tools/lib/bpf/libbpf.c

   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 <limits.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <endian.h>
  25 #include <fcntl.h>
  26 #include <errno.h>
  27 #include <ctype.h>
  28 #include <asm/unistd.h>
  29 #include <linux/err.h>
  30 #include <linux/kernel.h>
  31 #include <linux/bpf.h>
  32 #include <linux/btf.h>
  33 #include <linux/filter.h>
  34 #include <linux/list.h>
  35 #include <linux/limits.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/ring_buffer.h>
  38 #include <linux/version.h>
  39 #include <sys/epoll.h>
  40 #include <sys/ioctl.h>
  41 #include <sys/mman.h>
  42 #include <sys/stat.h>
  43 #include <sys/types.h>
  44 #include <sys/vfs.h>
  45 #include <sys/utsname.h>
  46 #include <sys/resource.h>
  47 #include <tools/libc_compat.h>
  48 #include <libelf.h>
  49 #include <gelf.h>
  50 #include <zlib.h>
  51
  52 #include "libbpf.h"
  53 #include "bpf.h"
  54 #include "btf.h"
  55 #include "str_error.h"
  56 #include "libbpf_internal.h"
  57 #include "hashmap.h"
  58
  59 /* make sure libbpf doesn't use kernel-only integer typedefs */
  60 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
  61
  62 #ifndef EM_BPF
  63 #define EM_BPF 247
  64 #endif
  65
  66 #ifndef BPF_FS_MAGIC
  67 #define BPF_FS_MAGIC            0xcafe4a11
  68 #endif
  69
  70 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  71  * compilation if user enables corresponding warning. Disable it explicitly.
  72  */
  73 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  74
  75 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  76
  77 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
  78 static struct bpf_program *bpf_object__find_prog_by_idx(struct bpf_object *obj,
  79                                                         int idx);
  80 static const struct btf_type *
  81 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id);
  82
  83 static int __base_pr(enum libbpf_print_level level, const char *format,
  84                      va_list args)
  85 {
  86         if (level == LIBBPF_DEBUG)
  87                 return 0;
  88
  89         return vfprintf(stderr, format, args);
  90 }
  91
  92 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  93
  94 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  95 {
  96         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  97
  98         __libbpf_pr = fn;
  99         return old_print_fn;
 100 }
 101
 102 __printf(2, 3)
 103 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 104 {
 105         va_list args;
 106
 107         if (!__libbpf_pr)
 108                 return;
 109
 110         va_start(args, format);
 111         __libbpf_pr(level, format, args);
 112         va_end(args);
 113 }
 114
 115 static void pr_perm_msg(int err)
 116 {
 117         struct rlimit limit;
 118         char buf[100];
 119
 120         if (err != -EPERM || geteuid() != 0)
 121                 return;
 122
 123         err = getrlimit(RLIMIT_MEMLOCK, &limit);
 124         if (err)
 125                 return;
 126
 127         if (limit.rlim_cur == RLIM_INFINITY)
 128                 return;
 129
 130         if (limit.rlim_cur < 1024)
 131                 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
 132         else if (limit.rlim_cur < 1024*1024)
 133                 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
 134         else
 135                 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
 136
 137         pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
 138                 buf);
 139 }
 140
 141 #define STRERR_BUFSIZE  128
 142
 143 /* Copied from tools/perf/util/util.h */
 144 #ifndef zfree
 145 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 146 #endif
 147
 148 #ifndef zclose
 149 # define zclose(fd) ({                  \
 150         int ___err = 0;                 \
 151         if ((fd) >= 0)                  \
 152                 ___err = close((fd));   \
 153         fd = -1;                        \
 154         ___err; })
 155 #endif
 156
 157 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 158 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 159 #else
 160 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 161 #endif
 162
 163 static inline __u64 ptr_to_u64(const void *ptr)
 164 {
 165         return (__u64) (unsigned long) ptr;
 166 }
 167
 168 struct bpf_capabilities {
 169         /* v4.14: kernel support for program & map names. */
 170         __u32 name:1;
 171         /* v5.2: kernel support for global data sections. */
 172         __u32 global_data:1;
 173         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 174         __u32 btf_func:1;
 175         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 176         __u32 btf_datasec:1;
 177         /* BPF_F_MMAPABLE is supported for arrays */
 178         __u32 array_mmap:1;
 179         /* BTF_FUNC_GLOBAL is supported */
 180         __u32 btf_func_global:1;
 181         /* kernel support for expected_attach_type in BPF_PROG_LOAD */
 182         __u32 exp_attach_type:1;
 183 };
 184
 185 enum reloc_type {
 186         RELO_LD64,
 187         RELO_CALL,
 188         RELO_DATA,
 189         RELO_EXTERN,
 190 };
 191
 192 struct reloc_desc {
 193         enum reloc_type type;
 194         int insn_idx;
 195         int map_idx;
 196         int sym_off;
 197 };
 198
 199 struct bpf_sec_def;
 200
 201 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
 202                                         struct bpf_program *prog);
 203
 204 struct bpf_sec_def {
 205         const char *sec;
 206         size_t len;
 207         enum bpf_prog_type prog_type;
 208         enum bpf_attach_type expected_attach_type;
 209         bool is_exp_attach_type_optional;
 210         bool is_attachable;
 211         bool is_attach_btf;
 212         attach_fn_t attach_fn;
 213 };
 214
 215 /*
 216  * bpf_prog should be a better name but it has been used in
 217  * linux/filter.h.
 218  */
 219 struct bpf_program {
 220         /* Index in elf obj file, for relocation use. */
 221         int idx;
 222         char *name;
 223         int prog_ifindex;
 224         char *section_name;
 225         const struct bpf_sec_def *sec_def;
 226         /* section_name with / replaced by _; makes recursive pinning
 227          * in bpf_object__pin_programs easier
 228          */
 229         char *pin_name;
 230         struct bpf_insn *insns;
 231         size_t insns_cnt, main_prog_cnt;
 232         enum bpf_prog_type type;
 233         bool load;
 234
 235         struct reloc_desc *reloc_desc;
 236         int nr_reloc;
 237         int log_level;
 238
 239         struct {
 240                 int nr;
 241                 int *fds;
 242         } instances;
 243         bpf_program_prep_t preprocessor;
 244
 245         struct bpf_object *obj;
 246         void *priv;
 247         bpf_program_clear_priv_t clear_priv;
 248
 249         enum bpf_attach_type expected_attach_type;
 250         __u32 attach_btf_id;
 251         __u32 attach_prog_fd;
 252         void *func_info;
 253         __u32 func_info_rec_size;
 254         __u32 func_info_cnt;
 255
 256         struct bpf_capabilities *caps;
 257
 258         void *line_info;
 259         __u32 line_info_rec_size;
 260         __u32 line_info_cnt;
 261         __u32 prog_flags;
 262 };
 263
 264 struct bpf_struct_ops {
 265         const char *tname;
 266         const struct btf_type *type;
 267         struct bpf_program **progs;
 268         __u32 *kern_func_off;
 269         /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
 270         void *data;
 271         /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
 272          *      btf_vmlinux's format.
 273          * struct bpf_struct_ops_tcp_congestion_ops {
 274          *      [... some other kernel fields ...]
 275          *      struct tcp_congestion_ops data;
 276          * }
 277          * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
 278          * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
 279          * from "data".
 280          */
 281         void *kern_vdata;
 282         __u32 type_id;
 283 };
 284
 285 #define DATA_SEC ".data"
 286 #define BSS_SEC ".bss"
 287 #define RODATA_SEC ".rodata"
 288 #define KCONFIG_SEC ".kconfig"
 289 #define KSYMS_SEC ".ksyms"
 290 #define STRUCT_OPS_SEC ".struct_ops"
 291
 292 enum libbpf_map_type {
 293         LIBBPF_MAP_UNSPEC,
 294         LIBBPF_MAP_DATA,
 295         LIBBPF_MAP_BSS,
 296         LIBBPF_MAP_RODATA,
 297         LIBBPF_MAP_KCONFIG,
 298 };
 299
 300 static const char * const libbpf_type_to_btf_name[] = {
 301         [LIBBPF_MAP_DATA]       = DATA_SEC,
 302         [LIBBPF_MAP_BSS]        = BSS_SEC,
 303         [LIBBPF_MAP_RODATA]     = RODATA_SEC,
 304         [LIBBPF_MAP_KCONFIG]    = KCONFIG_SEC,
 305 };
 306
 307 struct bpf_map {
 308         char *name;
 309         int fd;
 310         int sec_idx;
 311         size_t sec_offset;
 312         int map_ifindex;
 313         int inner_map_fd;
 314         struct bpf_map_def def;
 315         __u32 numa_node;
 316         __u32 btf_var_idx;
 317         __u32 btf_key_type_id;
 318         __u32 btf_value_type_id;
 319         __u32 btf_vmlinux_value_type_id;
 320         void *priv;
 321         bpf_map_clear_priv_t clear_priv;
 322         enum libbpf_map_type libbpf_type;
 323         void *mmaped;
 324         struct bpf_struct_ops *st_ops;
 325         struct bpf_map *inner_map;
 326         void **init_slots;
 327         int init_slots_sz;
 328         char *pin_path;
 329         bool pinned;
 330         bool reused;
 331 };
 332
 333 enum extern_type {
 334         EXT_UNKNOWN,
 335         EXT_KCFG,
 336         EXT_KSYM,
 337 };
 338
 339 enum kcfg_type {
 340         KCFG_UNKNOWN,
 341         KCFG_CHAR,
 342         KCFG_BOOL,
 343         KCFG_INT,
 344         KCFG_TRISTATE,
 345         KCFG_CHAR_ARR,
 346 };
 347
 348 struct extern_desc {
 349         enum extern_type type;
 350         int sym_idx;
 351         int btf_id;
 352         int sec_btf_id;
 353         const char *name;
 354         bool is_set;
 355         bool is_weak;
 356         union {
 357                 struct {
 358                         enum kcfg_type type;
 359                         int sz;
 360                         int align;
 361                         int data_off;
 362                         bool is_signed;
 363                 } kcfg;
 364                 struct {
 365                         unsigned long long addr;
 366                 } ksym;
 367         };
 368 };
 369
 370 static LIST_HEAD(bpf_objects_list);
 371
 372 struct bpf_object {
 373         char name[BPF_OBJ_NAME_LEN];
 374         char license[64];
 375         __u32 kern_version;
 376
 377         struct bpf_program *programs;
 378         size_t nr_programs;
 379         struct bpf_map *maps;
 380         size_t nr_maps;
 381         size_t maps_cap;
 382
 383         char *kconfig;
 384         struct extern_desc *externs;
 385         int nr_extern;
 386         int kconfig_map_idx;
 387
 388         bool loaded;
 389         bool has_pseudo_calls;
 390
 391         /*
 392          * Information when doing elf related work. Only valid if fd
 393          * is valid.
 394          */
 395         struct {
 396                 int fd;
 397                 const void *obj_buf;
 398                 size_t obj_buf_sz;
 399                 Elf *elf;
 400                 GElf_Ehdr ehdr;
 401                 Elf_Data *symbols;
 402                 Elf_Data *data;
 403                 Elf_Data *rodata;
 404                 Elf_Data *bss;
 405                 Elf_Data *st_ops_data;
 406                 size_t strtabidx;
 407                 struct {
 408                         GElf_Shdr shdr;
 409                         Elf_Data *data;
 410                 } *reloc_sects;
 411                 int nr_reloc_sects;
 412                 int maps_shndx;
 413                 int btf_maps_shndx;
 414                 __u32 btf_maps_sec_btf_id;
 415                 int text_shndx;
 416                 int symbols_shndx;
 417                 int data_shndx;
 418                 int rodata_shndx;
 419                 int bss_shndx;
 420                 int st_ops_shndx;
 421         } efile;
 422         /*
 423          * All loaded bpf_object is linked in a list, which is
 424          * hidden to caller. bpf_objects__<func> handlers deal with
 425          * all objects.
 426          */
 427         struct list_head list;
 428
 429         struct btf *btf;
 430         /* Parse and load BTF vmlinux if any of the programs in the object need
 431          * it at load time.
 432          */
 433         struct btf *btf_vmlinux;
 434         struct btf_ext *btf_ext;
 435
 436         void *priv;
 437         bpf_object_clear_priv_t clear_priv;
 438
 439         struct bpf_capabilities caps;
 440
 441         char path[];
 442 };
 443 #define obj_elf_valid(o)        ((o)->efile.elf)
 444
 445 void bpf_program__unload(struct bpf_program *prog)
 446 {
 447         int i;
 448
 449         if (!prog)
 450                 return;
 451
 452         /*
 453          * If the object is opened but the program was never loaded,
 454          * it is possible that prog->instances.nr == -1.
 455          */
 456         if (prog->instances.nr > 0) {
 457                 for (i = 0; i < prog->instances.nr; i++)
 458                         zclose(prog->instances.fds[i]);
 459         } else if (prog->instances.nr != -1) {
 460                 pr_warn("Internal error: instances.nr is %d\n",
 461                         prog->instances.nr);
 462         }
 463
 464         prog->instances.nr = -1;
 465         zfree(&prog->instances.fds);
 466
 467         zfree(&prog->func_info);
 468         zfree(&prog->line_info);
 469 }
 470
 471 static void bpf_program__exit(struct bpf_program *prog)
 472 {
 473         if (!prog)
 474                 return;
 475
 476         if (prog->clear_priv)
 477                 prog->clear_priv(prog, prog->priv);
 478
 479         prog->priv = NULL;
 480         prog->clear_priv = NULL;
 481
 482         bpf_program__unload(prog);
 483         zfree(&prog->name);
 484         zfree(&prog->section_name);
 485         zfree(&prog->pin_name);
 486         zfree(&prog->insns);
 487         zfree(&prog->reloc_desc);
 488
 489         prog->nr_reloc = 0;
 490         prog->insns_cnt = 0;
 491         prog->idx = -1;
 492 }
 493
 494 static char *__bpf_program__pin_name(struct bpf_program *prog)
 495 {
 496         char *name, *p;
 497
 498         name = p = strdup(prog->section_name);
 499         while ((p = strchr(p, '/')))
 500                 *p = '_';
 501
 502         return name;
 503 }
 504
 505 static int
 506 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 507                   struct bpf_program *prog)
 508 {
 509         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 510
 511         if (size == 0 || size % bpf_insn_sz) {
 512                 pr_warn("corrupted section '%s', size: %zu\n",
 513                         section_name, size);
 514                 return -EINVAL;
 515         }
 516
 517         memset(prog, 0, sizeof(*prog));
 518
 519         prog->section_name = strdup(section_name);
 520         if (!prog->section_name) {
 521                 pr_warn("failed to alloc name for prog under section(%d) %s\n",
 522                         idx, section_name);
 523                 goto errout;
 524         }
 525
 526         prog->pin_name = __bpf_program__pin_name(prog);
 527         if (!prog->pin_name) {
 528                 pr_warn("failed to alloc pin name for prog under section(%d) %s\n",
 529                         idx, section_name);
 530                 goto errout;
 531         }
 532
 533         prog->insns = malloc(size);
 534         if (!prog->insns) {
 535                 pr_warn("failed to alloc insns for prog under section %s\n",
 536                         section_name);
 537                 goto errout;
 538         }
 539         prog->insns_cnt = size / bpf_insn_sz;
 540         memcpy(prog->insns, data, size);
 541         prog->idx = idx;
 542         prog->instances.fds = NULL;
 543         prog->instances.nr = -1;
 544         prog->type = BPF_PROG_TYPE_UNSPEC;
 545         prog->load = true;
 546
 547         return 0;
 548 errout:
 549         bpf_program__exit(prog);
 550         return -ENOMEM;
 551 }
 552
 553 static int
 554 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 555                         char *section_name, int idx)
 556 {
 557         struct bpf_program prog, *progs;
 558         int nr_progs, err;
 559
 560         err = bpf_program__init(data, size, section_name, idx, &prog);
 561         if (err)
 562                 return err;
 563
 564         prog.caps = &obj->caps;
 565         progs = obj->programs;
 566         nr_progs = obj->nr_programs;
 567
 568         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 569         if (!progs) {
 570                 /*
 571                  * In this case the original obj->programs
 572                  * is still valid, so don't need special treat for
 573                  * bpf_close_object().
 574                  */
 575                 pr_warn("failed to alloc a new program under section '%s'\n",
 576                         section_name);
 577                 bpf_program__exit(&prog);
 578                 return -ENOMEM;
 579         }
 580
 581         pr_debug("found program %s\n", prog.section_name);
 582         obj->programs = progs;
 583         obj->nr_programs = nr_progs + 1;
 584         prog.obj = obj;
 585         progs[nr_progs] = prog;
 586         return 0;
 587 }
 588
 589 static int
 590 bpf_object__init_prog_names(struct bpf_object *obj)
 591 {
 592         Elf_Data *symbols = obj->efile.symbols;
 593         struct bpf_program *prog;
 594         size_t pi, si;
 595
 596         for (pi = 0; pi < obj->nr_programs; pi++) {
 597                 const char *name = NULL;
 598
 599                 prog = &obj->programs[pi];
 600
 601                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 602                      si++) {
 603                         GElf_Sym sym;
 604
 605                         if (!gelf_getsym(symbols, si, &sym))
 606                                 continue;
 607                         if (sym.st_shndx != prog->idx)
 608                                 continue;
 609                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 610                                 continue;
 611
 612                         name = elf_strptr(obj->efile.elf,
 613                                           obj->efile.strtabidx,
 614                                           sym.st_name);
 615                         if (!name) {
 616                                 pr_warn("failed to get sym name string for prog %s\n",
 617                                         prog->section_name);
 618                                 return -LIBBPF_ERRNO__LIBELF;
 619                         }
 620                 }
 621
 622                 if (!name && prog->idx == obj->efile.text_shndx)
 623                         name = ".text";
 624
 625                 if (!name) {
 626                         pr_warn("failed to find sym for prog %s\n",
 627                                 prog->section_name);
 628                         return -EINVAL;
 629                 }
 630
 631                 prog->name = strdup(name);
 632                 if (!prog->name) {
 633                         pr_warn("failed to allocate memory for prog sym %s\n",
 634                                 name);
 635                         return -ENOMEM;
 636                 }
 637         }
 638
 639         return 0;
 640 }
 641
 642 static __u32 get_kernel_version(void)
 643 {
 644         __u32 major, minor, patch;
 645         struct utsname info;
 646
 647         uname(&info);
 648         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 649                 return 0;
 650         return KERNEL_VERSION(major, minor, patch);
 651 }
 652
 653 static const struct btf_member *
 654 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
 655 {
 656         struct btf_member *m;
 657         int i;
 658
 659         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 660                 if (btf_member_bit_offset(t, i) == bit_offset)
 661                         return m;
 662         }
 663
 664         return NULL;
 665 }
 666
 667 static const struct btf_member *
 668 find_member_by_name(const struct btf *btf, const struct btf_type *t,
 669                     const char *name)
 670 {
 671         struct btf_member *m;
 672         int i;
 673
 674         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 675                 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
 676                         return m;
 677         }
 678
 679         return NULL;
 680 }
 681
 682 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
 683 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
 684                                    const char *name, __u32 kind);
 685
 686 static int
 687 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
 688                            const struct btf_type **type, __u32 *type_id,
 689                            const struct btf_type **vtype, __u32 *vtype_id,
 690                            const struct btf_member **data_member)
 691 {
 692         const struct btf_type *kern_type, *kern_vtype;
 693         const struct btf_member *kern_data_member;
 694         __s32 kern_vtype_id, kern_type_id;
 695         __u32 i;
 696
 697         kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
 698         if (kern_type_id < 0) {
 699                 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
 700                         tname);
 701                 return kern_type_id;
 702         }
 703         kern_type = btf__type_by_id(btf, kern_type_id);
 704
 705         /* Find the corresponding "map_value" type that will be used
 706          * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 707          * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 708          * btf_vmlinux.
