Bonding von Ethernetkarten unter Gentoo

khenske · n00b Joined: 19 Dec 2003 Posts: 23

Hallo Leute,

ich werde in loser Reiherfolge einiges von meinen Erfahrungen mit Linux hier iniederschreiben. Und der erste Beitrag handelt darum, wie ich unter Gentoo zwei psysikalische Nettzwerkkarten zu einer logischen zusammenfasse, um damit den Durchsatz zu erhöhen. Dieses nennt man Bonding Nicht vergessen, das die Gegenstelle das auch unterstützen sollte. Viele managed Switche unterstützen das.

Die Beispielkonfiguration, an der ich das erkläre ist ein PC mit Tycon mit zwei Onboard-Gigabit-Schnittstellen und einer dritten Netzwerkkarte für das DSL-Modem. Als Bound laufen die eingebauten Netzwerkkarten , die durch das Kernelmodul tg3 atkiviert sind und die Name eth0 und eth1 haben. Die dritte Karte ist eth2.

Damit das Bonding auch klappt, muss das in der Kernelkonfiigurationdatei .config folgendes unter Network testing folgendes stehen "CONFIG_BONDING=m stehen,. Ich empfehle, das bonding als Modul zu verwenden, da es vor den Netzwerkkarten gelanden wender muss, wobei die später im Bondverbund laufenen dann danach geladen werden sollten.

Wenn ihr make menuconfig verwendet, dann sollte folgendes aktiviert, bzw als Modul erstellt werden:

Code:

Device Drivers ---> Networking support ---> Networking options ---> QoS and/or fair queueing --->
[*] QoS and/or fair queueing
Device Drivers ---> Networking support --->
<M> Bonding driver support
<M> EQL (serial line load balancing) support

Wichtig ist auch dabei, das Proggram ifenslave unter /usr/src/linux/Documentation/networking zu erstellen (script mkifenslave.sh):

Code:

#!/bin/bash
cd /usr/src/linux/Documentation/networking
gcc -Wall -Wstrict-prototypes -O -I/usr/src/linux/include ifenslave.c -o ifenslave
mv ifenslave /sbin/ifenslave

Wichtig ist dabei, das bei einem Kernel-Update dieses Programm auch neu erstellt werden muss, denn es llegt jedem neuen Kernel bei..

Dann fügt man die Datei bond0 in das Verzeichnis /etc/modules.d bei:

Code:

### bonding module alias
alias bond0 bonding

Danach einmal modules-update aufrufen.

Dann werden noch folgende Zeilen in die entsprechende kernel-2.x Datei eingefügt:

Code:

bonding
tg3
8139too

Wichtig ist dabei, das die Netzwerkkartentreiber, die miteinander verbunden werden, nach bonding geladen werden. In diesen Beispiel handelt es sich um das tg3-Modul.

Dann werden noch ein paar motifizierte gentoo-Dateien verwendet. In der Datei /etc/conf.d/net habe ich folgende Ergänzung gemacht:

Code:

iface_bond0="192.168.30.1 broadcast 192.168.30.255 netmask 255.255.255.0"
# added for the slaves iface by Klaus Henske
slaves_bond0="eth0 eth1"
iface_eth2="up"

Wir ihr seht, habe ich eine neue Variable eingefügt: slaves_bond0, in der ich schreibe, welche beiden psysikalischen Ethernetdevices im Slavemode dem Ethernetdevice bond0 zugeordnet werden. Natürlich muss dann auch das Script /etc/init.d/net.bond0 entsprechend angepasst werden. Nachfolgend das gesamte Script:

Code:

#!/sbin/runscript
# Copyright 1999-2004 Gentoo Technologies, Inc.
# Distributed under the terms of the GNU General Public License v2
# $Header: /home/cvsroot/gentoo-src/rc-scripts/init.d/net.eth0,v 1.41 2004/05/10 14:16:35 agriffis Exp $

#NB: Config is in /etc/conf.d/net

if [[ -n $NET_DEBUG ]]; then
set -x
devnull=/dev/stderr
else
devnull=/dev/null
fi

# For pcmcia users. note that pcmcia must be added to the same
# runlevel as the net.* script that needs it.
depend() {
use hotplug pcmcia
}

checkconfig() {
if [[ -z "${ifconfig_IFACE}" ]]; then
   eerror "Please make sure that /etc/conf.d/net has \$ifconfig_$IFACE set"
   eerror "(or \$iface_$IFACE for old-style configuration)"
   return 1
fi
if [[ -n "${vlans_IFACE}" && ! -x /sbin/vconfig ]]; then
   eerror "For VLAN (802.1q) support, emerge net-misc/vconfig"
   return 1
fi
}

