set -e
ONLY=${ONLY:-"$*"}
+
# bug number for skipped test:
ALWAYS_EXCEPT="$SANITY_LNET_EXCEPT "
-[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+# skip the grant tests for ARM until they are fixed
+if [[ $(uname -m) = aarch64 ]]; then
+ # bug number: LU-14067
+ ALWAYS_EXCEPT+=" 300"
+fi
+
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
+
LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
. $LUSTRE/tests/test-framework.sh
build_test_filter
-export LNETCTL=${LNETCTL:-"$LUSTRE/../lnet/utils/lnetctl"}
-[ ! -f "$LNETCTL" ] &&
- export LNETCTL=$(which lnetctl 2> /dev/null)
[[ -z $LNETCTL ]] && skip "Need lnetctl"
restore_mounts=false
cleanup_lnet() {
echo "Cleaning up LNet"
- $LNETCTL lnet unconfigure 2>/dev/null
+ lsmod | grep -q lnet &&
+ $LNETCTL lnet unconfigure 2>/dev/null
unload_modules
}
cleanup_testsuite() {
trap "" EXIT
- rm -f $TMP/sanity-dlc*
+ # Cleanup any tmp files created by the sub tests
+ rm -f $TMP/sanity-lnet*
cleanup_netns
cleanup_lnet
if $restore_mounts; then
# variable to remote nodes
unset MODOPTS_LIBCFS
- set_default_debug
+ set_default_debug "neterror net nettrace malloc"
load_module ../lnet/lnet/lnet "$@"
LNDPATH=${LNDPATH:-"../lnet/klnds"}
ip netns exec $TESTNS "$@"
}
+setup_fakeif() {
+ local netns="$1"
+
+ local netns_arg=""
+ [[ -n $netns ]] &&
+ netns_arg="netns $netns"
+
+ ip link add 'test1pl' type veth peer name $FAKE_IF $netns_arg
+ ip link set 'test1pl' up
+ if [[ -n $netns ]]; then
+ do_ns ip addr add "${FAKE_IP}/31" dev $FAKE_IF
+ do_ns ip link set $FAKE_IF up
+ else
+ ip addr add "${FAKE_IP}/31" dev $FAKE_IF
+ ip link set $FAKE_IF up
+ fi
+}
+
+cleanup_fakeif() {
+ ip link show test1pl >& /dev/null && ip link del test1pl || return 0
+}
+
setup_netns() {
cleanup_netns
ip netns add $TESTNS
- ip link add 'test1pl' type veth peer name $FAKE_IF netns $TESTNS
- ip link set 'test1pl' up
- do_ns ip addr add "${FAKE_IP}/31" dev $FAKE_IF
- do_ns ip link set $FAKE_IF up
+ setup_fakeif $TESTNS
}
cleanup_netns() {
(ip netns list | grep -q $TESTNS) && ip netns del $TESTNS
- if ip link show test1pl >/dev/null 2>&1; then
- ip link del test1pl
- fi
+ cleanup_fakeif
}
configure_dlc() {
- nid: 25@gni
EOF
append_global_yaml
- echo"Add peer with nidrange (gni)"
+ echo "Add peer with nidrange (gni)"
compare_peer_add "21@gni" "[22-25]@gni" || error
echo "Add peer with nidrange that overlaps primary nid (gni)"
compare_peer_add "21@gni" "[21-25]@gni"
do_lnetctl peer del --prim_nid 1.1.1.1@o2ib &&
error "Command should have failed"
- echo "Don't provide mandatory arguments peer del"
+ echo "Don't provide mandatory argument for peer del"
do_lnetctl peer del --nid 1.1.1.1@tcp &&
error "Command should have failed"
+ echo "Don't provide mandatory argument for peer add"
+ do_lnetctl peer add --nid 1.1.1.1@tcp &&
+ error "Command should have failed"
+
echo "Don't provide mandatory arguments peer add"
do_lnetctl peer add &&
error "Command should have failed"
local nidstr
for nidstr in ${invalid_strs}; do
echo "Check invalid nidstring - '$nidstr'"
- do_lnetctl peer add --nid $nidstr &&
+ do_lnetctl peer add --prim_nid 1.1.1.1@tcp --nid $nidstr &&
error "Command should have failed"
done
local net="$1"
local if="$2"
- reinit_dlc || return $?
