xref: /openbmc/linux/Documentation/bpf/btf.rst (revision 3ce7547e)
1=====================
2BPF Type Format (BTF)
3=====================
4
51. Introduction
6===============
7
8BTF (BPF Type Format) is the metadata format which encodes the debug info
9related to BPF program/map. The name BTF was used initially to describe data
10types. The BTF was later extended to include function info for defined
11subroutines, and line info for source/line information.
12
13The debug info is used for map pretty print, function signature, etc. The
14function signature enables better bpf program/function kernel symbol. The line
15info helps generate source annotated translated byte code, jited code and
16verifier log.
17
18The BTF specification contains two parts,
19  * BTF kernel API
20  * BTF ELF file format
21
22The kernel API is the contract between user space and kernel. The kernel
23verifies the BTF info before using it. The ELF file format is a user space
24contract between ELF file and libbpf loader.
25
26The type and string sections are part of the BTF kernel API, describing the
27debug info (mostly types related) referenced by the bpf program. These two
28sections are discussed in details in :ref:`BTF_Type_String`.
29
30.. _BTF_Type_String:
31
322. BTF Type and String Encoding
33===============================
34
35The file ``include/uapi/linux/btf.h`` provides high-level definition of how
36types/strings are encoded.
37
38The beginning of data blob must be::
39
40    struct btf_header {
41        __u16   magic;
42        __u8    version;
43        __u8    flags;
44        __u32   hdr_len;
45
46        /* All offsets are in bytes relative to the end of this header */
47        __u32   type_off;       /* offset of type section       */
48        __u32   type_len;       /* length of type section       */
49        __u32   str_off;        /* offset of string section     */
50        __u32   str_len;        /* length of string section     */
51    };
52
53The magic is ``0xeB9F``, which has different encoding for big and little
54endian systems, and can be used to test whether BTF is generated for big- or
55little-endian target. The ``btf_header`` is designed to be extensible with
56``hdr_len`` equal to ``sizeof(struct btf_header)`` when a data blob is
57generated.
58
592.1 String Encoding
60-------------------
61
62The first string in the string section must be a null string. The rest of
63string table is a concatenation of other null-terminated strings.
64
652.2 Type Encoding
66-----------------
67
68The type id ``0`` is reserved for ``void`` type. The type section is parsed
69sequentially and type id is assigned to each recognized type starting from id
70``1``. Currently, the following types are supported::
71
72    #define BTF_KIND_INT            1       /* Integer      */
73    #define BTF_KIND_PTR            2       /* Pointer      */
74    #define BTF_KIND_ARRAY          3       /* Array        */
75    #define BTF_KIND_STRUCT         4       /* Struct       */
76    #define BTF_KIND_UNION          5       /* Union        */
77    #define BTF_KIND_ENUM           6       /* Enumeration up to 32-bit values */
78    #define BTF_KIND_FWD            7       /* Forward      */
79    #define BTF_KIND_TYPEDEF        8       /* Typedef      */
80    #define BTF_KIND_VOLATILE       9       /* Volatile     */
81    #define BTF_KIND_CONST          10      /* Const        */
82    #define BTF_KIND_RESTRICT       11      /* Restrict     */
83    #define BTF_KIND_FUNC           12      /* Function     */
84    #define BTF_KIND_FUNC_PROTO     13      /* Function Proto       */
85    #define BTF_KIND_VAR            14      /* Variable     */
86    #define BTF_KIND_DATASEC        15      /* Section      */
87    #define BTF_KIND_FLOAT          16      /* Floating point       */
88    #define BTF_KIND_DECL_TAG       17      /* Decl Tag     */
89    #define BTF_KIND_TYPE_TAG       18      /* Type Tag     */
90    #define BTF_KIND_ENUM64         19      /* Enumeration up to 64-bit values */
91
92Note that the type section encodes debug info, not just pure types.
93``BTF_KIND_FUNC`` is not a type, and it represents a defined subprogram.
94
95Each type contains the following common data::
96
97    struct btf_type {
98        __u32 name_off;
99        /* "info" bits arrangement
100         * bits  0-15: vlen (e.g. # of struct's members)
101         * bits 16-23: unused
102         * bits 24-28: kind (e.g. int, ptr, array...etc)
103         * bits 29-30: unused
104         * bit     31: kind_flag, currently used by
105         *             struct, union, fwd, enum and enum64.
106         */
107        __u32 info;
108        /* "size" is used by INT, ENUM, STRUCT, UNION and ENUM64.
109         * "size" tells the size of the type it is describing.
110         *
111         * "type" is used by PTR, TYPEDEF, VOLATILE, CONST, RESTRICT,
112         * FUNC, FUNC_PROTO, DECL_TAG and TYPE_TAG.
113         * "type" is a type_id referring to another type.
114         */
115        union {
116                __u32 size;
117                __u32 type;
118        };
119    };
120
121For certain kinds, the common data are followed by kind-specific data. The
122``name_off`` in ``struct btf_type`` specifies the offset in the string table.
123The following sections detail encoding of each kind.
124
1252.2.1 BTF_KIND_INT
126~~~~~~~~~~~~~~~~~~
127
128``struct btf_type`` encoding requirement:
129 * ``name_off``: any valid offset
130 * ``info.kind_flag``: 0
131 * ``info.kind``: BTF_KIND_INT
132 * ``info.vlen``: 0
133 * ``size``: the size of the int type in bytes.
134
135``btf_type`` is followed by a ``u32`` with the following bits arrangement::
136
137  #define BTF_INT_ENCODING(VAL)   (((VAL) & 0x0f000000) >> 24)
138  #define BTF_INT_OFFSET(VAL)     (((VAL) & 0x00ff0000) >> 16)
139  #define BTF_INT_BITS(VAL)       ((VAL)  & 0x000000ff)
140
141The ``BTF_INT_ENCODING`` has the following attributes::
142
143  #define BTF_INT_SIGNED  (1 << 0)
144  #define BTF_INT_CHAR    (1 << 1)
145  #define BTF_INT_BOOL    (1 << 2)
146
147The ``BTF_INT_ENCODING()`` provides extra information: signedness, char, or
148bool, for the int type. The char and bool encoding are mostly useful for
149pretty print. At most one encoding can be specified for the int type.
150
151The ``BTF_INT_BITS()`` specifies the number of actual bits held by this int
152type. For example, a 4-bit bitfield encodes ``BTF_INT_BITS()`` equals to 4.
153The ``btf_type.size * 8`` must be equal to or greater than ``BTF_INT_BITS()``
154for the type. The maximum value of ``BTF_INT_BITS()`` is 128.
155
156The ``BTF_INT_OFFSET()`` specifies the starting bit offset to calculate values
157for this int. For example, a bitfield struct member has:
158
159 * btf member bit offset 100 from the start of the structure,
160 * btf member pointing to an int type,
161 * the int type has ``BTF_INT_OFFSET() = 2`` and ``BTF_INT_BITS() = 4``
162
163Then in the struct memory layout, this member will occupy ``4`` bits starting
164from bits ``100 + 2 = 102``.
165
166Alternatively, the bitfield struct member can be the following to access the
167same bits as the above:
168
169 * btf member bit offset 102,
170 * btf member pointing to an int type,
171 * the int type has ``BTF_INT_OFFSET() = 0`` and ``BTF_INT_BITS() = 4``
172
173The original intention of ``BTF_INT_OFFSET()`` is to provide flexibility of
174bitfield encoding. Currently, both llvm and pahole generate
175``BTF_INT_OFFSET() = 0`` for all int types.
176
1772.2.2 BTF_KIND_PTR
178~~~~~~~~~~~~~~~~~~
179
180``struct btf_type`` encoding requirement:
181  * ``name_off``: 0
182  * ``info.kind_flag``: 0
183  * ``info.kind``: BTF_KIND_PTR
184  * ``info.vlen``: 0
185  * ``type``: the pointee type of the pointer
186
187No additional type data follow ``btf_type``.
188
1892.2.3 BTF_KIND_ARRAY
190~~~~~~~~~~~~~~~~~~~~
191
192``struct btf_type`` encoding requirement:
193  * ``name_off``: 0
194  * ``info.kind_flag``: 0
195  * ``info.kind``: BTF_KIND_ARRAY
196  * ``info.vlen``: 0
197  * ``size/type``: 0, not used
198
199``btf_type`` is followed by one ``struct btf_array``::
200
201    struct btf_array {
202        __u32   type;
203        __u32   index_type;
204        __u32   nelems;
205    };
206
207The ``struct btf_array`` encoding:
208  * ``type``: the element type
209  * ``index_type``: the index type
210  * ``nelems``: the number of elements for this array (``0`` is also allowed).
211
212The ``index_type`` can be any regular int type (``u8``, ``u16``, ``u32``,
213``u64``, ``unsigned __int128``). The original design of including
214``index_type`` follows DWARF, which has an ``index_type`` for its array type.
215Currently in BTF, beyond type verification, the ``index_type`` is not used.
216
217The ``struct btf_array`` allows chaining through element type to represent
218multidimensional arrays. For example, for ``int a[5][6]``, the following type
219information illustrates the chaining:
220
221  * [1]: int
222  * [2]: array, ``btf_array.type = [1]``, ``btf_array.nelems = 6``
223  * [3]: array, ``btf_array.type = [2]``, ``btf_array.nelems = 5``
224
225Currently, both pahole and llvm collapse multidimensional array into
226one-dimensional array, e.g., for ``a[5][6]``, the ``btf_array.nelems`` is
227equal to ``30``. This is because the original use case is map pretty print
228where the whole array is dumped out so one-dimensional array is enough. As
229more BTF usage is explored, pahole and llvm can be changed to generate proper
230chained representation for multidimensional arrays.
231
2322.2.4 BTF_KIND_STRUCT
233~~~~~~~~~~~~~~~~~~~~~
2342.2.5 BTF_KIND_UNION
235~~~~~~~~~~~~~~~~~~~~
236
237``struct btf_type`` encoding requirement:
238  * ``name_off``: 0 or offset to a valid C identifier
239  * ``info.kind_flag``: 0 or 1
240  * ``info.kind``: BTF_KIND_STRUCT or BTF_KIND_UNION
241  * ``info.vlen``: the number of struct/union members
242  * ``info.size``: the size of the struct/union in bytes
243
244``btf_type`` is followed by ``info.vlen`` number of ``struct btf_member``.::
245
246    struct btf_member {
247        __u32   name_off;
248        __u32   type;
249        __u32   offset;
250    };
251
252``struct btf_member`` encoding:
253  * ``name_off``: offset to a valid C identifier
254  * ``type``: the member type
255  * ``offset``: <see below>
256
257If the type info ``kind_flag`` is not set, the offset contains only bit offset
258of the member. Note that the base type of the bitfield can only be int or enum
259type. If the bitfield size is 32, the base type can be either int or enum
260type. If the bitfield size is not 32, the base type must be int, and int type
261``BTF_INT_BITS()`` encodes the bitfield size.
262
263If the ``kind_flag`` is set, the ``btf_member.offset`` contains both member
264bitfield size and bit offset. The bitfield size and bit offset are calculated
265as below.::
266
267  #define BTF_MEMBER_BITFIELD_SIZE(val)   ((val) >> 24)
268  #define BTF_MEMBER_BIT_OFFSET(val)      ((val) & 0xffffff)
269
270In this case, if the base type is an int type, it must be a regular int type:
271
272  * ``BTF_INT_OFFSET()`` must be 0.
273  * ``BTF_INT_BITS()`` must be equal to ``{1,2,4,8,16} * 8``.
274
275The following kernel patch introduced ``kind_flag`` and explained why both
276modes exist:
277
278  https://github.com/torvalds/linux/commit/9d5f9f701b1891466fb3dbb1806ad97716f95cc3#diff-fa650a64fdd3968396883d2fe8215ff3
279
2802.2.6 BTF_KIND_ENUM
281~~~~~~~~~~~~~~~~~~~
282
283``struct btf_type`` encoding requirement:
284  * ``name_off``: 0 or offset to a valid C identifier
285  * ``info.kind_flag``: 0 for unsigned, 1 for signed
286  * ``info.kind``: BTF_KIND_ENUM
287  * ``info.vlen``: number of enum values
288  * ``size``: 1/2/4/8
289
290``btf_type`` is followed by ``info.vlen`` number of ``struct btf_enum``.::
291
292    struct btf_enum {
293        __u32   name_off;
294        __s32   val;
295    };
296
297The ``btf_enum`` encoding:
298  * ``name_off``: offset to a valid C identifier
299  * ``val``: any value
300
301If the original enum value is signed and the size is less than 4,
302that value will be sign extended into 4 bytes. If the size is 8,
303the value will be truncated into 4 bytes.
304
3052.2.7 BTF_KIND_FWD
306~~~~~~~~~~~~~~~~~~
307
308``struct btf_type`` encoding requirement:
309  * ``name_off``: offset to a valid C identifier
310  * ``info.kind_flag``: 0 for struct, 1 for union
311  * ``info.kind``: BTF_KIND_FWD
312  * ``info.vlen``: 0
313  * ``type``: 0
314
315No additional type data follow ``btf_type``.
316
3172.2.8 BTF_KIND_TYPEDEF
318~~~~~~~~~~~~~~~~~~~~~~
319
320``struct btf_type`` encoding requirement:
321  * ``name_off``: offset to a valid C identifier
322  * ``info.kind_flag``: 0
323  * ``info.kind``: BTF_KIND_TYPEDEF
324  * ``info.vlen``: 0
325  * ``type``: the type which can be referred by name at ``name_off``
326
327No additional type data follow ``btf_type``.
328
3292.2.9 BTF_KIND_VOLATILE
330~~~~~~~~~~~~~~~~~~~~~~~
331
332``struct btf_type`` encoding requirement:
333  * ``name_off``: 0
334  * ``info.kind_flag``: 0
335  * ``info.kind``: BTF_KIND_VOLATILE
336  * ``info.vlen``: 0
337  * ``type``: the type with ``volatile`` qualifier
338
339No additional type data follow ``btf_type``.
340
3412.2.10 BTF_KIND_CONST
342~~~~~~~~~~~~~~~~~~~~~
343
344``struct btf_type`` encoding requirement:
345  * ``name_off``: 0
346  * ``info.kind_flag``: 0
347  * ``info.kind``: BTF_KIND_CONST
348  * ``info.vlen``: 0
349  * ``type``: the type with ``const`` qualifier
350
351No additional type data follow ``btf_type``.
352
3532.2.11 BTF_KIND_RESTRICT
354~~~~~~~~~~~~~~~~~~~~~~~~
355
356``struct btf_type`` encoding requirement:
357  * ``name_off``: 0
358  * ``info.kind_flag``: 0
359  * ``info.kind``: BTF_KIND_RESTRICT
360  * ``info.vlen``: 0
361  * ``type``: the type with ``restrict`` qualifier
362
363No additional type data follow ``btf_type``.
364
3652.2.12 BTF_KIND_FUNC
366~~~~~~~~~~~~~~~~~~~~
367
368``struct btf_type`` encoding requirement:
369  * ``name_off``: offset to a valid C identifier
370  * ``info.kind_flag``: 0
371  * ``info.kind``: BTF_KIND_FUNC
372  * ``info.vlen``: 0
373  * ``type``: a BTF_KIND_FUNC_PROTO type
374
375No additional type data follow ``btf_type``.
376
377A BTF_KIND_FUNC defines not a type, but a subprogram (function) whose
378signature is defined by ``type``. The subprogram is thus an instance of that
379type. The BTF_KIND_FUNC may in turn be referenced by a func_info in the
380:ref:`BTF_Ext_Section` (ELF) or in the arguments to :ref:`BPF_Prog_Load`
381(ABI).
382
3832.2.13 BTF_KIND_FUNC_PROTO
384~~~~~~~~~~~~~~~~~~~~~~~~~~
385
386``struct btf_type`` encoding requirement:
387  * ``name_off``: 0
388  * ``info.kind_flag``: 0
389  * ``info.kind``: BTF_KIND_FUNC_PROTO
390  * ``info.vlen``: # of parameters
391  * ``type``: the return type
392
393``btf_type`` is followed by ``info.vlen`` number of ``struct btf_param``.::
394
395    struct btf_param {
396        __u32   name_off;
397        __u32   type;
398    };
399
400If a BTF_KIND_FUNC_PROTO type is referred by a BTF_KIND_FUNC type, then
401``btf_param.name_off`` must point to a valid C identifier except for the
402possible last argument representing the variable argument. The btf_param.type
403refers to parameter type.
404
405If the function has variable arguments, the last parameter is encoded with
406``name_off = 0`` and ``type = 0``.
407
4082.2.14 BTF_KIND_VAR
409~~~~~~~~~~~~~~~~~~~
410
411``struct btf_type`` encoding requirement:
412  * ``name_off``: offset to a valid C identifier
413  * ``info.kind_flag``: 0
414  * ``info.kind``: BTF_KIND_VAR
415  * ``info.vlen``: 0
416  * ``type``: the type of the variable
417
418``btf_type`` is followed by a single ``struct btf_variable`` with the
419following data::
420
421    struct btf_var {
422        __u32   linkage;
423    };
424
425``struct btf_var`` encoding:
426  * ``linkage``: currently only static variable 0, or globally allocated
427                 variable in ELF sections 1
428
429Not all type of global variables are supported by LLVM at this point.
430The following is currently available:
431
432  * static variables with or without section attributes
433  * global variables with section attributes
434
435The latter is for future extraction of map key/value type id's from a
436map definition.
437
4382.2.15 BTF_KIND_DATASEC
439~~~~~~~~~~~~~~~~~~~~~~~
440
441``struct btf_type`` encoding requirement:
442  * ``name_off``: offset to a valid name associated with a variable or
443                  one of .data/.bss/.rodata
444  * ``info.kind_flag``: 0
445  * ``info.kind``: BTF_KIND_DATASEC
446  * ``info.vlen``: # of variables
447  * ``size``: total section size in bytes (0 at compilation time, patched
448              to actual size by BPF loaders such as libbpf)
449
450``btf_type`` is followed by ``info.vlen`` number of ``struct btf_var_secinfo``.::
451
452    struct btf_var_secinfo {
453        __u32   type;
454        __u32   offset;
455        __u32   size;
456    };
457
458``struct btf_var_secinfo`` encoding:
459  * ``type``: the type of the BTF_KIND_VAR variable
460  * ``offset``: the in-section offset of the variable
461  * ``size``: the size of the variable in bytes
462
4632.2.16 BTF_KIND_FLOAT
464~~~~~~~~~~~~~~~~~~~~~
465
466``struct btf_type`` encoding requirement:
467 * ``name_off``: any valid offset
468 * ``info.kind_flag``: 0
469 * ``info.kind``: BTF_KIND_FLOAT
470 * ``info.vlen``: 0
471 * ``size``: the size of the float type in bytes: 2, 4, 8, 12 or 16.
472
473No additional type data follow ``btf_type``.
474
4752.2.17 BTF_KIND_DECL_TAG
476~~~~~~~~~~~~~~~~~~~~~~~~
477
478``struct btf_type`` encoding requirement:
479 * ``name_off``: offset to a non-empty string
480 * ``info.kind_flag``: 0
481 * ``info.kind``: BTF_KIND_DECL_TAG
482 * ``info.vlen``: 0
483 * ``type``: ``struct``, ``union``, ``func``, ``var`` or ``typedef``
484
485``btf_type`` is followed by ``struct btf_decl_tag``.::
486
487    struct btf_decl_tag {
488        __u32   component_idx;
489    };
490
491The ``name_off`` encodes btf_decl_tag attribute string.
492The ``type`` should be ``struct``, ``union``, ``func``, ``var`` or ``typedef``.
493For ``var`` or ``typedef`` type, ``btf_decl_tag.component_idx`` must be ``-1``.
494For the other three types, if the btf_decl_tag attribute is
495applied to the ``struct``, ``union`` or ``func`` itself,
496``btf_decl_tag.component_idx`` must be ``-1``. Otherwise,
497the attribute is applied to a ``struct``/``union`` member or
498a ``func`` argument, and ``btf_decl_tag.component_idx`` should be a
499valid index (starting from 0) pointing to a member or an argument.