 709          */
 710         kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
 711                                                 tname, BTF_KIND_STRUCT);
 712         if (kern_vtype_id < 0) {
 713                 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
 714                         STRUCT_OPS_VALUE_PREFIX, tname);
 715                 return kern_vtype_id;
 716         }
 717         kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 718
 719         /* Find "struct tcp_congestion_ops" from
 720          * struct bpf_struct_ops_tcp_congestion_ops {
 721          *      [ ... ]
 722          *      struct tcp_congestion_ops data;
 723          * }
 724          */
 725         kern_data_member = btf_members(kern_vtype);
 726         for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
 727                 if (kern_data_member->type == kern_type_id)
 728                         break;
 729         }
 730         if (i == btf_vlen(kern_vtype)) {
 731                 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
 732                         tname, STRUCT_OPS_VALUE_PREFIX, tname);
 733                 return -EINVAL;
 734         }
 735
 736         *type = kern_type;
 737         *type_id = kern_type_id;
 738         *vtype = kern_vtype;
 739         *vtype_id = kern_vtype_id;
 740         *data_member = kern_data_member;
 741
 742         return 0;
 743 }
 744
 745 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
 746 {
 747         return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
 748 }
 749
 750 /* Init the map's fields that depend on kern_btf */
 751 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
 752                                          const struct btf *btf,
 753                                          const struct btf *kern_btf)
 754 {
 755         const struct btf_member *member, *kern_member, *kern_data_member;
 756         const struct btf_type *type, *kern_type, *kern_vtype;
 757         __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
 758         struct bpf_struct_ops *st_ops;
 759         void *data, *kern_data;
 760         const char *tname;
 761         int err;
 762
 763         st_ops = map->st_ops;
 764         type = st_ops->type;
 765         tname = st_ops->tname;
 766         err = find_struct_ops_kern_types(kern_btf, tname,
 767                                          &kern_type, &kern_type_id,
 768                                          &kern_vtype, &kern_vtype_id,
 769                                          &kern_data_member);
 770         if (err)
 771                 return err;
 772
 773         pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
 774                  map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
 775
 776         map->def.value_size = kern_vtype->size;
 777         map->btf_vmlinux_value_type_id = kern_vtype_id;
 778
 779         st_ops->kern_vdata = calloc(1, kern_vtype->size);
 780         if (!st_ops->kern_vdata)
 781                 return -ENOMEM;
 782
 783         data = st_ops->data;
 784         kern_data_off = kern_data_member->offset / 8;
 785         kern_data = st_ops->kern_vdata + kern_data_off;
 786
 787         member = btf_members(type);
 788         for (i = 0; i < btf_vlen(type); i++, member++) {
 789                 const struct btf_type *mtype, *kern_mtype;
 790                 __u32 mtype_id, kern_mtype_id;
 791                 void *mdata, *kern_mdata;
 792                 __s64 msize, kern_msize;
 793                 __u32 moff, kern_moff;
 794                 __u32 kern_member_idx;
 795                 const char *mname;
 796
 797                 mname = btf__name_by_offset(btf, member->name_off);
 798                 kern_member = find_member_by_name(kern_btf, kern_type, mname);
 799                 if (!kern_member) {
 800                         pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
 801                                 map->name, mname);
 802                         return -ENOTSUP;
 803                 }
 804
 805                 kern_member_idx = kern_member - btf_members(kern_type);
 806                 if (btf_member_bitfield_size(type, i) ||
 807                     btf_member_bitfield_size(kern_type, kern_member_idx)) {
 808                         pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
 809                                 map->name, mname);
 810                         return -ENOTSUP;
 811                 }
 812
 813                 moff = member->offset / 8;
 814                 kern_moff = kern_member->offset / 8;
 815
 816                 mdata = data + moff;
 817                 kern_mdata = kern_data + kern_moff;
 818
 819                 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
 820                 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
 821                                                     &kern_mtype_id);
 822                 if (BTF_INFO_KIND(mtype->info) !=
 823                     BTF_INFO_KIND(kern_mtype->info)) {
 824                         pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
 825                                 map->name, mname, BTF_INFO_KIND(mtype->info),
 826                                 BTF_INFO_KIND(kern_mtype->info));
 827                         return -ENOTSUP;
 828                 }
 829
 830                 if (btf_is_ptr(mtype)) {
 831                         struct bpf_program *prog;
 832
 833                         mtype = skip_mods_and_typedefs(btf, mtype->type, &mtype_id);
 834                         kern_mtype = skip_mods_and_typedefs(kern_btf,
 835                                                             kern_mtype->type,
 836                                                             &kern_mtype_id);
 837                         if (!btf_is_func_proto(mtype) ||
 838                             !btf_is_func_proto(kern_mtype)) {
 839                                 pr_warn("struct_ops init_kern %s: non func ptr %s is not supported\n",
 840                                         map->name, mname);
 841                                 return -ENOTSUP;
 842                         }
 843
 844                         prog = st_ops->progs[i];
 845                         if (!prog) {
 846                                 pr_debug("struct_ops init_kern %s: func ptr %s is not set\n",
 847                                          map->name, mname);
 848                                 continue;
 849                         }
 850
 851                         prog->attach_btf_id = kern_type_id;
 852                         prog->expected_attach_type = kern_member_idx;
 853
 854                         st_ops->kern_func_off[i] = kern_data_off + kern_moff;
 855
 856                         pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
 857                                  map->name, mname, prog->name, moff,
 858                                  kern_moff);
 859
 860                         continue;
 861                 }
 862
 863                 msize = btf__resolve_size(btf, mtype_id);
 864                 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
 865                 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
 866                         pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
 867                                 map->name, mname, (ssize_t)msize,
 868                                 (ssize_t)kern_msize);
 869                         return -ENOTSUP;
 870                 }
 871
 872                 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
 873                          map->name, mname, (unsigned int)msize,
 874                          moff, kern_moff);
 875                 memcpy(kern_mdata, mdata, msize);
 876         }
 877
 878         return 0;
 879 }
 880
 881 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
 882 {
 883         struct bpf_map *map;
 884         size_t i;
 885         int err;
 886
 887         for (i = 0; i < obj->nr_maps; i++) {
 888                 map = &obj->maps[i];
 889
 890                 if (!bpf_map__is_struct_ops(map))
 891                         continue;
 892
 893                 err = bpf_map__init_kern_struct_ops(map, obj->btf,
 894                                                     obj->btf_vmlinux);
 895                 if (err)
 896                         return err;
 897         }
 898
 899         return 0;
 900 }
 901
 902 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
 903 {
 904         const struct btf_type *type, *datasec;
 905         const struct btf_var_secinfo *vsi;
 906         struct bpf_struct_ops *st_ops;
 907         const char *tname, *var_name;
 908         __s32 type_id, datasec_id;
 909         const struct btf *btf;
 910         struct bpf_map *map;
 911         __u32 i;
 912
 913         if (obj->efile.st_ops_shndx == -1)
 914                 return 0;
 915
 916         btf = obj->btf;
 917         datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
 918                                             BTF_KIND_DATASEC);
 919         if (datasec_id < 0) {
 920                 pr_warn("struct_ops init: DATASEC %s not found\n",
 921                         STRUCT_OPS_SEC);
 922                 return -EINVAL;
 923         }
 924
 925         datasec = btf__type_by_id(btf, datasec_id);
 926         vsi = btf_var_secinfos(datasec);
 927         for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
 928                 type = btf__type_by_id(obj->btf, vsi->type);
 929                 var_name = btf__name_by_offset(obj->btf, type->name_off);
 930
 931                 type_id = btf__resolve_type(obj->btf, vsi->type);
 932                 if (type_id < 0) {
 933                         pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
 934                                 vsi->type, STRUCT_OPS_SEC);
 935                         return -EINVAL;
 936                 }
 937
 938                 type = btf__type_by_id(obj->btf, type_id);
 939                 tname = btf__name_by_offset(obj->btf, type->name_off);
 940                 if (!tname[0]) {
 941                         pr_warn("struct_ops init: anonymous type is not supported\n");
 942                         return -ENOTSUP;
 943                 }
 944                 if (!btf_is_struct(type)) {
 945                         pr_warn("struct_ops init: %s is not a struct\n", tname);
 946                         return -EINVAL;
 947                 }
 948
 949                 map = bpf_object__add_map(obj);
 950                 if (IS_ERR(map))
 951                         return PTR_ERR(map);
 952
 953                 map->sec_idx = obj->efile.st_ops_shndx;
 954                 map->sec_offset = vsi->offset;
 955                 map->name = strdup(var_name);
 956                 if (!map->name)
 957                         return -ENOMEM;
 958
 959                 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
 960                 map->def.key_size = sizeof(int);
 961                 map->def.value_size = type->size;
 962                 map->def.max_entries = 1;
 963
 964                 map->st_ops = calloc(1, sizeof(*map->st_ops));
 965                 if (!map->st_ops)
 966                         return -ENOMEM;
 967                 st_ops = map->st_ops;
 968                 st_ops->data = malloc(type->size);
 969                 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
 970                 st_ops->kern_func_off = malloc(btf_vlen(type) *
 971                                                sizeof(*st_ops->kern_func_off));
 972                 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
 973                         return -ENOMEM;
 974
 975                 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
 976                         pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
 977                                 var_name, STRUCT_OPS_SEC);
 978                         return -EINVAL;
 979                 }
 980
 981                 memcpy(st_ops->data,
 982                        obj->efile.st_ops_data->d_buf + vsi->offset,
 983                        type->size);
 984                 st_ops->tname = tname;
 985                 st_ops->type = type;
 986                 st_ops->type_id = type_id;
 987
 988                 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
 989                          tname, type_id, var_name, vsi->offset);
 990         }
 991
 992         return 0;
 993 }
 994
 995 static struct bpf_object *bpf_object__new(const char *path,
 996                                           const void *obj_buf,
 997                                           size_t obj_buf_sz,
 998                                           const char *obj_name)
 999 {
1000         struct bpf_object *obj;
1001         char *end;
1002
1003         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1004         if (!obj) {
1005                 pr_warn("alloc memory failed for %s\n", path);
1006                 return ERR_PTR(-ENOMEM);
1007         }
1008
1009         strcpy(obj->path, path);
1010         if (obj_name) {
1011                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
1012                 obj->name[sizeof(obj->name) - 1] = 0;
1013         } else {
1014                 /* Using basename() GNU version which doesn't modify arg. */
1015                 strncpy(obj->name, basename((void *)path),
1016                         sizeof(obj->name) - 1);
1017                 end = strchr(obj->name, '.');
1018                 if (end)
1019                         *end = 0;
1020         }
1021
1022         obj->efile.fd = -1;
1023         /*
1024          * Caller of this function should also call
1025          * bpf_object__elf_finish() after data collection to return
1026          * obj_buf to user. If not, we should duplicate the buffer to
1027          * avoid user freeing them before elf finish.
1028          */
1029         obj->efile.obj_buf = obj_buf;
1030         obj->efile.obj_buf_sz = obj_buf_sz;
1031         obj->efile.maps_shndx = -1;
1032         obj->efile.btf_maps_shndx = -1;
1033         obj->efile.data_shndx = -1;
1034         obj->efile.rodata_shndx = -1;
1035         obj->efile.bss_shndx = -1;
1036         obj->efile.st_ops_shndx = -1;
1037         obj->kconfig_map_idx = -1;
1038
1039         obj->kern_version = get_kernel_version();
1040         obj->loaded = false;
1041
1042         INIT_LIST_HEAD(&obj->list);
1043         list_add(&obj->list, &bpf_objects_list);
1044         return obj;
1045 }
1046
1047 static void bpf_object__elf_finish(struct bpf_object *obj)
1048 {
1049         if (!obj_elf_valid(obj))
1050                 return;
1051
1052         if (obj->efile.elf) {
1053                 elf_end(obj->efile.elf);
1054                 obj->efile.elf = NULL;
1055         }
1056         obj->efile.symbols = NULL;
1057         obj->efile.data = NULL;
1058         obj->efile.rodata = NULL;
1059         obj->efile.bss = NULL;
1060         obj->efile.st_ops_data = NULL;
1061
1062         zfree(&obj->efile.reloc_sects);
1063         obj->efile.nr_reloc_sects = 0;
1064         zclose(obj->efile.fd);
1065         obj->efile.obj_buf = NULL;
1066         obj->efile.obj_buf_sz = 0;
1067 }
1068
1069 static int bpf_object__elf_init(struct bpf_object *obj)
1070 {
1071         int err = 0;
1072         GElf_Ehdr *ep;
1073
1074         if (obj_elf_valid(obj)) {
1075                 pr_warn("elf init: internal error\n");
1076                 return -LIBBPF_ERRNO__LIBELF;
1077         }
1078
1079         if (obj->efile.obj_buf_sz > 0) {
1080                 /*
1081                  * obj_buf should have been validated by
1082                  * bpf_object__open_buffer().
1083                  */
1084                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
1085                                             obj->efile.obj_buf_sz);
1086         } else {
1087                 obj->efile.fd = open(obj->path, O_RDONLY);
1088                 if (obj->efile.fd < 0) {
1089                         char errmsg[STRERR_BUFSIZE], *cp;
1090
1091                         err = -errno;
1092                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1093                         pr_warn("failed to open %s: %s\n", obj->path, cp);
1094                         return err;
1095                 }
1096
1097                 obj->efile.elf = elf_begin(obj->efile.fd,
1098                                            LIBBPF_ELF_C_READ_MMAP, NULL);
1099         }
1100
1101         if (!obj->efile.elf) {
1102                 pr_warn("failed to open %s as ELF file\n", obj->path);
1103                 err = -LIBBPF_ERRNO__LIBELF;
1104                 goto errout;
1105         }
1106
1107         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
1108                 pr_warn("failed to get EHDR from %s\n", obj->path);
1109                 err = -LIBBPF_ERRNO__FORMAT;
1110                 goto errout;
1111         }
1112         ep = &obj->efile.ehdr;
1113
1114         /* Old LLVM set e_machine to EM_NONE */
1115         if (ep->e_type != ET_REL ||
1116             (ep->e_machine && ep->e_machine != EM_BPF)) {
1117                 pr_warn("%s is not an eBPF object file\n", obj->path);
1118                 err = -LIBBPF_ERRNO__FORMAT;
1119                 goto errout;
1120         }
1121
1122         return 0;
1123 errout:
1124         bpf_object__elf_finish(obj);
1125         return err;
1126 }
1127
1128 static int bpf_object__check_endianness(struct bpf_object *obj)
1129 {
1130 #if __BYTE_ORDER == __LITTLE_ENDIAN
1131         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
1132                 return 0;
1133 #elif __BYTE_ORDER == __BIG_ENDIAN
1134         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
1135                 return 0;
1136 #else
1137 # error "Unrecognized __BYTE_ORDER__"
1138 #endif
1139         pr_warn("endianness mismatch.\n");
1140         return -LIBBPF_ERRNO__ENDIAN;
1141 }
1142
1143 static int
1144 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1145 {
1146         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
1147         pr_debug("license of %s is %s\n", obj->path, obj->license);
1148         return 0;
1149 }
1150
1151 static int
1152 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1153 {
1154         __u32 kver;
1155
1156         if (size != sizeof(kver)) {
1157                 pr_warn("invalid kver section in %s\n", obj->path);
1158                 return -LIBBPF_ERRNO__FORMAT;
1159         }
1160         memcpy(&kver, data, sizeof(kver));
1161         obj->kern_version = kver;
1162         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1163         return 0;
1164 }
1165
1166 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1167 {
1168         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1169             type == BPF_MAP_TYPE_HASH_OF_MAPS)
1170                 return true;
1171         return false;
1172 }
1173
1174 static int bpf_object_search_section_size(const struct bpf_object *obj,
1175                                           const char *name, size_t *d_size)
1176 {
1177         const GElf_Ehdr *ep = &obj->efile.ehdr;
1178         Elf *elf = obj->efile.elf;
1179         Elf_Scn *scn = NULL;
1180         int idx = 0;
1181
1182         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1183                 const char *sec_name;
1184                 Elf_Data *data;
1185                 GElf_Shdr sh;
1186
1187                 idx++;
1188                 if (gelf_getshdr(scn, &sh) != &sh) {
1189                         pr_warn("failed to get section(%d) header from %s\n",
1190                                 idx, obj->path);
1191                         return -EIO;
1192                 }
1193
1194                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1195                 if (!sec_name) {
1196                         pr_warn("failed to get section(%d) name from %s\n",
1197                                 idx, obj->path);
1198                         return -EIO;
1199                 }
1200
1201                 if (strcmp(name, sec_name))
1202                         continue;
1203
1204                 data = elf_getdata(scn, 0);
1205                 if (!data) {
1206                         pr_warn("failed to get section(%d) data from %s(%s)\n",
1207                                 idx, name, obj->path);
1208                         return -EIO;
1209                 }
1210
1211                 *d_size = data->d_size;
1212                 return 0;
1213         }
1214
1215         return -ENOENT;
1216 }
1217
1218 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
1219                              __u32 *size)
1220 {
1221         int ret = -ENOENT;
1222         size_t d_size;
1223
1224         *size = 0;
1225         if (!name) {
1226                 return -EINVAL;
1227         } else if (!strcmp(name, DATA_SEC)) {
1228                 if (obj->efile.data)
1229                         *size = obj->efile.data->d_size;
1230         } else if (!strcmp(name, BSS_SEC)) {
1231                 if (obj->efile.bss)
1232                         *size = obj->efile.bss->d_size;
1233         } else if (!strcmp(name, RODATA_SEC)) {
1234                 if (obj->efile.rodata)
1235                         *size = obj->efile.rodata->d_size;
1236         } else if (!strcmp(name, STRUCT_OPS_SEC)) {
1237                 if (obj->efile.st_ops_data)
1238                         *size = obj->efile.st_ops_data->d_size;
1239         } else {
1240                 ret = bpf_object_search_section_size(obj, name, &d_size);
1241                 if (!ret)
1242                         *size = d_size;
1243         }
1244
1245         return *size ? 0 : ret;
1246 }
1247
1248 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
1249                                 __u32 *off)
1250 {
1251         Elf_Data *symbols = obj->efile.symbols;
1252         const char *sname;
1253         size_t si;
1254
1255         if (!name || !off)
1256                 return -EINVAL;
1257
1258         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
1259                 GElf_Sym sym;
1260
1261                 if (!gelf_getsym(symbols, si, &sym))
1262                         continue;
1263                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1264                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
1265                         continue;
1266
1267                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1268                                    sym.st_name);
1269                 if (!sname) {
1270                         pr_warn("failed to get sym name string for var %s\n",
1271                                 name);
1272                         return -EIO;
1273                 }
1274                 if (strcmp(name, sname) == 0) {
1275                         *off = sym.st_value;
1276                         return 0;
1277                 }
1278         }
1279
1280         return -ENOENT;
1281 }
1282
1283 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1284 {
1285         struct bpf_map *new_maps;
1286         size_t new_cap;
1287         int i;
1288
1289         if (obj->nr_maps < obj->maps_cap)
1290                 return &obj->maps[obj->nr_maps++];
1291
1292         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1293         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
1294         if (!new_maps) {
1295                 pr_warn("alloc maps for object failed\n");
1296                 return ERR_PTR(-ENOMEM);
1297         }
1298
1299         obj->maps_cap = new_cap;
1300         obj->maps = new_maps;
1301
1302         /* zero out new maps */
1303         memset(obj->maps + obj->nr_maps, 0,
1304                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1305         /*
1306          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1307          * when failure (zclose won't close negative fd)).
1308          */
1309         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1310                 obj->maps[i].fd = -1;
1311                 obj->maps[i].inner_map_fd = -1;
1312         }
1313
1314         return &obj->maps[obj->nr_maps++];
1315 }
1316
1317 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1318 {
1319         long page_sz = sysconf(_SC_PAGE_SIZE);
1320         size_t map_sz;
1321
1322         map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1323         map_sz = roundup(map_sz, page_sz);
1324         return map_sz;
1325 }
1326
1327 static char *internal_map_name(struct bpf_object *obj,
1328                                enum libbpf_map_type type)
1329 {
1330         char map_name[BPF_OBJ_NAME_LEN], *p;
1331         const char *sfx = libbpf_type_to_btf_name[type];
1332         int sfx_len = max((size_t)7, strlen(sfx));
1333         int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
1334                           strlen(obj->name));
1335
1336         snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1337                  sfx_len, libbpf_type_to_btf_name[type]);
1338
1339         /* sanitise map name to characters allowed by kernel */
1340         for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1341                 if (!isalnum(*p) && *p != '_' && *p != '.')
1342                         *p = '_';
1343
1344         return strdup(map_name);
1345 }
1346
1347 static int
1348 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1349                               int sec_idx, void *data, size_t data_sz)
1350 {
1351         struct bpf_map_def *def;
1352         struct bpf_map *map;
1353         int err;
1354
1355         map = bpf_object__add_map(obj);
1356         if (IS_ERR(map))
1357                 return PTR_ERR(map);
1358
1359         map->libbpf_type = type;
1360         map->sec_idx = sec_idx;
1361         map->sec_offset = 0;
1362         map->name = internal_map_name(obj, type);
1363         if (!map->name) {
1364                 pr_warn("failed to alloc map name\n");
1365                 return -ENOMEM;
1366         }
1367
1368         def = &map->def;
1369         def->type = BPF_MAP_TYPE_ARRAY;
1370         def->key_size = sizeof(int);
1371         def->value_size = data_sz;
1372         def->max_entries = 1;
1373         def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1374                          ? BPF_F_RDONLY_PROG : 0;
1375         def->map_flags |= BPF_F_MMAPABLE;
1376
1377         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1378                  map->name, map->sec_idx, map->sec_offset, def->map_flags);
1379
1380         map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1381                            MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1382         if (map->mmaped == MAP_FAILED) {
1383                 err = -errno;
1384                 map->mmaped = NULL;
1385                 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1386                         map->name, err);
1387                 zfree(&map->name);
1388                 return err;
1389         }
1390
1391         if (data)
1392                 memcpy(map->mmaped, data, data_sz);
1393
1394         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1395         return 0;
1396 }
1397
1398 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1399 {
1400         int err;
1401
1402         /*
1403          * Populate obj->maps with libbpf internal maps.
1404          */
1405         if (obj->efile.data_shndx >= 0) {
1406                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1407                                                     obj->efile.data_shndx,
1408                                                     obj->efile.data->d_buf,
1409                                                     obj->efile.data->d_size);
1410                 if (err)
1411                         return err;
1412         }
1413         if (obj->efile.rodata_shndx >= 0) {
1414                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1415                                                     obj->efile.rodata_shndx,
1416                                                     obj->efile.rodata->d_buf,
1417                                                     obj->efile.rodata->d_size);
1418                 if (err)
1419                         return err;
1420         }
1421         if (obj->efile.bss_shndx >= 0) {
1422                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1423                                                     obj->efile.bss_shndx,
1424                                                     NULL,
1425                                                     obj->efile.bss->d_size);
1426                 if (err)
1427                         return err;
1428         }
1429         return 0;
1430 }
1431
1432
1433 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1434                                                const void *name)
1435 {
1436         int i;
1437
1438         for (i = 0; i < obj->nr_extern; i++) {
1439                 if (strcmp(obj->externs[i].name, name) == 0)
1440                         return &obj->externs[i];
1441         }
1442         return NULL;
1443 }
1444
1445 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1446                               char value)
1447 {
1448         switch (ext->kcfg.type) {
1449         case KCFG_BOOL:
1450                 if (value == 'm') {
1451                         pr_warn("extern (kcfg) %s=%c should be tristate or char\n",
1452                                 ext->name, value);
1453                         return -EINVAL;
1454                 }
1455                 *(bool *)ext_val = value == 'y' ? true : false;
1456                 break;
1457         case KCFG_TRISTATE:
1458                 if (value == 'y')
1459                         *(enum libbpf_tristate *)ext_val = TRI_YES;
1460                 else if (value == 'm')
1461                         *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1462                 else /* value == 'n' */
1463                         *(enum libbpf_tristate *)ext_val = TRI_NO;
1464                 break;
1465         case KCFG_CHAR:
1466                 *(char *)ext_val = value;
1467                 break;
1468         case KCFG_UNKNOWN:
1469         case KCFG_INT:
1470         case KCFG_CHAR_ARR:
1471         default:
1472                 pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n",
1473                         ext->name, value);
1474                 return -EINVAL;
1475         }
1476         ext->is_set = true;
1477         return 0;
1478 }
1479
1480 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1481                               const char *value)
1482 {
1483         size_t len;
1484
1485         if (ext->kcfg.type != KCFG_CHAR_ARR) {
1486                 pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value);
1487                 return -EINVAL;
1488         }
1489
1490         len = strlen(value);
1491         if (value[len - 1] != '"') {
1492                 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1493                         ext->name, value);
1494                 return -EINVAL;
1495         }
1496
1497         /* strip quotes */
1498         len -= 2;
1499         if (len >= ext->kcfg.sz) {
1500                 pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1501                         ext->name, value, len, ext->kcfg.sz - 1);
1502                 len = ext->kcfg.sz - 1;
1503         }
1504         memcpy(ext_val, value + 1, len);
1505         ext_val[len] = '\0';
1506         ext->is_set = true;
1507         return 0;
1508 }
1509
1510 static int parse_u64(const char *value, __u64 *res)
1511 {
1512         char *value_end;
1513         int err;
1514
1515         errno = 0;
1516         *res = strtoull(value, &value_end, 0);
1517         if (errno) {
1518                 err = -errno;
1519                 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1520                 return err;
1521         }
1522         if (*value_end) {
1523                 pr_warn("failed to parse '%s' as integer completely\n", value);
1524                 return -EINVAL;
1525         }
1526         return 0;
1527 }
1528
1529 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1530 {
1531         int bit_sz = ext->kcfg.sz * 8;
1532
1533         if (ext->kcfg.sz == 8)
1534                 return true;
1535
1536         /* Validate that value stored in u64 fits in integer of `ext->sz`
1537          * bytes size without any loss of information. If the target integer
1538          * is signed, we rely on the following limits of integer type of
1539          * Y bits and subsequent transformation:
1540          *
1541          *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1542          *            0 <= X + 2^(Y-1) <= 2^Y - 1
1543          *            0 <= X + 2^(Y-1) <  2^Y
1544          *
1545          *  For unsigned target integer, check that all the (64 - Y) bits are
1546          *  zero.
