1/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */ 2#ifndef __BPF_CORE_READ_H__ 3#define __BPF_CORE_READ_H__ 4 5#include <bpf/bpf_helpers.h> 6 7/* 8 * enum bpf_field_info_kind is passed as a second argument into 9 * __builtin_preserve_field_info() built-in to get a specific aspect of 10 * a field, captured as a first argument. __builtin_preserve_field_info(field, 11 * info_kind) returns __u32 integer and produces BTF field relocation, which 12 * is understood and processed by libbpf during BPF object loading. See 13 * selftests/bpf for examples. 14 */ 15enum bpf_field_info_kind { 16 BPF_FIELD_BYTE_OFFSET = 0, /* field byte offset */ 17 BPF_FIELD_BYTE_SIZE = 1, 18 BPF_FIELD_EXISTS = 2, /* field existence in target kernel */ 19 BPF_FIELD_SIGNED = 3, 20 BPF_FIELD_LSHIFT_U64 = 4, 21 BPF_FIELD_RSHIFT_U64 = 5, 22}; 23 24/* second argument to __builtin_btf_type_id() built-in */ 25enum bpf_type_id_kind { 26 BPF_TYPE_ID_LOCAL = 0, /* BTF type ID in local program */ 27 BPF_TYPE_ID_TARGET = 1, /* BTF type ID in target kernel */ 28}; 29 30/* second argument to __builtin_preserve_type_info() built-in */ 31enum bpf_type_info_kind { 32 BPF_TYPE_EXISTS = 0, /* type existence in target kernel */ 33 BPF_TYPE_SIZE = 1, /* type size in target kernel */ 34 BPF_TYPE_MATCHES = 2, /* type match in target kernel */ 35}; 36 37/* second argument to __builtin_preserve_enum_value() built-in */ 38enum bpf_enum_value_kind { 39 BPF_ENUMVAL_EXISTS = 0, /* enum value existence in kernel */ 40 BPF_ENUMVAL_VALUE = 1, /* enum value value relocation */ 41}; 42 43#define __CORE_RELO(src, field, info) \ 44 __builtin_preserve_field_info((src)->field, BPF_FIELD_##info) 45 46#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 47#define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \ 48 bpf_probe_read_kernel( \ 49 (void *)dst, \ 50 __CORE_RELO(src, fld, BYTE_SIZE), \ 51 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET)) 52#else 53/* semantics of LSHIFT_64 assumes loading values into low-ordered bytes, so 54 * for big-endian we need to adjust destination pointer accordingly, based on 55 * field byte size 56 */ 57#define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \ 58 bpf_probe_read_kernel( \ 59 (void *)dst + (8 - __CORE_RELO(src, fld, BYTE_SIZE)), \ 60 __CORE_RELO(src, fld, BYTE_SIZE), \ 61 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET)) 62#endif 63 64/* 65 * Extract bitfield, identified by s->field, and return its value as u64. 66 * All this is done in relocatable manner, so bitfield changes such as 67 * signedness, bit size, offset changes, this will be handled automatically. 68 * This version of macro is using bpf_probe_read_kernel() to read underlying 69 * integer storage. Macro functions as an expression and its return type is 70 * bpf_probe_read_kernel()'s return value: 0, on success, <0 on error. 71 */ 72#define BPF_CORE_READ_BITFIELD_PROBED(s, field) ({ \ 73 unsigned long long val = 0; \ 74 \ 75 __CORE_BITFIELD_PROBE_READ(&val, s, field); \ 76 val <<= __CORE_RELO(s, field, LSHIFT_U64); \ 77 if (__CORE_RELO(s, field, SIGNED)) \ 78 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \ 79 else \ 80 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \ 81 val; \ 82}) 83 84/* 85 * Extract bitfield, identified by s->field, and return its value as u64. 86 * This version of macro is using direct memory reads and should be used from 87 * BPF program types that support such functionality (e.g., typed raw 88 * tracepoints). 89 */ 90#define BPF_CORE_READ_BITFIELD(s, field) ({ \ 91 const void *p = (const void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \ 92 unsigned long long val; \ 93 \ 94 /* This is a so-called barrier_var() operation that makes specified \ 95 * variable "a black box" for optimizing compiler. \ 96 * It forces compiler to perform BYTE_OFFSET relocation on p and use \ 97 * its calculated value in the switch below, instead of applying \ 98 * the same relocation 4 times for each individual memory load. \ 99 */ \ 100 asm volatile("" : "=r"(p) : "0"(p)); \ 101 \ 102 switch (__CORE_RELO(s, field, BYTE_SIZE)) { \ 103 case 1: val = *(const unsigned char *)p; break; \ 104 case 2: val = *(const unsigned short *)p; break; \ 105 case 4: val = *(const unsigned int *)p; break; \ 106 case 8: val = *(const unsigned long long *)p; break; \ 107 } \ 108 val <<= __CORE_RELO(s, field, LSHIFT_U64); \ 109 if (__CORE_RELO(s, field, SIGNED)) \ 110 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \ 111 else \ 112 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \ 113 val; \ 114}) 115 116/* 117 * Write to a bitfield, identified by s->field. 118 * This is the inverse of BPF_CORE_WRITE_BITFIELD(). 119 */ 120#define BPF_CORE_WRITE_BITFIELD(s, field, new_val) ({ \ 121 void *p = (void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \ 122 unsigned int byte_size = __CORE_RELO(s, field, BYTE_SIZE); \ 123 unsigned int lshift = __CORE_RELO(s, field, LSHIFT_U64); \ 124 unsigned int rshift = __CORE_RELO(s, field, RSHIFT_U64); \ 125 unsigned long long mask, val, nval = new_val; \ 126 unsigned int rpad = rshift - lshift; \ 127 \ 128 asm volatile("" : "+r"(p)); \ 129 \ 130 switch (byte_size) { \ 131 case 1: val = *(unsigned char *)p; break; \ 132 case 2: val = *(unsigned short *)p; break; \ 133 case 4: val = *(unsigned int *)p; break; \ 134 case 8: val = *(unsigned long long *)p; break; \ 135 } \ 136 \ 137 mask = (~0ULL << rshift) >> lshift; \ 138 val = (val & ~mask) | ((nval << rpad) & mask); \ 139 \ 140 switch (byte_size) { \ 141 case 1: *(unsigned char *)p = val; break; \ 142 case 2: *(unsigned short *)p = val; break; \ 143 case 4: *(unsigned int *)p = val; break; \ 144 case 8: *(unsigned long long *)p = val; break; \ 145 } \ 146}) 147 148/* Differentiator between compilers builtin implementations. This is a 149 * requirement due to the compiler parsing differences where GCC optimizes 150 * early in parsing those constructs of type pointers to the builtin specific 151 * type, resulting in not being possible to collect the required type 152 * information in the builtin expansion. 153 */ 154#ifdef __clang__ 155#define ___bpf_typeof(type) ((typeof(type) *) 0) 156#else 157#define ___bpf_typeof1(type, NR) ({ \ 158 extern typeof(type) *___concat(bpf_type_tmp_, NR); \ 159 ___concat(bpf_type_tmp_, NR); \ 160}) 161#define ___bpf_typeof(type) ___bpf_typeof1(type, __COUNTER__) 162#endif 163 164#ifdef __clang__ 165#define ___bpf_field_ref1(field) (field) 166#define ___bpf_field_ref2(type, field) (___bpf_typeof(type)->field) 167#else 168#define ___bpf_field_ref1(field) (&(field)) 169#define ___bpf_field_ref2(type, field) (&(___bpf_typeof(type)->field)) 170#endif 171#define ___bpf_field_ref(args...) \ 172 ___bpf_apply(___bpf_field_ref, ___bpf_narg(args))(args) 173 174/* 175 * Convenience macro to check that field actually exists in target kernel's. 176 * Returns: 177 * 1, if matching field is present in target kernel; 178 * 0, if no matching field found. 179 * 180 * Supports two forms: 181 * - field reference through variable access: 182 * bpf_core_field_exists(p->my_field); 183 * - field reference through type and field names: 184 * bpf_core_field_exists(struct my_type, my_field). 185 */ 186#define bpf_core_field_exists(field...) \ 187 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_EXISTS) 188 189/* 190 * Convenience macro to get the byte size of a field. Works for integers, 191 * struct/unions, pointers, arrays, and enums. 192 * 193 * Supports two forms: 194 * - field reference through variable access: 195 * bpf_core_field_size(p->my_field); 196 * - field reference through type and field names: 197 * bpf_core_field_size(struct my_type, my_field). 