Documentation/bpf/drgn.rst

   1 .. SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2
   3 ==============
   4 BPF drgn tools
   5 ==============
   6
   7 drgn scripts is a convenient and easy to use mechanism to retrieve arbitrary
   8 kernel data structures. drgn is not relying on kernel UAPI to read the data.
   9 Instead it's reading directly from ``/proc/kcore`` or vmcore and pretty prints
  10 the data based on DWARF debug information from vmlinux.
  11
  12 This document describes BPF related drgn tools.
  13
  14 See `drgn/tools`_ for all tools available at the moment and `drgn/doc`_ for
  15 more details on drgn itself.
  16
  17 bpf_inspect.py
  18 --------------
  19
  20 Description
  21 ===========
  22
  23 `bpf_inspect.py`_ is a tool intended to inspect BPF programs and maps. It can
  24 iterate over all programs and maps in the system and print basic information
  25 about these objects, including id, type and name.
  26
  27 The main use-case `bpf_inspect.py`_ covers is to show BPF programs of types
  28 ``BPF_PROG_TYPE_EXT`` and ``BPF_PROG_TYPE_TRACING`` attached to other BPF
  29 programs via ``freplace``/``fentry``/``fexit`` mechanisms, since there is no
  30 user-space API to get this information.
  31
  32 Getting started
  33 ===============
  34
  35 List BPF programs (full names are obtained from BTF)::
  36
  37     % sudo bpf_inspect.py prog
  38         27: BPF_PROG_TYPE_TRACEPOINT         tracepoint__tcp__tcp_send_reset
  39       4632: BPF_PROG_TYPE_CGROUP_SOCK_ADDR   tw_ipt_bind
  40      49464: BPF_PROG_TYPE_RAW_TRACEPOINT     raw_tracepoint__sched_process_exit
  41
  42 List BPF maps::
  43
  44       % sudo bpf_inspect.py map
  45         2577: BPF_MAP_TYPE_HASH                tw_ipt_vips
  46         4050: BPF_MAP_TYPE_STACK_TRACE         stack_traces
  47         4069: BPF_MAP_TYPE_PERCPU_ARRAY        ned_dctcp_cntr
  48
  49 Find BPF programs attached to BPF program ``test_pkt_access``::
  50
  51       % sudo bpf_inspect.py p | grep test_pkt_access
  52          650: BPF_PROG_TYPE_SCHED_CLS          test_pkt_access
  53          654: BPF_PROG_TYPE_TRACING            test_main                        linked:[650->25: BPF_TRAMP_FEXIT test_pkt_access->test_pkt_access()]
  54          655: BPF_PROG_TYPE_TRACING            test_subprog1                    linked:[650->29: BPF_TRAMP_FEXIT test_pkt_access->test_pkt_access_subprog1()]
  55          656: BPF_PROG_TYPE_TRACING            test_subprog2                    linked:[650->31: BPF_TRAMP_FEXIT test_pkt_access->test_pkt_access_subprog2()]
  56          657: BPF_PROG_TYPE_TRACING            test_subprog3                    linked:[650->21: BPF_TRAMP_FEXIT test_pkt_access->test_pkt_access_subprog3()]
  57          658: BPF_PROG_TYPE_EXT                new_get_skb_len                  linked:[650->16: BPF_TRAMP_REPLACE test_pkt_access->get_skb_len()]
  58          659: BPF_PROG_TYPE_EXT                new_get_skb_ifindex              linked:[650->23: BPF_TRAMP_REPLACE test_pkt_access->get_skb_ifindex()]
  59          660: BPF_PROG_TYPE_EXT                new_get_constant                 linked:[650->19: BPF_TRAMP_REPLACE test_pkt_access->get_constant()]
  60
  61 It can be seen that there is a program ``test_pkt_access``, id 650 and there
  62 are multiple other tracing and ext programs attached to functions in
  63 ``test_pkt_access``.
  64
  65 For example the line::
  66
  67          658: BPF_PROG_TYPE_EXT                new_get_skb_len                  linked:[650->16: BPF_TRAMP_REPLACE test_pkt_access->get_skb_len()]
  68
  69 , means that BPF program id 658, type ``BPF_PROG_TYPE_EXT``, name
  70 ``new_get_skb_len`` replaces (``BPF_TRAMP_REPLACE``) function ``get_skb_len()``
  71 that has BTF id 16 in BPF program id 650, name ``test_pkt_access``.
  72
  73 Getting help:
  74
  75 .. code-block:: none
  76
  77     % sudo bpf_inspect.py
  78     usage: bpf_inspect.py [-h] {prog,p,map,m} ...
  79
  80     drgn script to list BPF programs or maps and their properties
  81     unavailable via kernel API.
  82
  83     See https://github.com/osandov/drgn/ for more details on drgn.
  84
  85     optional arguments:
  86       -h, --help      show this help message and exit
  87
  88     subcommands:
  89       {prog,p,map,m}
  90         prog (p)      list BPF programs
  91         map (m)       list BPF maps
  92
  93 Customization
  94 =============
  95
  96 The script is intended to be customized by developers to print relevant
  97 information about BPF programs, maps and other objects.