1547          */
1548         if (ext->kcfg.is_signed)
1549                 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1550         else
1551                 return (v >> bit_sz) == 0;
1552 }
1553
1554 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1555                               __u64 value)
1556 {
1557         if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1558                 pr_warn("extern (kcfg) %s=%llu should be integer\n",
1559                         ext->name, (unsigned long long)value);
1560                 return -EINVAL;
1561         }
1562         if (!is_kcfg_value_in_range(ext, value)) {
1563                 pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n",
1564                         ext->name, (unsigned long long)value, ext->kcfg.sz);
1565                 return -ERANGE;
1566         }
1567         switch (ext->kcfg.sz) {
1568                 case 1: *(__u8 *)ext_val = value; break;
1569                 case 2: *(__u16 *)ext_val = value; break;
1570                 case 4: *(__u32 *)ext_val = value; break;
1571                 case 8: *(__u64 *)ext_val = value; break;
1572                 default:
1573                         return -EINVAL;
1574         }
1575         ext->is_set = true;
1576         return 0;
1577 }
1578
1579 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1580                                             char *buf, void *data)
1581 {
1582         struct extern_desc *ext;
1583         char *sep, *value;
1584         int len, err = 0;
1585         void *ext_val;
1586         __u64 num;
1587
1588         if (strncmp(buf, "CONFIG_", 7))
1589                 return 0;
1590
1591         sep = strchr(buf, '=');
1592         if (!sep) {
1593                 pr_warn("failed to parse '%s': no separator\n", buf);
1594                 return -EINVAL;
1595         }
1596
1597         /* Trim ending '\n' */
1598         len = strlen(buf);
1599         if (buf[len - 1] == '\n')
1600                 buf[len - 1] = '\0';
1601         /* Split on '=' and ensure that a value is present. */
1602         *sep = '\0';
1603         if (!sep[1]) {
1604                 *sep = '=';
1605                 pr_warn("failed to parse '%s': no value\n", buf);
1606                 return -EINVAL;
1607         }
1608
1609         ext = find_extern_by_name(obj, buf);
1610         if (!ext || ext->is_set)
1611                 return 0;
1612
1613         ext_val = data + ext->kcfg.data_off;
1614         value = sep + 1;
1615
1616         switch (*value) {
1617         case 'y': case 'n': case 'm':
1618                 err = set_kcfg_value_tri(ext, ext_val, *value);
1619                 break;
1620         case '"':
1621                 err = set_kcfg_value_str(ext, ext_val, value);
1622                 break;
1623         default:
1624                 /* assume integer */
1625                 err = parse_u64(value, &num);
1626                 if (err) {
1627                         pr_warn("extern (kcfg) %s=%s should be integer\n",
1628                                 ext->name, value);
1629                         return err;
1630                 }
1631                 err = set_kcfg_value_num(ext, ext_val, num);
1632                 break;
1633         }
1634         if (err)
1635                 return err;
1636         pr_debug("extern (kcfg) %s=%s\n", ext->name, value);
1637         return 0;
1638 }
1639
1640 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1641 {
1642         char buf[PATH_MAX];
1643         struct utsname uts;
1644         int len, err = 0;
1645         gzFile file;
1646
1647         uname(&uts);
1648         len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1649         if (len < 0)
1650                 return -EINVAL;
1651         else if (len >= PATH_MAX)
1652                 return -ENAMETOOLONG;
1653
1654         /* gzopen also accepts uncompressed files. */
1655         file = gzopen(buf, "r");
1656         if (!file)
1657                 file = gzopen("/proc/config.gz", "r");
1658
1659         if (!file) {
1660                 pr_warn("failed to open system Kconfig\n");
1661                 return -ENOENT;
1662         }
1663
1664         while (gzgets(file, buf, sizeof(buf))) {
1665                 err = bpf_object__process_kconfig_line(obj, buf, data);
1666                 if (err) {
1667                         pr_warn("error parsing system Kconfig line '%s': %d\n",
1668                                 buf, err);
1669                         goto out;
1670                 }
1671         }
1672
1673 out:
1674         gzclose(file);
1675         return err;
1676 }
1677
1678 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1679                                         const char *config, void *data)
1680 {
1681         char buf[PATH_MAX];
1682         int err = 0;
1683         FILE *file;
1684
1685         file = fmemopen((void *)config, strlen(config), "r");
1686         if (!file) {
1687                 err = -errno;
1688                 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1689                 return err;
1690         }
1691
1692         while (fgets(buf, sizeof(buf), file)) {
1693                 err = bpf_object__process_kconfig_line(obj, buf, data);
1694                 if (err) {
1695                         pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1696                                 buf, err);
1697                         break;
1698                 }
1699         }
1700
1701         fclose(file);
1702         return err;
1703 }
1704
1705 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1706 {
1707         struct extern_desc *last_ext = NULL, *ext;
1708         size_t map_sz;
1709         int i, err;
1710
1711         for (i = 0; i < obj->nr_extern; i++) {
1712                 ext = &obj->externs[i];
1713                 if (ext->type == EXT_KCFG)
1714                         last_ext = ext;
1715         }
1716
1717         if (!last_ext)
1718                 return 0;
1719
1720         map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
1721         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1722                                             obj->efile.symbols_shndx,
1723                                             NULL, map_sz);
1724         if (err)
1725                 return err;
1726
1727         obj->kconfig_map_idx = obj->nr_maps - 1;
1728
1729         return 0;
1730 }
1731
1732 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1733 {
1734         Elf_Data *symbols = obj->efile.symbols;
1735         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1736         Elf_Data *data = NULL;
1737         Elf_Scn *scn;
1738
1739         if (obj->efile.maps_shndx < 0)
1740                 return 0;
1741
1742         if (!symbols)
1743                 return -EINVAL;
1744
1745         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
1746         if (scn)
1747                 data = elf_getdata(scn, NULL);
1748         if (!scn || !data) {
1749                 pr_warn("failed to get Elf_Data from map section %d\n",
1750                         obj->efile.maps_shndx);
1751                 return -EINVAL;
1752         }
1753
1754         /*
1755          * Count number of maps. Each map has a name.
1756          * Array of maps is not supported: only the first element is
1757          * considered.
1758          *
1759          * TODO: Detect array of map and report error.
1760          */
1761         nr_syms = symbols->d_size / sizeof(GElf_Sym);
1762         for (i = 0; i < nr_syms; i++) {
1763                 GElf_Sym sym;
1764
1765                 if (!gelf_getsym(symbols, i, &sym))
1766                         continue;
1767                 if (sym.st_shndx != obj->efile.maps_shndx)
1768                         continue;
1769                 nr_maps++;
1770         }
1771         /* Assume equally sized map definitions */
1772         pr_debug("maps in %s: %d maps in %zd bytes\n",
1773                  obj->path, nr_maps, data->d_size);
1774
1775         if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1776                 pr_warn("unable to determine map definition size section %s, %d maps in %zd bytes\n",
1777                         obj->path, nr_maps, data->d_size);
1778                 return -EINVAL;
1779         }
1780         map_def_sz = data->d_size / nr_maps;
1781
1782         /* Fill obj->maps using data in "maps" section.  */
1783         for (i = 0; i < nr_syms; i++) {
1784                 GElf_Sym sym;
1785                 const char *map_name;
1786                 struct bpf_map_def *def;
1787                 struct bpf_map *map;
1788
1789                 if (!gelf_getsym(symbols, i, &sym))
1790                         continue;
1791                 if (sym.st_shndx != obj->efile.maps_shndx)
1792                         continue;
1793
1794                 map = bpf_object__add_map(obj);
1795                 if (IS_ERR(map))
1796                         return PTR_ERR(map);
1797
1798                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1799                                       sym.st_name);
1800                 if (!map_name) {
1801                         pr_warn("failed to get map #%d name sym string for obj %s\n",
1802                                 i, obj->path);
1803                         return -LIBBPF_ERRNO__FORMAT;
1804                 }
1805
1806                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1807                 map->sec_idx = sym.st_shndx;
1808                 map->sec_offset = sym.st_value;
1809                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1810                          map_name, map->sec_idx, map->sec_offset);
1811                 if (sym.st_value + map_def_sz > data->d_size) {
1812                         pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1813                                 obj->path, map_name);
1814                         return -EINVAL;
1815                 }
1816
1817                 map->name = strdup(map_name);
1818                 if (!map->name) {
1819                         pr_warn("failed to alloc map name\n");
1820                         return -ENOMEM;
1821                 }
1822                 pr_debug("map %d is \"%s\"\n", i, map->name);
1823                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1824                 /*
1825                  * If the definition of the map in the object file fits in
1826                  * bpf_map_def, copy it.  Any extra fields in our version
1827                  * of bpf_map_def will default to zero as a result of the
1828                  * calloc above.
1829                  */
1830                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1831                         memcpy(&map->def, def, map_def_sz);
1832                 } else {
1833                         /*
1834                          * Here the map structure being read is bigger than what
1835                          * we expect, truncate if the excess bits are all zero.
1836                          * If they are not zero, reject this map as
1837                          * incompatible.
1838                          */
1839                         char *b;
1840
1841                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1842                              b < ((char *)def) + map_def_sz; b++) {
1843                                 if (*b != 0) {
1844                                         pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1845                                                 obj->path, map_name);
1846                                         if (strict)
1847                                                 return -EINVAL;
1848                                 }
1849                         }
1850                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1851                 }
1852         }
1853         return 0;
1854 }
1855
1856 static const struct btf_type *
1857 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1858 {
1859         const struct btf_type *t = btf__type_by_id(btf, id);
1860
1861         if (res_id)
1862                 *res_id = id;
1863
1864         while (btf_is_mod(t) || btf_is_typedef(t)) {
1865                 if (res_id)
1866                         *res_id = t->type;
1867                 t = btf__type_by_id(btf, t->type);
1868         }
1869
1870         return t;
1871 }
1872
1873 static const struct btf_type *
1874 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
1875 {
1876         const struct btf_type *t;
1877
1878         t = skip_mods_and_typedefs(btf, id, NULL);
1879         if (!btf_is_ptr(t))
1880                 return NULL;
1881
1882         t = skip_mods_and_typedefs(btf, t->type, res_id);
1883
1884         return btf_is_func_proto(t) ? t : NULL;
1885 }
1886
1887 /*
1888  * Fetch integer attribute of BTF map definition. Such attributes are
1889  * represented using a pointer to an array, in which dimensionality of array
1890  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1891  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1892  * type definition, while using only sizeof(void *) space in ELF data section.
1893  */
1894 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1895                               const struct btf_member *m, __u32 *res)
1896 {
1897         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1898         const char *name = btf__name_by_offset(btf, m->name_off);
1899         const struct btf_array *arr_info;
1900         const struct btf_type *arr_t;
1901
1902         if (!btf_is_ptr(t)) {
1903                 pr_warn("map '%s': attr '%s': expected PTR, got %u.\n",
1904                         map_name, name, btf_kind(t));
1905                 return false;
1906         }
1907
1908         arr_t = btf__type_by_id(btf, t->type);
1909         if (!arr_t) {
1910                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1911                         map_name, name, t->type);
1912                 return false;
1913         }
1914         if (!btf_is_array(arr_t)) {
1915                 pr_warn("map '%s': attr '%s': expected ARRAY, got %u.\n",
1916                         map_name, name, btf_kind(arr_t));
1917                 return false;
1918         }
1919         arr_info = btf_array(arr_t);
1920         *res = arr_info->nelems;
1921         return true;
1922 }
1923
1924 static int build_map_pin_path(struct bpf_map *map, const char *path)
1925 {
1926         char buf[PATH_MAX];
1927         int err, len;
1928
1929         if (!path)
1930                 path = "/sys/fs/bpf";
1931
1932         len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1933         if (len < 0)
1934                 return -EINVAL;
1935         else if (len >= PATH_MAX)
1936                 return -ENAMETOOLONG;
1937
1938         err = bpf_map__set_pin_path(map, buf);
1939         if (err)
1940                 return err;
1941
1942         return 0;
1943 }
1944
1945
1946 static int parse_btf_map_def(struct bpf_object *obj,
1947                              struct bpf_map *map,
1948                              const struct btf_type *def,
1949                              bool strict, bool is_inner,
1950                              const char *pin_root_path)
1951 {
1952         const struct btf_type *t;
1953         const struct btf_member *m;
1954         int vlen, i;
1955
1956         vlen = btf_vlen(def);
1957         m = btf_members(def);
1958         for (i = 0; i < vlen; i++, m++) {
1959                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1960
1961                 if (!name) {
1962                         pr_warn("map '%s': invalid field #%d.\n", map->name, i);
1963                         return -EINVAL;
1964                 }
1965                 if (strcmp(name, "type") == 0) {
1966                         if (!get_map_field_int(map->name, obj->btf, m,
1967                                                &map->def.type))
1968                                 return -EINVAL;
1969                         pr_debug("map '%s': found type = %u.\n",
1970                                  map->name, map->def.type);
1971                 } else if (strcmp(name, "max_entries") == 0) {
1972                         if (!get_map_field_int(map->name, obj->btf, m,
1973                                                &map->def.max_entries))
1974                                 return -EINVAL;
1975                         pr_debug("map '%s': found max_entries = %u.\n",
1976                                  map->name, map->def.max_entries);
1977                 } else if (strcmp(name, "map_flags") == 0) {
1978                         if (!get_map_field_int(map->name, obj->btf, m,
1979                                                &map->def.map_flags))
1980                                 return -EINVAL;
1981                         pr_debug("map '%s': found map_flags = %u.\n",
1982                                  map->name, map->def.map_flags);
1983                 } else if (strcmp(name, "numa_node") == 0) {
1984                         if (!get_map_field_int(map->name, obj->btf, m, &map->numa_node))
1985                                 return -EINVAL;
1986                         pr_debug("map '%s': found numa_node = %u.\n", map->name, map->numa_node);
1987                 } else if (strcmp(name, "key_size") == 0) {
1988                         __u32 sz;
1989
1990                         if (!get_map_field_int(map->name, obj->btf, m, &sz))
1991                                 return -EINVAL;
1992                         pr_debug("map '%s': found key_size = %u.\n",
1993                                  map->name, sz);
1994                         if (map->def.key_size && map->def.key_size != sz) {
1995                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
1996                                         map->name, map->def.key_size, sz);
1997                                 return -EINVAL;
1998                         }
1999                         map->def.key_size = sz;
2000                 } else if (strcmp(name, "key") == 0) {
2001                         __s64 sz;
2002
2003                         t = btf__type_by_id(obj->btf, m->type);
2004                         if (!t) {
2005                                 pr_warn("map '%s': key type [%d] not found.\n",
2006                                         map->name, m->type);
2007                                 return -EINVAL;
2008                         }
2009                         if (!btf_is_ptr(t)) {
2010                                 pr_warn("map '%s': key spec is not PTR: %u.\n",
2011                                         map->name, btf_kind(t));
2012                                 return -EINVAL;
2013                         }
2014                         sz = btf__resolve_size(obj->btf, t->type);
2015                         if (sz < 0) {
2016                                 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2017                                         map->name, t->type, (ssize_t)sz);
2018                                 return sz;
2019                         }
2020                         pr_debug("map '%s': found key [%u], sz = %zd.\n",
2021                                  map->name, t->type, (ssize_t)sz);
2022                         if (map->def.key_size && map->def.key_size != sz) {
2023                                 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2024                                         map->name, map->def.key_size, (ssize_t)sz);
2025                                 return -EINVAL;
2026                         }
2027                         map->def.key_size = sz;
2028                         map->btf_key_type_id = t->type;
2029                 } else if (strcmp(name, "value_size") == 0) {
2030                         __u32 sz;
2031
2032                         if (!get_map_field_int(map->name, obj->btf, m, &sz))
2033                                 return -EINVAL;
2034                         pr_debug("map '%s': found value_size = %u.\n",
2035                                  map->name, sz);
2036                         if (map->def.value_size && map->def.value_size != sz) {
2037                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
2038                                         map->name, map->def.value_size, sz);
2039                                 return -EINVAL;
2040                         }
2041                         map->def.value_size = sz;
2042                 } else if (strcmp(name, "value") == 0) {
2043                         __s64 sz;
2044
2045                         t = btf__type_by_id(obj->btf, m->type);
2046                         if (!t) {
2047                                 pr_warn("map '%s': value type [%d] not found.\n",
2048                                         map->name, m->type);
2049                                 return -EINVAL;
2050                         }
2051                         if (!btf_is_ptr(t)) {
2052                                 pr_warn("map '%s': value spec is not PTR: %u.\n",
2053                                         map->name, btf_kind(t));
2054                                 return -EINVAL;
2055                         }
2056                         sz = btf__resolve_size(obj->btf, t->type);
2057                         if (sz < 0) {
2058                                 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2059                                         map->name, t->type, (ssize_t)sz);
2060                                 return sz;
2061                         }
2062                         pr_debug("map '%s': found value [%u], sz = %zd.\n",
2063                                  map->name, t->type, (ssize_t)sz);
2064                         if (map->def.value_size && map->def.value_size != sz) {
2065                                 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2066                                         map->name, map->def.value_size, (ssize_t)sz);
2067                                 return -EINVAL;
2068                         }
2069                         map->def.value_size = sz;
2070                         map->btf_value_type_id = t->type;
2071                 }
2072                 else if (strcmp(name, "values") == 0) {
2073                         int err;
2074
2075                         if (is_inner) {
2076                                 pr_warn("map '%s': multi-level inner maps not supported.\n",
2077                                         map->name);
2078                                 return -ENOTSUP;
2079                         }
2080                         if (i != vlen - 1) {
2081                                 pr_warn("map '%s': '%s' member should be last.\n",
2082                                         map->name, name);
2083                                 return -EINVAL;
2084                         }
2085                         if (!bpf_map_type__is_map_in_map(map->def.type)) {
2086                                 pr_warn("map '%s': should be map-in-map.\n",
2087                                         map->name);
2088                                 return -ENOTSUP;
2089                         }
2090                         if (map->def.value_size && map->def.value_size != 4) {
2091                                 pr_warn("map '%s': conflicting value size %u != 4.\n",
2092                                         map->name, map->def.value_size);
2093                                 return -EINVAL;
2094                         }
2095                         map->def.value_size = 4;
2096                         t = btf__type_by_id(obj->btf, m->type);
2097                         if (!t) {
2098                                 pr_warn("map '%s': map-in-map inner type [%d] not found.\n",
2099                                         map->name, m->type);
2100                                 return -EINVAL;
2101                         }
2102                         if (!btf_is_array(t) || btf_array(t)->nelems) {
2103                                 pr_warn("map '%s': map-in-map inner spec is not a zero-sized array.\n",
2104                                         map->name);
2105                                 return -EINVAL;
2106                         }
2107                         t = skip_mods_and_typedefs(obj->btf, btf_array(t)->type,
2108                                                    NULL);
2109                         if (!btf_is_ptr(t)) {
2110                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %u.\n",
2111                                         map->name, btf_kind(t));
2112                                 return -EINVAL;
2113                         }
2114                         t = skip_mods_and_typedefs(obj->btf, t->type, NULL);
2115                         if (!btf_is_struct(t)) {
2116                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %u.\n",
2117                                         map->name, btf_kind(t));
2118                                 return -EINVAL;
2119                         }
2120
2121                         map->inner_map = calloc(1, sizeof(*map->inner_map));
2122                         if (!map->inner_map)
2123                                 return -ENOMEM;
2124                         map->inner_map->sec_idx = obj->efile.btf_maps_shndx;
2125                         map->inner_map->name = malloc(strlen(map->name) +
2126                                                       sizeof(".inner") + 1);
2127                         if (!map->inner_map->name)
2128                                 return -ENOMEM;
2129                         sprintf(map->inner_map->name, "%s.inner", map->name);
2130
2131                         err = parse_btf_map_def(obj, map->inner_map, t, strict,
2132                                                 true /* is_inner */, NULL);
2133                         if (err)
2134                                 return err;
2135                 } else if (strcmp(name, "pinning") == 0) {
2136                         __u32 val;
2137                         int err;
2138
2139                         if (is_inner) {
2140                                 pr_debug("map '%s': inner def can't be pinned.\n",
2141                                          map->name);
2142                                 return -EINVAL;
2143                         }
2144                         if (!get_map_field_int(map->name, obj->btf, m, &val))
2145                                 return -EINVAL;
2146                         pr_debug("map '%s': found pinning = %u.\n",
2147                                  map->name, val);
2148
2149                         if (val != LIBBPF_PIN_NONE &&
2150                             val != LIBBPF_PIN_BY_NAME) {
2151                                 pr_warn("map '%s': invalid pinning value %u.\n",
2152                                         map->name, val);
2153                                 return -EINVAL;
2154                         }
2155                         if (val == LIBBPF_PIN_BY_NAME) {
2156                                 err = build_map_pin_path(map, pin_root_path);
2157                                 if (err) {
2158                                         pr_warn("map '%s': couldn't build pin path.\n",
2159                                                 map->name);
2160                                         return err;
2161                                 }
2162                         }
2163                 } else {
2164                         if (strict) {
2165                                 pr_warn("map '%s': unknown field '%s'.\n",
2166                                         map->name, name);
2167                                 return -ENOTSUP;
2168                         }
2169                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
2170                                  map->name, name);
2171                 }
2172         }
2173
2174         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
2175                 pr_warn("map '%s': map type isn't specified.\n", map->name);
2176                 return -EINVAL;
2177         }
2178
2179         return 0;
2180 }
2181
2182 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2183                                          const struct btf_type *sec,
2184                                          int var_idx, int sec_idx,
2185                                          const Elf_Data *data, bool strict,
2186                                          const char *pin_root_path)
2187 {
2188         const struct btf_type *var, *def;
2189         const struct btf_var_secinfo *vi;
2190         const struct btf_var *var_extra;
2191         const char *map_name;
2192         struct bpf_map *map;
2193
2194         vi = btf_var_secinfos(sec) + var_idx;
2195         var = btf__type_by_id(obj->btf, vi->type);
2196         var_extra = btf_var(var);
2197         map_name = btf__name_by_offset(obj->btf, var->name_off);
2198
2199         if (map_name == NULL || map_name[0] == '\0') {
2200                 pr_warn("map #%d: empty name.\n", var_idx);
2201                 return -EINVAL;
2202         }
2203         if ((__u64)vi->offset + vi->size > data->d_size) {
2204                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2205                 return -EINVAL;
2206         }
2207         if (!btf_is_var(var)) {
2208                 pr_warn("map '%s': unexpected var kind %u.\n",
2209                         map_name, btf_kind(var));
2210                 return -EINVAL;
2211         }
2212         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2213             var_extra->linkage != BTF_VAR_STATIC) {
2214                 pr_warn("map '%s': unsupported var linkage %u.\n",
2215                         map_name, var_extra->linkage);
2216                 return -EOPNOTSUPP;
2217         }
2218
2219         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2220         if (!btf_is_struct(def)) {
2221                 pr_warn("map '%s': unexpected def kind %u.\n",
2222                         map_name, btf_kind(var));
2223                 return -EINVAL;
2224         }
2225         if (def->size > vi->size) {
2226                 pr_warn("map '%s': invalid def size.\n", map_name);
2227                 return -EINVAL;
2228         }
2229
2230         map = bpf_object__add_map(obj);
2231         if (IS_ERR(map))
2232                 return PTR_ERR(map);
2233         map->name = strdup(map_name);
2234         if (!map->name) {
2235                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2236                 return -ENOMEM;
2237         }
2238         map->libbpf_type = LIBBPF_MAP_UNSPEC;
2239         map->def.type = BPF_MAP_TYPE_UNSPEC;
2240         map->sec_idx = sec_idx;
2241         map->sec_offset = vi->offset;
2242         map->btf_var_idx = var_idx;
2243         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2244                  map_name, map->sec_idx, map->sec_offset);
2245
2246         return parse_btf_map_def(obj, map, def, strict, false, pin_root_path);
2247 }
2248
2249 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2250                                           const char *pin_root_path)
2251 {
2252         const struct btf_type *sec = NULL;
2253         int nr_types, i, vlen, err;
2254         const struct btf_type *t;
2255         const char *name;
2256         Elf_Data *data;
2257         Elf_Scn *scn;
2258
2259         if (obj->efile.btf_maps_shndx < 0)
2260                 return 0;
2261
2262         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
2263         if (scn)
2264                 data = elf_getdata(scn, NULL);
2265         if (!scn || !data) {
2266                 pr_warn("failed to get Elf_Data from map section %d (%s)\n",
2267                         obj->efile.btf_maps_shndx, MAPS_ELF_SEC);
2268                 return -EINVAL;
2269         }
2270
2271         nr_types = btf__get_nr_types(obj->btf);
2272         for (i = 1; i <= nr_types; i++) {
2273                 t = btf__type_by_id(obj->btf, i);
2274                 if (!btf_is_datasec(t))
2275                         continue;
2276                 name = btf__name_by_offset(obj->btf, t->name_off);
2277                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2278                         sec = t;
2279                         obj->efile.btf_maps_sec_btf_id = i;
2280                         break;
2281                 }
2282         }
2283
2284         if (!sec) {
2285                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2286                 return -ENOENT;
2287         }
2288
2289         vlen = btf_vlen(sec);
2290         for (i = 0; i < vlen; i++) {
2291                 err = bpf_object__init_user_btf_map(obj, sec, i,
2292                                                     obj->efile.btf_maps_shndx,
2293                                                     data, strict,
2294                                                     pin_root_path);
2295                 if (err)
2296                         return err;
2297         }
2298
2299         return 0;
2300 }
2301
2302 static int bpf_object__init_maps(struct bpf_object *obj,
2303                                  const struct bpf_object_open_opts *opts)
2304 {
2305         const char *pin_root_path;
2306         bool strict;
2307         int err;
2308
2309         strict = !OPTS_GET(opts, relaxed_maps, false);
2310         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2311
2312         err = bpf_object__init_user_maps(obj, strict);
2313         err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2314         err = err ?: bpf_object__init_global_data_maps(obj);
2315         err = err ?: bpf_object__init_kconfig_map(obj);
2316         err = err ?: bpf_object__init_struct_ops_maps(obj);
2317         if (err)
2318                 return err;
2319
2320         return 0;
2321 }
2322
2323 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2324 {
2325         Elf_Scn *scn;
2326         GElf_Shdr sh;
2327
2328         scn = elf_getscn(obj->efile.elf, idx);
2329         if (!scn)
2330                 return false;
2331
2332         if (gelf_getshdr(scn, &sh) != &sh)
2333                 return false;
2334
2335         if (sh.sh_flags & SHF_EXECINSTR)
2336                 return true;
2337
2338         return false;
2339 }
2340
2341 static bool btf_needs_sanitization(struct bpf_object *obj)
2342 {
2343         bool has_func_global = obj->caps.btf_func_global;
2344         bool has_datasec = obj->caps.btf_datasec;
2345         bool has_func = obj->caps.btf_func;
2346
2347         return !has_func || !has_datasec || !has_func_global;
2348 }
2349
2350 static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2351 {
2352         bool has_func_global = obj->caps.btf_func_global;
2353         bool has_datasec = obj->caps.btf_datasec;
2354         bool has_func = obj->caps.btf_func;
2355         struct btf_type *t;
2356         int i, j, vlen;
2357
2358         for (i = 1; i <= btf__get_nr_types(btf); i++) {
2359                 t = (struct btf_type *)btf__type_by_id(btf, i);
2360
2361                 if (!has_datasec && btf_is_var(t)) {
2362                         /* replace VAR with INT */
2363                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2364                         /*
2365                          * using size = 1 is the safest choice, 4 will be too
2366                          * big and cause kernel BTF validation failure if
2367                          * original variable took less than 4 bytes
2368                          */
2369                         t->size = 1;
2370                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2371                 } else if (!has_datasec && btf_is_datasec(t)) {
2372                         /* replace DATASEC with STRUCT */
2373                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
2374                         struct btf_member *m = btf_members(t);
2375                         struct btf_type *vt;
2376                         char *name;
2377
2378                         name = (char *)btf__name_by_offset(btf, t->name_off);
2379                         while (*name) {
2380                                 if (*name == '.')
