Handbook:PPC64/Networking/Advanced

From Gentoo Wiki
Jump to:navigation Jump to:search
This page is a translated version of the page Handbook:PPC64/Networking/Advanced and the translation is 100% complete.
PPC64 Handbook
Instalacja
O instalacji
Wybór medium instalacyjnego
Konfiguracja sieci
Przygotowanie dysków
Instalacja etapu 3
Instalacja systemu podstawowego
Konfiguracja jądra
Konfiguracja systemu
Instalacja narzędzi
Instalacja systemu rozruchowego
Finalizacja
Praca z Gentoo
Wstęp do Portage
Flagi USE
Funkcje portage
System initscript
Zmienne środowiskowe
Praca z Portage
Pliki i katalogi
Zmienne
Mieszanie działów oprogramowania
Dodatkowe narzędzia
Custom package repository
Funkcje zaawansowane
Konfiguracja sieci
Zaczynamy
Zaawansowana konfiguracja
Sieć modularna
Sieć bezprzewodowa
Dodawanie funkcjonalności
Dynamiczne zarządzanie

The config_eth0 variable is the heart of an interface configuration. It is a high level instruction list for configuring the interface (eth0 in this case). Each command in the instruction list is performed sequentially. The interface is deemed OK if at least one command works.

Here's a list of built-in instructions:

Value Description
null Do nothing.
noop If the interface is up and there is an address then abort configuration successfully.
An IPv4 or IPv6 address Add the address to the interface.
dhcp, adsl, or apipa (or a custom value from a 3rd party module) Run the module which provides the command. For example dhcp will run a module that provides DHCP which can be served by dhcpcd, dhclient, or pump.

If a command fails, specify a fallback value. The fallback has to match the config structure exactly.

It is possible to chain these values together. Here are some real world examples:

FILE /etc/conf.d/netConfiguration examples
# Adding three IPv4 addresses
config_eth0="192.168.0.2/24
192.168.0.3/24
192.168.0.4/24"
  
# Adding an IPv4 address and two IPv6 addresses
config_eth0="192.168.0.2/24
4321:0:1:2:3:4:567:89ab
4321:0:1:2:3:4:567:89ac"
  
# Keep our kernel assigned address, unless the interface goes
# down so assign another via DHCP. If DHCP fails then add a
# static address determined by APIPA
config_eth0="noop
dhcp"
fallback_eth0="null
apipa"
Informacja
When using the ifconfig module and adding more than one address, interface aliases are created for each extra address. So with the above two examples users will get interfaces eth0, eth0:1 and eth0:2. It is not possible to do anything special with these interfaces as the kernel and other programs will just treat eth0:1 and eth0:2 as eth0.
Ważne
The fallback order is important! If the null option was not specified then apipa would only be run if noop failed.
Informacja
APIPA (handbook) and DHCP (handbook) are discussed in the modular networking section.

Network dependencies

Init scripts in /etc/init.d/ can depend on a specific network interface or just "net". All network interfaces in Gentoo's init system provide what is called "net".

If, in /etc/rc.conf, the rc_depend_strict variable is set to YES, then all network interfaces that provide "net" must be active before a dependency on "net" is assumed to be met. In other words, if a system has a net.eth0 and net.eth1 and an init script depends on "net", then both must be enabled.

On the other hand, if rc_depend_strict="NO" is set, then the "net" dependency is marked as resolved the moment at least one network interface is brought up.

But what about net.br0 depending on net.eth0 and net.eth1? net.eth1 may be a wireless or PPP device that needs configuration before it can be added to the bridge. This cannot be done in /etc/init.d/net.br0 as that's a symbolic link to net.lo.

The answer is to define a rc_net_{interface}_need setting in /etc/conf.d/net:

FILE /etc/conf.d/netAdding a net.br0 dependency
rc_net_br0_need="net.eth0 net.eth1"

That alone, however, is not sufficient. Gentoo's networking init scripts use a virtual dependency called "net" to inform the system when networking is available. Clearly, in the above case, networking should only be marked as available when net.br0 is up, not when the others are. So we need to tell that in /etc/conf.d/net as well:

FILE /etc/conf.d/netUpdating virtual dependencies and provisions for networking
rc_net_eth0_provide="!net"
rc_net_eth1_provide="!net"

For a more detailed discussion about dependency, consult the section on writing initscripts in the Gentoo Handbook. More information about /etc/rc.conf is available as comments within that file.

Variable names and values

Variable names are dynamic. They normally follow the structure of variable_${interface|mac|essid|apmac}. For example, the variable dhcpcd_eth0 holds the value for dhcpcd options for eth0 and dhcpcd_essid holds the value for dhcpcd options when any interface connects to the ESSID "essid".

