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 */ 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 91Note that the type section encodes debug info, not just pure types. 92``BTF_KIND_FUNC`` is not a type, and it represents a defined subprogram. 93 94Each type contains the following common data:: 95 96 struct btf_type { 97 __u32 name_off; 98 /* "info" bits arrangement 99 * bits 0-15: vlen (e.g. # of struct's members) 100 * bits 16-23: unused 101 * bits 24-28: kind (e.g. int, ptr, array...etc) 102 * bits 29-30: unused 103 * bit 31: kind_flag, currently used by 104 * struct, union and fwd 105 */ 106 __u32 info; 107 /* "size" is used by INT, ENUM, STRUCT and UNION. 108 * "size" tells the size of the type it is describing. 109 * 110 * "type" is used by PTR, TYPEDEF, VOLATILE, CONST, RESTRICT, 111 * FUNC, FUNC_PROTO, DECL_TAG and TYPE_TAG. 112 * "type" is a type_id referring to another type. 113 */ 114 union { 115 __u32 size; 116 __u32 type; 117 }; 118 }; 119 120For certain kinds, the common data are followed by kind-specific data. The 121``name_off`` in ``struct btf_type`` specifies the offset in the string table. 122The following sections detail encoding of each kind. 123 1242.2.1 BTF_KIND_INT 125~~~~~~~~~~~~~~~~~~ 126 127``struct btf_type`` encoding requirement: 128 * ``name_off``: any valid offset 129 * ``info.kind_flag``: 0 130 * ``info.kind``: BTF_KIND_INT 131 * ``info.vlen``: 0 132 * ``size``: the size of the int type in bytes. 133 134``btf_type`` is followed by a ``u32`` with the following bits arrangement:: 135 136 #define BTF_INT_ENCODING(VAL) (((VAL) & 0x0f000000) >> 24) 137 #define BTF_INT_OFFSET(VAL) (((VAL) & 0x00ff0000) >> 16) 138 #define BTF_INT_BITS(VAL) ((VAL) & 0x000000ff) 139 140The ``BTF_INT_ENCODING`` has the following attributes:: 141 142 #define BTF_INT_SIGNED (1 << 0) 143 #define BTF_INT_CHAR (1 << 1) 144 #define BTF_INT_BOOL (1 << 2) 145 146The ``BTF_INT_ENCODING()`` provides extra information: signedness, char, or 147bool, for the int type. The char and bool encoding are mostly useful for 148pretty print. At most one encoding can be specified for the int type. 149 150The ``BTF_INT_BITS()`` specifies the number of actual bits held by this int 151type. For example, a 4-bit bitfield encodes ``BTF_INT_BITS()`` equals to 4. 152The ``btf_type.size * 8`` must be equal to or greater than ``BTF_INT_BITS()`` 153for the type. The maximum value of ``BTF_INT_BITS()`` is 128. 154 155The ``BTF_INT_OFFSET()`` specifies the starting bit offset to calculate values 156for this int. For example, a bitfield struct member has: 157 158 * btf member bit offset 100 from the start of the structure, 159 * btf member pointing to an int type, 160 * the int type has ``BTF_INT_OFFSET() = 2`` and ``BTF_INT_BITS() = 4`` 161 162Then in the struct memory layout, this member will occupy ``4`` bits starting 163from bits ``100 + 2 = 102``. 164 165Alternatively, the bitfield struct member can be the following to access the 166same bits as the above: 167 168 * btf member bit offset 102, 169 * btf member pointing to an int type, 170 * the int type has ``BTF_INT_OFFSET() = 0`` and ``BTF_INT_BITS() = 4`` 171 172The original intention of ``BTF_INT_OFFSET()`` is to provide flexibility of 173bitfield encoding. Currently, both llvm and pahole generate 174``BTF_INT_OFFSET() = 0`` for all int types. 175 1762.2.2 BTF_KIND_PTR 177~~~~~~~~~~~~~~~~~~ 178 179``struct btf_type`` encoding requirement: 180 * ``name_off``: 0 181 * ``info.kind_flag``: 0 182 * ``info.kind``: BTF_KIND_PTR 183 * ``info.vlen``: 0 184 * ``type``: the pointee type of the pointer 185 186No additional type data follow ``btf_type``. 187 1882.2.3 BTF_KIND_ARRAY 189~~~~~~~~~~~~~~~~~~~~ 190 191``struct btf_type`` encoding requirement: 192 * ``name_off``: 0 193 * ``info.kind_flag``: 0 194 * ``info.kind``: BTF_KIND_ARRAY 195 * ``info.vlen``: 0 196 * ``size/type``: 0, not used 197 198``btf_type`` is followed by one ``struct btf_array``:: 199 200 struct btf_array { 201 __u32 type; 202 __u32 index_type; 203 __u32 nelems; 204 }; 205 206The ``struct btf_array`` encoding: 207 * ``type``: the element type 208 * ``index_type``: the index type 209 * ``nelems``: the number of elements for this array (``0`` is also allowed). 210 211The ``index_type`` can be any regular int type (``u8``, ``u16``, ``u32``, 212``u64``, ``unsigned __int128``). The original design of including 213``index_type`` follows DWARF, which has an ``index_type`` for its array type. 214Currently in BTF, beyond type verification, the ``index_type`` is not used. 215 216The ``struct btf_array`` allows chaining through element type to represent 217multidimensional arrays. For example, for ``int a[5][6]``, the following type 218information illustrates the chaining: 219 220 * [1]: int 221 * [2]: array, ``btf_array.type = [1]``, ``btf_array.nelems = 6`` 222 * [3]: array, ``btf_array.type = [2]``, ``btf_array.nelems = 5`` 223 224Currently, both pahole and llvm collapse multidimensional array into 225one-dimensional array, e.g., for ``a[5][6]``, the ``btf_array.nelems`` is 226equal to ``30``. This is because the original use case is map pretty print 227where the whole array is dumped out so one-dimensional array is enough. As 228more BTF usage is explored, pahole and llvm can be changed to generate proper 229chained representation for multidimensional arrays. 230 2312.2.4 BTF_KIND_STRUCT 232~~~~~~~~~~~~~~~~~~~~~ 2332.2.5 BTF_KIND_UNION 234~~~~~~~~~~~~~~~~~~~~ 235 236``struct btf_type`` encoding requirement: 237 * ``name_off``: 0 or offset to a valid C identifier 238 * ``info.kind_flag``: 0 or 1 239 * ``info.kind``: BTF_KIND_STRUCT or BTF_KIND_UNION 240 * ``info.vlen``: the number of struct/union members 241 * ``info.size``: the size of the struct/union in bytes 242 243``btf_type`` is followed by ``info.vlen`` number of ``struct btf_member``.:: 244 245 struct btf_member { 246 __u32 name_off; 247 __u32 type; 248 __u32 offset; 249 }; 250 251``struct btf_member`` encoding: 252 * ``name_off``: offset to a valid C identifier 253 * ``type``: the member type 254 * ``offset``: <see below> 255 256If the type info ``kind_flag`` is not set, the offset contains only bit offset 257of the member. Note that the base type of the bitfield can only be int or enum 258type. If the bitfield size is 32, the base type can be either int or enum 259type. If the bitfield size is not 32, the base type must be int, and int type 260``BTF_INT_BITS()`` encodes the bitfield size. 261 262If the ``kind_flag`` is set, the ``btf_member.offset`` contains both member 263bitfield size and bit offset. The bitfield size and bit offset are calculated 264as below.:: 265 266 #define BTF_MEMBER_BITFIELD_SIZE(val) ((val) >> 24) 267 #define BTF_MEMBER_BIT_OFFSET(val) ((val) & 0xffffff) 268 269In this case, if the base type is an int type, it must be a regular int type: 270 271 * ``BTF_INT_OFFSET()`` must be 0. 272 * ``BTF_INT_BITS()`` must be equal to ``{1,2,4,8,16} * 8``. 