- load_module ../lnet/klnds/socklnd/ksocklnd ||
- error "Can't load ksocklnd.ko"
+ if ! lsmod | grep -q ksocklnd ; then
+ load_module ../lnet/klnds/socklnd/ksocklnd ||
+ error "Can't load ksocklnd.ko"
+ fi
+
do_lnetctl net add --net ${net} --if ${if} ||
error "Failed to add net ${net} on if ${if}"
}
test_100() {
have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
add_net "tcp" "eth0"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
peer_credits: 8
peer_buffer_credits: 0
credits: 256
+ lnd tunables:
+ conns_per_peer: 1
route:
- net: tcp7
gateway: 7.7.7.7@tcp
health_sensitivity: 1
peer:
- primary nid: 7.7.7.7@tcp
- Multi-Rail: True
+ Multi-Rail: False
peer ni:
- nid: 7.7.7.7@tcp
EOF
test_101() {
have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
add_net "tcp" "eth0"
cat <<EOF > $TMP/sanity-lnet-$testnum-expected.yaml
net:
peer_credits: 8
peer_buffer_credits: 0
credits: 256
+ lnd tunables:
+ conns_per_peer: 1
route:
- net: tcp8
gateway: 8.8.8.10@tcp
health_sensitivity: 1
peer:
- primary nid: 8.8.8.9@tcp
- Multi-Rail: True
+ Multi-Rail: False
peer ni:
- nid: 8.8.8.9@tcp
- primary nid: 8.8.8.10@tcp
- Multi-Rail: True
+ Multi-Rail: False
peer ni:
- nid: 8.8.8.10@tcp
- primary nid: 8.8.8.8@tcp
- Multi-Rail: True
+ Multi-Rail: False
peer ni:
- nid: 8.8.8.8@tcp
EOF
test_102() {
have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
add_net "tcp" "eth0"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl route add --net tcp102 --gateway 102.102.102.102@tcp ||
test_103() {
have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
add_net "tcp" "eth0"
$LNETCTL export --backup > $TMP/sanity-lnet-$testnum-expected.yaml
do_lnetctl route add --net tcp103 \
}
run_test 103 "Delete route with multiple gw (tcp)"
+test_104() {
+ local tyaml="$TMP/sanity-lnet-$testnum-expected.yaml"
+
+ reinit_dlc || return $?
+
+ # Default value is '3'
+ local val=$($LNETCTL global show | awk '/response_tracking/{print $NF}')
+ [[ $val -ne 3 ]] &&
+ error "Expect 3 found $val"
+
+ echo "Set < 0; Should fail"
+ do_lnetctl set response_tracking -1 &&
+ error "should have failed $?"
+
+ reinit_dlc || return $?
+ cat <<EOF > $tyaml
+global:
+ response_tracking: -10
+EOF
+ do_lnetctl import < $tyaml &&
+ error "should have failed $?"
+
+ echo "Check valid values; Should succeed"
+ local i
+ for ((i = 0; i < 4; i++)); do
+ reinit_dlc || return $?
+ do_lnetctl set response_tracking $i ||
+ error "should have succeeded $?"
+ $LNETCTL global show | grep -q "response_tracking: $i" ||
+ error "Failed to set response_tracking to $i"
+ reinit_dlc || return $?
+ cat <<EOF > $tyaml
+global:
+ response_tracking: $i
+EOF
+ do_lnetctl import < $tyaml ||
+ error "should have succeeded $?"
+ $LNETCTL global show | grep -q "response_tracking: $i" ||
+ error "Failed to set response_tracking to $i"
+ done
+
+ reinit_dlc || return $?
+ echo "Set > 3; Should fail"
+ do_lnetctl set response_tracking 4 &&
+ error "should have failed $?"
+
+ reinit_dlc || return $?
+ cat <<EOF > $tyaml
+global:
+ response_tracking: 10
+EOF
+ do_lnetctl import < $tyaml &&
+ error "should have failed $?"
