TCP btl segfaulting with ipv6

[PeterGottesman]

Following #5328, all v2.x tests are segfaulting on MTT when configured with --enable-ipv6. I confirmed that this always happens with ipv6 enabled, and was not present before that PR was merged.

I am unsure of the cause, but in gdb I see that at line 365 index is 4294951708. I suspect is incorrect as the size of the array is 8(MAX_KERNEL_INTERFACES) as defined on line 283, confirmed through gdb.

Configure args:

--enable-picky --enable-debug --enable-mpirun-prefix-by-default --enable-mpi-cxx
--disable-dlopen --enable-ipv6

$ mpirun --hetero-nodes -np 2 --mca orte_startup_timeout 10000 --mca oob tcp --mca btl tcp,self ./c_ring 

Sample output: https://mtt.open-mpi.org/index.php?do_redir=2643

(gdb) bt
#0  0x00002aaaabc44d9f in mca_btl_tcp_retrieve_local_interfaces (proc_data=0x7fffffffc280) at btl_tcp_proc.c:365
#1  0x00002aaaabc4514d in mca_btl_tcp_proc_insert (btl_proc=0x751450, btl_endpoint=0x755a70) at btl_tcp_proc.c:453
#2  0x00002aaaabc3b02a in mca_btl_tcp_add_procs (btl=0x726050, nprocs=2, procs=0x74aa70, peers=0x74aab0, reachable=0x7fffffffc630) at btl_tcp.c:118
#3  0x00002aaaaadb2002 in mca_bml_r2_add_procs (nprocs=2, procs=0x74aa70, reachable=0x7fffffffc630) at bml_r2.c:521
#4  0x00002aaaaaec6489 in mca_pml_ob1_add_procs (procs=0x74ccf0, nprocs=2) at pml_ob1.c:302
#5  0x00002aaaaad42e80 in ompi_mpi_init (argc=1, argv=0x7fffffffc888, requested=0, provided=0x7fffffffc73c, reinit_ok=false) at runtime/ompi_mpi_init.c:777
#6  0x00002aaaaad76c9a in PMPI_Init (argc=0x7fffffffc76c, argv=0x7fffffffc760) at pinit.c:66
#7  0x00000000004009e3 in main ()

This issue is present in #5330 and #5331, which have not been merged yet, as well.

[PeterGottesman]

@jsquyres @hppritcha

[jsquyres]

Fixed in all the branches.