198 */ 199#define bpf_core_field_size(field...) \ 200 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_SIZE) 201 202/* 203 * Convenience macro to get field's byte offset. 204 * 205 * Supports two forms: 206 * - field reference through variable access: 207 * bpf_core_field_offset(p->my_field); 208 * - field reference through type and field names: 209 * bpf_core_field_offset(struct my_type, my_field). 210 */ 211#define bpf_core_field_offset(field...) \ 212 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_OFFSET) 213 214/* 215 * Convenience macro to get BTF type ID of a specified type, using a local BTF 216 * information. Return 32-bit unsigned integer with type ID from program's own 217 * BTF. Always succeeds. 218 */ 219#define bpf_core_type_id_local(type) \ 220 __builtin_btf_type_id(*___bpf_typeof(type), BPF_TYPE_ID_LOCAL) 221 222/* 223 * Convenience macro to get BTF type ID of a target kernel's type that matches 224 * specified local type. 225 * Returns: 226 * - valid 32-bit unsigned type ID in kernel BTF; 227 * - 0, if no matching type was found in a target kernel BTF. 228 */ 229#define bpf_core_type_id_kernel(type) \ 230 __builtin_btf_type_id(*___bpf_typeof(type), BPF_TYPE_ID_TARGET) 231 232/* 233 * Convenience macro to check that provided named type 234 * (struct/union/enum/typedef) exists in a target kernel. 235 * Returns: 236 * 1, if such type is present in target kernel's BTF; 237 * 0, if no matching type is found. 238 */ 239#define bpf_core_type_exists(type) \ 240 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_EXISTS) 241 242/* 243 * Convenience macro to check that provided named type 244 * (struct/union/enum/typedef) "matches" that in a target kernel. 245 * Returns: 246 * 1, if the type matches in the target kernel's BTF; 247 * 0, if the type does not match any in the target kernel 248 */ 249#define bpf_core_type_matches(type) \ 250 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_MATCHES) 251 252/* 253 * Convenience macro to get the byte size of a provided named type 254 * (struct/union/enum/typedef) in a target kernel. 255 * Returns: 256 * >= 0 size (in bytes), if type is present in target kernel's BTF; 257 * 0, if no matching type is found. 258 */ 259#define bpf_core_type_size(type) \ 260 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_SIZE) 261 262/* 263 * Convenience macro to check that provided enumerator value is defined in 264 * a target kernel. 265 * Returns: 266 * 1, if specified enum type and its enumerator value are present in target 267 * kernel's BTF; 268 * 0, if no matching enum and/or enum value within that enum is found. 269 */ 270#ifdef __clang__ 271#define bpf_core_enum_value_exists(enum_type, enum_value) \ 272 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_EXISTS) 273#else 274#define bpf_core_enum_value_exists(enum_type, enum_value) \ 275 __builtin_preserve_enum_value(___bpf_typeof(enum_type), enum_value, BPF_ENUMVAL_EXISTS) 276#endif 277 278/* 279 * Convenience macro to get the integer value of an enumerator value in 280 * a target kernel. 281 * Returns: 282 * 64-bit value, if specified enum type and its enumerator value are 283 * present in target kernel's BTF; 284 * 0, if no matching enum and/or enum value within that enum is found. 285 */ 286#ifdef __clang__ 287#define bpf_core_enum_value(enum_type, enum_value) \ 288 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE) 289#else 290#define bpf_core_enum_value(enum_type, enum_value) \ 291 __builtin_preserve_enum_value(___bpf_typeof(enum_type), enum_value, BPF_ENUMVAL_VALUE) 292#endif 293 294/* 295 * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures 296 * offset relocation for source address using __builtin_preserve_access_index() 297 * built-in, provided by Clang. 