Suppression of IP Stack ICMP UDP Destination Port Not Found

[bob2oneil]

I am developing an application which utilizes raw packet sockets for a range of pseudo-multicast addresses, but also need to support standard UDP ports using different UPD destination ports. To support raw packet sockets for the multicast addresses, where I need to retain L2/L3 headers, I am using a raw packet socket.

This raw socket provides me the host packet with everything intact, so I can use that read opportunity to dispatch the UDP packet also on to specific handlers for the individual UDP destination ports, simply by removing the UDP header itself.

Currently when host data is driven into the raw socket interface will not find a UDP socket bound for 2 of the 3 UDP destination ports, and therefore will send an ICMP message to the data originator. The call stack for this is shown below.

I thought I could solve this by creating 2 additional UDP sockets, each bound to the different UDP destination ports, or perhaps one UPD socket bound to any port. But this approach might introduce some inefficiencies, since I already am receiving the full packet.

Is there a more efficient way of suppressing the Zephyr to host ICMP for the unresolved driven UDP destination ports than what I have described?

I did add a couple of UDP sockets bounds to the sent destination port. However, this resulted in leaks, perhaps requiring each of these sockets having a reader rather than just being passive listeners? Since I already have the packet as L2, it is not technically necessary that I read them at the UDP socket level, but that is a viable option.

Since I want UDP support for a single port (1 of 3), I do not want to simply remove UDP support for the kernel build.

The callstack for the ICMP error response is as follows:

net_icmpv4_send_error(struct net_pkt * orig, uint8_t type, uint8_t code) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/icmpv4.c:581)
conn_send_icmp_error(struct net_pkt * pkt) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/connection.c:830)
net_conn_input(struct net_pkt * pkt, union net_ip_header * ip_hdr, uint8_t proto, union net_proto_header * proto_hdr) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/connection.c:830)
net_ipv4_input(struct net_pkt * pkt) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/ipv4.c:375)
process_data(_Bool is_loopback, struct net_pkt * pkt) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/net_core.c:126)
processing_data(struct net_pkt * pkt, _Bool is_loopback) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/net_core.c:141)
net_rx(struct net_if * iface, struct net_pkt * pkt) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/net_core.c:361)
net_process_rx_packet(struct net_pkt * pkt) (/home/robocom/zephyrproject/zephyr/include/net/net_pkt.h:280)
tc_rx_handler(struct k_fifo * fifo) (/home/robocom/zephyrproject/zephyr/subsys/net/ip/net_tc.c:250)
z_thread_entry(k_thread_entry_t entry, void * p1, void * p2, void * p3) (/home/robocom/zephyrproject/zephyr/lib/os/thread_entry.c:29)
posix_thread_starter(void * arg) (/home/robocom/zephyrproject/zephyr/arch/posix/core/posix_core.c:305)
libpthread.so.0!start_thread(void * arg) (/build/glibc-eX1tMB/glibc-2.31/nptl/pthread_create.c:477)
libc.so.6!clone() (/build/glibc-eX1tMB/glibc-2.31/sysdeps/unix/sysv/linux/x86_64/clone.S:95)

Implementation is raw packet sockets using zsock_select() for multiplexed i/o.

Kernel configuration is as follows:

# Project Configuration
CONFIG_CPLUSPLUS=y
CONFIG_LIB_CPLUSPLUS=y

# This setting in necessary for QEMU-RISCV, but will
# default to 'n' or external for NATIVE-POSIX
CONFIG_NEWLIB_LIBC=y

# General kernel options
CONFIG_INIT_STACKS=y

# Size of heap memory pool
CONFIG_HEAP_MEM_POOL_SIZE=16384

# Networking config
# Need TCP for Telnet session for shell interface
CONFIG_NETWORKING=y
CONFIG_NET_UDP=y
CONFIG_NET_TCP=y
CONFIG_NET_IPV6=n
CONFIG_NET_IPV4=y
CONFIG_NET_DHCPV4=n
CONFIG_NET_ARP=y
CONFIG_NET_NATIVE=y
CONFIG_NET_L2_ETHERNET=y

# Network drivers buffers
CONFIG_NET_PKT_RX_COUNT=800
CONFIG_NET_PKT_TX_COUNT=800
CONFIG_NET_BUF_RX_COUNT=800
CONFIG_NET_BUF_TX_COUNT=800

# IP Address ranges
CONFIG_NET_IF_UNICAST_IPV4_ADDR_COUNT=258
CONFIG_NET_IF_MCAST_IPV4_ADDR_COUNT=0
CONFIG_NET_MAX_CONTEXTS=10

# To support prj.conf based IP setup
# We will be assigning IPs dynamically runtime
# so will not use the configuration based approach
CONFIG_NET_CONFIG_SETTINGS=y
CONFIG_NET_CONFIG_NEED_IPV6=n
CONFIG_NET_CONFIG_NEED_IPV4=n

