root/tools/lib/bpf/libbpf.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. __base_pr
  2. libbpf_set_print
  3. __printf
  4. ptr_to_u64
  5. bpf_program__unload
  6. bpf_program__exit
  7. __bpf_program__pin_name
  8. bpf_program__init
  9. bpf_object__add_program
  10. bpf_object__init_prog_names
  11. bpf_object__new
  12. bpf_object__elf_finish
  13. bpf_object__elf_init
  14. bpf_object__check_endianness
  15. bpf_object__init_license
  16. bpf_object__init_kversion
  17. compare_bpf_map
  18. bpf_map_type__is_map_in_map
  19. bpf_object_search_section_size
  20. bpf_object__section_size
  21. bpf_object__variable_offset
  22. bpf_object__add_map
  23. bpf_object__init_internal_map
  24. bpf_object__init_global_data_maps
  25. bpf_object__init_user_maps
  26. skip_mods_and_typedefs
  27. get_map_field_int
  28. bpf_object__init_user_btf_map
  29. bpf_object__init_user_btf_maps
  30. bpf_object__init_maps
  31. section_have_execinstr
  32. bpf_object__sanitize_btf
  33. bpf_object__sanitize_btf_ext
  34. bpf_object__is_btf_mandatory
  35. bpf_object__init_btf
  36. bpf_object__sanitize_and_load_btf
  37. bpf_object__elf_collect
  38. bpf_object__find_prog_by_idx
  39. bpf_object__find_program_by_title
  40. bpf_object__shndx_is_data
  41. bpf_object__shndx_is_maps
  42. bpf_object__relo_in_known_section
  43. bpf_object__section_to_libbpf_map_type
  44. bpf_program__collect_reloc
  45. bpf_map_find_btf_info
  46. bpf_map__reuse_fd
  47. bpf_map__resize
  48. bpf_object__probe_name
  49. bpf_object__probe_global_data
  50. bpf_object__probe_btf_func
  51. bpf_object__probe_btf_datasec
  52. bpf_object__probe_caps
  53. bpf_object__populate_internal_map
  54. bpf_object__create_maps
  55. check_btf_ext_reloc_err
  56. bpf_program_reloc_btf_ext
  57. str_is_empty
  58. bpf_core_spec_parse
  59. bpf_core_is_flavor_sep
  60. bpf_core_essential_name_len
  61. bpf_core_free_cands
  62. bpf_core_find_cands
  63. bpf_core_fields_are_compat
  64. bpf_core_match_member
  65. bpf_core_spec_match
  66. bpf_core_reloc_insn
  67. btf_load_raw
  68. bpf_core_find_kernel_btf
  69. bpf_core_dump_spec
  70. bpf_core_hash_fn
  71. bpf_core_equal_fn
  72. u32_as_hash_key
  73. bpf_core_reloc_offset
  74. bpf_core_reloc_offsets
  75. bpf_object__relocate_core
  76. bpf_program__reloc_text
  77. bpf_program__relocate
  78. bpf_object__relocate
  79. bpf_object__collect_reloc
  80. load_program
  81. bpf_program__load
  82. bpf_program__is_function_storage
  83. bpf_object__load_progs
  84. bpf_prog_type__needs_kver
  85. bpf_object__validate
  86. __bpf_object__open
  87. __bpf_object__open_xattr
  88. bpf_object__open_xattr
  89. bpf_object__open
  90. bpf_object__open_buffer
  91. bpf_object__unload
  92. bpf_object__load_xattr
  93. bpf_object__load
  94. check_path
  95. bpf_program__pin_instance
  96. bpf_program__unpin_instance
  97. make_dir
  98. bpf_program__pin
  99. bpf_program__unpin
  100. bpf_map__pin
  101. bpf_map__unpin
  102. bpf_object__pin_maps
  103. bpf_object__unpin_maps
  104. bpf_object__pin_programs
  105. bpf_object__unpin_programs
  106. bpf_object__pin
  107. bpf_object__close
  108. bpf_object__next
  109. bpf_object__name
  110. bpf_object__kversion
  111. bpf_object__btf
  112. bpf_object__btf_fd
  113. bpf_object__set_priv
  114. bpf_object__priv
  115. __bpf_program__iter
  116. bpf_program__next
  117. bpf_program__prev
  118. bpf_program__set_priv
  119. bpf_program__priv
  120. bpf_program__set_ifindex
  121. bpf_program__title
  122. bpf_program__fd
  123. bpf_program__set_prep
  124. bpf_program__nth_fd
  125. bpf_program__set_type
  126. bpf_program__is_type
  127. bpf_program__set_expected_attach_type
  128. libbpf_get_type_names
  129. libbpf_prog_type_by_name
  130. libbpf_attach_type_by_name
  131. bpf_program__identify_section
  132. bpf_map__fd
  133. bpf_map__def
  134. bpf_map__name
  135. bpf_map__btf_key_type_id
  136. bpf_map__btf_value_type_id
  137. bpf_map__set_priv
  138. bpf_map__priv
  139. bpf_map__is_offload_neutral
  140. bpf_map__is_internal
  141. bpf_map__set_ifindex
  142. bpf_map__set_inner_map_fd
  143. __bpf_map__iter
  144. bpf_map__next
  145. bpf_map__prev
  146. bpf_object__find_map_by_name
  147. bpf_object__find_map_fd_by_name
  148. bpf_object__find_map_by_offset
  149. libbpf_get_error
  150. bpf_prog_load
  151. bpf_prog_load_xattr
  152. bpf_link__destroy
  153. bpf_link__destroy_perf_event
  154. bpf_program__attach_perf_event
  155. parse_uint_from_file
  156. determine_kprobe_perf_type
  157. determine_uprobe_perf_type
  158. determine_kprobe_retprobe_bit
  159. determine_uprobe_retprobe_bit
  160. perf_event_open_probe
  161. bpf_program__attach_kprobe
  162. bpf_program__attach_uprobe
  163. determine_tracepoint_id
  164. perf_event_open_tracepoint
  165. bpf_program__attach_tracepoint
  166. bpf_link__destroy_fd
  167. bpf_program__attach_raw_tracepoint
  168. bpf_perf_event_read_simple
  169. perf_buffer__free_cpu_buf
  170. perf_buffer__free
  171. perf_buffer__open_cpu_buf
  172. perf_buffer__new
  173. perf_buffer__new_raw
  174. __perf_buffer__new
  175. perf_buffer__process_record
  176. perf_buffer__process_records
  177. perf_buffer__poll
  178. bpf_prog_info_read_offset_u32
  179. bpf_prog_info_read_offset_u64
  180. bpf_prog_info_set_offset_u32
  181. bpf_prog_info_set_offset_u64
  182. bpf_program__get_prog_info_linear
  183. bpf_program__bpil_addr_to_offs
  184. bpf_program__bpil_offs_to_addr
  185. parse_cpu_mask_str
  186. parse_cpu_mask_file
  187. libbpf_num_possible_cpus

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2 
   3 /*
   4  * Common eBPF ELF object loading operations.
   5  *
   6  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
   7  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
   8  * Copyright (C) 2015 Huawei Inc.
   9  * Copyright (C) 2017 Nicira, Inc.
  10  * Copyright (C) 2019 Isovalent, Inc.
  11  */
  12 
  13 #ifndef _GNU_SOURCE
  14 #define _GNU_SOURCE
  15 #endif
  16 #include <stdlib.h>
  17 #include <stdio.h>
  18 #include <stdarg.h>
  19 #include <libgen.h>
  20 #include <inttypes.h>
  21 #include <string.h>
  22 #include <unistd.h>
  23 #include <endian.h>
  24 #include <fcntl.h>
  25 #include <errno.h>
  26 #include <asm/unistd.h>
  27 #include <linux/err.h>
  28 #include <linux/kernel.h>
  29 #include <linux/bpf.h>
  30 #include <linux/btf.h>
  31 #include <linux/filter.h>
  32 #include <linux/list.h>
  33 #include <linux/limits.h>
  34 #include <linux/perf_event.h>
  35 #include <linux/ring_buffer.h>
  36 #include <sys/epoll.h>
  37 #include <sys/ioctl.h>
  38 #include <sys/mman.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/vfs.h>
  42 #include <sys/utsname.h>
  43 #include <tools/libc_compat.h>
  44 #include <libelf.h>
  45 #include <gelf.h>
  46 
  47 #include "libbpf.h"
  48 #include "bpf.h"
  49 #include "btf.h"
  50 #include "str_error.h"
  51 #include "libbpf_internal.h"
  52 #include "hashmap.h"
  53 
  54 #ifndef EM_BPF
  55 #define EM_BPF 247
  56 #endif
  57 
  58 #ifndef BPF_FS_MAGIC
  59 #define BPF_FS_MAGIC            0xcafe4a11
  60 #endif
  61 
  62 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  63  * compilation if user enables corresponding warning. Disable it explicitly.
  64  */
  65 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  66 
  67 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  68 
  69 static int __base_pr(enum libbpf_print_level level, const char *format,
  70                      va_list args)
  71 {
  72         if (level == LIBBPF_DEBUG)
  73                 return 0;
  74 
  75         return vfprintf(stderr, format, args);
  76 }
  77 
  78 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  79 
  80 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  81 {
  82         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  83 
  84         __libbpf_pr = fn;
  85         return old_print_fn;
  86 }
  87 
  88 __printf(2, 3)
  89 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  90 {
  91         va_list args;
  92 
  93         if (!__libbpf_pr)
  94                 return;
  95 
  96         va_start(args, format);
  97         __libbpf_pr(level, format, args);
  98         va_end(args);
  99 }
 100 
 101 #define STRERR_BUFSIZE  128
 102 
 103 #define CHECK_ERR(action, err, out) do {        \
 104         err = action;                   \
 105         if (err)                        \
 106                 goto out;               \
 107 } while(0)
 108 
 109 
 110 /* Copied from tools/perf/util/util.h */
 111 #ifndef zfree
 112 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 113 #endif
 114 
 115 #ifndef zclose
 116 # define zclose(fd) ({                  \
 117         int ___err = 0;                 \
 118         if ((fd) >= 0)                  \
 119                 ___err = close((fd));   \
 120         fd = -1;                        \
 121         ___err; })
 122 #endif
 123 
 124 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 125 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 126 #else
 127 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 128 #endif
 129 
 130 static inline __u64 ptr_to_u64(const void *ptr)
 131 {
 132         return (__u64) (unsigned long) ptr;
 133 }
 134 
 135 struct bpf_capabilities {
 136         /* v4.14: kernel support for program & map names. */
 137         __u32 name:1;
 138         /* v5.2: kernel support for global data sections. */
 139         __u32 global_data:1;
 140         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 141         __u32 btf_func:1;
 142         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 143         __u32 btf_datasec:1;
 144 };
 145 
 146 /*
 147  * bpf_prog should be a better name but it has been used in
 148  * linux/filter.h.
 149  */
 150 struct bpf_program {
 151         /* Index in elf obj file, for relocation use. */
 152         int idx;
 153         char *name;
 154         int prog_ifindex;
 155         char *section_name;
 156         /* section_name with / replaced by _; makes recursive pinning
 157          * in bpf_object__pin_programs easier
 158          */
 159         char *pin_name;
 160         struct bpf_insn *insns;
 161         size_t insns_cnt, main_prog_cnt;
 162         enum bpf_prog_type type;
 163 
 164         struct reloc_desc {
 165                 enum {
 166                         RELO_LD64,
 167                         RELO_CALL,
 168                         RELO_DATA,
 169                 } type;
 170                 int insn_idx;
 171                 union {
 172                         int map_idx;
 173                         int text_off;
 174                 };
 175         } *reloc_desc;
 176         int nr_reloc;
 177         int log_level;
 178 
 179         struct {
 180                 int nr;
 181                 int *fds;
 182         } instances;
 183         bpf_program_prep_t preprocessor;
 184 
 185         struct bpf_object *obj;
 186         void *priv;
 187         bpf_program_clear_priv_t clear_priv;
 188 
 189         enum bpf_attach_type expected_attach_type;
 190         void *func_info;
 191         __u32 func_info_rec_size;
 192         __u32 func_info_cnt;
 193 
 194         struct bpf_capabilities *caps;
 195 
 196         void *line_info;
 197         __u32 line_info_rec_size;
 198         __u32 line_info_cnt;
 199         __u32 prog_flags;
 200 };
 201 
 202 enum libbpf_map_type {
 203         LIBBPF_MAP_UNSPEC,
 204         LIBBPF_MAP_DATA,
 205         LIBBPF_MAP_BSS,
 206         LIBBPF_MAP_RODATA,
 207 };
 208 
 209 static const char * const libbpf_type_to_btf_name[] = {
 210         [LIBBPF_MAP_DATA]       = ".data",
 211         [LIBBPF_MAP_BSS]        = ".bss",
 212         [LIBBPF_MAP_RODATA]     = ".rodata",
 213 };
 214 
 215 struct bpf_map {
 216         int fd;
 217         char *name;
 218         int sec_idx;
 219         size_t sec_offset;
 220         int map_ifindex;
 221         int inner_map_fd;
 222         struct bpf_map_def def;
 223         __u32 btf_key_type_id;
 224         __u32 btf_value_type_id;
 225         void *priv;
 226         bpf_map_clear_priv_t clear_priv;
 227         enum libbpf_map_type libbpf_type;
 228 };
 229 
 230 struct bpf_secdata {
 231         void *rodata;
 232         void *data;
 233 };
 234 
 235 static LIST_HEAD(bpf_objects_list);
 236 
 237 struct bpf_object {
 238         char name[BPF_OBJ_NAME_LEN];
 239         char license[64];
 240         __u32 kern_version;
 241 
 242         struct bpf_program *programs;
 243         size_t nr_programs;
 244         struct bpf_map *maps;
 245         size_t nr_maps;
 246         size_t maps_cap;
 247         struct bpf_secdata sections;
 248 
 249         bool loaded;
 250         bool has_pseudo_calls;
 251 
 252         /*
 253          * Information when doing elf related work. Only valid if fd
 254          * is valid.
 255          */
 256         struct {
 257                 int fd;
 258                 void *obj_buf;
 259                 size_t obj_buf_sz;
 260                 Elf *elf;
 261                 GElf_Ehdr ehdr;
 262                 Elf_Data *symbols;
 263                 Elf_Data *data;
 264                 Elf_Data *rodata;
 265                 Elf_Data *bss;
 266                 size_t strtabidx;
 267                 struct {
 268                         GElf_Shdr shdr;
 269                         Elf_Data *data;
 270                 } *reloc;
 271                 int nr_reloc;
 272                 int maps_shndx;
 273                 int btf_maps_shndx;
 274                 int text_shndx;
 275                 int data_shndx;
 276                 int rodata_shndx;
 277                 int bss_shndx;
 278         } efile;
 279         /*
 280          * All loaded bpf_object is linked in a list, which is
 281          * hidden to caller. bpf_objects__<func> handlers deal with
 282          * all objects.
 283          */
 284         struct list_head list;
 285 
 286         struct btf *btf;
 287         struct btf_ext *btf_ext;
 288 
 289         void *priv;
 290         bpf_object_clear_priv_t clear_priv;
 291 
 292         struct bpf_capabilities caps;
 293 
 294         char path[];
 295 };
 296 #define obj_elf_valid(o)        ((o)->efile.elf)
 297 
 298 void bpf_program__unload(struct bpf_program *prog)
 299 {
 300         int i;
 301 
 302         if (!prog)
 303                 return;
 304 
 305         /*
 306          * If the object is opened but the program was never loaded,
 307          * it is possible that prog->instances.nr == -1.
 308          */
 309         if (prog->instances.nr > 0) {
 310                 for (i = 0; i < prog->instances.nr; i++)
 311                         zclose(prog->instances.fds[i]);
 312         } else if (prog->instances.nr != -1) {
 313                 pr_warning("Internal error: instances.nr is %d\n",
 314                            prog->instances.nr);
 315         }
 316 
 317         prog->instances.nr = -1;
 318         zfree(&prog->instances.fds);
 319 
 320         zfree(&prog->func_info);
 321         zfree(&prog->line_info);
 322 }
 323 
 324 static void bpf_program__exit(struct bpf_program *prog)
 325 {
 326         if (!prog)
 327                 return;
 328 
 329         if (prog->clear_priv)
 330                 prog->clear_priv(prog, prog->priv);
 331 
 332         prog->priv = NULL;
 333         prog->clear_priv = NULL;
 334 
 335         bpf_program__unload(prog);
 336         zfree(&prog->name);
 337         zfree(&prog->section_name);
 338         zfree(&prog->pin_name);
 339         zfree(&prog->insns);
 340         zfree(&prog->reloc_desc);
 341 
 342         prog->nr_reloc = 0;
 343         prog->insns_cnt = 0;
 344         prog->idx = -1;
 345 }
 346 
 347 static char *__bpf_program__pin_name(struct bpf_program *prog)
 348 {
 349         char *name, *p;
 350 
 351         name = p = strdup(prog->section_name);
 352         while ((p = strchr(p, '/')))
 353                 *p = '_';
 354 
 355         return name;
 356 }
 357 
 358 static int
 359 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 360                   struct bpf_program *prog)
 361 {
 362         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 363 
 364         if (size == 0 || size % bpf_insn_sz) {
 365                 pr_warning("corrupted section '%s', size: %zu\n",
 366                            section_name, size);
 367                 return -EINVAL;
 368         }
 369 
 370         memset(prog, 0, sizeof(*prog));
 371 
 372         prog->section_name = strdup(section_name);
 373         if (!prog->section_name) {
 374                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 375                            idx, section_name);
 376                 goto errout;
 377         }
 378 
 379         prog->pin_name = __bpf_program__pin_name(prog);
 380         if (!prog->pin_name) {
 381                 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
 382                            idx, section_name);
 383                 goto errout;
 384         }
 385 
 386         prog->insns = malloc(size);
 387         if (!prog->insns) {
 388                 pr_warning("failed to alloc insns for prog under section %s\n",
 389                            section_name);
 390                 goto errout;
 391         }
 392         prog->insns_cnt = size / bpf_insn_sz;
 393         memcpy(prog->insns, data, size);
 394         prog->idx = idx;
 395         prog->instances.fds = NULL;
 396         prog->instances.nr = -1;
 397         prog->type = BPF_PROG_TYPE_UNSPEC;
 398 
 399         return 0;
 400 errout:
 401         bpf_program__exit(prog);
 402         return -ENOMEM;
 403 }
 404 
 405 static int
 406 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 407                         char *section_name, int idx)
 408 {
 409         struct bpf_program prog, *progs;
 410         int nr_progs, err;
 411 
 412         err = bpf_program__init(data, size, section_name, idx, &prog);
 413         if (err)
 414                 return err;
 415 
 416         prog.caps = &obj->caps;
 417         progs = obj->programs;
 418         nr_progs = obj->nr_programs;
 419 
 420         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 421         if (!progs) {
 422                 /*
 423                  * In this case the original obj->programs
 424                  * is still valid, so don't need special treat for
 425                  * bpf_close_object().
 426                  */
 427                 pr_warning("failed to alloc a new program under section '%s'\n",
 428                            section_name);
 429                 bpf_program__exit(&prog);
 430                 return -ENOMEM;
 431         }
 432 
 433         pr_debug("found program %s\n", prog.section_name);
 434         obj->programs = progs;
 435         obj->nr_programs = nr_progs + 1;
 436         prog.obj = obj;
 437         progs[nr_progs] = prog;
 438         return 0;
 439 }
 440 
 441 static int
 442 bpf_object__init_prog_names(struct bpf_object *obj)
 443 {
 444         Elf_Data *symbols = obj->efile.symbols;
 445         struct bpf_program *prog;
 446         size_t pi, si;
 447 
 448         for (pi = 0; pi < obj->nr_programs; pi++) {
 449                 const char *name = NULL;
 450 
 451                 prog = &obj->programs[pi];
 452 
 453                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 454                      si++) {
 455                         GElf_Sym sym;
 456 
 457                         if (!gelf_getsym(symbols, si, &sym))
 458                                 continue;
 459                         if (sym.st_shndx != prog->idx)
 460                                 continue;
 461                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 462                                 continue;
 463 
 464                         name = elf_strptr(obj->efile.elf,
 465                                           obj->efile.strtabidx,
 466                                           sym.st_name);
 467                         if (!name) {
 468                                 pr_warning("failed to get sym name string for prog %s\n",
 469                                            prog->section_name);
 470                                 return -LIBBPF_ERRNO__LIBELF;
 471                         }
 472                 }
 473 
 474                 if (!name && prog->idx == obj->efile.text_shndx)
 475                         name = ".text";
 476 
 477                 if (!name) {
 478                         pr_warning("failed to find sym for prog %s\n",
 479                                    prog->section_name);
 480                         return -EINVAL;
 481                 }
 482 
 483                 prog->name = strdup(name);
 484                 if (!prog->name) {
 485                         pr_warning("failed to allocate memory for prog sym %s\n",
 486                                    name);
 487                         return -ENOMEM;
 488                 }
 489         }
 490 
 491         return 0;
 492 }
 493 
 494 static struct bpf_object *bpf_object__new(const char *path,
 495                                           void *obj_buf,
 496                                           size_t obj_buf_sz)
 497 {
 498         struct bpf_object *obj;
 499         char *end;
 500 
 501         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 502         if (!obj) {
 503                 pr_warning("alloc memory failed for %s\n", path);
 504                 return ERR_PTR(-ENOMEM);
 505         }
 506 
 507         strcpy(obj->path, path);
 508         /* Using basename() GNU version which doesn't modify arg. */
 509         strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
 510         end = strchr(obj->name, '.');
 511         if (end)
 512                 *end = 0;
 513 
 514         obj->efile.fd = -1;
 515         /*
 516          * Caller of this function should also call
 517          * bpf_object__elf_finish() after data collection to return
 518          * obj_buf to user. If not, we should duplicate the buffer to
 519          * avoid user freeing them before elf finish.
 520          */
 521         obj->efile.obj_buf = obj_buf;
 522         obj->efile.obj_buf_sz = obj_buf_sz;
 523         obj->efile.maps_shndx = -1;
 524         obj->efile.btf_maps_shndx = -1;
 525         obj->efile.data_shndx = -1;
 526         obj->efile.rodata_shndx = -1;
 527         obj->efile.bss_shndx = -1;
 528 
 529         obj->loaded = false;
 530 
 531         INIT_LIST_HEAD(&obj->list);
 532         list_add(&obj->list, &bpf_objects_list);
 533         return obj;
 534 }
 535 
 536 static void bpf_object__elf_finish(struct bpf_object *obj)
 537 {
 538         if (!obj_elf_valid(obj))
 539                 return;
 540 
 541         if (obj->efile.elf) {
 542                 elf_end(obj->efile.elf);
 543                 obj->efile.elf = NULL;
 544         }
 545         obj->efile.symbols = NULL;
 546         obj->efile.data = NULL;
 547         obj->efile.rodata = NULL;
 548         obj->efile.bss = NULL;
 549 
 550         zfree(&obj->efile.reloc);
 551         obj->efile.nr_reloc = 0;
 552         zclose(obj->efile.fd);
 553         obj->efile.obj_buf = NULL;
 554         obj->efile.obj_buf_sz = 0;
 555 }
 556 
 557 static int bpf_object__elf_init(struct bpf_object *obj)
 558 {
 559         int err = 0;
 560         GElf_Ehdr *ep;
 561 
 562         if (obj_elf_valid(obj)) {
 563                 pr_warning("elf init: internal error\n");
 564                 return -LIBBPF_ERRNO__LIBELF;
 565         }
 566 
 567         if (obj->efile.obj_buf_sz > 0) {
 568                 /*
 569                  * obj_buf should have been validated by
 570                  * bpf_object__open_buffer().
