[VPP-1240] NAT memory leak - identity mapping fo GbE IP

[vvalderrv]

Hi Matus,

 

We tried to add the identity mapping for GbE IP (192.168.16.10 in my case) over the binary API, but VPP will always crash after just few seconds of runtime:

Backtrace:

 

#0  0x00007f301f6dc428 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:54

#1  0x00007f301f6de02a in __GI_abort () at abort.c:89

#2  0x0000000000406a24 in os_panic () at /home/mlenco/dev/vpp/build-data/../src/vpp/vnet/main.c:268

#3  0x00007f301ff7148d in os_out_of_memory () at /home/mlenco/dev/vpp/build-data/../src/vppinfra/unix-misc.c:221

#4  0x00007f301ff4e90f in clib_mem_alloc_aligned_at_offset (size=568693920, align=4, align_offset=4,

    os_out_of_memory_on_failure=1) at /home/mlenco/dev/vpp/build-data/../src/vppinfra/mem.h:105

#5  0x00007f301ff4ed26 in vec_resize_allocate_memory (v=0x7f30009c57a0, length_increment=1, data_bytes=379129288,

    header_bytes=4, data_align=4) at /home/mlenco/dev/vpp/build-data/../src/vppinfra/vec.c:84

#6  0x00007f3020f9fd5a in _vec_resize (v=0x7f30009c57a0, length_increment=1, data_bytes=379129284, header_bytes=0,

    data_align=0) at /home/mlenco/dev/vpp/build-data/../src/vppinfra/vec.h:142

#7  0x00007f3020fa38c8 in vlib_put_next_frame (vm=0x7f3021246480 <vlib_global_main>, r=0x7f2fe0293c80,

    next_index=2, n_vectors_left=255) at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:501

#8  0x00007f302045a83e in ip4_lookup_inline (vm=0x7f3021246480 <vlib_global_main>, node=0x7f2fe0293c80,

    frame=0x7f2fe0316ac0, lookup_for_responses_to_locally_received_packets=0)

    at /home/mlenco/dev/vpp/build-data/../src/vnet/ip/ip4_forward.c:420

#9  0x00007f302045a8e6 in ip4_lookup (vm=0x7f3021246480 <vlib_global_main>, node=0x7f2fe0293c80,

    frame=0x7f2fe0316ac0) at /home/mlenco/dev/vpp/build-data/../src/vnet/ip/ip4_forward.c:465

#10 0x00007f3020fa4ca4 in dispatch_node (vm=0x7f3021246480 <vlib_global_main>, node=0x7f2fe0293c80,

    type=VLIB_NODE_TYPE_INTERNAL, dispatch_state=VLIB_NODE_STATE_POLLING, frame=0x7f2fe0316ac0,

    last_time_stamp=230993216616) at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:988

#11 0x00007f3020fa525d in dispatch_pending_node (vm=0x7f3021246480 <vlib_global_main>,

    pending_frame_index=31594105, last_time_stamp=230993216616)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:1138

#12 0x00007f3020fa7439 in vlib_main_or_worker_loop (vm=0x7f3021246480 <vlib_global_main>, is_main=1)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:1609

#13 0x00007f3020fa74e7 in vlib_main_loop (vm=0x7f3021246480 <vlib_global_main>)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:1628

#14 0x00007f3020fa7d5d in vlib_main (vm=0x7f3021246480 <vlib_global_main>, input=0x7f2fe012ffb0)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/main.c:1783

#15 0x00007f302100ed11 in thread0 (arg=139844691190912)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/unix/main.c:567

#16 0x00007f301fee6dc4 in clib_calljmp () at /home/mlenco/dev/vpp/build-data/../src/vppinfra/longjmp.S:110

#17 0x00007ffd63ccdb00 in ?? ()

#18 0x00007f302100f172 in vlib_unix_main (argc=17, argv=0x2570010)

    at /home/mlenco/dev/vpp/build-data/../src/vlib/unix/main.c:631

#19 0x000000000040670c in main (argc=17, argv=0x2570010)

    at /home/mlenco/dev/vpp/build-data/../src/vpp/vnet/main.c:207

  

Attached is compressed full core dump, api post-mortem dump and a screenshot of installed mappings, captured quickly just before VPP had crashed.

It appears the VPP had crashed in ip4-lookup with OOM. Without that extra identity mapping the VPP runs OK, so perhaps a memory leak in the NAT plugin?

VPP version: vpp-dev/vpp@88bfc1e

 

Thanks,

Milan