500
5012.2.18 BTF_KIND_TYPE_TAG
502~~~~~~~~~~~~~~~~~~~~~~~~
503
504``struct btf_type`` encoding requirement:
505 * ``name_off``: offset to a non-empty string
506 * ``info.kind_flag``: 0
507 * ``info.kind``: BTF_KIND_TYPE_TAG
508 * ``info.vlen``: 0
509 * ``type``: the type with ``btf_type_tag`` attribute
510
511Currently, ``BTF_KIND_TYPE_TAG`` is only emitted for pointer types.
512It has the following btf type chain:
513::
514
515  ptr -> [type_tag]*
516      -> [const | volatile | restrict | typedef]*
517      -> base_type
518
519Basically, a pointer type points to zero or more
520type_tag, then zero or more const/volatile/restrict/typedef
521and finally the base type. The base type is one of
522int, ptr, array, struct, union, enum, func_proto and float types.
523
5242.2.19 BTF_KIND_ENUM64
525~~~~~~~~~~~~~~~~~~~~~~
526
527``struct btf_type`` encoding requirement:
528  * ``name_off``: 0 or offset to a valid C identifier
529  * ``info.kind_flag``: 0 for unsigned, 1 for signed
530  * ``info.kind``: BTF_KIND_ENUM64
531  * ``info.vlen``: number of enum values
532  * ``size``: 1/2/4/8
533
534``btf_type`` is followed by ``info.vlen`` number of ``struct btf_enum64``.::
535
536    struct btf_enum64 {
537        __u32   name_off;
538        __u32   val_lo32;
539        __u32   val_hi32;
540    };
541
542The ``btf_enum64`` encoding:
543  * ``name_off``: offset to a valid C identifier
544  * ``val_lo32``: lower 32-bit value for a 64-bit value
545  * ``val_hi32``: high 32-bit value for a 64-bit value
546
547If the original enum value is signed and the size is less than 8,
548that value will be sign extended into 8 bytes.
549
5503. BTF Kernel API
551=================
552
553The following bpf syscall command involves BTF:
554   * BPF_BTF_LOAD: load a blob of BTF data into kernel
555   * BPF_MAP_CREATE: map creation with btf key and value type info.
556   * BPF_PROG_LOAD: prog load with btf function and line info.
557   * BPF_BTF_GET_FD_BY_ID: get a btf fd
558   * BPF_OBJ_GET_INFO_BY_FD: btf, func_info, line_info
559     and other btf related info are returned.
560
561The workflow typically looks like:
562::
563
564  Application:
565      BPF_BTF_LOAD
566          |
567          v
568      BPF_MAP_CREATE and BPF_PROG_LOAD
569          |
570          V
571      ......
572
573  Introspection tool:
574      ......
575      BPF_{PROG,MAP}_GET_NEXT_ID (get prog/map id's)
576          |
577          V
578      BPF_{PROG,MAP}_GET_FD_BY_ID (get a prog/map fd)
579          |
580          V
581      BPF_OBJ_GET_INFO_BY_FD (get bpf_prog_info/bpf_map_info with btf_id)
582          |                                     |
583          V                                     |
584      BPF_BTF_GET_FD_BY_ID (get btf_fd)         |
585          |                                     |
586          V                                     |
587      BPF_OBJ_GET_INFO_BY_FD (get btf)          |
588          |                                     |
589          V                                     V
590      pretty print types, dump func signatures and line info, etc.
591
592
5933.1 BPF_BTF_LOAD
594----------------
595
596Load a blob of BTF data into kernel. A blob of data, described in
597:ref:`BTF_Type_String`, can be directly loaded into the kernel. A ``btf_fd``
598is returned to a userspace.
599
6003.2 BPF_MAP_CREATE
601------------------
602
603A map can be created with ``btf_fd`` and specified key/value type id.::
604
605    __u32   btf_fd;         /* fd pointing to a BTF type data */
606    __u32   btf_key_type_id;        /* BTF type_id of the key */
607    __u32   btf_value_type_id;      /* BTF type_id of the value */
608
609In libbpf, the map can be defined with extra annotation like below:
610::
611
612    struct {
613        __uint(type, BPF_MAP_TYPE_ARRAY);
614        __type(key, int);
615        __type(value, struct ipv_counts);
616        __uint(max_entries, 4);
617    } btf_map SEC(".maps");
618
619During ELF parsing, libbpf is able to extract key/value type_id's and assign
620them to BPF_MAP_CREATE attributes automatically.
621
622.. _BPF_Prog_Load:
623
6243.3 BPF_PROG_LOAD
625-----------------
626
627During prog_load, func_info and line_info can be passed to kernel with proper
628values for the following attributes:
629::
630
631    __u32           insn_cnt;
632    __aligned_u64   insns;
633    ......
634    __u32           prog_btf_fd;    /* fd pointing to BTF type data */
635    __u32           func_info_rec_size;     /* userspace bpf_func_info size */
636    __aligned_u64   func_info;      /* func info */
637    __u32           func_info_cnt;  /* number of bpf_func_info records */
638    __u32           line_info_rec_size;     /* userspace bpf_line_info size */
639    __aligned_u64   line_info;      /* line info */
640    __u32           line_info_cnt;  /* number of bpf_line_info records */
641
642The func_info and line_info are an array of below, respectively.::
643
644    struct bpf_func_info {
645        __u32   insn_off; /* [0, insn_cnt - 1] */
646        __u32   type_id;  /* pointing to a BTF_KIND_FUNC type */
647    };
648    struct bpf_line_info {
649        __u32   insn_off; /* [0, insn_cnt - 1] */
650        __u32   file_name_off; /* offset to string table for the filename */
651        __u32   line_off; /* offset to string table for the source line */
652        __u32   line_col; /* line number and column number */
653    };
654
655func_info_rec_size is the size of each func_info record, and
656line_info_rec_size is the size of each line_info record. Passing the record
657size to kernel make it possible to extend the record itself in the future.
658
659Below are requirements for func_info:
660  * func_info[0].insn_off must be 0.
661  * the func_info insn_off is in strictly increasing order and matches
662    bpf func boundaries.
663
664Below are requirements for line_info:
665  * the first insn in each func must have a line_info record pointing to it.
666  * the line_info insn_off is in strictly increasing order.
667
668For line_info, the line number and column number are defined as below:
669::
670
671    #define BPF_LINE_INFO_LINE_NUM(line_col)        ((line_col) >> 10)
672    #define BPF_LINE_INFO_LINE_COL(line_col)        ((line_col) & 0x3ff)
673
6743.4 BPF_{PROG,MAP}_GET_NEXT_ID
675------------------------------
676
677In kernel, every loaded program, map or btf has a unique id. The id won't
678change during the lifetime of a program, map, or btf.
679
680The bpf syscall command BPF_{PROG,MAP}_GET_NEXT_ID returns all id's, one for
681each command, to user space, for bpf program or maps, respectively, so an
682inspection tool can inspect all programs and maps.
683
6843.5 BPF_{PROG,MAP}_GET_FD_BY_ID
685-------------------------------
686
687An introspection tool cannot use id to get details about program or maps.
688A file descriptor needs to be obtained first for reference-counting purpose.
689
6903.6 BPF_OBJ_GET_INFO_BY_FD
691--------------------------
692
693Once a program/map fd is acquired, an introspection tool can get the detailed
694information from kernel about this fd, some of which are BTF-related. For
695example, ``bpf_map_info`` returns ``btf_id`` and key/value type ids.
696``bpf_prog_info`` returns ``btf_id``, func_info, and line info for translated
697bpf byte codes, and jited_line_info.
698
6993.7 BPF_BTF_GET_FD_BY_ID
700------------------------
701
702With ``btf_id`` obtained in ``bpf_map_info`` and ``bpf_prog_info``, bpf
703syscall command BPF_BTF_GET_FD_BY_ID can retrieve a btf fd. Then, with
704command BPF_OBJ_GET_INFO_BY_FD, the btf blob, originally loaded into the
705kernel with BPF_BTF_LOAD, can be retrieved.
706
707With the btf blob, ``bpf_map_info``, and ``bpf_prog_info``, an introspection
708tool has full btf knowledge and is able to pretty print map key/values, dump
709func signatures and line info, along with byte/jit codes.
710
7114. ELF File Format Interface
712============================
713
7144.1 .BTF section
715----------------
716
717The .BTF section contains type and string data. The format of this section is
718same as the one describe in :ref:`BTF_Type_String`.
719
720.. _BTF_Ext_Section:
721
7224.2 .BTF.ext section
723--------------------
724
725The .BTF.ext section encodes func_info and line_info which needs loader
726manipulation before loading into the kernel.
727
728The specification for .BTF.ext section is defined at ``tools/lib/bpf/btf.h``
729and ``tools/lib/bpf/btf.c``.
730
731The current header of .BTF.ext section::
732
733    struct btf_ext_header {
734        __u16   magic;
735        __u8    version;
736        __u8    flags;
737        __u32   hdr_len;
738
739        /* All offsets are in bytes relative to the end of this header */
740        __u32   func_info_off;
741        __u32   func_info_len;
742        __u32   line_info_off;
743        __u32   line_info_len;
744    };
745
746It is very similar to .BTF section. Instead of type/string section, it
747contains func_info and line_info section. See :ref:`BPF_Prog_Load` for details
748about func_info and line_info record format.
749
750The func_info is organized as below.::
751
752     func_info_rec_size
753     btf_ext_info_sec for section #1 /* func_info for section #1 */
754     btf_ext_info_sec for section #2 /* func_info for section #2 */
755     ...
756
757``func_info_rec_size`` specifies the size of ``bpf_func_info`` structure when
758.BTF.ext is generated. ``btf_ext_info_sec``, defined below, is a collection of
759func_info for each specific ELF section.::
760
761     struct btf_ext_info_sec {
762        __u32   sec_name_off; /* offset to section name */
763        __u32   num_info;
764        /* Followed by num_info * record_size number of bytes */
765        __u8    data[0];
766     };
767
768Here, num_info must be greater than 0.
769
770The line_info is organized as below.::
771
772     line_info_rec_size
773     btf_ext_info_sec for section #1 /* line_info for section #1 */
774     btf_ext_info_sec for section #2 /* line_info for section #2 */
775     ...
