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 "relo_core.h" 19 20 /* make sure libbpf doesn't use kernel-only integer typedefs */ 21 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64 22 23 /* prevent accidental re-addition of reallocarray() */ 24 #pragma GCC poison reallocarray 25 26 #include "libbpf.h" 27 #include "btf.h" 28 29 #ifndef EM_BPF 30 #define EM_BPF 247 31 #endif 32 33 #ifndef R_BPF_64_64 34 #define R_BPF_64_64 1 35 #endif 36 #ifndef R_BPF_64_ABS64 37 #define R_BPF_64_ABS64 2 38 #endif 39 #ifndef R_BPF_64_ABS32 40 #define R_BPF_64_ABS32 3 41 #endif 42 #ifndef R_BPF_64_32 43 #define R_BPF_64_32 10 44 #endif 45 46 #ifndef SHT_LLVM_ADDRSIG 47 #define SHT_LLVM_ADDRSIG 0x6FFF4C03 48 #endif 49 50 /* if libelf is old and doesn't support mmap(), fall back to read() */ 51 #ifndef ELF_C_READ_MMAP 52 #define ELF_C_READ_MMAP ELF_C_READ 53 #endif 54 55 /* Older libelf all end up in this expression, for both 32 and 64 bit */ 56 #ifndef ELF64_ST_VISIBILITY 57 #define ELF64_ST_VISIBILITY(o) ((o) & 0x03) 58 #endif 59 60 #define BTF_INFO_ENC(kind, kind_flag, vlen) \ 61 ((!!(kind_flag) << 31) | ((kind) << 24) | ((vlen) & BTF_MAX_VLEN)) 62 #define BTF_TYPE_ENC(name, info, size_or_type) (name), (info), (size_or_type) 63 #define BTF_INT_ENC(encoding, bits_offset, nr_bits) \ 64 ((encoding) << 24 | (bits_offset) << 16 | (nr_bits)) 65 #define BTF_TYPE_INT_ENC(name, encoding, bits_offset, bits, sz) \ 66 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_INT, 0, 0), sz), \ 67 BTF_INT_ENC(encoding, bits_offset, bits) 68 #define BTF_MEMBER_ENC(name, type, bits_offset) (name), (type), (bits_offset) 69 #define BTF_PARAM_ENC(name, type) (name), (type) 70 #define BTF_VAR_SECINFO_ENC(type, offset, size) (type), (offset), (size) 71 #define BTF_TYPE_FLOAT_ENC(name, sz) \ 72 BTF_TYPE_ENC(name, BTF_INFO_ENC(BTF_KIND_FLOAT, 0, 0), sz) 73 #define BTF_TYPE_DECL_TAG_ENC(value, type, component_idx) \ 74 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_DECL_TAG, 0, 0), type), (component_idx) 75 #define BTF_TYPE_TYPE_TAG_ENC(value, type) \ 76 BTF_TYPE_ENC(value, BTF_INFO_ENC(BTF_KIND_TYPE_TAG, 0, 0), type) 77 78 #ifndef likely 79 #define likely(x) __builtin_expect(!!(x), 1) 80 #endif 81 #ifndef unlikely 82 #define unlikely(x) __builtin_expect(!!(x), 0) 83 #endif 84 #ifndef min 85 # define min(x, y) ((x) < (y) ? (x) : (y)) 86 #endif 87 #ifndef max 88 # define max(x, y) ((x) < (y) ? (y) : (x)) 89 #endif 90 #ifndef offsetofend 91 # define offsetofend(TYPE, FIELD) \ 92 (offsetof(TYPE, FIELD) + sizeof(((TYPE *)0)->FIELD)) 93 #endif 94 #ifndef __alias 95 #define __alias(symbol) __attribute__((alias(#symbol))) 96 #endif 97 98 /* Check whether a string `str` has prefix `pfx`, regardless if `pfx` is 99 * a string literal known at compilation time or char * pointer known only at 100 * runtime. 101 */ 102 #define str_has_pfx(str, pfx) \ 103 (strncmp(str, pfx, __builtin_constant_p(pfx) ? sizeof(pfx) - 1 : strlen(pfx)) == 0) 104 105 /* suffix check */ 106 static inline bool str_has_sfx(const char *str, const char *sfx) 107 { 108 size_t str_len = strlen(str); 109 size_t sfx_len = strlen(sfx); 110 111 if (sfx_len > str_len) 112 return false; 113 return strcmp(str + str_len - sfx_len, sfx) == 0; 114 } 115 116 /* Symbol versioning is different between static and shared library. 117 * Properly versioned symbols are needed for shared library, but 118 * only the symbol of the new version is needed for static library. 119 * Starting with GNU C 10, use symver attribute instead of .symver assembler 120 * directive, which works better with GCC LTO builds. 121 */ 122 #if defined(SHARED) && defined(__GNUC__) && __GNUC__ >= 10 123 124 #define DEFAULT_VERSION(internal_name, api_name, version) \ 125 __attribute__((symver(#api_name "@@" #version))) 126 #define COMPAT_VERSION(internal_name, api_name, version) \ 127 __attribute__((symver(#api_name "@" #version))) 128 129 #elif defined(SHARED) 130 131 #define COMPAT_VERSION(internal_name, api_name, version) \ 132 asm(".symver " #internal_name "," #api_name "@" #version); 133 #define DEFAULT_VERSION(internal_name, api_name, version) \ 134 asm(".symver " #internal_name "," #api_name "@@" #version); 135 136 #else /* !SHARED */ 137 138 #define COMPAT_VERSION(internal_name, api_name, version) 139 #define DEFAULT_VERSION(internal_name, api_name, version) \ 140 extern typeof(internal_name) api_name \ 141 __attribute__((alias(#internal_name))); 142 143 #endif 144 145 extern void libbpf_print(enum libbpf_print_level level, 146 const char *format, ...) 147 __attribute__((format(printf, 2, 3))); 148 149 #define __pr(level, fmt, ...) \ 150 do { \ 151 libbpf_print(level, "libbpf: " fmt, ##__VA_ARGS__); \ 152 } while (0) 153 154 #define pr_warn(fmt, ...) __pr(LIBBPF_WARN, fmt, ##__VA_ARGS__) 155 #define pr_info(fmt, ...) __pr(LIBBPF_INFO, fmt, ##__VA_ARGS__) 156 #define pr_debug(fmt, ...) __pr(LIBBPF_DEBUG, fmt, ##__VA_ARGS__) 157 158 #ifndef __has_builtin 159 #define __has_builtin(x) 0 160 #endif 161 162 struct bpf_link { 163 int (*detach)(struct bpf_link *link); 164 void (*dealloc)(struct bpf_link *link); 165 char *pin_path; /* NULL, if not pinned */ 166 int fd; /* hook FD, -1 if not applicable */ 167 bool disconnected; 168 }; 169 170 /* 171 * Re-implement glibc's reallocarray() for libbpf internal-only use. 172 * reallocarray(), unfortunately, is not available in all versions of glibc, 173 * so requires extra feature detection and using reallocarray() stub from 174 * <tools/libc_compat.h> and COMPAT_NEED_REALLOCARRAY. All this complicates 175 * build of libbpf unnecessarily and is just a maintenance burden. Instead, 176 * it's trivial to implement libbpf-specific internal version and use it 177 * throughout libbpf. 178 */ 179 static inline void *libbpf_reallocarray(void *ptr, size_t nmemb, size_t size) 180 { 181 size_t total; 182 183 #if __has_builtin(__builtin_mul_overflow) 184 if (unlikely(__builtin_mul_overflow(nmemb, size, &total))) 185 return NULL; 186 #else 187 if (size == 0 || nmemb > ULONG_MAX / size) 188 return NULL; 189 total = nmemb * size; 190 #endif 191 return realloc(ptr, total); 192 } 193 194 /* Copy up to sz - 1 bytes from zero-terminated src string and ensure that dst 195 * is zero-terminated string no matter what (unless sz == 0, in which case 196 * it's a no-op). It's conceptually close to FreeBSD's strlcpy(), but differs 197 * in what is returned. Given this is internal helper, it's trivial to extend 198 * this, when necessary. Use this instead of strncpy inside libbpf source code. 199 */ 200 static inline void libbpf_strlcpy(char *dst, const char *src, size_t sz) 201 { 202 size_t i; 203 204 if (sz == 0) 205 return; 206 207 sz--; 208 for (i = 0; i < sz && src[i]; i++) 209 dst[i] = src[i]; 210 dst[i] = '\0'; 211 } 212 213 __u32 get_kernel_version(void); 214 215 struct btf; 216 struct btf_type; 217 218 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id); 219 const char *btf_kind_str(const struct btf_type *t); 220 const struct btf_type *skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id); 221 222 static inline enum btf_func_linkage btf_func_linkage(const struct btf_type *t) 223 { 224 return (enum btf_func_linkage)(int)btf_vlen(t); 225 } 226 227 static inline __u32 