 571                  */
 572                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 573                                             obj->efile.obj_buf_sz);
 574         } else {
 575                 obj->efile.fd = open(obj->path, O_RDONLY);
 576                 if (obj->efile.fd < 0) {
 577                         char errmsg[STRERR_BUFSIZE], *cp;
 578 
 579                         err = -errno;
 580                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 581                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 582                         return err;
 583                 }
 584 
 585                 obj->efile.elf = elf_begin(obj->efile.fd,
 586                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 587         }
 588 
 589         if (!obj->efile.elf) {
 590                 pr_warning("failed to open %s as ELF file\n", obj->path);
 591                 err = -LIBBPF_ERRNO__LIBELF;
 592                 goto errout;
 593         }
 594 
 595         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 596                 pr_warning("failed to get EHDR from %s\n", obj->path);
 597                 err = -LIBBPF_ERRNO__FORMAT;
 598                 goto errout;
 599         }
 600         ep = &obj->efile.ehdr;
 601 
 602         /* Old LLVM set e_machine to EM_NONE */
 603         if (ep->e_type != ET_REL ||
 604             (ep->e_machine && ep->e_machine != EM_BPF)) {
 605                 pr_warning("%s is not an eBPF object file\n", obj->path);
 606                 err = -LIBBPF_ERRNO__FORMAT;
 607                 goto errout;
 608         }
 609 
 610         return 0;
 611 errout:
 612         bpf_object__elf_finish(obj);
 613         return err;
 614 }
 615 
 616 static int bpf_object__check_endianness(struct bpf_object *obj)
 617 {
 618 #if __BYTE_ORDER == __LITTLE_ENDIAN
 619         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 620                 return 0;
 621 #elif __BYTE_ORDER == __BIG_ENDIAN
 622         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 623                 return 0;
 624 #else
 625 # error "Unrecognized __BYTE_ORDER__"
 626 #endif
 627         pr_warning("endianness mismatch.\n");
 628         return -LIBBPF_ERRNO__ENDIAN;
 629 }
 630 
 631 static int
 632 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 633 {
 634         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 635         pr_debug("license of %s is %s\n", obj->path, obj->license);
 636         return 0;
 637 }
 638 
 639 static int
 640 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 641 {
 642         __u32 kver;
 643 
 644         if (size != sizeof(kver)) {
 645                 pr_warning("invalid kver section in %s\n", obj->path);
 646                 return -LIBBPF_ERRNO__FORMAT;
 647         }
 648         memcpy(&kver, data, sizeof(kver));
 649         obj->kern_version = kver;
 650         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 651         return 0;
 652 }
 653 
 654 static int compare_bpf_map(const void *_a, const void *_b)
 655 {
 656         const struct bpf_map *a = _a;
 657         const struct bpf_map *b = _b;
 658 
 659         if (a->sec_idx != b->sec_idx)
 660                 return a->sec_idx - b->sec_idx;
 661         return a->sec_offset - b->sec_offset;
 662 }
 663 
 664 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 665 {
 666         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 667             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 668                 return true;
 669         return false;
 670 }
 671 
 672 static int bpf_object_search_section_size(const struct bpf_object *obj,
 673                                           const char *name, size_t *d_size)
 674 {
 675         const GElf_Ehdr *ep = &obj->efile.ehdr;
 676         Elf *elf = obj->efile.elf;
 677         Elf_Scn *scn = NULL;
 678         int idx = 0;
 679 
 680         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 681                 const char *sec_name;
 682                 Elf_Data *data;
 683                 GElf_Shdr sh;
 684 
 685                 idx++;
 686                 if (gelf_getshdr(scn, &sh) != &sh) {
 687                         pr_warning("failed to get section(%d) header from %s\n",
 688                                    idx, obj->path);
 689                         return -EIO;
 690                 }
 691 
 692                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 693                 if (!sec_name) {
 694                         pr_warning("failed to get section(%d) name from %s\n",
 695                                    idx, obj->path);
 696                         return -EIO;
 697                 }
 698 
 699                 if (strcmp(name, sec_name))
 700                         continue;
 701 
 702                 data = elf_getdata(scn, 0);
 703                 if (!data) {
 704                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 705                                    idx, name, obj->path);
 706                         return -EIO;
 707                 }
 708 
 709                 *d_size = data->d_size;
 710                 return 0;
 711         }
 712 
 713         return -ENOENT;
 714 }
 715 
 716 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 717                              __u32 *size)
 718 {
 719         int ret = -ENOENT;
 720         size_t d_size;
 721 
 722         *size = 0;
 723         if (!name) {
 724                 return -EINVAL;
 725         } else if (!strcmp(name, ".data")) {
 726                 if (obj->efile.data)
 727                         *size = obj->efile.data->d_size;
 728         } else if (!strcmp(name, ".bss")) {
 729                 if (obj->efile.bss)
 730                         *size = obj->efile.bss->d_size;
 731         } else if (!strcmp(name, ".rodata")) {
 732                 if (obj->efile.rodata)
 733                         *size = obj->efile.rodata->d_size;
 734         } else {
 735                 ret = bpf_object_search_section_size(obj, name, &d_size);
 736                 if (!ret)
 737                         *size = d_size;
 738         }
 739 
 740         return *size ? 0 : ret;
 741 }
 742 
 743 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 744                                 __u32 *off)
 745 {
 746         Elf_Data *symbols = obj->efile.symbols;
 747         const char *sname;
 748         size_t si;
 749 
 750         if (!name || !off)
 751                 return -EINVAL;
 752 
 753         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 754                 GElf_Sym sym;
 755 
 756                 if (!gelf_getsym(symbols, si, &sym))
 757                         continue;
 758                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 759                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 760                         continue;
 761 
 762                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 763                                    sym.st_name);
 764                 if (!sname) {
 765                         pr_warning("failed to get sym name string for var %s\n",
 766                                    name);
 767                         return -EIO;
 768                 }
 769                 if (strcmp(name, sname) == 0) {
 770                         *off = sym.st_value;
 771                         return 0;
 772                 }
 773         }
 774 
 775         return -ENOENT;
 776 }
 777 
 778 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 779 {
 780         struct bpf_map *new_maps;
 781         size_t new_cap;
 782         int i;
 783 
 784         if (obj->nr_maps < obj->maps_cap)
 785                 return &obj->maps[obj->nr_maps++];
 786 
 787         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 788         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 789         if (!new_maps) {
 790                 pr_warning("alloc maps for object failed\n");
 791                 return ERR_PTR(-ENOMEM);
 792         }
 793 
 794         obj->maps_cap = new_cap;
 795         obj->maps = new_maps;
 796 
 797         /* zero out new maps */
 798         memset(obj->maps + obj->nr_maps, 0,
 799                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 800         /*
 801          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 802          * when failure (zclose won't close negative fd)).
 803          */
 804         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 805                 obj->maps[i].fd = -1;
 806                 obj->maps[i].inner_map_fd = -1;
 807         }
 808 
 809         return &obj->maps[obj->nr_maps++];
 810 }
 811 
 812 static int
 813 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 814                               int sec_idx, Elf_Data *data, void **data_buff)
 815 {
 816         char map_name[BPF_OBJ_NAME_LEN];
 817         struct bpf_map_def *def;
 818         struct bpf_map *map;
 819 
 820         map = bpf_object__add_map(obj);
 821         if (IS_ERR(map))
 822                 return PTR_ERR(map);
 823 
 824         map->libbpf_type = type;
 825         map->sec_idx = sec_idx;
 826         map->sec_offset = 0;
 827         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 828                  libbpf_type_to_btf_name[type]);
 829         map->name = strdup(map_name);
 830         if (!map->name) {
 831                 pr_warning("failed to alloc map name\n");
 832                 return -ENOMEM;
 833         }
 834         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
 835                  map_name, map->sec_idx, map->sec_offset);
 836 
 837         def = &map->def;
 838         def->type = BPF_MAP_TYPE_ARRAY;
 839         def->key_size = sizeof(int);
 840         def->value_size = data->d_size;
 841         def->max_entries = 1;
 842         def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
 843         if (data_buff) {
 844                 *data_buff = malloc(data->d_size);
 845                 if (!*data_buff) {
 846                         zfree(&map->name);
 847                         pr_warning("failed to alloc map content buffer\n");
 848                         return -ENOMEM;
 849                 }
 850                 memcpy(*data_buff, data->d_buf, data->d_size);
 851         }
 852 
 853         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 854         return 0;
 855 }
 856 
 857 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 858 {
 859         int err;
 860 
 861         if (!obj->caps.global_data)
 862                 return 0;
 863         /*
 864          * Populate obj->maps with libbpf internal maps.
 865          */
 866         if (obj->efile.data_shndx >= 0) {
 867                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 868                                                     obj->efile.data_shndx,
 869                                                     obj->efile.data,
 870                                                     &obj->sections.data);
 871                 if (err)
 872                         return err;
 873         }
 874         if (obj->efile.rodata_shndx >= 0) {
 875                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 876                                                     obj->efile.rodata_shndx,
 877                                                     obj->efile.rodata,
 878                                                     &obj->sections.rodata);
 879                 if (err)
 880                         return err;
 881         }
 882         if (obj->efile.bss_shndx >= 0) {
 883                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 884                                                     obj->efile.bss_shndx,
 885                                                     obj->efile.bss, NULL);
 886                 if (err)
 887                         return err;
 888         }
 889         return 0;
 890 }
 891 
 892 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
 893 {
 894         Elf_Data *symbols = obj->efile.symbols;
 895         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
 896         Elf_Data *data = NULL;
 897         Elf_Scn *scn;
 898 
 899         if (obj->efile.maps_shndx < 0)
 900                 return 0;
 901 
 902         if (!symbols)
 903                 return -EINVAL;
 904 
 905         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 906         if (scn)
 907                 data = elf_getdata(scn, NULL);
 908         if (!scn || !data) {
 909                 pr_warning("failed to get Elf_Data from map section %d\n",
 910                            obj->efile.maps_shndx);
 911                 return -EINVAL;
 912         }
 913 
 914         /*
 915          * Count number of maps. Each map has a name.
 916          * Array of maps is not supported: only the first element is
 917          * considered.
 918          *
 919          * TODO: Detect array of map and report error.
 920          */
 921         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 922         for (i = 0; i < nr_syms; i++) {
 923                 GElf_Sym sym;
 924 
 925                 if (!gelf_getsym(symbols, i, &sym))
 926                         continue;
 927                 if (sym.st_shndx != obj->efile.maps_shndx)
 928                         continue;
 929                 nr_maps++;
 930         }
 931         /* Assume equally sized map definitions */
 932         pr_debug("maps in %s: %d maps in %zd bytes\n",
 933                  obj->path, nr_maps, data->d_size);
 934 
 935         map_def_sz = data->d_size / nr_maps;
 936         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 937                 pr_warning("unable to determine map definition size "
 938                            "section %s, %d maps in %zd bytes\n",
 939                            obj->path, nr_maps, data->d_size);
 940                 return -EINVAL;
 941         }
 942 
 943         /* Fill obj->maps using data in "maps" section.  */
 944         for (i = 0; i < nr_syms; i++) {
 945                 GElf_Sym sym;
 946                 const char *map_name;
 947                 struct bpf_map_def *def;
 948                 struct bpf_map *map;
 949 
 950                 if (!gelf_getsym(symbols, i, &sym))
 951                         continue;
 952                 if (sym.st_shndx != obj->efile.maps_shndx)
 953                         continue;
 954 
 955                 map = bpf_object__add_map(obj);
 956                 if (IS_ERR(map))
 957                         return PTR_ERR(map);
 958 
 959                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 960                                       sym.st_name);
 961                 if (!map_name) {
 962                         pr_warning("failed to get map #%d name sym string for obj %s\n",
 963                                    i, obj->path);
 964                         return -LIBBPF_ERRNO__FORMAT;
 965                 }
 966 
 967                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
 968                 map->sec_idx = sym.st_shndx;
 969                 map->sec_offset = sym.st_value;
 970                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
 971                          map_name, map->sec_idx, map->sec_offset);
 972                 if (sym.st_value + map_def_sz > data->d_size) {
 973                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 974                                    obj->path, map_name);
 975                         return -EINVAL;
 976                 }
 977 
 978                 map->name = strdup(map_name);
 979                 if (!map->name) {
 980                         pr_warning("failed to alloc map name\n");
 981                         return -ENOMEM;
 982                 }
 983                 pr_debug("map %d is \"%s\"\n", i, map->name);
 984                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 985                 /*
 986                  * If the definition of the map in the object file fits in
 987                  * bpf_map_def, copy it.  Any extra fields in our version
 988                  * of bpf_map_def will default to zero as a result of the
 989                  * calloc above.
 990                  */
 991                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 992                         memcpy(&map->def, def, map_def_sz);
 993                 } else {
 994                         /*
 995                          * Here the map structure being read is bigger than what
 996                          * we expect, truncate if the excess bits are all zero.
 997                          * If they are not zero, reject this map as
 998                          * incompatible.
 999                          */
1000                         char *b;
1001                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1002                              b < ((char *)def) + map_def_sz; b++) {
1003                                 if (*b != 0) {
1004                                         pr_warning("maps section in %s: \"%s\" "
1005                                                    "has unrecognized, non-zero "
1006                                                    "options\n",
1007                                                    obj->path, map_name);
1008                                         if (strict)
1009                                                 return -EINVAL;
1010                                 }
1011                         }
1012                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1013                 }
1014         }
1015         return 0;
1016 }
1017 
1018 static const struct btf_type *
1019 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1020 {
1021         const struct btf_type *t = btf__type_by_id(btf, id);
1022 
1023         if (res_id)
1024                 *res_id = id;
1025 
1026         while (btf_is_mod(t) || btf_is_typedef(t)) {
1027                 if (res_id)
1028                         *res_id = t->type;
1029                 t = btf__type_by_id(btf, t->type);
1030         }
1031 
1032         return t;
1033 }
1034 
1035 /*
1036  * Fetch integer attribute of BTF map definition. Such attributes are
1037  * represented using a pointer to an array, in which dimensionality of array
1038  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1039  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1040  * type definition, while using only sizeof(void *) space in ELF data section.
1041  */
1042 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1043                               const struct btf_type *def,
1044                               const struct btf_member *m, __u32 *res) {
1045         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1046         const char *name = btf__name_by_offset(btf, m->name_off);
1047         const struct btf_array *arr_info;
1048         const struct btf_type *arr_t;
1049 
1050         if (!btf_is_ptr(t)) {
1051                 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1052                            map_name, name, btf_kind(t));
1053                 return false;
1054         }
1055 
1056         arr_t = btf__type_by_id(btf, t->type);
1057         if (!arr_t) {
1058                 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1059                            map_name, name, t->type);
1060                 return false;
1061         }
1062         if (!btf_is_array(arr_t)) {
1063                 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1064                            map_name, name, btf_kind(arr_t));
1065                 return false;
1066         }
1067         arr_info = btf_array(arr_t);
1068         *res = arr_info->nelems;
1069         return true;
1070 }
1071 
1072 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1073                                          const struct btf_type *sec,
1074                                          int var_idx, int sec_idx,
1075                                          const Elf_Data *data, bool strict)
1076 {
1077         const struct btf_type *var, *def, *t;
1078         const struct btf_var_secinfo *vi;
1079         const struct btf_var *var_extra;
1080         const struct btf_member *m;
1081         const char *map_name;
1082         struct bpf_map *map;
1083         int vlen, i;
1084 
1085         vi = btf_var_secinfos(sec) + var_idx;
1086         var = btf__type_by_id(obj->btf, vi->type);
1087         var_extra = btf_var(var);
1088         map_name = btf__name_by_offset(obj->btf, var->name_off);
1089         vlen = btf_vlen(var);
1090 
1091         if (map_name == NULL || map_name[0] == '\0') {
1092                 pr_warning("map #%d: empty name.\n", var_idx);
1093                 return -EINVAL;
1094         }
1095         if ((__u64)vi->offset + vi->size > data->d_size) {
1096                 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1097                 return -EINVAL;
1098         }
1099         if (!btf_is_var(var)) {
1100                 pr_warning("map '%s': unexpected var kind %u.\n",
1101                            map_name, btf_kind(var));
1102                 return -EINVAL;
1103         }
1104         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1105             var_extra->linkage != BTF_VAR_STATIC) {
1106                 pr_warning("map '%s': unsupported var linkage %u.\n",
1107                            map_name, var_extra->linkage);
1108                 return -EOPNOTSUPP;
1109         }
1110 
1111         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1112         if (!btf_is_struct(def)) {
1113                 pr_warning("map '%s': unexpected def kind %u.\n",
1114                            map_name, btf_kind(var));
1115                 return -EINVAL;
1116         }
1117         if (def->size > vi->size) {
1118                 pr_warning("map '%s': invalid def size.\n", map_name);
1119                 return -EINVAL;
1120         }
1121 
1122         map = bpf_object__add_map(obj);
1123         if (IS_ERR(map))
1124                 return PTR_ERR(map);
1125         map->name = strdup(map_name);
1126         if (!map->name) {
1127                 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1128                 return -ENOMEM;
1129         }
1130         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1131         map->def.type = BPF_MAP_TYPE_UNSPEC;
1132         map->sec_idx = sec_idx;
1133         map->sec_offset = vi->offset;
1134         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1135                  map_name, map->sec_idx, map->sec_offset);
1136 
1137         vlen = btf_vlen(def);
1138         m = btf_members(def);
1139         for (i = 0; i < vlen; i++, m++) {
1140                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1141 
1142                 if (!name) {
1143                         pr_warning("map '%s': invalid field #%d.\n",
1144                                    map_name, i);
1145                         return -EINVAL;
1146                 }
1147                 if (strcmp(name, "type") == 0) {
1148                         if (!get_map_field_int(map_name, obj->btf, def, m,
1149                                                &map->def.type))
1150                                 return -EINVAL;
1151                         pr_debug("map '%s': found type = %u.\n",
1152                                  map_name, map->def.type);
1153                 } else if (strcmp(name, "max_entries") == 0) {
1154                         if (!get_map_field_int(map_name, obj->btf, def, m,
1155                                                &map->def.max_entries))
1156                                 return -EINVAL;
1157                         pr_debug("map '%s': found max_entries = %u.\n",
1158                                  map_name, map->def.max_entries);
1159                 } else if (strcmp(name, "map_flags") == 0) {
1160                         if (!get_map_field_int(map_name, obj->btf, def, m,
1161                                                &map->def.map_flags))
1162                                 return -EINVAL;
1163                         pr_debug("map '%s': found map_flags = %u.\n",
1164                                  map_name, map->def.map_flags);
1165                 } else if (strcmp(name, "key_size") == 0) {
1166                         __u32 sz;
1167 
1168                         if (!get_map_field_int(map_name, obj->btf, def, m,
1169                                                &sz))
1170                                 return -EINVAL;
1171                         pr_debug("map '%s': found key_size = %u.\n",
1172                                  map_name, sz);
1173                         if (map->def.key_size && map->def.key_size != sz) {
1174                                 pr_warning("map '%s': conflicting key size %u != %u.\n",
1175                                            map_name, map->def.key_size, sz);
1176                                 return -EINVAL;
1177                         }
1178                         map->def.key_size = sz;
1179                 } else if (strcmp(name, "key") == 0) {
1180                         __s64 sz;
1181 
1182                         t = btf__type_by_id(obj->btf, m->type);
1183                         if (!t) {
1184                                 pr_warning("map '%s': key type [%d] not found.\n",
1185                                            map_name, m->type);
1186                                 return -EINVAL;
1187                         }
1188                         if (!btf_is_ptr(t)) {
1189                                 pr_warning("map '%s': key spec is not PTR: %u.\n",
1190                                            map_name, btf_kind(t));
1191                                 return -EINVAL;
1192                         }
1193                         sz = btf__resolve_size(obj->btf, t->type);
1194                         if (sz < 0) {
1195                                 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1196                                            map_name, t->type, sz);
1197                                 return sz;
1198                         }
1199                         pr_debug("map '%s': found key [%u], sz = %lld.\n",
1200                                  map_name, t->type, sz);
1201                         if (map->def.key_size && map->def.key_size != sz) {
1202                                 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1203                                            map_name, map->def.key_size, sz);
1204                                 return -EINVAL;
1205                         }
1206                         map->def.key_size = sz;
1207                         map->btf_key_type_id = t->type;
1208                 } else if (strcmp(name, "value_size") == 0) {
1209                         __u32 sz;
1210 
1211                         if (!get_map_field_int(map_name, obj->btf, def, m,
1212                                                &sz))
1213                                 return -EINVAL;
1214                         pr_debug("map '%s': found value_size = %u.\n",
1215                                  map_name, sz);
1216                         if (map->def.value_size && map->def.value_size != sz) {
1217                                 pr_warning("map '%s': conflicting value size %u != %u.\n",
1218                                            map_name, map->def.value_size, sz);
1219                                 return -EINVAL;
1220                         }
1221                         map->def.value_size = sz;
1222                 } else if (strcmp(name, "value") == 0) {
1223                         __s64 sz;
1224 
1225                         t = btf__type_by_id(obj->btf, m->type);
1226                         if (!t) {
1227                                 pr_warning("map '%s': value type [%d] not found.\n",
1228                                            map_name, m->type);
1229                                 return -EINVAL;
1230                         }
1231                         if (!btf_is_ptr(t)) {
1232                                 pr_warning("map '%s': value spec is not PTR: %u.\n",
1233                                            map_name, btf_kind(t));
1234                                 return -EINVAL;
1235                         }
1236                         sz = btf__resolve_size(obj->btf, t->type);
1237                         if (sz < 0) {
1238                                 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1239                                            map_name, t->type, sz);
1240                                 return sz;
1241                         }
1242                         pr_debug("map '%s': found value [%u], sz = %lld.\n",
1243                                  map_name, t->type, sz);
1244                         if (map->def.value_size && map->def.value_size != sz) {
1245                                 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1246                                            map_name, map->def.value_size, sz);
1247                                 return -EINVAL;
1248                         }
1249                         map->def.value_size = sz;
1250                         map->btf_value_type_id = t->type;
1251                 } else {
1252                         if (strict) {
1253                                 pr_warning("map '%s': unknown field '%s'.\n",
1254                                            map_name, name);
1255                                 return -ENOTSUP;
1256                         }
1257                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1258                                  map_name, name);
1259                 }
1260         }
1261 
1262         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1263                 pr_warning("map '%s': map type isn't specified.\n", map_name);
1264                 return -EINVAL;
1265         }
1266 
1267         return 0;
1268 }
1269 
1270 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1271 {
1272         const struct btf_type *sec = NULL;
1273         int nr_types, i, vlen, err;
1274         const struct btf_type *t;
1275         const char *name;
1276         Elf_Data *data;
1277         Elf_Scn *scn;
1278 
1279         if (obj->efile.btf_maps_shndx < 0)
1280                 return 0;
1281 
1282         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1283         if (scn)
1284                 data = elf_getdata(scn, NULL);
1285         if (!scn || !data) {
1286                 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1287                            obj->efile.maps_shndx, MAPS_ELF_SEC);
1288                 return -EINVAL;
1289         }
1290 
1291         nr_types = btf__get_nr_types(obj->btf);
1292         for (i = 1; i <= nr_types; i++) {
1293                 t = btf__type_by_id(obj->btf, i);
1294                 if (!btf_is_datasec(t))
1295                         continue;
1296                 name = btf__name_by_offset(obj->btf, t->name_off);
1297                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1298                         sec = t;
1299                         break;
1300                 }
1301         }
1302 
1303         if (!sec) {
1304                 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1305                 return -ENOENT;
1306         }
1307 
1308         vlen = btf_vlen(sec);
1309         for (i = 0; i < vlen; i++) {
1310                 err = bpf_object__init_user_btf_map(obj, sec, i,
1311                                                     obj->efile.btf_maps_shndx,
1312                                                     data, strict);
1313                 if (err)
1314                         return err;
1315         }
1316 
1317         return 0;
1318 }
1319 
1320 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1321 {
1322         bool strict = !(flags & MAPS_RELAX_COMPAT);
1323         int err;
1324 
1325         err = bpf_object__init_user_maps(obj, strict);
1326         if (err)
1327                 return err;
1328 
1329         err = bpf_object__init_user_btf_maps(obj, strict);
1330         if (err)
1331                 return err;
1332 
1333         err = bpf_object__init_global_data_maps(obj);
1334         if (err)
1335                 return err;
1336 
1337         if (obj->nr_maps) {
1338                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1339                       compare_bpf_map);
1340         }
1341         return 0;
1342 }
1343 
1344 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1345 {
1346         Elf_Scn *scn;
1347         GElf_Shdr sh;
1348 
1349         scn = elf_getscn(obj->efile.elf, idx);
1350         if (!scn)
1351                 return false;
1352 
1353         if (gelf_getshdr(scn, &sh) != &sh)
1354                 return false;
1355 
1356         if (sh.sh_flags & SHF_EXECINSTR)
1357                 return true;
1358 
1359         return false;
1360 }
1361 
1362 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1363 {
1364         bool has_datasec = obj->caps.btf_datasec;
1365         bool has_func = obj->caps.btf_func;
1366         struct btf *btf = obj->btf;
1367         struct btf_type *t;
1368         int i, j, vlen;
1369 
1370         if (!obj->btf || (has_func && has_datasec))
1371                 return;
1372 
1373         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1374                 t = (struct btf_type *)btf__type_by_id(btf, i);
1375 
1376                 if (!has_datasec && btf_is_var(t)) {
1377                         /* replace VAR with INT */
1378                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1379                         /*
1380                          * using size = 1 is the safest choice, 4 will be too
1381                          * big and cause kernel BTF validation failure if
1382                          * original variable took less than 4 bytes
1383                          */
1384                         t->size = 1;
1385                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1386                 } else if (!has_datasec && btf_is_datasec(t)) {
1387                         /* replace DATASEC with STRUCT */
1388                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
1389                         struct btf_member *m = btf_members(t);
1390                         struct btf_type *vt;
1391                         char *name;
1392 
1393                         name = (char *)btf__name_by_offset(btf, t->name_off);
1394                         while (*name) {
1395                                 if (*name == '.')
1396                                         *name = '_';
1397                                 name++;
1398                         }
1399 
1400                         vlen = btf_vlen(t);
1401                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1402                         for (j = 0; j < vlen; j++, v++, m++) {
1403                                 /* order of field assignments is important */
1404                                 m->offset = v->offset * 8;
1405                                 m->type = v->type;
1406                                 /* preserve variable name as member name */
1407                                 vt = (void *)btf__type_by_id(btf, v->type);
1408                                 m->name_off = vt->name_off;
1409                         }
1410                 } else if (!has_func && btf_is_func_proto(t)) {
1411                         /* replace FUNC_PROTO with ENUM */
1412                         vlen = btf_vlen(t);
1413                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1414                         t->size = sizeof(__u32); /* kernel enforced */
1415                 } else if (!has_func && btf_is_func(t)) {
1416                         /* replace FUNC with TYPEDEF */
1417                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1418                 }
1419         }
1420 }
1421 
1422 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1423 {
1424         if (!obj->btf_ext)
1425                 return;
1426 
1427         if (!obj->caps.btf_func) {
1428                 btf_ext__free(obj->btf_ext);
1429                 obj->btf_ext = NULL;
1430         }
1431 }
1432 
1433 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1434 {
1435         return obj->efile.btf_maps_shndx >= 0;
1436 }
1437 
1438 static int bpf_object__init_btf(struct bpf_object *obj,
1439                                 Elf_Data *btf_data,
1440                                 Elf_Data *btf_ext_data)
1441 {
1442         bool btf_required = bpf_object__is_btf_mandatory(obj);
1443         int err = 0;
1444 
1445         if (btf_data) {
1446                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1447                 if (IS_ERR(obj->btf)) {
1448                         pr_warning("Error loading ELF section %s: %d.\n",
1449                                    BTF_ELF_SEC, err);
1450                         goto out;
1451                 }
1452                 err = btf__finalize_data(obj, obj->btf);
1453                 if (err) {
1454                         pr_warning("Error finalizing %s: %d.\n",
1455                                    BTF_ELF_SEC, err);
1456                         goto out;
1457                 }
1458         }
1459         if (btf_ext_data) {
1460                 if (!obj->btf) {
1461                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1462                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1463                         goto out;
1464                 }
1465                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1466                                             btf_ext_data->d_size);
1467                 if (IS_ERR(obj->btf_ext)) {
1468                         pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1469                                    BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1470                         obj->btf_ext = NULL;
1471                         goto out;
1472                 }
1473         }
1474 out:
1475         if (err || IS_ERR(obj->btf)) {
1476                 if (btf_required)
1477                         err = err ? : PTR_ERR(obj->btf);
1478                 else
1479                         err = 0;
1480                 if (!IS_ERR_OR_NULL(obj->btf))
1481                         btf__free(obj->btf);
1482                 obj->btf = NULL;
1483         }
1484         if (btf_required && !obj->btf) {
1485                 pr_warning("BTF is required, but is missing or corrupted.\n");
1486                 return err == 0 ? -ENOENT : err;
1487         }
1488         return 0;
1489 }
1490 
1491 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1492 {
1493         int err = 0;
1494 
1495         if (!obj->btf)
1496                 return 0;
1497 
1498         bpf_object__sanitize_btf(obj);
1499         bpf_object__sanitize_btf_ext(obj);
1500 
1501         err = btf__load(obj->btf);
1502         if (err) {
1503                 pr_warning("Error loading %s into kernel: %d.\n",
1504                            BTF_ELF_SEC, err);
1505                 btf__free(obj->btf);
1506                 obj->btf = NULL;
1507                 /* btf_ext can't exist without btf, so free it as well */
1508                 if (obj->btf_ext) {
1509                         btf_ext__free(obj->btf_ext);
1510                         obj->btf_ext = NULL;
1511                 }
1512 
1513                 if (bpf_object__is_btf_mandatory(obj))
1514                         return err;
1515         }
1516         return 0;
1517 }
1518 
1519 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1520 {
1521         Elf *elf = obj->efile.elf;
1522         GElf_Ehdr *ep = &obj->efile.ehdr;
1523         Elf_Data *btf_ext_data = NULL;
1524         Elf_Data *btf_data = NULL;
1525         Elf_Scn *scn = NULL;
1526         int idx = 0, err = 0;
1527 
1528         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1529         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1530                 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1531                 return -LIBBPF_ERRNO__FORMAT;
1532         }
1533 
1534         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1535                 char *name;
1536                 GElf_Shdr sh;
1537                 Elf_Data *data;
1538 
1539                 idx++;
1540                 if (gelf_getshdr(scn, &sh) != &sh) {
1541                         pr_warning("failed to get section(%d) header from %s\n",
1542                                    idx, obj->path);
1543                         return -LIBBPF_ERRNO__FORMAT;
1544                 }
1545 
1546                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1547                 if (!name) {
1548                         pr_warning("failed to get section(%d) name from %s\n",
1549                                    idx, obj->path);
1550                         return -LIBBPF_ERRNO__FORMAT;
1551                 }
1552 
1553                 data = elf_getdata(scn, 0);
1554                 if (!data) {
1555                         pr_warning("failed to get section(%d) data from %s(%s)\n",
1556                                    idx, name, obj->path);
1557                         return -LIBBPF_ERRNO__FORMAT;
1558                 }
1559                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1560                          idx, name, (unsigned long)data->d_size,
1561                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1562                          (int)sh.sh_type);
1563 
1564                 if (strcmp(name, "license") == 0) {
1565                         err = bpf_object__init_license(obj,
1566                                                        data->d_buf,
1567                                                        data->d_size);
1568                         if (err)
1569                                 return err;
1570                 } else if (strcmp(name, "version") == 0) {
1571                         err = bpf_object__init_kversion(obj,
1572                                                         data->d_buf,
1573                                                         data->d_size);
1574                         if (err)
1575                                 return err;
1576                 } else if (strcmp(name, "maps") == 0) {
1577                         obj->efile.maps_shndx = idx;
1578                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1579                         obj->efile.btf_maps_shndx = idx;
1580                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1581                         btf_data = data;
1582                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1583                         btf_ext_data = data;
1584                 } else if (sh.sh_type == SHT_SYMTAB) {
1585                         if (obj->efile.symbols) {
1586                                 pr_warning("bpf: multiple SYMTAB in %s\n",
1587                                            obj->path);
1588                                 return -LIBBPF_ERRNO__FORMAT;
1589                         }
1590                         obj->efile.symbols = data;
1591                         obj->efile.strtabidx = sh.sh_link;
1592                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1593                         if (sh.sh_flags & SHF_EXECINSTR) {
1594                                 if (strcmp(name, ".text") == 0)
1595                                         obj->efile.text_shndx = idx;
1596                                 err = bpf_object__add_program(obj, data->d_buf,
1597                                                               data->d_size, name, idx);
1598                                 if (err) {
1599                                         char errmsg[STRERR_BUFSIZE];
1600                                         char *cp = libbpf_strerror_r(-err, errmsg,
1601                                                                      sizeof(errmsg));
1602 
1603                                         pr_warning("failed to alloc program %s (%s): %s",
1604                                                    name, obj->path, cp);
1605                                         return err;
1606                                 }
1607                         } else if (strcmp(name, ".data") == 0) {
1608                                 obj->efile.data = data;
1609                                 obj->efile.data_shndx = idx;
1610                         } else if (strcmp(name, ".rodata") == 0) {
1611                                 obj->efile.rodata = data;
1612                                 obj->efile.rodata_shndx = idx;
1613                         } else {
1614                                 pr_debug("skip section(%d) %s\n", idx, name);
1615                         }
1616                 } else if (sh.sh_type == SHT_REL) {
1617                         int nr_reloc = obj->efile.nr_reloc;
1618                         void *reloc = obj->efile.reloc;
1619                         int sec = sh.sh_info; /* points to other section */
1620 
1621                         /* Only do relo for section with exec instructions */
1622                         if (!section_have_execinstr(obj, sec)) {
1623                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1624                                          name, idx, sec);
1625                                 continue;
1626                         }
1627 
1628                         reloc = reallocarray(reloc, nr_reloc + 1,
1629                                              sizeof(*obj->efile.reloc));
1630                         if (!reloc) {
1631                                 pr_warning("realloc failed\n");
1632                                 return -ENOMEM;
1633                         }
1634 
1635                         obj->efile.reloc = reloc;
1636                         obj->efile.nr_reloc++;
1637 
1638                         obj->efile.reloc[nr_reloc].shdr = sh;
1639                         obj->efile.reloc[nr_reloc].data = data;
1640                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1641                         obj->efile.bss = data;
1642                         obj->efile.bss_shndx = idx;
1643                 } else {
1644                         pr_debug("skip section(%d) %s\n", idx, name);
1645                 }
1646         }
1647 
1648         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1649                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1650                 return -LIBBPF_ERRNO__FORMAT;
1651         }
1652         err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1653         if (!err)
1654                 err = bpf_object__init_maps(obj, flags);
1655         if (!err)
1656                 err = bpf_object__sanitize_and_load_btf(obj);
1657         if (!err)
1658                 err = bpf_object__init_prog_names(obj);
1659         return err;
1660 }
1661 
1662 static struct bpf_program *
1663 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1664 {
1665         struct bpf_program *prog;
1666         size_t i;
1667 
1668         for (i = 0; i < obj->nr_programs; i++) {
1669                 prog = &obj->programs[i];
1670                 if (prog->idx == idx)
1671                         return prog;
1672         }
1673         return NULL;
1674 }
1675 
1676 struct bpf_program *
1677 bpf_object__find_program_by_title(const struct bpf_object *obj,
1678                                   const char *title)
1679 {
1680         struct bpf_program *pos;
1681 
1682         bpf_object__for_each_program(pos, obj) {
1683                 if (pos->section_name && !strcmp(pos->section_name, title))
1684                         return pos;
1685         }
1686         return NULL;
1687 }
1688 
1689 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1690                                       int shndx)
1691 {
1692         return shndx == obj->efile.data_shndx ||
1693                shndx == obj->efile.bss_shndx ||
1694                shndx == obj->efile.rodata_shndx;
1695 }
1696 
1697 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1698                                       int shndx)
1699 {
1700         return shndx == obj->efile.maps_shndx ||
1701                shndx == obj->efile.btf_maps_shndx;
1702 }
1703 
1704 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1705                                               int shndx)
1706 {
1707         return shndx == obj->efile.text_shndx ||
1708                bpf_object__shndx_is_maps(obj, shndx) ||
1709                bpf_object__shndx_is_data(obj, shndx);
1710 }
1711 
1712 static enum libbpf_map_type
1713 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1714 {
1715         if (shndx == obj->efile.data_shndx)
1716                 return LIBBPF_MAP_DATA;
1717         else if (shndx == obj->efile.bss_shndx)
1718                 return LIBBPF_MAP_BSS;
1719         else if (shndx == obj->efile.rodata_shndx)
1720                 return LIBBPF_MAP_RODATA;
1721         else
1722                 return LIBBPF_MAP_UNSPEC;
1723 }
1724 
1725 static int
1726 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1727                            Elf_Data *data, struct bpf_object *obj)
1728 {
1729         Elf_Data *symbols = obj->efile.symbols;
1730         struct bpf_map *maps = obj->maps;
1731         size_t nr_maps = obj->nr_maps;
1732         int i, nrels;
1733 
1734         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1735         nrels = shdr->sh_size / shdr->sh_entsize;
1736 
1737         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1738         if (!prog->reloc_desc) {
1739                 pr_warning("failed to alloc memory in relocation\n");
1740                 return -ENOMEM;
1741         }
1742         prog->nr_reloc = nrels;
1743 
1744         for (i = 0; i < nrels; i++) {
1745                 struct bpf_insn *insns = prog->insns;
1746                 enum libbpf_map_type type;
1747                 unsigned int insn_idx;
1748                 unsigned int shdr_idx;
1749                 const char *name;
1750                 size_t map_idx;
1751                 GElf_Sym sym;
1752                 GElf_Rel rel;
1753 
1754                 if (!gelf_getrel(data, i, &rel)) {
1755                         pr_warning("relocation: failed to get %d reloc\n", i);
1756                         return -LIBBPF_ERRNO__FORMAT;
1757                 }
1758 
1759                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1760                         pr_warning("relocation: symbol %"PRIx64" not found\n",
1761                                    GELF_R_SYM(rel.r_info));
1762                         return -LIBBPF_ERRNO__FORMAT;
1763                 }
1764 
1765                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1766                                   sym.st_name) ? : "<?>";
1767 
1768                 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1769                          (long long) (rel.r_info >> 32),
1770                          (long long) sym.st_value, sym.st_name, name);
1771 
1772                 shdr_idx = sym.st_shndx;
1773                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1774                 pr_debug("relocation: insn_idx=%u, shdr_idx=%u\n",
1775                          insn_idx, shdr_idx);
1776 
1777                 if (shdr_idx >= SHN_LORESERVE) {
1778                         pr_warning("relocation: not yet supported relo for non-static global \'%s\' variable in special section (0x%x) found in insns[%d].code 0x%x\n",
1779                                    name, shdr_idx, insn_idx,
1780                                    insns[insn_idx].code);
1781                         return -LIBBPF_ERRNO__RELOC;
1782                 }
1783                 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1784                         pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1785                                    prog->section_name, shdr_idx);
1786                         return -LIBBPF_ERRNO__RELOC;
1787                 }
1788 
1789                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1790                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1791                                 pr_warning("incorrect bpf_call opcode\n");
1792                                 return -LIBBPF_ERRNO__RELOC;
1793                         }
1794                         prog->reloc_desc[i].type = RELO_CALL;
1795                         prog->reloc_desc[i].insn_idx = insn_idx;
1796                         prog->reloc_desc[i].text_off = sym.st_value;
1797                         obj->has_pseudo_calls = true;
1798                         continue;
1799                 }
1800 
1801                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1802                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1803                                    insn_idx, insns[insn_idx].code);
1804                         return -LIBBPF_ERRNO__RELOC;
1805                 }
1806 
1807                 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1808                     bpf_object__shndx_is_data(obj, shdr_idx)) {
1809                         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1810                         if (type != LIBBPF_MAP_UNSPEC) {
1811                                 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1812                                         pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1813                                                    name, insn_idx, insns[insn_idx].code);
1814                                         return -LIBBPF_ERRNO__RELOC;
1815                                 }
1816                                 if (!obj->caps.global_data) {
1817                                         pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1818                                                    name, insn_idx);
1819                                         return -LIBBPF_ERRNO__RELOC;
1820                                 }
1821                         }
1822 
1823                         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1824                                 if (maps[map_idx].libbpf_type != type)
1825                                         continue;
1826                                 if (type != LIBBPF_MAP_UNSPEC ||
1827                                     (maps[map_idx].sec_idx == sym.st_shndx &&
1828                                      maps[map_idx].sec_offset == sym.st_value)) {
1829                                         pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1830                                                  map_idx, maps[map_idx].name,
1831                                                  maps[map_idx].sec_idx,
1832                                                  maps[map_idx].sec_offset,
1833                                                  insn_idx);
1834                                         break;
1835                                 }
1836                         }
1837 
1838                         if (map_idx >= nr_maps) {
1839                                 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1840                                            (int)map_idx, (int)nr_maps - 1);
1841                                 return -LIBBPF_ERRNO__RELOC;
1842                         }
1843 
1844                         prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1845                                                    RELO_DATA : RELO_LD64;
1846                         prog->reloc_desc[i].insn_idx = insn_idx;
1847                         prog->reloc_desc[i].map_idx = map_idx;
1848                 }
1849         }
1850         return 0;
1851 }
1852 
1853 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1854 {
1855         struct bpf_map_def *def = &map->def;
1856         __u32 key_type_id = 0, value_type_id = 0;
1857         int ret;
1858 
1859         /* if it's BTF-defined map, we don't need to search for type IDs */
1860         if (map->sec_idx == obj->efile.btf_maps_shndx)
1861                 return 0;
1862 
1863         if (!bpf_map__is_internal(map)) {
1864                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1865                                            def->value_size, &key_type_id,
1866                                            &value_type_id);
1867         } else {
1868                 /*
1869                  * LLVM annotates global data differently in BTF, that is,
1870                  * only as '.data', '.bss' or '.rodata'.
1871                  */
1872                 ret = btf__find_by_name(obj->btf,
1873                                 libbpf_type_to_btf_name[map->libbpf_type]);
1874         }
1875         if (ret < 0)
1876                 return ret;
1877 
1878         map->btf_key_type_id = key_type_id;
1879         map->btf_value_type_id = bpf_map__is_internal(map) ?
1880                                  ret : value_type_id;
1881         return 0;
1882 }
1883 
1884 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1885 {
1886         struct bpf_map_info info = {};
1887         __u32 len = sizeof(info);
1888         int new_fd, err;
1889         char *new_name;
1890 
1891         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1892         if (err)
1893                 return err;
1894 
1895         new_name = strdup(info.name);
1896         if (!new_name)
1897                 return -errno;
1898 
1899         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1900         if (new_fd < 0) {
1901                 err = -errno;
1902                 goto err_free_new_name;
1903         }
1904 
1905         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1906         if (new_fd < 0) {
1907                 err = -errno;
1908                 goto err_close_new_fd;
1909         }
1910 
1911         err = zclose(map->fd);
1912         if (err) {
1913                 err = -errno;
1914                 goto err_close_new_fd;
1915         }
1916         free(map->name);
1917 
1918         map->fd = new_fd;
1919         map->name = new_name;
1920         map->def.type = info.type;
1921         map->def.key_size = info.key_size;
1922         map->def.value_size = info.value_size;
1923         map->def.max_entries = info.max_entries;
1924         map->def.map_flags = info.map_flags;
1925         map->btf_key_type_id = info.btf_key_type_id;
1926         map->btf_value_type_id = info.btf_value_type_id;
1927 
1928         return 0;
1929 
1930 err_close_new_fd:
1931         close(new_fd);
1932 err_free_new_name:
1933         free(new_name);
1934         return err;
1935 }
1936 
1937 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1938 {
1939         if (!map || !max_entries)
1940                 return -EINVAL;
1941 
1942         /* If map already created, its attributes can't be changed. */
1943         if (map->fd >= 0)
1944                 return -EBUSY;
1945 
1946         map->def.max_entries = max_entries;
1947 
1948         return 0;
1949 }
1950 
1951 static int
1952 bpf_object__probe_name(struct bpf_object *obj)
1953 {
1954         struct bpf_load_program_attr attr;
1955         char *cp, errmsg[STRERR_BUFSIZE];
1956         struct bpf_insn insns[] = {
1957                 BPF_MOV64_IMM(BPF_REG_0, 0),
1958                 BPF_EXIT_INSN(),
1959         };
1960         int ret;
1961 
1962         /* make sure basic loading works */
1963 
1964         memset(&attr, 0, sizeof(attr));
1965         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1966         attr.insns = insns;
1967         attr.insns_cnt = ARRAY_SIZE(insns);
1968         attr.license = "GPL";
1969 
1970         ret = bpf_load_program_xattr(&attr, NULL, 0);
1971         if (ret < 0) {
1972                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1973                 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1974                            __func__, cp, errno);
1975                 return -errno;
1976         }
1977         close(ret);
1978 
1979         /* now try the same program, but with the name */
1980 
1981         attr.name = "test";
1982         ret = bpf_load_program_xattr(&attr, NULL, 0);
1983         if (ret >= 0) {
1984                 obj->caps.name = 1;
1985                 close(ret);
1986         }
1987 
1988         return 0;
1989 }
1990 
1991 static int
1992 bpf_object__probe_global_data(struct bpf_object *obj)
1993 {
1994         struct bpf_load_program_attr prg_attr;
1995         struct bpf_create_map_attr map_attr;
1996         char *cp, errmsg[STRERR_BUFSIZE];
1997         struct bpf_insn insns[] = {
1998                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1999                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
2000                 BPF_MOV64_IMM(BPF_REG_0, 0),
2001                 BPF_EXIT_INSN(),
2002         };
2003         int ret, map;
2004 
2005         memset(&map_attr, 0, sizeof(map_attr));
2006         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2007         map_attr.key_size = sizeof(int);
2008         map_attr.value_size = 32;
2009         map_attr.max_entries = 1;
2010 
2011         map = bpf_create_map_xattr(&map_attr);
2012         if (map < 0) {
2013                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2014                 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2015                            __func__, cp, errno);
2016                 return -errno;
2017         }
2018 
2019         insns[0].imm = map;
2020 
2021         memset(&prg_attr, 0, sizeof(prg_attr));
2022         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2023         prg_attr.insns = insns;
2024         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2025         prg_attr.license = "GPL";
2026 
2027         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2028         if (ret >= 0) {
2029                 obj->caps.global_data = 1;
2030                 close(ret);
2031         }
2032 
2033         close(map);
2034         return 0;
2035 }
2036 
2037 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2038 {
2039         const char strs[] = "\0int\0x\0a";
2040         /* void x(int a) {} */
2041         __u32 types[] = {
2042                 /* int */
2043                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2044                 /* FUNC_PROTO */                                /* [2] */
2045                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2046                 BTF_PARAM_ENC(7, 1),
2047                 /* FUNC x */                                    /* [3] */
2048                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2049         };
2050         int btf_fd;
2051 
2052         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2053                                       strs, sizeof(strs));
2054         if (btf_fd >= 0) {
2055                 obj->caps.btf_func = 1;
2056                 close(btf_fd);
2057                 return 1;
2058         }
2059 
2060         return 0;
2061 }
2062 
2063 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2064 {
2065         const char strs[] = "\0x\0.data";
2066         /* static int a; */
2067         __u32 types[] = {
2068                 /* int */
2069                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2070                 /* VAR x */                                     /* [2] */
2071                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2072                 BTF_VAR_STATIC,
2073                 /* DATASEC val */                               /* [3] */
2074                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2075                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2076         };
2077         int btf_fd;
2078 
2079         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2080                                       strs, sizeof(strs));
2081         if (btf_fd >= 0) {
2082                 obj->caps.btf_datasec = 1;
2083                 close(btf_fd);
2084                 return 1;
2085         }
2086 
2087         return 0;
2088 }
2089 
2090 static int
2091 bpf_object__probe_caps(struct bpf_object *obj)
2092 {
2093         int (*probe_fn[])(struct bpf_object *obj) = {
2094                 bpf_object__probe_name,
2095                 bpf_object__probe_global_data,
2096                 bpf_object__probe_btf_func,
2097                 bpf_object__probe_btf_datasec,
2098         };
2099         int i, ret;
2100 
2101         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2102                 ret = probe_fn[i](obj);
2103                 if (ret < 0)
2104                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2105         }
2106 
2107         return 0;
2108 }
2109 
2110 static int
2111 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2112 {
2113         char *cp, errmsg[STRERR_BUFSIZE];
2114         int err, zero = 0;
2115         __u8 *data;
2116 
2117         /* Nothing to do here since kernel already zero-initializes .bss map. */
2118         if (map->libbpf_type == LIBBPF_MAP_BSS)
2119                 return 0;
2120 
2121         data = map->libbpf_type == LIBBPF_MAP_DATA ?
