On 11/20/20 4:37 PM, Dumitru Ceara wrote: > When a port binding of type "l3gateway" is claimed its remote peer > port_binding is also stored in local_datapath.peer_ports[].remote. > > If the remote peer port_binding is deleted first (i.e., before the local > "l3gateway" one) then we need to remove the complete > local_datapath.peer_ports[] entry in order to avoid ending up using > dangling pointers to already freed port bindings. > > Also, properly reset local_datapath->has_local_l3gateway in > remove_pb_from_local_datapath(). > > Ilya reported this issue found by AddressSanitizer during his testing: > > ==1816017==ERROR: AddressSanitizer: heap-use-after-free on address > 0x6140000cb170 at pc 0x0000005ab574 bp 0x7fff68925a30 sp 0x7fff68925a28 > READ of size 8 at 0x6140000cb170 thread T0 > #0 0x5ab573 in put_replace_chassis_mac_flows > git/ovn/controller/physical.c:550:9 > #1 0x5a65eb in consider_port_binding git/ovn/controller/physical.c:1168:13 > #2 0x5a8764 in physical_run git/ovn/controller/physical.c:1607:9 > #3 0x5a0064 in flow_output_physical_flow_changes_handler > git/ovn/controller/ovn-controller.c:2127:9 > #4 0x5db423 in engine_compute git/ovn/lib/inc-proc-eng.c:306:18 > #5 0x5dae1f in engine_run_node git/ovn/lib/inc-proc-eng.c:352:14 > #6 0x5dac74 in engine_run git/ovn/lib/inc-proc-eng.c:377:9 > #7 0x59ad64 in main git/ovn/controller/ovn-controller.c > #8 0x7f39fa6491a2 in __libc_start_main (/lib64/libc.so.6+0x271a2) > #9 0x480b2d in _start (git/ovn/controller/ovn-controller+0x480b2d) > > 0x6140000cb170 is located 304 bytes inside of 408-byte region > [0x6140000cb040,0x6140000cb1d8) > freed by thread T0 here: > #0 0x520d07 in free (git/ovn/controller/ovn-controller+0x520d07) > #1 0x712de7 in ovsdb_idl_db_track_clear git/ovs/lib/ovsdb-idl.c:1984:21 > #2 0x59b5cd in main git/ovn/controller/ovn-controller.c:2762:9 > #3 0x7f39fa6491a2 in __libc_start_main (/lib64/libc.so.6+0x271a2) > > Reported-by: Ilya Maximets <i.maxim...@ovn.org> > Fixes: 354bdba51abf ("ovn-controller: I-P for SB port binding and OVS > interface in runtime_data.") > Signed-off-by: Dumitru Ceara <dce...@redhat.com> > ---
I do not know this part of the code well enough to ack, but this patch fixes the issue for me. Thanks! Tested-by: Ilya Maximets <i.maxim...@ovn.org> Best regards, Ilya Maximets. _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev