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 /* 114 * Convenience macro to check that field actually exists in target kernel's. 115 * Returns: 116 * 1, if matching field is present in target kernel; 117 * 0, if no matching field found. 118 */ 119 #define bpf_core_field_exists(field) \ 120 __builtin_preserve_field_info(field, BPF_FIELD_EXISTS) 121 122 /* 123 * Convenience macro to get the byte size of a field. Works for integers, 124 * struct/unions, pointers, arrays, and enums. 125 */ 126 #define bpf_core_field_size(field) \ 127 __builtin_preserve_field_info(field, BPF_FIELD_BYTE_SIZE) 128 129 /* 130 * Convenience macro to get BTF type ID of a specified type, using a local BTF 131 * information. Return 32-bit unsigned integer with type ID from program's own 132 * BTF. Always succeeds. 133 */ 134 #define bpf_core_type_id_local(type) \ 135 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_LOCAL) 136 137 /* 138 * Convenience macro to get BTF type ID of a target kernel's type that matches 139 * specified local type. 140 * Returns: 141 * - valid 32-bit unsigned type ID in kernel BTF; 142 * - 0, if no matching type was found in a target kernel BTF. 143 */ 144 #define bpf_core_type_id_kernel(type) \ 145 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_TARGET) 146 147 /* 148 * Convenience macro to check that provided named type 149 * (struct/union/enum/typedef) exists in a target kernel. 150 * Returns: 151 * 1, if such type is present in target kernel's BTF; 152 * 0, if no matching type is found. 153 */ 154 #define bpf_core_type_exists(type) \ 155 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_EXISTS) 156 157 /* 158 * Convenience macro to get the byte size of a provided named type 159 * (struct/union/enum/typedef) in a target kernel. 160 * Returns: 161 * >= 0 size (in bytes), if type is present in target kernel's BTF; 162 * 0, if no matching type is found. 163 */ 164 #define bpf_core_type_size(type) \ 165 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_SIZE) 166 167 /* 168 * Convenience macro to check that provided enumerator value is defined in 169 * a target kernel. 170 * Returns: 171 * 1, if specified enum type and its enumerator value are present in target 172 * kernel's BTF; 173 * 0, if no matching enum and/or enum value within that enum is found. 174 */ 175 #define bpf_core_enum_value_exists(enum_type, enum_value) \ 176 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_EXISTS) 177 178 /* 179 * Convenience macro to get the integer value of an enumerator value in 180 * a target kernel. 181 * Returns: 182 * 64-bit value, if specified enum type and its enumerator value are 183 * present in target kernel's BTF; 184 * 0, if no matching enum and/or enum value within that enum is found. 185 */ 186 #define bpf_core_enum_value(enum_type, enum_value) \ 187 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE) 188 189 /* 190 * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures 191 * offset relocation for source address using __builtin_preserve_access_index() 192 * built-in, provided by Clang. 193 * 194 * __builtin_preserve_access_index() takes as an argument an expression of 195 * taking an address of a field within struct/union. It makes compiler emit 196 * a relocation, which records BTF type ID describing root struct/union and an 197 * accessor string which describes exact embedded field that was used to take 198 * an address. See detailed description of this relocation format and 199 * semantics in comments to struct bpf_field_reloc in libbpf_internal.h. 200 * 201 * This relocation allows libbpf to adjust BPF instruction to use correct 202 * actual field offset, based on target kernel BTF type that matches original 203 * (local) BTF, used to record relocation. 