298 * 299 * __builtin_preserve_access_index() takes as an argument an expression of 300 * taking an address of a field within struct/union. It makes compiler emit 301 * a relocation, which records BTF type ID describing root struct/union and an 302 * accessor string which describes exact embedded field that was used to take 303 * an address. See detailed description of this relocation format and 304 * semantics in comments to struct bpf_core_relo in include/uapi/linux/bpf.h. 305 * 306 * This relocation allows libbpf to adjust BPF instruction to use correct 307 * actual field offset, based on target kernel BTF type that matches original 308 * (local) BTF, used to record relocation. 309 */ 310#define bpf_core_read(dst, sz, src) \ 311 bpf_probe_read_kernel(dst, sz, (const void *)__builtin_preserve_access_index(src)) 312 313/* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 314#define bpf_core_read_user(dst, sz, src) \ 315 bpf_probe_read_user(dst, sz, (const void *)__builtin_preserve_access_index(src)) 316/* 317 * bpf_core_read_str() is a thin wrapper around bpf_probe_read_str() 318 * additionally emitting BPF CO-RE field relocation for specified source 319 * argument. 320 */ 321#define bpf_core_read_str(dst, sz, src) \ 322 bpf_probe_read_kernel_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 323 324/* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 325#define bpf_core_read_user_str(dst, sz, src) \ 326 bpf_probe_read_user_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 327 328extern void *bpf_rdonly_cast(const void *obj, __u32 btf_id) __ksym __weak; 329 330/* 331 * Cast provided pointer *ptr* into a pointer to a specified *type* in such 332 * a way that BPF verifier will become aware of associated kernel-side BTF 333 * type. This allows to access members of kernel types directly without the 334 * need to use BPF_CORE_READ() macros. 335 */ 336#define bpf_core_cast(ptr, type) \ 337 ((typeof(type) *)bpf_rdonly_cast((ptr), bpf_core_type_id_kernel(type))) 338 339#define ___concat(a, b) a ## b 340#define ___apply(fn, n) ___concat(fn, n) 341#define ___nth(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, __11, N, ...) N 342 343/* 344 * return number of provided arguments; used for switch-based variadic macro 345 * definitions (see ___last, ___arrow, etc below) 346 */ 347#define ___narg(...) ___nth(_, ##__VA_ARGS__, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0) 348/* 349 * return 0 if no arguments are passed, N - otherwise; used for 350 * recursively-defined macros to specify termination (0) case, and generic 351 * (N) case (e.g., ___read_ptrs, ___core_read) 352 */ 353#define ___empty(...) ___nth(_, ##__VA_ARGS__, N, N, N, N, N, N, N, N, N, N, 0) 354 355#define ___last1(x) x 356#define ___last2(a, x) x 357#define ___last3(a, b, x) x 358#define ___last4(a, b, c, x) x 359#define ___last5(a, b, c, d, x) x 360#define ___last6(a, b, c, d, e, x) x 361#define ___last7(a, b, c, d, e, f, x) x 362#define ___last8(a, b, c, d, e, f, g, x) x 363#define ___last9(a, b, c, d, e, f, g, h, x) x 364#define ___last10(a, b, c, d, e, f, g, h, i, x) x 365#define ___last(...) ___apply(___last, ___narg(__VA_ARGS__))(__VA_ARGS__) 366 367#define ___nolast2(a, _) a 368#define ___nolast3(a, b, _) a, b 369#define ___nolast4(a, b, c, _) a, b, c 370#define ___nolast5(a, b, c, d, _) a, b, c, d 371#define ___nolast6(a, b, c, d, e, _) a, b, c, d, e 372#define ___nolast7(a, b, c, d, e, f, _) a, b, c, d, e, f 373#define ___nolast8(a, b, c, d, e, f, g, _) a, b, c, d, e, f, g 374#define ___nolast9(a, b, c, d, e, f, g, h, _) a, b, c, d, e, f, g, h 375#define ___nolast10(a, b, c, d, e, f, g, h, i, _) a, b, c, d, e, f, g, h, i 376#define ___nolast(...) ___apply(___nolast, ___narg(__VA_ARGS__))(__VA_ARGS__) 377 378#define ___arrow1(a) a 379#define ___arrow2(a, b) a->b 380#define ___arrow3(a, b, c) a->b->c 381#define ___arrow4(a, b, c, d) a->b->c->d 382#define ___arrow5(a, b, c, d, e) a->b->c->d->e 383#define ___arrow6(a, b, c, d, e, f) a->b->c->d->e->f 384#define ___arrow7(a, b, c, d, e, f, g) a->b->c->d->e->f->g 385#define ___arrow8(a, b, c, d, e, f, g, h) a->b->c->d->e->f->g->h 386#define ___arrow9(a, b, c, d, e, f, g, h, i) a->b->c->d->e->f->g->h->i 387#define ___arrow10(a, b, c, d, e, f, g, h, i, j) a->b->c->d->e->f->g->h->i->j 388#define ___arrow(...) ___apply(___arrow, ___narg(__VA_ARGS__))(__VA_ARGS__) 389 390#define ___type(...) typeof(___arrow(__VA_ARGS__)) 391 392#define ___read(read_fn, dst, src_type, src, accessor) \ 393 read_fn((void *)(dst), sizeof(*(dst)), &((src_type)(src))->accessor) 394 395/* "recursively" read a sequence of inner pointers using local __t var */ 396#define ___rd_first(fn, src, a) ___read(fn, &__t, ___type(src), src, a); 397#define ___rd_last(fn, ...) \ 398 ___read(fn, &__t, ___type(___nolast(__VA_ARGS__)), __t, ___last(__VA_ARGS__)); 399#define ___rd_p1(fn, ...) const void *__t; ___rd_first(fn, __VA_ARGS__) 400#define ___rd_p2(fn, ...) ___rd_p1(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 401#define ___rd_p3(fn, ...) ___rd_p2(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 402#define ___rd_p4(fn, ...) ___rd_p3(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 403#define ___rd_p5(fn, ...) ___rd_p4(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 404#define ___rd_p6(fn, ...) ___rd_p5(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 405#define ___rd_p7(fn, ...) ___rd_p6(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 406#define ___rd_p8(fn, ...) ___rd_p7(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 407#define ___rd_p9(fn, ...) ___rd_p8(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 408#define ___read_ptrs(fn, src, ...) \ 409 ___apply(___rd_p, ___narg(__VA_ARGS__))(fn, src, __VA_ARGS__) 410 411#define ___core_read0(fn, fn_ptr, dst, src, a) \ 412 ___read(fn, dst, ___type(src), src, a); 413#define ___core_readN(fn, fn_ptr, dst, src, ...) \ 414 ___read_ptrs(fn_ptr, src, ___nolast(__VA_ARGS__)) \ 415 ___read(fn, dst, ___type(src, ___nolast(__VA_ARGS__)), __t, \ 416 ___last(__VA_ARGS__)); 417#define ___core_read(fn, fn_ptr, dst, src, a, ...) \ 418 ___apply(___core_read, ___empty(__VA_ARGS__))(fn, fn_ptr, dst, \ 419 src, a, ##__VA_ARGS__) 420 421/* 422 * BPF_CORE_READ_INTO() is a more performance-conscious variant of 423 * BPF_CORE_READ(), in which final field is read into user-provided storage. 424 * See BPF_CORE_READ() below for more details on general usage. 425 */ 426#define BPF_CORE_READ_INTO(dst, src, a, ...) ({ \ 427 ___core_read(bpf_core_read, bpf_core_read, \ 428 dst, (src), a, ##__VA_ARGS__) \ 429}) 430 431/* 432 * Variant of BPF_CORE_READ_INTO() for reading from user-space memory. 433 * 434 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 435 */ 436#define BPF_CORE_READ_USER_INTO(dst, src, a, ...) ({ \ 437 ___core_read(bpf_core_read_user, bpf_core_read_user, \ 438 dst, (src), a, ##__VA_ARGS__) \ 439}) 440 441/* Non-CO-RE variant of BPF_CORE_READ_INTO() */ 442#define BPF_PROBE_READ_INTO(dst, src, a, ...) ({ \ 443 ___core_read(bpf_probe_read_kernel, bpf_probe_read_kernel, \ 444 dst, (src), a, ##__VA_ARGS__) \ 445}) 446 447/* Non-CO-RE variant of BPF_CORE_READ_USER_INTO(). 