2122                obj->sections.data : obj->sections.rodata;
2123 
2124         err = bpf_map_update_elem(map->fd, &zero, data, 0);
2125         /* Freeze .rodata map as read-only from syscall side. */
2126         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2127                 err = bpf_map_freeze(map->fd);
2128                 if (err) {
2129                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2130                         pr_warning("Error freezing map(%s) as read-only: %s\n",
2131                                    map->name, cp);
2132                         err = 0;
2133                 }
2134         }
2135         return err;
2136 }
2137 
2138 static int
2139 bpf_object__create_maps(struct bpf_object *obj)
2140 {
2141         struct bpf_create_map_attr create_attr = {};
2142         int nr_cpus = 0;
2143         unsigned int i;
2144         int err;
2145 
2146         for (i = 0; i < obj->nr_maps; i++) {
2147                 struct bpf_map *map = &obj->maps[i];
2148                 struct bpf_map_def *def = &map->def;
2149                 char *cp, errmsg[STRERR_BUFSIZE];
2150                 int *pfd = &map->fd;
2151 
2152                 if (map->fd >= 0) {
2153                         pr_debug("skip map create (preset) %s: fd=%d\n",
2154                                  map->name, map->fd);
2155                         continue;
2156                 }
2157 
2158                 if (obj->caps.name)
2159                         create_attr.name = map->name;
2160                 create_attr.map_ifindex = map->map_ifindex;
2161                 create_attr.map_type = def->type;
2162                 create_attr.map_flags = def->map_flags;
2163                 create_attr.key_size = def->key_size;
2164                 create_attr.value_size = def->value_size;
2165                 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2166                     !def->max_entries) {
2167                         if (!nr_cpus)
2168                                 nr_cpus = libbpf_num_possible_cpus();
2169                         if (nr_cpus < 0) {
2170                                 pr_warning("failed to determine number of system CPUs: %d\n",
2171                                            nr_cpus);
2172                                 err = nr_cpus;
2173                                 goto err_out;
2174                         }
2175                         pr_debug("map '%s': setting size to %d\n",
2176                                  map->name, nr_cpus);
2177                         create_attr.max_entries = nr_cpus;
2178                 } else {
2179                         create_attr.max_entries = def->max_entries;
2180                 }
2181                 create_attr.btf_fd = 0;
2182                 create_attr.btf_key_type_id = 0;
2183                 create_attr.btf_value_type_id = 0;
2184                 if (bpf_map_type__is_map_in_map(def->type) &&
2185                     map->inner_map_fd >= 0)
2186                         create_attr.inner_map_fd = map->inner_map_fd;
2187 
2188                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2189                         create_attr.btf_fd = btf__fd(obj->btf);
2190                         create_attr.btf_key_type_id = map->btf_key_type_id;
2191                         create_attr.btf_value_type_id = map->btf_value_type_id;
2192                 }
2193 
2194                 *pfd = bpf_create_map_xattr(&create_attr);
2195                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2196                                  create_attr.btf_value_type_id)) {
2197                         err = -errno;
2198                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2199                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2200                                    map->name, cp, err);
2201                         create_attr.btf_fd = 0;
2202                         create_attr.btf_key_type_id = 0;
2203                         create_attr.btf_value_type_id = 0;
2204                         map->btf_key_type_id = 0;
2205                         map->btf_value_type_id = 0;
2206                         *pfd = bpf_create_map_xattr(&create_attr);
2207                 }
2208 
2209                 if (*pfd < 0) {
2210                         size_t j;
2211 
2212                         err = -errno;
2213 err_out:
2214                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2215                         pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2216                                    map->name, cp, err);
2217                         for (j = 0; j < i; j++)
2218                                 zclose(obj->maps[j].fd);
2219                         return err;
2220                 }
2221 
2222                 if (bpf_map__is_internal(map)) {
2223                         err = bpf_object__populate_internal_map(obj, map);
2224                         if (err < 0) {
2225                                 zclose(*pfd);
2226                                 goto err_out;
2227                         }
2228                 }
2229 
2230                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2231         }
2232 
2233         return 0;
2234 }
2235 
2236 static int
2237 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2238                         void *btf_prog_info, const char *info_name)
2239 {
2240         if (err != -ENOENT) {
2241                 pr_warning("Error in loading %s for sec %s.\n",
2242                            info_name, prog->section_name);
2243                 return err;
2244         }
2245 
2246         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2247 
2248         if (btf_prog_info) {
2249                 /*
2250                  * Some info has already been found but has problem
2251                  * in the last btf_ext reloc. Must have to error out.
2252                  */
2253                 pr_warning("Error in relocating %s for sec %s.\n",
2254                            info_name, prog->section_name);
2255                 return err;
2256         }
2257 
2258         /* Have problem loading the very first info. Ignore the rest. */
2259         pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2260                    info_name, prog->section_name, info_name);
2261         return 0;
2262 }
2263 
2264 static int
2265 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2266                           const char *section_name,  __u32 insn_offset)
2267 {
2268         int err;
2269 
2270         if (!insn_offset || prog->func_info) {
2271                 /*
2272                  * !insn_offset => main program
2273                  *
2274                  * For sub prog, the main program's func_info has to
2275                  * be loaded first (i.e. prog->func_info != NULL)
2276                  */
2277                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2278                                                section_name, insn_offset,
2279                                                &prog->func_info,
2280                                                &prog->func_info_cnt);
2281                 if (err)
2282                         return check_btf_ext_reloc_err(prog, err,
2283                                                        prog->func_info,
2284                                                        "bpf_func_info");
2285 
2286                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2287         }
2288 
2289         if (!insn_offset || prog->line_info) {
2290                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2291                                                section_name, insn_offset,
2292                                                &prog->line_info,
2293                                                &prog->line_info_cnt);
2294                 if (err)
2295                         return check_btf_ext_reloc_err(prog, err,
2296                                                        prog->line_info,
2297                                                        "bpf_line_info");
2298 
2299                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2300         }
2301 
2302         return 0;
2303 }
2304 
2305 #define BPF_CORE_SPEC_MAX_LEN 64
2306 
2307 /* represents BPF CO-RE field or array element accessor */
2308 struct bpf_core_accessor {
2309         __u32 type_id;          /* struct/union type or array element type */
2310         __u32 idx;              /* field index or array index */
2311         const char *name;       /* field name or NULL for array accessor */
2312 };
2313 
2314 struct bpf_core_spec {
2315         const struct btf *btf;
2316         /* high-level spec: named fields and array indices only */
2317         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2318         /* high-level spec length */
2319         int len;
2320         /* raw, low-level spec: 1-to-1 with accessor spec string */
2321         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2322         /* raw spec length */
2323         int raw_len;
2324         /* field byte offset represented by spec */
2325         __u32 offset;
2326 };
2327 
2328 static bool str_is_empty(const char *s)
2329 {
2330         return !s || !s[0];
2331 }
2332 
2333 /*
2334  * Turn bpf_offset_reloc into a low- and high-level spec representation,
2335  * validating correctness along the way, as well as calculating resulting
2336  * field offset (in bytes), specified by accessor string. Low-level spec
2337  * captures every single level of nestedness, including traversing anonymous
2338  * struct/union members. High-level one only captures semantically meaningful
2339  * "turning points": named fields and array indicies.
2340  * E.g., for this case:
2341  *
2342  *   struct sample {
2343  *       int __unimportant;
2344  *       struct {
2345  *           int __1;
2346  *           int __2;
2347  *           int a[7];
2348  *       };
2349  *   };
2350  *
2351  *   struct sample *s = ...;
2352  *
2353  *   int x = &s->a[3]; // access string = '0:1:2:3'
2354  *
2355  * Low-level spec has 1:1 mapping with each element of access string (it's
2356  * just a parsed access string representation): [0, 1, 2, 3].
2357  *
2358  * High-level spec will capture only 3 points:
2359  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2360  *   - field 'a' access (corresponds to '2' in low-level spec);
2361  *   - array element #3 access (corresponds to '3' in low-level spec).
2362  *
2363  */
2364 static int bpf_core_spec_parse(const struct btf *btf,
2365                                __u32 type_id,
2366                                const char *spec_str,
2367                                struct bpf_core_spec *spec)
2368 {
2369         int access_idx, parsed_len, i;
2370         const struct btf_type *t;
2371         const char *name;
2372         __u32 id;
2373         __s64 sz;
2374 
2375         if (str_is_empty(spec_str) || *spec_str == ':')
2376                 return -EINVAL;
2377 
2378         memset(spec, 0, sizeof(*spec));
2379         spec->btf = btf;
2380 
2381         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2382         while (*spec_str) {
2383                 if (*spec_str == ':')
2384                         ++spec_str;
2385                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2386                         return -EINVAL;
2387                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2388                         return -E2BIG;
2389                 spec_str += parsed_len;
2390                 spec->raw_spec[spec->raw_len++] = access_idx;
2391         }
2392 
2393         if (spec->raw_len == 0)
2394                 return -EINVAL;
2395 
2396         /* first spec value is always reloc type array index */
2397         t = skip_mods_and_typedefs(btf, type_id, &id);
2398         if (!t)
2399                 return -EINVAL;
2400 
2401         access_idx = spec->raw_spec[0];
2402         spec->spec[0].type_id = id;
2403         spec->spec[0].idx = access_idx;
2404         spec->len++;
2405 
2406         sz = btf__resolve_size(btf, id);
2407         if (sz < 0)
2408                 return sz;
2409         spec->offset = access_idx * sz;
2410 
2411         for (i = 1; i < spec->raw_len; i++) {
2412                 t = skip_mods_and_typedefs(btf, id, &id);
2413                 if (!t)
2414                         return -EINVAL;
2415 
2416                 access_idx = spec->raw_spec[i];
2417 
2418                 if (btf_is_composite(t)) {
2419                         const struct btf_member *m;
2420                         __u32 offset;
2421 
2422                         if (access_idx >= btf_vlen(t))
2423                                 return -EINVAL;
2424                         if (btf_member_bitfield_size(t, access_idx))
2425                                 return -EINVAL;
2426 
2427                         offset = btf_member_bit_offset(t, access_idx);
2428                         if (offset % 8)
2429                                 return -EINVAL;
2430                         spec->offset += offset / 8;
2431 
2432                         m = btf_members(t) + access_idx;
2433                         if (m->name_off) {
2434                                 name = btf__name_by_offset(btf, m->name_off);
2435                                 if (str_is_empty(name))
2436                                         return -EINVAL;
2437 
2438                                 spec->spec[spec->len].type_id = id;
2439                                 spec->spec[spec->len].idx = access_idx;
2440                                 spec->spec[spec->len].name = name;
2441                                 spec->len++;
2442                         }
2443 
2444                         id = m->type;
2445                 } else if (btf_is_array(t)) {
2446                         const struct btf_array *a = btf_array(t);
2447 
2448                         t = skip_mods_and_typedefs(btf, a->type, &id);
2449                         if (!t || access_idx >= a->nelems)
2450                                 return -EINVAL;
2451 
2452                         spec->spec[spec->len].type_id = id;
2453                         spec->spec[spec->len].idx = access_idx;
2454                         spec->len++;
2455 
2456                         sz = btf__resolve_size(btf, id);
2457                         if (sz < 0)
2458                                 return sz;
2459                         spec->offset += access_idx * sz;
2460                 } else {
2461                         pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2462                                    type_id, spec_str, i, id, btf_kind(t));
2463                         return -EINVAL;
2464                 }
2465         }
2466 
2467         return 0;
2468 }
2469 
2470 static bool bpf_core_is_flavor_sep(const char *s)
2471 {
2472         /* check X___Y name pattern, where X and Y are not underscores */
2473         return s[0] != '_' &&                                 /* X */
2474                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
2475                s[4] != '_';                                   /* Y */
2476 }
2477 
2478 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2479  * before last triple underscore. Struct name part after last triple
2480  * underscore is ignored by BPF CO-RE relocation during relocation matching.
2481  */
2482 static size_t bpf_core_essential_name_len(const char *name)
2483 {
2484         size_t n = strlen(name);
2485         int i;
2486 
2487         for (i = n - 5; i >= 0; i--) {
2488                 if (bpf_core_is_flavor_sep(name + i))
2489                         return i + 1;
2490         }
2491         return n;
2492 }
2493 
2494 /* dynamically sized list of type IDs */
2495 struct ids_vec {
2496         __u32 *data;
2497         int len;
2498 };
2499 
2500 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2501 {
2502         free(cand_ids->data);
2503         free(cand_ids);
2504 }
2505 
2506 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2507                                            __u32 local_type_id,
2508                                            const struct btf *targ_btf)
2509 {
2510         size_t local_essent_len, targ_essent_len;
2511         const char *local_name, *targ_name;
2512         const struct btf_type *t;
2513         struct ids_vec *cand_ids;
2514         __u32 *new_ids;
2515         int i, err, n;
2516 
2517         t = btf__type_by_id(local_btf, local_type_id);
2518         if (!t)
2519                 return ERR_PTR(-EINVAL);
2520 
2521         local_name = btf__name_by_offset(local_btf, t->name_off);
2522         if (str_is_empty(local_name))
2523                 return ERR_PTR(-EINVAL);
2524         local_essent_len = bpf_core_essential_name_len(local_name);
2525 
2526         cand_ids = calloc(1, sizeof(*cand_ids));
2527         if (!cand_ids)
2528                 return ERR_PTR(-ENOMEM);
2529 
2530         n = btf__get_nr_types(targ_btf);
2531         for (i = 1; i <= n; i++) {
2532                 t = btf__type_by_id(targ_btf, i);
2533                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2534                 if (str_is_empty(targ_name))
2535                         continue;
2536 
2537                 targ_essent_len = bpf_core_essential_name_len(targ_name);
2538                 if (targ_essent_len != local_essent_len)
2539                         continue;
2540 
2541                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2542                         pr_debug("[%d] %s: found candidate [%d] %s\n",
2543                                  local_type_id, local_name, i, targ_name);
2544                         new_ids = reallocarray(cand_ids->data,
2545                                                cand_ids->len + 1,
2546                                                sizeof(*cand_ids->data));
2547                         if (!new_ids) {
2548                                 err = -ENOMEM;
2549                                 goto err_out;
2550                         }
2551                         cand_ids->data = new_ids;
2552                         cand_ids->data[cand_ids->len++] = i;
2553                 }
2554         }
2555         return cand_ids;
2556 err_out:
2557         bpf_core_free_cands(cand_ids);
2558         return ERR_PTR(err);
2559 }
2560 
2561 /* Check two types for compatibility, skipping const/volatile/restrict and
2562  * typedefs, to ensure we are relocating offset to the compatible entities:
2563  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
2564  *   - any two FWDs are compatible;
2565  *   - any two PTRs are always compatible;
2566  *   - for ENUMs, check sizes, names are ignored;
2567  *   - for INT, size and bitness should match, signedness is ignored;
2568  *   - for ARRAY, dimensionality is ignored, element types are checked for
2569  *     compatibility recursively;
2570  *   - everything else shouldn't be ever a target of relocation.
2571  * These rules are not set in stone and probably will be adjusted as we get
2572  * more experience with using BPF CO-RE relocations.
2573  */
2574 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2575                                       __u32 local_id,
2576                                       const struct btf *targ_btf,
2577                                       __u32 targ_id)
2578 {
2579         const struct btf_type *local_type, *targ_type;
2580 
2581 recur:
2582         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2583         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2584         if (!local_type || !targ_type)
2585                 return -EINVAL;
2586 
2587         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2588                 return 1;
2589         if (btf_kind(local_type) != btf_kind(targ_type))
2590                 return 0;
2591 
2592         switch (btf_kind(local_type)) {
2593         case BTF_KIND_FWD:
2594         case BTF_KIND_PTR:
2595                 return 1;
2596         case BTF_KIND_ENUM:
2597                 return local_type->size == targ_type->size;
2598         case BTF_KIND_INT:
2599                 return btf_int_offset(local_type) == 0 &&
2600                        btf_int_offset(targ_type) == 0 &&
2601                        local_type->size == targ_type->size &&
2602                        btf_int_bits(local_type) == btf_int_bits(targ_type);
2603         case BTF_KIND_ARRAY:
2604                 local_id = btf_array(local_type)->type;
2605                 targ_id = btf_array(targ_type)->type;
2606                 goto recur;
2607         default:
2608                 pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
2609                            btf_kind(local_type), local_id, targ_id);
2610                 return 0;
2611         }
2612 }
2613 
2614 /*
2615  * Given single high-level named field accessor in local type, find
2616  * corresponding high-level accessor for a target type. Along the way,
2617  * maintain low-level spec for target as well. Also keep updating target
2618  * offset.
2619  *
2620  * Searching is performed through recursive exhaustive enumeration of all
2621  * fields of a struct/union. If there are any anonymous (embedded)
2622  * structs/unions, they are recursively searched as well. If field with
2623  * desired name is found, check compatibility between local and target types,
2624  * before returning result.
2625  *
2626  * 1 is returned, if field is found.
2627  * 0 is returned if no compatible field is found.
2628  * <0 is returned on error.
2629  */
2630 static int bpf_core_match_member(const struct btf *local_btf,
2631                                  const struct bpf_core_accessor *local_acc,
2632                                  const struct btf *targ_btf,
2633                                  __u32 targ_id,
2634                                  struct bpf_core_spec *spec,
2635                                  __u32 *next_targ_id)
2636 {
2637         const struct btf_type *local_type, *targ_type;
2638         const struct btf_member *local_member, *m;
2639         const char *local_name, *targ_name;
2640         __u32 local_id;
2641         int i, n, found;
2642 
2643         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2644         if (!targ_type)
2645                 return -EINVAL;
2646         if (!btf_is_composite(targ_type))
2647                 return 0;
2648 
2649         local_id = local_acc->type_id;
2650         local_type = btf__type_by_id(local_btf, local_id);
2651         local_member = btf_members(local_type) + local_acc->idx;
2652         local_name = btf__name_by_offset(local_btf, local_member->name_off);
2653 
2654         n = btf_vlen(targ_type);
2655         m = btf_members(targ_type);
2656         for (i = 0; i < n; i++, m++) {
2657                 __u32 offset;
2658 
2659                 /* bitfield relocations not supported */
2660                 if (btf_member_bitfield_size(targ_type, i))
2661                         continue;
2662                 offset = btf_member_bit_offset(targ_type, i);
2663                 if (offset % 8)
2664                         continue;
2665 
2666                 /* too deep struct/union/array nesting */
2667                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2668                         return -E2BIG;
2669 
2670                 /* speculate this member will be the good one */
2671                 spec->offset += offset / 8;
2672                 spec->raw_spec[spec->raw_len++] = i;
2673 
2674                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2675                 if (str_is_empty(targ_name)) {
2676                         /* embedded struct/union, we need to go deeper */
2677                         found = bpf_core_match_member(local_btf, local_acc,
2678                                                       targ_btf, m->type,
2679                                                       spec, next_targ_id);
2680                         if (found) /* either found or error */
2681                                 return found;
2682                 } else if (strcmp(local_name, targ_name) == 0) {
2683                         /* matching named field */
2684                         struct bpf_core_accessor *targ_acc;
2685 
2686                         targ_acc = &spec->spec[spec->len++];
2687                         targ_acc->type_id = targ_id;
2688                         targ_acc->idx = i;
2689                         targ_acc->name = targ_name;
2690 
2691                         *next_targ_id = m->type;
2692                         found = bpf_core_fields_are_compat(local_btf,
2693                                                            local_member->type,
2694                                                            targ_btf, m->type);
2695                         if (!found)
2696                                 spec->len--; /* pop accessor */
2697                         return found;
2698                 }
2699                 /* member turned out not to be what we looked for */
2700                 spec->offset -= offset / 8;
2701                 spec->raw_len--;
2702         }
2703 
2704         return 0;
2705 }
2706 
2707 /*
2708  * Try to match local spec to a target type and, if successful, produce full
2709  * target spec (high-level, low-level + offset).
2710  */
2711 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2712                                const struct btf *targ_btf, __u32 targ_id,
2713                                struct bpf_core_spec *targ_spec)
2714 {
2715         const struct btf_type *targ_type;
2716         const struct bpf_core_accessor *local_acc;
2717         struct bpf_core_accessor *targ_acc;
2718         int i, sz, matched;
2719 
2720         memset(targ_spec, 0, sizeof(*targ_spec));
2721         targ_spec->btf = targ_btf;
2722 
2723         local_acc = &local_spec->spec[0];
2724         targ_acc = &targ_spec->spec[0];
2725 
2726         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2727                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2728                                                    &targ_id);
2729                 if (!targ_type)
2730                         return -EINVAL;
2731 
2732                 if (local_acc->name) {
2733                         matched = bpf_core_match_member(local_spec->btf,
2734                                                         local_acc,
2735                                                         targ_btf, targ_id,
2736                                                         targ_spec, &targ_id);
2737                         if (matched <= 0)
2738                                 return matched;
2739                 } else {
2740                         /* for i=0, targ_id is already treated as array element
2741                          * type (because it's the original struct), for others
2742                          * we should find array element type first
2743                          */
2744                         if (i > 0) {
2745                                 const struct btf_array *a;
2746 
2747                                 if (!btf_is_array(targ_type))
2748                                         return 0;
2749 
2750                                 a = btf_array(targ_type);
2751                                 if (local_acc->idx >= a->nelems)
2752                                         return 0;
2753                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
2754                                                             &targ_id))
2755                                         return -EINVAL;
2756                         }
2757 
2758                         /* too deep struct/union/array nesting */
2759                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2760                                 return -E2BIG;
2761 
2762                         targ_acc->type_id = targ_id;
2763                         targ_acc->idx = local_acc->idx;
2764                         targ_acc->name = NULL;
2765                         targ_spec->len++;
2766                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2767                         targ_spec->raw_len++;
2768 
2769                         sz = btf__resolve_size(targ_btf, targ_id);
2770                         if (sz < 0)
2771                                 return sz;
2772                         targ_spec->offset += local_acc->idx * sz;
2773                 }
2774         }
2775 
2776         return 1;
2777 }
2778 
2779 /*
2780  * Patch relocatable BPF instruction.
2781  * Expected insn->imm value is provided for validation, as well as the new
2782  * relocated value.
2783  *
2784  * Currently three kinds of BPF instructions are supported:
2785  * 1. rX = <imm> (assignment with immediate operand);
2786  * 2. rX += <imm> (arithmetic operations with immediate operand);
2787  * 3. *(rX) = <imm> (indirect memory assignment with immediate operand).
2788  *
2789  * If actual insn->imm value is wrong, bail out.
2790  */
2791 static int bpf_core_reloc_insn(struct bpf_program *prog, int insn_off,
2792                                __u32 orig_off, __u32 new_off)
2793 {
2794         struct bpf_insn *insn;
2795         int insn_idx;
2796         __u8 class;
2797 
2798         if (insn_off % sizeof(struct bpf_insn))
2799                 return -EINVAL;
2800         insn_idx = insn_off / sizeof(struct bpf_insn);
2801 
2802         insn = &prog->insns[insn_idx];
2803         class = BPF_CLASS(insn->code);
2804 
2805         if (class == BPF_ALU || class == BPF_ALU64) {
2806                 if (BPF_SRC(insn->code) != BPF_K)
2807                         return -EINVAL;
2808                 if (insn->imm != orig_off)
2809                         return -EINVAL;
2810                 insn->imm = new_off;
2811                 pr_debug("prog '%s': patched insn #%d (ALU/ALU64) imm %d -> %d\n",
2812                          bpf_program__title(prog, false),
2813                          insn_idx, orig_off, new_off);
2814         } else {
2815                 pr_warning("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
2816                            bpf_program__title(prog, false),
2817                            insn_idx, insn->code, insn->src_reg, insn->dst_reg,
2818                            insn->off, insn->imm);
2819                 return -EINVAL;
2820         }
2821         return 0;
2822 }
2823 
2824 static struct btf *btf_load_raw(const char *path)
2825 {
2826         struct btf *btf;
2827         size_t read_cnt;
2828         struct stat st;
2829         void *data;
2830         FILE *f;
2831 
2832         if (stat(path, &st))
2833                 return ERR_PTR(-errno);
2834 
2835         data = malloc(st.st_size);
2836         if (!data)
2837                 return ERR_PTR(-ENOMEM);
2838 
2839         f = fopen(path, "rb");
2840         if (!f) {
2841                 btf = ERR_PTR(-errno);
2842                 goto cleanup;
2843         }
2844 
2845         read_cnt = fread(data, 1, st.st_size, f);
2846         fclose(f);
2847         if (read_cnt < st.st_size) {
2848                 btf = ERR_PTR(-EBADF);
2849                 goto cleanup;
2850         }
2851 
2852         btf = btf__new(data, read_cnt);
2853 
2854 cleanup:
2855         free(data);
2856         return btf;
2857 }
2858 
2859 /*
2860  * Probe few well-known locations for vmlinux kernel image and try to load BTF
2861  * data out of it to use for target BTF.