btf_type_info(int kind, int vlen, int kflag) 228 { 229 return (kflag << 31) | (kind << 24) | vlen; 230 } 231 232 enum map_def_parts { 233 MAP_DEF_MAP_TYPE = 0x001, 234 MAP_DEF_KEY_TYPE = 0x002, 235 MAP_DEF_KEY_SIZE = 0x004, 236 MAP_DEF_VALUE_TYPE = 0x008, 237 MAP_DEF_VALUE_SIZE = 0x010, 238 MAP_DEF_MAX_ENTRIES = 0x020, 239 MAP_DEF_MAP_FLAGS = 0x040, 240 MAP_DEF_NUMA_NODE = 0x080, 241 MAP_DEF_PINNING = 0x100, 242 MAP_DEF_INNER_MAP = 0x200, 243 MAP_DEF_MAP_EXTRA = 0x400, 244 245 MAP_DEF_ALL = 0x7ff, /* combination of all above */ 246 }; 247 248 struct btf_map_def { 249 enum map_def_parts parts; 250 __u32 map_type; 251 __u32 key_type_id; 252 __u32 key_size; 253 __u32 value_type_id; 254 __u32 value_size; 255 __u32 max_entries; 256 __u32 map_flags; 257 __u32 numa_node; 258 __u32 pinning; 259 __u64 map_extra; 260 }; 261 262 int parse_btf_map_def(const char *map_name, struct btf *btf, 263 const struct btf_type *def_t, bool strict, 264 struct btf_map_def *map_def, struct btf_map_def *inner_def); 265 266 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz, 267 size_t cur_cnt, size_t max_cnt, size_t add_cnt); 268 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt); 269 270 static inline bool libbpf_is_mem_zeroed(const char *p, ssize_t len) 271 { 272 while (len > 0) { 273 if (*p) 274 return false; 275 p++; 276 len--; 277 } 278 return true; 279 } 280 281 static inline bool libbpf_validate_opts(const char *opts, 282 size_t opts_sz, size_t user_sz, 283 const char *type_name) 284 { 285 if (user_sz < sizeof(size_t)) { 286 pr_warn("%s size (%zu) is too small\n", type_name, user_sz); 287 return false; 288 } 289 if (!libbpf_is_mem_zeroed(opts + opts_sz, (ssize_t)user_sz - opts_sz)) { 290 pr_warn("%s has non-zero extra bytes\n", type_name); 291 return false; 292 } 293 return true; 294 } 295 296 #define OPTS_VALID(opts, type) \ 297 (!(opts) || libbpf_validate_opts((const char *)opts, \ 298 offsetofend(struct type, \ 299 type##__last_field), \ 300 (opts)->sz, #type)) 301 #define OPTS_HAS(opts, field) \ 302 ((opts) && opts->sz >= offsetofend(typeof(*(opts)), field)) 303 #define OPTS_GET(opts, field, fallback_value) \ 304 (OPTS_HAS(opts, field) ? (opts)->field : fallback_value) 305 #define OPTS_SET(opts, field, value) \ 306 do { \ 307 if (OPTS_HAS(opts, field)) \ 308 (opts)->field = value; \ 309 } while (0) 310 311 #define OPTS_ZEROED(opts, last_nonzero_field) \ 312 ({ \ 313 ssize_t __off = offsetofend(typeof(*(opts)), last_nonzero_field); \ 314 !(opts) || libbpf_is_mem_zeroed((const void *)opts + __off, \ 315 (opts)->sz - __off); \ 316 }) 317 318 enum kern_feature_id { 319 /* v4.14: kernel support for program & map names. */ 320 FEAT_PROG_NAME, 321 /* v5.2: kernel support for global data sections. */ 322 FEAT_GLOBAL_DATA, 323 /* BTF support */ 324 FEAT_BTF, 325 /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */ 326 FEAT_BTF_FUNC, 327 /* BTF_KIND_VAR and BTF_KIND_DATASEC support */ 328 FEAT_BTF_DATASEC, 329 /* BTF_FUNC_GLOBAL is supported */ 330 FEAT_BTF_GLOBAL_FUNC, 331 /* BPF_F_MMAPABLE is supported for arrays */ 332 FEAT_ARRAY_MMAP, 333 /* kernel support for expected_attach_type in BPF_PROG_LOAD */ 334 FEAT_EXP_ATTACH_TYPE, 335 /* bpf_probe_read_{kernel,user}[_str] helpers */ 336 FEAT_PROBE_READ_KERN, 337 /* BPF_PROG_BIND_MAP is supported */ 338 FEAT_PROG_BIND_MAP, 339 /* Kernel support for module BTFs */ 340 FEAT_MODULE_BTF, 341 /* BTF_KIND_FLOAT support */ 342 FEAT_BTF_FLOAT, 343 /* BPF perf link support */ 344 FEAT_PERF_LINK, 345 /* BTF_KIND_DECL_TAG support */ 346 FEAT_BTF_DECL_TAG, 347 /* BTF_KIND_TYPE_TAG support */ 348 FEAT_BTF_TYPE_TAG, 349 /* memcg-based