  98
  99 For example, to print ``struct bpf_prog_aux`` for BPF program id 53077:
 100
 101 .. code-block:: none
 102
 103     % git diff
 104     diff --git a/tools/bpf_inspect.py b/tools/bpf_inspect.py
 105     index 650e228..aea2357 100755
 106     --- a/tools/bpf_inspect.py
 107     +++ b/tools/bpf_inspect.py
 108     @@ -112,7 +112,9 @@ def list_bpf_progs(args):
 109              if linked:
 110                  linked = f" linked:[{linked}]"
 111
 112     -        print(f"{id_:>6}: {type_:32} {name:32} {linked}")
 113     +        if id_ == 53077:
 114     +            print(f"{id_:>6}: {type_:32} {name:32}")
 115     +            print(f"{bpf_prog.aux}")
 116
 117
 118      def list_bpf_maps(args):
 119
 120 It produces the output::
 121
 122     % sudo bpf_inspect.py p
 123      53077: BPF_PROG_TYPE_XDP                tw_xdp_policer
 124     *(struct bpf_prog_aux *)0xffff8893fad4b400 = {
 125             .refcnt = (atomic64_t){
 126                     .counter = (long)58,
 127             },
 128             .used_map_cnt = (u32)1,
 129             .max_ctx_offset = (u32)8,
 130             .max_pkt_offset = (u32)15,
 131             .max_tp_access = (u32)0,
 132             .stack_depth = (u32)8,
 133             .id = (u32)53077,
 134             .func_cnt = (u32)0,
 135             .func_idx = (u32)0,
 136             .attach_btf_id = (u32)0,
 137             .linked_prog = (struct bpf_prog *)0x0,
 138             .verifier_zext = (bool)0,
 139             .offload_requested = (bool)0,
 140             .attach_btf_trace = (bool)0,
 141             .func_proto_unreliable = (bool)0,
 142             .trampoline_prog_type = (enum bpf_tramp_prog_type)BPF_TRAMP_FENTRY,
 143             .trampoline = (struct bpf_trampoline *)0x0,
 144             .tramp_hlist = (struct hlist_node){
 145                     .next = (struct hlist_node *)0x0,
 146                     .pprev = (struct hlist_node **)0x0,
 147             },
 148             .attach_func_proto = (const struct btf_type *)0x0,
 149             .attach_func_name = (const char *)0x0,
 150             .func = (struct bpf_prog **)0x0,
 151             .jit_data = (void *)0x0,
 152             .poke_tab = (struct bpf_jit_poke_descriptor *)0x0,
 153             .size_poke_tab = (u32)0,
 154             .ksym_tnode = (struct latch_tree_node){
 155                     .node = (struct rb_node [2]){
 156                             {
 157                                     .__rb_parent_color = (unsigned long)18446612956263126665,
 158                                     .rb_right = (struct rb_node *)0x0,
 159                                     .rb_left = (struct rb_node *)0xffff88a0be3d0088,
 160                             },
 161                             {
 162                                     .__rb_parent_color = (unsigned long)18446612956263126689,
 163                                     .rb_right = (struct rb_node *)0x0,
 164                                     .rb_left = (struct rb_node *)0xffff88a0be3d00a0,
 165                             },
 166                     },
 167             },
 168             .ksym_lnode = (struct list_head){
 169                     .next = (struct list_head *)0xffff88bf481830b8,
 170                     .prev = (struct list_head *)0xffff888309f536b8,
 171             },
 172             .ops = (const struct bpf_prog_ops *)xdp_prog_ops+0x0 = 0xffffffff820fa350,
 173             .used_maps = (struct bpf_map **)0xffff889ff795de98,
 174             .prog = (struct bpf_prog *)0xffffc9000cf2d000,
 175             .user = (struct user_struct *)root_user+0x0 = 0xffffffff82444820,
 176             .load_time = (u64)2408348759285319,
 177             .cgroup_storage = (struct bpf_map *[2]){},
 178             .name = (char [16])"tw_xdp_policer",
 179             .security = (void *)0xffff889ff795d548,
 180             .offload = (struct bpf_prog_offload *)0x0,
 181             .btf = (struct btf *)0xffff8890ce6d0580,
 182             .func_info = (struct bpf_func_info *)0xffff889ff795d240,
 183             .func_info_aux = (struct bpf_func_info_aux *)0xffff889ff795de20,
 184             .linfo = (struct bpf_line_info *)0xffff888a707afc00,
 185             .jited_linfo = (void **)0xffff8893fad48600,
 186             .func_info_cnt = (u32)1,
 187             .nr_linfo = (u32)37,
 188             .linfo_idx = (u32)0,
 189             .num_exentries = (u32)0,
 190             .extable = (struct exception_table_entry *)0xffffffffa032d950,
 191             .stats = (struct bpf_prog_stats *)0x603fe3a1f6d0,
 192             .work = (struct work_struct){
 193                     .data = (atomic_long_t){
 194                             .counter = (long)0,
 195                     },
 196                     .entry = (struct list_head){
 197                             .next = (struct list_head *)0x0,
 198                             .prev = (struct list_head *)0x0,
 199                     },
 200                     .func = (work_func_t)0x0,
 201             },
 202             .rcu = (struct callback_head){
 203                     .next = (struct callback_head *)0x0,
 204                     .func = (void (*)(struct callback_head *))0x0,
 205             },
 206     }
 207
 208
 209 .. Links
 210 .. _drgn/doc: https://drgn.readthedocs.io/en/latest/
 211 .. _drgn/tools: https://github.com/osandov/drgn/tree/master/tools
 212 .. _bpf_inspect.py:
 213    https://github.com/osandov/drgn/blob/master/tools/bpf_inspect.py