 address 2001:4860:b002::1. All the streams will be in server mode and managed by lcore 3 (FE lcore). In this case, the UDP server will send the incoming data back to the sender. 4.4 TCP, "echo" mode, IPv4-only, Multicore, RX-Offload This example shows receiving data from a IPv4 stream and sending the data back through the same IPv4 stream. The TLDK TCP server runs on multicore where BE runs on cpu core 2 and FE runs on cpu core 3. The BE also uses receive offload features of the NIC. be.cfg file contains: port=0,masklen=24,addr=192.168.1.0,mac=AA:BB:CC:DD:EE:F2 fe.cfg file contains (tcp server listening to port 6000): lcore=3,op=echo,laddr=192.168.1.1,lport=6000,raddr=0.0.0.0,rport=0 run the l4fwd application as below (DPDK port 0 (pci 01:00.0)): l4fwd --lcores='2,3' -w 01:00.0 -- \ --promisc --rbufs 0x100 --sbufs 0x100 --streams 0x100 --fecfg fe.cfg \ --becfg be.cfg -T -L port=0,lcore=2,rx_offload=0xf,tx_offload=0,\ ipv4=192.168.1.1 This will create TLDK TCP context on lcore=2 (BE lcore) to manage DPDK port 0. The port 0 will have IPv4 address 192.168.1.1. The following DPDK RX HW offloads will be enabled on that port. DEV_RX_OFFLOAD_VLAN_STRIP, DEV_RX_OFFLOAD_IPV4_CKSUM, DEV_RX_OFFLOAD_UDP_CKSUM, DEV_RX_OFFLOAD_TCP_CKSUM No HW TX offloads will be enabled. All the streams will be in server mode and managed by lcore 3 (FE core). In this case, the TCP server will send the incoming data back to the sender. 4.5 TCP, "fwd" (proxy) mode, IPv4-to-IPv6, Multi-core, RX-Offload This example shows receiving data from a IPv4 stream and forwarding the data to a IPv6 stream. The TLDK TCP server runs on multicore where BE runs on cpu core 2 and FE runs on cpu core 3. The BE also uses receive offload features of the NIC. be.cfg file contains: port=0,masklen=24,addr=192.168.1.0,mac=AA:BB:CC:DD:EE:F2 fe.cfg file contains (tcp server listening to port 6000): lcore=3,op=fwd,laddr=192.168.1.1,lport=6000,raddr=0.0.0.0,rport=0,\ rladdr=::,lport=0,raddr=2001:4860:b002::2,rport=7000 run the l4fwd application as below (DPDK port 0 (pci 01:00.0)): l4fwd --lcores='2,3' -w 01:00.0 -- \ --promisc --rbufs 0x100 --sbufs 0x100 --streams 0x100 --fecfg fe.cfg \ --becfg be.cfg -T -L port=0,lcore=2,rx_offload=0xf,tx_offload=0,\ ipv4=192.168.1.1,ipv6=2001:4860:b002::1 This will create TLDK TCP context on lcore=2 (BE lcore) to manage DPDK port 0. The port 0 will have IPv4 address 192.168.1.1. The following DPDK RX HW offloads will be enabled on that port. DEV_RX_OFFLOAD_VLAN_STRIP, DEV_RX_OFFLOAD_IPV4_CKSUM, DEV_RX_OFFLOAD_UDP_CKSUM, DEV_RX_OFFLOAD_TCP_CKSUM No HW TX offloads will be enabled. All the streams will be in server mode and managed by lcore 3 (FE core). In this case, the IPv4 TCP server will forward the incoming data to the IPv6 TCP server 2001:4860:b002::2 listening to port 7000. 4.6 TCP, "echo" mode, RSS, IPv4-only, Multicore, RX-Offload This example shows receiving data from a IPv4 stream and sending the data back through the same IPv4 stream. The TLDK TCP server runs on multicore where BE runs on cpu cores 1-2 and FE runs on cpu core 3. As BE runs on multicore, Receive Side Scaling (RSS) feature will be automatically enabled. The BE also uses receive offload features of the NIC. be.cfg file contains: port=0,masklen=24,addr=192.168.1.0,mac=AA:BB:CC:DD:EE:F2 fe.cfg file contains (tcp server listening to port 6000): lcore=3,op=echo,laddr=192.168.1.1,lport=6000,raddr=0.0.0.0,rport=0 lcore=3,op=echo,laddr=192.168.1.1,lport=6001,raddr=0.0.0.0,rport=0 run the l4fwd application as below (DPDK port 0 (pci 01:00.0)): l4fwd --lcores='1,2,3' -w 01:00.0 -- \ --promisc --rbufs 0x100 --sbufs 0x100 --streams 0x100 --fecfg fe.cfg \ --becfg be.cfg -T -L port=0,lcore="1-2",rx_offload=0xf,tx_offload=0,\ ipv4=192.168.1.1 This will create TLDK TCP context on lcore=1-2 (BE lcore) to manage DPDK port 0. The port 0 will have IPv4 address 192.168.1.1. The following DPDK RX HW offloads will be enabled on that port. DEV_RX_OFFLOAD_VLAN_STRIP, DEV_RX_OFFLOAD_IPV4_CKSUM, DEV_RX_OFFLOAD_UDP_CKSUM, DEV_RX_OFFLOAD_TCP_CKSUM No HW TX offloads will be enabled. All the streams will be in server mode and managed by lcore 3 (FE core). In this case, the TCP server will send the incoming data back to the sender. As RSS is enabled, all the packets with destination port 6000 and 6001 will be managed by HW queue 0 and queue 1 respectively. Please note that RSS is not supported on the interface when both IPv4 and IPv6 are enabled. Only one of IPv4 or IPv6 has to be enabled in the port.