xref: /linux/tools/bpf/bpftool/Documentation/bpftool-gen.rst (revision 68a052239fc4b351e961f698b824f7654a346091)
1.. SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2
3================
4bpftool-gen
5================
6-------------------------------------------------------------------------------
7tool for BPF code-generation
8-------------------------------------------------------------------------------
9
10:Manual section: 8
11
12.. include:: substitutions.rst
13
14SYNOPSIS
15========
16
17**bpftool** [*OPTIONS*] **gen** *COMMAND*
18
19*OPTIONS* := { |COMMON_OPTIONS| | { **-L** | **--use-loader** } | [ { **-S** | **--sign** } {**-k** <private_key.pem>} **-i** <certificate.x509> ] }
20
21*COMMAND* := { **object** | **skeleton** | **help** }
22
23GEN COMMANDS
24=============
25
26| **bpftool** **gen object** *OUTPUT_FILE* *INPUT_FILE* [*INPUT_FILE*...]
27| **bpftool** **gen skeleton** *FILE* [**name** *OBJECT_NAME*]
28| **bpftool** **gen subskeleton** *FILE* [**name** *OBJECT_NAME*]
29| **bpftool** **gen min_core_btf** *INPUT* *OUTPUT* *OBJECT* [*OBJECT*...]
30| **bpftool** **gen help**
31
32DESCRIPTION
33===========
34bpftool gen object *OUTPUT_FILE* *INPUT_FILE* [*INPUT_FILE*...]
35    Statically link (combine) together one or more *INPUT_FILE*'s into a single
36    resulting *OUTPUT_FILE*. All the files involved are BPF ELF object files.
37
38    The rules of BPF static linking are mostly the same as for user-space
39    object files, but in addition to combining data and instruction sections,
40    .BTF and .BTF.ext (if present in any of the input files) data are combined
41    together. .BTF data is deduplicated, so all the common types across
42    *INPUT_FILE*'s will only be represented once in the resulting BTF
43    information.
44
45    BPF static linking allows to partition BPF source code into individually
46    compiled files that are then linked into a single resulting BPF object
47    file, which can be used to generated BPF skeleton (with **gen skeleton**
48    command) or passed directly into **libbpf** (using **bpf_object__open()**
49    family of APIs).
50
51bpftool gen skeleton *FILE*
52    Generate BPF skeleton C header file for a given *FILE*.
53
54    BPF skeleton is an alternative interface to existing libbpf APIs for
55    working with BPF objects. Skeleton code is intended to significantly
56    shorten and simplify code to load and work with BPF programs from userspace
57    side. Generated code is tailored to specific input BPF object *FILE*,
58    reflecting its structure by listing out available maps, program, variables,
59    etc. Skeleton eliminates the need to lookup mentioned components by name.
60    Instead, if skeleton instantiation succeeds, they are populated in skeleton
61    structure as valid libbpf types (e.g., **struct bpf_map** pointer) and can
62    be passed to existing generic libbpf APIs.
63
64    In addition to simple and reliable access to maps and programs, skeleton
65    provides a storage for BPF links (**struct bpf_link**) for each BPF program
66    within BPF object. When requested, supported BPF programs will be
67    automatically attached and resulting BPF links stored for further use by
68    user in pre-allocated fields in skeleton struct. For BPF programs that
69    can't be automatically attached by libbpf, user can attach them manually,
70    but store resulting BPF link in per-program link field. All such set up
71    links will be automatically destroyed on BPF skeleton destruction. This
72    eliminates the need for users to manage links manually and rely on libbpf
73    support to detach programs and free up resources.
74
75    Another facility provided by BPF skeleton is an interface to global
76    variables of all supported kinds: mutable, read-only, as well as extern
77    ones. This interface allows to pre-setup initial values of variables before
78    BPF object is loaded and verified by kernel. For non-read-only variables,
79    the same interface can be used to fetch values of global variables on
80    userspace side, even if they are modified by BPF code.
81
82    During skeleton generation, contents of source BPF object *FILE* is
83    embedded within generated code and is thus not necessary to keep around.
84    This ensures skeleton and BPF object file are matching 1-to-1 and always
85    stay in sync. Generated code is dual-licensed under LGPL-2.1 and
86    BSD-2-Clause licenses.
87
88    It is a design goal and guarantee that skeleton interfaces are
89    interoperable with generic libbpf APIs. User should always be able to use
90    skeleton API to create and load BPF object, and later use libbpf APIs to
91    keep working with specific maps, programs, etc.
