ctdb/config/statd-callout

   1 #!/bin/sh
   2
   3 # This must run as root as CTDB tool commands need to access CTDB socket
   4 [ "$(id -u)" -eq 0 ] || exec sudo "$0" "$@"
   5
   6 # this script needs to be installed so that statd points to it with the -H
   7 # command line argument. The easiest way to do that is to put something like this in
   8 # /etc/sysconfig/nfs:
   9 #   STATD_HOSTNAME="myhostname -H /etc/ctdb/statd-callout"
  10
  11 [ -n "$CTDB_BASE" ] || \
  12     CTDB_BASE=$(d=$(dirname "$0") ; cd -P "$d" ; echo "$PWD")
  13
  14 . "${CTDB_BASE}/functions"
  15
  16 # Overwrite this so we get some logging
  17 die ()
  18 {
  19     script_log "statd-callout" "$@"
  20     exit 1
  21 }
  22
  23 load_system_config "nfs"
  24
  25 [ -n "$NFS_HOSTNAME" ] || \
  26     die "NFS_HOSTNAME is not configured. statd-callout failed"
  27
  28 ############################################################
  29
  30 ctdb_setup_state_dir "service" "nfs"
  31
  32 # script_state_dir set by ctdb_setup_state_dir()
  33 # shellcheck disable=SC2154
  34 d="${script_state_dir}/statd-callout"
  35
  36 mkdir -p "$d" || die "Failed to create directory \"${d}\""
  37 cd "$d" || die "Failed to change directory to \"${d}\""
  38
  39 pnn=$(ctdb_get_pnn)
  40
  41 ############################################################
  42
  43 send_notifies ()
  44 {
  45         _smnotify="${CTDB_HELPER_BINDIR}/smnotify"
  46
  47         # State must monotonically increase, across the entire
  48         # cluster.  Use seconds since epoch and hope the time is in
  49         # sync across nodes.  Even numbers mean service is shut down,
  50         # odd numbers mean service is started.
  51
  52         # Intentionally round to an even number
  53         # shellcheck disable=SC2017
  54         _state_even=$(( $(date '+%s') / 2 * 2))
  55
  56         _prev=""
  57         while read _sip _cip ; do
  58                 # NOTE: Consider optimising smnotify to read all the
  59                 # data from stdin and then run it in the background.
  60
  61                 # Reset stateval for each serverip
  62                 if [ "$_sip" != "$_prev" ] ; then
  63                         _stateval="$_state_even"
  64                 fi
  65
  66                 # Send notifies for server shutdown
  67                 "$_smnotify" --client="$_cip" --ip="$_sip" \
  68                              --server="$_sip" --stateval="$_stateval"
  69                 "$_smnotify" --client="$_cip" --ip="$_sip" \
  70                              --server="$NFS_HOSTNAME" --stateval="$_stateval"
  71
  72                 # Send notifies for server startup
  73                 _stateval=$((_stateval + 1))
  74                 "$_smnotify" --client="$_cip" --ip="$_sip" \
  75                              --server="$_sip" --stateval="$_stateval"
  76                 "$_smnotify" --client="$_cip" --ip="$_sip" \
  77                              --server="$NFS_HOSTNAME" --stateval="$_stateval"
  78         done
  79 }
  80
  81 delete_records ()
  82 {
  83         while read _sip _cip ; do
  84                 _key="statd-state@${_sip}@${_cip}"
  85                 echo "\"${_key}\" \"\""
  86         done | $CTDB ptrans "ctdb.tdb"
  87 }
  88
  89 ############################################################
  90
  91 case "$1" in
  92     # Keep a single file to keep track of the last "add-client" or
  93     # "del-client'.  These get pushed to ctdb.tdb during "update",
  94     # which will generally be run once each "monitor" cycle.  In this
  95     # way we avoid scalability problems with flood of persistent
  96     # transactions after a "notify" when all the clients re-take their
  97     # locks.
  98
  99     add-client)
 100         # statd does not tell us to which IP the client connected so
 101         # we must add it to all the IPs that we serve
 102         cip="$2"
 103         date=$(date '+%s')
 104         # x is intentionally ignored
 105         # shellcheck disable=SC2034
 106         $CTDB ip -X |
 107         tail -n +2 |
 108         while IFS="|" read x sip node x ; do
 109             [ "$node" = "$pnn" ] || continue # not us
 110             key="statd-state@${sip}@${cip}"
 111             echo "\"${key}\" \"${date}\"" >"$key"
 112         done
 113         ;;
 114
 115     del-client)
 116         # statd does not tell us from which IP the client disconnected
 117         # so we must add it to all the IPs that we serve
 118         cip="$2"
 119         # x is intentionally ignored
 120         # shellcheck disable=SC2034
 121         $CTDB ip -X |
 122         tail -n +2 |
 123         while IFS="|" read x sip node x ; do
 124             [ "$node" = "$pnn" ] || continue # not us
 125             key="statd-state@${sip}@${cip}"
 126             echo "\"${key}\" \"\"" >"$key"
 127         done
 128         ;;
 129
 130     update)
 131         files=$(echo statd-state@*)
 132         if [ "$files" = "statd-state@*" ] ; then
 133             # No files!