+ return 0
+}
+run_test 104 "Set/check response_tracking param"
+
### load lnet in default namespace, configure in target namespace
test_200() {
}
run_test 203 "add a network using an interface in the non-default namespace"
+LNET_PARAMS_FILE="$TMP/$TESTSUITE.parameters"
+function save_lnet_params() {
+ $LNETCTL global show | egrep -v '^global:$' |
+ sed 's/://' > $LNET_PARAMS_FILE
+}
+
+function restore_lnet_params() {
+ local param value
+ while read param value; do
+ [[ $param == max_intf ]] && continue
+ [[ $param == lnd_timeout ]] && continue
+ $LNETCTL set ${param} ${value} ||
+ error "Failed to restore ${param} to ${value}"
+ done < $LNET_PARAMS_FILE
+}
+
+function lnet_health_pre() {
+ save_lnet_params
+
+ # Lower transaction timeout to speed up test execution
+ $LNETCTL set transaction_timeout 10 ||
+ error "Failed to set transaction_timeout $?"
+
+ # Increase recovery interval so we have time to capture health values
+ $LNETCTL set recovery_interval 20 ||
+ error "Failed to set recovery_interval $?"
+
+ RETRY_PARAM=$($LNETCTL global show | awk '/retry_count/{print $NF}')
+ RSND_PRE=$($LNETCTL stats show | awk '/resend_count/{print $NF}')
+ LO_HVAL_PRE=$($LNETCTL net show -v 2 | awk '/health value/{print $NF}' |
+ xargs echo | sed 's/ /+/g' | bc -l)
+
+ local my_nid=$($LCTL list_nids | head -n 1)
+
+ RMT_HVAL_PRE=$($LNETCTL peer show --nid $my_nid -v 2 2>/dev/null |
+ awk '/health value/{print $NF}' | xargs echo |
+ sed 's/ /+/g' | bc -l)
+
+ # Might not have any peers so initialize to zero.
+ RMT_HVAL_PRE=${RMT_HVAL_PRE:-0}
+
+ return 0
+}
+
+function lnet_health_post() {
+ RSND_POST=$($LNETCTL stats show | awk '/resend_count/{print $NF}')
+ LO_HVAL_POST=$($LNETCTL net show -v 2 |
+ awk '/health value/{print $NF}' |
+ xargs echo | sed 's/ /+/g' | bc -l)
+
+ local my_nid=$($LCTL list_nids | head -n 1)
+
+ RMT_HVAL_POST=$($LNETCTL peer show --nid $my_nid -v 2 2>/dev/null |
+ awk '/health value/{print $NF}' | xargs echo |
+ sed 's/ /+/g' | bc -l)
+
+ # Might not have any peers so initialize to zero.
+ RMT_HVAL_POST=${RMT_HVAL_POST:-0}
+
+ ${VERBOSE} &&
+ echo "Pre resends: $RSND_PRE" &&
+ echo "Post resends: $RSND_POST" &&
+ echo "Resends delta: $((RSND_POST - RSND_PRE))" &&
+ echo "Pre local health: $LO_HVAL_PRE" &&
+ echo "Post local health: $LO_HVAL_POST" &&
+ echo "Pre remote health: $RMT_HVAL_PRE" &&
+ echo "Post remote health: $RMT_HVAL_POST"
+
+ restore_lnet_params
+
+ return 0
+}
+
+function check_no_resends() {
+ echo "Check that no resends took place"
+ [[ $RSND_POST -ne $RSND_PRE ]] &&
+ error "Found resends: $RSND_POST != $RSND_PRE"
+
+ return 0
+}
+
+function check_resends() {
+ local delta=$((RSND_POST - RSND_PRE))
+
+ echo "Check that $RETRY_PARAM resends took place"
+ [[ $delta -ne $RETRY_PARAM ]] &&
+ error "Expected $RETRY_PARAM resends found $delta"
+
+ return 0
+}
+
+function check_no_local_health() {
+ echo "Check that local NI health is unchanged"
+ [[ $LO_HVAL_POST -ne $LO_HVAL_PRE ]] &&
+ error "Local health changed: $LO_HVAL_POST != $LO_HVAL_PRE"
+
+ return 0
+}
+
+function check_local_health() {
+ echo "Check that local NI health has been changed"
+ [[ $LO_HVAL_POST -eq $LO_HVAL_PRE ]] &&
+ error "Local health unchanged: $LO_HVAL_POST == $LO_HVAL_PRE"
+
+ return 0
+}
+
+function check_no_remote_health() {
+ echo "Check that remote NI health is unchanged"
+ [[ $RMT_HVAL_POST -ne $RMT_HVAL_PRE ]] &&
+ error "Remote health changed: $RMT_HVAL_POST != $RMT_HVAL_PRE"
+
+ return 0
+}
+
+function check_remote_health() {
+ echo "Check that remote NI health has been changed"
+ [[ $RMT_HVAL_POST -eq $RMT_HVAL_PRE ]] &&
+ error "Remote health unchanged: $RMT_HVAL_POST == $RMT_HVAL_PRE"
+
+ return 0
+}
+
+# See lnet/lnet/lib-msg.c:lnet_health_check()
+LNET_LOCAL_RESEND_STATUSES="local_interrupt local_dropped local_aborted"
+LNET_LOCAL_RESEND_STATUSES+=" local_no_route local_timeout"
+LNET_LOCAL_NO_RESEND_STATUSES="local_error"
+test_204() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+
+ lnet_health_pre || return $?