# Fix bug 50039 (init.d/net.eth0 localization)
# Some other commands in this script might need to be wrapped, but
# we'll get them one-by-one. Note that LC_ALL trumps LC_anything_else
# according to locale(7)
ifconfig() {
LC_ALL=C /sbin/ifconfig "$@"
}

# setup_vars: setup variables based on $1 and content of /etc/conf.d/net
# The following variables are set, which should be declared local by
# the calling routine.
# status_IFACE       (up or '')
# vlans_IFACE          (space-separated list)
# ifconfig_IFACE       (array of ifconfig lines, replaces iface_IFACE)
# dhcpcd_IFACE       (command-line args for dhcpcd)
# routes_IFACE       (array of route lines)
# inet6_IFACE          (array of inet6 lines)
# ifconfig_fallback_IFACE (fallback ifconfig if dhcp fails)
# slaves_IFACE (slaves of bonding interface added by Klaus Henske)
setup_vars() {
local i iface="${1//\./_}"

status_IFACE="$(ifconfig ${1} 2>${devnull} | gawk '$1 == "UP" {print "up"}')"
eval vlans_IFACE=\"\$\{iface_${iface}_vlans\}\"
eval ifconfig_IFACE=( \"\$\{ifconfig_$iface\[@\]\}\" )
eval dhcpcd_IFACE=\"\$\{dhcpcd_$iface\}\"
eval routes_IFACE=( \"\$\{routes_$iface\[@\]\}\" )
eval slaves_IFACE=( \"\$\{slaves_$iface\[@\]\}\" )
eval inet6_IFACE=( \"\$\{inet6_$iface\[@\]\}\" )
eval ifconfig_fallback_IFACE=( \"\$\{ifconfig_fallback_$iface\[@\]\}\" )

# BACKWARD COMPATIBILITY: populate the ifconfig_IFACE array
# if iface_IFACE is set (fex. iface_eth0 instead of ifconfig_eth0)
eval local iface_IFACE=\"\$\{iface_$iface\}\"
if [[ -n ${iface_IFACE} && -z ${ifconfig_IFACE} ]]; then
   # Make sure these get evaluated as arrays
   local -a aliases broadcasts netmasks

   # Start with the primary interface
   ifconfig_IFACE=( "${iface_IFACE}" )

   # ..then add aliases
   eval aliases=( \$\{alias_$iface\} )
   eval broadcasts=( \$\{broadcast_$iface\} )
   eval netmasks=( \$\{netmask_$iface\} )
   for ((i = 0; i < ${#aliases[@]}; i = i + 1)); do
      ifconfig_IFACE[i+1]="${aliases[i]} ${broadcasts[i]:+broadcast ${broadcasts[i]}} ${netmasks[i]:+netmask ${netmasks[i]}}"
   done
fi

# BACKWARD COMPATIBILITY: check for space-separated inet6 addresses
if [[ ${#inet6_IFACE[@]} == 1 && ${inet6_IFACE} == *' '* ]]; then
   inet6_IFACE=( ${inet6_IFACE} )
fi
}

iface_start() {
local IFACE=${1} i x retval
checkconfig || return 1

if [[ ${ifconfig_IFACE} != dhcp ]]; then
   # Show the address, but catch if this interface will be inet6 only
   i=${ifconfig_IFACE%% *}
   if [[ ${i} == *.*.*.* ]]; then
      ebegin "Bringing ${IFACE} up (${i})"
   else
      ebegin "Bringing ${IFACE} up"
   fi
   # ifconfig does not always return failure ..
   ifconfig ${IFACE} ${ifconfig_IFACE} >${devnull} && \
   ifconfig ${IFACE} up &>${devnull}
## added for bond support by Klaus Henske
   if [[ -n ${slaves_IFACE} ]]; then
      /sbin/ifenslave ${IFACE} ${slaves_IFACE} &>${devnull}
   fi
##
   eend $? || return $?
else
   # Check that eth0 was not brought up by the kernel ...
   if [[ ${status_IFACE} == up ]]; then
      einfo "Keeping kernel configuration for ${IFACE}"
   else
      ebegin "Bringing ${IFACE} up via DHCP"
      /sbin/dhcpcd ${dhcpcd_IFACE} ${IFACE}
      retval=$?
      eend $retval
      if [[ $retval == 0 ]]; then
         # DHCP succeeded, show address retrieved
         i=$(ifconfig ${IFACE} | grep -m1 -o 'inet addr:[^ ]*' |
            cut -d: -f2)
         [[ -n ${i} ]] && einfo " ${IFACE} received address ${i}"
      elif [[ -n "${ifconfig_fallback_IFACE}" ]]; then
         # DHCP failed, try fallback.
         # Show the address, but catch if this interface will be inet6 only
         i=${ifconfig_fallback_IFACE%% *}
         if [[ ${i} == *.*.*.* ]]; then
            ebegin "Using fallback configuration (${i}) for ${IFACE}"
         else
            ebegin "Using fallback configuration for ${IFACE}"
         fi
         ifconfig ${IFACE} ${ifconfig_fallback_IFACE} >${devnull} && \
         ifconfig ${IFACE} up &>${devnull}
         eend $? || return $?
      else
         return $retval
      fi
   fi
fi

