1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */ 2 3 /* 4 * Internal libbpf helpers. 5 * 6 * Copyright (c) 2019 Facebook 7 */ 8 9 #ifndef __LIBBPF_LIBBPF_INTERNAL_H 10 #define __LIBBPF_LIBBPF_INTERNAL_H 11 12 #include <stdlib.h> 13 #include <limits.h> 14 #include <errno.h> 15 #include <linux/err.h> 16 #include <fcntl.h> 17 #include <unistd.h> 18 #include "libbpf_legacy.h" 19 #include "relo_core.h" 20 21 /* make sure libbpf doesn't use kernel-only integer typedefs */ 22 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64 23 24 /* prevent accidental re-addition of reallocarray() */ 25 #pragma GCC poison reallocarray 26 27 #include "libbpf.h" 28 #include "btf.h" 29 30 #ifndef EM_BPF 31 #define EM_BPF 247 32 #endif 33 34 #ifndef R_BPF_64_64 35 #define R_BPF_64_64 1 36 #endif 37 #ifndef R_BPF_64_ABS64 38 #define R_BPF_64_ABS64 2 39 #endif 40 #ifndef R_BPF_64_ABS32 41 #define R_BPF_64_ABS32 3 42 #endif 43 #ifndef R_BPF_64_32 44 #define R_BPF_64_32 10 45 #endif 46 47 #ifndef SHT_LLVM_ADDRSIG 48 #define SHT_LLVM_ADDRSIG 0x6FFF4C03 49 #endif 50 51 /* if libelf is old and doesn't support mmap(), fall back to read() */ 52 #ifndef ELF_C_READ_MMAP 53 #define ELF_C_READ_MMAP ELF_C_READ 54 #endif 55 56 /* Older libelf all end up in this expression, for both 32 and 64 bit */ 57 #ifndef ELF64_ST_VISIBILITY 58 #define ELF64_ST_VISIBILITY(o) ((o) & 0x03) 59 #endif 60 61 #define BTF_INFO_ENC(kind, kind_flag, vlen) \ 62 ((!!(kind_flag) << 31) | ((kind) << 24) | ((vlen) & BTF_MAX_VLEN)) 63 #define BTF_TYPE_ENC(name, info, size_or_type) (name), (info), (size_or_type) 64 #define BTF_INT_ENC(encoding, bits_offset, nr_bits) \ 65 ((encoding) << 24 | (bits_offset) << 16 | (nr_bits)) 66 #define BTF_TYPE_INT_ENC(name, encoding, bits_offset, bits, sz) \ 67 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_INT, 0, 0), sz), \ 68 BTF_INT_ENC(encoding, bits_offset, bits) 69 #define BTF_MEMBER_ENC(name, type, bits_offset) (name), (type), (bits_offset) 70 #define BTF_PARAM_ENC(name, type) (name), (type) 71 #define BTF_VAR_SECINFO_ENC(type, offset, size) (type), (offset), (size) 72 #define BTF_TYPE_FLOAT_ENC(name, sz) \ 73 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_FLOAT, 0, 0), sz) 74 #define BTF_TYPE_DECL_TAG_ENC(value, type, component_idx) \ 75 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_DECL_TAG, 0, 0), type), (component_idx) 76 #define BTF_TYPE_TYPE_TAG_ENC(value, type) \ 77 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_TYPE_TAG, 0, 0), type) 78 79 #ifndef likely 80 #define likely(x) __builtin_expect(!!(x), 1) 81 #endif 82 #ifndef unlikely 83 #define unlikely(x) __builtin_expect(!!(x), 0) 84 #endif 85 #ifndef min 86 # define min(x, y) ((x) < (y) ? (x) : (y)) 87 #endif 88 #ifndef max 89 # define max(x, y) ((x) < (y) ? (y) : (x)) 90 #endif 91 #ifndef offsetofend 92 # define offsetofend(TYPE, FIELD) \ 93 (offsetof(TYPE, FIELD) + sizeof(((TYPE *)0)->FIELD)) 94 #endif 95 96 /* Check whether a string `str` has prefix `pfx`, regardless if `pfx` is 97 * a string literal known at compilation time or char * pointer known only at 98 * runtime. 99 */ 100 #define str_has_pfx(str, pfx) \ 101 (strncmp(str, pfx, __builtin_constant_p(pfx) ? sizeof(pfx) - 1 : strlen(pfx)) == 0) 102 103 /* Symbol versioning is different between static and shared library. 