448 * 449 * As no CO-RE relocations are emitted, source types can be arbitrary and are 450 * not restricted to kernel types only. 451 */ 452#define BPF_PROBE_READ_USER_INTO(dst, src, a, ...) ({ \ 453 ___core_read(bpf_probe_read_user, bpf_probe_read_user, \ 454 dst, (src), a, ##__VA_ARGS__) \ 455}) 456 457/* 458 * BPF_CORE_READ_STR_INTO() does same "pointer chasing" as 459 * BPF_CORE_READ() for intermediate pointers, but then executes (and returns 460 * corresponding error code) bpf_core_read_str() for final string read. 461 */ 462#define BPF_CORE_READ_STR_INTO(dst, src, a, ...) ({ \ 463 ___core_read(bpf_core_read_str, bpf_core_read, \ 464 dst, (src), a, ##__VA_ARGS__) \ 465}) 466 467/* 468 * Variant of BPF_CORE_READ_STR_INTO() for reading from user-space memory. 469 * 470 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 471 */ 472#define BPF_CORE_READ_USER_STR_INTO(dst, src, a, ...) ({ \ 473 ___core_read(bpf_core_read_user_str, bpf_core_read_user, \ 474 dst, (src), a, ##__VA_ARGS__) \ 475}) 476 477/* Non-CO-RE variant of BPF_CORE_READ_STR_INTO() */ 478#define BPF_PROBE_READ_STR_INTO(dst, src, a, ...) ({ \ 479 ___core_read(bpf_probe_read_kernel_str, bpf_probe_read_kernel, \ 480 dst, (src), a, ##__VA_ARGS__) \ 481}) 482 483/* 484 * Non-CO-RE variant of BPF_CORE_READ_USER_STR_INTO(). 485 * 486 * As no CO-RE relocations are emitted, source types can be arbitrary and are 487 * not restricted to kernel types only. 488 */ 489#define BPF_PROBE_READ_USER_STR_INTO(dst, src, a, ...) ({ \ 490 ___core_read(bpf_probe_read_user_str, bpf_probe_read_user, \ 491 dst, (src), a, ##__VA_ARGS__) \ 492}) 493 494/* 495 * BPF_CORE_READ() is used to simplify BPF CO-RE relocatable read, especially 496 * when there are few pointer chasing steps. 497 * E.g., what in non-BPF world (or in BPF w/ BCC) would be something like: 498 * int x = s->a.b.c->d.e->f->g; 499 * can be succinctly achieved using BPF_CORE_READ as: 500 * int x = BPF_CORE_READ(s, a.b.c, d.e, f, g); 501 * 502 * BPF_CORE_READ will decompose above statement into 4 bpf_core_read (BPF 503 * CO-RE relocatable bpf_probe_read_kernel() wrapper) calls, logically 504 * equivalent to: 505 * 1. const void *__t = s->a.b.c; 506 * 2. __t = __t->d.e; 507 * 3. __t = __t->f; 508 * 4. return __t->g; 509 * 510 * Equivalence is logical, because there is a heavy type casting/preservation 511 * involved, as well as all the reads are happening through 512 * bpf_probe_read_kernel() calls using __builtin_preserve_access_index() to 513 * emit CO-RE relocations. 514 * 515 * N.B. Only up to 9 "field accessors" are supported, which should be more 516 * than enough for any practical purpose. 517 */ 518#define BPF_CORE_READ(src, a, ...) ({ \ 519 ___type((src), a, ##__VA_ARGS__) __r; \ 520 BPF_CORE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 521 __r; \ 522}) 523 524/* 525 * Variant of BPF_CORE_READ() for reading from user-space memory. 526 * 527 * NOTE: all the source types involved are still *kernel types* and need to 528 * exist in kernel (or kernel module) BTF, otherwise CO-RE relocation will 529 * fail. Custom user types are not relocatable with CO-RE. 530 * The typical situation in which BPF_CORE_READ_USER() might be used is to 531 * read kernel UAPI types from the user-space memory passed in as a syscall 532 * input argument. 533 */ 534#define BPF_CORE_READ_USER(src, a, ...) ({ \ 535 ___type((src), a, ##__VA_ARGS__) __r; \ 536 BPF_CORE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 537 __r; \ 538}) 539 540/* Non-CO-RE variant of BPF_CORE_READ() */ 541#define BPF_PROBE_READ(src, a, ...) ({ \ 542 ___type((src), a, ##__VA_ARGS__) __r; \ 543 BPF_PROBE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 544 __r; \ 545}) 546 547/* 548 * Non-CO-RE variant of BPF_CORE_READ_USER(). 549 * 550 * As no CO-RE relocations are emitted, source types can be arbitrary and are 551 * not restricted to kernel types only. 552 */ 553#define BPF_PROBE_READ_USER(src, a, ...) ({ \ 554 ___type((src), a, ##__VA_ARGS__) __r; \ 555 BPF_PROBE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 556 __r; \ 557}) 558 559#endif 560 561