+
+ local hstatus
+ for hstatus in ${LNET_LOCAL_RESEND_STATUSES} \
+ ${LNET_LOCAL_NO_RESEND_STATUSES}; do
+ echo "Simulate $hstatus"
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+ done
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_no_local_health || return $?
+
+ return 0
+}
+run_test 204 "Check no health or resends for single-rail local failures"
+
+test_205() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ local hstatus
+ for hstatus in ${LNET_LOCAL_RESEND_STATUSES}; do
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ echo "Simulate $hstatus"
+ lnet_health_pre
+
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_resends || return $?
+ check_local_health || return $?
+ done
+
+ for hstatus in ${LNET_LOCAL_NO_RESEND_STATUSES}; do
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ echo "Simulate $hstatus"
+ lnet_health_pre || return $?
+
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_local_health || return $?
+ done
+
+ return 0
+}
+run_test 205 "Check health and resends for multi-rail local failures"
+
+# See lnet/lnet/lib-msg.c:lnet_health_check()
+LNET_REMOTE_RESEND_STATUSES="remote_dropped"
+LNET_REMOTE_NO_RESEND_STATUSES="remote_error remote_timeout"
+test_206() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+
+ do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+ error "failed to discover myself"
+
+ lnet_health_pre || return $?
+
+ local hstatus
+ for hstatus in ${LNET_REMOTE_RESEND_STATUSES} \
+ ${LNET_REMOTE_NO_RESEND_STATUSES}; do
+ echo "Simulate $hstatus"
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+ done
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_no_local_health || return $?
+ check_no_remote_health || return $?
+
+ return 0
+}
+run_test 206 "Check no health or resends for single-rail remote failures"
+
+test_207() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ local hstatus
+ for hstatus in ${LNET_REMOTE_RESEND_STATUSES}; do
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+ error "failed to discover myself"
+
+ echo "Simulate $hstatus"
+ lnet_health_pre || return $?
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_resends || return $?
+ check_no_local_health || return $?
+ check_remote_health || return $?
+ done
+ for hstatus in ${LNET_REMOTE_NO_RESEND_STATUSES}; do
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+ error "failed to discover myself"
+
+ echo "Simulate $hstatus"
+ lnet_health_pre || return $?
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e ${hstatus}
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e ${hstatus}
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_no_local_health || return $?
+ check_remote_health || return $?