104 * Properly versioned symbols are needed for shared library, but 105 * only the symbol of the new version is needed for static library. 106 * Starting with GNU C 10, use symver attribute instead of .symver assembler 107 * directive, which works better with GCC LTO builds. 108 */ 109 #if defined(SHARED) && defined(__GNUC__) && __GNUC__ >= 10 110 111 #define DEFAULT_VERSION(internal_name, api_name, version) \ 112 __attribute__((symver(#api_name "@@" #version))) 113 #define COMPAT_VERSION(internal_name, api_name, version) \ 114 __attribute__((symver(#api_name "@" #version))) 115 116 #elif defined(SHARED) 117 118 #define COMPAT_VERSION(internal_name, api_name, version) \ 119 asm(".symver " #internal_name "," #api_name "@" #version); 120 #define DEFAULT_VERSION(internal_name, api_name, version) \ 121 asm(".symver " #internal_name "," #api_name "@@" #version); 122 123 #else /* !SHARED */ 124 125 #define COMPAT_VERSION(internal_name, api_name, version) 126 #define DEFAULT_VERSION(internal_name, api_name, version) \ 127 extern typeof(internal_name) api_name \ 128 __attribute__((alias(#internal_name))); 129 130 #endif 131 132 extern void libbpf_print(enum libbpf_print_level level, 133 const char *format, ...) 134 __attribute__((format(printf, 2, 3))); 135 136 #define __pr(level, fmt, ...) \ 137 do { \ 138 libbpf_print(level, "libbpf: " fmt, ##__VA_ARGS__); \ 139 } while (0) 140 141 #define pr_warn(fmt, ...) __pr(LIBBPF_WARN, fmt, ##__VA_ARGS__) 142 #define pr_info(fmt, ...) __pr(LIBBPF_INFO, fmt, ##__VA_ARGS__) 143 #define pr_debug(fmt, ...) __pr(LIBBPF_DEBUG, fmt, ##__VA_ARGS__) 144 145 #ifndef __has_builtin 146 #define __has_builtin(x) 0 147 #endif 148 /* 149 * Re-implement glibc's reallocarray() for libbpf internal-only use. 150 * reallocarray(), unfortunately, is not available in all versions of glibc, 151 * so requires extra feature detection and using reallocarray() stub from 152 * <tools/libc_compat.h> and COMPAT_NEED_REALLOCARRAY. All this complicates 153 * build of libbpf unnecessarily and is just a maintenance burden. Instead, 154 * it's trivial to implement libbpf-specific internal version and use it 155 * throughout libbpf. 156 */ 157 static inline void *libbpf_reallocarray(void *ptr, size_t nmemb, size_t size) 158 { 159 size_t total; 160 161 #if __has_builtin(__builtin_mul_overflow) 162 if (unlikely(__builtin_mul_overflow(nmemb, size, &total))) 163 return NULL; 164 #else 165 if (size == 0 || nmemb > ULONG_MAX / size) 166 return NULL; 167 total = nmemb * size; 168 #endif 169 return realloc(ptr, total); 170 } 171 172 /* Copy up to sz - 1 bytes from zero-terminated src string and ensure that dst 173 * is zero-terminated string no matter what (unless sz == 0, in which case 174 * it's a no-op). It's conceptually close to FreeBSD's strlcpy(), but differs 175 * in what is returned. Given this is internal helper, it's trivial to extend 176 * this, when necessary. Use this instead of strncpy inside libbpf source code. 177 */ 178 static inline void libbpf_strlcpy(char *dst, const char *src, size_t sz) 179 { 180 size_t i; 181 182 if (sz == 0) 183 return; 184 185 sz--; 186 for (i = 0; i < sz && src[i]; i++) 187 dst[i] = src[i]; 188 dst[i] = '\0'; 189 } 190 191 __u32 get_kernel_version(void); 192 193 struct btf; 194 struct btf_type; 195 196 