When the master SM failovers to the slave SM, the following sequence of
events cause segfault in ibacm. We are using Mellanox CX3 in the
virtualized environment.
1. IBV_EVENT_CLIENT_REREGISTRATION
2. IBV_EVENT_GID_CHANGE
During the client reregistration, a multicast join request is sent. Then,
when handling the GID table change async event, the acm logical port is
reset(the acm port is shutdown and set back to up again). Due to the fact
that the physical IB port is ACTIVE, the multicast join response can come
back before the acm logical port has transitioned back to up state. In the
handling of the multicast join response while the acm logical port is not
up, the acm process hits into a null pointer derefence when it tries to
associate the gid with the associate acm port via acm_gid_index. To avoid
this, this patch delays the handling of the MAD until the next event if
there is no acm port provider in the associated IB port.
ibacm[39381]: segfault at bc ip 0000000000402e59 sp 00007fc77c267bf8 error
4 in ibacm[400000+c000]
back trace
==========
0 acm_gid_index (port=0x0, gid=0x1088b30) at src/acm.c:374
1 0x00007f1e0b18f93c in acmp_record_mc_av (sa_mad=0x1088aa0) at
prov/acmp/src/acmp.c:658
2 acmp_process_join_resp (sa_mad=0x1088aa0) at prov/acmp/src/acmp.c:734
3 0x00000000004038c1 in acmc_recv_mad (port=0x107bc78) at src/acm.c:2888
4 0x0000000000403b0d in acm_sa_handler (context=<value optimized out>) at
src/acm.c:2926
Signed-off-by: Wei Lin Guay <wei.lin.guay@xxxxxxxxxx>
Reviewed-by: Håkon Bugge <haakon.bugge@xxxxxxxxxx>
---
ibacm/src/acm.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/ibacm/src/acm.c b/ibacm/src/acm.c
index 59385fb..22a7055 100644
--- a/ibacm/src/acm.c
+++ b/ibacm/src/acm.c
@@ -2855,6 +2855,11 @@ static void acmc_recv_mad(struct acmc_port *port)
int ret, len, found;
struct umad_hdr *hdr;
+ if (!port->prov) {
+ acm_log(1, "no provider assigned to port\n");
+ return;
+ }
+
acm_log(2, "\n");
len = sizeof(resp.sa_mad);
ret = umad_recv(port->mad_portid, &resp.umad, &len, 0);
--
2.4.3
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