92
93    As part of skeleton, few custom functions are generated. Each of them is
94    prefixed with object name. Object name can either be derived from object
95    file name, i.e., if BPF object file name is **example.o**, BPF object name
96    will be **example**. Object name can be also specified explicitly through
97    **name** *OBJECT_NAME* parameter. The following custom functions are
98    provided (assuming **example** as the object name):
99
100    - **example__open** and **example__open_opts**.
101      These functions are used to instantiate skeleton. It corresponds to
102      libbpf's **bpf_object__open**\ () API. **_opts** variants accepts extra
103      **bpf_object_open_opts** options.
104
105    - **example__load**.
106      This function creates maps, loads and verifies BPF programs, initializes
107      global data maps. It corresponds to libbpf's **bpf_object__load**\ ()
108      API.
109
110    - **example__open_and_load** combines **example__open** and
111      **example__load** invocations in one commonly used operation.
112
113    - **example__attach** and **example__detach**.
114      This pair of functions allow to attach and detach, correspondingly,
115      already loaded BPF object. Only BPF programs of types supported by libbpf
116      for auto-attachment will be auto-attached and their corresponding BPF
117      links instantiated. For other BPF programs, user can manually create a
118      BPF link and assign it to corresponding fields in skeleton struct.
119      **example__detach** will detach both links created automatically, as well
120      as those populated by user manually.
121
122    - **example__destroy**.
123      Detach and unload BPF programs, free up all the resources used by
124      skeleton and BPF object.
125
126    If BPF object has global variables, corresponding structs with memory
127    layout corresponding to global data data section layout will be created.
128    Currently supported ones are: *.data*, *.bss*, *.rodata*, and *.kconfig*
129    structs/data sections. These data sections/structs can be used to set up
130    initial values of variables, if set before **example__load**. Afterwards,
131    if target kernel supports memory-mapped BPF arrays, same structs can be
132    used to fetch and update (non-read-only) data from userspace, with same
133    simplicity as for BPF side.
134
135bpftool gen subskeleton *FILE*
136    Generate BPF subskeleton C header file for a given *FILE*.
137
138    Subskeletons are similar to skeletons, except they do not own the
139    corresponding maps, programs, or global variables. They require that the
140    object file used to generate them is already loaded into a *bpf_object* by
141    some other means.
142
143    This functionality is useful when a library is included into a larger BPF
144    program. A subskeleton for the library would have access to all objects and
145    globals defined in it, without having to know about the larger program.
146
147    Consequently, there are only two functions defined for subskeletons:
148
149    - **example__open(bpf_object\*)**.
150      Instantiates a subskeleton from an already opened (but not necessarily
151      loaded) **bpf_object**.
152
153    - **example__destroy()**.
154      Frees the storage for the subskeleton but *does not* unload any BPF
155      programs or maps.
156
157bpftool gen min_core_btf *INPUT* *OUTPUT* *OBJECT* [*OBJECT*...]
158    Generate a minimum BTF file as *OUTPUT*, derived from a given *INPUT* BTF
159    file, containing all needed BTF types so one, or more, given eBPF objects
160    CO-RE relocations may be satisfied.
161
162    When kernels aren't compiled with CONFIG_DEBUG_INFO_BTF, libbpf, when
163    loading an eBPF object, has to rely on external BTF files to be able to
164    calculate CO-RE relocations.
165
166    Usually, an external BTF file is built from existing kernel DWARF data
167    using pahole. It contains all the types used by its respective kernel image
168    and, because of that, is big.
169
170    The min_core_btf feature builds smaller BTF files, customized to one or
171    multiple eBPF objects, so they can be distributed together with an eBPF
172    CO-RE based application, turning the application portable to different
173    kernel versions.
174
175    Check examples below for more information on how to use it.
176
177bpftool gen help
178    Print short help message.
179
180OPTIONS
181=======
182.. include:: common_options.rst
183
184-L, --use-loader
185    For skeletons, generate a "light" skeleton (also known as "loader"
186    skeleton). A light skeleton contains a loader eBPF program. It does not use
187    the majority of the libbpf infrastructure, and does not need libelf.
188
189-S, --sign
190    For skeletons, generate a signed skeleton. This option must be used with
191    **-k** and **-i**. Using this flag implicitly enables **--use-loader**.
192
193-k <private_key.pem>
194    Path to the private key file in PEM format, required for signing.
195
196-i <certificate.x509>
197    Path to the X.509 certificate file in PEM or DER format, required for
198    signing.