+ done
+
+ return 0
+}
+run_test 207 "Check health and resends for multi-rail remote errors"
+
+test_208_load_and_check_lnet() {
+ local ip2nets="$1"
+ local p_nid="$2"
+ local s_nid="$3"
+ local num_expected=1
+
+ load_lnet "networks=\"\" ip2nets=\"${ip2nets_str}\""
+
+ $LCTL net up ||
+ error "Failed to load LNet with ip2nets \"${ip2nets_str}\""
+
+ [[ -n $s_nid ]] &&
+ num_expected=2
+
+ declare -a nids
+ nids=( $($LCTL list_nids) )
+
+ [[ ${#nids[@]} -ne ${num_expected} ]] &&
+ error "Expect ${num_expected} NIDs found ${#nids[@]}"
+
+ [[ ${nids[0]} == ${p_nid} ]] ||
+ error "Expect NID \"${p_nid}\" found \"${nids[0]}\""
+
+ [[ -n $s_nid ]] && [[ ${nids[1]} != ${s_nid} ]] &&
+ error "Expect second NID \"${s_nid}\" found \"${nids[1]}\""
+
+ $LCTL net down &>/dev/null
+ cleanup_lnet
+}
+
+test_208() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ cleanup_netns || error "Failed to cleanup netns before test execution"
+ cleanup_lnet || error "Failed to unload modules before test execution"
+ setup_fakeif || error "Failed to add fake IF"
+
+ have_interface "$FAKE_IF" ||
+ error "Expect $FAKE_IF configured but not found"
+
+ local eth0_ip=$(ip --oneline addr show dev eth0 |
+ awk '/inet /{print $4}' |
+ sed 's:/.*::')
+ local ip2nets_str="tcp(eth0) $eth0_ip"
+
+ echo "Configure single NID \"$ip2nets_str\""
+ test_208_load_and_check_lnet "${ip2nets_str}" "${eth0_ip}@tcp"
+
+ ip2nets_str="tcp(eth0) $eth0_ip; tcp1($FAKE_IF) $FAKE_IP"
+ echo "Configure two NIDs; two NETs \"$ip2nets_str\""
+ test_208_load_and_check_lnet "${ip2nets_str}" "${eth0_ip}@tcp" \
+ "${FAKE_IP}@tcp1"
+
+ ip2nets_str="tcp(eth0) $eth0_ip; tcp($FAKE_IF) $FAKE_IP"
+ echo "Configure two NIDs; one NET \"$ip2nets_str\""
+ test_208_load_and_check_lnet "${ip2nets_str}" "${eth0_ip}@tcp" \
+ "${FAKE_IP}@tcp"
+ local addr1=( ${eth0_ip//./ } )
+ local addr2=( ${FAKE_IP//./ } )
+ local range="[${addr1[0]},${addr2[0]}]"
+
+ local i
+ for i in $(seq 1 3); do
+ range+=".[${addr1[$i]},${addr2[$i]}]"
+ done
+ ip2nets_str="tcp(eth0,${FAKE_IF}) ${range}"
+
+ echo "Configured two NIDs; one NET alt syntax \"$ip2nets_str\""
+ test_208_load_and_check_lnet "${ip2nets_str}" "${eth0_ip}@tcp" \
+ "${FAKE_IP}@tcp"
+
+ cleanup_fakeif
+
+ echo "alt syntax with missing IF \"$ip2nets_str\""
+ load_lnet "networks=\"\" ip2nets=\"${ip2nets_str}\""
+
+ echo "$LCTL net up should fail"
+ $LCTL net up &&
+ error "LNet bringup should have failed"
+
+ cleanup_lnet
+}
+run_test 208 "Test various kernel ip2nets configurations"
+
+test_209() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+
+ do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+ error "failed to discover myself"
+
+ echo "Simulate network_timeout w/SR config"
+ lnet_health_pre
+
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e network_timeout
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_no_local_health || return $?
+ check_no_remote_health || return $?
+
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ do_lnetctl discover $($LCTL list_nids | head -n 1) ||
+ error "failed to discover myself"
+
+ echo "Simulate network_timeout w/MR config"
+ lnet_health_pre
+
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e network_timeout
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e network_timeout
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Should have failed"
+ $LCTL net_drop_del -a
+
+ lnet_health_post
+
+ check_no_resends || return $?
+ check_local_health || return $?
+ check_remote_health || return $?
+
+ return 0
+}
+run_test 209 "Check health, but not resends, for network timeout"
+
+check_nid_in_recovq() {
+ local recovq=$($LNETCTL debug recovery $1)
+ local expect="$2"
+ local nids=$($LCTL list_nids | xargs echo)
+ local found=false
+ local nid=""
+
+ echo "Check recovery queue"
+ echo "$recovq"
+ if [[ $(grep -c 'nid-'<<<$recovq) -ne $expect ]]; then
+ error "Expect $expect NIDs found: \"$recovq\""
+ fi
+
+ [[ $expect -eq 0 ]] && return 0
+
+ for nid in ${nids}; do
+ grep -q "nid-0: $nid"<<<$recovq &&
+ found=true
+ done
+
+ if ! $found; then
+ error "Didn't find local NIDs in recovery queue: \"$recovq\""
+ fi
+
+ return 0
+}
+
+# First enqueue happens at time 0.
+# 2nd at 0 + 2^0 = 1
+# 3rd at 1 + 2^1 = 3
+# 4th at 3 + 2^2 = 7
+# 5th at 7 + 2^3 = 15
+# e.g. after 10 seconds we would expect to have seen the 4th enqueue,
+# (3 pings sent, 4th about to happen) and the 5th enqueue is yet to
+# happen
+# If the recovery limit is 10 seconds, then when the 5th enqueue happens
+# we expect the peer NI to have aged out, so it will not actually be
+# queued.
+check_ping_count() {
+ local queue="$1"
+ local expect="$2"
+
+ echo "Check ping counts:"
+ local ping_count
+ if [[ $queue == "ni" ]]; then
+ $LNETCTL net show -v 2 | egrep 'nid|health value|ping'
+ ping_count=( $($LNETCTL net show -v 2 |
+ awk '/ping_count/{print $NF}') )
+ elif [[ $queue == "peer_ni" ]]; then
+ $LNETCTL peer show -v 2 | egrep 'nid|health value|ping'
+ ping_count=( $($LNETCTL peer show -v 2 |
+ awk '/ping_count/{print $NF}') )
+ else
+ error "Unrecognized queue \"$queue\""
+ return 1
+ fi
+
+ local count
+ local found=false
+ for count in ${ping_count[@]}; do
+ if [[ $count -eq $expect ]]; then
+ if [[ $expect -ne 0 ]] && $found ; then
+ error "Found more than one interface matching \"$expect\" ping count"
+ return 1
+ else
+ echo "Expect ping count \"$expect\" found \"$count\""
+ found=true;
+ fi
+ elif [[ $count -ne 0 ]]; then
+ error "Found interface with ping count \"$count\" but expect \"$expect\""
+ return 1
+ fi
+ done
+
+ return 0
+}
+
+test_210() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ local prim_nid=$($LCTL list_nids | head -n 1)
+
+ do_lnetctl discover $prim_nid ||
+ error "failed to discover myself"
+
+ # Set recovery limit to 10 seconds.
+ do_lnetctl set recovery_limit 10 ||
+ error "failed to set recovery_limit"
+
+ $LCTL set_param debug=+net
+ # Use local_error so LNet doesn't attempt to resend the discovery ping
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e local_error
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e local_error
+ do_lnetctl discover $($LCTL list_nids | head -n 1) &&
+ error "Expected discovery to fail"
+
+ sleep 5
+ check_nid_in_recovq "-l" 1
+ check_ping_count "ni" "2"
+
+ sleep 5
+
+ check_nid_in_recovq "-l" 1
+ check_ping_count "ni" "3"
+
+ $LCTL net_drop_del -a
+
+ return 0
+}
+run_test 210 "Local NI recovery checks"
+
+test_211() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp1" "eth0" || return $?
+
+ local prim_nid=$($LCTL list_nids | head -n 1)
+
+ do_lnetctl discover $prim_nid ||
+ error "failed to discover myself"
+
+ # Set recovery limit to 10 seconds.
+ do_lnetctl set recovery_limit 10 ||
+ error "failed to set recovery_limit"
+
+ $LCTL net_drop_add -s *@tcp -d *@tcp -m GET -r 1 -e remote_error
+ $LCTL net_drop_add -s *@tcp1 -d *@tcp1 -m GET -r 1 -e remote_error
+
+ # Set health to 0 on one interface. This forces it onto the recovery
+ # queue.
+ $LNETCTL peer set --nid $prim_nid --health 0
+
+ # After 5 seconds, we expect the peer NI to still be in recovery
+ sleep 5
+ check_nid_in_recovq "-p" 1
+ check_ping_count "peer_ni" "2"
+
+ # After 15 seconds, the peer NI should have been fully processed out of
+ # the recovery queue. We'll allow a total of 17 seconds to account for
+ # differences in sleeping for whole seconds vs. the more accurate time
+ # keeping that is done in the recovery code.
+ sleep 12
+ check_nid_in_recovq "-p" 0
+ check_ping_count "peer_ni" "4"
+
+ $LCTL net_drop_del -a
+
+ # Set health to force it back onto the recovery queue. Set to 500 means
+ # in 5 seconds it should be back at maximum value. We'll wait a couple
+ # more seconds than that to be safe.