if [[ ${#ifconfig_IFACE[@]} -gt 1 ]]; then
   einfo " Adding aliases"
   for ((i = 1; i < ${#ifconfig_IFACE[@]}; i = i + 1)); do
      ebegin " ${IFACE}:${i} (${ifconfig_IFACE[i]%% *})"
      ifconfig ${IFACE}:${i} ${ifconfig_IFACE[i]}
      eend $?
   done
fi

if [[ -n ${inet6_IFACE} ]]; then
   einfo " Adding inet6 addresses"
   for ((i = 0; i < ${#inet6_IFACE[@]}; i = i + 1)); do
      ebegin " ${IFACE} inet6 add ${inet6_IFACE[i]}"
      ifconfig ${IFACE} inet6 add ${inet6_IFACE[i]} >${devnull}
      eend $?
   done
fi

# Set static routes
if [[ -n ${routes_IFACE} ]]; then
   einfo " Adding routes"
   for ((i = 0; i < ${#routes_IFACE[@]}; i = i + 1)); do
      ebegin " ${routes_IFACE[i]}"
      /sbin/route add ${routes_IFACE[i]}
      eend $?
   done
fi

# Set default route if applicable to this interface
if [[ ${gateway} == ${IFACE}/* ]]; then
   local ogw=$(/bin/netstat -rn | awk '$1 == "0.0.0.0" {print $2}')
   local gw=${gateway#*/}
   if [[ ${ogw} != ${gw} ]]; then
      ebegin " Setting default gateway ($gw)"

      # First delete any existing route if it was setup by kernel...
      /sbin/route del default dev ${IFACE} &>${devnull}

      # Second delete old gateway if it was set...
      /sbin/route del default gw ${ogw} &>${devnull}

      # Third add our new default gateway
      /sbin/route add default gw ${gw} >${devnull}
      eend $? || {
         true # need to have some command in here
         # Note: This originally called stop, which is obviously
         # wrong since it's calling with a local version of IFACE.
         # The below code works correctly to abort configuration of
         # the interface, but is commented because we're assuming
         # that default route failure should not cause the interface
         # to be unconfigured.
         #local error=$?
         #ewarn "Aborting configuration of ${IFACE}"
         #iface_stop ${IFACE}
         #return ${error}
      }
   fi
fi

# Enabling rp_filter causes wacky packets to be auto-dropped by
# the kernel. Note that we only do this if it is not set via
# /etc/sysctl.conf ...
if [[ -e /proc/sys/net/ipv4/conf/${IFACE}/rp_filter && \
      -z "$(grep -s '^[^#]*rp_filter' /etc/sysctl.conf)" ]]; then
   echo -n 1 > /proc/sys/net/ipv4/conf/${IFACE}/rp_filter
fi
}

# iface_stop: bring down an interface. Don't trust information in
# /etc/conf.d/net since the configuration might have changed since
# iface_start ran. Instead query for current configuration and bring
# down the interface.
iface_stop() {
local IFACE=${1} i x aliases inet6 count

# Try to do a simple down (no aliases, no inet6, no dhcp)
aliases="$(ifconfig | grep -o "^$IFACE:[0-9]*" | tac)"
inet6="$(ifconfig ${IFACE} | awk '$1 == "inet6" {print $2}')"
if [[ -z ${aliases} && -z ${inet6} && ! -e /var/run/dhcpcd-${IFACE}.pid ]]; then
   ebegin "Bringing ${IFACE} down"
   ifconfig ${IFACE} down &>/dev/null
   eend 0
   return 0
fi

einfo "Bringing ${IFACE} down"