273 274The following kernel patch introduced ``kind_flag`` and explained why both 275modes exist: 276 277 https://github.com/torvalds/linux/commit/9d5f9f701b1891466fb3dbb1806ad97716f95cc3#diff-fa650a64fdd3968396883d2fe8215ff3 278 2792.2.6 BTF_KIND_ENUM 280~~~~~~~~~~~~~~~~~~~ 281 282``struct btf_type`` encoding requirement: 283 * ``name_off``: 0 or offset to a valid C identifier 284 * ``info.kind_flag``: 0 285 * ``info.kind``: BTF_KIND_ENUM 286 * ``info.vlen``: number of enum values 287 * ``size``: 4 288 289``btf_type`` is followed by ``info.vlen`` number of ``struct btf_enum``.:: 290 291 struct btf_enum { 292 __u32 name_off; 293 __s32 val; 294 }; 295 296The ``btf_enum`` encoding: 297 * ``name_off``: offset to a valid C identifier 298 * ``val``: any value 299 3002.2.7 BTF_KIND_FWD 301~~~~~~~~~~~~~~~~~~ 302 303``struct btf_type`` encoding requirement: 304 * ``name_off``: offset to a valid C identifier 305 * ``info.kind_flag``: 0 for struct, 1 for union 306 * ``info.kind``: BTF_KIND_FWD 307 * ``info.vlen``: 0 308 * ``type``: 0 309 310No additional type data follow ``btf_type``. 311 3122.2.8 BTF_KIND_TYPEDEF 313~~~~~~~~~~~~~~~~~~~~~~ 314 315``struct btf_type`` encoding requirement: 316 * ``name_off``: offset to a valid C identifier 317 * ``info.kind_flag``: 0 318 * ``info.kind``: BTF_KIND_TYPEDEF 319 * ``info.vlen``: 0 320 * ``type``: the type which can be referred by name at ``name_off`` 321 322No additional type data follow ``btf_type``. 323 3242.2.9 BTF_KIND_VOLATILE 325~~~~~~~~~~~~~~~~~~~~~~~ 326 327``struct btf_type`` encoding requirement: 328 * ``name_off``: 0 329 * ``info.kind_flag``: 0 330 * ``info.kind``: BTF_KIND_VOLATILE 331 * ``info.vlen``: 0 332 * ``type``: the type with ``volatile`` qualifier 333 334No additional type data follow ``btf_type``. 335 3362.2.10 BTF_KIND_CONST 337~~~~~~~~~~~~~~~~~~~~~ 338 339``struct btf_type`` encoding requirement: 340 * ``name_off``: 0 341 * ``info.kind_flag``: 0 342 * ``info.kind``: BTF_KIND_CONST 343 * ``info.vlen``: 0 344 * ``type``: the type with ``const`` qualifier 345 346No additional type data follow ``btf_type``. 347 3482.2.11 BTF_KIND_RESTRICT 349~~~~~~~~~~~~~~~~~~~~~~~~ 350 351``struct btf_type`` encoding requirement: 352 * ``name_off``: 0 353 * ``info.kind_flag``: 0 354 * ``info.kind``: BTF_KIND_RESTRICT 355 * ``info.vlen``: 0 356 * ``type``: the type with ``restrict`` qualifier 357 358No additional type data follow ``btf_type``. 359 3602.2.12 BTF_KIND_FUNC 361~~~~~~~~~~~~~~~~~~~~ 362 363``struct btf_type`` encoding requirement: 364 * ``name_off``: offset to a valid C identifier 365 * ``info.kind_flag``: 0 366 * ``info.kind``: BTF_KIND_FUNC 367 * ``info.vlen``: 0 368 * ``type``: a BTF_KIND_FUNC_PROTO type 369 370No additional type data follow ``btf_type``. 371 372A BTF_KIND_FUNC defines not a type, but a subprogram (function) whose 373signature is defined by ``type``. The subprogram is thus an instance of that 374type. The BTF_KIND_FUNC may in turn be referenced by a func_info in the 375:ref:`BTF_Ext_Section` (ELF) or in the arguments to :ref:`BPF_Prog_Load` 376(ABI). 377 3782.2.13 BTF_KIND_FUNC_PROTO 379~~~~~~~~~~~~~~~~~~~~~~~~~~ 380 381``struct btf_type`` encoding requirement: 382 * ``name_off``: 0 383 * ``info.kind_flag``: 0 384 * ``info.kind``: BTF_KIND_FUNC_PROTO 385 * ``info.vlen``: # of parameters 386 * ``type``: the return type 387 388``btf_type`` is followed by ``info.vlen`` number of ``struct btf_param``.:: 389 390 struct btf_param { 391 __u32 name_off; 392 __u32 type; 393 }; 394 395If a BTF_KIND_FUNC_PROTO type is referred by a BTF_KIND_FUNC type, then 396``btf_param.name_off`` must point to a valid C identifier except for the 397possible last argument representing the variable argument. The btf_param.type 398refers to parameter type. 399 400If the function has variable arguments, the last parameter is encoded with 401``name_off = 0`` and ``type = 0``. 402 4032.2.14 BTF_KIND_VAR 404~~~~~~~~~~~~~~~~~~~ 405 406``struct btf_type`` encoding requirement: 407 * ``name_off``: offset to a valid C identifier 408 * ``info.kind_flag``: 0 409 * ``info.kind``: BTF_KIND_VAR 410 * ``info.vlen``: 0 411 * ``type``: the type of the variable 412 413``btf_type`` is followed by a single ``struct btf_variable`` with the 414following data:: 415 416 struct btf_var { 417 __u32 linkage; 418 }; 419 420``struct btf_var`` encoding: 421 * ``linkage``: currently only static variable 0, or globally allocated 422 variable in ELF sections 1 423 424Not all type of global variables are supported by LLVM at this point. 