However, there is no hard and fast rule that states interface names must be ethx. In fact, many wireless interfaces have names like wlanx, rax as well as ethx. Also, some user defined interfaces such as bridges can be given any name. To make life more interesting, wireless Access Points can have names with non alpha-numeric characters in them - this is important because users can configure networking parameters per ESSID.

The downside of all this is that Gentoo uses bash variables for networking - and bash cannot use anything outside of English alpha-numerics. To get around this limitation we change every character that is not an English alpha-numeric into an _ (underscore) character.

Another downside of bash is the content of variables - some characters need to be escaped. This can be achieved by placing the \ (backslash) character in front of the character that needs to be escaped. The following list of characters needs to be escaped in this way: ", ' and \.

In this example we use wireless ESSID as they can contain the widest scope of characters. We shall use the ESSID My "\ NET:

FILE /etc/conf.d/netVariable names
# This does work, but the domain is invalid
dns_domain_My____NET="My \"\\ NET"

The above sets the DNS domain to My "\ NET when a wireless card connects to an AP whose ESSID is My "\ NET.

Network interface naming

How it works

Network interface names are not chosen arbitrarily: the Linux kernel and the device manager (most systems have udev as their device manager although others are available as well) choose the interface name through a fixed set of rules.

When an interface card is detected on a system, the Linux kernel gathers the necessary data about this card. This includes:

  • The onboard (on the interface itself) registered name of the network card, which is later seen through the ID_NET_NAME_ONBOARD value.
  • The slot in which the network card is plugged in, which is later seen through the ID_NET_NAME_SLOT value.
  • The path through which the network card device can be accessed, which is later seen through the ID_NET_NAME_PATH value.
  • The (vendor-provided) MAC address of the card, which is later seen through the ID_NET_NAME_MAC value.

Based on this information, the device manager decides how to name the interface on the system. By default, it uses the first hit of the first three variables above (ID_NET_NAME_ONBOARD, _SLOT or _PATH). For instance, if ID_NET_NAME_ONBOARD is found and set to eno1, then the interface will be called eno1.

Given an active interface name, the values of the provided variables can be shown using udevadm:

root #udevadm test-builtin net_id /sys/class/net/enp3s0 2>/dev/null
ID_NET_NAME_MAC=enxc80aa9429d76
ID_OUI_FROM_DATABASE=Quanta Computer Inc.
ID_NET_NAME_PATH=enp3s0

As the first (and actually only) hit of the top three variables is ID_NET_NAME_PATH, its value is used as the interface name. If none of the variables contain values, then the system reverts back to the kernel-provided naming (eth0, eth1, etc.)

Using the old-style kernel naming

Before this change, network interface cards were named by the Linux kernel itself, depending on the order that drivers are loaded (amongst other, possibly more obscure reasons). This behavior can still be enabled by setting the net.ifnames=0 boot parameter in the boot loader.

Using custom names

The entire idea behind the change in naming is not to confuse people, but to make changing the names easier. Suppose a system has two interfaces that are otherwise called eth0 and eth1. One is meant to access the network through a wire, the other one is for wireless access. With the support for interface naming, users can have these called lan0 (wired) and wifi0 (wireless - it is best to avoid using the previously well-known names like eth* and wlan* as those can still collide with the suggested names).

Find out what the parameters are for the cards and then use this information to set up a custom own naming rule:

root #udevadm test-builtin net_id /sys/class/net/eth0 2>/dev/null
ID_NET_NAME_MAC=enxc80aa9429d76
ID_OUI_FROM_DATABASE=Quanta Computer Inc.
root #vim /etc/udev/rules.d/70-net-name-use-custom.rules
# First one uses MAC information, and 70- number to be before other net rules
SUBSYSTEM=="net", ACTION=="add", ATTR{address}=="c8:0a:a9:42:9d:76", NAME="lan0"
root #vim /etc/udev/rules.d/76-net-name-use-custom.rules
# Second one uses ID_NET_NAME_PATH information, and 76- number to be between
# 75-net-*.rules and 80-net-*.rules
SUBSYSTEM=="net", ACTION=="add", ENV{ID_NET_NAME_PATH}=="enp3s0", NAME="wifi0"

Because the rules are triggered before the default one (rules are triggered in alphanumerical order, so 70 comes before 80) the names provided in the rule file will be used instead of the default ones. The number granted to the file should be between 76 and 79 (the environment variables are defined by a rule start starts with 75 and the fallback naming is done in a rule numbered 80).