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 == __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 switch (__CORE_RELO(s, field, BYTE_SIZE)) { \ 92 case 1: val = *(const unsigned char *)p; \ 93 case 2: val = *(const unsigned short *)p; \ 94 case 4: val = *(const unsigned int *)p; \ 95 case 8: val = *(const unsigned long long *)p; \ 96 } \ 97 val <<= __CORE_RELO(s, field, LSHIFT_U64); \ 98 if (__CORE_RELO(s, field, SIGNED)) \ 99 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \ 100 else \ 101 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \ 102 val; \ 103 }) 104 105 /* 106 * Convenience macro to check that field actually exists in target kernel's. 107 * Returns: 108 * 1, if matching field is present in target kernel; 109 * 0, if no matching field found. 110 */ 111 #define bpf_core_field_exists(field) \ 112 __builtin_preserve_field_info(field, BPF_FIELD_EXISTS) 113 114 /* 115 * Convenience macro to get the byte size of a field. Works for integers, 116 * struct/unions, pointers, arrays, and enums. 117 */ 118 #define bpf_core_field_size(field) \ 119 __builtin_preserve_field_info(field, BPF_FIELD_BYTE_SIZE) 120 121 /* 122 * Convenience macro to get BTF type ID of a specified type, using a local BTF 123 * information. Return 32-bit unsigned integer with type ID from program's own 124 * BTF. Always succeeds. 125 */ 126 #define bpf_core_type_id_local(type) \ 127 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_LOCAL) 128 129 /* 130 * Convenience macro to get BTF type ID of a target kernel's type that matches 131 * specified local type. 132 * Returns: 133 * - valid 32-bit unsigned type ID in kernel BTF; 134 * - 0, if no matching type was found in a target kernel BTF. 135 */ 136 #define bpf_core_type_id_kernel(type) \ 137 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_TARGET) 138 139 /* 140 * Convenience macro to check that provided named type 141 * (struct/union/enum/typedef) exists in a target kernel. 142 * Returns: 143 * 1, if such type is present in target kernel's BTF; 144 * 0, if no matching type is found. 145 */ 146 #define bpf_core_type_exists(type) \ 147 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_EXISTS) 148 149 /* 150 * Convenience macro to get the byte size of a provided named type 151 * (struct/union/enum/typedef) in a target kernel. 152 * Returns: 153 * >= 0 size (in bytes), if type is present in target kernel's BTF; 154 * 0, if no matching type is found. 155 */ 156 #define bpf_core_type_size(type) \ 157 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_SIZE) 158 159 /* 160 * Convenience macro to check that provided enumerator value is defined in 161 * a target kernel. 162 * Returns: 163 * 1, if specified enum type and its enumerator value are present in target 164 * kernel's BTF; 165 * 0, if no matching enum and/or enum value within that enum is found. 166 */ 167 #define bpf_core_enum_value_exists(enum_type, enum_value) \ 168 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_EXISTS) 169 170 /* 171 * Convenience macro to get the integer value of an enumerator value in 172 * a target kernel. 173 * Returns: 174 * 64-bit value, if specified enum type and its enumerator value are 175 * present in target kernel's BTF; 176 * 0, if no matching enum and/or enum value within that enum is found. 177 */ 178 #define bpf_core_enum_value(enum_type, enum_value) \ 179 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE) 180 181 /* 182 * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures 183 * offset relocation for source address using __builtin_preserve_access_index() 184 * built-in, provided by Clang. 