425The following is currently available: 426 427 * static variables with or without section attributes 428 * global variables with section attributes 429 430The latter is for future extraction of map key/value type id's from a 431map definition. 432 4332.2.15 BTF_KIND_DATASEC 434~~~~~~~~~~~~~~~~~~~~~~~ 435 436``struct btf_type`` encoding requirement: 437 * ``name_off``: offset to a valid name associated with a variable or 438 one of .data/.bss/.rodata 439 * ``info.kind_flag``: 0 440 * ``info.kind``: BTF_KIND_DATASEC 441 * ``info.vlen``: # of variables 442 * ``size``: total section size in bytes (0 at compilation time, patched 443 to actual size by BPF loaders such as libbpf) 444 445``btf_type`` is followed by ``info.vlen`` number of ``struct btf_var_secinfo``.:: 446 447 struct btf_var_secinfo { 448 __u32 type; 449 __u32 offset; 450 __u32 size; 451 }; 452 453``struct btf_var_secinfo`` encoding: 454 * ``type``: the type of the BTF_KIND_VAR variable 455 * ``offset``: the in-section offset of the variable 456 * ``size``: the size of the variable in bytes 457 4582.2.16 BTF_KIND_FLOAT 459~~~~~~~~~~~~~~~~~~~~~ 460 461``struct btf_type`` encoding requirement: 462 * ``name_off``: any valid offset 463 * ``info.kind_flag``: 0 464 * ``info.kind``: BTF_KIND_FLOAT 465 * ``info.vlen``: 0 466 * ``size``: the size of the float type in bytes: 2, 4, 8, 12 or 16. 467 468No additional type data follow ``btf_type``. 469 4702.2.17 BTF_KIND_DECL_TAG 471~~~~~~~~~~~~~~~~~~~~~~~~ 472 473``struct btf_type`` encoding requirement: 474 * ``name_off``: offset to a non-empty string 475 * ``info.kind_flag``: 0 476 * ``info.kind``: BTF_KIND_DECL_TAG 477 * ``info.vlen``: 0 478 * ``type``: ``struct``, ``union``, ``func``, ``var`` or ``typedef`` 479 480``btf_type`` is followed by ``struct btf_decl_tag``.:: 481 482 struct btf_decl_tag { 483 __u32 component_idx; 484 }; 485 486The ``name_off`` encodes btf_decl_tag attribute string. 487The ``type`` should be ``struct``, ``union``, ``func``, ``var`` or ``typedef``. 488For ``var`` or ``typedef`` type, ``btf_decl_tag.component_idx`` must be ``-1``. 489For the other three types, if the btf_decl_tag attribute is 490applied to the ``struct``, ``union`` or ``func`` itself, 491``btf_decl_tag.component_idx`` must be ``-1``. Otherwise, 492the attribute is applied to a ``struct``/``union`` member or 493a ``func`` argument, and ``btf_decl_tag.component_idx`` should be a 494valid index (starting from 0) pointing to a member or an argument. 495 4962.2.17 BTF_KIND_TYPE_TAG 497~~~~~~~~~~~~~~~~~~~~~~~~ 498 499``struct btf_type`` encoding requirement: 500 * ``name_off``: offset to a non-empty string 501 * ``info.kind_flag``: 0 502 * ``info.kind``: BTF_KIND_TYPE_TAG 503 * ``info.vlen``: 0 504 * ``type``: the type with ``btf_type_tag`` attribute 505 5063. BTF Kernel API 507================= 508 509The following bpf syscall command involves BTF: 510 * BPF_BTF_LOAD: load a blob of BTF data into kernel 511 * BPF_MAP_CREATE: map creation with btf key and value type info. 512 * BPF_PROG_LOAD: prog load with btf function and line info. 513 * BPF_BTF_GET_FD_BY_ID: get a btf fd 514 * BPF_OBJ_GET_INFO_BY_FD: btf, func_info, line_info 515 and other btf related info are returned. 516 517The workflow typically looks like: 518:: 519 520 Application: 521 BPF_BTF_LOAD 522 | 523 v 524 BPF_MAP_CREATE and BPF_PROG_LOAD 525 | 526 V 527 ...... 528 529 Introspection tool: 530 ...... 