update-ramdisk-node

#!/bin/sh

# (c) Robin Humble 2009
# licensed under the GPLv3 or later

# run this script on a oneSIS client node to update its OS ramdisk image.

# firstly read /proc/cmdline to figure out how we were booted using oneSIS and
# update the ramdisk portions (if any) of this node's OS to be current.
#
# after the OS files are updated we then automatically run 'update-node' to
# reconfig the oneSIS setup to the latest version in the OS image.
#
# this script understands: rsyncroot, halfroot (a tweak on pure lustreroot)

forceroot=""
if [ $# -gt 1 ]; then
   forceroot=$1
   shift
fi

if [ $# -ne 1 ]; then
   echo "Usage: `basename $0` [root-image-location] dry-run|commit" >&2
   exit 1
fi

case "$1" in
   dry-run)
      debug=1
      dryrun=--dry-run
      echo running in --dry-run mode
      ;;
   commit)
      debug=0
      dryrun=
      ;;
   *)
      echo "Usage: `basename $0` dry-run|commit" >&2
      exit 1
      ;;
esac

# see if we're doing rsync:// with a secret
RSYNC_PW_OPT=""
cmdline=`cat /proc/cmdline | grep "rsync://"`
forcerootline=`echo $forceroot | grep "rsync://"`
if [ -n "$cmdline" -o -n "$forcerootline" ]; then
   if [ -r /etc/rsyncd.passwd ]; then
      RSYNC_PW_OPT="--password-file=/etc/rsyncd.passwd"
   else
      echo "Error! rsync:// found, but failed to find a secret file."
      exit 1
   fi
fi

# determine if we are running on a full 'rsyncroot' node, or on a 'halfroot' node

# find out which root-on-ramdisk image we are by parsing the boot line
cmdline1=`cat /proc/cmdline |grep ' rsyncroot='`
cmdline2=`cat /proc/cmdline |grep ' halfroot '`
if [ -n "$cmdline1" -a -z "$cmdline2" ]; then
   # check that root is on a ramdisk, otherwise we might be trying to run this on the wrong node
   # expect just the one line for a tmpfs root:  "rootfs on / type rootfs (rw)"
   if [ `mount | grep ' / ' | wc -l` -ne 1 ]; then
      echo root is not on a ramdisk. bye bye
      exit 1
   fi
   if [ `mount | grep ' / ' | cut -d' ' -f1` != "rootfs" ]; then
      echo root is not on a ramdisk. bye bye
      exit 1
   fi

   if [ "$forceroot" = "" ]; then
      # eg.
      #  rsyncroot=10.7.10.242:/var/lib/oneSIS/centos-5.2-minimal-lustre-1.6.6/
      # or
      #  rsyncroot=rsync://boot@10.7.10.242:/images/centos-5.2-minimal-lustre-1.6.6/
      root=`echo $cmdline1|sed -e "s/.* rsyncroot=//" -e "s/ .*//"`
      if [ -z "$root" ]; then
         echo cannot parse rsyncroot in /proc/cmdline - exiting
         exit 1
      fi
   else
      root=$forceroot
   fi

   # dirty hack for r nodes that booted over 10.10 gige but then shutdown gige and use 10.9 IPoIB
   cluster=`cat /proc/cmdline | sed -e "s/.* cluster=//" -e "s/ .*//"`
   if [ "$cluster" = "r" -a "`echo $root | grep @10.10.`" != "" ]; then
      if [ "`ifconfig 2>&1 | grep 10.10`" = "" ]; then # no 10.10 up
         root=`echo $root | sed 's/@10.10/@10.9/'`
      fi
   fi