2862  */
2863 static struct btf *bpf_core_find_kernel_btf(void)
2864 {
2865         struct {
2866                 const char *path_fmt;
2867                 bool raw_btf;
2868         } locations[] = {
2869                 /* try canonical vmlinux BTF through sysfs first */
2870                 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
2871                 /* fall back to trying to find vmlinux ELF on disk otherwise */
2872                 { "/boot/vmlinux-%1$s" },
2873                 { "/lib/modules/%1$s/vmlinux-%1$s" },
2874                 { "/lib/modules/%1$s/build/vmlinux" },
2875                 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
2876                 { "/usr/lib/debug/boot/vmlinux-%1$s" },
2877                 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
2878                 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
2879         };
2880         char path[PATH_MAX + 1];
2881         struct utsname buf;
2882         struct btf *btf;
2883         int i;
2884 
2885         uname(&buf);
2886 
2887         for (i = 0; i < ARRAY_SIZE(locations); i++) {
2888                 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
2889 
2890                 if (access(path, R_OK))
2891                         continue;
2892 
2893                 if (locations[i].raw_btf)
2894                         btf = btf_load_raw(path);
2895                 else
2896                         btf = btf__parse_elf(path, NULL);
2897 
2898                 pr_debug("loading kernel BTF '%s': %ld\n",
2899                          path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
2900                 if (IS_ERR(btf))
2901                         continue;
2902 
2903                 return btf;
2904         }
2905 
2906         pr_warning("failed to find valid kernel BTF\n");
2907         return ERR_PTR(-ESRCH);
2908 }
2909 
2910 /* Output spec definition in the format:
2911  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
2912  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
2913  */
2914 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
2915 {
2916         const struct btf_type *t;
2917         const char *s;
2918         __u32 type_id;
2919         int i;
2920 
2921         type_id = spec->spec[0].type_id;
2922         t = btf__type_by_id(spec->btf, type_id);
2923         s = btf__name_by_offset(spec->btf, t->name_off);
2924         libbpf_print(level, "[%u] %s + ", type_id, s);
2925 
2926         for (i = 0; i < spec->raw_len; i++)
2927                 libbpf_print(level, "%d%s", spec->raw_spec[i],
2928                              i == spec->raw_len - 1 ? " => " : ":");
2929 
2930         libbpf_print(level, "%u @ &x", spec->offset);
2931 
2932         for (i = 0; i < spec->len; i++) {
2933                 if (spec->spec[i].name)
2934                         libbpf_print(level, ".%s", spec->spec[i].name);
2935                 else
2936                         libbpf_print(level, "[%u]", spec->spec[i].idx);
2937         }
2938 
2939 }
2940 
2941 static size_t bpf_core_hash_fn(const void *key, void *ctx)
2942 {
2943         return (size_t)key;
2944 }
2945 
2946 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
2947 {
2948         return k1 == k2;
2949 }
2950 
2951 static void *u32_as_hash_key(__u32 x)
2952 {
2953         return (void *)(uintptr_t)x;
2954 }
2955 
2956 /*
2957  * CO-RE relocate single instruction.
2958  *
2959  * The outline and important points of the algorithm:
2960  * 1. For given local type, find corresponding candidate target types.
2961  *    Candidate type is a type with the same "essential" name, ignoring
2962  *    everything after last triple underscore (___). E.g., `sample`,
2963  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
2964  *    for each other. Names with triple underscore are referred to as
2965  *    "flavors" and are useful, among other things, to allow to
2966  *    specify/support incompatible variations of the same kernel struct, which
2967  *    might differ between different kernel versions and/or build
2968  *    configurations.
2969  *
2970  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
2971  *    converter, when deduplicated BTF of a kernel still contains more than
2972  *    one different types with the same name. In that case, ___2, ___3, etc
2973  *    are appended starting from second name conflict. But start flavors are
2974  *    also useful to be defined "locally", in BPF program, to extract same
2975  *    data from incompatible changes between different kernel
2976  *    versions/configurations. For instance, to handle field renames between
2977  *    kernel versions, one can use two flavors of the struct name with the
2978  *    same common name and use conditional relocations to extract that field,
2979  *    depending on target kernel version.
2980  * 2. For each candidate type, try to match local specification to this
2981  *    candidate target type. Matching involves finding corresponding
2982  *    high-level spec accessors, meaning that all named fields should match,
2983  *    as well as all array accesses should be within the actual bounds. Also,
2984  *    types should be compatible (see bpf_core_fields_are_compat for details).
2985  * 3. It is supported and expected that there might be multiple flavors
2986  *    matching the spec. As long as all the specs resolve to the same set of
2987  *    offsets across all candidates, there is not error. If there is any
2988  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
2989  *    imprefection of BTF deduplication, which can cause slight duplication of
2990  *    the same BTF type, if some directly or indirectly referenced (by
2991  *    pointer) type gets resolved to different actual types in different
2992  *    object files. If such situation occurs, deduplicated BTF will end up
2993  *    with two (or more) structurally identical types, which differ only in
2994  *    types they refer to through pointer. This should be OK in most cases and
2995  *    is not an error.
2996  * 4. Candidate types search is performed by linearly scanning through all
2997  *    types in target BTF. It is anticipated that this is overall more
2998  *    efficient memory-wise and not significantly worse (if not better)
2999  *    CPU-wise compared to prebuilding a map from all local type names to
3000  *    a list of candidate type names. It's also sped up by caching resolved
3001  *    list of matching candidates per each local "root" type ID, that has at
3002  *    least one bpf_offset_reloc associated with it. This list is shared
3003  *    between multiple relocations for the same type ID and is updated as some
3004  *    of the candidates are pruned due to structural incompatibility.
3005  */
3006 static int bpf_core_reloc_offset(struct bpf_program *prog,
3007                                  const struct bpf_offset_reloc *relo,
3008                                  int relo_idx,
3009                                  const struct btf *local_btf,
3010                                  const struct btf *targ_btf,
3011                                  struct hashmap *cand_cache)
3012 {
3013         const char *prog_name = bpf_program__title(prog, false);
3014         struct bpf_core_spec local_spec, cand_spec, targ_spec;
3015         const void *type_key = u32_as_hash_key(relo->type_id);
3016         const struct btf_type *local_type, *cand_type;
3017         const char *local_name, *cand_name;
3018         struct ids_vec *cand_ids;
3019         __u32 local_id, cand_id;
3020         const char *spec_str;
3021         int i, j, err;
3022 
3023         local_id = relo->type_id;
3024         local_type = btf__type_by_id(local_btf, local_id);
3025         if (!local_type)
3026                 return -EINVAL;
3027 
3028         local_name = btf__name_by_offset(local_btf, local_type->name_off);
3029         if (str_is_empty(local_name))
3030                 return -EINVAL;
3031 
3032         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3033         if (str_is_empty(spec_str))
3034                 return -EINVAL;
3035 
3036         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3037         if (err) {
3038                 pr_warning("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3039                            prog_name, relo_idx, local_id, local_name, spec_str,
3040                            err);
3041                 return -EINVAL;
3042         }
3043 
3044         pr_debug("prog '%s': relo #%d: spec is ", prog_name, relo_idx);
3045         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3046         libbpf_print(LIBBPF_DEBUG, "\n");
3047 
3048         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3049                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3050                 if (IS_ERR(cand_ids)) {
3051                         pr_warning("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3052                                    prog_name, relo_idx, local_id, local_name,
3053                                    PTR_ERR(cand_ids));
3054                         return PTR_ERR(cand_ids);
3055                 }
3056                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3057                 if (err) {
3058                         bpf_core_free_cands(cand_ids);
3059                         return err;
3060                 }
3061         }
3062 
3063         for (i = 0, j = 0; i < cand_ids->len; i++) {
3064                 cand_id = cand_ids->data[i];
3065                 cand_type = btf__type_by_id(targ_btf, cand_id);
3066                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3067 
3068                 err = bpf_core_spec_match(&local_spec, targ_btf,
3069                                           cand_id, &cand_spec);
3070                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3071                          prog_name, relo_idx, i, cand_name);
3072                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3073                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3074                 if (err < 0) {
3075                         pr_warning("prog '%s': relo #%d: matching error: %d\n",
3076                                    prog_name, relo_idx, err);
3077                         return err;
3078                 }
3079                 if (err == 0)
3080                         continue;
3081 
3082                 if (j == 0) {
3083                         targ_spec = cand_spec;
3084                 } else if (cand_spec.offset != targ_spec.offset) {
3085                         /* if there are many candidates, they should all
3086                          * resolve to the same offset
3087                          */
3088                         pr_warning("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3089                                    prog_name, relo_idx, cand_spec.offset,
3090                                    targ_spec.offset);
3091                         return -EINVAL;
3092                 }
3093 
3094                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3095         }
3096 
3097         cand_ids->len = j;
3098         if (cand_ids->len == 0) {
3099                 pr_warning("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3100                            prog_name, relo_idx, local_id, local_name, spec_str);
3101                 return -ESRCH;
3102         }
3103 
3104         err = bpf_core_reloc_insn(prog, relo->insn_off,
3105                                   local_spec.offset, targ_spec.offset);
3106         if (err) {
3107                 pr_warning("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3108                            prog_name, relo_idx, relo->insn_off, err);
3109                 return -EINVAL;
3110         }
3111 
3112         return 0;
3113 }
3114 
3115 static int
3116 bpf_core_reloc_offsets(struct bpf_object *obj, const char *targ_btf_path)
3117 {
3118         const struct btf_ext_info_sec *sec;
3119         const struct bpf_offset_reloc *rec;
3120         const struct btf_ext_info *seg;
3121         struct hashmap_entry *entry;
3122         struct hashmap *cand_cache = NULL;
3123         struct bpf_program *prog;
3124         struct btf *targ_btf;
3125         const char *sec_name;
3126         int i, err = 0;
3127 
3128         if (targ_btf_path)
3129                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3130         else
3131                 targ_btf = bpf_core_find_kernel_btf();
3132         if (IS_ERR(targ_btf)) {
3133                 pr_warning("failed to get target BTF: %ld\n",
3134                            PTR_ERR(targ_btf));
3135                 return PTR_ERR(targ_btf);
3136         }
3137 
3138         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3139         if (IS_ERR(cand_cache)) {
3140                 err = PTR_ERR(cand_cache);
3141                 goto out;
3142         }
3143 
3144         seg = &obj->btf_ext->offset_reloc_info;
3145         for_each_btf_ext_sec(seg, sec) {
3146                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3147                 if (str_is_empty(sec_name)) {
3148                         err = -EINVAL;
3149                         goto out;
3150                 }
3151                 prog = bpf_object__find_program_by_title(obj, sec_name);
3152                 if (!prog) {
3153                         pr_warning("failed to find program '%s' for CO-RE offset relocation\n",
3154                                    sec_name);
3155                         err = -EINVAL;
3156                         goto out;
3157                 }
3158 
3159                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3160                          sec_name, sec->num_info);
3161 
3162                 for_each_btf_ext_rec(seg, sec, i, rec) {
3163                         err = bpf_core_reloc_offset(prog, rec, i, obj->btf,
3164                                                     targ_btf, cand_cache);
3165                         if (err) {
3166                                 pr_warning("prog '%s': relo #%d: failed to relocate: %d\n",
3167                                            sec_name, i, err);
3168                                 goto out;
3169                         }
3170                 }
3171         }
3172 
3173 out:
3174         btf__free(targ_btf);
3175         if (!IS_ERR_OR_NULL(cand_cache)) {
3176                 hashmap__for_each_entry(cand_cache, entry, i) {
3177                         bpf_core_free_cands(entry->value);
3178                 }
3179                 hashmap__free(cand_cache);
3180         }
3181         return err;
3182 }
3183 
3184 static int
3185 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3186 {
3187         int err = 0;
3188 
3189         if (obj->btf_ext->offset_reloc_info.len)
3190                 err = bpf_core_reloc_offsets(obj, targ_btf_path);
3191 
3192         return err;
3193 }
3194 
3195 static int
3196 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3197                         struct reloc_desc *relo)
3198 {
3199         struct bpf_insn *insn, *new_insn;
3200         struct bpf_program *text;
3201         size_t new_cnt;
3202         int err;
3203 
3204         if (relo->type != RELO_CALL)
3205                 return -LIBBPF_ERRNO__RELOC;
3206 
3207         if (prog->idx == obj->efile.text_shndx) {
3208                 pr_warning("relo in .text insn %d into off %d\n",
3209                            relo->insn_idx, relo->text_off);
3210                 return -LIBBPF_ERRNO__RELOC;
3211         }
3212 
3213         if (prog->main_prog_cnt == 0) {
3214                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3215                 if (!text) {
3216                         pr_warning("no .text section found yet relo into text exist\n");
3217                         return -LIBBPF_ERRNO__RELOC;
3218                 }
3219                 new_cnt = prog->insns_cnt + text->insns_cnt;
3220                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3221                 if (!new_insn) {
3222                         pr_warning("oom in prog realloc\n");
3223                         return -ENOMEM;
3224                 }
3225                 prog->insns = new_insn;
3226 
3227                 if (obj->btf_ext) {
3228                         err = bpf_program_reloc_btf_ext(prog, obj,
3229                                                         text->section_name,
3230                                                         prog->insns_cnt);
3231                         if (err)
3232                                 return err;
3233                 }
3234 
3235                 memcpy(new_insn + prog->insns_cnt, text->insns,
3236                        text->insns_cnt * sizeof(*insn));
3237                 prog->main_prog_cnt = prog->insns_cnt;
3238                 prog->insns_cnt = new_cnt;
3239                 pr_debug("added %zd insn from %s to prog %s\n",
3240                          text->insns_cnt, text->section_name,
3241                          prog->section_name);
3242         }
3243         insn = &prog->insns[relo->insn_idx];
3244         insn->imm += prog->main_prog_cnt - relo->insn_idx;
3245         return 0;
3246 }
3247 
3248 static int
3249 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3250 {
3251         int i, err;
3252 
3253         if (!prog)
3254                 return 0;
3255 
3256         if (obj->btf_ext) {
3257                 err = bpf_program_reloc_btf_ext(prog, obj,
3258                                                 prog->section_name, 0);
3259                 if (err)
3260                         return err;
3261         }
3262 
3263         if (!prog->reloc_desc)
3264                 return 0;
3265 
3266         for (i = 0; i < prog->nr_reloc; i++) {
3267                 if (prog->reloc_desc[i].type == RELO_LD64 ||
3268                     prog->reloc_desc[i].type == RELO_DATA) {
3269                         bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
3270                         struct bpf_insn *insns = prog->insns;
3271                         int insn_idx, map_idx;
3272 
3273                         insn_idx = prog->reloc_desc[i].insn_idx;
3274                         map_idx = prog->reloc_desc[i].map_idx;
3275 
3276                         if (insn_idx + 1 >= (int)prog->insns_cnt) {
3277                                 pr_warning("relocation out of range: '%s'\n",
3278                                            prog->section_name);
3279                                 return -LIBBPF_ERRNO__RELOC;
3280                         }
3281 
3282                         if (!relo_data) {
3283                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
3284                         } else {
3285                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
3286                                 insns[insn_idx + 1].imm = insns[insn_idx].imm;
3287                         }
3288                         insns[insn_idx].imm = obj->maps[map_idx].fd;
3289                 } else if (prog->reloc_desc[i].type == RELO_CALL) {
3290                         err = bpf_program__reloc_text(prog, obj,
3291                                                       &prog->reloc_desc[i]);
3292                         if (err)
3293                                 return err;
3294                 }
3295         }
3296 
3297         zfree(&prog->reloc_desc);
3298         prog->nr_reloc = 0;
3299         return 0;
3300 }
3301 
3302 static int
3303 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3304 {
3305         struct bpf_program *prog;
3306         size_t i;
3307         int err;
3308 
3309         if (obj->btf_ext) {
3310                 err = bpf_object__relocate_core(obj, targ_btf_path);
3311                 if (err) {
3312                         pr_warning("failed to perform CO-RE relocations: %d\n",
3313                                    err);
3314                         return err;
3315                 }
3316         }
3317         for (i = 0; i < obj->nr_programs; i++) {
3318                 prog = &obj->programs[i];
3319 
3320                 err = bpf_program__relocate(prog, obj);
3321                 if (err) {
3322                         pr_warning("failed to relocate '%s'\n",
3323                                    prog->section_name);
3324                         return err;
3325                 }
3326         }
3327         return 0;
3328 }
3329 
3330 static int bpf_object__collect_reloc(struct bpf_object *obj)
3331 {
3332         int i, err;
3333 
3334         if (!obj_elf_valid(obj)) {
3335                 pr_warning("Internal error: elf object is closed\n");
3336                 return -LIBBPF_ERRNO__INTERNAL;
3337         }
3338 
3339         for (i = 0; i < obj->efile.nr_reloc; i++) {
3340                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
3341                 Elf_Data *data = obj->efile.reloc[i].data;
3342                 int idx = shdr->sh_info;
3343                 struct bpf_program *prog;
3344 
3345                 if (shdr->sh_type != SHT_REL) {
3346                         pr_warning("internal error at %d\n", __LINE__);
3347                         return -LIBBPF_ERRNO__INTERNAL;
3348                 }
3349 
3350                 prog = bpf_object__find_prog_by_idx(obj, idx);
3351                 if (!prog) {
3352                         pr_warning("relocation failed: no section(%d)\n", idx);
3353                         return -LIBBPF_ERRNO__RELOC;
3354                 }
3355 
3356                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3357                 if (err)
3358                         return err;
3359         }
3360         return 0;
3361 }
3362 
3363 static int
3364 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3365              char *license, __u32 kern_version, int *pfd)
3366 {
3367         struct bpf_load_program_attr load_attr;
3368         char *cp, errmsg[STRERR_BUFSIZE];
3369         int log_buf_size = BPF_LOG_BUF_SIZE;
3370         char *log_buf;
3371         int btf_fd, ret;
3372 
3373         if (!insns || !insns_cnt)
3374                 return -EINVAL;
3375 
3376         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3377         load_attr.prog_type = prog->type;
3378         load_attr.expected_attach_type = prog->expected_attach_type;
3379         if (prog->caps->name)
3380                 load_attr.name = prog->name;
3381         load_attr.insns = insns;
3382         load_attr.insns_cnt = insns_cnt;
3383         load_attr.license = license;
3384         load_attr.kern_version = kern_version;
3385         load_attr.prog_ifindex = prog->prog_ifindex;
3386         /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
3387         if (prog->obj->btf_ext)
3388                 btf_fd = bpf_object__btf_fd(prog->obj);
3389         else
3390                 btf_fd = -1;
3391         load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
3392         load_attr.func_info = prog->func_info;
3393         load_attr.func_info_rec_size = prog->func_info_rec_size;
3394         load_attr.func_info_cnt = prog->func_info_cnt;
3395         load_attr.line_info = prog->line_info;
3396         load_attr.line_info_rec_size = prog->line_info_rec_size;
3397         load_attr.line_info_cnt = prog->line_info_cnt;
3398         load_attr.log_level = prog->log_level;
3399         load_attr.prog_flags = prog->prog_flags;
3400 
3401 retry_load:
3402         log_buf = malloc(log_buf_size);
3403         if (!log_buf)
3404                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
3405 
3406         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3407 
3408         if (ret >= 0) {
3409                 if (load_attr.log_level)
3410                         pr_debug("verifier log:\n%s", log_buf);
3411                 *pfd = ret;
3412                 ret = 0;
3413                 goto out;
3414         }
3415 
3416         if (errno == ENOSPC) {
3417                 log_buf_size <<= 1;
3418                 free(log_buf);
3419                 goto retry_load;
3420         }
3421         ret = -LIBBPF_ERRNO__LOAD;
3422         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3423         pr_warning("load bpf program failed: %s\n", cp);
3424 
3425         if (log_buf && log_buf[0] != '\0') {
3426                 ret = -LIBBPF_ERRNO__VERIFY;
3427                 pr_warning("-- BEGIN DUMP LOG ---\n");
3428                 pr_warning("\n%s\n", log_buf);
3429                 pr_warning("-- END LOG --\n");
3430         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3431                 pr_warning("Program too large (%zu insns), at most %d insns\n",
3432                            load_attr.insns_cnt, BPF_MAXINSNS);
3433                 ret = -LIBBPF_ERRNO__PROG2BIG;
3434         } else {
3435                 /* Wrong program type? */
3436                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3437                         int fd;
3438 
3439                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3440                         load_attr.expected_attach_type = 0;
3441                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3442                         if (fd >= 0) {
3443                                 close(fd);
3444                                 ret = -LIBBPF_ERRNO__PROGTYPE;
3445                                 goto out;
3446                         }
3447                 }
3448 
3449                 if (log_buf)
3450                         ret = -LIBBPF_ERRNO__KVER;
3451         }
3452 
3453 out:
3454         free(log_buf);
3455         return ret;
3456 }
3457 
3458 int
3459 bpf_program__load(struct bpf_program *prog,
3460                   char *license, __u32 kern_version)
3461 {
3462         int err = 0, fd, i;
3463 
3464         if (prog->instances.nr < 0 || !prog->instances.fds) {
3465                 if (prog->preprocessor) {
3466                         pr_warning("Internal error: can't load program '%s'\n",
3467                                    prog->section_name);
3468                         return -LIBBPF_ERRNO__INTERNAL;
3469                 }
3470 
3471                 prog->instances.fds = malloc(sizeof(int));
3472                 if (!prog->instances.fds) {
3473                         pr_warning("Not enough memory for BPF fds\n");
3474                         return -ENOMEM;
3475                 }
3476                 prog->instances.nr = 1;
3477                 prog->instances.fds[0] = -1;
3478         }
3479 
3480         if (!prog->preprocessor) {
3481                 if (prog->instances.nr != 1) {
3482                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
3483                                    prog->section_name, prog->instances.nr);
3484                 }
3485                 err = load_program(prog, prog->insns, prog->insns_cnt,
3486                                    license, kern_version, &fd);
3487                 if (!err)
3488                         prog->instances.fds[0] = fd;
3489                 goto out;
3490         }
3491 
3492         for (i = 0; i < prog->instances.nr; i++) {
3493                 struct bpf_prog_prep_result result;
3494                 bpf_program_prep_t preprocessor = prog->preprocessor;
3495 
3496                 memset(&result, 0, sizeof(result));
3497                 err = preprocessor(prog, i, prog->insns,
3498                                    prog->insns_cnt, &result);
3499                 if (err) {
3500                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
3501                                    i, prog->section_name);
3502                         goto out;
3503                 }
3504 
3505                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3506                         pr_debug("Skip loading the %dth instance of program '%s'\n",
3507                                  i, prog->section_name);
3508                         prog->instances.fds[i] = -1;
3509                         if (result.pfd)
3510                                 *result.pfd = -1;
3511                         continue;
3512                 }
3513 
3514                 err = load_program(prog, result.new_insn_ptr,
3515                                    result.new_insn_cnt,
3516                                    license, kern_version, &fd);
3517 
3518                 if (err) {
3519                         pr_warning("Loading the %dth instance of program '%s' failed\n",
3520                                         i, prog->section_name);
3521                         goto out;
3522                 }
3523 
3524                 if (result.pfd)
3525                         *result.pfd = fd;
3526                 prog->instances.fds[i] = fd;
3527         }
3528 out:
3529         if (err)
3530                 pr_warning("failed to load program '%s'\n",
3531                            prog->section_name);
3532         zfree(&prog->insns);
3533         prog->insns_cnt = 0;
3534         return err;
3535 }
3536 
3537 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3538                                              const struct bpf_object *obj)
3539 {
3540         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3541 }
3542 
3543 static int
3544 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3545 {
3546         size_t i;
3547         int err;
3548 
3549         for (i = 0; i < obj->nr_programs; i++) {
3550                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3551                         continue;
3552                 obj->programs[i].log_level |= log_level;
3553                 err = bpf_program__load(&obj->programs[i],
3554                                         obj->license,
3555                                         obj->kern_version);
3556                 if (err)
3557                         return err;
3558         }