199
200EXAMPLES
201========
202**$ cat example1.bpf.c**
203
204::
205
206  #include <stdbool.h>
207  #include <linux/ptrace.h>
208  #include <linux/bpf.h>
209  #include <bpf/bpf_helpers.h>
210
211  const volatile int param1 = 42;
212  bool global_flag = true;
213  struct { int x; } data = {};
214
215  SEC("raw_tp/sys_enter")
216  int handle_sys_enter(struct pt_regs *ctx)
217  {
218  	static long my_static_var;
219  	if (global_flag)
220  		my_static_var++;
221  	else
222  		data.x += param1;
223  	return 0;
224  }
225
226**$ cat example2.bpf.c**
227
228::
229
230  #include <linux/ptrace.h>
231  #include <linux/bpf.h>
232  #include <bpf/bpf_helpers.h>
233
234  struct {
235  	__uint(type, BPF_MAP_TYPE_HASH);
236  	__uint(max_entries, 128);
237  	__type(key, int);
238  	__type(value, long);
239  } my_map SEC(".maps");
240
241  SEC("raw_tp/sys_exit")
242  int handle_sys_exit(struct pt_regs *ctx)
243  {
244  	int zero = 0;
245  	bpf_map_lookup_elem(&my_map, &zero);
246  	return 0;
247  }
248
249**$ cat example3.bpf.c**
250
251::
252
253  #include <linux/ptrace.h>
254  #include <linux/bpf.h>
255  #include <bpf/bpf_helpers.h>
256  /* This header file is provided by the bpf_testmod module. */
257  #include "bpf_testmod.h"
258
259  int test_2_result = 0;
260
261  /* bpf_Testmod.ko calls this function, passing a "4"
262   * and testmod_map->data.
263   */
264  SEC("struct_ops/test_2")
265  void BPF_PROG(test_2, int a, int b)
266  {
267	test_2_result = a + b;
268  }
269
270  SEC(".struct_ops")
271  struct bpf_testmod_ops testmod_map = {
272	.test_2 = (void *)test_2,
273	.data = 0x1,
274  };
275
276This is example BPF application with three BPF programs and a mix of BPF
277maps and global variables. Source code is split across three source code
278files.
279
280**$ clang --target=bpf -g example1.bpf.c -o example1.bpf.o**
281
282**$ clang --target=bpf -g example2.bpf.c -o example2.bpf.o**
283
284**$ clang --target=bpf -g example3.bpf.c -o example3.bpf.o**
285
286**$ bpftool gen object example.bpf.o example1.bpf.o example2.bpf.o example3.bpf.o**
287
288This set of commands compiles *example1.bpf.c*, *example2.bpf.c* and
289*example3.bpf.c* individually and then statically links respective object
290files into the final BPF ELF object file *example.bpf.o*.
291
292**$ bpftool gen skeleton example.bpf.o name example | tee example.skel.h**
293
294::
295
296  /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
297
298  /* THIS FILE IS AUTOGENERATED! */
299  #ifndef __EXAMPLE_SKEL_H__
300  #define __EXAMPLE_SKEL_H__
301
302  #include <stdlib.h>
303  #include <bpf/libbpf.h>
304
305  struct example {
306  	struct bpf_object_skeleton *skeleton;
307  	struct bpf_object *obj;
308  	struct {
309  		struct bpf_map *rodata;
310  		struct bpf_map *data;
311  		struct bpf_map *bss;
312  		struct bpf_map *my_map;
313		struct bpf_map *testmod_map;
314  	} maps;
315	struct {
316		struct example__testmod_map__bpf_testmod_ops {
317			const struct bpf_program *test_1;
318			const struct bpf_program *test_2;
319			int data;
320		} *testmod_map;
321	} struct_ops;
322  	struct {
323  		struct bpf_program *handle_sys_enter;
324  		struct bpf_program *handle_sys_exit;
325  	} progs;
326  	struct {
327  		struct bpf_link *handle_sys_enter;
328  		struct bpf_link *handle_sys_exit;
329  	} links;
330  	struct example__bss {
331  		struct {
332  			int x;
333  		} data;
334		int test_2_result;
335  	} *bss;
336  	struct example__data {
337  		_Bool global_flag;
338  		long int handle_sys_enter_my_static_var;
339  	} *data;
340  	struct example__rodata {
341  		int param1;
342  	} *rodata;
343  };
344
345  static void example__destroy(struct example *obj);
346  static inline struct example *example__open_opts(
347                const struct bpf_object_open_opts *opts);
348  static inline struct example *example__open();
349  static inline int example__load(struct example *obj);
350  static inline struct example *example__open_and_load();