531 BPF_{PROG,MAP}_GET_NEXT_ID (get prog/map id's) 532 | 533 V 534 BPF_{PROG,MAP}_GET_FD_BY_ID (get a prog/map fd) 535 | 536 V 537 BPF_OBJ_GET_INFO_BY_FD (get bpf_prog_info/bpf_map_info with btf_id) 538 | | 539 V | 540 BPF_BTF_GET_FD_BY_ID (get btf_fd) | 541 | | 542 V | 543 BPF_OBJ_GET_INFO_BY_FD (get btf) | 544 | | 545 V V 546 pretty print types, dump func signatures and line info, etc. 547 548 5493.1 BPF_BTF_LOAD 550---------------- 551 552Load a blob of BTF data into kernel. A blob of data, described in 553:ref:`BTF_Type_String`, can be directly loaded into the kernel. A ``btf_fd`` 554is returned to a userspace. 555 5563.2 BPF_MAP_CREATE 557------------------ 558 559A map can be created with ``btf_fd`` and specified key/value type id.:: 560 561 __u32 btf_fd; /* fd pointing to a BTF type data */ 562 __u32 btf_key_type_id; /* BTF type_id of the key */ 563 __u32 btf_value_type_id; /* BTF type_id of the value */ 564 565In libbpf, the map can be defined with extra annotation like below: 566:: 567 568 struct bpf_map_def SEC("maps") btf_map = { 569 .type = BPF_MAP_TYPE_ARRAY, 570 .key_size = sizeof(int), 571 .value_size = sizeof(struct ipv_counts), 572 .max_entries = 4, 573 }; 574 BPF_ANNOTATE_KV_PAIR(btf_map, int, struct ipv_counts); 575 576Here, the parameters for macro BPF_ANNOTATE_KV_PAIR are map name, key and 577value types for the map. During ELF parsing, libbpf is able to extract 578key/value type_id's and assign them to BPF_MAP_CREATE attributes 579automatically. 580 581.. _BPF_Prog_Load: 582 5833.3 BPF_PROG_LOAD 584----------------- 585 586During prog_load, func_info and line_info can be passed to kernel with proper 587values for the following attributes: 588:: 589 590 __u32 insn_cnt; 591 __aligned_u64 insns; 592 ...... 593 __u32 prog_btf_fd; /* fd pointing to BTF type data */ 594 __u32 func_info_rec_size; /* userspace bpf_func_info size */ 595 __aligned_u64 func_info; /* func info */ 596 __u32 func_info_cnt; /* number of bpf_func_info records */ 597 __u32 line_info_rec_size; /* userspace bpf_line_info size */ 598 __aligned_u64 line_info; /* line info */ 599 __u32 line_info_cnt; /* number of bpf_line_info records */ 600 601The func_info and line_info are an array of below, respectively.:: 602 603 struct bpf_func_info { 604 __u32 insn_off; /* [0, insn_cnt - 1] */ 605 __u32 type_id; /* pointing to a BTF_KIND_FUNC type */ 606 }; 607 struct bpf_line_info { 608 __u32 insn_off; /* [0, insn_cnt - 1] */ 609 __u32 file_name_off; /* offset to string table for the filename */ 610 __u32 line_off; /* offset to string table for the source line */ 611 __u32 line_col; /* line number and column number */ 612 }; 613 614func_info_rec_size is the size of each func_info record, and 615line_info_rec_size is the size of each line_info record. Passing the record 616size to kernel make it possible to extend the record itself in the future. 617 618Below are requirements for func_info: 619 * func_info[0].insn_off must be 0. 620 * the func_info insn_off is in strictly increasing order and matches 621 bpf func boundaries. 622 623Below are requirements for line_info: 624 * the first insn in each func must have a line_info record pointing to it. 625 * the line_info insn_off is in strictly increasing order. 626 627For line_info, the line number and column number are defined as below: 628:: 629 630 #define BPF_LINE_INFO_LINE_NUM(line_col) ((line_col) >> 10) 631 #define BPF_LINE_INFO_LINE_COL(line_col) ((line_col) & 0x3ff) 632 6333.4 BPF_{PROG,MAP}_GET_NEXT_ID 634------------------------------ 635 636In kernel, every loaded program, map or btf has a unique id. The id won't 637change during the lifetime of a program, map, or btf. 638 639The bpf syscall command BPF_{PROG,MAP}_GET_NEXT_ID returns all id's, one for 640each command, to user space, for bpf program or maps, respectively, so an 641inspection tool can inspect all programs and maps. 642 6433.5 BPF_{PROG,MAP}_GET_FD_BY_ID 644------------------------------- 645 646An introspection tool cannot use id to get details about program or maps. 647A file descriptor needs to be obtained first for reference-counting purpose. 