X-Git-Url: http://git.onelab.eu/?p=build.git;a=blobdiff_plain;f=lbuild-initvm.sh;h=5dedf7f8b31797f2762ccd6a7c17ca47d74b4b68;hp=c355c6ca577aea5be7f0fd10ed82cb3b9ecc91a8;hb=9770ae10ecdfae9bdcaa1232ac47c94d4b1891ee;hpb=1b7c5a2156560d67e5dbc738d24d76deeb8ddbce diff --git a/lbuild-initvm.sh b/lbuild-initvm.sh index c355c6ca..5dedf7f8 100755 --- a/lbuild-initvm.sh +++ b/lbuild-initvm.sh @@ -13,16 +13,27 @@ BUILD_DIR=$(pwd) # pkgs parsing utilities + lbuild-bridge.sh export PATH=$(dirname $0):$PATH -# old guests have e.g. mount in /bin but this is no longer part of +# old guests have e.g. mount in /bin but this is no longer part of # the standard PATH in recent hosts after usrmove, so let's keep it simple export PATH=$PATH:/bin:/sbin . build.common -DEFAULT_FCDISTRO=f20 +# xxx fixme : we pass $lxc around in functions, +# but in fact then we use lxc_root which is a global.. +# it works, but this really is poor practice +# we should have an lxc_root function instead +function lxcroot () { + local lxc=$1; shift + echo /vservers/$lxc +} + +# XXX fixme : when creating a 32bits VM we need to call linux32 as appropriate...s + +DEFAULT_FCDISTRO=f29 DEFAULT_PLDISTRO=lxc DEFAULT_PERSONALITY=linux64 -DEFAULT_MEMORY=512 +DEFAULT_MEMORY=3072 ########## # constant @@ -32,80 +43,82 @@ PUBLIC_BRIDGE=br0 VIF_GUEST=eth0 ########## -FEDORA_PREINSTALLED="yum initscripts passwd rsyslog vim-minimal dhclient chkconfig rootfiles policycoreutils openssh-server openssh-clients" +FEDORA_MIRROR="http://mirror.onelab.eu/" +FEDORA_MIRROR_KEYS="http://mirror.onelab.eu/keys/" +FEDORA_PREINSTALLED="dnf dnf-yum passwd rsyslog vim-minimal dhclient chkconfig rootfiles policycoreutils openssh-server openssh-clients" DEBIAN_PREINSTALLED="openssh-server openssh-client" ########## networking utilities function gethostbyname () { - hostname=$1 - python -c "import socket; print socket.gethostbyname('"$hostname"')" 2> /dev/null + local hostname=$1 + python3 -c "import socket; print(socket.gethostbyname('"$hostname"'))" 2> /dev/null } # e.g. 21 -> 255.255.248.0 function masklen_to_netmask () { - masklen=$1; shift - python <=1 and masklen<=32): - print "Wrong masklen",masklen +masklen = $masklen +if not (1 <= masklen <= 32): + print("Wrong masklen", masklen) exit(1) -result=[] +result = [] for i in range(4): - if masklen>=8: + if masklen >= 8: result.append(8) - masklen-=8 + masklen -= 8 else: result.append(masklen) - masklen=0 -print ".".join([ str(256-2**(8-i)) for i in result ]) - + masklen = 0 +print(".".join([ str(256-2**(8-i)) for i in result ])) EOF } ############################## -# return yum or debootstrap +# return dnf or debootstrap function package_method () { - fcdistro=$1; shift + local fcdistro=$1; shift case $fcdistro in - f[0-9]*|centos[0-9]*|sl[0-9]*) echo yum ;; - squeeze|wheezy|jessie|oneiric|precise|quantal|raring|saucy) echo debootstrap ;; - *) echo Unknown distro $fcdistro ;; - esac + f[0-9]*|centos[0-9]*|sl[0-9]*) + echo dnf ;; + wheezy|jessie|precise|trusty|utopic|vivid|wily|xenial) + echo debootstrap ;; + *) + echo Unknown distro $fcdistro ;; + esac } # return arch from debian distro and personality function canonical_arch () { - personality=$1; shift - fcdistro=$1; shift + local personality=$1; shift + local fcdistro=$1; shift case $(package_method $fcdistro) in - yum) - case $personality in *32) echo i386 ;; *64) echo x86_64 ;; *) echo Unknown-arch-1 ;; esac ;; - debootstrap) - case $personality in *32) echo i386 ;; *64) echo amd64 ;; *) echo Unknown-arch-2 ;; esac ;; - *) - echo Unknown-arch-3 ;; + dnf) + case $personality in + *32) echo i386 ;; + *64) echo x86_64 ;; + *) echo Unknown-arch-1 ;; + esac ;; + debootstrap) + case $personality in + *32) echo i386 ;; + *64) echo amd64 ;; + *) echo Unknown-arch-2 ;; + esac ;; + *) + echo Unknown-arch-3 ;; esac } # the new test framework creates /timestamp in /vservers/ *before* populating it -function almost_empty () { - dir="$1"; shift ; +function almost_empty () { + local dir="$1"; shift ; # non existing is fine - [ ! -d $dir ] && return 0; + [ ! -d $dir ] && return 0; # need to have at most one file - count=$(cd $dir; ls | wc -l); [ $count -le 1 ]; -} - -############################## -function check_yum_installed () { - package=$1; shift - rpm -q $package >& /dev/null || yum -y install $package -} - -function check_yumgroup_installed () { - group="$1"; shift - yum grouplist "$group" | grep -q Installed || { yum -y groupinstall "$group" ; } + local count=$(cd $dir; ls | wc -l) + [ $count -le 1 ] } ############################## @@ -113,36 +126,42 @@ function fedora_install() { set -x set -e - cache=/var/cache/lxc/fedora/$arch/$release - - mkdir -p /var/lock/subsys/ + local lxc=$1; shift + local lxc_root=$(lxcroot $lxc) + + local cache=/var/cache/lxc/fedora/$arch/${fedora_release} + mkdir -p $cache + ( - flock -n -x 200 || { echo "Cache repository is busy." ; return 1 ; } + flock --exclusive --timeout 60 200 || { echo "Cache repository is busy." ; return 1 ; } if [ ! -e "$cache/rootfs" ]; then echo "Getting cache download in $cache/rootfs ... " - fedora_download || { echo "Failed to download 'fedora base'"; return 1; } + fedora_download $cache || { echo "Failed to download 'fedora base'"; return 1; } else echo "Updating cache $cache/rootfs ..." - if ! yum --installroot $cache/rootfs -y --nogpgcheck update ; then + if ! dnf --installroot $cache/rootfs --releasever=${fedora_release} -y --nogpgcheck update ; then echo "Failed to update 'fedora base', continuing with last known good cache" else echo "Update finished" fi fi - echo "Copy $cache/rootfs to $lxc_root ... " - rsync -a $cache/rootfs/ $lxc_root/ - + echo "Filling $lxc_root from $cache/rootfs ... " + rsync -a $cache/rootfs/ $lxc_root/ + return 0 - ) 200>/var/lock/subsys/lxc + ) 200> $cache/lock return $? } function fedora_download() { set -x + + local cache=$1; shift + # check the mini fedora was not already downloaded INSTALL_ROOT=$cache/partial echo $INSTALL_ROOT @@ -152,54 +171,62 @@ function fedora_download() { mkdir -p $INSTALL_ROOT || { echo "Failed to create '$INSTALL_ROOT' directory" ; return 1; } - mkdir -p $INSTALL_ROOT/etc/yum.repos.d + mkdir -p $INSTALL_ROOT/etc/yum.repos.d mkdir -p $INSTALL_ROOT/dev mknod -m 0444 $INSTALL_ROOT/dev/random c 1 8 mknod -m 0444 $INSTALL_ROOT/dev/urandom c 1 9 # copy yum config and repo files cp /etc/yum.conf $INSTALL_ROOT/etc/ - cp /etc/yum.repos.d/fedora* $INSTALL_ROOT/etc/yum.repos.d/ + cp /etc/yum.repos.d/fedora{,-updates}.repo $INSTALL_ROOT/etc/yum.repos.d/ - # append fedora repo files with desired $release and $basearch + # append fedora repo files with hardwired releasever and basearch for f in $INSTALL_ROOT/etc/yum.repos.d/* ; do - sed -i "s/\$basearch/$arch/g; s/\$releasever/$release/g;" $f - done - - MIRROR_URL=http://mirror.onelab.eu/fedora/releases/$release/Everything/$arch/os - RELEASE_URL1="$MIRROR_URL/Packages/fedora-release-$release-1.noarch.rpm" - # with fedora18 the rpms are scattered by first name - RELEASE_URL2="$MIRROR_URL/Packages/f/fedora-release-$release-1.noarch.rpm" - RELEASE_TARGET=$INSTALL_ROOT/fedora-release-$release.noarch.rpm - found="" - for attempt in $RELEASE_URL1 $RELEASE_URL2; do - if curl -f $attempt -o $RELEASE_TARGET ; then - echo "Retrieved $attempt" - found=true - break - else - echo "Failed attempt $attempt" - fi + sed -i "s/\$basearch/$arch/g; s/\$releasever/${fedora_release}/g;" $f done - [ -n "$found" ] || { echo "Could not retrieve fedora-release rpm - exiting" ; exit 1; } - + +# looks like all this business about fetching fedora-release is not needed +# it does +# MIRROR_URL=$FEDORA_MIRROR/fedora/releases/${fedora_release}/Everything/$arch/os +# # since fedora18 the rpms are scattered by first name +# # first try the second version of fedora-release first +# RELEASE_URLS="" +# local subindex +# for subindex in 3 2 1; do +# RELEASE_URLS="$RELEASE_URLS $MIRROR_URL/Packages/f/fedora-release-${fedora_release}-${subindex}.noarch.rpm" +# done +# +# RELEASE_TARGET=$INSTALL_ROOT/fedora-release-${fedora_release}.noarch.rpm +# local found="" +# local attempt +# for attempt in $RELEASE_URLS; do +# if curl --silent --fail $attempt -o $RELEASE_TARGET; then +# echo "Successfully Retrieved $attempt" +# found=true +# break +# else +# echo "Failed (not to worry about) with attempt $attempt" +# fi +# done +# [ -n "$found" ] || { echo "Could not retrieve fedora-release rpm - exiting" ; exit 1; } + mkdir -p $INSTALL_ROOT/var/lib/rpm rpm --root $INSTALL_ROOT --initdb # when installing f12 this apparently is already present, so ignore result - rpm --root $INSTALL_ROOT -ivh $INSTALL_ROOT/fedora-release-$release.noarch.rpm || : +# rpm --root $INSTALL_ROOT -ivh $INSTALL_ROOT/fedora-release-${fedora_release}.noarch.rpm || : # however f12 root images won't get created on a f18 host # (the issue here is the same as the one we ran into when dealing with a vs-box) # in a nutshell, in f12 the glibc-common and filesystem rpms have an apparent conflict - # >>> file /usr/lib/locale from install of glibc-common-2.11.2-3.x86_64 conflicts + # >>> file /usr/lib/locale from install of glibc-common-2.11.2-3.x86_64 conflicts # with file from package filesystem-2.4.30-2.fc12.x86_64 - # in fact this was - of course - allowed by f12's rpm but later on a fix was made + # in fact this was - of course - allowed by f12's rpm but later on a fix was made # http://rpm.org/gitweb?p=rpm.git;a=commitdiff;h=cf1095648194104a81a58abead05974a5bfa3b9a # So ideally if we want to be able to build f12 images from f18 we need an rpm that has # this patch undone, like we have in place on our f14 boxes (our f14 boxes need a f18-like rpm) - YUM="yum --installroot=$INSTALL_ROOT --nogpgcheck -y" - echo "$YUM install $FEDORA_PREINSTALLED" - $YUM install $FEDORA_PREINSTALLED || { echo "Failed to download rootfs, aborting." ; return 1; } + DNF="dnf --installroot=$INSTALL_ROOT --nogpgcheck -y" + echo "$DNF install $FEDORA_PREINSTALLED" + $DNF install $FEDORA_PREINSTALLED || { echo "Failed to download rootfs, aborting." ; return 1; } mv "$INSTALL_ROOT" "$cache/rootfs" echo "Download complete." @@ -213,30 +240,23 @@ function fedora_configure() { set -x set -e + local lxc=$1; shift + local fcdistro=$1; shift + local lxc_root=$(lxcroot $lxc) + # disable selinux in fedora mkdir -p $lxc_root/selinux echo 0 > $lxc_root/selinux/enforce - # set the hostname - case "$fcdistro" in - f18|f2?) - cat < ${lxc_root}/etc/hostname -$GUEST_HOSTNAME -EOF - echo ;; - *) - cat < ${lxc_root}/etc/sysconfig/network + # enable networking and set hostname + cat < ${lxc_root}/etc/sysconfig/network NETWORKING=yes -HOSTNAME=$GUEST_HOSTNAME EOF - # set minimal hosts - cat < $lxc_root/etc/hosts -127.0.0.1 localhost $GUEST_HOSTNAME + cat < ${lxc_root}/etc/hostname +$GUEST_HOSTNAME EOF - echo ;; - esac - dev_path="${lxc_root}/dev" + local dev_path="${lxc_root}/dev" rm -rf $dev_path mkdir -p $dev_path mknod -m 666 ${dev_path}/null c 1 3 @@ -256,159 +276,146 @@ EOF mknod -m 600 ${dev_path}/initctl p mknod -m 666 ${dev_path}/ptmx c 5 2 - if [ "$(echo $fcdistro | cut -d"f" -f2)" -le "14" ]; then - fedora_configure_init - else - fedora_configure_systemd - fi + fedora_configure_systemd $lxc - guest_ifcfg=${lxc_root}/etc/sysconfig/network-scripts/ifcfg-$VIF_GUEST - ( [ -n "$BUILD_MODE" ] && write_guest_ifcfg_build || write_guest_ifcfg_test ) > $guest_ifcfg + local guest_ifcfg=${lxc_root}/etc/sysconfig/network-scripts/ifcfg-$VIF_GUEST + mkdir -p $(dirname ${guest_ifcfg}) + # starting with f27, we go for NetworkManager + # no more NM_CONTROLLED nonsense + if [ -n "$NAT_MODE" ]; then + write_guest_ifcfg_natip + else + write_guest_ifcfg_publicip + fi > $guest_ifcfg [ -z "$IMAGE" ] && fedora_configure_yum $lxc $fcdistro $pldistro return 0 } -function fedora_configure_init() { - set -e - set -x - sed -i 's|.sbin.start_udev||' ${lxc_root}/etc/rc.sysinit - sed -i 's|.sbin.start_udev||' ${lxc_root}/etc/rc.d/rc.sysinit - # don't mount devpts, for pete's sake - sed -i 's/^.*dev.pts.*$/#\0/' ${lxc_root}/etc/rc.sysinit - sed -i 's/^.*dev.pts.*$/#\0/' ${lxc_root}/etc/rc.d/rc.sysinit - chroot ${lxc_root} chkconfig udev-post off - chroot ${lxc_root} chkconfig network on -} - # this code of course is for guests that do run on systemd function fedora_configure_systemd() { set -e set -x - # so ignore if we can't find /etc/systemd at all + local lxc=$1; shift + local lxc_root=$(lxcroot $lxc) + + # so ignore if we can't find /etc/systemd at all [ -d ${lxc_root}/etc/systemd ] || return 0 # otherwise let's proceed ln -sf /lib/systemd/system/multi-user.target ${lxc_root}/etc/systemd/system/default.target touch ${lxc_root}/etc/fstab ln -sf /dev/null ${lxc_root}/etc/systemd/system/udev.service -# Thierry - Feb 2013 -# this was intended for f16 initially, in order to enable getty that otherwise would not start -# having a getty running is helpful only if ssh won't start though, and we see a correlation between -# VM's that refuse to lxc-stop and VM's that run crazy getty's +# Thierry - Feb 2013 relying on getty is looking for trouble # so, turning getty off for now instead -# #dependency on a device unit fails it specially that we disabled udev # sed -i 's/After=dev-%i.device/After=/' ${lxc_root}/lib/systemd/system/getty\@.service ln -sf /dev/null ${lxc_root}/etc/systemd/system/"getty@.service" rm -f ${lxc_root}/etc/systemd/system/getty.target.wants/*service || : # can't seem to handle this one with systemctl - chroot ${lxc_root} chkconfig network on + chroot ${lxc_root} $personality chkconfig network on } # overwrite container yum config function fedora_configure_yum () { - set -x - set -e + set -x + set -e trap failure ERR INT - lxc=$1; shift - fcdistro=$1; shift - pldistro=$1; shift + local lxc=$1; shift + local fcdistro=$1; shift + local pldistro=$1; shift + local lxc_root=$(lxcroot $lxc) # rpm --rebuilddb - chroot $lxc_root rpm --rebuilddb + chroot ${lxc_root} $personality rpm --rebuilddb echo "Initializing yum.repos.d in $lxc" rm -f $lxc_root/etc/yum.repos.d/* - cat > $lxc_root/etc/yum.repos.d/building.repo < $lxc_root/etc/yum.repos.d/myplc.repo < $lxc_root/etc/yum.repos.d/myplc.repo < ${lxc_root}/etc/hostname $GUEST_HOSTNAME EOF - + } function debian_configure () { - guest_interfaces=${lxc_root}/etc/network/interfaces - ( [ -n "$BUILD_MODE" ] && write_guest_interfaces_build || write_guest_interfaces_test ) > $guest_interfaces + local guest_interfaces=${lxc_root}/etc/network/interfaces + ( [ -n "$NAT_MODE" ] && write_guest_interfaces_natip || write_guest_interfaces_publicip ) > $guest_interfaces } -function write_guest_interfaces_build () { +function write_guest_interfaces_natip () { cat < $lxc_root/etc/hosts - + # grant ssh access from host to guest mkdir -p $lxc_root/root/.ssh cat /root/.ssh/id_rsa.pub >> $lxc_root/root/.ssh/authorized_keys @@ -466,17 +482,27 @@ function setup_lxc() { chmod 600 $lxc_root/root/.ssh/authorized_keys # don't keep the input xml, this can be retrieved at all times with virsh dumpxml - config_xml=/tmp/$lxc.xml - ( [ -n "$BUILD_MODE" ] && write_lxc_xml_build $lxc || write_lxc_xml_test $lxc ) > $config_xml - + local config_xml=/tmp/$lxc.xml + ( [ -n "$NAT_MODE" ] && write_lxc_xml_natip $lxc || write_lxc_xml_publicip $lxc ) > $config_xml + # define lxc container for libvirt virsh -c lxc:/// define $config_xml return 0 } -function write_lxc_xml_test () { - lxc=$1; shift +# this part does not belong in a domain any more +# but goes in a network object of its own existence +# +# host-bridge +# +# +# +# + +function write_lxc_xml_publicip () { + local lxc=$1; shift + local lxc_root=$(lxcroot $lxc) cat < $lxc @@ -505,17 +531,14 @@ function write_lxc_xml_test () { - - host-bridge - - - EOF } -function write_lxc_xml_build () { - lxc=$1; shift +# grant build guests the ability to do mknods +function write_lxc_xml_natip () { + local lxc=$1; shift + local lxc_root=$(lxcroot $lxc) cat < $lxc @@ -526,6 +549,9 @@ function write_lxc_xml_build () { + + + 1 @@ -548,19 +574,18 @@ EOF } # this one is dhcp-based -function write_guest_ifcfg_build () { +function write_guest_ifcfg_natip () { cat < $lxc_root/init-lxc.rpms - ;; - debootstrap) - # for ubuntu - if grep -iq ubuntu /vservers/$lxc/etc/lsb-release 2> /dev/null; then - # on ubuntu, at this point we end up with a single feed in /etc/apt/sources.list - # we need at least to add the 'universe' feed for python-rpm - ( cd /vservers/$lxc/etc/apt ; head -1 sources.list | sed -e s,main,universe, > sources.list.d/universe.list ) - # also adding a link to updates sounds about right - ( cd /vservers/$lxc/etc/apt ; head -1 sources.list | sed -e 's, main,-updates main,' > sources.list.d/updates.list ) - # tell apt about the changes - chroot /vservers/$lxc apt-get update - fi - for package in $packages ; do - # container not started yet - #virsh -c lxc:/// lxc-enter-namespace $lxc /bin/bash -c "apt-get install -y $package" || : - chroot $lxc_root apt-get install -y $package || : - done - ### xxx todo install groups with apt.. - ;; - *) - echo "unknown pkg_method $pkg_method" - ;; + dnf) + # --allowerasing required starting with fedora24 + # + local has_dnf="" + chroot ${lxc_root} $personality dnf --version && has_dnf=true + if [ -n "$has_dnf" ]; then + echo "container has dnf - invoking with --allowerasing" + local pkg_installer="dnf -y install --allowerasing" + local grp_installer="dnf -y groupinstall --allowerasing" + else + echo "container has only dnf" + local pkg_installer="dnf -y install" + local grp_installer="dnf -y groupinstall" + fi + [ -n "$packages" ] && chroot ${lxc_root} $personality $pkg_installer $packages + for group_plus in $groups; do + local group=$(echo $group_plus | sed -e "s,+++, ,g") + chroot ${lxc_root} $personality $grp_installer "$group" + done + # store current rpm list in /init-lxc.rpms in case we need to check the contents + chroot ${lxc_root} $personality rpm -aq > $lxc_root/init-lxc.rpms + ;; + debootstrap) + # for ubuntu + if grep -iq ubuntu /vservers/$lxc/etc/lsb-release 2> /dev/null; then + # on ubuntu, at this point we end up with a single feed in /etc/apt/sources.list + # we need at least to add the 'universe' feed for python-rpm + ( cd /vservers/$lxc/etc/apt ; head -1 sources.list | sed -e s,main,universe, > sources.list.d/universe.list ) + # also adding a link to updates sounds about right + ( cd /vservers/$lxc/etc/apt ; head -1 sources.list | sed -e 's, main,-updates main,' > sources.list.d/updates.list ) + # tell apt about the changes + chroot /vservers/$lxc apt-get update + fi + for package in $packages ; do + # container not started yet + #virsh -c lxc:/// lxc-enter-namespace $lxc /usr/bin/$personality /bin/bash -c "apt-get install -y $package" || : + chroot ${lxc_root} $personality apt-get install -y $package || : + done + ### xxx todo install groups with apt.. + ;; + *) + echo "unknown pkg_method $pkg_method" + ;; esac return 0 } function post_install () { - lxc=$1; shift - personality=$1; shift - if [ -n "$BUILD_MODE" ] ; then - post_install_build $lxc $personality - virsh -c lxc:/// start $lxc - # manually run dhclient in guest - somehow this network won't start on its own - virsh -c lxc:/// lxc-enter-namespace $lxc /bin/bash -c "dhclient $VIF_GUEST" - else - post_install_myplc $lxc $personality - virsh -c lxc:/// start $lxc -# it sounds like we don't need ssh per se any more -# it still makes sense to wait for network readiness though -# some day maybe... - wait_for_ssh $lxc - fi + local lxc=$1; shift + local personality=$1; shift + local lxc_root=$(lxcroot $lxc) # setup localtime from the host cp /etc/localtime $lxc_root/etc/localtime + sshd_disable_password_auth $lxc + # post install hook + [ -n "$NAT_MODE" ] && post_install_natip $lxc $personality || post_install_myplc $lxc $personality + # start the VM unless specified otherwise + if [ -n "$START_VM" ] ; then + echo Starting guest $lxc + virsh -c lxc:/// start $lxc + if [ -n "$NAT_MODE" ] ; then + wait_for_ssh $lxc + else + wait_for_ssh $lxc $GUEST_IP + fi + fi } -function post_install_build () { +# just in case, let's stay on the safe side +function sshd_disable_password_auth () { + local lxc=$1; shift + local lxc_root=$(lxcroot $lxc) + sed --in-place=.password -e 's,^#\?PasswordAuthentication.*,PasswordAuthentication no,' \ + $lxc_root/etc/ssh/sshd_config +} - set -x - set -e +function post_install_natip () { + + set -x + set -e trap failure ERR INT - lxc=$1; shift - personality=$1; shift + local lxc=$1; shift + local personality=$1; shift + local lxc_root=$(lxcroot $lxc) ### From myplc-devel-native.spec # be careful to backslash $ in this, otherwise it's the root context that's going to do the evaluation - cat << EOF | chroot $lxc_root bash -x - + cat << EOF | chroot ${lxc_root} $personality bash -x + # customize root's prompt /bin/cat << PROFILE > /root/.profile export PS1="[$lxc] \\w # " PROFILE EOF - + } function post_install_myplc () { - set -x - set -e + set -x + set -e trap failure ERR INT - lxc=$1; shift - personality=$1; shift + local lxc=$1; shift + local personality=$1; shift + local lxc_root=$(lxcroot $lxc) # be careful to backslash $ in this, otherwise it's the root context that's going to do the evaluation - cat << EOF | chroot $lxc_root bash -x + cat << EOF | chroot ${lxc_root} $personality bash -x # create /etc/sysconfig/network if missing [ -f /etc/sysconfig/network ] || /bin/echo NETWORKING=yes > /etc/sysconfig/network @@ -703,31 +753,51 @@ PROFILE EOF } +######################################## +# workaround for broken lxc-enter-namespace +# 1st version was relying on virsh net-dhcp-leases +# however this was too fragile, would not work for fedora14 containers +# WARNING: this code is duplicated in lbuild-nightly.sh +function guest_ipv4() { + local lxc=$1; shift + + local mac=$(virsh -c lxc:/// domiflist $lxc | egrep 'network|bridge' | awk '{print $5;}') + # sanity check + [ -z "$mac" ] && return 0 + arp -en | grep "$mac" | awk '{print $1;}' +} + function wait_for_ssh () { set -x set -e - #trap failure ERR INT - lxc=$1; shift - - echo network in guest is up, waiting for ssh... + local lxc=$1; shift + + # if run in public_ip mode, we know the IP of the guest and it is specified here + [ -n "$1" ] && { guest_ip=$1; shift; } + + #wait max 2 min for sshd to start + local success="" + local current_time=$(date +%s) + local stop_time=$(($current_time + 120)) - #wait max 2 min for sshd to start - ssh_up="" - current_time=$(date +%s) - stop_time=$(($current_time + 120)) - - counter=1 + local counter=1 while [ "$current_time" -lt "$stop_time" ] ; do - echo "$counter-th attempt to reach sshd in container $lxc ..." - ssh -o "StrictHostKeyChecking no" $GUEST_IP 'uname -i' && { ssh_up=true; echo "SSHD in container $lxc is UP"; break ; } || : - sleep 10 - current_time=$(($current_time + 10)) - counter=$(($counter+1)) + echo "$counter-th attempt to reach sshd in container $lxc ..." + [ -z "$guest_ip" ] && guest_ip=$(guest_ipv4 $lxc) + [ -n "$guest_ip" ] && ssh -o "StrictHostKeyChecking no" $guest_ip 'uname -i' && { + success=true; echo "SSHD in container $lxc is UP on IP $guest_ip"; break ; } || : + # some of our boxes have gone through a long upgrade historically, and + # so they don't end up with the same gid mapping for the ssh_keys + # group as the ones in the guest that result from a fresh install + virsh -c lxc:/// lxc-enter-namespace $lxc /bin/bash -c "chown root:ssh_keys /etc/ssh/*_key" || : + counter=$(($counter+1)) + sleep 10 + current_time=$(date +%s) done # Thierry: this is fatal, let's just exit with a failure here - [ -z $ssh_up ] && { echo "SSHD in container $lxc is not running" ; exit 1 ; } + [ -z $success ] && { echo "SSHD in container $lxc could not be reached (guest_ip=$guest_ip)" ; exit 1 ; } return 0 } @@ -738,7 +808,7 @@ function failure () { } function usage () { - set +x + set +x echo "Usage: $COMMAND [options] lxc-name (aka build mode)" echo "Usage: $COMMAND -n hostname [options] lxc-name (aka test mode)" echo "Description:" @@ -756,11 +826,12 @@ function usage () { echo " by default we use devel.pkgs (build mode) or runtime.pkgs (test mode)" echo " -i image - the location of the rootfs" echo " -m memory - the amount of allocated memory in MB - defaults to $DEFAULT_MEMORY MB" + echo " -s do not start VM" echo " -v be verbose" exit 1 } -### parse args and +### parse args and function main () { #set -e @@ -771,83 +842,90 @@ function main () { exit 1 fi - while getopts "n:f:d:p:r:P:i:m:v" opt ; do - case $opt in - n) GUEST_HOSTNAME=$OPTARG;; - f) fcdistro=$OPTARG;; - d) pldistro=$OPTARG;; - p) personality=$OPTARG;; - r) REPO_URL=$OPTARG;; - P) PREINSTALLED=$OPTARG;; + START_VM=true + while getopts "n:f:d:p:r:P:i:m:sv" opt ; do + case $opt in + n) GUEST_HOSTNAME=$OPTARG;; + f) fcdistro=$OPTARG;; + d) pldistro=$OPTARG;; + p) personality=$OPTARG;; + r) REPO_URL=$OPTARG;; + P) PREINSTALLED=$OPTARG;; i) IMAGE=$OPTARG;; m) MEMORY=$OPTARG;; - v) VERBOSE=true; set -x;; - *) usage ;; - esac + s) START_VM= ;; + v) VERBOSE=true; set -x;; + *) usage ;; + esac done - + shift $(($OPTIND - 1)) # parse fixed arguments [[ -z "$@" ]] && usage - lxc=$1 ; shift - lxc_root=/vservers/$lxc + local lxc=$1 ; shift + local lxc_root=$(lxcroot $lxc) # rainchecks + # when using with the -i option, checking that $lxc_root is void + # is a little too much stress.. almost_empty $lxc_root || \ - { echo "container $lxc already exists in $lxc_root - exiting" ; exit 1 ; } + { echo "container $lxc already exists in $lxc_root - exiting" ; exit 1 ; } virsh -c lxc:/// domuuid $lxc >& /dev/null && \ - { echo "container $lxc already exists in libvirt - exiting" ; exit 1 ; } + { echo "container $lxc already exists in libvirt - exiting" ; exit 