3559         return 0;
3560 }
3561 
3562 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
3563 {
3564         switch (type) {
3565         case BPF_PROG_TYPE_SOCKET_FILTER:
3566         case BPF_PROG_TYPE_SCHED_CLS:
3567         case BPF_PROG_TYPE_SCHED_ACT:
3568         case BPF_PROG_TYPE_XDP:
3569         case BPF_PROG_TYPE_CGROUP_SKB:
3570         case BPF_PROG_TYPE_CGROUP_SOCK:
3571         case BPF_PROG_TYPE_LWT_IN:
3572         case BPF_PROG_TYPE_LWT_OUT:
3573         case BPF_PROG_TYPE_LWT_XMIT:
3574         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
3575         case BPF_PROG_TYPE_SOCK_OPS:
3576         case BPF_PROG_TYPE_SK_SKB:
3577         case BPF_PROG_TYPE_CGROUP_DEVICE:
3578         case BPF_PROG_TYPE_SK_MSG:
3579         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3580         case BPF_PROG_TYPE_LIRC_MODE2:
3581         case BPF_PROG_TYPE_SK_REUSEPORT:
3582         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3583         case BPF_PROG_TYPE_UNSPEC:
3584         case BPF_PROG_TYPE_TRACEPOINT:
3585         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3586         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3587         case BPF_PROG_TYPE_PERF_EVENT:
3588         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3589         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3590                 return false;
3591         case BPF_PROG_TYPE_KPROBE:
3592         default:
3593                 return true;
3594         }
3595 }
3596 
3597 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
3598 {
3599         if (needs_kver && obj->kern_version == 0) {
3600                 pr_warning("%s doesn't provide kernel version\n",
3601                            obj->path);
3602                 return -LIBBPF_ERRNO__KVERSION;
3603         }
3604         return 0;
3605 }
3606 
3607 static struct bpf_object *
3608 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
3609                    bool needs_kver, int flags)
3610 {
3611         struct bpf_object *obj;
3612         int err;
3613 
3614         if (elf_version(EV_CURRENT) == EV_NONE) {
3615                 pr_warning("failed to init libelf for %s\n", path);
3616                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3617         }
3618 
3619         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
3620         if (IS_ERR(obj))
3621                 return obj;
3622 
3623         CHECK_ERR(bpf_object__elf_init(obj), err, out);
3624         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3625         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3626         CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
3627         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3628         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
3629 
3630         bpf_object__elf_finish(obj);
3631         return obj;
3632 out:
3633         bpf_object__close(obj);
3634         return ERR_PTR(err);
3635 }
3636 
3637 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
3638                                             int flags)
3639 {
3640         /* param validation */
3641         if (!attr->file)
3642                 return NULL;
3643 
3644         pr_debug("loading %s\n", attr->file);
3645 
3646         return __bpf_object__open(attr->file, NULL, 0,
3647                                   bpf_prog_type__needs_kver(attr->prog_type),
3648                                   flags);
3649 }
3650 
3651 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
3652 {
3653         return __bpf_object__open_xattr(attr, 0);
3654 }
3655 
3656 struct bpf_object *bpf_object__open(const char *path)
3657 {
3658         struct bpf_object_open_attr attr = {
3659                 .file           = path,
3660                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
3661         };
3662 
3663         return bpf_object__open_xattr(&attr);
3664 }
3665 
3666 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
3667                                            size_t obj_buf_sz,
3668                                            const char *name)
3669 {
3670         char tmp_name[64];
3671 
3672         /* param validation */
3673         if (!obj_buf || obj_buf_sz <= 0)
3674                 return NULL;
3675 
3676         if (!name) {
3677                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3678                          (unsigned long)obj_buf,
3679                          (unsigned long)obj_buf_sz);
3680                 name = tmp_name;
3681         }
3682         pr_debug("loading object '%s' from buffer\n", name);
3683 
3684         return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
3685 }
3686 
3687 int bpf_object__unload(struct bpf_object *obj)
3688 {
3689         size_t i;
3690 
3691         if (!obj)
3692                 return -EINVAL;
3693 
3694         for (i = 0; i < obj->nr_maps; i++)
3695                 zclose(obj->maps[i].fd);
3696 
3697         for (i = 0; i < obj->nr_programs; i++)
3698                 bpf_program__unload(&obj->programs[i]);
3699 
3700         return 0;
3701 }
3702 
3703 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
3704 {
3705         struct bpf_object *obj;
3706         int err;
3707 
3708         if (!attr)
3709                 return -EINVAL;
3710         obj = attr->obj;
3711         if (!obj)
3712                 return -EINVAL;
3713 
3714         if (obj->loaded) {
3715                 pr_warning("object should not be loaded twice\n");
3716                 return -EINVAL;
3717         }
3718 
3719         obj->loaded = true;
3720 
3721         CHECK_ERR(bpf_object__create_maps(obj), err, out);
3722         CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
3723         CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
3724 
3725         return 0;
3726 out:
3727         bpf_object__unload(obj);
3728         pr_warning("failed to load object '%s'\n", obj->path);
3729         return err;
3730 }
3731 
3732 int bpf_object__load(struct bpf_object *obj)
3733 {
3734         struct bpf_object_load_attr attr = {
3735                 .obj = obj,
3736         };
3737 
3738         return bpf_object__load_xattr(&attr);
3739 }
3740 
3741 static int check_path(const char *path)
3742 {
3743         char *cp, errmsg[STRERR_BUFSIZE];
3744         struct statfs st_fs;
3745         char *dname, *dir;
3746         int err = 0;
3747 
3748         if (path == NULL)
3749                 return -EINVAL;
3750 
3751         dname = strdup(path);
3752         if (dname == NULL)
3753                 return -ENOMEM;
3754 
3755         dir = dirname(dname);
3756         if (statfs(dir, &st_fs)) {
3757                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3758                 pr_warning("failed to statfs %s: %s\n", dir, cp);
3759                 err = -errno;
3760         }
3761         free(dname);
3762 
3763         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
3764                 pr_warning("specified path %s is not on BPF FS\n", path);
3765                 err = -EINVAL;
3766         }
3767 
3768         return err;
3769 }
3770 
3771 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
3772                               int instance)
3773 {
3774         char *cp, errmsg[STRERR_BUFSIZE];
3775         int err;
3776 
3777         err = check_path(path);
3778         if (err)
3779                 return err;
3780 
3781         if (prog == NULL) {
3782                 pr_warning("invalid program pointer\n");
3783                 return -EINVAL;
3784         }
3785 
3786         if (instance < 0 || instance >= prog->instances.nr) {
3787                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3788                            instance, prog->section_name, prog->instances.nr);
3789                 return -EINVAL;
3790         }
3791 
3792         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
3793                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3794                 pr_warning("failed to pin program: %s\n", cp);
3795                 return -errno;
3796         }
3797         pr_debug("pinned program '%s'\n", path);
3798 
3799         return 0;
3800 }
3801 
3802 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
3803                                 int instance)
3804 {
3805         int err;
3806 
3807         err = check_path(path);
3808         if (err)
3809                 return err;
3810 
3811         if (prog == NULL) {
3812                 pr_warning("invalid program pointer\n");
3813                 return -EINVAL;
3814         }
3815 
3816         if (instance < 0 || instance >= prog->instances.nr) {
3817                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3818                            instance, prog->section_name, prog->instances.nr);
3819                 return -EINVAL;
3820         }
3821 
3822         err = unlink(path);
3823         if (err != 0)
3824                 return -errno;
3825         pr_debug("unpinned program '%s'\n", path);
3826 
3827         return 0;
3828 }
3829 
3830 static int make_dir(const char *path)
3831 {
3832         char *cp, errmsg[STRERR_BUFSIZE];
3833         int err = 0;
3834 
3835         if (mkdir(path, 0700) && errno != EEXIST)
3836                 err = -errno;
3837 
3838         if (err) {
3839                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3840                 pr_warning("failed to mkdir %s: %s\n", path, cp);
3841         }
3842         return err;
3843 }
3844 
3845 int bpf_program__pin(struct bpf_program *prog, const char *path)
3846 {
3847         int i, err;
3848 
3849         err = check_path(path);
3850         if (err)
3851                 return err;
3852 
3853         if (prog == NULL) {
3854                 pr_warning("invalid program pointer\n");
3855                 return -EINVAL;
3856         }
3857 
3858         if (prog->instances.nr <= 0) {
3859                 pr_warning("no instances of prog %s to pin\n",
3860                            prog->section_name);
3861                 return -EINVAL;
3862         }
3863 
3864         if (prog->instances.nr == 1) {
3865                 /* don't create subdirs when pinning single instance */
3866                 return bpf_program__pin_instance(prog, path, 0);
3867         }
3868 
3869         err = make_dir(path);
3870         if (err)
3871                 return err;
3872 
3873         for (i = 0; i < prog->instances.nr; i++) {
3874                 char buf[PATH_MAX];
3875                 int len;
3876 
3877                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3878                 if (len < 0) {
3879                         err = -EINVAL;
3880                         goto err_unpin;
3881                 } else if (len >= PATH_MAX) {
3882                         err = -ENAMETOOLONG;
3883                         goto err_unpin;
3884                 }
3885 
3886                 err = bpf_program__pin_instance(prog, buf, i);
3887                 if (err)
3888                         goto err_unpin;
3889         }
3890 
3891         return 0;
3892 
3893 err_unpin:
3894         for (i = i - 1; i >= 0; i--) {
3895                 char buf[PATH_MAX];
3896                 int len;
3897 
3898                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3899                 if (len < 0)
3900                         continue;
3901                 else if (len >= PATH_MAX)
3902                         continue;
3903 
3904                 bpf_program__unpin_instance(prog, buf, i);
3905         }
3906 
3907         rmdir(path);
3908 
3909         return err;
3910 }
3911 
3912 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3913 {
3914         int i, err;
3915 
3916         err = check_path(path);
3917         if (err)
3918                 return err;
3919 
3920         if (prog == NULL) {
3921                 pr_warning("invalid program pointer\n");
3922                 return -EINVAL;
3923         }
3924 
3925         if (prog->instances.nr <= 0) {
3926                 pr_warning("no instances of prog %s to pin\n",
3927                            prog->section_name);
3928                 return -EINVAL;
3929         }
3930 
3931         if (prog->instances.nr == 1) {
3932                 /* don't create subdirs when pinning single instance */
3933                 return bpf_program__unpin_instance(prog, path, 0);
3934         }
3935 
3936         for (i = 0; i < prog->instances.nr; i++) {
3937                 char buf[PATH_MAX];
3938                 int len;
3939 
3940                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3941                 if (len < 0)
3942                         return -EINVAL;
3943                 else if (len >= PATH_MAX)
3944                         return -ENAMETOOLONG;
3945 
3946                 err = bpf_program__unpin_instance(prog, buf, i);
3947                 if (err)
3948                         return err;
3949         }
3950 
3951         err = rmdir(path);
3952         if (err)
3953                 return -errno;
3954 
3955         return 0;
3956 }
3957 
3958 int bpf_map__pin(struct bpf_map *map, const char *path)
3959 {
3960         char *cp, errmsg[STRERR_BUFSIZE];
3961         int err;
3962 
3963         err = check_path(path);
3964         if (err)
3965                 return err;
3966 
3967         if (map == NULL) {
3968                 pr_warning("invalid map pointer\n");
3969                 return -EINVAL;
3970         }
3971 
3972         if (bpf_obj_pin(map->fd, path)) {
3973                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3974                 pr_warning("failed to pin map: %s\n", cp);
3975                 return -errno;
3976         }
3977 
3978         pr_debug("pinned map '%s'\n", path);
3979 
3980         return 0;
3981 }
3982 
3983 int bpf_map__unpin(struct bpf_map *map, const char *path)
3984 {
3985         int err;
3986 
3987         err = check_path(path);
3988         if (err)
3989                 return err;
3990 
3991         if (map == NULL) {
3992                 pr_warning("invalid map pointer\n");
3993                 return -EINVAL;
3994         }
3995 
3996         err = unlink(path);
3997         if (err != 0)
3998                 return -errno;
3999         pr_debug("unpinned map '%s'\n", path);
4000 
4001         return 0;
4002 }
4003 
4004 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
4005 {
4006         struct bpf_map *map;
4007         int err;
4008 
4009         if (!obj)
4010                 return -ENOENT;
4011 
4012         if (!obj->loaded) {
4013                 pr_warning("object not yet loaded; load it first\n");
4014                 return -ENOENT;
4015         }
4016 
4017         err = make_dir(path);
4018         if (err)
4019                 return err;
4020 
4021         bpf_object__for_each_map(map, obj) {
4022                 char buf[PATH_MAX];
4023                 int len;
4024 
4025                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4026                                bpf_map__name(map));
4027                 if (len < 0) {
4028                         err = -EINVAL;
4029                         goto err_unpin_maps;
4030                 } else if (len >= PATH_MAX) {
4031                         err = -ENAMETOOLONG;
4032                         goto err_unpin_maps;
4033                 }
4034 
4035                 err = bpf_map__pin(map, buf);
4036                 if (err)
4037                         goto err_unpin_maps;
4038         }
4039 
4040         return 0;
4041 
4042 err_unpin_maps:
4043         while ((map = bpf_map__prev(map, obj))) {
4044                 char buf[PATH_MAX];
4045                 int len;
4046 
4047                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4048                                bpf_map__name(map));
4049                 if (len < 0)
4050                         continue;
4051                 else if (len >= PATH_MAX)
4052                         continue;
4053 
4054                 bpf_map__unpin(map, buf);
4055         }
4056 
4057         return err;
4058 }
4059 
4060 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4061 {
4062         struct bpf_map *map;
4063         int err;
4064 
4065         if (!obj)
4066                 return -ENOENT;
4067 
4068         bpf_object__for_each_map(map, obj) {
4069                 char buf[PATH_MAX];
4070                 int len;
4071 
4072                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4073                                bpf_map__name(map));
4074                 if (len < 0)
4075                         return -EINVAL;
4076                 else if (len >= PATH_MAX)
4077                         return -ENAMETOOLONG;
4078 
4079                 err = bpf_map__unpin(map, buf);
4080                 if (err)
4081                         return err;
4082         }
4083 
4084         return 0;
4085 }
4086 
4087 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4088 {
4089         struct bpf_program *prog;
4090         int err;
4091 
4092         if (!obj)
4093                 return -ENOENT;
4094 
4095         if (!obj->loaded) {
4096                 pr_warning("object not yet loaded; load it first\n");
4097                 return -ENOENT;
4098         }
4099 
4100         err = make_dir(path);
4101         if (err)
4102                 return err;
4103 
4104         bpf_object__for_each_program(prog, obj) {
4105                 char buf[PATH_MAX];
4106                 int len;
4107 
4108                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4109                                prog->pin_name);
4110                 if (len < 0) {
4111                         err = -EINVAL;
4112                         goto err_unpin_programs;
4113                 } else if (len >= PATH_MAX) {
4114                         err = -ENAMETOOLONG;
4115                         goto err_unpin_programs;
4116                 }
4117 
4118                 err = bpf_program__pin(prog, buf);
4119                 if (err)
4120                         goto err_unpin_programs;
4121         }
4122 
4123         return 0;
4124 
4125 err_unpin_programs:
4126         while ((prog = bpf_program__prev(prog, obj))) {
4127                 char buf[PATH_MAX];
4128                 int len;
4129 
4130                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4131                                prog->pin_name);
4132                 if (len < 0)
4133                         continue;
4134                 else if (len >= PATH_MAX)
4135                         continue;
4136 
4137                 bpf_program__unpin(prog, buf);
4138         }
4139 
4140         return err;
4141 }
4142 
4143 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4144 {
4145         struct bpf_program *prog;
4146         int err;
4147 
4148         if (!obj)
4149                 return -ENOENT;
4150 
4151         bpf_object__for_each_program(prog, obj) {
4152                 char buf[PATH_MAX];
4153                 int len;
4154 
4155                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4156                                prog->pin_name);
4157                 if (len < 0)
4158                         return -EINVAL;
4159                 else if (len >= PATH_MAX)
4160                         return -ENAMETOOLONG;
4161 
4162                 err = bpf_program__unpin(prog, buf);
4163                 if (err)
4164                         return err;
4165         }
4166 
4167         return 0;
4168 }
4169 
4170 int bpf_object__pin(struct bpf_object *obj, const char *path)
4171 {
4172         int err;
4173 
4174         err = bpf_object__pin_maps(obj, path);
4175         if (err)
4176                 return err;
4177 
4178         err = bpf_object__pin_programs(obj, path);
4179         if (err) {
4180                 bpf_object__unpin_maps(obj, path);
4181                 return err;
4182         }
4183 
4184         return 0;
4185 }
4186 
4187 void bpf_object__close(struct bpf_object *obj)
4188 {
4189         size_t i;
4190 
4191         if (!obj)
4192                 return;
4193 
4194         if (obj->clear_priv)
4195                 obj->clear_priv(obj, obj->priv);
4196 
4197         bpf_object__elf_finish(obj);
4198         bpf_object__unload(obj);
4199         btf__free(obj->btf);
4200         btf_ext__free(obj->btf_ext);
4201 
4202         for (i = 0; i < obj->nr_maps; i++) {
4203                 zfree(&obj->maps[i].name);
4204                 if (obj->maps[i].clear_priv)
4205                         obj->maps[i].clear_priv(&obj->maps[i],
4206                                                 obj->maps[i].priv);
4207                 obj->maps[i].priv = NULL;
4208                 obj->maps[i].clear_priv = NULL;
4209         }
4210 
4211         zfree(&obj->sections.rodata);
4212         zfree(&obj->sections.data);
4213         zfree(&obj->maps);
4214         obj->nr_maps = 0;
4215 
4216         if (obj->programs && obj->nr_programs) {
4217                 for (i = 0; i < obj->nr_programs; i++)
4218                         bpf_program__exit(&obj->programs[i]);
4219         }
4220         zfree(&obj->programs);
4221 
4222         list_del(&obj->list);
4223         free(obj);
4224 }
4225 
4226 struct bpf_object *
4227 bpf_object__next(struct bpf_object *prev)
4228 {
4229         struct bpf_object *next;
4230 
4231         if (!prev)
4232                 next = list_first_entry(&bpf_objects_list,
4233                                         struct bpf_object,
4234                                         list);
4235         else
4236                 next = list_next_entry(prev, list);
4237 
4238         /* Empty list is noticed here so don't need checking on entry. */
4239         if (&next->list == &bpf_objects_list)
4240                 return NULL;
4241 
4242         return next;
4243 }
4244 
4245 const char *bpf_object__name(const struct bpf_object *obj)
4246 {
4247         return obj ? obj->path : ERR_PTR(-EINVAL);
4248 }
4249 
4250 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4251 {
4252         return obj ? obj->kern_version : 0;
4253 }
4254 
4255 struct btf *bpf_object__btf(const struct bpf_object *obj)
4256 {
4257         return obj ? obj->btf : NULL;
4258 }
4259 
4260 int bpf_object__btf_fd(const struct bpf_object *obj)
4261 {
4262         return obj->btf ? btf__fd(obj->btf) : -1;
4263 }
4264 
4265 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4266                          bpf_object_clear_priv_t clear_priv)
4267 {
4268         if (obj->priv && obj->clear_priv)
4269                 obj->clear_priv(obj, obj->priv);
4270 
4271         obj->priv = priv;
4272         obj->clear_priv = clear_priv;
4273         return 0;
4274 }
4275 
4276 void *bpf_object__priv(const struct bpf_object *obj)
4277 {
4278         return obj ? obj->priv : ERR_PTR(-EINVAL);
4279 }
4280 
4281 static struct bpf_program *
4282 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4283                     bool forward)
4284 {
4285         size_t nr_programs = obj->nr_programs;
4286         ssize_t idx;
4287 
4288         if (!nr_programs)
4289                 return NULL;
4290 
4291         if (!p)
4292                 /* Iter from the beginning */
4293                 return forward ? &obj->programs[0] :
4294                         &obj->programs[nr_programs - 1];
4295 
4296         if (p->obj != obj) {
4297                 pr_warning("error: program handler doesn't match object\n");
4298                 return NULL;
4299         }
4300 
4301         idx = (p - obj->programs) + (forward ? 1 : -1);
4302         if (idx >= obj->nr_programs || idx < 0)
4303                 return NULL;
4304         return &obj->programs[idx];
4305 }
4306 
4307 struct bpf_program *
4308 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4309 {
4310         struct bpf_program *prog = prev;
4311 
4312         do {
4313                 prog = __bpf_program__iter(prog, obj, true);
4314         } while (prog && bpf_program__is_function_storage(prog, obj));
4315 
4316         return prog;
4317 }
4318 
4319 struct bpf_program *
4320 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4321 {
4322         struct bpf_program *prog = next;
4323 
4324         do {
4325                 prog = __bpf_program__iter(prog, obj, false);
4326         } while (prog && bpf_program__is_function_storage(prog, obj));
4327 
4328         return prog;
4329 }
4330 
4331 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4332                           bpf_program_clear_priv_t clear_priv)
4333 {
4334         if (prog->priv && prog->clear_priv)
4335                 prog->clear_priv(prog, prog->priv);
4336 
4337         prog->priv = priv;
4338         prog->clear_priv = clear_priv;
4339         return 0;
4340 }
4341 
4342 void *bpf_program__priv(const struct bpf_program *prog)
4343 {
4344         return prog ? prog->priv : ERR_PTR(-EINVAL);
4345 }
4346 
4347 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4348 {
4349         prog->prog_ifindex = ifindex;
4350 }
4351 
4352 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4353 {
4354         const char *title;
4355 
4356         title = prog->section_name;
4357         if (needs_copy) {
4358                 title = strdup(title);
4359                 if (!title) {
4360                         pr_warning("failed to strdup program title\n");
4361                         return ERR_PTR(-ENOMEM);
4362                 }
4363         }
4364 
4365         return title;
4366 }
4367 
4368 int bpf_program__fd(const struct bpf_program *prog)
4369 {
4370         return bpf_program__nth_fd(prog, 0);
4371 }
4372 
4373 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4374                           bpf_program_prep_t prep)
4375 {
4376         int *instances_fds;
4377 
4378         if (nr_instances <= 0 || !prep)
4379                 return -EINVAL;
4380 
4381         if (prog->instances.nr > 0 || prog->instances.fds) {
4382                 pr_warning("Can't set pre-processor after loading\n");
4383                 return -EINVAL;
4384         }
4385 
4386         instances_fds = malloc(sizeof(int) * nr_instances);
4387         if (!instances_fds) {
4388                 pr_warning("alloc memory failed for fds\n");
4389                 return -ENOMEM;
4390         }
4391 
4392         /* fill all fd with -1 */
4393         memset(instances_fds, -1, sizeof(int) * nr_instances);
4394 
4395         prog->instances.nr = nr_instances;
4396         prog->instances.fds = instances_fds;
4397         prog->preprocessor = prep;
4398         return 0;
4399 }
4400 
4401 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4402 {
4403         int fd;
4404 
4405         if (!prog)
4406                 return -EINVAL;
4407 
4408         if (n >= prog->instances.nr || n < 0) {
4409                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
4410                            n, prog->section_name, prog->instances.nr);
4411                 return -EINVAL;
4412         }
4413 
4414         fd = prog->instances.fds[n];
4415         if (fd < 0) {
4416                 pr_warning("%dth instance of program '%s' is invalid\n",
4417                            n, prog->section_name);
4418                 return -ENOENT;
4419         }
4420 
4421         return fd;
4422 }
4423 
4424 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4425 {
4426         prog->type = type;
4427 }
4428 
4429 static bool bpf_program__is_type(const struct bpf_program *prog,
4430                                  enum bpf_prog_type type)
4431 {
4432         return prog ? (prog->type == type) : false;
4433 }
4434 
4435 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
4436 int bpf_program__set_##NAME(struct bpf_program *prog)           \
4437 {                                                               \
4438         if (!prog)                                              \
4439                 return -EINVAL;                                 \
4440         bpf_program__set_type(prog, TYPE);                      \
4441         return 0;                                               \
4442 }                                                               \
4443                                                                 \
4444 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
4445 {                                                               \
4446         return bpf_program__is_type(prog, TYPE);                \
4447 }                                                               \
4448 
4449 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4450 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4451 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4452 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4453 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4454 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4455 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4456 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4457 
4458 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4459                                            enum bpf_attach_type type)
4460 {
4461         prog->expected_attach_type = type;
4462 }
4463 
4464 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
4465         { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
4466 
4467 /* Programs that can NOT be attached. */
4468 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
4469 
4470 /* Programs that can be attached. */
4471 #define BPF_APROG_SEC(string, ptype, atype) \
4472         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
4473 
4474 /* Programs that must specify expected attach type at load time. */
4475 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4476         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
4477 
4478 /* Programs that can be attached but attach type can't be identified by section
4479  * name. Kept for backward compatibility.