   # rsync over any changes
   rsync -avxHS $RSYNC_PW_OPT $dryrun --exclude=init.log --exclude=/root/ --exclude=/lib/modules/ --exclude=/usr/share/doc --exclude=/mnt/ --exclude=/tmp/ --exclude=/ram/ --exclude=/etc/aliases.db --exclude=/var/log.default/wtmp --exclude=/var/run.default/utmp --exclude=/usr/java/jdk1.5.0_16/demo/ --exclude=/usr/java/jdk1.5.0_16/sample/ --exclude=/dev/ --exclude=/etc/blkid.tab --exclude=/etc/lvm/cache/ --exclude=/etc/lvm/archive/ --exclude=/etc/lvm/backup --delete $root/ /
   rsync -avxHS $RSYNC_PW_OPT $dryrun --delete $root/lib/modules/`uname -r`/ /lib/modules/`uname -r`/

   # update oneSIS links if they've changed
   if [ $debug -eq 1 ]; then
      echo dry-run update-node:
      update-node -d
   else
      update-node -r
   fi

elif [ -n "$cmdline2" -a -z "$cmdline1" ]; then
   cmdline=`cat /proc/cmdline |grep ' lustreroot='`
   if [ -z "$cmdline" ]; then
      echo halfroot but no lustreroot. confused. bye bye
      exit 1
   fi

   # check that root is on a ramdisk, otherwise we might be trying to run this on the wrong node
   # expect 2 lines eg.
   #   rootfs on / type rootfs (rw)
   #   none on / type ramfs (rw)
   if [ `mount | grep ' / ' | wc -l` -ne 2 ]; then
      echo root is not on a ramdisk. bye bye
      exit 1
   fi
   if [ `mount | grep ' / ' | cut -d' ' -f1 | head -1` != "rootfs" ]; then
      echo root is not on a ramdisk - rootfs. bye bye
      exit 1
   fi
   if [ `mount | grep ' / ' | cut -d' ' -f1 | tail -1` != "none" -o `mount | grep ' / ' | cut -d' ' -f5 | tail -1` != "ramfs" ]; then
      echo root is not on a ramdisk - ramfs. bye bye
      exit 1
   fi

   # we expect the 3rd mount entry to point at the rootfs on lustre. eg.
   #   10.8.30.244@o2ib:10.8.30.245@o2ib:/system on /whenPenguins/ruledTheEarth type lustre (ro)              # compute
   #   10.8.30.244@o2ib:10.8.30.245@o2ib:/system on /whenPenguins/ruledTheEarth type lustre (ro,localflock)   # front-end
   if [ "`mount | head -3 | tail -1 | cut -d' ' -f5-`" = "lustre (ro)" -o "`mount | head -3 | tail -1 | cut -d' ' -f5-`" = "lustre (ro,localflock)" -o "`mount | head -3 | tail -1 | cut -d' ' -f5-`" = "lustre (ro,relatime,localflock)" -o "`mount | head -3 | tail -1 | cut -d' ' -f5-`" = "lustre (ro,nosuid,relatime,localflock)" ]; then
      continue
   else
      echo 3rd mount is not the read-only root on lustre
      exit 1
   fi
   root=`mount | head -3 | tail -1 | cut -d' ' -f3`

   # check that eg. /usr is linked to this
   if [ ! -h /usr ]; then
      echo /usr is not a symlink
      exit 1
   fi
   if [ `ls -l /usr | awk '{print $11}'` != "$root/usr" ]; then
      echo root $root and link for /usr do not match
      exit 1
   fi

   # rsync over any changes
   for d in lib64 bin sbin; do
      echo $d
      # a couple more checks...
      if [ ! -d /$d ]; then
         echo $d is not a real directory
         exit 1
      fi
      df1=`df /$d | tail -1 | awk '{print $1}'`
      df6=`df /$d | tail -1 | awk '{print $6}'`
      if [ $df1 != "none" -o $df6 != "/" ]; then
         echo $d not on ramfs
      fi

      # go..
      rsync -avxHS $dryrun --delete $root/$d/ /$d/
   done

else
   echo no rsyncroot or halfroot in /proc/cmdline - exiting
   exit 1
fi