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id); 197 const char *btf_kind_str(const struct btf_type *t); 198 const struct btf_type *skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id); 199 200 static inline enum btf_func_linkage btf_func_linkage(const struct btf_type *t) 201 { 202 return (enum btf_func_linkage)(int)btf_vlen(t); 203 } 204 205 static inline __u32 btf_type_info(int kind, int vlen, int kflag) 206 { 207 return (kflag << 31) | (kind << 24) | vlen; 208 } 209 210 enum map_def_parts { 211 MAP_DEF_MAP_TYPE = 0x001, 212 MAP_DEF_KEY_TYPE = 0x002, 213 MAP_DEF_KEY_SIZE = 0x004, 214 MAP_DEF_VALUE_TYPE = 0x008, 215 MAP_DEF_VALUE_SIZE = 0x010, 216 MAP_DEF_MAX_ENTRIES = 0x020, 217 MAP_DEF_MAP_FLAGS = 0x040, 218 MAP_DEF_NUMA_NODE = 0x080, 219 MAP_DEF_PINNING = 0x100, 220 MAP_DEF_INNER_MAP = 0x200, 221 MAP_DEF_MAP_EXTRA = 0x400, 222 223 MAP_DEF_ALL = 0x7ff, /* combination of all above */ 224 }; 225 226 struct btf_map_def { 227 enum map_def_parts parts; 228 __u32 map_type; 229 __u32 key_type_id; 230 __u32 key_size; 231 __u32 value_type_id; 232 __u32 value_size; 233 __u32 max_entries; 234 __u32 map_flags; 235 __u32 numa_node; 236 __u32 pinning; 237 __u64 map_extra; 238 }; 239 240 int parse_btf_map_def(const char *map_name, struct btf *btf, 241 const struct btf_type *def_t, bool strict, 242 struct btf_map_def *map_def, struct btf_map_def *inner_def); 243 244 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz, 245 size_t cur_cnt, size_t max_cnt, size_t add_cnt); 246 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt); 247 248 static inline bool libbpf_is_mem_zeroed(const char *p, ssize_t len) 249 { 250 while (len > 0) { 251 if (*p) 252 return false; 253 p++; 254 len--; 255 } 256 return true; 257 } 258 259 static inline bool libbpf_validate_opts(const char *opts, 260 size_t opts_sz, size_t user_sz, 261 const char *type_name) 262 { 263 if (user_sz < sizeof(size_t)) { 264 pr_warn("%s size (%zu) is too small\n", type_name, user_sz); 265 return false; 266 } 267 if (!libbpf_is_mem_zeroed(opts + opts_sz, (ssize_t)user_sz - opts_sz)) { 268 pr_warn("%s has non-zero extra bytes\n", type_name); 269 return false; 270 } 271 return true; 272 } 273 274 #define OPTS_VALID(opts, type) \ 275 (!(opts) || libbpf_validate_opts((const char *)opts, \ 276 offsetofend(struct type, \ 277 type##__last_field), \ 278 (opts)->sz, #type)) 279 #define OPTS_HAS(opts, field) \ 280 ((opts) && opts->sz >= offsetofend(typeof(*(opts)), field)) 281 #define OPTS_GET(opts, field, fallback_value) \ 282 (OPTS_HAS(opts, field) ? (opts)->field : fallback_value) 283 #define OPTS_SET(opts, field, value) \ 284 do { \ 285 if (OPTS_HAS(opts, field)) \ 286 (opts)->field = value; \ 287 } while (0) 288 289 #define OPTS_ZEROED(opts, last_nonzero_field) \ 290 ({ \ 291 ssize_t __off = offsetofend(typeof(*(opts)), last_nonzero_field); \ 292 !