# Stop aliases before primary interface.
# Note this must be done in reverse order, since ifconfig eth0:1
# will remove eth0:2, etc. It might be sufficient to simply remove
# the base interface but we're being safe here.
for i in ${aliases} ${IFACE}; do

   # Delete all the inet6 addresses for this interface
   inet6="$(ifconfig ${i} | awk '$1 == "inet6" {print $3}')"
   if [[ -n ${inet6} ]]; then
      einfo " Removing inet6 addresses"
      for x in ${inet6}; do
         ebegin " ${IFACE} inet6 del ${x}"
         ifconfig ${i} inet6 del ${x}
         eend $?
      done
   fi

   # Stop DHCP (should be N/A for aliases)
   # Don't trust current configuration... investigate ourselves
   if /sbin/dhcpcd -z ${i} &>${devnull}; then
      ebegin " Releasing DHCP lease for ${IFACE}"
      for ((count = 0; count < 9; count = count + 1)); do
         /sbin/dhcpcd -z ${i} &>${devnull} || break
         sleep 1
      done
      [[ ${count} -lt 9 ]]
      eend $? "Timed out"
   fi
   ebegin " Stopping ${i}"
   ifconfig ${i} down &>${devnull}
   eend 0
done

return 0
}

start() {
# These variables are set by setup_vars
local status_IFACE vlans_IFACE dhcpcd_IFACE
local -a ifconfig_IFACE routes_IFACE inet6_IFACE

# Call user-defined preup function if it exists
if [[ $(type -t preup) == function ]]; then
   einfo "Running preup function"
   preup ${IFACE} || {
      eerror "preup ${IFACE} failed"
      return 1
   }
fi

# Start the primary interface and aliases
setup_vars ${IFACE}
iface_start ${IFACE} || return 1

# Start vlans
local vlan
for vlan in ${vlans_IFACE}; do
   /sbin/vconfig add ${IFACE} ${vlan} >${devnull}
   setup_vars ${IFACE}.${vlan}
   iface_start ${IFACE}.${vlan}
done

# Call user-defined postup function if it exists
if [[ $(type -t postup) == function ]]; then
   einfo "Running postup function"
   postup ${IFACE}
fi
}

stop() {
# Call user-defined predown function if it exists
if [[ $(type -t predown) == function ]]; then
   einfo "Running predown function"
   predown ${IFACE}
fi

# Don't depend on setup_vars since configuration might have changed.
# Investigate current configuration instead.
local vlan
for vlan in $(ifconfig | grep -o "^${IFACE}\.[^ ]*"); do
   iface_stop ${vlan}
   /sbin/vconfig rem ${vlan} >${devnull}
done

iface_stop ${IFACE} || return 1 # always succeeds, btw

# Call user-defined postdown function if it exists
if [[ $(type -t postdown) == function ]]; then
   einfo "Running postdown function"
   postdown ${IFACE}
fi
}

# vim:ts=4

Durch dieses Vorgehen gewährleistet, das das Device bond0 von jeden Dienst korrekt angesprochen wird. Natürlich nicht vergessen, überall dort, wo Bezug darauf genommen wird, bond0 als Ethernetdevice anzugeben, wie zum Beispiel in der Datei /etc/conf.d/dhcp:

Code:

IFACE="bond0"

Wenn man mehrere Netzwerkkarten zu einer zusammenschliesen will, kein Prob, eben die entsprechenen slaves_IFACE Anweisung dazuschreiben.

So das wärs für heute. Dann viel Spass beim Bonding

ossi · Apprentice Joined: 11 Apr 2004 Posts: 221 Location: DDR

habe ich jetzt ein Deja-Vu :roll:

oder hatten wir das hier schon mal :?:

khenske · n00b Joined: 19 Dec 2003 Posts: 23

Kann sein, so funktiioniert es jedenfalls, und ich habe es geschrieben, weil es sonst Probleme gibt mit nachfolgenen Anwedungen (dhcp, bind). So gibt es sie nicht

Warum, liegt daran, das es wichtig für die internen Kerneltabellen.

Damit die Adresstable so ist, ist es gut, das die Initiialisierung so vor sich geht:

In amderem Falle kann es passieren, das bond0 nicht den richtigen Tabellenplatz bekommt.

Deswegen das in dieser Ausführlichkeit.

Earthwings · Posted: Thu Jan 06, 2005 1:23 am Post subject: