LU-17760 lnet: Crash caused by uninitialized interface name

When adding an interface with ip2net, a duplicate configuration of an
already existing interface can cause a crash or misconfiguration of
lnet. Incoming interface names have to be checked if they are null and
furthermore duplicate interface configurations have to be removed.
When a duplicate is detected add has to be added to a list to be able
to shut it down otherwise shutdown would assert.
The problem can be repoduced on tcp and o2ib networks.
Steps that were used to reproduce the problem in the original
configuration, but it is reproducable in other variations and
in tcp networks.
modprobe lnet
lnetctl lnet configure
lnetctl net add --net  o2ib --if mlxib1
lnetctl net add --net  o2ib --if mlxib1
       --ip2net "o2ib 172.30.12.*"

Test-Parameters: trivial testlist=sanity-lnet
Signed-off-by: Frank Sehr <fsehr@whamcloud.com>
Change-Id: Ie76d97cc52855ab897a9e07a3697483189d4b19e
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/54859
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Reviewed-by: Cyril Bordage <cbordage@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lnet/lnet/api-ni.c b/lnet/lnet/api-ni.c
index cf96e8e..1ce3aad 100644 (file)
--- a/lnet/lnet/api-ni.c
+++ b/lnet/lnet/api-ni.c
@@ -2537,6 +2537,14 @@ lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
                goto failed0;
        }
 
+       lnet_net_lock(0);
+       if (lnet_nid_to_ni_locked(&ni->ni_nid, 0)) {
+               lnet_ni_addref_locked(ni, 0);
+               lnet_net_unlock(0);
+               return -EEXIST;
+       }
+       lnet_net_unlock(0);
+
        /* We keep a reference on the loopback net through the loopback NI */
        if (net->net_lnd->lnd_type == LOLND) {
                lnet_ni_addref(ni);
@@ -2687,7 +2695,9 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
 
                /* make sure that the the NI we're about to start
                 * up is actually unique. if it's not fail. */
-               if (!lnet_ni_unique_net(&net_l->net_ni_list,
+
+               if (ni->ni_interface &&
+                   !lnet_ni_unique_net(&net_l->net_ni_list,
                                        ni->ni_interface)) {
                        rc = -EEXIST;
                        goto failed1;
@@ -2699,6 +2709,9 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
 
                rc = lnet_startup_lndni(ni, tun);
 
+               if (rc == -EEXIST)
+                       list_add_tail(&ni->ni_netlist, &local_ni_list);
+
                if (rc != 0)
                        goto failed1;
 

diff --git a/lustre/tests/sanity-lnet.sh b/lustre/tests/sanity-lnet.sh
index bbbaebe..14e198b 100755 (executable)
--- a/lustre/tests/sanity-lnet.sh
+++ b/lustre/tests/sanity-lnet.sh
@@ -326,6 +326,7 @@ if [[ $NETTYPE =~ (tcp|o2ib)[0-9]* ]]; then
        if $FORCE_LARGE_NID; then
                always_except LU-14288 101
                always_except LU-14288 103
+               always_except LU-17457 199
                always_except LU-17457 208
                always_except LU-9680 213
                always_except LU-17458 220
@@ -1824,6 +1825,27 @@ test_111() {
 }
 run_test 111 "Test many routes"
 
+test_199() {
+       [[ ${NETTYPE} == tcp* || ${NETTYPE} == o2ib* ]] ||
+               skip "Need tcp or o2ib NETTYPE"
+
+       reinit_dlc || return $?
+
+       echo "Add interface ${INTERFACES[0]}"
+       do_lnetctl net add --net ${NETTYPE} --if ${INTERFACES[0]} ||
+               error "Failed to add net ${NETTYPE} with ${INTERFACES[0]}"
+
+       local ipaddress=$(ip --oneline addr show dev ${INTERFACES[0]} |
+                         awk '/inet /{print $4}' |
+                         sed 's:/.*::')
+
+       echo "Add IP address ${ipaddress} for interface ${INTERFACES[0]}"
+       do_lnetctl net add --ip2net "${NETTYPE} ${ipaddress}" || return 0
+
+       error "Failed to add net ${NETTYPE} with ${INTERFACES[0]}"
+}
+run_test 199 "load lnet w/o module option, configure interface twice"
+
 test_200() {
        [[ ${NETTYPE} == tcp* ]] ||
                skip "Need tcp NETTYPE"