4480  */
4481 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4482 
4483 static const struct {
4484         const char *sec;
4485         size_t len;
4486         enum bpf_prog_type prog_type;
4487         enum bpf_attach_type expected_attach_type;
4488         int is_attachable;
4489         enum bpf_attach_type attach_type;
4490 } section_names[] = {
4491         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
4492         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
4493         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
4494         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
4495         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
4496         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
4497         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
4498         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
4499         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
4500         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
4501         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
4502         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
4503         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
4504         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
4505                                                 BPF_CGROUP_INET_INGRESS),
4506         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
4507                                                 BPF_CGROUP_INET_EGRESS),
4508         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
4509         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
4510                                                 BPF_CGROUP_INET_SOCK_CREATE),
4511         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
4512                                                 BPF_CGROUP_INET4_POST_BIND),
4513         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
4514                                                 BPF_CGROUP_INET6_POST_BIND),
4515         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
4516                                                 BPF_CGROUP_DEVICE),
4517         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
4518                                                 BPF_CGROUP_SOCK_OPS),
4519         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
4520                                                 BPF_SK_SKB_STREAM_PARSER),
4521         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
4522                                                 BPF_SK_SKB_STREAM_VERDICT),
4523         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
4524         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
4525                                                 BPF_SK_MSG_VERDICT),
4526         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
4527                                                 BPF_LIRC_MODE2),
4528         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
4529                                                 BPF_FLOW_DISSECTOR),
4530         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4531                                                 BPF_CGROUP_INET4_BIND),
4532         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4533                                                 BPF_CGROUP_INET6_BIND),
4534         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4535                                                 BPF_CGROUP_INET4_CONNECT),
4536         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4537                                                 BPF_CGROUP_INET6_CONNECT),
4538         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4539                                                 BPF_CGROUP_UDP4_SENDMSG),
4540         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4541                                                 BPF_CGROUP_UDP6_SENDMSG),
4542         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4543                                                 BPF_CGROUP_UDP4_RECVMSG),
4544         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4545                                                 BPF_CGROUP_UDP6_RECVMSG),
4546         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
4547                                                 BPF_CGROUP_SYSCTL),
4548         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
4549                                                 BPF_CGROUP_GETSOCKOPT),
4550         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
4551                                                 BPF_CGROUP_SETSOCKOPT),
4552 };
4553 
4554 #undef BPF_PROG_SEC_IMPL
4555 #undef BPF_PROG_SEC
4556 #undef BPF_APROG_SEC
4557 #undef BPF_EAPROG_SEC
4558 #undef BPF_APROG_COMPAT
4559 
4560 #define MAX_TYPE_NAME_SIZE 32
4561 
4562 static char *libbpf_get_type_names(bool attach_type)
4563 {
4564         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
4565         char *buf;
4566 
4567         buf = malloc(len);
4568         if (!buf)
4569                 return NULL;
4570 
4571         buf[0] = '\0';
4572         /* Forge string buf with all available names */
4573         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4574                 if (attach_type && !section_names[i].is_attachable)
4575                         continue;
4576 
4577                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
4578                         free(buf);
4579                         return NULL;
4580                 }
4581                 strcat(buf, " ");
4582                 strcat(buf, section_names[i].sec);
4583         }
4584 
4585         return buf;
4586 }
4587 
4588 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
4589                              enum bpf_attach_type *expected_attach_type)
4590 {
4591         char *type_names;
4592         int i;
4593 
4594         if (!name)
4595                 return -EINVAL;
4596 
4597         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4598                 if (strncmp(name, section_names[i].sec, section_names[i].len))
4599                         continue;
4600                 *prog_type = section_names[i].prog_type;
4601                 *expected_attach_type = section_names[i].expected_attach_type;
4602                 return 0;
4603         }
4604         pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
4605         type_names = libbpf_get_type_names(false);
4606         if (type_names != NULL) {
4607                 pr_info("supported section(type) names are:%s\n", type_names);
4608                 free(type_names);
4609         }
4610 
4611         return -EINVAL;
4612 }
4613 
4614 int libbpf_attach_type_by_name(const char *name,
4615                                enum bpf_attach_type *attach_type)
4616 {
4617         char *type_names;
4618         int i;
4619 
4620         if (!name)
4621                 return -EINVAL;
4622 
4623         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4624                 if (strncmp(name, section_names[i].sec, section_names[i].len))
4625                         continue;
4626                 if (!section_names[i].is_attachable)
4627                         return -EINVAL;
4628                 *attach_type = section_names[i].attach_type;
4629                 return 0;
4630         }
4631         pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
4632         type_names = libbpf_get_type_names(true);
4633         if (type_names != NULL) {
4634                 pr_info("attachable section(type) names are:%s\n", type_names);
4635                 free(type_names);
4636         }
4637 
4638         return -EINVAL;
4639 }
4640 
4641 static int
4642 bpf_program__identify_section(struct bpf_program *prog,
4643                               enum bpf_prog_type *prog_type,
4644                               enum bpf_attach_type *expected_attach_type)
4645 {
4646         return libbpf_prog_type_by_name(prog->section_name, prog_type,
4647                                         expected_attach_type);
4648 }
4649 
4650 int bpf_map__fd(const struct bpf_map *map)
4651 {
4652         return map ? map->fd : -EINVAL;
4653 }
4654 
4655 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
4656 {
4657         return map ? &map->def : ERR_PTR(-EINVAL);
4658 }
4659 
4660 const char *bpf_map__name(const struct bpf_map *map)
4661 {
4662         return map ? map->name : NULL;
4663 }
4664 
4665 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
4666 {
4667         return map ? map->btf_key_type_id : 0;
4668 }
4669 
4670 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
4671 {
4672         return map ? map->btf_value_type_id : 0;
4673 }
4674 
4675 int bpf_map__set_priv(struct bpf_map *map, void *priv,
4676                      bpf_map_clear_priv_t clear_priv)
4677 {
4678         if (!map)
4679                 return -EINVAL;
4680 
4681         if (map->priv) {
4682                 if (map->clear_priv)
4683                         map->clear_priv(map, map->priv);
4684         }
4685 
4686         map->priv = priv;
4687         map->clear_priv = clear_priv;
4688         return 0;
4689 }
4690 
4691 void *bpf_map__priv(const struct bpf_map *map)
4692 {
4693         return map ? map->priv : ERR_PTR(-EINVAL);
4694 }
4695 
4696 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
4697 {
4698         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
4699 }
4700 
4701 bool bpf_map__is_internal(const struct bpf_map *map)
4702 {
4703         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
4704 }
4705 
4706 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
4707 {
4708         map->map_ifindex = ifindex;
4709 }
4710 
4711 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
4712 {
4713         if (!bpf_map_type__is_map_in_map(map->def.type)) {
4714                 pr_warning("error: unsupported map type\n");
4715                 return -EINVAL;
4716         }
4717         if (map->inner_map_fd != -1) {
4718                 pr_warning("error: inner_map_fd already specified\n");
4719                 return -EINVAL;
4720         }
4721         map->inner_map_fd = fd;
4722         return 0;
4723 }
4724 
4725 static struct bpf_map *
4726 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
4727 {
4728         ssize_t idx;
4729         struct bpf_map *s, *e;
4730 
4731         if (!obj || !obj->maps)
4732                 return NULL;
4733 
4734         s = obj->maps;
4735         e = obj->maps + obj->nr_maps;
4736 
4737         if ((m < s) || (m >= e)) {
4738                 pr_warning("error in %s: map handler doesn't belong to object\n",
4739                            __func__);
4740                 return NULL;
4741         }
4742 
4743         idx = (m - obj->maps) + i;
4744         if (idx >= obj->nr_maps || idx < 0)
4745                 return NULL;
4746         return &obj->maps[idx];
4747 }
4748 
4749 struct bpf_map *
4750 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
4751 {
4752         if (prev == NULL)
4753                 return obj->maps;
4754 
4755         return __bpf_map__iter(prev, obj, 1);
4756 }
4757 
4758 struct bpf_map *
4759 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
4760 {
4761         if (next == NULL) {
4762                 if (!obj->nr_maps)
4763                         return NULL;
4764                 return obj->maps + obj->nr_maps - 1;
4765         }
4766 
4767         return __bpf_map__iter(next, obj, -1);
4768 }
4769 
4770 struct bpf_map *
4771 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
4772 {
4773         struct bpf_map *pos;
4774 
4775         bpf_object__for_each_map(pos, obj) {
4776                 if (pos->name && !strcmp(pos->name, name))
4777                         return pos;
4778         }
4779         return NULL;
4780 }
4781 
4782 int
4783 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
4784 {
4785         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
4786 }
4787 
4788 struct bpf_map *
4789 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
4790 {
4791         return ERR_PTR(-ENOTSUP);
4792 }
4793 
4794 long libbpf_get_error(const void *ptr)
4795 {
4796         return PTR_ERR_OR_ZERO(ptr);
4797 }
4798 
4799 int bpf_prog_load(const char *file, enum bpf_prog_type type,
4800                   struct bpf_object **pobj, int *prog_fd)
4801 {
4802         struct bpf_prog_load_attr attr;
4803 
4804         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
4805         attr.file = file;
4806         attr.prog_type = type;
4807         attr.expected_attach_type = 0;
4808 
4809         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
4810 }
4811 
4812 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
4813                         struct bpf_object **pobj, int *prog_fd)
4814 {
4815         struct bpf_object_open_attr open_attr = {};
4816         struct bpf_program *prog, *first_prog = NULL;
4817         enum bpf_attach_type expected_attach_type;
4818         enum bpf_prog_type prog_type;
4819         struct bpf_object *obj;
4820         struct bpf_map *map;
4821         int err;
4822 
4823         if (!attr)
4824                 return -EINVAL;
4825         if (!attr->file)
4826                 return -EINVAL;
4827 
4828         open_attr.file = attr->file;
4829         open_attr.prog_type = attr->prog_type;
4830 
4831         obj = bpf_object__open_xattr(&open_attr);
4832         if (IS_ERR_OR_NULL(obj))
4833                 return -ENOENT;
4834 
4835         bpf_object__for_each_program(prog, obj) {
4836                 /*
4837                  * If type is not specified, try to guess it based on
4838                  * section name.
4839                  */
4840                 prog_type = attr->prog_type;
4841                 prog->prog_ifindex = attr->ifindex;
4842                 expected_attach_type = attr->expected_attach_type;
4843                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
4844                         err = bpf_program__identify_section(prog, &prog_type,
4845                                                             &expected_attach_type);
4846                         if (err < 0) {
4847                                 bpf_object__close(obj);
4848                                 return -EINVAL;
4849                         }
4850                 }
4851 
4852                 bpf_program__set_type(prog, prog_type);
4853                 bpf_program__set_expected_attach_type(prog,
4854                                                       expected_attach_type);
4855 
4856                 prog->log_level = attr->log_level;
4857                 prog->prog_flags = attr->prog_flags;
4858                 if (!first_prog)
4859                         first_prog = prog;
4860         }
4861 
4862         bpf_object__for_each_map(map, obj) {
4863                 if (!bpf_map__is_offload_neutral(map))
4864                         map->map_ifindex = attr->ifindex;
4865         }
4866 
4867         if (!first_prog) {
4868                 pr_warning("object file doesn't contain bpf program\n");
4869                 bpf_object__close(obj);
4870                 return -ENOENT;
4871         }
4872 
4873         err = bpf_object__load(obj);
4874         if (err) {
4875                 bpf_object__close(obj);
4876                 return -EINVAL;
4877         }
4878 
4879         *pobj = obj;
4880         *prog_fd = bpf_program__fd(first_prog);
4881         return 0;
4882 }
4883 
4884 struct bpf_link {
4885         int (*destroy)(struct bpf_link *link);
4886 };
4887 
4888 int bpf_link__destroy(struct bpf_link *link)
4889 {
4890         int err;
4891 
4892         if (!link)
4893                 return 0;
4894 
4895         err = link->destroy(link);
4896         free(link);
4897 
4898         return err;
4899 }
4900 
4901 struct bpf_link_fd {
4902         struct bpf_link link; /* has to be at the top of struct */
4903         int fd; /* hook FD */
4904 };
4905 
4906 static int bpf_link__destroy_perf_event(struct bpf_link *link)
4907 {
4908         struct bpf_link_fd *l = (void *)link;
4909         int err;
4910 
4911         err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4912         if (err)
4913                 err = -errno;
4914 
4915         close(l->fd);
4916         return err;
4917 }
4918 
4919 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4920                                                 int pfd)
4921 {
4922         char errmsg[STRERR_BUFSIZE];
4923         struct bpf_link_fd *link;
4924         int prog_fd, err;
4925 
4926         if (pfd < 0) {
4927                 pr_warning("program '%s': invalid perf event FD %d\n",
4928                            bpf_program__title(prog, false), pfd);
4929                 return ERR_PTR(-EINVAL);
4930         }
4931         prog_fd = bpf_program__fd(prog);
4932         if (prog_fd < 0) {
4933                 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
4934                            bpf_program__title(prog, false));
4935                 return ERR_PTR(-EINVAL);
4936         }
4937 
4938         link = malloc(sizeof(*link));
4939         if (!link)
4940                 return ERR_PTR(-ENOMEM);
4941         link->link.destroy = &bpf_link__destroy_perf_event;
4942         link->fd = pfd;
4943 
4944         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
4945                 err = -errno;
4946                 free(link);
4947                 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
4948                            bpf_program__title(prog, false), pfd,
4949                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4950                 return ERR_PTR(err);
4951         }
4952         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4953                 err = -errno;
4954                 free(link);
4955                 pr_warning("program '%s': failed to enable pfd %d: %s\n",
4956                            bpf_program__title(prog, false), pfd,
4957                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4958                 return ERR_PTR(err);
4959         }
4960         return (struct bpf_link *)link;
4961 }
4962 
4963 /*
4964  * this function is expected to parse integer in the range of [0, 2^31-1] from
4965  * given file using scanf format string fmt. If actual parsed value is
4966  * negative, the result might be indistinguishable from error
4967  */
4968 static int parse_uint_from_file(const char *file, const char *fmt)
4969 {
4970         char buf[STRERR_BUFSIZE];
4971         int err, ret;
4972         FILE *f;
4973 
4974         f = fopen(file, "r");
4975         if (!f) {
4976                 err = -errno;
4977                 pr_debug("failed to open '%s': %s\n", file,
4978                          libbpf_strerror_r(err, buf, sizeof(buf)));
4979                 return err;
4980         }
4981         err = fscanf(f, fmt, &ret);
4982         if (err != 1) {
4983                 err = err == EOF ? -EIO : -errno;
4984                 pr_debug("failed to parse '%s': %s\n", file,
4985                         libbpf_strerror_r(err, buf, sizeof(buf)));
4986                 fclose(f);
4987                 return err;
4988         }
4989         fclose(f);
4990         return ret;
4991 }
4992 
4993 static int determine_kprobe_perf_type(void)
4994 {
4995         const char *file = "/sys/bus/event_source/devices/kprobe/type";
4996 
4997         return parse_uint_from_file(file, "%d\n");
4998 }
4999 
5000 static int determine_uprobe_perf_type(void)
5001 {
5002         const char *file = "/sys/bus/event_source/devices/uprobe/type";
5003 
5004         return parse_uint_from_file(file, "%d\n");
5005 }
5006 
5007 static int determine_kprobe_retprobe_bit(void)
5008 {
5009         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5010 
5011         return parse_uint_from_file(file, "config:%d\n");
5012 }
5013 
5014 static int determine_uprobe_retprobe_bit(void)
5015 {
5016         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5017 
5018         return parse_uint_from_file(file, "config:%d\n");
5019 }
5020 
5021 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5022                                  uint64_t offset, int pid)
5023 {
5024         struct perf_event_attr attr = {};
5025         char errmsg[STRERR_BUFSIZE];
5026         int type, pfd, err;
5027 
5028         type = uprobe ? determine_uprobe_perf_type()
5029                       : determine_kprobe_perf_type();
5030         if (type < 0) {
5031                 pr_warning("failed to determine %s perf type: %s\n",
5032                            uprobe ? "uprobe" : "kprobe",
5033                            libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5034                 return type;
5035         }
5036         if (retprobe) {
5037                 int bit = uprobe ? determine_uprobe_retprobe_bit()
5038                                  : determine_kprobe_retprobe_bit();
5039 
5040                 if (bit < 0) {
5041                         pr_warning("failed to determine %s retprobe bit: %s\n",
5042                                    uprobe ? "uprobe" : "kprobe",
5043                                    libbpf_strerror_r(bit, errmsg,
5044                                                      sizeof(errmsg)));
5045                         return bit;
5046                 }
5047                 attr.config |= 1 << bit;
5048         }
5049         attr.size = sizeof(attr);
5050         attr.type = type;
5051         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5052         attr.config2 = offset;           /* kprobe_addr or probe_offset */
5053 
5054         /* pid filter is meaningful only for uprobes */
5055         pfd = syscall(__NR_perf_event_open, &attr,
5056                       pid < 0 ? -1 : pid /* pid */,
5057                       pid == -1 ? 0 : -1 /* cpu */,
5058                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5059         if (pfd < 0) {
5060                 err = -errno;
5061                 pr_warning("%s perf_event_open() failed: %s\n",
5062                            uprobe ? "uprobe" : "kprobe",
5063                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5064                 return err;
5065         }
5066         return pfd;
5067 }
5068 
5069 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5070                                             bool retprobe,
5071                                             const char *func_name)
5072 {
5073         char errmsg[STRERR_BUFSIZE];
5074         struct bpf_link *link;
5075         int pfd, err;
5076 
5077         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5078                                     0 /* offset */, -1 /* pid */);
5079         if (pfd < 0) {
5080                 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
5081                            bpf_program__title(prog, false),
5082                            retprobe ? "kretprobe" : "kprobe", func_name,
5083                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5084                 return ERR_PTR(pfd);
5085         }
5086         link = bpf_program__attach_perf_event(prog, pfd);
5087         if (IS_ERR(link)) {
5088                 close(pfd);
5089                 err = PTR_ERR(link);
5090                 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
5091                            bpf_program__title(prog, false),
5092                            retprobe ? "kretprobe" : "kprobe", func_name,
5093                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5094                 return link;
5095         }
5096         return link;
5097 }
5098 
5099 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5100                                             bool retprobe, pid_t pid,
5101                                             const char *binary_path,
5102                                             size_t func_offset)
5103 {
5104         char errmsg[STRERR_BUFSIZE];
5105         struct bpf_link *link;
5106         int pfd, err;
5107 
5108         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5109                                     binary_path, func_offset, pid);
5110         if (pfd < 0) {
5111                 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5112                            bpf_program__title(prog, false),
5113                            retprobe ? "uretprobe" : "uprobe",
5114                            binary_path, func_offset,
5115                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5116                 return ERR_PTR(pfd);
5117         }
5118         link = bpf_program__attach_perf_event(prog, pfd);
5119         if (IS_ERR(link)) {
5120                 close(pfd);
5121                 err = PTR_ERR(link);
5122                 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5123                            bpf_program__title(prog, false),
5124                            retprobe ? "uretprobe" : "uprobe",
5125                            binary_path, func_offset,
5126                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5127                 return link;
5128         }
5129         return link;
5130 }
5131 
5132 static int determine_tracepoint_id(const char *tp_category,
5133                                    const char *tp_name)
5134 {
5135         char file[PATH_MAX];
5136         int ret;
5137 
5138         ret = snprintf(file, sizeof(file),
5139                        "/sys/kernel/debug/tracing/events/%s/%s/id",
5140                        tp_category, tp_name);
5141         if (ret < 0)
5142                 return -errno;
5143         if (ret >= sizeof(file)) {
5144                 pr_debug("tracepoint %s/%s path is too long\n",
5145                          tp_category, tp_name);
5146                 return -E2BIG;
5147         }
5148         return parse_uint_from_file(file, "%d\n");
5149 }
5150 
5151 static int perf_event_open_tracepoint(const char *tp_category,
5152                                       const char *tp_name)
5153 {
5154         struct perf_event_attr attr = {};
5155         char errmsg[STRERR_BUFSIZE];
5156         int tp_id, pfd, err;
5157 
5158         tp_id = determine_tracepoint_id(tp_category, tp_name);
5159         if (tp_id < 0) {
5160                 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5161                            tp_category, tp_name,
5162                            libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5163                 return tp_id;
5164         }
5165 
5166         attr.type = PERF_TYPE_TRACEPOINT;
5167         attr.size = sizeof(attr);
5168         attr.config = tp_id;
5169 
5170         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5171                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5172         if (pfd < 0) {
5173                 err = -errno;
5174                 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5175                            tp_category, tp_name,
5176                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5177                 return err;
5178         }
5179         return pfd;
5180 }
5181 
5182 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5183                                                 const char *tp_category,
5184                                                 const char *tp_name)
5185 {
5186         char errmsg[STRERR_BUFSIZE];
5187         struct bpf_link *link;
5188         int pfd, err;
5189 
5190         pfd = perf_event_open_tracepoint(tp_category, tp_name);
5191         if (pfd < 0) {
5192                 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5193                            bpf_program__title(prog, false),
5194                            tp_category, tp_name,
5195                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5196                 return ERR_PTR(pfd);
5197         }
5198         link = bpf_program__attach_perf_event(prog, pfd);
5199         if (IS_ERR(link)) {
5200                 close(pfd);
5201                 err = PTR_ERR(link);
5202                 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5203                            bpf_program__title(prog, false),
5204                            tp_category, tp_name,
5205                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5206                 return link;
5207         }
5208         return link;
5209 }
5210 
5211 static int bpf_link__destroy_fd(struct bpf_link *link)
5212 {
5213         struct bpf_link_fd *l = (void *)link;
5214 
5215         return close(l->fd);
5216 }
5217 
5218 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5219                                                     const char *tp_name)
5220 {
5221         char errmsg[STRERR_BUFSIZE];
5222         struct bpf_link_fd *link;
5223         int prog_fd, pfd;
5224 
5225         prog_fd = bpf_program__fd(prog);
5226         if (prog_fd < 0) {
5227                 pr_warning("program '%s': can't attach before loaded\n",
5228                            bpf_program__title(prog, false));
5229                 return ERR_PTR(-EINVAL);
5230         }
5231 
5232         link = malloc(sizeof(*link));
5233         if (!link)
5234                 return ERR_PTR(-ENOMEM);
5235         link->link.destroy = &bpf_link__destroy_fd;
5236 
5237         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5238         if (pfd < 0) {
5239                 pfd = -errno;
5240                 free(link);
5241                 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5242                            bpf_program__title(prog, false), tp_name,
