[PKG-Openstack-devel] Bug#813116: /usr/sbin/ovs-vswitchd: segfault when push_mpls flow action is applied

[Ferenc Wágner]

Hi,

Yesterday an upstream switch reboot segfaulted all our ovs-vswitchd
instances at the same time.  There's no MPLS or flow config in our OVS
setups, they run a bridge over a bond and several 802.1q interfaces.  I
mention it here, though, as the backtrace starts similarly, so maybe
this is a more general issue:

Core was generated by `ovs-vswitchd unix:/var/run/openvswitch/db.sock -vconsole:emer -vsyslog:err -vfi'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  nl_attr_get_size (nla=0x1ffffffffffff) at ../lib/netlink.c:516
516	../lib/netlink.c: No such file or directory.
(gdb) bt
#0  nl_attr_get_size (nla=0x1ffffffffffff) at ../lib/netlink.c:516
#1  0x000000000046fd76 in parse_l2_5_onward (attrs=0x7f954eb165b0, present_attrs=2097216, 
    out_of_range_attr=0, expected_attrs=64, flow=0x7f954eb16c90, key=0x7f954eb18fa4, key_len=136, 
    src_flow=0x7f954eb16c90) at ../lib/odp-util.c:2970
#2  0x0000000000470ba8 in parse_8021q_onward (src_flow=<optimized out>, key_len=<optimized out>, 
    key=<optimized out>, flow=<optimized out>, expected_attrs=64, out_of_range_attr=0, 
    present_attrs=2097216, attrs=<optimized out>) at ../lib/odp-util.c:3284
#3  odp_flow_key_to_flow__ (key=0x7f954eb18fa4, key_len=136, flow=0x7f954eb16c90, 
    src_flow=0x7f954eb16c90) at ../lib/odp-util.c:3377
#4  0x0000000000470f68 in odp_flow_key_to_flow (key=key at entry=0x7f954eb18fa4, 
    key_len=key_len at entry=136, flow=flow at entry=0x7f954eb16c90) at ../lib/odp-util.c:3410
#5  0x0000000000434bcd in xlate_receive (backer=0xb3aa40, packet=0x7f954eb18f00, key=0x7f954eb18fa4, 
    key_len=136, flow=0x7f954eb16c90, ofproto=0x7f954eb168b8, ipfix=0x7f954eb168c8, 
    sflow=0x7f954eb168c0, netflow=0x0, odp_in_port=0x7f954eb168b4)
    at ../ofproto/ofproto-dpif-xlate.c:640
#6  0x000000000042e728 in read_upcalls (handler=<optimized out>, handler=<optimized out>, 
    misses=<optimized out>, miss_buf=<optimized out>, upcalls=<optimized out>)
    at ../ofproto/ofproto-dpif-upcall.c:788
#7  udpif_upcall_handler (arg=0x7f954eb18ef0) at ../ofproto/ofproto-dpif-upcall.c:525
#8  0x0000000000494eb1 in ovsthread_wrapper (aux_=<optimized out>) at ../lib/ovs-thread.c:322
#9  0x00007f95505f3064 in start_thread (arg=0x7f954eb30700) at pthread_create.c:309
#10 0x00007f954fe1f62d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

The following restart attempts also segfaulted.  The above core file is
probably from the last attempt.  Later a full OVS restart fixed the
issue.

Identically positioned 2.5.2-0ubuntu0.16.04.3~cloud0 OVS versions (in
the same topology) had not problem at the same time.
-- 
Regards,
Feri