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