351  static inline int example__attach(struct example *obj);
352  static inline void example__detach(struct example *obj);
353
354  #endif /* __EXAMPLE_SKEL_H__ */
355
356**$ cat example.c**
357
358::
359
360  #include "example.skel.h"
361
362  int main()
363  {
364  	struct example *skel;
365  	int err = 0;
366
367  	skel = example__open();
368  	if (!skel)
369  		goto cleanup;
370
371  	skel->rodata->param1 = 128;
372
373	/* Change the value through the pointer of shadow type */
374	skel->struct_ops.testmod_map->data = 13;
375
376  	err = example__load(skel);
377  	if (err)
378  		goto cleanup;
379
380	/* The result of the function test_2() */
381	printf("test_2_result: %d\n", skel->bss->test_2_result);
382
383  	err = example__attach(skel);
384  	if (err)
385  		goto cleanup;
386
387  	/* all libbpf APIs are usable */
388  	printf("my_map name: %s\n", bpf_map__name(skel->maps.my_map));
389  	printf("sys_enter prog FD: %d\n",
390  	       bpf_program__fd(skel->progs.handle_sys_enter));
391
392  	/* detach and re-attach sys_exit program */
393  	bpf_link__destroy(skel->links.handle_sys_exit);
394  	skel->links.handle_sys_exit =
395  		bpf_program__attach(skel->progs.handle_sys_exit);
396
397  	printf("my_static_var: %ld\n",
398  	       skel->bss->handle_sys_enter_my_static_var);
399
400  cleanup:
401  	example__destroy(skel);
402  	return err;
403  }
404
405**# ./example**
406
407::
408
409  test_2_result: 17
410  my_map name: my_map
411  sys_enter prog FD: 8
412  my_static_var: 7
413
414This is a stripped-out version of skeleton generated for above example code.
415
416min_core_btf
417------------
418
419**$ bpftool btf dump file 5.4.0-example.btf format raw**
420
421::
422
423  [1] INT 'long unsigned int' size=8 bits_offset=0 nr_bits=64 encoding=(none)
424  [2] CONST '(anon)' type_id=1
425  [3] VOLATILE '(anon)' type_id=1
426  [4] ARRAY '(anon)' type_id=1 index_type_id=21 nr_elems=2
427  [5] PTR '(anon)' type_id=8
428  [6] CONST '(anon)' type_id=5
429  [7] INT 'char' size=1 bits_offset=0 nr_bits=8 encoding=(none)
430  [8] CONST '(anon)' type_id=7
431  [9] INT 'unsigned int' size=4 bits_offset=0 nr_bits=32 encoding=(none)
432  <long output>
433
434**$ bpftool btf dump file one.bpf.o format raw**
435
436::
437
438  [1] PTR '(anon)' type_id=2
439  [2] STRUCT 'trace_event_raw_sys_enter' size=64 vlen=4
440        'ent' type_id=3 bits_offset=0
441        'id' type_id=7 bits_offset=64
442        'args' type_id=9 bits_offset=128
443        '__data' type_id=12 bits_offset=512
444  [3] STRUCT 'trace_entry' size=8 vlen=4
445        'type' type_id=4 bits_offset=0
446        'flags' type_id=5 bits_offset=16
447        'preempt_count' type_id=5 bits_offset=24
448  <long output>
449
450**$ bpftool gen min_core_btf 5.4.0-example.btf 5.4.0-smaller.btf one.bpf.o**
451
452**$ bpftool btf dump file 5.4.0-smaller.btf format raw**
453
454::
455
456  [1] TYPEDEF 'pid_t' type_id=6
457  [2] STRUCT 'trace_event_raw_sys_enter' size=64 vlen=1
458        'args' type_id=4 bits_offset=128
459  [3] STRUCT 'task_struct' size=9216 vlen=2
460        'pid' type_id=1 bits_offset=17920
461        'real_parent' type_id=7 bits_offset=18048
462  [4] ARRAY '(anon)' type_id=5 index_type_id=8 nr_elems=6
463  [5] INT 'long unsigned int' size=8 bits_offset=0 nr_bits=64 encoding=(none)
464  [6] TYPEDEF '__kernel_pid_t' type_id=8
465  [7] PTR '(anon)' type_id=3
466  [8] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
467  <end>
468
469Now, the "5.4.0-smaller.btf" file may be used by libbpf as an external BTF file
470when loading the "one.bpf.o" object into the "5.4.0-example" kernel. Note that
471the generated BTF file won't allow other eBPF objects to be loaded, just the
472ones given to min_core_btf.
473
474::
475
476  LIBBPF_OPTS(bpf_object_open_opts, opts, .btf_custom_path = "5.4.0-smaller.btf");
477  struct bpf_object *obj;
478
479  obj = bpf_object__open_file("one.bpf.o", &opts);
480
481  ...
482