204 */ 205 #define bpf_core_read(dst, sz, src) \ 206 bpf_probe_read_kernel(dst, sz, (const void *)__builtin_preserve_access_index(src)) 207 208 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 209 #define bpf_core_read_user(dst, sz, src) \ 210 bpf_probe_read_user(dst, sz, (const void *)__builtin_preserve_access_index(src)) 211 /* 212 * bpf_core_read_str() is a thin wrapper around bpf_probe_read_str() 213 * additionally emitting BPF CO-RE field relocation for specified source 214 * argument. 215 */ 216 #define bpf_core_read_str(dst, sz, src) \ 217 bpf_probe_read_kernel_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 218 219 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */ 220 #define bpf_core_read_user_str(dst, sz, src) \ 221 bpf_probe_read_user_str(dst, sz, (const void *)__builtin_preserve_access_index(src)) 222 223 #define ___concat(a, b) a ## b 224 #define ___apply(fn, n) ___concat(fn, n) 225 #define ___nth(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, __11, N, ...) N 226 227 /* 228 * return number of provided arguments; used for switch-based variadic macro 229 * definitions (see ___last, ___arrow, etc below) 230 */ 231 #define ___narg(...) ___nth(_, ##__VA_ARGS__, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0) 232 /* 233 * return 0 if no arguments are passed, N - otherwise; used for 234 * recursively-defined macros to specify termination (0) case, and generic 235 * (N) case (e.g., ___read_ptrs, ___core_read) 236 */ 237 #define ___empty(...) ___nth(_, ##__VA_ARGS__, N, N, N, N, N, N, N, N, N, N, 0) 238 239 #define ___last1(x) x 240 #define ___last2(a, x) x 241 #define ___last3(a, b, x) x 242 #define ___last4(a, b, c, x) x 243 #define ___last5(a, b, c, d, x) x 244 #define ___last6(a, b, c, d, e, x) x 245 #define ___last7(a, b, c, d, e, f, x) x 246 #define ___last8(a, b, c, d, e, f, g, x) x 247 #define ___last9(a, b, c, d, e, f, g, h, x) x 248 #define ___last10(a, b, c, d, e, f, g, h, i, x) x 249 #define ___last(...) ___apply(___last, ___narg(__VA_ARGS__))(__VA_ARGS__) 250 251 #define ___nolast2(a, _) a 252 #define ___nolast3(a, b, _) a, b 253 #define ___nolast4(a, b, c, _) a, b, c 254 #define ___nolast5(a, b, c, d, _) a, b, c, d 255 #define ___nolast6(a, b, c, d, e, _) a, b, c, d, e 256 #define ___nolast7(a, b, c, d, e, f, _) a, b, c, d, e, f 257 #define ___nolast8(a, b, c, d, e, f, g, _) a, b, c, d, e, f, g 258 #define ___nolast9(a, b, c, d, e, f, g, h, _) a, b, c, d, e, f, g, h 259 #define ___nolast10(a, b, c, d, e, f, g, h, i, _) a, b, c, d, e, f, g, h, i 260 #define ___nolast(...) ___apply(___nolast, ___narg(__VA_ARGS__))(__VA_ARGS__) 261 262 #define ___arrow1(a) a 263 #define ___arrow2(a, b) a->b 264 #define ___arrow3(a, b, c) a->b->c 265 #define ___arrow4(a, b, c, d) a->b->c->d 266 #define ___arrow5(a, b, c, d, e) a->b->c->d->e 267 #define ___arrow6(a, b, c, d, e, f) a->b->c->d->e->f 268 #define ___arrow7(a, b, c, d, e, f, g) a->b->c->d->e->f->g 269 #define ___arrow8(a, b, c, d, e, f, g, h) a->b->c->d->e->f->g->h 270 #define ___arrow9(a, b, c, d, e, f, g, h, i) a->b->c->d->e->f->g->h->i 271 #define ___arrow10(a, b, c, d, e, f, g, h, i, j) a->b->c->d->e->f->g->h->i->j 272 #define ___arrow(...) ___apply(___arrow, ___narg(__VA_ARGS__))(__VA_ARGS__) 