776
777``line_info_rec_size`` specifies the size of ``bpf_line_info`` structure when
778.BTF.ext is generated.
779
780The interpretation of ``bpf_func_info->insn_off`` and
781``bpf_line_info->insn_off`` is different between kernel API and ELF API. For
782kernel API, the ``insn_off`` is the instruction offset in the unit of ``struct
783bpf_insn``. For ELF API, the ``insn_off`` is the byte offset from the
784beginning of section (``btf_ext_info_sec->sec_name_off``).
785
7864.2 .BTF_ids section
787--------------------
788
789The .BTF_ids section encodes BTF ID values that are used within the kernel.
790
791This section is created during the kernel compilation with the help of
792macros defined in ``include/linux/btf_ids.h`` header file. Kernel code can
793use them to create lists and sets (sorted lists) of BTF ID values.
794
795The ``BTF_ID_LIST`` and ``BTF_ID`` macros define unsorted list of BTF ID values,
796with following syntax::
797
798  BTF_ID_LIST(list)
799  BTF_ID(type1, name1)
800  BTF_ID(type2, name2)
801
802resulting in following layout in .BTF_ids section::
803
804  __BTF_ID__type1__name1__1:
805  .zero 4
806  __BTF_ID__type2__name2__2:
807  .zero 4
808
809The ``u32 list[];`` variable is defined to access the list.
810
811The ``BTF_ID_UNUSED`` macro defines 4 zero bytes. It's used when we
812want to define unused entry in BTF_ID_LIST, like::
813
814      BTF_ID_LIST(bpf_skb_output_btf_ids)
815      BTF_ID(struct, sk_buff)
816      BTF_ID_UNUSED
817      BTF_ID(struct, task_struct)
818
819The ``BTF_SET_START/END`` macros pair defines sorted list of BTF ID values
820and their count, with following syntax::
821
822  BTF_SET_START(set)
823  BTF_ID(type1, name1)
824  BTF_ID(type2, name2)
825  BTF_SET_END(set)
826
827resulting in following layout in .BTF_ids section::
828
829  __BTF_ID__set__set:
830  .zero 4
831  __BTF_ID__type1__name1__3:
832  .zero 4
833  __BTF_ID__type2__name2__4:
834  .zero 4
835
836The ``struct btf_id_set set;`` variable is defined to access the list.
837
838The ``typeX`` name can be one of following::
839
840   struct, union, typedef, func
841
842and is used as a filter when resolving the BTF ID value.
843
844All the BTF ID lists and sets are compiled in the .BTF_ids section and
845resolved during the linking phase of kernel build by ``resolve_btfids`` tool.
846
8475. Using BTF
848============
849
8505.1 bpftool map pretty print
851----------------------------
852
853With BTF, the map key/value can be printed based on fields rather than simply
854raw bytes. This is especially valuable for large structure or if your data
855structure has bitfields. For example, for the following map,::
856
857      enum A { A1, A2, A3, A4, A5 };
858      typedef enum A ___A;
859      struct tmp_t {
860           char a1:4;
861           int  a2:4;
862           int  :4;
863           __u32 a3:4;
864           int b;
865           ___A b1:4;
866           enum A b2:4;
867      };
868      struct {
869           __uint(type, BPF_MAP_TYPE_ARRAY);
870           __type(key, int);
871           __type(value, struct tmp_t);
872           __uint(max_entries, 1);
873      } tmpmap SEC(".maps");
874
875bpftool is able to pretty print like below:
876::
877
878      [{
879            "key": 0,
880            "value": {
881                "a1": 0x2,
882                "a2": 0x4,
883                "a3": 0x6,
884                "b": 7,
885                "b1": 0x8,
886                "b2": 0xa
887            }
888        }
889      ]
890
8915.2 bpftool prog dump
892---------------------
893
894The following is an example showing how func_info and line_info can help prog
895dump with better kernel symbol names, function prototypes and line
896information.::
897
898    $ bpftool prog dump jited pinned /sys/fs/bpf/test_btf_haskv
899    [...]
900    int test_long_fname_2(struct dummy_tracepoint_args * arg):
901    bpf_prog_44a040bf25481309_test_long_fname_2:
902    ; static int test_long_fname_2(struct dummy_tracepoint_args *arg)
903       0:   push   %rbp
904       1:   mov    %rsp,%rbp
905       4:   sub    $0x30,%rsp
906       b:   sub    $0x28,%rbp
907       f:   mov    %rbx,0x0(%rbp)
908      13:   mov    %r13,0x8(%rbp)
909      17:   mov    %r14,0x10(%rbp)
910      1b:   mov    %r15,0x18(%rbp)
911      1f:   xor    %eax,%eax
912      21:   mov    %rax,0x20(%rbp)
913      25:   xor    %esi,%esi
914    ; int key = 0;
915      27:   mov    %esi,-0x4(%rbp)
916    ; if (!arg->sock)
917      2a:   mov    0x8(%rdi),%rdi
918    ; if (!arg->sock)
919      2e:   cmp    $0x0,%rdi
920      32:   je     0x0000000000000070
921      34:   mov    %rbp,%rsi
922    ; counts = bpf_map_lookup_elem(&btf_map, &key);
923    [...]