+ # NB: we reset the recovery limit to 0 (indefinite) so the peer NI is
+ # eligible again
+ do_lnetctl set recovery_limit 0 ||
+ error "failed to set recovery_limit"
+
+ $LNETCTL peer set --nid $prim_nid --health 500
+
+ check_nid_in_recovq "-p" 1
+ check_ping_count "peer_ni" "2"
+
+ sleep 7
+
+ check_nid_in_recovq "-p" 0
+ check_ping_count "peer_ni" "0"
+
+ return 0
+}
+run_test 211 "Remote NI recovery checks"
+
+test_212() {
+ local rnodes=$(remote_nodes_list)
+ [[ -z $rnodes ]] && skip "Need at least 1 remote node"
+
+ cleanup_lnet || error "Failed to cleanup before test execution"
+
+ # Loading modules should configure LNet with the appropriate
+ # test-framework configuration
+ load_modules || error "Failed to load modules"
+
+ local my_nid=$($LCTL list_nids | head -n 1)
+ [[ -z $my_nid ]] &&
+ error "Failed to get primary NID for local host $HOSTNAME"
+
+ local rnode=$(awk '{print $1}' <<<$rnodes)
+ local rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
+ local rloaded=false
+
+ if [[ -z $rnodenids ]]; then
+ do_rpc_nodes $rnode load_modules_local
+ rloaded=true
+ rnodenids=$(do_node $rnode $LCTL list_nids | xargs echo)
+ fi
+
+ local rnodepnid=$(awk '{print $1}' <<< $rnodenids)
+
+ [[ -z $rnodepnid ]] &&
+ error "Failed to get primary NID for remote host $rnode"
+
+ log "Initial discovery"
+ do_lnetctl discover --force $rnodepnid ||
+ error "Failed to discover $rnodepnid"
+
+ do_node $rnode "$LNETCTL discover --force $my_nid" ||
+ error "$rnode failed to discover $my_nid"
+
+ log "Fail local discover ping to set LNET_PEER_REDISCOVER flag"
+ $LCTL net_drop_add -s "*@$NETTYPE" -d "*@$NETTYPE" -r 1 -e local_error
+ do_lnetctl discover --force $rnodepnid &&
+ error "Discovery should have failed"
+ $LCTL net_drop_del -a
+
+ local nid
+ for nid in $rnodenids; do
+ # We need GET (PING) delay just long enough so we can trigger
+ # discovery on the remote peer
+ $LCTL net_delay_add -s "*@$NETTYPE" -d $nid -r 1 -m GET -l 3
+ $LCTL net_drop_add -s "*@$NETTYPE" -d $nid -r 1 -m GET -e local_error
+ # We need PUT (PUSH) delay just long enough so we can process
+ # the PING failure
+ $LCTL net_delay_add -s "*@$NETTYPE" -d $nid -r 1 -m PUT -l 6
+ done
+
+ log "Force $HOSTNAME to discover $rnodepnid (in background)"
+ # We want to get a PING sent that we know will eventually fail.
+ # The delay rules we added will ensure the ping is not sent until
+ # the PUSH is also in flight (see below), and the drop rule ensures that
+ # when the PING is eventually sent it will error out
+ do_lnetctl discover --force $rnodepnid &
+ local pid1=$!
+
+ # We want a discovery PUSH from rnode to put rnode back on our
+ # discovery queue. This should cause us to try and send a PUSH to rnode
+ # while the PING is still outstanding.
+ log "Force $rnode to discover $my_nid"
+ do_node $rnode $LNETCTL discover --force $my_nid
+
+ # At this point we'll have both PING_SENT and PUSH_SENT set for the
+ # rnode peer. Wait for the PING to error out which should terminate the
+ # discovery process that we backgrounded.
+ log "Wait for $pid1"
+ wait $pid1
+ log "Finished wait on $pid1"
+
+ # The PING send failure clears the PING_SENT flag and puts the peer back
+ # on the discovery queue. When discovery thread processes the peer it
+ # will mistakenly clear the PUSH_SENT flag (and set PUSH_FAILED).
+ # Discovery will then complete for this peer even though we have an
+ # outstanding PUSH.