185 * 186 * __builtin_preserve_access_index() takes as an argument an expression of 187 * taking an address of a field within struct/union. It makes compiler emit 188 * a relocation, which records BTF type ID describing root struct/union and an 189 * accessor string which describes exact embedded field that was used to take 190 * an address. See detailed description of this relocation format and 191 * semantics in comments to struct bpf_field_reloc in libbpf_internal.h. 192 * 193 * This relocation allows libbpf to adjust BPF instruction to use correct 194 * actual field offset, based on target kernel BTF type that matches original 195 * (local) BTF, used to record relocation. 196 */ 197 #define bpf_core_read(dst, sz, src) \ 198 bpf_probe_read_kernel(dst, sz, (const void *)__builtin_preserve_access_index(src)) 199 200 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 201 #define bpf_core_read_user(dst, sz, src) \ 202 bpf_probe_read_user(dst, sz, (const void *)__builtin_preserve_access_index(src)) 203 /* 204 * bpf_core_read_str() is a thin wrapper around bpf_probe_read_str() 205 * additionally emitting BPF CO-RE field relocation for specified source 206 * argument. 207 */ 208 #define bpf_core_read_str(dst, sz, src) \ 209 bpf_probe_read_kernel_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 210 211 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 212 #define bpf_core_read_user_str(dst, sz, src) \ 213 bpf_probe_read_user_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 214 215 #define ___concat(a, b) a ## b 216 #define ___apply(fn, n) ___concat(fn, n) 217 #define ___nth(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, __11, N, ...) N 218 219 /* 220 * return number of provided arguments; used for switch-based variadic macro 221 * definitions (see ___last, ___arrow, etc below) 222 */ 223 #define ___narg(...) ___nth(_, ##__VA_ARGS__, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0) 224 /* 225 * return 0 if no arguments are passed, N - otherwise; used for 226 * recursively-defined macros to specify termination (0) case, and generic 227 * (N) case (e.g., ___read_ptrs, ___core_read) 228 */ 229 #define ___empty(...) ___nth(_, ##__VA_ARGS__, N, N, N, N, N, N, N, N, N, N, 0) 230 231 #define ___last1(x) x 232 #define ___last2(a, x) x 233 #define ___last3(a, b, x) x 234 #define ___last4(a, b, c, x) x 235 #define ___last5(a, b, c, d, x) x 236 #define ___last6(a, b, c, d, e, x) x 237 #define ___last7(a, b, c, d, e, f, x) x 238 #define ___last8(a, b, c, d, e, f, g, x) x 239 #define ___last9(a, b, c, d, e, f, g, h, x) x 240 #define ___last10(a, b, c, d, e, f, g, h, i, x) x 241 #define ___last(...) ___apply(___last, ___narg(__VA_ARGS__))(__VA_ARGS__) 242 243 #define ___nolast2(a, _) a 244 #define ___nolast3(a, b, _) a, b 245 #define ___nolast4(a, b, c, _) a, b, c 246 #define ___nolast5(a, b, c, d, _) a, b, c, d 247 #define ___nolast6(a, b, c, d, e, _) a, b, c, d, e 248 #define ___nolast7(a, b, c, d, e, f, _) a, b, c, d, e, f 249 #define ___nolast8(a, b, c, d, e, f, g, _) a, b, c, d, e, f, g 250 #define ___nolast9(a, b, c, d, e, f, g, h, _) a, b, c, d, e, f, g, h 251 #define ___nolast10(a, b, c, d, e, f, g, h, i, _) a, b, c, d, e, f, g, h, i 252 #define ___nolast(...) ___apply(___nolast, ___narg(__VA_ARGS__))(__VA_ARGS__) 253 254 #define ___arrow1(a) a 255 #define ___arrow2(a, b) a->b 256 #define ___arrow3(a, b, c) a->b->c 257 #define ___arrow4(a, b, c, d) a->b->c->d 258 #define ___arrow5(a, b, c, d, e) a->b->c->d->e 259 #define ___arrow6(a, b, c, d, e, f) a->b->c->d->e->f 260 #define ___arrow7(a, b, c, d, e, f, g) a->b->c->d->e->f->g 261 #define ___arrow8(a, b, c, d, e, f, g, h) a->b->c->d->e->f->g->h 262 #define ___arrow9(a, b, c, d, e, f, g, h, i) a->b->c->d->e->f->g->h->i 263 #define ___arrow10(a, b, c, d, e, f, g, h, i, j) a->b->c->d->e->f->g->h->i->j 264 #define ___arrow(...) ___apply(___arrow, ___narg(__VA_ARGS__))(__VA_ARGS__) 265 266 #define ___type(...) typeof(___arrow(__VA_ARGS__)) 267 268 #define ___read(read_fn, dst, src_type, src, accessor) \ 269 read_fn((void *)(dst), sizeof(*(dst)), &((src_type)(src))->accessor) 270 271 /* "recursively" read a sequence of inner pointers using local __t var */ 272 #define ___rd_first(fn, src, a) ___read(fn, &__t, ___type(src), src, a); 273 #define ___rd_last(fn, ...) \ 274 ___read(fn, &__t, ___type(___nolast(__VA_ARGS__)), __t, ___last(__VA_ARGS__)); 275 #define ___rd_p1(fn, ...) const void *__t; ___rd_first(fn, __VA_ARGS__) 276 #define ___rd_p2(fn, ...) ___rd_p1(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 277 #define ___rd_p3(fn, ...) ___rd_p2(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 278 #define ___rd_p4(fn, ...) ___rd_p3(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 279 #define ___rd_p5(fn, ...) ___rd_p4(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 280 #define ___rd_p6(fn, ...) ___rd_p5(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 281 #define ___rd_p7(fn, ...) ___rd_p6(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 282 #define ___rd_p8(fn, ...) ___rd_p7(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 283 #define ___rd_p9(fn, ...) ___rd_p8(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 284 #define ___read_ptrs(fn, src, ...) \ 285 ___apply(___rd_p, ___narg(__VA_ARGS__))(fn, src, __VA_ARGS__) 286 287 #define ___core_read0(fn, fn_ptr, dst, src, a) \ 288 ___read(fn, dst, ___type(src), src, a); 289 #define ___core_readN(fn, fn_ptr, dst, src, ...) \ 290 ___read_ptrs(fn_ptr, src, ___nolast(__VA_ARGS__)) \ 291 ___read(fn, dst, ___type(src, ___nolast(__VA_ARGS__)), __t, \ 292 ___last(__VA_ARGS__)); 293 #define ___core_read(fn, fn_ptr, dst, src, a, ...) \ 294 ___apply(___core_read, ___empty(__VA_ARGS__))(fn, fn_ptr, dst, \ 295 src, a, ##__VA_ARGS__) 296 297 /* 298 * BPF_CORE_READ_INTO() is a more performance-conscious variant of 299 * BPF_CORE_READ(), in which final field is read into user-provided storage. 300 * See BPF_CORE_READ() below for more details on general usage. 301 */ 302 #define BPF_CORE_READ_INTO(dst, src, a, ...) ({ \ 303 ___core_read(bpf_core_read, bpf_core_read, \ 304 dst, (src), a, ##__VA_ARGS__) \ 305 }) 306 307 /* 308 * Variant of BPF_CORE_READ_INTO() for reading from user-space memory. 309 * 310 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 311 */ 312 #define BPF_CORE_READ_USER_INTO(dst, src, a, ...) ({ \ 313 ___core_read(bpf_core_read_user, bpf_core_read_user, \ 314 dst, (src), a, ##__VA_ARGS__) \ 315 }) 316 317 /* Non-CO-RE variant of BPF_CORE_READ_INTO() */ 318 #define BPF_PROBE_READ_INTO(dst, src, a, ...) ({ \ 319 ___core_read(bpf_probe_read, bpf_probe_read, \ 320 dst, (src), a, ##__VA_ARGS__) \ 321 }) 322 323 /* Non-CO-RE variant of BPF_CORE_READ_USER_INTO(). 324 * 325 * As no CO-RE relocations are emitted, source types can be arbitrary and are 326 * not restricted to kernel types only. 327 */ 328 #define BPF_PROBE_READ_USER_INTO(dst, src, a, ...) ({ \ 329 ___core_read(bpf_probe_read_user, bpf_probe_read_user, \ 330 dst, (src), a, ##__VA_ARGS__) \ 331 }) 332 333 /* 334 * BPF_CORE_READ_STR_INTO() does same "pointer chasing" as 335 * BPF_CORE_READ() for intermediate pointers, but then executes (and returns 336 * corresponding error code) bpf_core_read_str() for final string read. 337 */ 338 #define BPF_CORE_READ_STR_INTO(dst, src, a, ...) ({ \ 339 ___core_read(bpf_core_read_str, bpf_core_read, \ 340 dst, (src), a, ##__VA_ARGS__) \ 341 }) 342 343 /* 344 * Variant of BPF_CORE_READ_STR_INTO() for reading from user-space memory. 