(opts) || libbpf_is_mem_zeroed((const void *)opts + __off, \ 293 (opts)->sz - __off); \ 294 }) 295 296 enum kern_feature_id { 297 /* v4.14: kernel support for program & map names. */ 298 FEAT_PROG_NAME, 299 /* v5.2: kernel support for global data sections. */ 300 FEAT_GLOBAL_DATA, 301 /* BTF support */ 302 FEAT_BTF, 303 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */ 304 FEAT_BTF_FUNC, 305 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */ 306 FEAT_BTF_DATASEC, 307 /* BTF_FUNC_GLOBAL is supported */ 308 FEAT_BTF_GLOBAL_FUNC, 309 /* BPF_F_MMAPABLE is supported for arrays */ 310 FEAT_ARRAY_MMAP, 311 /* kernel support for expected_attach_type in BPF_PROG_LOAD */ 312 FEAT_EXP_ATTACH_TYPE, 313 /* bpf_probe_read_{kernel,user}[_str] helpers */ 314 FEAT_PROBE_READ_KERN, 315 /* BPF_PROG_BIND_MAP is supported */ 316 FEAT_PROG_BIND_MAP, 317 /* Kernel support for module BTFs */ 318 FEAT_MODULE_BTF, 319 /* BTF_KIND_FLOAT support */ 320 FEAT_BTF_FLOAT, 321 /* BPF perf link support */ 322 FEAT_PERF_LINK, 323 /* BTF_KIND_DECL_TAG support */ 324 FEAT_BTF_DECL_TAG, 325 /* BTF_KIND_TYPE_TAG support */ 326 FEAT_BTF_TYPE_TAG, 327 /* memcg-based accounting for BPF maps and progs */ 328 FEAT_MEMCG_ACCOUNT, 329 __FEAT_CNT, 330 }; 331 332 int probe_memcg_account(void); 333 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id); 334 int bump_rlimit_memlock(void); 335 336 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz); 337 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz); 338 int libbpf__load_raw_btf(const char *raw_types, size_t types_len, 339 const char *str_sec, size_t str_len); 340 int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level); 341 342 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf); 343 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, 344 const char **prefix, int *kind); 345 346 struct btf_ext_info { 347 /* 348 * info points to the individual info section (e.g. func_info and 349 * line_info) from the .BTF.ext. It does not include the __u32 rec_size. 350 */ 351 void *info; 352 __u32 rec_size; 353 __u32 len; 354 }; 355 356 #define for_each_btf_ext_sec(seg, sec) \ 357 for (sec = (seg)->info; \ 358 (void *)sec < (seg)->info + (seg)->len; \ 359 sec = (void *)sec + sizeof(struct btf_ext_info_sec) + \ 360 (seg)->rec_size * sec->num_info) 361 362 #define for_each_btf_ext_rec(seg, sec, i, rec) \ 363 for (i = 0, rec = (void *)&(sec)->data; \ 364 i < (sec)->num_info; \ 365 i++, rec = (void *)rec + (seg)->rec_size) 366 367 /* 368 * The .BTF.ext ELF section layout defined as 369 * struct btf_ext_header 370 * func_info subsection 371 * 372 * The func_info subsection layout: 373 * record size for struct bpf_func_info in the func_info subsection 374 * struct btf_sec_func_info for section #1 375 * a list of bpf_func_info records for section #1 376 * where struct bpf_func_info mimics one in include/uapi/linux/bpf.h 377 * but may not be identical 378 * struct btf_sec_func_info for section #2 379 * a list of bpf_func_info records for section #2 380 * ...... 381 * 382 * Note that the bpf_func_info record size in .BTF.ext may not 383 * be the same as the one defined in include/uapi/linux/bpf.h. 384 * The loader should ensure that record_size meets minimum 385 * requirement and pass the record as is to the kernel. The 386 * kernel will handle the func_info properly based on its contents. 