2381                                         *name = '_';
2382                                 name++;
2383                         }
2384
2385                         vlen = btf_vlen(t);
2386                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2387                         for (j = 0; j < vlen; j++, v++, m++) {
2388                                 /* order of field assignments is important */
2389                                 m->offset = v->offset * 8;
2390                                 m->type = v->type;
2391                                 /* preserve variable name as member name */
2392                                 vt = (void *)btf__type_by_id(btf, v->type);
2393                                 m->name_off = vt->name_off;
2394                         }
2395                 } else if (!has_func && btf_is_func_proto(t)) {
2396                         /* replace FUNC_PROTO with ENUM */
2397                         vlen = btf_vlen(t);
2398                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2399                         t->size = sizeof(__u32); /* kernel enforced */
2400                 } else if (!has_func && btf_is_func(t)) {
2401                         /* replace FUNC with TYPEDEF */
2402                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2403                 } else if (!has_func_global && btf_is_func(t)) {
2404                         /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2405                         t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2406                 }
2407         }
2408 }
2409
2410 static bool libbpf_needs_btf(const struct bpf_object *obj)
2411 {
2412         return obj->efile.btf_maps_shndx >= 0 ||
2413                obj->efile.st_ops_shndx >= 0 ||
2414                obj->nr_extern > 0;
2415 }
2416
2417 static bool kernel_needs_btf(const struct bpf_object *obj)
2418 {
2419         return obj->efile.st_ops_shndx >= 0;
2420 }
2421
2422 static int bpf_object__init_btf(struct bpf_object *obj,
2423                                 Elf_Data *btf_data,
2424                                 Elf_Data *btf_ext_data)
2425 {
2426         int err = -ENOENT;
2427
2428         if (btf_data) {
2429                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2430                 if (IS_ERR(obj->btf)) {
2431                         err = PTR_ERR(obj->btf);
2432                         obj->btf = NULL;
2433                         pr_warn("Error loading ELF section %s: %d.\n",
2434                                 BTF_ELF_SEC, err);
2435                         goto out;
2436                 }
2437                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2438                 btf__set_pointer_size(obj->btf, 8);
2439                 err = 0;
2440         }
2441         if (btf_ext_data) {
2442                 if (!obj->btf) {
2443                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2444                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2445                         goto out;
2446                 }
2447                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
2448                                             btf_ext_data->d_size);
2449                 if (IS_ERR(obj->btf_ext)) {
2450                         pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
2451                                 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
2452                         obj->btf_ext = NULL;
2453                         goto out;
2454                 }
2455         }
2456 out:
2457         if (err && libbpf_needs_btf(obj)) {
2458                 pr_warn("BTF is required, but is missing or corrupted.\n");
2459                 return err;
2460         }
2461         return 0;
2462 }
2463
2464 static int bpf_object__finalize_btf(struct bpf_object *obj)
2465 {
2466         int err;
2467
2468         if (!obj->btf)
2469                 return 0;
2470
2471         err = btf__finalize_data(obj, obj->btf);
2472         if (err) {
2473                 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2474                 return err;
2475         }
2476
2477         return 0;
2478 }
2479
2480 static inline bool libbpf_prog_needs_vmlinux_btf(struct bpf_program *prog)
2481 {
2482         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2483             prog->type == BPF_PROG_TYPE_LSM)
2484                 return true;
2485
2486         /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2487          * also need vmlinux BTF
2488          */
2489         if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2490                 return true;
2491
2492         return false;
2493 }
2494
2495 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj)
2496 {
2497         bool need_vmlinux_btf = false;
2498         struct bpf_program *prog;
2499         int err;
2500
2501         /* CO-RE relocations need kernel BTF */
2502         if (obj->btf_ext && obj->btf_ext->field_reloc_info.len)
2503                 need_vmlinux_btf = true;
2504
2505         bpf_object__for_each_program(prog, obj) {
2506                 if (!prog->load)
2507                         continue;
2508                 if (libbpf_prog_needs_vmlinux_btf(prog)) {
2509                         need_vmlinux_btf = true;
2510                         break;
2511                 }
2512         }
2513
2514         if (!need_vmlinux_btf)
2515                 return 0;
2516
2517         obj->btf_vmlinux = libbpf_find_kernel_btf();
2518         if (IS_ERR(obj->btf_vmlinux)) {
2519                 err = PTR_ERR(obj->btf_vmlinux);
2520                 pr_warn("Error loading vmlinux BTF: %d\n", err);
2521                 obj->btf_vmlinux = NULL;
2522                 return err;
2523         }
2524         return 0;
2525 }
2526
2527 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2528 {
2529         struct btf *kern_btf = obj->btf;
2530         bool btf_mandatory, sanitize;
2531         int err = 0;
2532
2533         if (!obj->btf)
2534                 return 0;
2535
2536         sanitize = btf_needs_sanitization(obj);
2537         if (sanitize) {
2538                 const void *raw_data;
2539                 __u32 sz;
2540
2541                 /* clone BTF to sanitize a copy and leave the original intact */
2542                 raw_data = btf__get_raw_data(obj->btf, &sz);
2543                 kern_btf = btf__new(raw_data, sz);
2544                 if (IS_ERR(kern_btf))
2545                         return PTR_ERR(kern_btf);
2546
2547                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2548                 btf__set_pointer_size(obj->btf, 8);
2549                 bpf_object__sanitize_btf(obj, kern_btf);
2550         }
2551
2552         err = btf__load(kern_btf);
2553         if (sanitize) {
2554                 if (!err) {
2555                         /* move fd to libbpf's BTF */
2556                         btf__set_fd(obj->btf, btf__fd(kern_btf));
2557                         btf__set_fd(kern_btf, -1);
2558                 }
2559                 btf__free(kern_btf);
2560         }
2561         if (err) {
2562                 btf_mandatory = kernel_needs_btf(obj);
2563                 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
2564                         btf_mandatory ? "BTF is mandatory, can't proceed."
2565                                       : "BTF is optional, ignoring.");
2566                 if (!btf_mandatory)
2567                         err = 0;
2568         }
2569         return err;
2570 }
2571
2572 static int bpf_object__elf_collect(struct bpf_object *obj)
2573 {
2574         Elf *elf = obj->efile.elf;
2575         GElf_Ehdr *ep = &obj->efile.ehdr;
2576         Elf_Data *btf_ext_data = NULL;
2577         Elf_Data *btf_data = NULL;
2578         Elf_Scn *scn = NULL;
2579         int idx = 0, err = 0;
2580
2581         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
2582         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
2583                 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
2584                 return -LIBBPF_ERRNO__FORMAT;
2585         }
2586
2587         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2588                 char *name;
2589                 GElf_Shdr sh;
2590                 Elf_Data *data;
2591
2592                 idx++;
2593                 if (gelf_getshdr(scn, &sh) != &sh) {
2594                         pr_warn("failed to get section(%d) header from %s\n",
2595                                 idx, obj->path);
2596                         return -LIBBPF_ERRNO__FORMAT;
2597                 }
2598
2599                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
2600                 if (!name) {
2601                         pr_warn("failed to get section(%d) name from %s\n",
2602                                 idx, obj->path);
2603                         return -LIBBPF_ERRNO__FORMAT;
2604                 }
2605
2606                 data = elf_getdata(scn, 0);
2607                 if (!data) {
2608                         pr_warn("failed to get section(%d) data from %s(%s)\n",
2609                                 idx, name, obj->path);
2610                         return -LIBBPF_ERRNO__FORMAT;
2611                 }
2612                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2613                          idx, name, (unsigned long)data->d_size,
2614                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
2615                          (int)sh.sh_type);
2616
2617                 if (strcmp(name, "license") == 0) {
2618                         err = bpf_object__init_license(obj,
2619                                                        data->d_buf,
2620                                                        data->d_size);
2621                         if (err)
2622                                 return err;
2623                 } else if (strcmp(name, "version") == 0) {
2624                         err = bpf_object__init_kversion(obj,
2625                                                         data->d_buf,
2626                                                         data->d_size);
2627                         if (err)
2628                                 return err;
2629                 } else if (strcmp(name, "maps") == 0) {
2630                         obj->efile.maps_shndx = idx;
2631                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2632                         obj->efile.btf_maps_shndx = idx;
2633                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2634                         btf_data = data;
2635                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
2636                         btf_ext_data = data;
2637                 } else if (sh.sh_type == SHT_SYMTAB) {
2638                         if (obj->efile.symbols) {
2639                                 pr_warn("bpf: multiple SYMTAB in %s\n",
2640                                         obj->path);
2641                                 return -LIBBPF_ERRNO__FORMAT;
2642                         }
2643                         obj->efile.symbols = data;
2644                         obj->efile.symbols_shndx = idx;
2645                         obj->efile.strtabidx = sh.sh_link;
2646                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2647                         if (sh.sh_flags & SHF_EXECINSTR) {
2648                                 if (strcmp(name, ".text") == 0)
2649                                         obj->efile.text_shndx = idx;
2650                                 err = bpf_object__add_program(obj, data->d_buf,
2651                                                               data->d_size,
2652                                                               name, idx);
2653                                 if (err) {
2654                                         char errmsg[STRERR_BUFSIZE];
2655                                         char *cp;
2656
2657                                         cp = libbpf_strerror_r(-err, errmsg,
2658                                                                sizeof(errmsg));
2659                                         pr_warn("failed to alloc program %s (%s): %s",
2660                                                 name, obj->path, cp);
2661                                         return err;
2662                                 }
2663                         } else if (strcmp(name, DATA_SEC) == 0) {
2664                                 obj->efile.data = data;
2665                                 obj->efile.data_shndx = idx;
2666                         } else if (strcmp(name, RODATA_SEC) == 0) {
2667                                 obj->efile.rodata = data;
2668                                 obj->efile.rodata_shndx = idx;
2669                         } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
2670                                 obj->efile.st_ops_data = data;
2671                                 obj->efile.st_ops_shndx = idx;
2672                         } else {
2673                                 pr_debug("skip section(%d) %s\n", idx, name);
2674                         }
2675                 } else if (sh.sh_type == SHT_REL) {
2676                         int nr_sects = obj->efile.nr_reloc_sects;
2677                         void *sects = obj->efile.reloc_sects;
2678                         int sec = sh.sh_info; /* points to other section */
2679
2680                         /* Only do relo for section with exec instructions */
2681                         if (!section_have_execinstr(obj, sec) &&
2682                             strcmp(name, ".rel" STRUCT_OPS_SEC) &&
2683                             strcmp(name, ".rel" MAPS_ELF_SEC)) {
2684                                 pr_debug("skip relo %s(%d) for section(%d)\n",
2685                                          name, idx, sec);
2686                                 continue;
2687                         }
2688
2689                         sects = reallocarray(sects, nr_sects + 1,
2690                                              sizeof(*obj->efile.reloc_sects));
2691                         if (!sects) {
2692                                 pr_warn("reloc_sects realloc failed\n");
2693                                 return -ENOMEM;
2694                         }
2695
2696                         obj->efile.reloc_sects = sects;
2697                         obj->efile.nr_reloc_sects++;
2698
2699                         obj->efile.reloc_sects[nr_sects].shdr = sh;
2700                         obj->efile.reloc_sects[nr_sects].data = data;
2701                 } else if (sh.sh_type == SHT_NOBITS &&
2702                            strcmp(name, BSS_SEC) == 0) {
2703                         obj->efile.bss = data;
2704                         obj->efile.bss_shndx = idx;
2705                 } else {
2706                         pr_debug("skip section(%d) %s\n", idx, name);
2707                 }
2708         }
2709
2710         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
2711                 pr_warn("Corrupted ELF file: index of strtab invalid\n");
2712                 return -LIBBPF_ERRNO__FORMAT;
2713         }
2714         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
2715 }
2716
2717 static bool sym_is_extern(const GElf_Sym *sym)
2718 {
2719         int bind = GELF_ST_BIND(sym->st_info);
2720         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2721         return sym->st_shndx == SHN_UNDEF &&
2722                (bind == STB_GLOBAL || bind == STB_WEAK) &&
2723                GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2724 }
2725
2726 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2727 {
2728         const struct btf_type *t;
2729         const char *var_name;
2730         int i, n;
2731
2732         if (!btf)
2733                 return -ESRCH;
2734
2735         n = btf__get_nr_types(btf);
2736         for (i = 1; i <= n; i++) {
2737                 t = btf__type_by_id(btf, i);
2738
2739                 if (!btf_is_var(t))
2740                         continue;
2741
2742                 var_name = btf__name_by_offset(btf, t->name_off);
2743                 if (strcmp(var_name, ext_name))
2744                         continue;
2745
2746                 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2747                         return -EINVAL;
2748
2749                 return i;
2750         }
2751
2752         return -ENOENT;
2753 }
2754
2755 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
2756         const struct btf_var_secinfo *vs;
2757         const struct btf_type *t;
2758         int i, j, n;
2759
2760         if (!btf)
2761                 return -ESRCH;
2762
2763         n = btf__get_nr_types(btf);
2764         for (i = 1; i <= n; i++) {
2765                 t = btf__type_by_id(btf, i);
2766
2767                 if (!btf_is_datasec(t))
2768                         continue;
2769
2770                 vs = btf_var_secinfos(t);
2771                 for (j = 0; j < btf_vlen(t); j++, vs++) {
2772                         if (vs->type == ext_btf_id)
2773                                 return i;
2774                 }
2775         }
2776
2777         return -ENOENT;
2778 }
2779
2780 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
2781                                      bool *is_signed)
2782 {
2783         const struct btf_type *t;
2784         const char *name;
2785
2786         t = skip_mods_and_typedefs(btf, id, NULL);
2787         name = btf__name_by_offset(btf, t->name_off);
2788
2789         if (is_signed)
2790                 *is_signed = false;
2791         switch (btf_kind(t)) {
2792         case BTF_KIND_INT: {
2793                 int enc = btf_int_encoding(t);
2794
2795                 if (enc & BTF_INT_BOOL)
2796                         return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
2797                 if (is_signed)
2798                         *is_signed = enc & BTF_INT_SIGNED;
2799                 if (t->size == 1)
2800                         return KCFG_CHAR;
2801                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
2802                         return KCFG_UNKNOWN;
2803                 return KCFG_INT;
2804         }
2805         case BTF_KIND_ENUM:
2806                 if (t->size != 4)
2807                         return KCFG_UNKNOWN;
2808                 if (strcmp(name, "libbpf_tristate"))
2809                         return KCFG_UNKNOWN;
2810                 return KCFG_TRISTATE;
2811         case BTF_KIND_ARRAY:
2812                 if (btf_array(t)->nelems == 0)
2813                         return KCFG_UNKNOWN;
2814                 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
2815                         return KCFG_UNKNOWN;
2816                 return KCFG_CHAR_ARR;
2817         default:
2818                 return KCFG_UNKNOWN;
2819         }
2820 }
2821
2822 static int cmp_externs(const void *_a, const void *_b)
2823 {
2824         const struct extern_desc *a = _a;
2825         const struct extern_desc *b = _b;
2826
2827         if (a->type != b->type)
2828                 return a->type < b->type ? -1 : 1;
2829
2830         if (a->type == EXT_KCFG) {
2831                 /* descending order by alignment requirements */
2832                 if (a->kcfg.align != b->kcfg.align)
2833                         return a->kcfg.align > b->kcfg.align ? -1 : 1;
2834                 /* ascending order by size, within same alignment class */
2835                 if (a->kcfg.sz != b->kcfg.sz)
2836                         return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
2837         }
2838
2839         /* resolve ties by name */
2840         return strcmp(a->name, b->name);
2841 }
2842
2843 static int find_int_btf_id(const struct btf *btf)
2844 {
2845         const struct btf_type *t;
2846         int i, n;
2847
2848         n = btf__get_nr_types(btf);
2849         for (i = 1; i <= n; i++) {
2850                 t = btf__type_by_id(btf, i);
2851
2852                 if (btf_is_int(t) && btf_int_bits(t) == 32)
2853                         return i;
2854         }
2855
2856         return 0;
2857 }
2858
2859 static int bpf_object__collect_externs(struct bpf_object *obj)
2860 {
2861         struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
2862         const struct btf_type *t;
2863         struct extern_desc *ext;
2864         int i, n, off;
2865         const char *ext_name, *sec_name;
2866         Elf_Scn *scn;
2867         GElf_Shdr sh;
2868
2869         if (!obj->efile.symbols)
2870                 return 0;
2871
2872         scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
2873         if (!scn)
2874                 return -LIBBPF_ERRNO__FORMAT;
2875         if (gelf_getshdr(scn, &sh) != &sh)
2876                 return -LIBBPF_ERRNO__FORMAT;
2877         n = sh.sh_size / sh.sh_entsize;
2878
2879         pr_debug("looking for externs among %d symbols...\n", n);
2880         for (i = 0; i < n; i++) {
2881                 GElf_Sym sym;
2882
2883                 if (!gelf_getsym(obj->efile.symbols, i, &sym))
2884                         return -LIBBPF_ERRNO__FORMAT;
2885                 if (!sym_is_extern(&sym))
2886                         continue;
2887                 ext_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2888                                       sym.st_name);
2889                 if (!ext_name || !ext_name[0])
2890                         continue;
2891
2892                 ext = obj->externs;
2893                 ext = reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
2894                 if (!ext)
2895                         return -ENOMEM;
2896                 obj->externs = ext;
2897                 ext = &ext[obj->nr_extern];
2898                 memset(ext, 0, sizeof(*ext));
2899                 obj->nr_extern++;
2900
2901                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
2902                 if (ext->btf_id <= 0) {
2903                         pr_warn("failed to find BTF for extern '%s': %d\n",
2904                                 ext_name, ext->btf_id);
2905                         return ext->btf_id;
2906                 }
2907                 t = btf__type_by_id(obj->btf, ext->btf_id);
2908                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
2909                 ext->sym_idx = i;
2910                 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
2911
2912                 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
2913                 if (ext->sec_btf_id <= 0) {
2914                         pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
2915                                 ext_name, ext->btf_id, ext->sec_btf_id);
2916                         return ext->sec_btf_id;
2917                 }
2918                 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
2919                 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
2920
2921                 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
2922                         kcfg_sec = sec;
2923                         ext->type = EXT_KCFG;
2924                         ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
2925                         if (ext->kcfg.sz <= 0) {
2926                                 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
2927                                         ext_name, ext->kcfg.sz);
2928                                 return ext->kcfg.sz;
2929                         }
2930                         ext->kcfg.align = btf__align_of(obj->btf, t->type);
2931                         if (ext->kcfg.align <= 0) {
2932                                 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
2933                                         ext_name, ext->kcfg.align);
2934                                 return -EINVAL;
2935                         }
2936                         ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
2937                                                         &ext->kcfg.is_signed);
2938                         if (ext->kcfg.type == KCFG_UNKNOWN) {
2939                                 pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
2940                                 return -ENOTSUP;
2941                         }
2942                 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
2943                         const struct btf_type *vt;
2944
2945                         ksym_sec = sec;
2946                         ext->type = EXT_KSYM;
2947
2948                         vt = skip_mods_and_typedefs(obj->btf, t->type, NULL);
2949                         if (!btf_is_void(vt)) {
2950                                 pr_warn("extern (ksym) '%s' is not typeless (void)\n", ext_name);
2951                                 return -ENOTSUP;
2952                         }
2953                 } else {
2954                         pr_warn("unrecognized extern section '%s'\n", sec_name);
2955                         return -ENOTSUP;
2956                 }
2957         }
2958         pr_debug("collected %d externs total\n", obj->nr_extern);
2959
2960         if (!obj->nr_extern)
2961                 return 0;
2962
2963         /* sort externs by type, for kcfg ones also by (align, size, name) */
2964         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
2965
2966         /* for .ksyms section, we need to turn all externs into allocated
2967          * variables in BTF to pass kernel verification; we do this by
2968          * pretending that each extern is a 8-byte variable
2969          */
2970         if (ksym_sec) {
2971                 /* find existing 4-byte integer type in BTF to use for fake
2972                  * extern variables in DATASEC
2973                  */
2974                 int int_btf_id = find_int_btf_id(obj->btf);
2975
2976                 for (i = 0; i < obj->nr_extern; i++) {
2977                         ext = &obj->externs[i];
2978                         if (ext->type != EXT_KSYM)
2979                                 continue;
2980                         pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
2981                                  i, ext->sym_idx, ext->name);
2982                 }
2983
2984                 sec = ksym_sec;
2985                 n = btf_vlen(sec);
2986                 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
2987                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
2988                         struct btf_type *vt;
2989
2990                         vt = (void *)btf__type_by_id(obj->btf, vs->type);
2991                         ext_name = btf__name_by_offset(obj->btf, vt->name_off);
2992                         ext = find_extern_by_name(obj, ext_name);
2993                         if (!ext) {
2994                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
2995                                         ext_name);
2996                                 return -ESRCH;
2997                         }
2998                         btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
2999                         vt->type = int_btf_id;
3000                         vs->offset = off;
3001                         vs->size = sizeof(int);
3002                 }
3003                 sec->size = off;
3004         }
3005
3006         if (kcfg_sec) {
3007                 sec = kcfg_sec;
3008                 /* for kcfg externs calculate their offsets within a .kconfig map */
3009                 off = 0;
3010                 for (i = 0; i < obj->nr_extern; i++) {
3011                         ext = &obj->externs[i];
3012                         if (ext->type != EXT_KCFG)
3013                                 continue;
3014
3015                         ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3016                         off = ext->kcfg.data_off + ext->kcfg.sz;
3017                         pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3018                                  i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3019                 }
3020                 sec->size = off;
3021                 n = btf_vlen(sec);
3022                 for (i = 0; i < n; i++) {
3023                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3024
3025                         t = btf__type_by_id(obj->btf, vs->type);
3026                         ext_name = btf__name_by_offset(obj->btf, t->name_off);