345 * 346 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 347 */ 348 #define BPF_CORE_READ_USER_STR_INTO(dst, src, a, ...) ({ \ 349 ___core_read(bpf_core_read_user_str, bpf_core_read_user, \ 350 dst, (src), a, ##__VA_ARGS__) \ 351 }) 352 353 /* Non-CO-RE variant of BPF_CORE_READ_STR_INTO() */ 354 #define BPF_PROBE_READ_STR_INTO(dst, src, a, ...) ({ \ 355 ___core_read(bpf_probe_read_str, bpf_probe_read, \ 356 dst, (src), a, ##__VA_ARGS__) \ 357 }) 358 359 /* 360 * Non-CO-RE variant of BPF_CORE_READ_USER_STR_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_STR_INTO(dst, src, a, ...) ({ \ 366 ___core_read(bpf_probe_read_user_str, bpf_probe_read_user, \ 367 dst, (src), a, ##__VA_ARGS__) \ 368 }) 369 370 /* 371 * BPF_CORE_READ() is used to simplify BPF CO-RE relocatable read, especially 372 * when there are few pointer chasing steps. 373 * E.g., what in non-BPF world (or in BPF w/ BCC) would be something like: 374 * int x = s->a.b.c->d.e->f->g; 375 * can be succinctly achieved using BPF_CORE_READ as: 376 * int x = BPF_CORE_READ(s, a.b.c, d.e, f, g); 377 * 378 * BPF_CORE_READ will decompose above statement into 4 bpf_core_read (BPF 379 * CO-RE relocatable bpf_probe_read_kernel() wrapper) calls, logically 380 * equivalent to: 381 * 1. const void *__t = s->a.b.c; 382 * 2. __t = __t->d.e; 383 * 3. __t = __t->f; 384 * 4. return __t->g; 385 * 386 * Equivalence is logical, because there is a heavy type casting/preservation 387 * involved, as well as all the reads are happening through 388 * bpf_probe_read_kernel() calls using __builtin_preserve_access_index() to 389 * emit CO-RE relocations. 390 * 391 * N.B. Only up to 9 "field accessors" are supported, which should be more 392 * than enough for any practical purpose. 393 */ 394 #define BPF_CORE_READ(src, a, ...) ({ \ 395 ___type((src), a, ##__VA_ARGS__) __r; \ 396 BPF_CORE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 397 __r; \ 398 }) 399 400 /* 401 * Variant of BPF_CORE_READ() for reading from user-space memory. 402 * 403 * NOTE: all the source types involved are still *kernel types* and need to 404 * exist in kernel (or kernel module) BTF, otherwise CO-RE relocation will 405 * fail. Custom user types are not relocatable with CO-RE. 406 * The typical situation in which BPF_CORE_READ_USER() might be used is to 407 * read kernel UAPI types from the user-space memory passed in as a syscall 408 * input argument. 409 */ 410 #define BPF_CORE_READ_USER(src, a, ...) ({ \ 411 ___type((src), a, ##__VA_ARGS__) __r; \ 412 BPF_CORE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 413 __r; \ 414 }) 415 416 /* Non-CO-RE variant of BPF_CORE_READ() */ 417 #define BPF_PROBE_READ(src, a, ...) ({ \ 418 ___type((src), a, ##__VA_ARGS__) __r; \ 419 BPF_PROBE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 420 __r; \ 421 }) 422 423 /* 424 * Non-CO-RE variant of BPF_CORE_READ_USER(). 425 * 426 * As no CO-RE relocations are emitted, source types can be arbitrary and are 427 * not restricted to kernel types only. 428 */ 429 #define BPF_PROBE_READ_USER(src, a, ...) ({ \ 430 ___type((src), a, ##__VA_ARGS__) __r; \ 431 BPF_PROBE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 432 __r; \ 433 }) 434 435 #endif 436 437