648 6493.6 BPF_OBJ_GET_INFO_BY_FD 650-------------------------- 651 652Once a program/map fd is acquired, an introspection tool can get the detailed 653information from kernel about this fd, some of which are BTF-related. For 654example, ``bpf_map_info`` returns ``btf_id`` and key/value type ids. 655``bpf_prog_info`` returns ``btf_id``, func_info, and line info for translated 656bpf byte codes, and jited_line_info. 657 6583.7 BPF_BTF_GET_FD_BY_ID 659------------------------ 660 661With ``btf_id`` obtained in ``bpf_map_info`` and ``bpf_prog_info``, bpf 662syscall command BPF_BTF_GET_FD_BY_ID can retrieve a btf fd. Then, with 663command BPF_OBJ_GET_INFO_BY_FD, the btf blob, originally loaded into the 664kernel with BPF_BTF_LOAD, can be retrieved. 665 666With the btf blob, ``bpf_map_info``, and ``bpf_prog_info``, an introspection 667tool has full btf knowledge and is able to pretty print map key/values, dump 668func signatures and line info, along with byte/jit codes. 669 6704. ELF File Format Interface 671============================ 672 6734.1 .BTF section 674---------------- 675 676The .BTF section contains type and string data. The format of this section is 677same as the one describe in :ref:`BTF_Type_String`. 678 679.. _BTF_Ext_Section: 680 6814.2 .BTF.ext section 682-------------------- 683 684The .BTF.ext section encodes func_info and line_info which needs loader 685manipulation before loading into the kernel. 686 687The specification for .BTF.ext section is defined at ``tools/lib/bpf/btf.h`` 688and ``tools/lib/bpf/btf.c``. 689 690The current header of .BTF.ext section:: 691 692 struct btf_ext_header { 693 __u16 magic; 694 __u8 version; 695 __u8 flags; 696 __u32 hdr_len; 697 698 /* All offsets are in bytes relative to the end of this header */ 699 __u32 func_info_off; 700 __u32 func_info_len; 701 __u32 line_info_off; 702 __u32 line_info_len; 703 }; 704 705It is very similar to .BTF section. Instead of type/string section, it 706contains func_info and line_info section. See :ref:`BPF_Prog_Load` for details 707about func_info and line_info record format. 708 709The func_info is organized as below.:: 710 711 func_info_rec_size 712 btf_ext_info_sec for section #1 /* func_info for section #1 */ 713 btf_ext_info_sec for section #2 /* func_info for section #2 */ 714 ... 715 716``func_info_rec_size`` specifies the size of ``bpf_func_info`` structure when 717.BTF.ext is generated. ``btf_ext_info_sec``, defined below, is a collection of 718func_info for each specific ELF section.:: 719 720 struct btf_ext_info_sec { 721 __u32 sec_name_off; /* offset to section name */ 722 __u32 num_info; 723 /* Followed by num_info * record_size number of bytes */ 724 __u8 data[0]; 725 }; 726 727Here, num_info must be greater than 0. 728 729The line_info is organized as below.:: 730 731 line_info_rec_size 732 btf_ext_info_sec for section #1 /* line_info for section #1 */ 733 btf_ext_info_sec for section #2 /* line_info for section #2 */ 734 ... 735 736``line_info_rec_size`` specifies the size of ``bpf_line_info`` structure when 737.BTF.ext is generated. 738 739The interpretation of ``bpf_func_info->insn_off`` and 740``bpf_line_info->insn_off`` is different between kernel API and ELF API. For 741kernel API, the ``insn_off`` is the instruction offset in the unit of ``struct 742bpf_insn``. For ELF API, the ``insn_off`` is the byte offset from the 743beginning of section (``btf_ext_info_sec->sec_name_off``). 