# IP address options, leave blank so can be assigned runtime
# Primary interface IP and peer IP
CONFIG_NET_CONFIG_MY_IPV4_ADDR=""
CONFIG_NET_CONFIG_PEER_IPV4_ADDR=""

# Statistics
CONFIG_NET_STATISTICS=y
CONFIG_NET_STATISTICS_USER_API=n
CONFIG_NET_STATISTICS_PERIODIC_OUTPUT=y
CONFIG_NET_STATISTICS_IPV4=y
CONFIG_NET_STATISTICS_IPV6=n
CONFIG_NET_STATISTICS_IPV6_ND=n
CONFIG_NET_STATISTICS_ICMP=y
CONFIG_NET_STATISTICS_UDP=y
CONFIG_NET_STATISTICS_TCP=y
CONFIG_NET_STATISTICS_MLD=n
CONFIG_NET_STATISTICS_ETHERNET=y
CONFIG_NET_STATISTICS_ETHERNET_VENDOR=n
CONFIG_NET_STATISTICS_LOG_LEVEL_DBG=y
CONFIG_NET_STATISTICS_PER_INTERFACE=y

# Enable promiscuous mode support on sockets for L2 access
# Not necessary for raw socket interface for IPs assigned
# to the primary wired interface, but needed for the
# wireless simulated RF network
CONFIG_NET_PROMISCUOUS_MODE=y

# Ethernet interface native POSIX
CONFIG_ETH_NATIVE_POSIX=y

# Random MAC is required for more than 1 POSIX interfaces
CONFIG_ETH_NATIVE_POSIX_RANDOM_MAC=y

# POSIX two interfaces, wired and wireless
CONFIG_ETH_NATIVE_POSIX_INTERFACE_COUNT=2

# Enable BSD sockets but don't use POSIX names for clashes with build machine
CONFIG_NET_SOCKETS=y
CONFIG_NET_SOCKETS_POSIX_NAMES=n

# Maximum file descriptors
CONFIG_POSIX_MAX_FDS=6

# Packet raw socket configuration to allow for raw sockets bound to iface
CONFIG_NET_SOCKETS_PACKET=y

# Random number generator
CONFIG_ENTROPY_GENERATOR=y
CONFIG_TEST_RANDOM_GENERATOR=y

# Logging
CONFIG_NET_LOG=y
CONFIG_LOG=y
CONFIG_LOG_RUNTIME_FILTERING=y
CONFIG_PRINTK=y

# - 0 OFF, do not write by default
# - 1 ERROR, default to only write LOG_LEVEL_ERR
# - 2 WARNING, default to write LOG_LEVEL_WRN
# - 3 INFO, default to write LOG_LEVEL_INFO
# - 4 DEBUG, default to write LOG_LEVEL_DBG
CONFIG_LOG_DEFAULT_LEVEL=4

# Prefix log output with function name
CONFIG_LOG_FUNC_NAME_PREFIX_ERR=y
CONFIG_LOG_FUNC_NAME_PREFIX_WRN=y
CONFIG_LOG_FUNC_NAME_PREFIX_INF=y
CONFIG_LOG_FUNC_NAME_PREFIX_DBG=y

# Additions to the Shell/Console Interface
CONFIG_SHELL_STACK_SIZE=4096
CONFIG_SHELL_CMDS=y
CONFIG_SHELL_TAB_AUTOCOMPLETION=y
CONFIG_SHELL_HISTORY=y
CONFIG_NET_SHELL=y
CONFIG_KERNEL_SHELL=y
CONFIG_NET_STATISTICS=y
CONFIG_DEVICE_SHELL=y

# Shell and logging device backends
CONFIG_SHELL_LOG_BACKEND=n
CONFIG_SHELL_BACKEND_TELNET=y
CONFIG_LOG_BACKEND_UART=y

# Debug Options for debugging
CONFIG_DEBUG=y
CONFIG_DEBUG_OPTIMIZATIONS=y

# Memory buffer leak verification shell extensions
CONFIG_NET_BUF_POOL_USAGE=y
CONFIG_NET_DEBUG_NET_PKT_ALLOC=y

# File System Shell
CONFIG_FILE_SYSTEM=y
CONFIG_FILE_SYSTEM_SHELL=y

# Gather network statistics
CONFIG_NET_STATISTICS_ETHERNET=y

# Name threads
CONFIG_THREAD_NAME=y

# Max number of IPv4 network interfaces in the system
# Will be a value of 1 for real radio
CONFIG_NET_IF_MAX_IPV4_COUNT=2

# How many Tx traffic classes to have for each network device
CONFIG_NET_TC_TX_COUNT=1

# How many network connections is supported (default is 4)
CONFIG_NET_MAX_CONN=6

The iperf command driving data into the wired interface is as follows:

iperf -c 192.0.1.1 -u -i1 -b5M -l188 -p49169 -t1

[bob2oneil]

I was able to solve this problem by not just having a BSD style UDP socket bound to specific ports, but I also needed to have the data read using this socket rather than just obtaining the same data, with L2 header intact at the raw socket level.