Add lflush script

Signed-off-by: Jim Garlick <garlick@llnl.gov>

lustre/doc/Makefile.am

@@ -50,7 +50,7 @@ MANFILES = lustre.7 lfs.1 mount.lustre.8 mkfs.lustre.8 tunefs.lustre.8 lctl.8 \
 	llapi_quotactl.3 llobdstat.8 llstat.8 \
 	plot-llstat.8 l_getgroups.8 lst.8 routerstat.8 lshowmount.8 \
 	ll_recover_lost_found_objs.8 llog_reader.8 llapi_file_open.3 \
-	llapi_file_create.3 llapi_file_get_stripe.3 liblustreapi.7 lpurge.8
+	llapi_file_create.3 llapi_file_get_stripe.3 liblustreapi.7 lpurge.8 lflush.8
 if UTILS
 man_MANS = $(MANFILES)
 endif

lustre/doc/lflush.8

@@ -0,0 +1,43 @@
+.TH LFLUSH 8 Lustre LLNL LFLUSH
+.SH NAME
+lflush \- revoke all client locks
+.SH SYNOPSIS
+.B "lflush [-v]"
+.br
+.SH DESCRIPTION
+.B lflush
+synchronously revokes all Lustre MDC, OSC, and MGC locks on the local client,
+returning them to the lock manager for reuse.  This causes all Lustre
+page cache pages to be flushed (if dirty) or discarded (if clean).
+If Lustre servers are unavailable,
+.B lflush
+will block until recovery is complete.
+.LP
+.B lflush
+can be called from a batch resource manager epilog script to ensure that the
+next job has all available memory.  In the case of slurm, the node will
+remain in COMPLETING state while the epilog script executes.
+.LP
+A server will be skipped if its lock_count, lock_unused_count, and
+resource_count proc files all contain zeros.
+.LP
+.B lflush
+must be run as root.
+.SH OPTIONS
+.B lflush
+accepts the following options:
+.TP
+.I "-v"
+causes
+.B lflush
+to list each server before it flushes the servers locks.
+.SH FILES
+/proc/fs/lustre/ldlm/namespaces/<server>/lru_size
+.br
+/proc/fs/lustre/ldlm/namespaces/<server>/lock_count
+.br
+/proc/fs/lustre/ldlm/namespaces/<server>/lock_unused_count
+.br
+/proc/fs/lustre/ldlm/namespaces/<server>/resource_count
+.SH AUTHOR
+Herb Wartens and Jim Garlick

lustre/utils/Makefile.am

@@ -6,7 +6,7 @@ AM_LDFLAGS := -L$(top_builddir)/lnet/utils
 
 LIBPTLCTL := $(top_builddir)/lnet/utils/libptlctl.a
 
-sbin_scripts =
+sbin_scripts = lflush
 noinst_scripts = lrun
 bin_scripts = llstat llobdstat plot-llstat
 

lustre/utils/lflush

@@ -0,0 +1,84 @@
+#!/bin/bash
+#
+# Revoke all client locks, which should write all dirty Lustre
+# page cache pages and discard the clean ones.
+#
+# Based on flush-lustre-client-locks script by Herb Wartens.
+#
+
+declare -r nsdir=/proc/fs/lustre/ldlm/namespaces
+declare -r prog=lflush
+
+vopt=0
+retval=0
+
+die()
+{
+    echo "${prog}: $1" >&2
+    exit 1
+}
+
+warn()
+{
+    echo "${prog}: $1" >&2
+}
+
+lock_count()
+{
+    local server=$1
+    local nlock nulock nres
+
+    pushd ${nsdir} >/dev/null || die "could not chdir to ${nsdir}"
+        [ "${server}" = "*" ] && die "could not find any servers"
+        nlock=$(cat ${server}/lock_count 2>/dev/null)
+        nulock=$(cat ${server}/lock_unused_count 2>/dev/null)
+        nres=$(cat ${server}/resource_count 2>/dev/null)
+    popd >/dev/null
+
+    return $((${nlock:-0}+${nulock:-0}+${nres:-0}))
+}
+
+# Flush all locks for all servers
+flush_locks()
+{
+    local server count
+
+    pushd ${nsdir} >/dev/null || die "could not chdir to ${nsdir}"
+        for server in *; do
+            [ "${server}" = "*" ] && die "could not find any servers"
+            if ! lock_count ${server}; then
+                [ ${vopt} = 1 ] && warn "flushing ${server}"
+                if ! echo clear >${server}/lru_size; then
+                    warn "problem flushing ${server}"
+                    retval=1
+                fi
+            fi
+	done
+    popd >/dev/null
+}
+
+##
+## MAIN
+##
+
+[ $(id -u) = 0 ] || die "must run as root"
+
+if [ ! -d /proc/fs/lustre/ldlm/namespaces ]; then
+    warn "lustre client is not active - nothing to do"
+    exit 0
+fi
+
+while getopts "v" opt; do
+    case ${opt} in
+        v) vopt=1 ;;
+        *) die "Usage: ${prog} [-v]" ;;
+    esac
+done
+
+if [ -x /bin/sync ]; then
+    /bin/sync
+fi
+
+flush_locks
+
+exit ${retval}