Networking/Bugs

Known networking bugs

Only platform bugs that can influence various application and/or the overall behavior of the system belong here.

Kernel

• Ethernet: noarp causes dst MAC to be the same as src MAC instead of broadcast
• Bridge/bond: Empty bond interface cannot be bridged unless you work around by manually setting the address
• Kernel removes ::1 but not 127.0.0.1 when setting lo down
• IPv4: Kernel can't store DHCP lifetime and cannot mark IPv4 addresses dynamic (in the DHCP sense) (due in 3.9)
• IPv6: IPv6 networking breaks when loopback is set down and up
• IPv6: Kernel ignores RA router preference
• IPv6: Tempaddr cycling breaks TCP connections prematurely
• IPv6: Kernel doesn't cache RDNSS and DNSSL
• IPv6: Kernel doesn't provide managable IPv6 autoconfiguration
• ?Netlink: Kernel doesn't solve race condition between rtnl commands and events
• Netlink: Kernel doesn't send some link change notifications (e.g. mtu) for devices that are down
• master/slave: Kernel enslaves a L3-configured interface even if it can't work that way (the same in the opposite order, reported by djasa)
• bridge: re-enslaving a slave to another bridge may cause two bridges to have duplicate MACs
• bridge: an empty bridge gets IFF_LOWER_UP when set up

libnl

• rtnl_link_bond_add() succeeds when device already exists
• rtnl_link_bridge_add() and rtnl_link_vlan_add missing

iproute

• ip link show won't distinguish bridges and bonds
• ip link show won't show vlan id
• ip -6 route shows cached routes (ip -4 route doesn't)

glibc

• IPv4: getaddrinfo(127.0.0.1, AI_ADDRCONFIG) fails on hosts without global IPv4 (e.g. notebooks not connected to any networks)
• IPv6: getaddrinfo(::1, AI_ADDRCONFIG) fails on hosts without any other IPv6 address
• IPv6: getaddrinfo(link-local-address, AI_ADDRCONFIG) fails on hosts without global IPv6 (hopefully solved)
• IPv6: GLIBC's nsswitch doesn't support overriding getaddrinfo which is requred to resolve link-local IPv6 addresses
• DNS: getaddrinfo doesn't support SRV records

More information about AI_ADDRCONFIG problems

Upstream bug reports for getaddrinfo

dhclient

• dhclient fails with large values of dhcp-renawal-time, notably the infinity (0xffffffff) value
• People say that dhclient exchange is slow, especially on wireless (I can't reproduce)

Avahi

• Avahi's nss-mdns returns link-local IPv6 addresses without scope_id and applications fail to connect
• Avahi's IPv6 support is disabled by default (this has been resolved in the past)

See Features/ZeroconfNetworking.

NetworkManager

• Various bugs in NetworkManager

Testing tools

• ping doesn't support IPv6 (you have to use ping6)
• netstat and ss won't mark non-v6only sockets

Various applications

• Assume that getaddrinfo() returns only one address and/or that such the first returned address works. See Networking/NameResolution.

Networking related standards

• RFC 6106 (RA DNS): Relies on a false assuption that every (even multicast) IP packet is delivered to its destination.
• RFC 4861 (NDP): Doesn't specify how zero lifetimes should be used and lets implementations rely on the same false assumption.
• RFC 3493 (socket interfce): Breaks both IPv4 and IPv6 localhost and link-local networking depending on available global addresses.
• POSIX.1-2008: Defines the AI_ADDRCONFIG in a bad way

Contacts

See Networking#Contacts.