744 7454.2 .BTF_ids section 746-------------------- 747 748The .BTF_ids section encodes BTF ID values that are used within the kernel. 749 750This section is created during the kernel compilation with the help of 751macros defined in ``include/linux/btf_ids.h`` header file. Kernel code can 752use them to create lists and sets (sorted lists) of BTF ID values. 753 754The ``BTF_ID_LIST`` and ``BTF_ID`` macros define unsorted list of BTF ID values, 755with following syntax:: 756 757 BTF_ID_LIST(list) 758 BTF_ID(type1, name1) 759 BTF_ID(type2, name2) 760 761resulting in following layout in .BTF_ids section:: 762 763 __BTF_ID__type1__name1__1: 764 .zero 4 765 __BTF_ID__type2__name2__2: 766 .zero 4 767 768The ``u32 list[];`` variable is defined to access the list. 769 770The ``BTF_ID_UNUSED`` macro defines 4 zero bytes. It's used when we 771want to define unused entry in BTF_ID_LIST, like:: 772 773 BTF_ID_LIST(bpf_skb_output_btf_ids) 774 BTF_ID(struct, sk_buff) 775 BTF_ID_UNUSED 776 BTF_ID(struct, task_struct) 777 778The ``BTF_SET_START/END`` macros pair defines sorted list of BTF ID values 779and their count, with following syntax:: 780 781 BTF_SET_START(set) 782 BTF_ID(type1, name1) 783 BTF_ID(type2, name2) 784 BTF_SET_END(set) 785 786resulting in following layout in .BTF_ids section:: 787 788 __BTF_ID__set__set: 789 .zero 4 790 __BTF_ID__type1__name1__3: 791 .zero 4 792 __BTF_ID__type2__name2__4: 793 .zero 4 794 795The ``struct btf_id_set set;`` variable is defined to access the list. 796 797The ``typeX`` name can be one of following:: 798 799 struct, union, typedef, func 800 801and is used as a filter when resolving the BTF ID value. 802 803All the BTF ID lists and sets are compiled in the .BTF_ids section and 804resolved during the linking phase of kernel build by ``resolve_btfids`` tool. 805 8065. Using BTF 807============ 808 8095.1 bpftool map pretty print 810---------------------------- 811 812With BTF, the map key/value can be printed based on fields rather than simply 813raw bytes. This is especially valuable for large structure or if your data 814structure has bitfields. For example, for the following map,:: 815 816 enum A { A1, A2, A3, A4, A5 }; 817 typedef enum A ___A; 818 struct tmp_t { 819 char a1:4; 820 int a2:4; 821 int :4; 822 __u32 a3:4; 823 int b; 824 ___A b1:4; 825 enum A b2:4; 826 }; 827 struct bpf_map_def SEC("maps") tmpmap = { 828 .type = BPF_MAP_TYPE_ARRAY, 829 .key_size = sizeof(__u32), 830 .value_size = sizeof(struct tmp_t), 831 .max_entries = 1, 832 }; 833 BPF_ANNOTATE_KV_PAIR(tmpmap, int, struct tmp_t); 834 835bpftool is able to pretty print like below: 836:: 837 838 [{ 839 "key": 0, 840 "value": { 841 "a1": 0x2, 842 "a2": 0x4, 843 "a3": 0x6, 844 "b": 7, 845 "b1": 0x8, 846 "b2": 0xa 847 } 848 } 849 ] 850 8515.2 bpftool prog dump 852--------------------- 853 854The following is an example showing how func_info and line_info can help prog 855dump with better kernel symbol names, function prototypes and line 856information.:: 857 858 $ bpftool prog dump jited pinned /sys/fs/bpf/test_btf_haskv 859 [...] 860 int test_long_fname_2(struct dummy_tracepoint_args * arg): 861 bpf_prog_44a040bf25481309_test_long_fname_2: 862 ; static int test_long_fname_2(struct dummy_tracepoint_args *arg) 863 0: push %rbp 864 1: mov %rsp,%rbp 865 4: sub $0x30,%rsp 866 b: sub $0x28,%rbp 867 f: mov %rbx,0x0(%rbp) 868 13: mov %r13,0x8(%rbp) 869 17: mov %r14,0x10(%rbp) 870 1b: mov %r15,0x18(%rbp) 871 1f: xor %eax,%eax 872 21: mov %rax,0x20(%rbp) 873 25: xor %esi,%esi 874 ; int key = 0; 875 27: mov %esi,-0x4(%rbp) 876 ; if (!arg->sock) 877 2a: mov 0x8(%rdi),%rdi 878 ; if (!arg->sock) 879 2e: cmp $0x0,%rdi 880 32: je 0x0000000000000070 881 34: mov %rbp,%rsi 882 ; counts = bpf_map_lookup_elem(&btf_map, &key); 883 [...] 