accounting for BPF maps and progs */ 350 FEAT_MEMCG_ACCOUNT, 351 /* BPF cookie (bpf_get_attach_cookie() BPF helper) support */ 352 FEAT_BPF_COOKIE, 353 /* BTF_KIND_ENUM64 support and BTF_KIND_ENUM kflag support */ 354 FEAT_BTF_ENUM64, 355 /* Kernel uses syscall wrapper (CONFIG_ARCH_HAS_SYSCALL_WRAPPER) */ 356 FEAT_SYSCALL_WRAPPER, 357 __FEAT_CNT, 358 }; 359 360 int probe_memcg_account(void); 361 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id); 362 int bump_rlimit_memlock(void); 363 364 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz); 365 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz); 366 int libbpf__load_raw_btf(const char *raw_types, size_t types_len, 367 const char *str_sec, size_t str_len); 368 int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level); 369 370 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf); 371 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type, 372 const char **prefix, int *kind); 373 374 struct btf_ext_info { 375 /* 376 * info points to the individual info section (e.g. func_info and 377 * line_info) from the .BTF.ext. It does not include the __u32 rec_size. 378 */ 379 void *info; 380 __u32 rec_size; 381 __u32 len; 382 /* optional (maintained internally by libbpf) mapping between .BTF.ext 383 * section and corresponding ELF section. This is used to join 384 * information like CO-RE relocation records with corresponding BPF 385 * programs defined in ELF sections 386 */ 387 __u32 *sec_idxs; 388 int sec_cnt; 389 }; 390 391 #define for_each_btf_ext_sec(seg, sec) \ 392 for (sec = (seg)->info; \ 393 (void *)sec < (seg)->info + (seg)->len; \ 394 sec = (void *)sec + sizeof(struct btf_ext_info_sec) + \ 395 (seg)->rec_size * sec->num_info) 396 397 #define for_each_btf_ext_rec(seg, sec, i, rec) \ 398 for (i = 0, rec = (void *)&(sec)->data; \ 399 i < (sec)->num_info; \ 400 i++, rec = (void *)rec + (seg)->rec_size) 401 402 /* 403 * The .BTF.ext ELF section layout defined as 404 * struct btf_ext_header 405 * func_info subsection 406 * 407 * The func_info subsection layout: 408 * record size for struct bpf_func_info in the func_info subsection 409 * struct btf_sec_func_info for section #1 410 * a list of bpf_func_info records for section #1 411 * where struct bpf_func_info mimics one in include/uapi/linux/bpf.h 412 * but may not be identical 413 * struct btf_sec_func_info for section #2 414 * a list of bpf_func_info records for section #2 415 * ...... 416 * 417 * Note that the bpf_func_info record size in .BTF.ext may not 418 * be the same as the one defined in include/uapi/linux/bpf.h. 419 * The loader should ensure that record_size meets minimum 420 * requirement and pass the record as is to the kernel. The 421 * kernel will handle the func_info properly based on its contents. 422 */ 423 struct btf_ext_header { 424 __u16 magic; 425 __u8 version; 426 __u8 flags; 427 __u32 hdr_len; 428 429 /* All offsets are in bytes relative to the end of this header */ 430 __u32 func_info_off; 431 __u32 func_info_len; 432 __u32 line_info_off; 433 __u32 line_info_len; 434 435 /* optional part of .BTF.ext header */ 436 __u32 core_relo_off; 437 __u32 core_relo_len; 438 }; 439 440 struct btf_ext { 441 union { 442 struct btf_ext_header *hdr; 443 void *data; 444 }; 445 struct btf_ext_info func_info; 446 struct btf_ext_info line_info; 447 struct btf_ext_info core_relo_info; 448 __u32 data_size; 449 }; 450 451 struct btf_ext_info_sec { 452 __u32 sec_name_off; 453 __u32 num_info; 454 /* Followed by num_info * record_size number of bytes */ 455 __u8 data[]; 456 }; 457 458 /* The minimum bpf_func_info checked by the loader */ 459 struct bpf_func_info_min { 460 __u32 insn_off; 461 __u32 type_id; 462 }; 463 464 /* The minimum bpf_line_info checked by the loader */ 465 struct bpf_line_info_min { 466 __u32 insn_off; 467 __u32 file_name_off; 468 __u32 line_off; 469 __u32 line_col; 470 }; 471 472 473 typedef int (*type_id_visit_fn)(__u32 *type_id, void *ctx); 474 typedef int (*str_off_visit_fn)(__u32 *str_off, void *ctx); 475 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx); 476 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx); 477 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx); 478 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx); 479 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name, 480 __u32 kind); 481 482 typedef int (*kallsyms_cb_t)(unsigned long long sym_addr, char sym_type, 483 const char *sym_name, void *ctx); 484 485 int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *arg); 486 487 /* handle direct returned errors */ 488 static inline int libbpf_err(int ret) 489 { 490 if (ret < 0) 491 errno = -ret; 492 return ret; 493 } 494 495 /* handle errno-based (e.g., syscall or libc) errors according to libbpf's 496 * strict mode settings 497 */ 498 static inline int libbpf_err_errno(int ret) 499 { 500 /* errno is already assumed to be set on error */ 501 return ret < 0 ? -errno : ret; 502 } 503 504 /* handle error for pointer-returning APIs, err is assumed to be < 0 always */ 505 static inline void *libbpf_err_ptr(int err) 506 { 507 /* set errno on error, this doesn't break anything */ 508 errno = -err; 509 return NULL; 510 } 511 512 /* handle pointer-returning APIs' error handling */ 513 static inline void *libbpf_ptr(void *ret) 514 { 515 /* set errno on error, this doesn't break anything */ 516 if (IS_ERR(ret)) 517 errno = -PTR_ERR(ret); 518 519 return IS_ERR(ret) ? NULL : ret; 520 } 521 522 static inline bool str_is_empty(const char *s) 523 { 524 return !s || !s[0]; 525 } 526 527 static inline bool is_ldimm64_insn(struct bpf_insn *insn) 528 { 529 return insn->code == (BPF_LD | BPF_IMM | BPF_DW); 530 } 531 532 /* if fd is stdin, stdout, or stderr, dup to a fd greater than 2 533 * Takes ownership of the fd passed in, and closes it if calling 534 * fcntl(fd, F_DUPFD_CLOEXEC, 3). 535 */ 536 static inline int ensure_good_fd(int fd) 537 { 538 int old_fd = fd, saved_errno; 539 540 if (fd < 0) 541 return fd; 542 if (fd < 3) { 543 fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); 544 saved_errno = errno; 545 close(old_fd); 546 errno = saved_errno; 547 if (fd < 0) { 548 pr_warn("failed to dup FD %d to FD > 2: %d\n", old_fd, -saved_errno); 549 errno = saved_errno; 550 } 551 } 552 return fd; 553 } 554 555 /* The following two functions are exposed to bpftool */ 556 int bpf_core_add_cands(struct bpf_core_cand *local_cand, 557 size_t local_essent_len, 558 const struct btf *targ_btf, 559 const char *targ_btf_name, 560 int targ_start_id, 561 struct bpf_core_cand_list *cands); 562 void bpf_core_free_cands(struct bpf_core_cand_list *cands); 563 564 struct usdt_manager *usdt_manager_new(struct bpf_object *obj); 565 void usdt_manager_free(struct usdt_manager *man); 566 struct bpf_link * usdt_manager_attach_usdt(struct usdt_manager *man, 567 const struct bpf_program *prog, 568 pid_t pid, const char *path, 569 const char *usdt_provider, const char *usdt_name, 570 __u64 usdt_cookie); 571 572 static inline bool is_pow_of_2(size_t x) 573 { 574 return x && (x & (x - 1)) == 0; 575 } 576 577 #define PROG_LOAD_ATTEMPTS 5 578 int sys_bpf_prog_load(union bpf_attr *attr, unsigned int size, int attempts); 579 580 #endif /* __LIBBPF_LIBBPF_INTERNAL_H */ 581