3027                         ext = find_extern_by_name(obj, ext_name);
3028                         if (!ext) {
3029                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
3030                                         ext_name);
3031                                 return -ESRCH;
3032                         }
3033                         btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3034                         vs->offset = ext->kcfg.data_off;
3035                 }
3036         }
3037         return 0;
3038 }
3039
3040 static struct bpf_program *
3041 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
3042 {
3043         struct bpf_program *prog;
3044         size_t i;
3045
3046         for (i = 0; i < obj->nr_programs; i++) {
3047                 prog = &obj->programs[i];
3048                 if (prog->idx == idx)
3049                         return prog;
3050         }
3051         return NULL;
3052 }
3053
3054 struct bpf_program *
3055 bpf_object__find_program_by_title(const struct bpf_object *obj,
3056                                   const char *title)
3057 {
3058         struct bpf_program *pos;
3059
3060         bpf_object__for_each_program(pos, obj) {
3061                 if (pos->section_name && !strcmp(pos->section_name, title))
3062                         return pos;
3063         }
3064         return NULL;
3065 }
3066
3067 struct bpf_program *
3068 bpf_object__find_program_by_name(const struct bpf_object *obj,
3069                                  const char *name)
3070 {
3071         struct bpf_program *prog;
3072
3073         bpf_object__for_each_program(prog, obj) {
3074                 if (!strcmp(prog->name, name))
3075                         return prog;
3076         }
3077         return NULL;
3078 }
3079
3080 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3081                                       int shndx)
3082 {
3083         return shndx == obj->efile.data_shndx ||
3084                shndx == obj->efile.bss_shndx ||
3085                shndx == obj->efile.rodata_shndx;
3086 }
3087
3088 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3089                                       int shndx)
3090 {
3091         return shndx == obj->efile.maps_shndx ||
3092                shndx == obj->efile.btf_maps_shndx;
3093 }
3094
3095 static enum libbpf_map_type
3096 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3097 {
3098         if (shndx == obj->efile.data_shndx)
3099                 return LIBBPF_MAP_DATA;
3100         else if (shndx == obj->efile.bss_shndx)
3101                 return LIBBPF_MAP_BSS;
3102         else if (shndx == obj->efile.rodata_shndx)
3103                 return LIBBPF_MAP_RODATA;
3104         else if (shndx == obj->efile.symbols_shndx)
3105                 return LIBBPF_MAP_KCONFIG;
3106         else
3107                 return LIBBPF_MAP_UNSPEC;
3108 }
3109
3110 static int bpf_program__record_reloc(struct bpf_program *prog,
3111                                      struct reloc_desc *reloc_desc,
3112                                      __u32 insn_idx, const char *name,
3113                                      const GElf_Sym *sym, const GElf_Rel *rel)
3114 {
3115         struct bpf_insn *insn = &prog->insns[insn_idx];
3116         size_t map_idx, nr_maps = prog->obj->nr_maps;
3117         struct bpf_object *obj = prog->obj;
3118         __u32 shdr_idx = sym->st_shndx;
3119         enum libbpf_map_type type;
3120         struct bpf_map *map;
3121
3122         /* sub-program call relocation */
3123         if (insn->code == (BPF_JMP | BPF_CALL)) {
3124                 if (insn->src_reg != BPF_PSEUDO_CALL) {
3125                         pr_warn("incorrect bpf_call opcode\n");
3126                         return -LIBBPF_ERRNO__RELOC;
3127                 }
3128                 /* text_shndx can be 0, if no default "main" program exists */
3129                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3130                         pr_warn("bad call relo against section %u\n", shdr_idx);
3131                         return -LIBBPF_ERRNO__RELOC;
3132                 }
3133                 if (sym->st_value % 8) {
3134                         pr_warn("bad call relo offset: %zu\n",
3135                                 (size_t)sym->st_value);
3136                         return -LIBBPF_ERRNO__RELOC;
3137                 }
3138                 reloc_desc->type = RELO_CALL;
3139                 reloc_desc->insn_idx = insn_idx;
3140                 reloc_desc->sym_off = sym->st_value;
3141                 obj->has_pseudo_calls = true;
3142                 return 0;
3143         }
3144
3145         if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
3146                 pr_warn("invalid relo for insns[%d].code 0x%x\n",
3147                         insn_idx, insn->code);
3148                 return -LIBBPF_ERRNO__RELOC;
3149         }
3150
3151         if (sym_is_extern(sym)) {
3152                 int sym_idx = GELF_R_SYM(rel->r_info);
3153                 int i, n = obj->nr_extern;
3154                 struct extern_desc *ext;
3155
3156                 for (i = 0; i < n; i++) {
3157                         ext = &obj->externs[i];
3158                         if (ext->sym_idx == sym_idx)
3159                                 break;
3160                 }
3161                 if (i >= n) {
3162                         pr_warn("extern relo failed to find extern for sym %d\n",
3163                                 sym_idx);
3164                         return -LIBBPF_ERRNO__RELOC;
3165                 }
3166                 pr_debug("found extern #%d '%s' (sym %d) for insn %u\n",
3167                          i, ext->name, ext->sym_idx, insn_idx);
3168                 reloc_desc->type = RELO_EXTERN;
3169                 reloc_desc->insn_idx = insn_idx;
3170                 reloc_desc->sym_off = i; /* sym_off stores extern index */
3171                 return 0;
3172         }
3173
3174         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3175                 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
3176                         name, shdr_idx);
3177                 return -LIBBPF_ERRNO__RELOC;
3178         }
3179
3180         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
3181
3182         /* generic map reference relocation */
3183         if (type == LIBBPF_MAP_UNSPEC) {
3184                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
3185                         pr_warn("bad map relo against section %u\n",
3186                                 shdr_idx);
3187                         return -LIBBPF_ERRNO__RELOC;
3188                 }
3189                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3190                         map = &obj->maps[map_idx];
3191                         if (map->libbpf_type != type ||
3192                             map->sec_idx != sym->st_shndx ||
3193                             map->sec_offset != sym->st_value)
3194                                 continue;
3195                         pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
3196                                  map_idx, map->name, map->sec_idx,
3197                                  map->sec_offset, insn_idx);
3198                         break;
3199                 }
3200                 if (map_idx >= nr_maps) {
3201                         pr_warn("map relo failed to find map for sec %u, off %zu\n",
3202                                 shdr_idx, (size_t)sym->st_value);
3203                         return -LIBBPF_ERRNO__RELOC;
3204                 }
3205                 reloc_desc->type = RELO_LD64;
3206                 reloc_desc->insn_idx = insn_idx;
3207                 reloc_desc->map_idx = map_idx;
3208                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
3209                 return 0;
3210         }
3211
3212         /* global data map relocation */
3213         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
3214                 pr_warn("bad data relo against section %u\n", shdr_idx);
3215                 return -LIBBPF_ERRNO__RELOC;
3216         }
3217         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3218                 map = &obj->maps[map_idx];
3219                 if (map->libbpf_type != type)
3220                         continue;
3221                 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
3222                          map_idx, map->name, map->sec_idx, map->sec_offset,
3223                          insn_idx);
3224                 break;
3225         }
3226         if (map_idx >= nr_maps) {
3227                 pr_warn("data relo failed to find map for sec %u\n",
3228                         shdr_idx);
3229                 return -LIBBPF_ERRNO__RELOC;
3230         }
3231
3232         reloc_desc->type = RELO_DATA;
3233         reloc_desc->insn_idx = insn_idx;
3234         reloc_desc->map_idx = map_idx;
3235         reloc_desc->sym_off = sym->st_value;
3236         return 0;
3237 }
3238
3239 static int
3240 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
3241                            Elf_Data *data, struct bpf_object *obj)
3242 {
3243         Elf_Data *symbols = obj->efile.symbols;
3244         int err, i, nrels;
3245
3246         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
3247         nrels = shdr->sh_size / shdr->sh_entsize;
3248
3249         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
3250         if (!prog->reloc_desc) {
3251                 pr_warn("failed to alloc memory in relocation\n");
3252                 return -ENOMEM;
3253         }
3254         prog->nr_reloc = nrels;
3255
3256         for (i = 0; i < nrels; i++) {
3257                 const char *name;
3258                 __u32 insn_idx;
3259                 GElf_Sym sym;
3260                 GElf_Rel rel;
3261
3262                 if (!gelf_getrel(data, i, &rel)) {
3263                         pr_warn("relocation: failed to get %d reloc\n", i);
3264                         return -LIBBPF_ERRNO__FORMAT;
3265                 }
3266                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
3267                         pr_warn("relocation: symbol %"PRIx64" not found\n",
3268                                 GELF_R_SYM(rel.r_info));
3269                         return -LIBBPF_ERRNO__FORMAT;
3270                 }
3271                 if (rel.r_offset % sizeof(struct bpf_insn))
3272                         return -LIBBPF_ERRNO__FORMAT;
3273
3274                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
3275                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
3276                                   sym.st_name) ? : "<?>";
3277
3278                 pr_debug("relo for shdr %u, symb %zu, value %zu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
3279                          (__u32)sym.st_shndx, (size_t)GELF_R_SYM(rel.r_info),
3280                          (size_t)sym.st_value, GELF_ST_TYPE(sym.st_info),
3281                          GELF_ST_BIND(sym.st_info), sym.st_name, name,
3282                          insn_idx);
3283
3284                 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
3285                                                 insn_idx, name, &sym, &rel);
3286                 if (err)
3287                         return err;
3288         }
3289         return 0;
3290 }
3291
3292 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
3293 {
3294         struct bpf_map_def *def = &map->def;
3295         __u32 key_type_id = 0, value_type_id = 0;
3296         int ret;
3297
3298         /* if it's BTF-defined map, we don't need to search for type IDs.
3299          * For struct_ops map, it does not need btf_key_type_id and
3300          * btf_value_type_id.
3301          */
3302         if (map->sec_idx == obj->efile.btf_maps_shndx ||
3303             bpf_map__is_struct_ops(map))
3304                 return 0;
3305
3306         if (!bpf_map__is_internal(map)) {
3307                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
3308                                            def->value_size, &key_type_id,
3309                                            &value_type_id);
3310         } else {
3311                 /*
3312                  * LLVM annotates global data differently in BTF, that is,
3313                  * only as '.data', '.bss' or '.rodata'.
3314                  */
3315                 ret = btf__find_by_name(obj->btf,
3316                                 libbpf_type_to_btf_name[map->libbpf_type]);
3317         }
3318         if (ret < 0)
3319                 return ret;
3320
3321         map->btf_key_type_id = key_type_id;
3322         map->btf_value_type_id = bpf_map__is_internal(map) ?
3323                                  ret : value_type_id;
3324         return 0;
3325 }
3326
3327 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3328 {
3329         struct bpf_map_info info = {};
3330         __u32 len = sizeof(info);
3331         int new_fd, err;
3332         char *new_name;
3333
3334         err = bpf_obj_get_info_by_fd(fd, &info, &len);
3335         if (err)
3336                 return err;
3337
3338         new_name = strdup(info.name);
3339         if (!new_name)
3340                 return -errno;
3341
3342         new_fd = open("/", O_RDONLY | O_CLOEXEC);
3343         if (new_fd < 0) {
3344                 err = -errno;
3345                 goto err_free_new_name;
3346         }
3347
3348         new_fd = dup3(fd, new_fd, O_CLOEXEC);
3349         if (new_fd < 0) {
3350                 err = -errno;
3351                 goto err_close_new_fd;
3352         }
3353
3354         err = zclose(map->fd);
3355         if (err) {
3356                 err = -errno;
3357                 goto err_close_new_fd;
3358         }
3359         free(map->name);
3360
3361         map->fd = new_fd;
3362         map->name = new_name;
3363         map->def.type = info.type;
3364         map->def.key_size = info.key_size;
3365         map->def.value_size = info.value_size;
3366         map->def.max_entries = info.max_entries;
3367         map->def.map_flags = info.map_flags;
3368         map->btf_key_type_id = info.btf_key_type_id;
3369         map->btf_value_type_id = info.btf_value_type_id;
3370         map->reused = true;
3371
3372         return 0;
3373
3374 err_close_new_fd:
3375         close(new_fd);
3376 err_free_new_name:
3377         free(new_name);
3378         return err;
3379 }
3380
3381 __u32 bpf_map__max_entries(const struct bpf_map *map)
3382 {
3383         return map->def.max_entries;
3384 }
3385
3386 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
3387 {
3388         if (map->fd >= 0)
3389                 return -EBUSY;
3390         map->def.max_entries = max_entries;
3391         return 0;
3392 }
3393
3394 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
3395 {
3396         if (!map || !max_entries)
3397                 return -EINVAL;
3398
3399         return bpf_map__set_max_entries(map, max_entries);
3400 }
3401
3402 static int
3403 bpf_object__probe_loading(struct bpf_object *obj)
3404 {
3405         struct bpf_load_program_attr attr;
3406         char *cp, errmsg[STRERR_BUFSIZE];
3407         struct bpf_insn insns[] = {
3408                 BPF_MOV64_IMM(BPF_REG_0, 0),
3409                 BPF_EXIT_INSN(),
3410         };
3411         int ret;
3412
3413         /* make sure basic loading works */
3414
3415         memset(&attr, 0, sizeof(attr));
3416         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3417         attr.insns = insns;
3418         attr.insns_cnt = ARRAY_SIZE(insns);
3419         attr.license = "GPL";
3420
3421         ret = bpf_load_program_xattr(&attr, NULL, 0);
3422         if (ret < 0) {
3423                 ret = errno;
3424                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3425                 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
3426                         "program. Make sure your kernel supports BPF "
3427                         "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
3428                         "set to big enough value.\n", __func__, cp, ret);
3429                 return -ret;
3430         }
3431         close(ret);
3432
3433         return 0;
3434 }
3435
3436 static int
3437 bpf_object__probe_name(struct bpf_object *obj)
3438 {
3439         struct bpf_load_program_attr attr;
3440         struct bpf_insn insns[] = {
3441                 BPF_MOV64_IMM(BPF_REG_0, 0),
3442                 BPF_EXIT_INSN(),
3443         };
3444         int ret;
3445
3446         /* make sure loading with name works */
3447
3448         memset(&attr, 0, sizeof(attr));
3449         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3450         attr.insns = insns;
3451         attr.insns_cnt = ARRAY_SIZE(insns);
3452         attr.license = "GPL";
3453         attr.name = "test";
3454         ret = bpf_load_program_xattr(&attr, NULL, 0);
3455         if (ret >= 0) {
3456                 obj->caps.name = 1;
3457                 close(ret);
3458         }
3459
3460         return 0;
3461 }
3462
3463 static int
3464 bpf_object__probe_global_data(struct bpf_object *obj)
3465 {
3466         struct bpf_load_program_attr prg_attr;
3467         struct bpf_create_map_attr map_attr;
3468         char *cp, errmsg[STRERR_BUFSIZE];
3469         struct bpf_insn insns[] = {
3470                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
3471                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
3472                 BPF_MOV64_IMM(BPF_REG_0, 0),
3473                 BPF_EXIT_INSN(),
3474         };
3475         int ret, map;
3476
3477         memset(&map_attr, 0, sizeof(map_attr));
3478         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3479         map_attr.key_size = sizeof(int);
3480         map_attr.value_size = 32;
3481         map_attr.max_entries = 1;
3482
3483         map = bpf_create_map_xattr(&map_attr);
3484         if (map < 0) {
3485                 ret = -errno;
3486                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3487                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3488                         __func__, cp, -ret);
3489                 return ret;
3490         }
3491
3492         insns[0].imm = map;
3493
3494         memset(&prg_attr, 0, sizeof(prg_attr));
3495         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3496         prg_attr.insns = insns;
3497         prg_attr.insns_cnt = ARRAY_SIZE(insns);
3498         prg_attr.license = "GPL";
3499
3500         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
3501         if (ret >= 0) {
3502                 obj->caps.global_data = 1;
3503                 close(ret);
3504         }
3505
3506         close(map);
3507         return 0;
3508 }
3509
3510 static int bpf_object__probe_btf_func(struct bpf_object *obj)
3511 {
3512         static const char strs[] = "\0int\0x\0a";
3513         /* void x(int a) {} */
3514         __u32 types[] = {
3515                 /* int */
3516                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3517                 /* FUNC_PROTO */                                /* [2] */
3518                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3519                 BTF_PARAM_ENC(7, 1),
3520                 /* FUNC x */                                    /* [3] */
3521                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
3522         };
3523         int btf_fd;
3524
3525         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3526                                       strs, sizeof(strs));
3527         if (btf_fd >= 0) {
3528                 obj->caps.btf_func = 1;
3529                 close(btf_fd);
3530                 return 1;
3531         }
3532
3533         return 0;
3534 }
3535
3536 static int bpf_object__probe_btf_func_global(struct bpf_object *obj)
3537 {
3538         static const char strs[] = "\0int\0x\0a";
3539         /* static void x(int a) {} */
3540         __u32 types[] = {
3541                 /* int */
3542                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3543                 /* FUNC_PROTO */                                /* [2] */
3544                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3545                 BTF_PARAM_ENC(7, 1),
3546                 /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
3547                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
3548         };
3549         int btf_fd;
3550
3551         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3552                                       strs, sizeof(strs));
3553         if (btf_fd >= 0) {
3554                 obj->caps.btf_func_global = 1;
3555                 close(btf_fd);
3556                 return 1;
3557         }
3558
3559         return 0;
3560 }
3561
3562 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
3563 {
3564         static const char strs[] = "\0x\0.data";
3565         /* static int a; */
3566         __u32 types[] = {
3567                 /* int */
3568                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3569                 /* VAR x */                                     /* [2] */
3570                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
3571                 BTF_VAR_STATIC,
3572                 /* DATASEC val */                               /* [3] */
3573                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
3574                 BTF_VAR_SECINFO_ENC(2, 0, 4),
3575         };
3576         int btf_fd;
3577
3578         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3579                                       strs, sizeof(strs));
3580         if (btf_fd >= 0) {
3581                 obj->caps.btf_datasec = 1;
3582                 close(btf_fd);
3583                 return 1;
3584         }
3585
3586         return 0;
3587 }
3588
3589 static int bpf_object__probe_array_mmap(struct bpf_object *obj)
3590 {
3591         struct bpf_create_map_attr attr = {
3592                 .map_type = BPF_MAP_TYPE_ARRAY,
3593                 .map_flags = BPF_F_MMAPABLE,
3594                 .key_size = sizeof(int),
3595                 .value_size = sizeof(int),
3596                 .max_entries = 1,
3597         };
3598         int fd;
3599
3600         fd = bpf_create_map_xattr(&attr);
3601         if (fd >= 0) {
3602                 obj->caps.array_mmap = 1;
3603                 close(fd);
3604                 return 1;
3605         }
3606
3607         return 0;
3608 }
3609
3610 static int
3611 bpf_object__probe_exp_attach_type(struct bpf_object *obj)
3612 {
3613         struct bpf_load_program_attr attr;
3614         struct bpf_insn insns[] = {
3615                 BPF_MOV64_IMM(BPF_REG_0, 0),
3616                 BPF_EXIT_INSN(),
3617         };
3618         int fd;
3619
3620         memset(&attr, 0, sizeof(attr));
3621         /* use any valid combination of program type and (optional)
3622          * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
3623          * to see if kernel supports expected_attach_type field for
3624          * BPF_PROG_LOAD command
3625          */
3626         attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
3627         attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE;
3628         attr.insns = insns;
3629         attr.insns_cnt = ARRAY_SIZE(insns);
3630         attr.license = "GPL";
3631
3632         fd = bpf_load_program_xattr(&attr, NULL, 0);
3633         if (fd >= 0) {
3634                 obj->caps.exp_attach_type = 1;
3635                 close(fd);
3636                 return 1;
3637         }
3638         return 0;
3639 }
3640
3641 static int
3642 bpf_object__probe_caps(struct bpf_object *obj)
3643 {
3644         int (*probe_fn[])(struct bpf_object *obj) = {
3645                 bpf_object__probe_name,
3646                 bpf_object__probe_global_data,
3647                 bpf_object__probe_btf_func,
3648                 bpf_object__probe_btf_func_global,
3649                 bpf_object__probe_btf_datasec,
3650                 bpf_object__probe_array_mmap,
3651                 bpf_object__probe_exp_attach_type,
3652         };
3653         int i, ret;
3654
3655         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
3656                 ret = probe_fn[i](obj);
3657                 if (ret < 0)
3658                         pr_debug("Probe #%d failed with %d.\n", i, ret);
3659         }
3660
3661         return 0;
3662 }
3663
3664 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
3665 {
3666         struct bpf_map_info map_info = {};
3667         char msg[STRERR_BUFSIZE];
3668         __u32 map_info_len;
3669
3670         map_info_len = sizeof(map_info);
3671
3672         if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
3673                 pr_warn("failed to get map info for map FD %d: %s\n",
3674                         map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
3675                 return false;
3676         }
3677
3678         return (map_info.type == map->def.type &&
3679                 map_info.key_size == map->def.key_size &&
3680                 map_info.value_size == map->def.value_size &&
3681                 map_info.max_entries == map->def.max_entries &&
3682                 map_info.map_flags == map->def.map_flags);
3683 }
3684
3685 static int
3686 bpf_object__reuse_map(struct bpf_map *map)
3687 {
3688         char *cp, errmsg[STRERR_BUFSIZE];
3689         int err, pin_fd;
3690
3691         pin_fd = bpf_obj_get(map->pin_path);
3692         if (pin_fd < 0) {
3693                 err = -errno;
3694                 if (err == -ENOENT) {
3695                         pr_debug("found no pinned map to reuse at '%s'\n",
3696                                  map->pin_path);
3697                         return 0;
3698                 }
3699
3700                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3701                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
3702                         map->pin_path, cp);
3703                 return err;
3704         }
3705
3706         if (!map_is_reuse_compat(map, pin_fd)) {
3707                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
3708                         map->pin_path);
3709                 close(pin_fd);
3710                 return -EINVAL;
3711         }
3712
3713         err = bpf_map__reuse_fd(map, pin_fd);
3714         if (err) {
3715                 close(pin_fd);
3716                 return err;
3717         }
3718         map->pinned = true;
3719         pr_debug("reused pinned map at '%s'\n", map->pin_path);
3720
3721         return 0;
3722 }
3723
3724 static int
3725 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
3726 {
3727         enum libbpf_map_type map_type = map->libbpf_type;
3728         char *cp, errmsg[STRERR_BUFSIZE];
3729         int err, zero = 0;
3730
3731         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
3732         if (err) {
3733                 err = -errno;
3734                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3735                 pr_warn("Error setting initial map(%s) contents: %s\n",
3736                         map->name, cp);
3737                 return err;
3738         }
3739
3740         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
3741         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
3742                 err = bpf_map_freeze(map->fd);
3743                 if (err) {
3744                         err = -errno;
3745                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3746                         pr_warn("Error freezing map(%s) as read-only: %s\n",
3747                                 map->name, cp);
3748                         return err;
3749                 }
3750         }
3751         return 0;
3752 }
3753
3754 static void bpf_map__destroy(struct bpf_map *map);
3755
3756 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map)
3757 {
3758         struct bpf_create_map_attr create_attr;
3759         struct bpf_map_def *def = &map->def;
3760
3761         memset(&create_attr, 0, sizeof(create_attr));
3762
3763         if (obj->caps.name)
3764                 create_attr.name = map->name;
3765         create_attr.map_ifindex = map->map_ifindex;
3766         create_attr.map_type = def->type;
3767         create_attr.map_flags = def->map_flags;
3768         create_attr.key_size = def->key_size;
3769         create_attr.value_size = def->value_size;