+ # When PUSH is actually unlinked it will be forced back onto the
+ # discovery queue, but we no longer have a ref on the peer. When
+ # discovery completes again, we'll trip the ASSERT in
+ # lnet_destroy_peer_locked()
+
+ # Delete the delay rules to send the PUSH
+ $LCTL net_delay_del -a
+ # Delete the drop rules
+ $LCTL net_drop_del -a
+
+ unload_modules ||
+ error "Failed to unload modules"
+ if $rloaded; then
+ do_rpc_nodes $rnode unload_modules_local ||
+ error "Failed to unload modules on $rnode"
+ fi
+
+ return 0
+}
+run_test 212 "Check discovery refcount loss bug (LU-14627)"
+
+test_213() {
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ cleanup_netns || error "Failed to cleanup netns before test execution"
+ cleanup_lnet || error "Failed to unload modules before test execution"
+
+ setup_fakeif || error "Failed to add fake IF"
+ have_interface "$FAKE_IF" ||
+ error "Expect $FAKE_IF configured but not found"
+
+ reinit_dlc || return $?
+
+ add_net "tcp" "eth0" || return $?
+ add_net "tcp" "$FAKE_IF" || return $?
+
+ local nid1=$(lctl list_nids | head -n 1)
+ local nid2=$(lctl list_nids | tail --lines 1)
+
+ [[ $(lctl which_nid $nid1 $nid2) == $nid1 ]] ||
+ error "Expect nid1 \"$nid1\" to be preferred"
+
+ [[ $(lctl which_nid $nid2 $nid1) == $nid2 ]] ||
+ error "Expect nid2 \"$nid2\" to be preferred"
+
+ return 0
+}
+run_test 213 "Check LNetDist calculation for multiple local NIDs"
+
+test_230() {
+ # LU-12815
+ have_interface "eth0" || skip "Need eth0 interface with ipv4 configured"
+
+ echo "Check valid values; Should succeed"
+ local i
+ local lnid
+ for ((i = 4; i < 16; i+=4)); do
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ do_lnetctl net set --all --conns-per-peer $i ||
+ error "should have succeeded $?"
+ $LNETCTL net show -v 1 | grep -q "conns_per_peer: $i" ||
+ error "failed to set conns-per-peer to $i"
+ lnid="$(lctl list_nids | head -n 1)"
+ do_lnetctl ping "$lnid" ||
+ error "failed to ping myself"
+ printf 'network tcp\nconn_list\n' | lctl | grep -c "$lnid" | grep -q $((2+i*2)) ||
+ error "expected number of tcp connections $((2+i*2))"
+ done
+
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ echo "Set > 127; Should fail"
+ do_lnetctl net set --all --conns-per-peer 128 &&
+ error "should have failed $?"
+
+ reinit_dlc || return $?
+ add_net "tcp" "eth0" || return $?
+ echo "Set < 0; Should be ignored"
+ do_lnetctl net set --all --conns-per-peer -1 ||
+ error "should have succeeded $?"
+ $LNETCTL net show -v 1 | grep -q "conns_per_peer: 1" ||
+ error "Did not stay at default"
+}
+run_test 230 "Test setting conns-per-peer"
+
+test_300() {
+ # LU-13274
+ local header
+ local out=$TMP/$tfile
+ local prefix=/usr/include/linux/lnet
+
+ # We use a hard coded prefix so that this test will not fail
+ # when run in tree.
+ CC=${CC:-cc}
+ if ! which $CC > /dev/null 2>&1; then
+ skip_env "$CC is not installed"
+ fi
+
+ cleanup_lnet || exit 1
+ load_lnet
+
+ local cc_args="-Wall -Werror -std=c99 -c -x c /dev/null -o $out"
+ if ! [[ -d $prefix ]]; then
+ # Assume we're running in tree and fixup the include path.
+ prefix=$LUSTRE/../lnet/include/uapi/linux/lnet
+ cc_args+=" -I $LUSTRE/../lnet/include/uapi"
+ fi
+
+ for header in $prefix/*.h; do
+ if ! [[ -f "$header" ]]; then
+ continue
+ fi
+
+ echo "$CC $cc_args -include $header"
+ $CC $cc_args -include $header ||
+ error "cannot compile '$header'"
+ done
+ rm -f $out
+}
+run_test 300 "packaged LNet UAPI headers can be compiled"
+
complete $SECONDS
cleanup_testsuite
git://git.whamcloud.com / fs / lustre-release.git / blobdiff