273 274 #define ___type(...) typeof(___arrow(__VA_ARGS__)) 275 276 #define ___read(read_fn, dst, src_type, src, accessor) \ 277 read_fn((void *)(dst), sizeof(*(dst)), &((src_type)(src))->accessor) 278 279 /* "recursively" read a sequence of inner pointers using local __t var */ 280 #define ___rd_first(fn, src, a) ___read(fn, &__t, ___type(src), src, a); 281 #define ___rd_last(fn, ...) \ 282 ___read(fn, &__t, ___type(___nolast(__VA_ARGS__)), __t, ___last(__VA_ARGS__)); 283 #define ___rd_p1(fn, ...) const void *__t; ___rd_first(fn, __VA_ARGS__) 284 #define ___rd_p2(fn, ...) ___rd_p1(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 285 #define ___rd_p3(fn, ...) ___rd_p2(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 286 #define ___rd_p4(fn, ...) ___rd_p3(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 287 #define ___rd_p5(fn, ...) ___rd_p4(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 288 #define ___rd_p6(fn, ...) ___rd_p5(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 289 #define ___rd_p7(fn, ...) ___rd_p6(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 290 #define ___rd_p8(fn, ...) ___rd_p7(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 291 #define ___rd_p9(fn, ...) ___rd_p8(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__) 292 #define ___read_ptrs(fn, src, ...) \ 293 ___apply(___rd_p, ___narg(__VA_ARGS__))(fn, src, __VA_ARGS__) 294 295 #define ___core_read0(fn, fn_ptr, dst, src, a) \ 296 ___read(fn, dst, ___type(src), src, a); 297 #define ___core_readN(fn, fn_ptr, dst, src, ...) \ 298 ___read_ptrs(fn_ptr, src, ___nolast(__VA_ARGS__)) \ 299 ___read(fn, dst, ___type(src, ___nolast(__VA_ARGS__)), __t, \ 300 ___last(__VA_ARGS__)); 301 #define ___core_read(fn, fn_ptr, dst, src, a, ...) \ 302 ___apply(___core_read, ___empty(__VA_ARGS__))(fn, fn_ptr, dst, \ 303 src, a, ##__VA_ARGS__) 304 305 /* 306 * BPF_CORE_READ_INTO() is a more performance-conscious variant of 307 * BPF_CORE_READ(), in which final field is read into user-provided storage. 308 * See BPF_CORE_READ() below for more details on general usage. 309 */ 310 #define BPF_CORE_READ_INTO(dst, src, a, ...) ({ \ 311 ___core_read(bpf_core_read, bpf_core_read, \ 312 dst, (src), a, ##__VA_ARGS__) \ 313 }) 314 315 /* 316 * Variant of BPF_CORE_READ_INTO() for reading from user-space memory. 317 * 318 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 319 */ 320 #define BPF_CORE_READ_USER_INTO(dst, src, a, ...) ({ \ 321 ___core_read(bpf_core_read_user, bpf_core_read_user, \ 322 dst, (src), a, ##__VA_ARGS__) \ 323 }) 324 325 /* Non-CO-RE variant of BPF_CORE_READ_INTO() */ 326 #define BPF_PROBE_READ_INTO(dst, src, a, ...) ({ \ 327 ___core_read(bpf_probe_read, bpf_probe_read, \ 328 dst, (src), a, ##__VA_ARGS__) \ 329 }) 330 331 /* Non-CO-RE variant of BPF_CORE_READ_USER_INTO(). 332 * 333 * As no CO-RE relocations are emitted, source types can be arbitrary and are 334 * not restricted to kernel types only. 335 */ 336 #define BPF_PROBE_READ_USER_INTO(dst, src, a, ...) ({ \ 337 ___core_read(bpf_probe_read_user, bpf_probe_read_user, \ 338 dst, (src), a, ##__VA_ARGS__) \ 339 }) 340 341 /* 342 * BPF_CORE_READ_STR_INTO() does same "pointer chasing" as 343 * BPF_CORE_READ() for intermediate pointers, but then executes (and returns 344 * corresponding error code) bpf_core_read_str() for final string read. 345 */ 346 #define BPF_CORE_READ_STR_INTO(dst, src, a, ...) ({ \ 347 ___core_read(bpf_core_read_str, bpf_core_read, \ 348 dst, (src), a, ##__VA_ARGS__) \ 349 }) 350 351 /* 352 * Variant of BPF_CORE_READ_STR_INTO() for reading from user-space memory. 