3770         create_attr.numa_node = map->numa_node;
3771
3772         if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) {
3773                 int nr_cpus;
3774
3775                 nr_cpus = libbpf_num_possible_cpus();
3776                 if (nr_cpus < 0) {
3777                         pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
3778                                 map->name, nr_cpus);
3779                         return nr_cpus;
3780                 }
3781                 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
3782                 create_attr.max_entries = nr_cpus;
3783         } else {
3784                 create_attr.max_entries = def->max_entries;
3785         }
3786
3787         if (bpf_map__is_struct_ops(map))
3788                 create_attr.btf_vmlinux_value_type_id =
3789                         map->btf_vmlinux_value_type_id;
3790
3791         create_attr.btf_fd = 0;
3792         create_attr.btf_key_type_id = 0;
3793         create_attr.btf_value_type_id = 0;
3794         if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) {
3795                 create_attr.btf_fd = btf__fd(obj->btf);
3796                 create_attr.btf_key_type_id = map->btf_key_type_id;
3797                 create_attr.btf_value_type_id = map->btf_value_type_id;
3798         }
3799
3800         if (bpf_map_type__is_map_in_map(def->type)) {
3801                 if (map->inner_map) {
3802                         int err;
3803
3804                         err = bpf_object__create_map(obj, map->inner_map);
3805                         if (err) {
3806                                 pr_warn("map '%s': failed to create inner map: %d\n",
3807                                         map->name, err);
3808                                 return err;
3809                         }
3810                         map->inner_map_fd = bpf_map__fd(map->inner_map);
3811                 }
3812                 if (map->inner_map_fd >= 0)
3813                         create_attr.inner_map_fd = map->inner_map_fd;
3814         }
3815
3816         map->fd = bpf_create_map_xattr(&create_attr);
3817         if (map->fd < 0 && (create_attr.btf_key_type_id ||
3818                             create_attr.btf_value_type_id)) {
3819                 char *cp, errmsg[STRERR_BUFSIZE];
3820                 int err = -errno;
3821
3822                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3823                 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
3824                         map->name, cp, err);
3825                 create_attr.btf_fd = 0;
3826                 create_attr.btf_key_type_id = 0;
3827                 create_attr.btf_value_type_id = 0;
3828                 map->btf_key_type_id = 0;
3829                 map->btf_value_type_id = 0;
3830                 map->fd = bpf_create_map_xattr(&create_attr);
3831         }
3832
3833         if (map->fd < 0)
3834                 return -errno;
3835
3836         if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
3837                 bpf_map__destroy(map->inner_map);
3838                 zfree(&map->inner_map);
3839         }
3840
3841         return 0;
3842 }
3843
3844 static int
3845 bpf_object__create_maps(struct bpf_object *obj)
3846 {
3847         struct bpf_map *map;
3848         char *cp, errmsg[STRERR_BUFSIZE];
3849         unsigned int i, j;
3850         int err;
3851
3852         for (i = 0; i < obj->nr_maps; i++) {
3853                 map = &obj->maps[i];
3854
3855                 if (map->pin_path) {
3856                         err = bpf_object__reuse_map(map);
3857                         if (err) {
3858                                 pr_warn("map '%s': error reusing pinned map\n",
3859                                         map->name);
3860                                 goto err_out;
3861                         }
3862                 }
3863
3864                 if (map->fd >= 0) {
3865                         pr_debug("map '%s': skipping creation (preset fd=%d)\n",
3866                                  map->name, map->fd);
3867                         continue;
3868                 }
3869
3870                 err = bpf_object__create_map(obj, map);
3871                 if (err)
3872                         goto err_out;
3873
3874                 pr_debug("map '%s': created successfully, fd=%d\n", map->name,
3875                          map->fd);
3876
3877                 if (bpf_map__is_internal(map)) {
3878                         err = bpf_object__populate_internal_map(obj, map);
3879                         if (err < 0) {
3880                                 zclose(map->fd);
3881                                 goto err_out;
3882                         }
3883                 }
3884
3885                 if (map->init_slots_sz) {
3886                         for (j = 0; j < map->init_slots_sz; j++) {
3887                                 const struct bpf_map *targ_map;
3888                                 int fd;
3889
3890                                 if (!map->init_slots[j])
3891                                         continue;
3892
3893                                 targ_map = map->init_slots[j];
3894                                 fd = bpf_map__fd(targ_map);
3895                                 err = bpf_map_update_elem(map->fd, &j, &fd, 0);
3896                                 if (err) {
3897                                         err = -errno;
3898                                         pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
3899                                                 map->name, j, targ_map->name,
3900                                                 fd, err);
3901                                         goto err_out;
3902                                 }
3903                                 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
3904                                          map->name, j, targ_map->name, fd);
3905                         }
3906                         zfree(&map->init_slots);
3907                         map->init_slots_sz = 0;
3908                 }
3909
3910                 if (map->pin_path && !map->pinned) {
3911                         err = bpf_map__pin(map, NULL);
3912                         if (err) {
3913                                 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
3914                                         map->name, map->pin_path, err);
3915                                 zclose(map->fd);
3916                                 goto err_out;
3917                         }
3918                 }
3919         }
3920
3921         return 0;
3922
3923 err_out:
3924         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3925         pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
3926         pr_perm_msg(err);
3927         for (j = 0; j < i; j++)
3928                 zclose(obj->maps[j].fd);
3929         return err;
3930 }
3931
3932 static int
3933 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
3934                         void *btf_prog_info, const char *info_name)
3935 {
3936         if (err != -ENOENT) {
3937                 pr_warn("Error in loading %s for sec %s.\n",
3938                         info_name, prog->section_name);
3939                 return err;
3940         }
3941
3942         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
3943
3944         if (btf_prog_info) {
3945                 /*
3946                  * Some info has already been found but has problem
3947                  * in the last btf_ext reloc. Must have to error out.
3948                  */
3949                 pr_warn("Error in relocating %s for sec %s.\n",
3950                         info_name, prog->section_name);
3951                 return err;
3952         }
3953
3954         /* Have problem loading the very first info. Ignore the rest. */
3955         pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
3956                 info_name, prog->section_name, info_name);
3957         return 0;
3958 }
3959
3960 static int
3961 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
3962                           const char *section_name,  __u32 insn_offset)
3963 {
3964         int err;
3965
3966         if (!insn_offset || prog->func_info) {
3967                 /*
3968                  * !insn_offset => main program
3969                  *
3970                  * For sub prog, the main program's func_info has to
3971                  * be loaded first (i.e. prog->func_info != NULL)
3972                  */
3973                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
3974                                                section_name, insn_offset,
3975                                                &prog->func_info,
3976                                                &prog->func_info_cnt);
3977                 if (err)
3978                         return check_btf_ext_reloc_err(prog, err,
3979                                                        prog->func_info,
3980                                                        "bpf_func_info");
3981
3982                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
3983         }
3984
3985         if (!insn_offset || prog->line_info) {
3986                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
3987                                                section_name, insn_offset,
3988                                                &prog->line_info,
3989                                                &prog->line_info_cnt);
3990                 if (err)
3991                         return check_btf_ext_reloc_err(prog, err,
3992                                                        prog->line_info,
3993                                                        "bpf_line_info");
3994
3995                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
3996         }
3997
3998         return 0;
3999 }
4000
4001 #define BPF_CORE_SPEC_MAX_LEN 64
4002
4003 /* represents BPF CO-RE field or array element accessor */
4004 struct bpf_core_accessor {
4005         __u32 type_id;          /* struct/union type or array element type */
4006         __u32 idx;              /* field index or array index */
4007         const char *name;       /* field name or NULL for array accessor */
4008 };
4009
4010 struct bpf_core_spec {
4011         const struct btf *btf;
4012         /* high-level spec: named fields and array indices only */
4013         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
4014         /* high-level spec length */
4015         int len;
4016         /* raw, low-level spec: 1-to-1 with accessor spec string */
4017         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
4018         /* raw spec length */
4019         int raw_len;
4020         /* field bit offset represented by spec */
4021         __u32 bit_offset;
4022 };
4023
4024 static bool str_is_empty(const char *s)
4025 {
4026         return !s || !s[0];
4027 }
4028
4029 static bool is_flex_arr(const struct btf *btf,
4030                         const struct bpf_core_accessor *acc,
4031                         const struct btf_array *arr)
4032 {
4033         const struct btf_type *t;
4034
4035         /* not a flexible array, if not inside a struct or has non-zero size */
4036         if (!acc->name || arr->nelems > 0)
4037                 return false;
4038
4039         /* has to be the last member of enclosing struct */
4040         t = btf__type_by_id(btf, acc->type_id);
4041         return acc->idx == btf_vlen(t) - 1;
4042 }
4043
4044 /*
4045  * Turn bpf_field_reloc into a low- and high-level spec representation,
4046  * validating correctness along the way, as well as calculating resulting
4047  * field bit offset, specified by accessor string. Low-level spec captures
4048  * every single level of nestedness, including traversing anonymous
4049  * struct/union members. High-level one only captures semantically meaningful
4050  * "turning points": named fields and array indicies.
4051  * E.g., for this case:
4052  *
4053  *   struct sample {
4054  *       int __unimportant;
4055  *       struct {
4056  *           int __1;
4057  *           int __2;
4058  *           int a[7];
4059  *       };
4060  *   };
4061  *
4062  *   struct sample *s = ...;
4063  *
4064  *   int x = &s->a[3]; // access string = '0:1:2:3'
4065  *
4066  * Low-level spec has 1:1 mapping with each element of access string (it's
4067  * just a parsed access string representation): [0, 1, 2, 3].
4068  *
4069  * High-level spec will capture only 3 points:
4070  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
4071  *   - field 'a' access (corresponds to '2' in low-level spec);
4072  *   - array element #3 access (corresponds to '3' in low-level spec).
4073  *
4074  */
4075 static int bpf_core_spec_parse(const struct btf *btf,
4076                                __u32 type_id,
4077                                const char *spec_str,
4078                                struct bpf_core_spec *spec)
4079 {
4080         int access_idx, parsed_len, i;
4081         struct bpf_core_accessor *acc;
4082         const struct btf_type *t;
4083         const char *name;
4084         __u32 id;
4085         __s64 sz;
4086
4087         if (str_is_empty(spec_str) || *spec_str == ':')
4088                 return -EINVAL;
4089
4090         memset(spec, 0, sizeof(*spec));
4091         spec->btf = btf;
4092
4093         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
4094         while (*spec_str) {
4095                 if (*spec_str == ':')
4096                         ++spec_str;
4097                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
4098                         return -EINVAL;
4099                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4100                         return -E2BIG;
4101                 spec_str += parsed_len;
4102                 spec->raw_spec[spec->raw_len++] = access_idx;
4103         }
4104
4105         if (spec->raw_len == 0)
4106                 return -EINVAL;
4107
4108         /* first spec value is always reloc type array index */
4109         t = skip_mods_and_typedefs(btf, type_id, &id);
4110         if (!t)
4111                 return -EINVAL;
4112
4113         access_idx = spec->raw_spec[0];
4114         spec->spec[0].type_id = id;
4115         spec->spec[0].idx = access_idx;
4116         spec->len++;
4117
4118         sz = btf__resolve_size(btf, id);
4119         if (sz < 0)
4120                 return sz;
4121         spec->bit_offset = access_idx * sz * 8;
4122
4123         for (i = 1; i < spec->raw_len; i++) {
4124                 t = skip_mods_and_typedefs(btf, id, &id);
4125                 if (!t)
4126                         return -EINVAL;
4127
4128                 access_idx = spec->raw_spec[i];
4129                 acc = &spec->spec[spec->len];
4130
4131                 if (btf_is_composite(t)) {
4132                         const struct btf_member *m;
4133                         __u32 bit_offset;
4134
4135                         if (access_idx >= btf_vlen(t))
4136                                 return -EINVAL;
4137
4138                         bit_offset = btf_member_bit_offset(t, access_idx);
4139                         spec->bit_offset += bit_offset;
4140
4141                         m = btf_members(t) + access_idx;
4142                         if (m->name_off) {
4143                                 name = btf__name_by_offset(btf, m->name_off);
4144                                 if (str_is_empty(name))
4145                                         return -EINVAL;
4146
4147                                 acc->type_id = id;
4148                                 acc->idx = access_idx;
4149                                 acc->name = name;
4150                                 spec->len++;
4151                         }
4152
4153                         id = m->type;
4154                 } else if (btf_is_array(t)) {
4155                         const struct btf_array *a = btf_array(t);
4156                         bool flex;
4157
4158                         t = skip_mods_and_typedefs(btf, a->type, &id);
4159                         if (!t)
4160                                 return -EINVAL;
4161
4162                         flex = is_flex_arr(btf, acc - 1, a);
4163                         if (!flex && access_idx >= a->nelems)
4164                                 return -EINVAL;
4165
4166                         spec->spec[spec->len].type_id = id;
4167                         spec->spec[spec->len].idx = access_idx;
4168                         spec->len++;
4169
4170                         sz = btf__resolve_size(btf, id);
4171                         if (sz < 0)
4172                                 return sz;
4173                         spec->bit_offset += access_idx * sz * 8;
4174                 } else {
4175                         pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
4176                                 type_id, spec_str, i, id, btf_kind(t));
4177                         return -EINVAL;
4178                 }
4179         }
4180
4181         return 0;
4182 }
4183
4184 static bool bpf_core_is_flavor_sep(const char *s)
4185 {
4186         /* check X___Y name pattern, where X and Y are not underscores */
4187         return s[0] != '_' &&                                 /* X */
4188                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
4189                s[4] != '_';                                   /* Y */
4190 }
4191
4192 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
4193  * before last triple underscore. Struct name part after last triple
4194  * underscore is ignored by BPF CO-RE relocation during relocation matching.
4195  */
4196 static size_t bpf_core_essential_name_len(const char *name)
4197 {
4198         size_t n = strlen(name);
4199         int i;
4200
4201         for (i = n - 5; i >= 0; i--) {
4202                 if (bpf_core_is_flavor_sep(name + i))
4203                         return i + 1;
4204         }
4205         return n;
4206 }
4207
4208 /* dynamically sized list of type IDs */
4209 struct ids_vec {
4210         __u32 *data;
4211         int len;
4212 };
4213
4214 static void bpf_core_free_cands(struct ids_vec *cand_ids)
4215 {
4216         free(cand_ids->data);
4217         free(cand_ids);
4218 }
4219
4220 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
4221                                            __u32 local_type_id,
4222                                            const struct btf *targ_btf)
4223 {
4224         size_t local_essent_len, targ_essent_len;
4225         const char *local_name, *targ_name;
4226         const struct btf_type *t;
4227         struct ids_vec *cand_ids;
4228         __u32 *new_ids;
4229         int i, err, n;
4230
4231         t = btf__type_by_id(local_btf, local_type_id);
4232         if (!t)
4233                 return ERR_PTR(-EINVAL);
4234
4235         local_name = btf__name_by_offset(local_btf, t->name_off);
4236         if (str_is_empty(local_name))
4237                 return ERR_PTR(-EINVAL);
4238         local_essent_len = bpf_core_essential_name_len(local_name);
4239
4240         cand_ids = calloc(1, sizeof(*cand_ids));
4241         if (!cand_ids)
4242                 return ERR_PTR(-ENOMEM);
4243
4244         n = btf__get_nr_types(targ_btf);
4245         for (i = 1; i <= n; i++) {
4246                 t = btf__type_by_id(targ_btf, i);
4247                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
4248                 if (str_is_empty(targ_name))
4249                         continue;
4250
4251                 t = skip_mods_and_typedefs(targ_btf, i, NULL);
4252                 if (!btf_is_composite(t) && !btf_is_array(t))
4253                         continue;
4254
4255                 targ_essent_len = bpf_core_essential_name_len(targ_name);
4256                 if (targ_essent_len != local_essent_len)
4257                         continue;
4258
4259                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
4260                         pr_debug("[%d] %s: found candidate [%d] %s\n",
4261                                  local_type_id, local_name, i, targ_name);
4262                         new_ids = reallocarray(cand_ids->data,
4263                                                cand_ids->len + 1,
4264                                                sizeof(*cand_ids->data));
4265                         if (!new_ids) {
4266                                 err = -ENOMEM;
4267                                 goto err_out;
4268                         }
4269                         cand_ids->data = new_ids;
4270                         cand_ids->data[cand_ids->len++] = i;
4271                 }
4272         }
4273         return cand_ids;
4274 err_out:
4275         bpf_core_free_cands(cand_ids);
4276         return ERR_PTR(err);
4277 }
4278
4279 /* Check two types for compatibility, skipping const/volatile/restrict and
4280  * typedefs, to ensure we are relocating compatible entities:
4281  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
4282  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
4283  *   - any two PTRs are always compatible;
4284  *   - for ENUMs, names should be the same (ignoring flavor suffix) or at
4285  *     least one of enums should be anonymous;
4286  *   - for ENUMs, check sizes, names are ignored;
4287  *   - for INT, size and signedness are ignored;
4288  *   - for ARRAY, dimensionality is ignored, element types are checked for
4289  *     compatibility recursively;
4290  *   - everything else shouldn't be ever a target of relocation.
4291  * These rules are not set in stone and probably will be adjusted as we get
4292  * more experience with using BPF CO-RE relocations.
4293  */
4294 static int bpf_core_fields_are_compat(const struct btf *local_btf,
4295                                       __u32 local_id,
4296                                       const struct btf *targ_btf,
4297                                       __u32 targ_id)
4298 {
4299         const struct btf_type *local_type, *targ_type;
4300
4301 recur:
4302         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
4303         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4304         if (!local_type || !targ_type)
4305                 return -EINVAL;
4306
4307         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
4308                 return 1;
4309         if (btf_kind(local_type) != btf_kind(targ_type))
4310                 return 0;
4311
4312         switch (btf_kind(local_type)) {
4313         case BTF_KIND_PTR:
4314                 return 1;
4315         case BTF_KIND_FWD:
4316         case BTF_KIND_ENUM: {
4317                 const char *local_name, *targ_name;
4318                 size_t local_len, targ_len;
4319
4320                 local_name = btf__name_by_offset(local_btf,
4321                                                  local_type->name_off);
4322                 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
4323                 local_len = bpf_core_essential_name_len(local_name);
4324                 targ_len = bpf_core_essential_name_len(targ_name);
4325                 /* one of them is anonymous or both w/ same flavor-less names */
4326                 return local_len == 0 || targ_len == 0 ||
4327                        (local_len == targ_len &&
4328                         strncmp(local_name, targ_name, local_len) == 0);
4329         }
4330         case BTF_KIND_INT:
4331                 /* just reject deprecated bitfield-like integers; all other
4332                  * integers are by default compatible between each other
4333                  */
4334                 return btf_int_offset(local_type) == 0 &&
4335                        btf_int_offset(targ_type) == 0;
4336         case BTF_KIND_ARRAY:
4337                 local_id = btf_array(local_type)->type;
4338                 targ_id = btf_array(targ_type)->type;
4339                 goto recur;
4340         default:
4341                 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
4342                         btf_kind(local_type), local_id, targ_id);
4343                 return 0;
4344         }
4345 }
4346
4347 /*
4348  * Given single high-level named field accessor in local type, find
4349  * corresponding high-level accessor for a target type. Along the way,
4350  * maintain low-level spec for target as well. Also keep updating target
4351  * bit offset.
4352  *
4353  * Searching is performed through recursive exhaustive enumeration of all
4354  * fields of a struct/union. If there are any anonymous (embedded)
4355  * structs/unions, they are recursively searched as well. If field with
4356  * desired name is found, check compatibility between local and target types,
4357  * before returning result.
4358  *
4359  * 1 is returned, if field is found.
4360  * 0 is returned if no compatible field is found.
4361  * <0 is returned on error.
4362  */
4363 static int bpf_core_match_member(const struct btf *local_btf,
4364                                  const struct bpf_core_accessor *local_acc,
4365                                  const struct btf *targ_btf,
4366                                  __u32 targ_id,
4367                                  struct bpf_core_spec *spec,
4368                                  __u32 *next_targ_id)
4369 {
4370         const struct btf_type *local_type, *targ_type;
4371         const struct btf_member *local_member, *m;
4372         const char *local_name, *targ_name;
4373         __u32 local_id;
4374         int i, n, found;
4375
4376         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4377         if (!targ_type)
4378                 return -EINVAL;
4379         if (!btf_is_composite(targ_type))
4380                 return 0;
4381
4382         local_id = local_acc->type_id;
4383         local_type = btf__type_by_id(local_btf, local_id);
4384         local_member = btf_members(local_type) + local_acc->idx;
4385         local_name = btf__name_by_offset(local_btf, local_member->name_off);
4386
4387         n = btf_vlen(targ_type);
4388         m = btf_members(targ_type);
4389         for (i = 0; i < n; i++, m++) {
4390                 __u32 bit_offset;
4391
4392                 bit_offset = btf_member_bit_offset(targ_type, i);
4393
4394                 /* too deep struct/union/array nesting */
4395                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4396                         return -E2BIG;
4397
4398                 /* speculate this member will be the good one */
4399                 spec->bit_offset += bit_offset;
4400                 spec->raw_spec[spec->raw_len++] = i;
4401
4402                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
4403                 if (str_is_empty(targ_name)) {
4404                         /* embedded struct/union, we need to go deeper */
4405                         found = bpf_core_match_member(local_btf, local_acc,
4406                                                       targ_btf, m->type,
4407                                                       spec, next_targ_id);
4408                         if (found) /* either found or error */
4409                                 return found;
4410                 } else if (strcmp(local_name, targ_name) == 0) {
4411                         /* matching named field */
4412                         struct bpf_core_accessor *targ_acc;
4413
4414                         targ_acc = &spec->spec[spec->len++];
4415                         targ_acc->type_id = targ_id;
4416                         targ_acc->idx = i;
4417                         targ_acc->name = targ_name;
4418
4419                         *next_targ_id = m->type;
4420                         found = bpf_core_fields_are_compat(local_btf,
4421                                                            local_member->type,
4422                                                            targ_btf, m->type);
4423                         if (!found)
4424                                 spec->len--; /* pop accessor */
4425                         return found;
4426                 }
4427                 /* member turned out not to be what we looked for */
4428                 spec->bit_offset -= bit_offset;
4429                 spec->raw_len--;
4430         }
4431
4432         return 0;
4433 }
4434
4435 /*
4436  * Try to match local spec to a target type and, if successful, produce full
4437  * target spec (high-level, low-level + bit offset).
