Page MenuHomeClusterLabs Projects
Files F4511780

svclib_nfslock
No OneTemporary
Actions

svclib_nfslock
View Options

#!/bin/bash
# Do reclaim-broadcasts when we kill lockd during shutdown/startup
# of a cluster service.
#
# Exported functions:
#
# notify_list_store
# notify_list_merge
# notify_list_broadcast
#
#
# Usage:
# statd_notify <directory> <hostname|ip>
#
# Copy out a list from <directory>, merge them with the system nfs lock
# list, and send them out as <hostname|ip> after generating a random
# state (needed so clients will reclaim their locks)
#
nfslock_statd_notify()
{
declare tmpdir=$(mktemp -d /tmp/statd-$2.XXXXXX)
declare nl_dir=$1
declare nl_ip=$2
declare command # Work around bugs in rpc.statd
declare pid_xxx # Work around bugs in rpc.statd
declare owner
[ -z "$lockd_pid" ] && return 0
if ! [ -d $nl_dir ]; then
return 0
fi
if [ -z "`ls $nl_dir/sm/* 2> /dev/null`" ]; then
ocf_log debug "No hosts to notify"
return 0
fi
# Ok, copy the HA directory to something we can use.
mkdir -p $tmpdir/sm
# Copy in our specified entries
cp -f $nl_dir/sm/* $tmpdir/sm
# Copy in our global entries
# XXX This might be what we just copied.
if [ -d "/var/lib/nfs/statd/sm" ]; then
owner=$(ls -dl /var/lib/nfs/statd/sm | awk '{print $3"."$4}')
cp -f /var/lib/nfs/statd/sm/* $tmpdir/sm
elif [ -d "/var/lib/nfs/sm" ]; then
owner=$(ls -dl /var/lib/nfs/statd/sm | awk '{print $3"."$4}')
cp -f /var/lib/nfs/sm/* $tmpdir/sm
fi
#
# Generate a random state file. If this ends up being what a client
# already has in its list, that's bad, but the chances of this
# are small - and relocations should be rare.
#
dd if=/dev/urandom of=$tmpdir/state bs=1 count=4 &> /dev/null
#
# Make sure we set permissions, or statd will not like it.
#
chown -R $owner $tmpdir
#
# Tell rpc.statd to notify clients. Don't go into background,
# because statd is buggy and won't exit like it's supposed to after
# sending the notifications out.
#
ocf_log info "Sending reclaim notifications via $nl_ip"
command="rpc.statd -NFP $tmpdir -n $nl_ip"
eval $command 2>&1 &
sleep 3 # XXX - the instance of rpc.statd we just spawned is supposed
# to exit after it finishes notifying clients.
# rpc.statd spawned which is still running handles the actual
# new SM_MON requests... we hope 3 seconds is enough time
# to get all the SM_NOTIFY messages out. rpc.statd = bugged
#
# clean up
#
pid_xxx=`ps auwwx | grep "$command" | grep -v grep | awk '{print $2}'`
kill $pid_xxx
rm -rf $tmpdir
return 0
}
#
# Copy of isSlave from svclib_ip and/or ip.sh
#
nfslock_isSlave()
{
declare intf=$1
declare line
if [ -z "$intf" ]; then
ocf_log err "usage: isSlave <I/F>"
return 1
fi
line=$(/sbin/ip link list dev $intf)
if [ $? -ne 0 ]; then
ocf_log err "$intf not found"
return 1
fi
if [ "$line" = "${line/<*SLAVE*>/}" ]; then
return 2
fi
# Yes, it is a slave device. Ignore.
return 0
}
#
# Get all the IPs on the system except loopback IPs
#
nfslock_ip_address_list()
{
declare idx dev family ifaddr
while read idx dev family ifaddr; do
if [ "$family" != "inet" ] && [ "$family" != "inet6" ]; then
continue
fi
if [ "$dev" = "lo" ]; then
# Ignore loopback
continue
fi
nfslock_isSlave $dev
if [ $? -ne 2 ]; then
continue
fi
idx=${idx/:/}
echo $dev $family ${ifaddr/\/*/} ${ifaddr/*\//}
done < <(/sbin/ip -o addr list | awk '{print $1,$2,$3,$4}')
return 0
}
#
# Usage: broadcast_notify <state_directory>
#
# Send the contents of <state_directory> out via all IPs on the system.
#
notify_list_broadcast()
{
declare dev family addr maskbits ip_name
declare lockd_pid=$(pidof lockd)
declare nl_dir=$1
# First of all, send lockd a SIGKILL. We hope nfsd is running.
# If it is, this will cause lockd to reset the grace period for
# lock reclaiming.
if [ -n "$lockd_pid" ]; then
ocf_log info "Asking lockd to drop locks (pid $lockd_pid)"
kill -9 $lockd_pid
else
ocf_log warning "lockd not running; cannot notify clients"
return 1
fi
while read dev family addr maskbits; do
if [ "$family" != "inet" ]; then
continue
fi
ip_name=$(clufindhostname -i $addr)
if [ -z "$ip_name" ]; then
nfslock_statd_notify $nl_dir $addr
else
nfslock_statd_notify $nl_dir $ip_name
fi
done < <(nfslock_ip_address_list)
}
#
# Store the lock monitor list from rpc.statd - do this during a teardown
# after the IP addresses of a service have been taken offline. Note that
# this should be done by HA-callout programs, but this feature is not in
# RHEL3.
#
notify_list_store()
{
declare nl_dir=$1
declare owner
mkdir -p $nl_dir/sm
if [ -d "/var/lib/nfs/statd/sm" ]; then
if [ -z "`ls /var/lib/nfs/statd/sm/* 2> /dev/null`" ]; then
return 1
# nothing to do!
fi
owner=$(ls -dl /var/lib/nfs/statd/sm | awk '{print $3"."$4}')
cp -af /var/lib/nfs/statd/sm/* $nl_dir/sm
chown -R $owner $nl_dir
return 0
elif [ -d "/var/lib/nfs/sm" ]; then
if [ -z "`ls /var/lib/nfs/sm/* 2> /dev/null`" ]; then
return 1
# nothing to do!
fi
owner=$(ls -dl /var/lib/nfs/sm | awk '{print $3"."$4}')
cp -af /var/lib/nfs/sm/* $nl_dir/sm
chown -R $owner $nl_dir
return 0
fi
return 1
}
#
# Merge the contents of <nl_dir>/sm with the system-wide list
# Make sure ownership is right, or statd will hiccup. This should not
# actually ever be needed because statd will, upon getting a SM_MON
# request, create all the entries in this list. It's mostly for
# housekeeping for next time we relocate the service.
#
notify_list_merge()
{
declare nl_dir=$1
declare owner
if [ -z "`ls $nl_dir/* 2> /dev/null`" ]; then
return 1
fi
if [ -d "/var/lib/nfs/statd/sm" ]; then
owner=$(ls -dl /var/lib/nfs/statd/sm | awk '{print $3"."$4}')
cp -af $nl_dir/sm/* /var/lib/nfs/statd/sm
chown -R $owner $nl_dir
return 0
elif [ -d "/var/lib/nfs/sm" ]; then
owner=$(ls -dl /var/lib/nfs/sm | awk '{print $3"."$4}')
cp -af $nl_dir/sm/* /var/lib/nfs/sm
chown -R $owner $nl_dir
return 0
fi
return 1
}

File Metadata

Mime Type
text/x-shellscript
Expires
Wed, Jun 25, 2:24 AM (8 h, 25 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1951876
Default Alt Text
svclib_nfslock (6 KB)

Event Timeline