353 * 354 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. 355 */ 356 #define BPF_CORE_READ_USER_STR_INTO(dst, src, a, ...) ({ \ 357 ___core_read(bpf_core_read_user_str, bpf_core_read_user, \ 358 dst, (src), a, ##__VA_ARGS__) \ 359 }) 360 361 /* Non-CO-RE variant of BPF_CORE_READ_STR_INTO() */ 362 #define BPF_PROBE_READ_STR_INTO(dst, src, a, ...) ({ \ 363 ___core_read(bpf_probe_read_str, bpf_probe_read, \ 364 dst, (src), a, ##__VA_ARGS__) \ 365 }) 366 367 /* 368 * Non-CO-RE variant of BPF_CORE_READ_USER_STR_INTO(). 369 * 370 * As no CO-RE relocations are emitted, source types can be arbitrary and are 371 * not restricted to kernel types only. 372 */ 373 #define BPF_PROBE_READ_USER_STR_INTO(dst, src, a, ...) ({ \ 374 ___core_read(bpf_probe_read_user_str, bpf_probe_read_user, \ 375 dst, (src), a, ##__VA_ARGS__) \ 376 }) 377 378 /* 379 * BPF_CORE_READ() is used to simplify BPF CO-RE relocatable read, especially 380 * when there are few pointer chasing steps. 381 * E.g., what in non-BPF world (or in BPF w/ BCC) would be something like: 382 * int x = s->a.b.c->d.e->f->g; 383 * can be succinctly achieved using BPF_CORE_READ as: 384 * int x = BPF_CORE_READ(s, a.b.c, d.e, f, g); 385 * 386 * BPF_CORE_READ will decompose above statement into 4 bpf_core_read (BPF 387 * CO-RE relocatable bpf_probe_read_kernel() wrapper) calls, logically 388 * equivalent to: 389 * 1. const void *__t = s->a.b.c; 390 * 2. __t = __t->d.e; 391 * 3. __t = __t->f; 392 * 4. return __t->g; 393 * 394 * Equivalence is logical, because there is a heavy type casting/preservation 395 * involved, as well as all the reads are happening through 396 * bpf_probe_read_kernel() calls using __builtin_preserve_access_index() to 397 * emit CO-RE relocations. 398 * 399 * N.B. Only up to 9 "field accessors" are supported, which should be more 400 * than enough for any practical purpose. 401 */ 402 #define BPF_CORE_READ(src, a, ...) ({ \ 403 ___type((src), a, ##__VA_ARGS__) __r; \ 404 BPF_CORE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 405 __r; \ 406 }) 407 408 /* 409 * Variant of BPF_CORE_READ() for reading from user-space memory. 410 * 411 * NOTE: all the source types involved are still *kernel types* and need to 412 * exist in kernel (or kernel module) BTF, otherwise CO-RE relocation will 413 * fail. Custom user types are not relocatable with CO-RE. 414 * The typical situation in which BPF_CORE_READ_USER() might be used is to 415 * read kernel UAPI types from the user-space memory passed in as a syscall 416 * input argument. 417 */ 418 #define BPF_CORE_READ_USER(src, a, ...) ({ \ 419 ___type((src), a, ##__VA_ARGS__) __r; \ 420 BPF_CORE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 421 __r; \ 422 }) 423 424 /* Non-CO-RE variant of BPF_CORE_READ() */ 425 #define BPF_PROBE_READ(src, a, ...) ({ \ 426 ___type((src), a, ##__VA_ARGS__) __r; \ 427 BPF_PROBE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \ 428 __r; \ 429 }) 430 431 /* 432 * Non-CO-RE variant of BPF_CORE_READ_USER(). 433 * 434 * As no CO-RE relocations are emitted, source types can be arbitrary and are 435 * not restricted to kernel types only. 436 */ 437 #define BPF_PROBE_READ_USER(src, a, ...) ({ \ 438 ___type((src), a, ##__VA_ARGS__) __r; \ 439 BPF_PROBE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \ 440 __r; \ 441 }) 442 443 #endif 444 445