bpf: implement btf handling and map annotation

bpf: implement btf handling and map annotation

Implement loading of .BTF section from object file and build up
internal table for retrieving key/value id related to maps in
the BPF program. Latter is done by setting up struct btf_type
table.

One of the issues is that there's a disconnect between the data
types used in the map and struct bpf_elf_map, meaning the underlying
types are unknown from the map description. One way to overcome
this is to add a annotation such that the loader will recognize
the relation to both. BPF_ANNOTATE_KV_PAIR(map_foo, struct key,
struct val); has been added to the API that programs can use.

The loader will then pick the corresponding key/value type ids and
attach it to the maps for creation. This can later on be dumped via
bpftool for introspection.

Example with test_xdp_noinline.o from kernel selftests:

  [...]

  struct ctl_value {
        union {
                __u64 value;
                __u32 ifindex;
                __u8 mac[6];
        };
  };

  struct bpf_map_def __attribute__ ((section("maps"), used)) ctl_array = {
        .type		= BPF_MAP_TYPE_ARRAY,
        .key_size	= sizeof(__u32),
        .value_size	= sizeof(struct ctl_value),
        .max_entries	= 16,
        .map_flags	= 0,
  };
  BPF_ANNOTATE_KV_PAIR(ctl_array, __u32, struct ctl_value);

  [...]

Above could also further be wrapped in a macro. Compiling through LLVM and
converting to BTF:

  # llc --version
  LLVM (http://llvm.org/):
    LLVM version 7.0.0svn
    Optimized build.
    Default target: x86_64-unknown-linux-gnu
    Host CPU: skylake

    Registered Targets:
      bpf    - BPF (host endian)
      bpfeb  - BPF (big endian)
      bpfel  - BPF (little endian)
  [...]

  # clang [...] -O2 -target bpf -g -emit-llvm -c test_xdp_noinline.c -o - |
    llc -march=bpf -mcpu=probe -mattr=dwarfris -filetype=obj -o test_xdp_noinline.o
  # pahole -J test_xdp_noinline.o

Checking pahole dump of BPF object file:

  # file test_xdp_noinline.o
  test_xdp_noinline.o: ELF 64-bit LSB relocatable, *unknown arch 0xf7* version 1 (SYSV), with debug_info, not stripped
  # pahole test_xdp_noinline.o
  [...]
  struct ctl_value {
	union {
		__u64              value;                /*     0     8 */
		__u32              ifindex;              /*     0     4 */
		__u8               mac[0];               /*     0     0 */
	};                                               /*     0     8 */

	/* size: 8, cachelines: 1, members: 1 */
	/* last cacheline: 8 bytes */
  };

Now loading into kernel and dumping the map via bpftool:

  # ip -force link set dev lo xdp obj test_xdp_noinline.o sec xdp-test
  # ip a
  1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 xdpgeneric/id:227 qdisc noqueue state UNKNOWN group default qlen 1000
      link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
      inet 127.0.0.1/8 scope host lo
         valid_lft forever preferred_lft forever
      inet6 ::1/128 scope host
         valid_lft forever preferred_lft forever
  [...]
  # bpftool prog show id 227
  227: xdp  tag a85e060c275c5616  gpl
      loaded_at 2018-07-17T14:41:29+0000  uid 0
      xlated 8152B  not jited  memlock 12288B  map_ids 381,385,386,382,384,383
  # bpftool map dump id 386
   [{
        "key": 0,
        "value": {
            "": {
                "value": 0,
                "ifindex": 0,
                "mac": []
            }
        }
    },{
        "key": 1,
        "value": {
            "": {
                "value": 0,
                "ifindex": 0,
                "mac": []
            }
        }
    },{
  [...]

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>