387 */ 388 struct btf_ext_header { 389 __u16 magic; 390 __u8 version; 391 __u8 flags; 392 __u32 hdr_len; 393 394 /* All offsets are in bytes relative to the end of this header */ 395 __u32 func_info_off; 396 __u32 func_info_len; 397 __u32 line_info_off; 398 __u32 line_info_len; 399 400 /* optional part of .BTF.ext header */ 401 __u32 core_relo_off; 402 __u32 core_relo_len; 403 }; 404 405 struct btf_ext { 406 union { 407 struct btf_ext_header *hdr; 408 void *data; 409 }; 410 struct btf_ext_info func_info; 411 struct btf_ext_info line_info; 412 struct btf_ext_info core_relo_info; 413 __u32 data_size; 414 }; 415 416 struct btf_ext_info_sec { 417 __u32 sec_name_off; 418 __u32 num_info; 419 /* Followed by num_info * record_size number of bytes */ 420 __u8 data[]; 421 }; 422 423 /* The minimum bpf_func_info checked by the loader */ 424 struct bpf_func_info_min { 425 __u32 insn_off; 426 __u32 type_id; 427 }; 428 429 /* The minimum bpf_line_info checked by the loader */ 430 struct bpf_line_info_min { 431 __u32 insn_off; 432 __u32 file_name_off; 433 __u32 line_off; 434 __u32 line_col; 435 }; 436 437 438 typedef int (*type_id_visit_fn)(__u32 *type_id, void *ctx); 439 typedef int (*str_off_visit_fn)(__u32 *str_off, void *ctx); 440 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx); 441 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx); 442 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx); 443 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx); 444 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name, 445 __u32 kind); 446 447 extern enum libbpf_strict_mode libbpf_mode; 448 449 /* handle direct returned errors */ 450 static inline int libbpf_err(int ret) 451 { 452 if (ret < 0) 453 errno = -ret; 454 return ret; 455 } 456 457 /* handle errno-based (e.g., syscall or libc) errors according to libbpf's 458 * strict mode settings 459 */ 460 static inline int libbpf_err_errno(int ret) 461 { 462 if (libbpf_mode & LIBBPF_STRICT_DIRECT_ERRS) 463 /* errno is already assumed to be set on error */ 464 return ret < 0 ? -errno : ret; 465 466 /* legacy: on error return -1 directly and don't touch errno */ 467 return ret; 468 } 469 470 /* handle error for pointer-returning APIs, err is assumed to be < 0 always */ 471 static inline void *libbpf_err_ptr(int err) 472 { 473 /* set errno on error, this doesn't break anything */ 474 errno = -err; 475 476 if (libbpf_mode & LIBBPF_STRICT_CLEAN_PTRS) 477 return NULL; 478 479 /* legacy: encode err as ptr */ 480 return ERR_PTR(err); 481 } 482 483 /* handle pointer-returning APIs' error handling */ 484 static inline void *libbpf_ptr(void *ret) 485 { 486 /* set errno on error, this doesn't break anything */ 487 if (IS_ERR(ret)) 488 errno = -PTR_ERR(ret); 489 490 if (libbpf_mode & LIBBPF_STRICT_CLEAN_PTRS) 491 return IS_ERR(ret) ? NULL : ret; 492 493 /* legacy: pass-through original pointer */ 494 return ret; 495 } 496 497 static inline bool str_is_empty(const char *s) 498 { 499 return !s || !s[0]; 500 } 501 502 static inline bool is_ldimm64_insn(struct bpf_insn *insn) 503 { 504 return insn->code == (BPF_LD | BPF_IMM | BPF_DW); 505 } 506 507 /* if fd is stdin, stdout, or stderr, dup to a fd greater than 2 508 * Takes ownership of the fd passed in, and closes it if calling 509 * fcntl(fd, F_DUPFD_CLOEXEC, 3). 510 */ 511 static inline int ensure_good_fd(int fd) 512 { 513 int old_fd = fd, saved_errno; 514 515 if (fd < 0) 516 return fd; 517 if (fd < 3) { 518 fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); 519 saved_errno = errno; 520 close(old_fd); 521 if (fd < 0) { 522 pr_warn("failed to dup FD %d to FD > 2: %d\n", old_fd, -saved_errno); 523 errno = saved_errno; 524 } 525 } 526 return fd; 527 } 528 529 #endif /* __LIBBPF_LIBBPF_INTERNAL_H */ 530