4438  */
4439 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
4440                                const struct btf *targ_btf, __u32 targ_id,
4441                                struct bpf_core_spec *targ_spec)
4442 {
4443         const struct btf_type *targ_type;
4444         const struct bpf_core_accessor *local_acc;
4445         struct bpf_core_accessor *targ_acc;
4446         int i, sz, matched;
4447
4448         memset(targ_spec, 0, sizeof(*targ_spec));
4449         targ_spec->btf = targ_btf;
4450
4451         local_acc = &local_spec->spec[0];
4452         targ_acc = &targ_spec->spec[0];
4453
4454         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
4455                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
4456                                                    &targ_id);
4457                 if (!targ_type)
4458                         return -EINVAL;
4459
4460                 if (local_acc->name) {
4461                         matched = bpf_core_match_member(local_spec->btf,
4462                                                         local_acc,
4463                                                         targ_btf, targ_id,
4464                                                         targ_spec, &targ_id);
4465                         if (matched <= 0)
4466                                 return matched;
4467                 } else {
4468                         /* for i=0, targ_id is already treated as array element
4469                          * type (because it's the original struct), for others
4470                          * we should find array element type first
4471                          */
4472                         if (i > 0) {
4473                                 const struct btf_array *a;
4474                                 bool flex;
4475
4476                                 if (!btf_is_array(targ_type))
4477                                         return 0;
4478
4479                                 a = btf_array(targ_type);
4480                                 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
4481                                 if (!flex && local_acc->idx >= a->nelems)
4482                                         return 0;
4483                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
4484                                                             &targ_id))
4485                                         return -EINVAL;
4486                         }
4487
4488                         /* too deep struct/union/array nesting */
4489                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4490                                 return -E2BIG;
4491
4492                         targ_acc->type_id = targ_id;
4493                         targ_acc->idx = local_acc->idx;
4494                         targ_acc->name = NULL;
4495                         targ_spec->len++;
4496                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
4497                         targ_spec->raw_len++;
4498
4499                         sz = btf__resolve_size(targ_btf, targ_id);
4500                         if (sz < 0)
4501                                 return sz;
4502                         targ_spec->bit_offset += local_acc->idx * sz * 8;
4503                 }
4504         }
4505
4506         return 1;
4507 }
4508
4509 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
4510                                     const struct bpf_field_reloc *relo,
4511                                     const struct bpf_core_spec *spec,
4512                                     __u32 *val, bool *validate)
4513 {
4514         const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
4515         const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
4516         __u32 byte_off, byte_sz, bit_off, bit_sz;
4517         const struct btf_member *m;
4518         const struct btf_type *mt;
4519         bool bitfield;
4520         __s64 sz;
4521
4522         /* a[n] accessor needs special handling */
4523         if (!acc->name) {
4524                 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
4525                         *val = spec->bit_offset / 8;
4526                 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
4527                         sz = btf__resolve_size(spec->btf, acc->type_id);
4528                         if (sz < 0)
4529                                 return -EINVAL;
4530                         *val = sz;
4531                 } else {
4532                         pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
4533                                 bpf_program__title(prog, false),
4534                                 relo->kind, relo->insn_off / 8);
4535                         return -EINVAL;
4536                 }
4537                 if (validate)
4538                         *validate = true;
4539                 return 0;
4540         }
4541
4542         m = btf_members(t) + acc->idx;
4543         mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
4544         bit_off = spec->bit_offset;
4545         bit_sz = btf_member_bitfield_size(t, acc->idx);
4546
4547         bitfield = bit_sz > 0;
4548         if (bitfield) {
4549                 byte_sz = mt->size;
4550                 byte_off = bit_off / 8 / byte_sz * byte_sz;
4551                 /* figure out smallest int size necessary for bitfield load */
4552                 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
4553                         if (byte_sz >= 8) {
4554                                 /* bitfield can't be read with 64-bit read */
4555                                 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
4556                                         bpf_program__title(prog, false),
4557                                         relo->kind, relo->insn_off / 8);
4558                                 return -E2BIG;
4559                         }
4560                         byte_sz *= 2;
4561                         byte_off = bit_off / 8 / byte_sz * byte_sz;
4562                 }
4563         } else {
4564                 sz = btf__resolve_size(spec->btf, m->type);
4565                 if (sz < 0)
4566                         return -EINVAL;
4567                 byte_sz = sz;
4568                 byte_off = spec->bit_offset / 8;
4569                 bit_sz = byte_sz * 8;
4570         }
4571
4572         /* for bitfields, all the relocatable aspects are ambiguous and we
4573          * might disagree with compiler, so turn off validation of expected
4574          * value, except for signedness
4575          */
4576         if (validate)
4577                 *validate = !bitfield;
4578
4579         switch (relo->kind) {
4580         case BPF_FIELD_BYTE_OFFSET:
4581                 *val = byte_off;
4582                 break;
4583         case BPF_FIELD_BYTE_SIZE:
4584                 *val = byte_sz;
4585                 break;
4586         case BPF_FIELD_SIGNED:
4587                 /* enums will be assumed unsigned */
4588                 *val = btf_is_enum(mt) ||
4589                        (btf_int_encoding(mt) & BTF_INT_SIGNED);
4590                 if (validate)
4591                         *validate = true; /* signedness is never ambiguous */
4592                 break;
4593         case BPF_FIELD_LSHIFT_U64:
4594 #if __BYTE_ORDER == __LITTLE_ENDIAN
4595                 *val = 64 - (bit_off + bit_sz - byte_off  * 8);
4596 #else
4597                 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
4598 #endif
4599                 break;
4600         case BPF_FIELD_RSHIFT_U64:
4601                 *val = 64 - bit_sz;
4602                 if (validate)
4603                         *validate = true; /* right shift is never ambiguous */
4604                 break;
4605         case BPF_FIELD_EXISTS:
4606         default:
4607                 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
4608                         bpf_program__title(prog, false),
4609                         relo->kind, relo->insn_off / 8);
4610                 return -EINVAL;
4611         }
4612
4613         return 0;
4614 }
4615
4616 /*
4617  * Patch relocatable BPF instruction.
4618  *
4619  * Patched value is determined by relocation kind and target specification.
4620  * For field existence relocation target spec will be NULL if field is not
4621  * found.
4622  * Expected insn->imm value is determined using relocation kind and local
4623  * spec, and is checked before patching instruction. If actual insn->imm value
4624  * is wrong, bail out with error.
4625  *
4626  * Currently three kinds of BPF instructions are supported:
4627  * 1. rX = <imm> (assignment with immediate operand);
4628  * 2. rX += <imm> (arithmetic operations with immediate operand);
4629  */
4630 static int bpf_core_reloc_insn(struct bpf_program *prog,
4631                                const struct bpf_field_reloc *relo,
4632                                int relo_idx,
4633                                const struct bpf_core_spec *local_spec,
4634                                const struct bpf_core_spec *targ_spec)
4635 {
4636         __u32 orig_val, new_val;
4637         struct bpf_insn *insn;
4638         bool validate = true;
4639         int insn_idx, err;
4640         __u8 class;
4641
4642         if (relo->insn_off % sizeof(struct bpf_insn))
4643                 return -EINVAL;
4644         insn_idx = relo->insn_off / sizeof(struct bpf_insn);
4645         insn = &prog->insns[insn_idx];
4646         class = BPF_CLASS(insn->code);
4647
4648         if (relo->kind == BPF_FIELD_EXISTS) {
4649                 orig_val = 1; /* can't generate EXISTS relo w/o local field */
4650                 new_val = targ_spec ? 1 : 0;
4651         } else if (!targ_spec) {
4652                 pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n",
4653                          bpf_program__title(prog, false), relo_idx, insn_idx);
4654                 insn->code = BPF_JMP | BPF_CALL;
4655                 insn->dst_reg = 0;
4656                 insn->src_reg = 0;
4657                 insn->off = 0;
4658                 /* if this instruction is reachable (not a dead code),
4659                  * verifier will complain with the following message:
4660                  * invalid func unknown#195896080
4661                  */
4662                 insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */
4663                 return 0;
4664         } else {
4665                 err = bpf_core_calc_field_relo(prog, relo, local_spec,
4666                                                &orig_val, &validate);
4667                 if (err)
4668                         return err;
4669                 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
4670                                                &new_val, NULL);
4671                 if (err)
4672                         return err;
4673         }
4674
4675         switch (class) {
4676         case BPF_ALU:
4677         case BPF_ALU64:
4678                 if (BPF_SRC(insn->code) != BPF_K)
4679                         return -EINVAL;
4680                 if (validate && insn->imm != orig_val) {
4681                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
4682                                 bpf_program__title(prog, false), relo_idx,
4683                                 insn_idx, insn->imm, orig_val, new_val);
4684                         return -EINVAL;
4685                 }
4686                 orig_val = insn->imm;
4687                 insn->imm = new_val;
4688                 pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n",
4689                          bpf_program__title(prog, false), relo_idx, insn_idx,
4690                          orig_val, new_val);
4691                 break;
4692         case BPF_LDX:
4693         case BPF_ST:
4694         case BPF_STX:
4695                 if (validate && insn->off != orig_val) {
4696                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (LD/LDX/ST/STX) value: got %u, exp %u -> %u\n",
4697                                 bpf_program__title(prog, false), relo_idx,
4698                                 insn_idx, insn->off, orig_val, new_val);
4699                         return -EINVAL;
4700                 }
4701                 if (new_val > SHRT_MAX) {
4702                         pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n",
4703                                 bpf_program__title(prog, false), relo_idx,
4704                                 insn_idx, new_val);
4705                         return -ERANGE;
4706                 }
4707                 orig_val = insn->off;
4708                 insn->off = new_val;
4709                 pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n",
4710                          bpf_program__title(prog, false), relo_idx, insn_idx,
4711                          orig_val, new_val);
4712                 break;
4713         default:
4714                 pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
4715                         bpf_program__title(prog, false), relo_idx,
4716                         insn_idx, insn->code, insn->src_reg, insn->dst_reg,
4717                         insn->off, insn->imm);
4718                 return -EINVAL;
4719         }
4720
4721         return 0;
4722 }
4723
4724 /* Output spec definition in the format:
4725  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
4726  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
4727  */
4728 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
4729 {
4730         const struct btf_type *t;
4731         const char *s;
4732         __u32 type_id;
4733         int i;
4734
4735         type_id = spec->spec[0].type_id;
4736         t = btf__type_by_id(spec->btf, type_id);
4737         s = btf__name_by_offset(spec->btf, t->name_off);
4738         libbpf_print(level, "[%u] %s + ", type_id, s);
4739
4740         for (i = 0; i < spec->raw_len; i++)
4741                 libbpf_print(level, "%d%s", spec->raw_spec[i],
4742                              i == spec->raw_len - 1 ? " => " : ":");
4743
4744         libbpf_print(level, "%u.%u @ &x",
4745                      spec->bit_offset / 8, spec->bit_offset % 8);
4746
4747         for (i = 0; i < spec->len; i++) {
4748                 if (spec->spec[i].name)
4749                         libbpf_print(level, ".%s", spec->spec[i].name);
4750                 else
4751                         libbpf_print(level, "[%u]", spec->spec[i].idx);
4752         }
4753
4754 }
4755
4756 static size_t bpf_core_hash_fn(const void *key, void *ctx)
4757 {
4758         return (size_t)key;
4759 }
4760
4761 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
4762 {
4763         return k1 == k2;
4764 }
4765
4766 static void *u32_as_hash_key(__u32 x)
4767 {
4768         return (void *)(uintptr_t)x;
4769 }
4770
4771 /*
4772  * CO-RE relocate single instruction.
4773  *
4774  * The outline and important points of the algorithm:
4775  * 1. For given local type, find corresponding candidate target types.
4776  *    Candidate type is a type with the same "essential" name, ignoring
4777  *    everything after last triple underscore (___). E.g., `sample`,
4778  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
4779  *    for each other. Names with triple underscore are referred to as
4780  *    "flavors" and are useful, among other things, to allow to
4781  *    specify/support incompatible variations of the same kernel struct, which
4782  *    might differ between different kernel versions and/or build
4783  *    configurations.
4784  *
4785  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
4786  *    converter, when deduplicated BTF of a kernel still contains more than
4787  *    one different types with the same name. In that case, ___2, ___3, etc
4788  *    are appended starting from second name conflict. But start flavors are
4789  *    also useful to be defined "locally", in BPF program, to extract same
4790  *    data from incompatible changes between different kernel
4791  *    versions/configurations. For instance, to handle field renames between
4792  *    kernel versions, one can use two flavors of the struct name with the
4793  *    same common name and use conditional relocations to extract that field,
4794  *    depending on target kernel version.
4795  * 2. For each candidate type, try to match local specification to this
4796  *    candidate target type. Matching involves finding corresponding
4797  *    high-level spec accessors, meaning that all named fields should match,
4798  *    as well as all array accesses should be within the actual bounds. Also,
4799  *    types should be compatible (see bpf_core_fields_are_compat for details).
4800  * 3. It is supported and expected that there might be multiple flavors
4801  *    matching the spec. As long as all the specs resolve to the same set of
4802  *    offsets across all candidates, there is no error. If there is any
4803  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
4804  *    imprefection of BTF deduplication, which can cause slight duplication of
4805  *    the same BTF type, if some directly or indirectly referenced (by
4806  *    pointer) type gets resolved to different actual types in different
4807  *    object files. If such situation occurs, deduplicated BTF will end up
4808  *    with two (or more) structurally identical types, which differ only in
4809  *    types they refer to through pointer. This should be OK in most cases and
4810  *    is not an error.
4811  * 4. Candidate types search is performed by linearly scanning through all
4812  *    types in target BTF. It is anticipated that this is overall more
4813  *    efficient memory-wise and not significantly worse (if not better)
4814  *    CPU-wise compared to prebuilding a map from all local type names to
4815  *    a list of candidate type names. It's also sped up by caching resolved
4816  *    list of matching candidates per each local "root" type ID, that has at
4817  *    least one bpf_field_reloc associated with it. This list is shared
4818  *    between multiple relocations for the same type ID and is updated as some
4819  *    of the candidates are pruned due to structural incompatibility.
4820  */
4821 static int bpf_core_reloc_field(struct bpf_program *prog,
4822                                  const struct bpf_field_reloc *relo,
4823                                  int relo_idx,
4824                                  const struct btf *local_btf,
4825                                  const struct btf *targ_btf,
4826                                  struct hashmap *cand_cache)
4827 {
4828         const char *prog_name = bpf_program__title(prog, false);
4829         struct bpf_core_spec local_spec, cand_spec, targ_spec;
4830         const void *type_key = u32_as_hash_key(relo->type_id);
4831         const struct btf_type *local_type, *cand_type;
4832         const char *local_name, *cand_name;
4833         struct ids_vec *cand_ids;
4834         __u32 local_id, cand_id;
4835         const char *spec_str;
4836         int i, j, err;
4837
4838         local_id = relo->type_id;
4839         local_type = btf__type_by_id(local_btf, local_id);
4840         if (!local_type)
4841                 return -EINVAL;
4842
4843         local_name = btf__name_by_offset(local_btf, local_type->name_off);
4844         if (str_is_empty(local_name))
4845                 return -EINVAL;
4846
4847         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
4848         if (str_is_empty(spec_str))
4849                 return -EINVAL;
4850
4851         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
4852         if (err) {
4853                 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
4854                         prog_name, relo_idx, local_id, local_name, spec_str,
4855                         err);
4856                 return -EINVAL;
4857         }
4858
4859         pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
4860                  relo->kind);
4861         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
4862         libbpf_print(LIBBPF_DEBUG, "\n");
4863
4864         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
4865                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
4866                 if (IS_ERR(cand_ids)) {
4867                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
4868                                 prog_name, relo_idx, local_id, local_name,
4869                                 PTR_ERR(cand_ids));
4870                         return PTR_ERR(cand_ids);
4871                 }
4872                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
4873                 if (err) {
4874                         bpf_core_free_cands(cand_ids);
4875                         return err;
4876                 }
4877         }
4878
4879         for (i = 0, j = 0; i < cand_ids->len; i++) {
4880                 cand_id = cand_ids->data[i];
4881                 cand_type = btf__type_by_id(targ_btf, cand_id);
4882                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
4883
4884                 err = bpf_core_spec_match(&local_spec, targ_btf,
4885                                           cand_id, &cand_spec);
4886                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
4887                          prog_name, relo_idx, i, cand_name);
4888                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
4889                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
4890                 if (err < 0) {
4891                         pr_warn("prog '%s': relo #%d: matching error: %d\n",
4892                                 prog_name, relo_idx, err);
4893                         return err;
4894                 }
4895                 if (err == 0)
4896                         continue;
4897
4898                 if (j == 0) {
4899                         targ_spec = cand_spec;
4900                 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
4901                         /* if there are many candidates, they should all
4902                          * resolve to the same bit offset
4903                          */
4904                         pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
4905                                 prog_name, relo_idx, cand_spec.bit_offset,
4906                                 targ_spec.bit_offset);
4907                         return -EINVAL;
4908                 }
4909
4910                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
4911         }
4912
4913         /*
4914          * For BPF_FIELD_EXISTS relo or when used BPF program has field
4915          * existence checks or kernel version/config checks, it's expected
4916          * that we might not find any candidates. In this case, if field
4917          * wasn't found in any candidate, the list of candidates shouldn't
4918          * change at all, we'll just handle relocating appropriately,
4919          * depending on relo's kind.
4920          */
4921         if (j > 0)
4922                 cand_ids->len = j;
4923
4924         /*
4925          * If no candidates were found, it might be both a programmer error,
4926          * as well as expected case, depending whether instruction w/
4927          * relocation is guarded in some way that makes it unreachable (dead
4928          * code) if relocation can't be resolved. This is handled in
4929          * bpf_core_reloc_insn() uniformly by replacing that instruction with
4930          * BPF helper call insn (using invalid helper ID). If that instruction
4931          * is indeed unreachable, then it will be ignored and eliminated by
4932          * verifier. If it was an error, then verifier will complain and point
4933          * to a specific instruction number in its log.