884 8855.3 Verifier Log 886---------------- 887 888The following is an example of how line_info can help debugging verification 889failure.:: 890 891 /* The code at tools/testing/selftests/bpf/test_xdp_noinline.c 892 * is modified as below. 893 */ 894 data = (void *)(long)xdp->data; 895 data_end = (void *)(long)xdp->data_end; 896 /* 897 if (data + 4 > data_end) 898 return XDP_DROP; 899 */ 900 *(u32 *)data = dst->dst; 901 902 $ bpftool prog load ./test_xdp_noinline.o /sys/fs/bpf/test_xdp_noinline type xdp 903 ; data = (void *)(long)xdp->data; 904 224: (79) r2 = *(u64 *)(r10 -112) 905 225: (61) r2 = *(u32 *)(r2 +0) 906 ; *(u32 *)data = dst->dst; 907 226: (63) *(u32 *)(r2 +0) = r1 908 invalid access to packet, off=0 size=4, R2(id=0,off=0,r=0) 909 R2 offset is outside of the packet 910 9116. BTF Generation 912================= 913 914You need latest pahole 915 916 https://git.kernel.org/pub/scm/devel/pahole/pahole.git/ 917 918or llvm (8.0 or later). The pahole acts as a dwarf2btf converter. It doesn't 919support .BTF.ext and btf BTF_KIND_FUNC type yet. For example,:: 920 921 -bash-4.4$ cat t.c 922 struct t { 923 int a:2; 924 int b:3; 925 int c:2; 926 } g; 927 -bash-4.4$ gcc -c -O2 -g t.c 928 -bash-4.4$ pahole -JV t.o 929 File t.o: 930 [1] STRUCT t kind_flag=1 size=4 vlen=3 931 a type_id=2 bitfield_size=2 bits_offset=0 932 b type_id=2 bitfield_size=3 bits_offset=2 933 c type_id=2 bitfield_size=2 bits_offset=5 934 [2] INT int size=4 bit_offset=0 nr_bits=32 encoding=SIGNED 935 936The llvm is able to generate .BTF and .BTF.ext directly with -g for bpf target 937only. The assembly code (-S) is able to show the BTF encoding in assembly 938format.:: 939 940 -bash-4.4$ cat t2.c 941 typedef int __int32; 942 struct t2 { 943 int a2; 944 int (*f2)(char q1, __int32 q2, ...); 945 int (*f3)(); 946 } g2; 947 int main() { return 0; } 948 int test() { return 0; } 949 -bash-4.4$ clang -c -g -O2 -target bpf t2.c 950 -bash-4.4$ readelf -S t2.o 951 ...... 952 [ 8] .BTF PROGBITS 0000000000000000 00000247 953 000000000000016e 0000000000000000 0 0 1 954 [ 9] .BTF.ext PROGBITS 0000000000000000 000003b5 955 0000000000000060 0000000000000000 0 0 1 956 [10] .rel.BTF.ext REL 0000000000000000 000007e0 957 0000000000000040 0000000000000010 16 9 8 958 ...... 959 -bash-4.4$ clang -S -g -O2 -target bpf t2.c 960 -bash-4.4$ cat t2.s 961 ...... 962 .section .BTF,"",@progbits 963 .short 60319 # 0xeb9f 964 .byte 1 965 .byte 0 966 .long 24 967 .long 0 968 .long 220 969 .long 220 970 .long 122 971 .long 0 # BTF_KIND_FUNC_PROTO(id = 1) 972 .long 218103808 # 0xd000000 973 .long 2 974 .long 83 # BTF_KIND_INT(id = 2) 975 .long 16777216 # 0x1000000 976 .long 4 977 .long 16777248 # 0x1000020 978 ...... 979 .byte 0 # string offset=0 980 .ascii ".text" # string offset=1 981 .byte 0 982 .ascii "/home/yhs/tmp-pahole/t2.c" # string offset=7 983 .byte 0 984 .ascii "int main() { return 0; }" # string offset=33 985 .byte 0 986 .ascii "int test() { return 0; }" # string offset=58 987 .byte 0 988 .ascii "int" # string offset=83 989 ...... 990 .section .BTF.ext,"",@progbits 991 .short 60319 # 0xeb9f 992 .byte 1 993 .byte 0 994 .long 24 995 .long 0 996 .long 28 997 .long 28 998 .long 44 999 .long 8 # FuncInfo 1000 .long 1 # FuncInfo section string offset=1 1001 .long 2 1002 .long .Lfunc_begin0 1003 .long 3 1004 .long .Lfunc_begin1 1005 .long 5 1006 .long 16 # LineInfo 1007 .long 1 # LineInfo section string offset=1 1008 .long 2 1009 .long .Ltmp0 1010 .long 7 1011 .long 33 1012 .long 7182 # Line 7 Col 14 1013 .long .Ltmp3 1014 .long 7 1015 .long 58 1016 .long 8206 # Line 8 Col 14 1017 10187. Testing 1019========== 1020 1021Kernel bpf selftest `test_btf.c` provides extensive set of BTF-related tests. 1022