924
9255.3 Verifier Log
926----------------
927
928The following is an example of how line_info can help debugging verification
929failure.::
930
931       /* The code at tools/testing/selftests/bpf/test_xdp_noinline.c
932        * is modified as below.
933        */
934       data = (void *)(long)xdp->data;
935       data_end = (void *)(long)xdp->data_end;
936       /*
937       if (data + 4 > data_end)
938               return XDP_DROP;
939       */
940       *(u32 *)data = dst->dst;
941
942    $ bpftool prog load ./test_xdp_noinline.o /sys/fs/bpf/test_xdp_noinline type xdp
943        ; data = (void *)(long)xdp->data;
944        224: (79) r2 = *(u64 *)(r10 -112)
945        225: (61) r2 = *(u32 *)(r2 +0)
946        ; *(u32 *)data = dst->dst;
947        226: (63) *(u32 *)(r2 +0) = r1
948        invalid access to packet, off=0 size=4, R2(id=0,off=0,r=0)
949        R2 offset is outside of the packet
950
9516. BTF Generation
952=================
953
954You need latest pahole
955
956  https://git.kernel.org/pub/scm/devel/pahole/pahole.git/
957
958or llvm (8.0 or later). The pahole acts as a dwarf2btf converter. It doesn't
959support .BTF.ext and btf BTF_KIND_FUNC type yet. For example,::
960
961      -bash-4.4$ cat t.c
962      struct t {
963        int a:2;
964        int b:3;
965        int c:2;
966      } g;
967      -bash-4.4$ gcc -c -O2 -g t.c
968      -bash-4.4$ pahole -JV t.o
969      File t.o:
970      [1] STRUCT t kind_flag=1 size=4 vlen=3
971              a type_id=2 bitfield_size=2 bits_offset=0
972              b type_id=2 bitfield_size=3 bits_offset=2
973              c type_id=2 bitfield_size=2 bits_offset=5
974      [2] INT int size=4 bit_offset=0 nr_bits=32 encoding=SIGNED
975
976The llvm is able to generate .BTF and .BTF.ext directly with -g for bpf target
977only. The assembly code (-S) is able to show the BTF encoding in assembly
978format.::
979
980    -bash-4.4$ cat t2.c
981    typedef int __int32;
982    struct t2 {
983      int a2;
984      int (*f2)(char q1, __int32 q2, ...);
985      int (*f3)();
986    } g2;
987    int main() { return 0; }
988    int test() { return 0; }
989    -bash-4.4$ clang -c -g -O2 -target bpf t2.c
990    -bash-4.4$ readelf -S t2.o
991      ......
992      [ 8] .BTF              PROGBITS         0000000000000000  00000247
993           000000000000016e  0000000000000000           0     0     1
994      [ 9] .BTF.ext          PROGBITS         0000000000000000  000003b5
995           0000000000000060  0000000000000000           0     0     1
996      [10] .rel.BTF.ext      REL              0000000000000000  000007e0
997           0000000000000040  0000000000000010          16     9     8
998      ......
999    -bash-4.4$ clang -S -g -O2 -target bpf t2.c
1000    -bash-4.4$ cat t2.s
1001      ......
1002            .section        .BTF,"",@progbits
1003            .short  60319                   # 0xeb9f
1004            .byte   1
1005            .byte   0
1006            .long   24
1007            .long   0
1008            .long   220
1009            .long   220
1010            .long   122
1011            .long   0                       # BTF_KIND_FUNC_PROTO(id = 1)
1012            .long   218103808               # 0xd000000
1013            .long   2
1014            .long   83                      # BTF_KIND_INT(id = 2)
1015            .long   16777216                # 0x1000000
1016            .long   4
1017            .long   16777248                # 0x1000020
1018      ......
1019            .byte   0                       # string offset=0
1020            .ascii  ".text"                 # string offset=1
1021            .byte   0
1022            .ascii  "/home/yhs/tmp-pahole/t2.c" # string offset=7
1023            .byte   0
1024            .ascii  "int main() { return 0; }" # string offset=33
1025            .byte   0
1026            .ascii  "int test() { return 0; }" # string offset=58
1027            .byte   0
1028            .ascii  "int"                   # string offset=83
1029      ......
1030            .section        .BTF.ext,"",@progbits
1031            .short  60319                   # 0xeb9f
1032            .byte   1
1033            .byte   0
1034            .long   24
1035            .long   0
1036            .long   28
1037            .long   28
1038            .long   44
1039            .long   8                       # FuncInfo
1040            .long   1                       # FuncInfo section string offset=1
1041            .long   2
1042            .long   .Lfunc_begin0
1043            .long   3
1044            .long   .Lfunc_begin1
1045            .long   5
1046            .long   16                      # LineInfo
1047            .long   1                       # LineInfo section string offset=1
1048            .long   2
1049            .long   .Ltmp0
1050            .long   7
1051            .long   33
1052            .long   7182                    # Line 7 Col 14
1053            .long   .Ltmp3
1054            .long   7
1055            .long   58
1056            .long   8206                    # Line 8 Col 14
1057
10587. Testing
1059==========
1060
1061Kernel bpf selftest `test_btf.c` provides extensive set of BTF-related tests.
1062