5243                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5244                 return ERR_PTR(pfd);
5245         }
5246         link->fd = pfd;
5247         return (struct bpf_link *)link;
5248 }
5249 
5250 enum bpf_perf_event_ret
5251 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5252                            void **copy_mem, size_t *copy_size,
5253                            bpf_perf_event_print_t fn, void *private_data)
5254 {
5255         struct perf_event_mmap_page *header = mmap_mem;
5256         __u64 data_head = ring_buffer_read_head(header);
5257         __u64 data_tail = header->data_tail;
5258         void *base = ((__u8 *)header) + page_size;
5259         int ret = LIBBPF_PERF_EVENT_CONT;
5260         struct perf_event_header *ehdr;
5261         size_t ehdr_size;
5262 
5263         while (data_head != data_tail) {
5264                 ehdr = base + (data_tail & (mmap_size - 1));
5265                 ehdr_size = ehdr->size;
5266 
5267                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5268                         void *copy_start = ehdr;
5269                         size_t len_first = base + mmap_size - copy_start;
5270                         size_t len_secnd = ehdr_size - len_first;
5271 
5272                         if (*copy_size < ehdr_size) {
5273                                 free(*copy_mem);
5274                                 *copy_mem = malloc(ehdr_size);
5275                                 if (!*copy_mem) {
5276                                         *copy_size = 0;
5277                                         ret = LIBBPF_PERF_EVENT_ERROR;
5278                                         break;
5279                                 }
5280                                 *copy_size = ehdr_size;
5281                         }
5282 
5283                         memcpy(*copy_mem, copy_start, len_first);
5284                         memcpy(*copy_mem + len_first, base, len_secnd);
5285                         ehdr = *copy_mem;
5286                 }
5287 
5288                 ret = fn(ehdr, private_data);
5289                 data_tail += ehdr_size;
5290                 if (ret != LIBBPF_PERF_EVENT_CONT)
5291                         break;
5292         }
5293 
5294         ring_buffer_write_tail(header, data_tail);
5295         return ret;
5296 }
5297 
5298 struct perf_buffer;
5299 
5300 struct perf_buffer_params {
5301         struct perf_event_attr *attr;
5302         /* if event_cb is specified, it takes precendence */
5303         perf_buffer_event_fn event_cb;
5304         /* sample_cb and lost_cb are higher-level common-case callbacks */
5305         perf_buffer_sample_fn sample_cb;
5306         perf_buffer_lost_fn lost_cb;
5307         void *ctx;
5308         int cpu_cnt;
5309         int *cpus;
5310         int *map_keys;
5311 };
5312 
5313 struct perf_cpu_buf {
5314         struct perf_buffer *pb;
5315         void *base; /* mmap()'ed memory */
5316         void *buf; /* for reconstructing segmented data */
5317         size_t buf_size;
5318         int fd;
5319         int cpu;
5320         int map_key;
5321 };
5322 
5323 struct perf_buffer {
5324         perf_buffer_event_fn event_cb;
5325         perf_buffer_sample_fn sample_cb;
5326         perf_buffer_lost_fn lost_cb;
5327         void *ctx; /* passed into callbacks */
5328 
5329         size_t page_size;
5330         size_t mmap_size;
5331         struct perf_cpu_buf **cpu_bufs;
5332         struct epoll_event *events;
5333         int cpu_cnt;
5334         int epoll_fd; /* perf event FD */
5335         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5336 };
5337 
5338 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5339                                       struct perf_cpu_buf *cpu_buf)
5340 {
5341         if (!cpu_buf)
5342                 return;
5343         if (cpu_buf->base &&
5344             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5345                 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5346         if (cpu_buf->fd >= 0) {
5347                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5348                 close(cpu_buf->fd);
5349         }
5350         free(cpu_buf->buf);
5351         free(cpu_buf);
5352 }
5353 
5354 void perf_buffer__free(struct perf_buffer *pb)
5355 {
5356         int i;
5357 
5358         if (!pb)
5359                 return;
5360         if (pb->cpu_bufs) {
5361                 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5362                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5363 
5364                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5365                         perf_buffer__free_cpu_buf(pb, cpu_buf);
5366                 }
5367                 free(pb->cpu_bufs);
5368         }
5369         if (pb->epoll_fd >= 0)
5370                 close(pb->epoll_fd);
5371         free(pb->events);
5372         free(pb);
5373 }
5374 
5375 static struct perf_cpu_buf *
5376 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5377                           int cpu, int map_key)
5378 {
5379         struct perf_cpu_buf *cpu_buf;
5380         char msg[STRERR_BUFSIZE];
5381         int err;
5382 
5383         cpu_buf = calloc(1, sizeof(*cpu_buf));
5384         if (!cpu_buf)
5385                 return ERR_PTR(-ENOMEM);
5386 
5387         cpu_buf->pb = pb;
5388         cpu_buf->cpu = cpu;
5389         cpu_buf->map_key = map_key;
5390 
5391         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
5392                               -1, PERF_FLAG_FD_CLOEXEC);
5393         if (cpu_buf->fd < 0) {
5394                 err = -errno;
5395                 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
5396                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5397                 goto error;
5398         }
5399 
5400         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
5401                              PROT_READ | PROT_WRITE, MAP_SHARED,
5402                              cpu_buf->fd, 0);
5403         if (cpu_buf->base == MAP_FAILED) {
5404                 cpu_buf->base = NULL;
5405                 err = -errno;
5406                 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
5407                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5408                 goto error;
5409         }
5410 
5411         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5412                 err = -errno;
5413                 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
5414                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5415                 goto error;
5416         }
5417 
5418         return cpu_buf;
5419 
5420 error:
5421         perf_buffer__free_cpu_buf(pb, cpu_buf);
5422         return (struct perf_cpu_buf *)ERR_PTR(err);
5423 }
5424 
5425 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5426                                               struct perf_buffer_params *p);
5427 
5428 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
5429                                      const struct perf_buffer_opts *opts)
5430 {
5431         struct perf_buffer_params p = {};
5432         struct perf_event_attr attr = { 0, };
5433 
5434         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
5435         attr.type = PERF_TYPE_SOFTWARE;
5436         attr.sample_type = PERF_SAMPLE_RAW;
5437         attr.sample_period = 1;
5438         attr.wakeup_events = 1;
5439 
5440         p.attr = &attr;
5441         p.sample_cb = opts ? opts->sample_cb : NULL;
5442         p.lost_cb = opts ? opts->lost_cb : NULL;
5443         p.ctx = opts ? opts->ctx : NULL;
5444 
5445         return __perf_buffer__new(map_fd, page_cnt, &p);
5446 }
5447 
5448 struct perf_buffer *
5449 perf_buffer__new_raw(int map_fd, size_t page_cnt,
5450                      const struct perf_buffer_raw_opts *opts)
5451 {
5452         struct perf_buffer_params p = {};
5453 
5454         p.attr = opts->attr;
5455         p.event_cb = opts->event_cb;
5456         p.ctx = opts->ctx;
5457         p.cpu_cnt = opts->cpu_cnt;
5458         p.cpus = opts->cpus;
5459         p.map_keys = opts->map_keys;
5460 
5461         return __perf_buffer__new(map_fd, page_cnt, &p);
5462 }
5463 
5464 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5465                                               struct perf_buffer_params *p)
5466 {
5467         struct bpf_map_info map = {};
5468         char msg[STRERR_BUFSIZE];
5469         struct perf_buffer *pb;
5470         __u32 map_info_len;
5471         int err, i;
5472 
5473         if (page_cnt & (page_cnt - 1)) {
5474                 pr_warning("page count should be power of two, but is %zu\n",
5475                            page_cnt);
5476                 return ERR_PTR(-EINVAL);
5477         }
5478 
5479         map_info_len = sizeof(map);
5480         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
5481         if (err) {
5482                 err = -errno;
5483                 pr_warning("failed to get map info for map FD %d: %s\n",
5484                            map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
5485                 return ERR_PTR(err);
5486         }
5487 
5488         if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
5489                 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
5490                            map.name);
5491                 return ERR_PTR(-EINVAL);
5492         }
5493 
5494         pb = calloc(1, sizeof(*pb));
5495         if (!pb)
5496                 return ERR_PTR(-ENOMEM);
5497 
5498         pb->event_cb = p->event_cb;
5499         pb->sample_cb = p->sample_cb;
5500         pb->lost_cb = p->lost_cb;
5501         pb->ctx = p->ctx;
5502 
5503         pb->page_size = getpagesize();
5504         pb->mmap_size = pb->page_size * page_cnt;
5505         pb->map_fd = map_fd;
5506 
5507         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
5508         if (pb->epoll_fd < 0) {
5509                 err = -errno;
5510                 pr_warning("failed to create epoll instance: %s\n",
5511                            libbpf_strerror_r(err, msg, sizeof(msg)));
5512                 goto error;
5513         }
5514 
5515         if (p->cpu_cnt > 0) {
5516                 pb->cpu_cnt = p->cpu_cnt;
5517         } else {
5518                 pb->cpu_cnt = libbpf_num_possible_cpus();
5519                 if (pb->cpu_cnt < 0) {
5520                         err = pb->cpu_cnt;
5521                         goto error;
5522                 }
5523                 if (map.max_entries < pb->cpu_cnt)
5524                         pb->cpu_cnt = map.max_entries;
5525         }
5526 
5527         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
5528         if (!pb->events) {
5529                 err = -ENOMEM;
5530                 pr_warning("failed to allocate events: out of memory\n");
5531                 goto error;
5532         }
5533         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
5534         if (!pb->cpu_bufs) {
5535                 err = -ENOMEM;
5536                 pr_warning("failed to allocate buffers: out of memory\n");
5537                 goto error;
5538         }
5539 
5540         for (i = 0; i < pb->cpu_cnt; i++) {
5541                 struct perf_cpu_buf *cpu_buf;
5542                 int cpu, map_key;
5543 
5544                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
5545                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
5546 
5547                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
5548                 if (IS_ERR(cpu_buf)) {
5549                         err = PTR_ERR(cpu_buf);
5550                         goto error;
5551                 }
5552 
5553                 pb->cpu_bufs[i] = cpu_buf;
5554 
5555                 err = bpf_map_update_elem(pb->map_fd, &map_key,
5556                                           &cpu_buf->fd, 0);
5557                 if (err) {
5558                         err = -errno;
5559                         pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
5560                                    cpu, map_key, cpu_buf->fd,
5561                                    libbpf_strerror_r(err, msg, sizeof(msg)));
5562                         goto error;
5563                 }
5564 
5565                 pb->events[i].events = EPOLLIN;
5566                 pb->events[i].data.ptr = cpu_buf;
5567                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
5568                               &pb->events[i]) < 0) {
5569                         err = -errno;
5570                         pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
5571                                    cpu, cpu_buf->fd,
5572                                    libbpf_strerror_r(err, msg, sizeof(msg)));
5573                         goto error;
5574                 }
5575         }
5576 
5577         return pb;
5578 
5579 error:
5580         if (pb)
5581                 perf_buffer__free(pb);
5582         return ERR_PTR(err);
5583 }
5584 
5585 struct perf_sample_raw {
5586         struct perf_event_header header;
5587         uint32_t size;
5588         char data[0];
5589 };
5590 
5591 struct perf_sample_lost {
5592         struct perf_event_header header;
5593         uint64_t id;
5594         uint64_t lost;
5595         uint64_t sample_id;
5596 };
5597 
5598 static enum bpf_perf_event_ret
5599 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
5600 {
5601         struct perf_cpu_buf *cpu_buf = ctx;
5602         struct perf_buffer *pb = cpu_buf->pb;
5603         void *data = e;
5604 
5605         /* user wants full control over parsing perf event */
5606         if (pb->event_cb)
5607                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
5608 
5609         switch (e->type) {
5610         case PERF_RECORD_SAMPLE: {
5611                 struct perf_sample_raw *s = data;
5612 
5613                 if (pb->sample_cb)
5614                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
5615                 break;
5616         }
5617         case PERF_RECORD_LOST: {
5618                 struct perf_sample_lost *s = data;
5619 
5620                 if (pb->lost_cb)
5621                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
5622                 break;
5623         }
5624         default:
5625                 pr_warning("unknown perf sample type %d\n", e->type);
5626                 return LIBBPF_PERF_EVENT_ERROR;
5627         }
5628         return LIBBPF_PERF_EVENT_CONT;
5629 }
5630 
5631 static int perf_buffer__process_records(struct perf_buffer *pb,
5632                                         struct perf_cpu_buf *cpu_buf)
5633 {
5634         enum bpf_perf_event_ret ret;
5635 
5636         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
5637                                          pb->page_size, &cpu_buf->buf,
5638                                          &cpu_buf->buf_size,
5639                                          perf_buffer__process_record, cpu_buf);
5640         if (ret != LIBBPF_PERF_EVENT_CONT)
5641                 return ret;
5642         return 0;
5643 }
5644 
5645 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
5646 {
5647         int i, cnt, err;
5648 
5649         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
5650         for (i = 0; i < cnt; i++) {
5651                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
5652 
5653                 err = perf_buffer__process_records(pb, cpu_buf);
5654                 if (err) {
5655                         pr_warning("error while processing records: %d\n", err);
5656                         return err;
5657                 }
5658         }
5659         return cnt < 0 ? -errno : cnt;
5660 }
5661 
5662 struct bpf_prog_info_array_desc {
5663         int     array_offset;   /* e.g. offset of jited_prog_insns */
5664         int     count_offset;   /* e.g. offset of jited_prog_len */
5665         int     size_offset;    /* > 0: offset of rec size,
5666                                  * < 0: fix size of -size_offset
5667                                  */
5668 };
5669 
5670 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
5671         [BPF_PROG_INFO_JITED_INSNS] = {
5672                 offsetof(struct bpf_prog_info, jited_prog_insns),
5673                 offsetof(struct bpf_prog_info, jited_prog_len),
5674                 -1,
5675         },
5676         [BPF_PROG_INFO_XLATED_INSNS] = {
5677                 offsetof(struct bpf_prog_info, xlated_prog_insns),
5678                 offsetof(struct bpf_prog_info, xlated_prog_len),
5679                 -1,
5680         },
5681         [BPF_PROG_INFO_MAP_IDS] = {
5682                 offsetof(struct bpf_prog_info, map_ids),
5683                 offsetof(struct bpf_prog_info, nr_map_ids),
5684                 -(int)sizeof(__u32),
5685         },
5686         [BPF_PROG_INFO_JITED_KSYMS] = {
5687                 offsetof(struct bpf_prog_info, jited_ksyms),
5688                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
5689                 -(int)sizeof(__u64),
5690         },
5691         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
5692                 offsetof(struct bpf_prog_info, jited_func_lens),
5693                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
5694                 -(int)sizeof(__u32),
5695         },
5696         [BPF_PROG_INFO_FUNC_INFO] = {
5697                 offsetof(struct bpf_prog_info, func_info),
5698                 offsetof(struct bpf_prog_info, nr_func_info),
5699                 offsetof(struct bpf_prog_info, func_info_rec_size),
5700         },
5701         [BPF_PROG_INFO_LINE_INFO] = {
5702                 offsetof(struct bpf_prog_info, line_info),
5703                 offsetof(struct bpf_prog_info, nr_line_info),
5704                 offsetof(struct bpf_prog_info, line_info_rec_size),
5705         },
5706         [BPF_PROG_INFO_JITED_LINE_INFO] = {
5707                 offsetof(struct bpf_prog_info, jited_line_info),
5708                 offsetof(struct bpf_prog_info, nr_jited_line_info),
5709                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
5710         },
5711         [BPF_PROG_INFO_PROG_TAGS] = {
5712                 offsetof(struct bpf_prog_info, prog_tags),
5713                 offsetof(struct bpf_prog_info, nr_prog_tags),
5714                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
5715         },
5716 
5717 };
5718 
5719 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
5720 {
5721         __u32 *array = (__u32 *)info;
5722 
5723         if (offset >= 0)
5724                 return array[offset / sizeof(__u32)];
5725         return -(int)offset;
5726 }
5727 
5728 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
5729 {
5730         __u64 *array = (__u64 *)info;
5731 
5732         if (offset >= 0)
5733                 return array[offset / sizeof(__u64)];
5734         return -(int)offset;
5735 }
5736 
5737 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
5738                                          __u32 val)
5739 {
5740         __u32 *array = (__u32 *)info;
5741 
5742         if (offset >= 0)
5743                 array[offset / sizeof(__u32)] = val;
5744 }
5745 
5746 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
5747                                          __u64 val)
5748 {
5749         __u64 *array = (__u64 *)info;
5750 
5751         if (offset >= 0)
5752                 array[offset / sizeof(__u64)] = val;
5753 }
5754 
5755 struct bpf_prog_info_linear *
5756 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
5757 {
5758         struct bpf_prog_info_linear *info_linear;
5759         struct bpf_prog_info info = {};
5760         __u32 info_len = sizeof(info);
5761         __u32 data_len = 0;
5762         int i, err;
5763         void *ptr;
5764 
5765         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
5766                 return ERR_PTR(-EINVAL);
5767 
5768         /* step 1: get array dimensions */
5769         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
5770         if (err) {
5771                 pr_debug("can't get prog info: %s", strerror(errno));
5772                 return ERR_PTR(-EFAULT);
5773         }
5774 
5775         /* step 2: calculate total size of all arrays */
5776         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5777                 bool include_array = (arrays & (1UL << i)) > 0;
5778                 struct bpf_prog_info_array_desc *desc;
5779                 __u32 count, size;
5780 
5781                 desc = bpf_prog_info_array_desc + i;
5782 
5783                 /* kernel is too old to support this field */
5784                 if (info_len < desc->array_offset + sizeof(__u32) ||
5785                     info_len < desc->count_offset + sizeof(__u32) ||
5786                     (desc->size_offset > 0 && info_len < desc->size_offset))
5787                         include_array = false;
5788 
5789                 if (!include_array) {
5790                         arrays &= ~(1UL << i);  /* clear the bit */
5791                         continue;
5792                 }
5793 
5794                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5795                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5796 
5797                 data_len += count * size;
5798         }
5799 
5800         /* step 3: allocate continuous memory */
5801         data_len = roundup(data_len, sizeof(__u64));
5802         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
5803         if (!info_linear)
5804                 return ERR_PTR(-ENOMEM);
5805 
5806         /* step 4: fill data to info_linear->info */
5807         info_linear->arrays = arrays;
5808         memset(&info_linear->info, 0, sizeof(info));
5809         ptr = info_linear->data;
5810 
5811         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5812                 struct bpf_prog_info_array_desc *desc;
5813                 __u32 count, size;
5814 
5815                 if ((arrays & (1UL << i)) == 0)
5816                         continue;
5817 
5818                 desc  = bpf_prog_info_array_desc + i;
5819                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5820                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5821                 bpf_prog_info_set_offset_u32(&info_linear->info,
5822                                              desc->count_offset, count);
5823                 bpf_prog_info_set_offset_u32(&info_linear->info,
5824                                              desc->size_offset, size);
5825                 bpf_prog_info_set_offset_u64(&info_linear->info,
5826                                              desc->array_offset,
5827                                              ptr_to_u64(ptr));
5828                 ptr += count * size;
5829         }
5830 
5831         /* step 5: call syscall again to get required arrays */
5832         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
5833         if (err) {
5834                 pr_debug("can't get prog info: %s", strerror(errno));
5835                 free(info_linear);
5836                 return ERR_PTR(-EFAULT);
5837         }
5838 
5839         /* step 6: verify the data */
5840         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5841                 struct bpf_prog_info_array_desc *desc;
5842                 __u32 v1, v2;
5843 
5844                 if ((arrays & (1UL << i)) == 0)
5845                         continue;
5846 
5847                 desc = bpf_prog_info_array_desc + i;
5848                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5849                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5850                                                    desc->count_offset);
5851                 if (v1 != v2)
5852                         pr_warning("%s: mismatch in element count\n", __func__);
5853 
5854                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5855                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5856                                                    desc->size_offset);
5857                 if (v1 != v2)
5858                         pr_warning("%s: mismatch in rec size\n", __func__);
5859         }
5860 
5861         /* step 7: update info_len and data_len */
5862         info_linear->info_len = sizeof(struct bpf_prog_info);
5863         info_linear->data_len = data_len;
5864 
5865         return info_linear;
5866 }
5867 
5868 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
5869 {
5870         int i;
5871 
5872         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5873                 struct bpf_prog_info_array_desc *desc;
5874                 __u64 addr, offs;
5875 
5876                 if ((info_linear->arrays & (1UL << i)) == 0)
5877                         continue;
5878 
5879                 desc = bpf_prog_info_array_desc + i;
5880                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
5881                                                      desc->array_offset);
5882                 offs = addr - ptr_to_u64(info_linear->data);
5883                 bpf_prog_info_set_offset_u64(&info_linear->info,
5884                                              desc->array_offset, offs);
5885         }
5886 }
5887 
5888 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
5889 {
5890         int i;
5891 
5892         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5893                 struct bpf_prog_info_array_desc *desc;
5894                 __u64 addr, offs;
5895 
5896                 if ((info_linear->arrays & (1UL << i)) == 0)
5897                         continue;
5898 
5899                 desc = bpf_prog_info_array_desc + i;
5900                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
5901                                                      desc->array_offset);
5902                 addr = offs + ptr_to_u64(info_linear->data);
5903                 bpf_prog_info_set_offset_u64(&info_linear->info,
5904                                              desc->array_offset, addr);
5905         }
5906 }
5907 
5908 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
5909 {
5910         int err = 0, n, len, start, end = -1;
5911         bool *tmp;
5912 
5913         *mask = NULL;
5914         *mask_sz = 0;
5915 
5916         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
5917         while (*s) {
5918                 if (*s == ',' || *s == '\n') {
5919                         s++;
5920                         continue;
5921                 }
5922                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
5923                 if (n <= 0 || n > 2) {
5924                         pr_warning("Failed to get CPU range %s: %d\n", s, n);
5925                         err = -EINVAL;
5926                         goto cleanup;
5927                 } else if (n == 1) {
5928                         end = start;
5929                 }
5930                 if (start < 0 || start > end) {
5931                         pr_warning("Invalid CPU range [%d,%d] in %s\n",
5932                                    start, end, s);
5933                         err = -EINVAL;
5934                         goto cleanup;
5935                 }
5936                 tmp = realloc(*mask, end + 1);
5937                 if (!tmp) {
5938                         err = -ENOMEM;
5939                         goto cleanup;
5940                 }
5941                 *mask = tmp;
5942                 memset(tmp + *mask_sz, 0, start - *mask_sz);
5943                 memset(tmp + start, 1, end - start + 1);
5944                 *mask_sz = end + 1;
5945                 s += len;
5946         }
5947         if (!*mask_sz) {
5948                 pr_warning("Empty CPU range\n");
5949                 return -EINVAL;
5950         }
5951         return 0;
5952 cleanup:
5953         free(*mask);
5954         *mask = NULL;
5955         return err;
5956 }
5957 
5958 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
5959 {
5960         int fd, err = 0, len;
5961         char buf[128];
5962 
5963         fd = open(fcpu, O_RDONLY);
5964         if (fd < 0) {
5965                 err = -errno;
5966                 pr_warning("Failed to open cpu mask file %s: %d\n", fcpu, err);
5967                 return err;
5968         }
5969         len = read(fd, buf, sizeof(buf));
5970         close(fd);
5971         if (len <= 0) {
5972                 err = len ? -errno : -EINVAL;
5973                 pr_warning("Failed to read cpu mask from %s: %d\n", fcpu, err);
5974                 return err;
5975         }
5976         if (len >= sizeof(buf)) {
5977                 pr_warning("CPU mask is too big in file %s\n", fcpu);
5978                 return -E2BIG;
5979         }
5980         buf[len] = '\0';
5981 
5982         return parse_cpu_mask_str(buf, mask, mask_sz);
5983 }
5984 
5985 int libbpf_num_possible_cpus(void)
5986 {
5987         static const char *fcpu = "/sys/devices/system/cpu/possible";
5988         static int cpus;
5989         int err, n, i, tmp_cpus;
5990         bool *mask;
5991 
5992         tmp_cpus = READ_ONCE(cpus);
5993         if (tmp_cpus > 0)
5994                 return tmp_cpus;
5995 
5996         err = parse_cpu_mask_file(fcpu, &mask, &n);
5997         if (err)
5998                 return err;
5999 
6000         tmp_cpus = 0;
6001         for (i = 0; i < n; i++) {
6002                 if (mask[i])
6003                         tmp_cpus++;
6004         }
6005         free(mask);
6006 
6007         WRITE_ONCE(cpus, tmp_cpus);
6008         return tmp_cpus;
6009 }

/* [<][>][^][v][top][bottom][index][help] */