1 ; } mkdir -p $lxc_root # if IMAGE, copy the provided rootfs to lxc_root if [ -n "$IMAGE" ] ; then - [ ! -d "$IMAGE" ] && \ - { echo "$IMAGE rootfs folder does not exist - exiting" ; exit 1 ; } - rsync -a $IMAGE/ $lxc_root/ + if [ ! -d "$IMAGE" ]; then + echo "$IMAGE rootfs folder does not exist - exiting" + exit 1 + else + echo "Copying $IMAGE into $lxc_root with rsync --archive --delete" + rsync --archive --delete $IMAGE/ $lxc_root/ + fi fi - # check we've exhausted the arguments [[ -n "$@" ]] && usage - # BUILD_MODE is true unless we specified a hostname - [ -n "$GUEST_HOSTNAME" ] || BUILD_MODE=true + # NAT_MODE is true unless we specified a hostname + [ -n "$GUEST_HOSTNAME" ] || NAT_MODE=true # set default values [ -z "$fcdistro" ] && fcdistro=$DEFAULT_FCDISTRO [ -z "$pldistro" ] && pldistro=$DEFAULT_PLDISTRO [ -z "$personality" ] && personality=$DEFAULT_PERSONALITY [ -z "$MEMORY" ] && MEMORY=$DEFAULT_MEMORY - + # set memory in KB MEMORY=$(($MEMORY * 1024)) - + # the set of preinstalled packages - depends on mode - if [ -z "$PREINSTALLED"] ; then - if [ -n "$BUILD_MODE" ] ; then - PREINSTALLED=devel.pkgs - else - PREINSTALLED=runtime.pkgs - fi + if [ -z "$PREINSTALLED" ] ; then + if [ -n "$NAT_MODE" ] ; then + PREINSTALLED=devel.pkgs + else + PREINSTALLED=runtime.pkgs + fi fi - if [ -n "$BUILD_MODE" ] ; then - # we can now set GUEST_HOSTNAME safely + if [ -n "$NAT_MODE" ] ; then + # we can now set GUEST_HOSTNAME safely [ -z "$GUEST_HOSTNAME" ] && GUEST_HOSTNAME=$(echo $lxc | sed -e 's,\.,-,g') else - # as this command can be used in other contexts, not specifying - # a repo is considered a warning - # use -r none to get rid of this warning - if [ "$REPO_URL" == "none" ] ; then - REPO_URL="" - elif [ -z "$REPO_URL" ] ; then - echo "WARNING -- setting up a yum repo is recommended" - fi + # as this command can be used in other contexts, not specifying + # a repo is considered a warning + # use -r none to get rid of this warning + if [ "$REPO_URL" == "none" ] ; then + REPO_URL="" + elif [ -z "$REPO_URL" ] ; then + echo "WARNING -- setting up a yum repo is recommended" + fi fi ########## - release=$(echo $fcdistro | cut -df -f2) + fedora_release=$(echo $fcdistro | cut -df -f2) if [ "$personality" == "linux32" ]; then arch=i386 @@ -861,26 +939,28 @@ function main () { # compute networking details for the test mode # (build mode relies entirely on dhcp on the private subnet) - if [ -z "$BUILD_MODE" ] ; then + if [ -z "$NAT_MODE" ] ; then - #create_bridge_if_needed $PUBLIC_BRIDGE - lbuild-bridge.sh $PUBLIC_BRIDGE + #create_bridge_if_needed $PUBLIC_BRIDGE + lbuild-bridge.sh $PUBLIC_BRIDGE - GUEST_IP=$(gethostbyname $GUEST_HOSTNAME) - # use same NETMASK as bridge interface br0 - masklen=$(ip addr show $PUBLIC_BRIDGE | grep -v inet6 | grep inet | awk '{print $2;}' | cut -d/ -f2) + GUEST_IP=$(gethostbyname $GUEST_HOSTNAME) + # use same NETMASK as bridge interface br0 + masklen=$(ip addr show $PUBLIC_BRIDGE | grep -v inet6 | grep inet | awk '{print $2;}' | cut -d/ -f2) NETMASK=$(masklen_to_netmask $masklen) GATEWAY=$(ip route show | grep default | awk '{print $3}' | head -1) - VIF_HOST="i$(echo $GUEST_HOSTNAME | cut -d. -f1)" + VIF_HOST="vif$(echo $GUEST_HOSTNAME | cut -d. -f1)" fi - setup_lxc $lxc $fcdistro $pldistro $personality + setup_lxc $lxc $fcdistro $pldistro $personality - devel_or_vtest_tools $lxc $fcdistro $pldistro $personality + # historically this command is for setting up a build or a test VM + # kind of patchy right now though + devel_or_test_tools $lxc $fcdistro $pldistro $personality # container gets started here post_install $lxc $personality - + echo $COMMAND Done exit 0