 134             exit 0
 135         fi
 136         # Filter out lines for any IP addresses that are not currently
 137         # hosted public IP addresses.
 138         ctdb_ips=$($CTDB ip | tail -n +2)
 139         sed_expr=$(echo "$ctdb_ips" |
 140             awk -v pnn="$pnn" 'pnn == $2 {
 141                 ip = $1; gsub(/\./, "\\.", ip);
 142                 printf "/statd-state@%s@/p\n", ip }')
 143         # Intentional multi-word expansion for multiple files
 144         # shellcheck disable=SC2086
 145         items=$(sed -n "$sed_expr" $files)
 146         if [ -n "$items" ] ; then
 147                 if echo "$items"  | $CTDB ptrans "ctdb.tdb" ; then
 148                         # shellcheck disable=SC2086
 149                         rm $files
 150                 fi
 151         fi
 152         ;;
 153
 154     notify)
 155         # we must restart the lockmanager (on all nodes) so that we get
 156         # a clusterwide grace period (so other clients don't take out
 157         # conflicting locks through other nodes before all locks have been
 158         # reclaimed)
 159
 160         # we need these settings to make sure that no tcp connections survive
 161         # across a very fast failover/failback
 162         #echo 10 > /proc/sys/net/ipv4/tcp_fin_timeout
 163         #echo 0 > /proc/sys/net/ipv4/tcp_max_tw_buckets
 164         #echo 0 > /proc/sys/net/ipv4/tcp_max_orphans
 165
 166         # Delete the notification list for statd, we don't want it to
 167         # ping any clients
 168         rm -f /var/lib/nfs/statd/sm/*
 169         rm -f /var/lib/nfs/statd/sm.bak/*
 170
 171         # We must also let some time pass between stopping and
 172         # restarting the lock manager.  Otherwise there is a window
 173         # where the lock manager will respond "strangely" immediately
 174         # after restarting it, which causes clients to fail to reclaim
 175         # their locks.
 176         nfs_callout_init
 177         "$CTDB_NFS_CALLOUT" "stop" "nlockmgr" >/dev/null 2>&1
 178         sleep 2
 179         "$CTDB_NFS_CALLOUT" "start" "nlockmgr" >/dev/null 2>&1
 180
 181         # we now need to send out additional statd notifications to ensure
 182         # that clients understand that the lockmanager has restarted.
 183         # we have three cases:
 184         # 1, clients that ignore the ip address the stat notification came from
 185         #    and ONLY care about the 'name' in the notify packet.
 186         #    these clients ONLY work with lock failover IFF that name
 187         #    can be resolved into an ipaddress that matches the one used
 188         #    to mount the share.  (==linux clients)
 189         #    This is handled when starting lockmanager above,  but those
 190         #    packets are sent from the "wrong" ip address, something linux
 191         #    clients are ok with, buth other clients will barf at.
 192         # 2, Some clients only accept statd packets IFF they come from the
 193         #    'correct' ip address.
 194         # 2a,Send out the notification using the 'correct' ip address and also
 195         #    specify the 'correct' hostname in the statd packet.
 196         #    Some clients require both the correct source address and also the
 197         #    correct name. (these clients also ONLY work if the ip addresses
 198         #    used to map the share can be resolved into the name returned in
 199         #    the notify packet.)
 200         # 2b,Other clients require that the source ip address of the notify
 201         #    packet matches the ip address used to take out the lock.
 202         #    I.e. that the correct source address is used.
 203         #    These clients also require that the statd notify packet contains
 204         #    the name as the ip address used when the lock was taken out.
 205         #
 206         # Both 2a and 2b are commonly used in lockmanagers since they maximize
 207         # probability that the client will accept the statd notify packet and
 208         # not just ignore it.
 209         # For all IPs we serve, collect info and push to the config database
 210
 211         # Construct a sed expression to take catdb output and produce pairs of:
 212         #   server-IP client-IP
 213         # but only for the server-IPs that are hosted on this node.
 214         ctdb_all_ips=$($CTDB ip all | tail -n +2)
 215         sed_expr=$(echo "$ctdb_all_ips" |
 216             awk -v pnn="$pnn" 'pnn == $2 {
 217                 ip = $1; gsub(/\./, "\\.", ip);
 218                 printf "s/^key.*=.*statd-state@\\(%s\\)@\\([^\"]*\\).*/\\1 \\2/p\n", ip }')
 219
 220         statd_state=$($CTDB catdb ctdb.tdb | sed -n "$sed_expr" | sort)
 221         [ -n "$statd_state" ] || exit 0
 222
 223         echo "$statd_state" | send_notifies
 224         echo "$statd_state" | delete_records
 225
 226         # Remove any stale touch files (i.e. for IPs not currently
 227         # hosted on this node and created since the last "update").
 228         # There's nothing else we can do with them at this stage.
 229         echo "$ctdb_all_ips" |
 230             awk -v pnn="$pnn" 'pnn != $2 { print $1 }' |
 231             while read sip ; do
 232                 rm -f "statd-state@${sip}@"*
 233             done
 234         ;;
 235 esac