4934          */
4935         if (j == 0)
4936                 pr_debug("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
4937                          prog_name, relo_idx, local_id, local_name, spec_str);
4938
4939         /* bpf_core_reloc_insn should know how to handle missing targ_spec */
4940         err = bpf_core_reloc_insn(prog, relo, relo_idx, &local_spec,
4941                                   j ? &targ_spec : NULL);
4942         if (err) {
4943                 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
4944                         prog_name, relo_idx, relo->insn_off, err);
4945                 return -EINVAL;
4946         }
4947
4948         return 0;
4949 }
4950
4951 static int
4952 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
4953 {
4954         const struct btf_ext_info_sec *sec;
4955         const struct bpf_field_reloc *rec;
4956         const struct btf_ext_info *seg;
4957         struct hashmap_entry *entry;
4958         struct hashmap *cand_cache = NULL;
4959         struct bpf_program *prog;
4960         struct btf *targ_btf;
4961         const char *sec_name;
4962         int i, err = 0;
4963
4964         if (targ_btf_path)
4965                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
4966         else
4967                 targ_btf = obj->btf_vmlinux;
4968         if (IS_ERR_OR_NULL(targ_btf)) {
4969                 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
4970                 return PTR_ERR(targ_btf);
4971         }
4972
4973         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
4974         if (IS_ERR(cand_cache)) {
4975                 err = PTR_ERR(cand_cache);
4976                 goto out;
4977         }
4978
4979         seg = &obj->btf_ext->field_reloc_info;
4980         for_each_btf_ext_sec(seg, sec) {
4981                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
4982                 if (str_is_empty(sec_name)) {
4983                         err = -EINVAL;
4984                         goto out;
4985                 }
4986                 prog = NULL;
4987                 for (i = 0; i < obj->nr_programs; i++) {
4988                         if (!strcmp(obj->programs[i].section_name, sec_name)) {
4989                                 prog = &obj->programs[i];
4990                                 break;
4991                         }
4992                 }
4993                 if (!prog) {
4994                         pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
4995                                 sec_name);
4996                         err = -EINVAL;
4997                         goto out;
4998                 }
4999
5000                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
5001                          sec_name, sec->num_info);
5002
5003                 for_each_btf_ext_rec(seg, sec, i, rec) {
5004                         err = bpf_core_reloc_field(prog, rec, i, obj->btf,
5005                                                    targ_btf, cand_cache);
5006                         if (err) {
5007                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5008                                         sec_name, i, err);
5009                                 goto out;
5010                         }
5011                 }
5012         }
5013
5014 out:
5015         /* obj->btf_vmlinux is freed at the end of object load phase */
5016         if (targ_btf != obj->btf_vmlinux)
5017                 btf__free(targ_btf);
5018         if (!IS_ERR_OR_NULL(cand_cache)) {
5019                 hashmap__for_each_entry(cand_cache, entry, i) {
5020                         bpf_core_free_cands(entry->value);
5021                 }
5022                 hashmap__free(cand_cache);
5023         }
5024         return err;
5025 }
5026
5027 static int
5028 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5029 {
5030         int err = 0;
5031
5032         if (obj->btf_ext->field_reloc_info.len)
5033                 err = bpf_core_reloc_fields(obj, targ_btf_path);
5034
5035         return err;
5036 }
5037
5038 static int
5039 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
5040                         struct reloc_desc *relo)
5041 {
5042         struct bpf_insn *insn, *new_insn;
5043         struct bpf_program *text;
5044         size_t new_cnt;
5045         int err;
5046
5047         if (prog->idx != obj->efile.text_shndx && prog->main_prog_cnt == 0) {
5048                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
5049                 if (!text) {
5050                         pr_warn("no .text section found yet relo into text exist\n");
5051                         return -LIBBPF_ERRNO__RELOC;
5052                 }
5053                 new_cnt = prog->insns_cnt + text->insns_cnt;
5054                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
5055                 if (!new_insn) {
5056                         pr_warn("oom in prog realloc\n");
5057                         return -ENOMEM;
5058                 }
5059                 prog->insns = new_insn;
5060
5061                 if (obj->btf_ext) {
5062                         err = bpf_program_reloc_btf_ext(prog, obj,
5063                                                         text->section_name,
5064                                                         prog->insns_cnt);
5065                         if (err)
5066                                 return err;
5067                 }
5068
5069                 memcpy(new_insn + prog->insns_cnt, text->insns,
5070                        text->insns_cnt * sizeof(*insn));
5071                 prog->main_prog_cnt = prog->insns_cnt;
5072                 prog->insns_cnt = new_cnt;
5073                 pr_debug("added %zd insn from %s to prog %s\n",
5074                          text->insns_cnt, text->section_name,
5075                          prog->section_name);
5076         }
5077
5078         insn = &prog->insns[relo->insn_idx];
5079         insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
5080         return 0;
5081 }
5082
5083 static int
5084 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
5085 {
5086         int i, err;
5087
5088         if (!prog)
5089                 return 0;
5090
5091         if (obj->btf_ext) {
5092                 err = bpf_program_reloc_btf_ext(prog, obj,
5093                                                 prog->section_name, 0);
5094                 if (err)
5095                         return err;
5096         }
5097
5098         if (!prog->reloc_desc)
5099                 return 0;
5100
5101         for (i = 0; i < prog->nr_reloc; i++) {
5102                 struct reloc_desc *relo = &prog->reloc_desc[i];
5103                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5104                 struct extern_desc *ext;
5105
5106                 if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
5107                         pr_warn("relocation out of range: '%s'\n",
5108                                 prog->section_name);
5109                         return -LIBBPF_ERRNO__RELOC;
5110                 }
5111
5112                 switch (relo->type) {
5113                 case RELO_LD64:
5114                         insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5115                         insn[0].imm = obj->maps[relo->map_idx].fd;
5116                         break;
5117                 case RELO_DATA:
5118                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5119                         insn[1].imm = insn[0].imm + relo->sym_off;
5120                         insn[0].imm = obj->maps[relo->map_idx].fd;
5121                         break;
5122                 case RELO_EXTERN:
5123                         ext = &obj->externs[relo->sym_off];
5124                         if (ext->type == EXT_KCFG) {
5125                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5126                                 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5127                                 insn[1].imm = ext->kcfg.data_off;
5128                         } else /* EXT_KSYM */ {
5129                                 insn[0].imm = (__u32)ext->ksym.addr;
5130                                 insn[1].imm = ext->ksym.addr >> 32;
5131                         }
5132                         break;
5133                 case RELO_CALL:
5134                         err = bpf_program__reloc_text(prog, obj, relo);
5135                         if (err)
5136                                 return err;
5137                         break;
5138                 default:
5139                         pr_warn("relo #%d: bad relo type %d\n", i, relo->type);
5140                         return -EINVAL;
5141                 }
5142         }
5143
5144         zfree(&prog->reloc_desc);
5145         prog->nr_reloc = 0;
5146         return 0;
5147 }
5148
5149 static int
5150 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
5151 {
5152         struct bpf_program *prog;
5153         size_t i;
5154         int err;
5155
5156         if (obj->btf_ext) {
5157                 err = bpf_object__relocate_core(obj, targ_btf_path);
5158                 if (err) {
5159                         pr_warn("failed to perform CO-RE relocations: %d\n",
5160                                 err);
5161                         return err;
5162                 }
5163         }
5164         /* ensure .text is relocated first, as it's going to be copied as-is
5165          * later for sub-program calls
5166          */
5167         for (i = 0; i < obj->nr_programs; i++) {
5168                 prog = &obj->programs[i];
5169                 if (prog->idx != obj->efile.text_shndx)
5170                         continue;
5171
5172                 err = bpf_program__relocate(prog, obj);
5173                 if (err) {
5174                         pr_warn("failed to relocate '%s'\n", prog->section_name);
5175                         return err;
5176                 }
5177                 break;
5178         }
5179         /* now relocate everything but .text, which by now is relocated
5180          * properly, so we can copy raw sub-program instructions as is safely
5181          */
5182         for (i = 0; i < obj->nr_programs; i++) {
5183                 prog = &obj->programs[i];
5184                 if (prog->idx == obj->efile.text_shndx)
5185                         continue;
5186
5187                 err = bpf_program__relocate(prog, obj);
5188                 if (err) {
5189                         pr_warn("failed to relocate '%s'\n", prog->section_name);
5190                         return err;
5191                 }
5192         }
5193         return 0;
5194 }
5195
5196 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
5197                                             GElf_Shdr *shdr, Elf_Data *data);
5198
5199 static int bpf_object__collect_map_relos(struct bpf_object *obj,
5200                                          GElf_Shdr *shdr, Elf_Data *data)
5201 {
5202         const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
5203         int i, j, nrels, new_sz;
5204         const struct btf_var_secinfo *vi = NULL;
5205         const struct btf_type *sec, *var, *def;
5206         const struct btf_member *member;
5207         struct bpf_map *map, *targ_map;
5208         const char *name, *mname;
5209         Elf_Data *symbols;
5210         unsigned int moff;
5211         GElf_Sym sym;
5212         GElf_Rel rel;
5213         void *tmp;
5214
5215         if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
5216                 return -EINVAL;
5217         sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
5218         if (!sec)
5219                 return -EINVAL;
5220
5221         symbols = obj->efile.symbols;
5222         nrels = shdr->sh_size / shdr->sh_entsize;
5223         for (i = 0; i < nrels; i++) {
5224                 if (!gelf_getrel(data, i, &rel)) {
5225                         pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
5226                         return -LIBBPF_ERRNO__FORMAT;
5227                 }
5228                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
5229                         pr_warn(".maps relo #%d: symbol %zx not found\n",
5230                                 i, (size_t)GELF_R_SYM(rel.r_info));
5231                         return -LIBBPF_ERRNO__FORMAT;
5232                 }
5233                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
5234                                   sym.st_name) ? : "<?>";
5235                 if (sym.st_shndx != obj->efile.btf_maps_shndx) {
5236                         pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
5237                                 i, name);
5238                         return -LIBBPF_ERRNO__RELOC;
5239                 }
5240
5241                 pr_debug(".maps relo #%d: for %zd value %zd rel.r_offset %zu name %d ('%s')\n",
5242                          i, (ssize_t)(rel.r_info >> 32), (size_t)sym.st_value,
5243                          (size_t)rel.r_offset, sym.st_name, name);
5244
5245                 for (j = 0; j < obj->nr_maps; j++) {
5246                         map = &obj->maps[j];
5247                         if (map->sec_idx != obj->efile.btf_maps_shndx)
5248                                 continue;
5249
5250                         vi = btf_var_secinfos(sec) + map->btf_var_idx;
5251                         if (vi->offset <= rel.r_offset &&
5252                             rel.r_offset + bpf_ptr_sz <= vi->offset + vi->size)
5253                                 break;
5254                 }
5255                 if (j == obj->nr_maps) {
5256                         pr_warn(".maps relo #%d: cannot find map '%s' at rel.r_offset %zu\n",
5257                                 i, name, (size_t)rel.r_offset);
5258                         return -EINVAL;
5259                 }
5260
5261                 if (!bpf_map_type__is_map_in_map(map->def.type))
5262                         return -EINVAL;
5263                 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
5264                     map->def.key_size != sizeof(int)) {
5265                         pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
5266                                 i, map->name, sizeof(int));
5267                         return -EINVAL;
5268                 }
5269
5270                 targ_map = bpf_object__find_map_by_name(obj, name);
5271                 if (!targ_map)
5272                         return -ESRCH;
5273
5274                 var = btf__type_by_id(obj->btf, vi->type);
5275                 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
5276                 if (btf_vlen(def) == 0)
5277                         return -EINVAL;
5278                 member = btf_members(def) + btf_vlen(def) - 1;
5279                 mname = btf__name_by_offset(obj->btf, member->name_off);
5280                 if (strcmp(mname, "values"))
5281                         return -EINVAL;
5282
5283                 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
5284                 if (rel.r_offset - vi->offset < moff)
5285                         return -EINVAL;
5286
5287                 moff = rel.r_offset - vi->offset - moff;
5288                 /* here we use BPF pointer size, which is always 64 bit, as we
5289                  * are parsing ELF that was built for BPF target
5290                  */
5291                 if (moff % bpf_ptr_sz)
5292                         return -EINVAL;
5293                 moff /= bpf_ptr_sz;
5294                 if (moff >= map->init_slots_sz) {
5295                         new_sz = moff + 1;
5296                         tmp = realloc(map->init_slots, new_sz * host_ptr_sz);
5297                         if (!tmp)
5298                                 return -ENOMEM;
5299                         map->init_slots = tmp;
5300                         memset(map->init_slots + map->init_slots_sz, 0,
5301                                (new_sz - map->init_slots_sz) * host_ptr_sz);
5302                         map->init_slots_sz = new_sz;
5303                 }
5304                 map->init_slots[moff] = targ_map;
5305
5306                 pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n",
5307                          i, map->name, moff, name);
5308         }
5309
5310         return 0;
5311 }
5312
5313 static int bpf_object__collect_reloc(struct bpf_object *obj)
5314 {
5315         int i, err;
5316
5317         if (!obj_elf_valid(obj)) {
5318                 pr_warn("Internal error: elf object is closed\n");
5319                 return -LIBBPF_ERRNO__INTERNAL;
5320         }
5321
5322         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
5323                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
5324                 Elf_Data *data = obj->efile.reloc_sects[i].data;
5325                 int idx = shdr->sh_info;
5326                 struct bpf_program *prog;
5327
5328                 if (shdr->sh_type != SHT_REL) {
5329                         pr_warn("internal error at %d\n", __LINE__);
5330                         return -LIBBPF_ERRNO__INTERNAL;
5331                 }
5332
5333                 if (idx == obj->efile.st_ops_shndx) {
5334                         err = bpf_object__collect_st_ops_relos(obj, shdr, data);
5335                 } else if (idx == obj->efile.btf_maps_shndx) {
5336                         err = bpf_object__collect_map_relos(obj, shdr, data);
5337                 } else {
5338                         prog = bpf_object__find_prog_by_idx(obj, idx);
5339                         if (!prog) {
5340                                 pr_warn("relocation failed: no prog in section(%d)\n", idx);
5341                                 return -LIBBPF_ERRNO__RELOC;
5342                         }
5343                         err = bpf_program__collect_reloc(prog, shdr, data, obj);
5344                 }
5345                 if (err)
5346                         return err;
5347         }
5348         return 0;
5349 }
5350
5351 static int
5352 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
5353              char *license, __u32 kern_version, int *pfd)
5354 {
5355         struct bpf_load_program_attr load_attr;
5356         char *cp, errmsg[STRERR_BUFSIZE];
5357         size_t log_buf_size = 0;
5358         char *log_buf = NULL;
5359         int btf_fd, ret;
5360
5361         if (!insns || !insns_cnt)
5362                 return -EINVAL;
5363
5364         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
5365         load_attr.prog_type = prog->type;
5366         /* old kernels might not support specifying expected_attach_type */
5367         if (!prog->caps->exp_attach_type && prog->sec_def &&
5368             prog->sec_def->is_exp_attach_type_optional)
5369                 load_attr.expected_attach_type = 0;
5370         else
5371                 load_attr.expected_attach_type = prog->expected_attach_type;
5372         if (prog->caps->name)
5373                 load_attr.name = prog->name;
5374         load_attr.insns = insns;
5375         load_attr.insns_cnt = insns_cnt;
5376         load_attr.license = license;
5377         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
5378             prog->type == BPF_PROG_TYPE_LSM) {
5379                 load_attr.attach_btf_id = prog->attach_btf_id;
5380         } else if (prog->type == BPF_PROG_TYPE_TRACING ||
5381                    prog->type == BPF_PROG_TYPE_EXT) {
5382                 load_attr.attach_prog_fd = prog->attach_prog_fd;
5383                 load_attr.attach_btf_id = prog->attach_btf_id;
5384         } else {
5385                 load_attr.kern_version = kern_version;
5386                 load_attr.prog_ifindex = prog->prog_ifindex;
5387         }
5388         /* specify func_info/line_info only if kernel supports them */
5389         btf_fd = bpf_object__btf_fd(prog->obj);
5390         if (btf_fd >= 0 && prog->obj->caps.btf_func) {
5391                 load_attr.prog_btf_fd = btf_fd;
5392                 load_attr.func_info = prog->func_info;
5393                 load_attr.func_info_rec_size = prog->func_info_rec_size;
5394                 load_attr.func_info_cnt = prog->func_info_cnt;
5395                 load_attr.line_info = prog->line_info;
5396                 load_attr.line_info_rec_size = prog->line_info_rec_size;
5397                 load_attr.line_info_cnt = prog->line_info_cnt;
5398         }
5399         load_attr.log_level = prog->log_level;
5400         load_attr.prog_flags = prog->prog_flags;
5401
5402 retry_load:
5403         if (log_buf_size) {
5404                 log_buf = malloc(log_buf_size);
5405                 if (!log_buf)
5406                         return -ENOMEM;
5407
5408                 *log_buf = 0;
5409         }
5410
5411         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
5412
5413         if (ret >= 0) {
5414                 if (log_buf && load_attr.log_level)
5415                         pr_debug("verifier log:\n%s", log_buf);
5416                 *pfd = ret;
5417                 ret = 0;
5418                 goto out;
5419         }
5420
5421         if (!log_buf || errno == ENOSPC) {
5422                 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE,
5423                                    log_buf_size << 1);
5424
5425                 free(log_buf);
5426                 goto retry_load;
5427         }
5428         ret = -errno;
5429         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5430         pr_warn("load bpf program failed: %s\n", cp);
5431         pr_perm_msg(ret);
5432
5433         if (log_buf && log_buf[0] != '\0') {
5434                 ret = -LIBBPF_ERRNO__VERIFY;
5435                 pr_warn("-- BEGIN DUMP LOG ---\n");
5436                 pr_warn("\n%s\n", log_buf);
5437                 pr_warn("-- END LOG --\n");
5438         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
5439                 pr_warn("Program too large (%zu insns), at most %d insns\n",
5440                         load_attr.insns_cnt, BPF_MAXINSNS);
5441                 ret = -LIBBPF_ERRNO__PROG2BIG;
5442         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
5443                 /* Wrong program type? */
5444                 int fd;
5445
5446                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
5447                 load_attr.expected_attach_type = 0;
5448                 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
5449                 if (fd >= 0) {
5450                         close(fd);
5451                         ret = -LIBBPF_ERRNO__PROGTYPE;
5452                         goto out;
5453                 }
5454         }
5455
5456 out:
5457         free(log_buf);
5458         return ret;
5459 }
5460
5461 static int libbpf_find_attach_btf_id(struct bpf_program *prog);
5462
5463 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
5464 {
5465         int err = 0, fd, i, btf_id;
5466
5467         if (prog->obj->loaded) {
5468                 pr_warn("prog '%s'('%s'): can't load after object was loaded\n",
5469                         prog->name, prog->section_name);
5470                 return -EINVAL;
5471         }
5472
5473         if ((prog->type == BPF_PROG_TYPE_TRACING ||
5474              prog->type == BPF_PROG_TYPE_LSM ||
5475              prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
5476                 btf_id = libbpf_find_attach_btf_id(prog);
5477                 if (btf_id <= 0)
5478                         return btf_id;
5479                 prog->attach_btf_id = btf_id;
5480         }
5481
5482         if (prog->instances.nr < 0 || !prog->instances.fds) {
5483                 if (prog->preprocessor) {
5484                         pr_warn("Internal error: can't load program '%s'\n",
5485                                 prog->section_name);
5486                         return -LIBBPF_ERRNO__INTERNAL;
5487                 }
5488
5489                 prog->instances.fds = malloc(sizeof(int));
5490                 if (!prog->instances.fds) {
5491                         pr_warn("Not enough memory for BPF fds\n");
5492                         return -ENOMEM;
5493                 }
5494                 prog->instances.nr = 1;
5495                 prog->instances.fds[0] = -1;
5496         }
5497
5498         if (!prog->preprocessor) {
5499                 if (prog->instances.nr != 1) {
5500                         pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
5501                                 prog->section_name, prog->instances.nr);
5502                 }
5503                 err = load_program(prog, prog->insns, prog->insns_cnt,
5504                                    license, kern_ver, &fd);
5505                 if (!err)
5506                         prog->instances.fds[0] = fd;
5507                 goto out;
5508         }
5509
5510         for (i = 0; i < prog->instances.nr; i++) {
5511                 struct bpf_prog_prep_result result;
5512                 bpf_program_prep_t preprocessor = prog->preprocessor;
5513
5514                 memset(&result, 0, sizeof(result));
5515                 err = preprocessor(prog, i, prog->insns,
5516                                    prog->insns_cnt, &result);
5517                 if (err) {
5518                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
5519                                 i, prog->section_name);
5520                         goto out;
5521                 }
5522
5523                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
5524                         pr_debug("Skip loading the %dth instance of program '%s'\n",
5525                                  i, prog->section_name);
5526                         prog->instances.fds[i] = -1;
5527                         if (result.pfd)
5528                                 *result.pfd = -1;
5529                         continue;
5530                 }
5531
5532                 err = load_program(prog, result.new_insn_ptr,
5533                                    result.new_insn_cnt, license, kern_ver, &fd);
5534                 if (err) {
5535                         pr_warn("Loading the %dth instance of program '%s' failed\n",
5536                                 i, prog->section_name);
5537                         goto out;
5538                 }
5539
5540                 if (result.pfd)
5541                         *result.pfd = fd;
5542                 prog->instances.fds[i] = fd;
5543         }
5544 out:
5545         if (err)
5546                 pr_warn("failed to load program '%s'\n", prog->section_name);
5547         zfree(&prog->insns);
5548         prog->insns_cnt = 0;
5549         return err;
5550 }
5551
5552 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
5553                                              const struct bpf_object *obj)
5554 {
5555         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
5556 }
5557
5558 static int
5559 bpf_object__load_progs(struct bpf_object *obj, int log_level)
5560 {
5561         struct bpf_program *prog;
5562         size_t i;
5563         int err;
5564
5565         for (i = 0; i < obj->nr_programs; i++) {
5566                 prog = &obj->programs[i];
5567                 if (bpf_program__is_function_storage(prog, obj))
5568                         continue;
5569                 if (!prog->load) {
5570                         pr_debug("prog '%s'('%s'): skipped loading\n",
5571                                  prog->name, prog->section_name);
5572                         continue;
5573                 }
5574                 prog->log_level |= log_level;
5575                 err = bpf_program__load(prog, obj->license, obj->kern_version);
5576                 if (err)
5577                         return err;
5578         }
5579         return 0;
5580 }
5581
5582 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
5583
5584 static struct bpf_object *
5585 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
5586                    const struct bpf_object_open_opts *opts)
5587 {
5588         const char *obj_name, *kconfig;
5589         struct bpf_program *prog;
5590         struct bpf_object *obj;
5591         char tmp_name[64];
5592         int err;
5593
5594         if (elf_version(EV_CURRENT) == EV_NONE) {
5595                 pr_warn("failed to init libelf for %s\n",
5596                         path ? : "(mem buf)");
5597                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
5598         }
5599
5600         if (!OPTS_VALID(opts, bpf_object_open_opts))
5601                 return ERR_PTR(-EINVAL);
5602
5603         obj_name = OPTS_GET(opts, object_name, NULL);
5604         if (obj_buf) {
5605                 if (!obj_name) {
5606                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
5607                                  (unsigned long)obj_buf,
5608                                  (unsigned long)obj_buf_sz);
5609                         obj_name = tmp_name;
5610                 }
5611                 path = obj_name;
5612                 pr_debug("loading object '%s' from buffer\n", obj_name);
5613         }
5614
5615         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
5616         if (IS_ERR(obj))
5617                 return obj;
5618
5619         kconfig = OPTS_GET(opts, kconfig, NULL);
5620         if (kconfig) {
5621                 obj->kconfig = strdup(kconfig);
5622                 if (!obj->kconfig)
5623                         return ERR_PTR(-ENOMEM);
5624         }
5625
5626         err = bpf_object__elf_init(obj);
5627         err = err ? : bpf_object__check_endianness(obj);
5628         err = err ? : bpf_object__elf_collect(obj);
5629         err = err ? : bpf_object__collect_externs(obj);
5630         err = err ? : bpf_object__finalize_btf(obj);
5631         err = err ? : bpf_object__init_maps(obj, opts);
5632         err = err ? : bpf_object__init_prog_names(obj);
5633         err = err ? : bpf_object__collect_reloc(obj);
5634         if (err)
5635                 goto out;
5636         bpf_object__elf_finish(obj);
5637
5638         bpf_object__for_each_program(prog, obj) {
5639                 prog->sec_def = find_sec_def(prog->section_name);
5640                 if (!prog->sec_def)
5641                         /* couldn't guess, but user might manually specify */
5642                         continue;
5643
5644                 bpf_program__set_type(prog, prog->sec_def->prog_type);
5645                 bpf_program__set_expected_attach_type(prog,
5646                                 prog->sec_def->expected_attach_type);
5647
5648                 if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
5649                     prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
5650                         prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
5651         }
5652
5653         return obj;
5654 out:
5655         bpf_object__close(obj);
5656         return ERR_PTR(err);
5657 }
5658
5659 static struct bpf_object *
5660 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
5661 {
5662         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5663                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
5664         );
5665
5666         /* param validation */
5667         if (!attr->file)
5668                 return NULL;
5669
5670         pr_debug("loading %s\n", attr->file);
5671         return __bpf_object__open(attr->file, NULL, 0, &opts);
5672 }
5673
5674 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
5675 {
5676         return __bpf_object__open_xattr(attr, 0);
5677 }
5678
5679 struct bpf_object *bpf_object__open(const char *path)
5680 {
5681         struct bpf_object_open_attr attr = {
5682                 .file           = path,
5683                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
5684         };
5685
5686         return bpf_object__open_xattr(&attr);
5687 }
5688
5689 struct bpf_object *
5690 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
5691 {
5692         if (!path)
5693                 return ERR_PTR(-EINVAL);
5694
5695         pr_debug("loading %s\n", path);
5696
5697         return __bpf_object__open(path, NULL, 0, opts);
5698 }
5699
5700 struct bpf_object *
5701 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
5702                      const struct bpf_object_open_opts *opts)
5703 {
5704         if (!obj_buf || obj_buf_sz == 0)
5705                 return ERR_PTR(-EINVAL);
5706
5707         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
5708 }
5709
5710 struct bpf_object *
5711 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
5712                         const char *name)
5713 {
5714         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5715                 .object_name = name,
5716                 /* wrong default, but backwards-compatible */
5717                 .relaxed_maps = true,
5718         );
5719
5720         /* returning NULL is wrong, but backwards-compatible */
5721         if (!obj_buf || obj_buf_sz == 0)
5722                 return NULL;
5723
5724         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
5725 }
5726
5727 int bpf_object__unload(struct bpf_object *obj)
5728 {
5729         size_t i;
5730
5731         if (!obj)
5732                 return -EINVAL;
5733
5734         for (i = 0; i < obj->nr_maps; i++) {
5735                 zclose(obj->maps[i].fd);
5736                 if (obj->maps[i].st_ops)
5737                         zfree(&obj->maps[i].st_ops->kern_vdata);
5738         }
5739
5740         for (i = 0; i < obj->nr_programs; i++)
5741                 bpf_program__unload(&obj->programs[i]);
5742
5743         return 0;
5744 }
5745
5746 static int bpf_object__sanitize_maps(struct bpf_object *obj)
5747 {
5748         struct bpf_map *m;
5749
5750         bpf_object__for_each_map(m, obj) {
5751                 if (!bpf_map__is_internal(m))
5752                         continue;
5753                 if (!obj->caps.global_data) {
5754                         pr_warn("kernel doesn't support global data\n");
5755                         return -ENOTSUP;
5756                 }
5757                 if (!obj->caps.array_mmap)
5758                         m->def.map_flags ^= BPF_F_MMAPABLE;
5759         }
5760
5761         return 0;
5762 }
5763
5764 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
5765 {
5766         char sym_type, sym_name[500];
5767         unsigned long long sym_addr;
5768         struct extern_desc *ext;
5769         int ret, err = 0;
5770         FILE *f;
5771
5772         f = fopen("/proc/kallsyms", "r");
5773         if (!f) {
5774                 err = -errno;
5775                 pr_warn("failed to open /proc/kallsyms: %d\n", err);
5776                 return err;
5777         }
5778
5779         while (true) {
5780                 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
5781                              &sym_addr, &sym_type, sym_name);
5782                 if (ret == EOF && feof(f))
5783                         break;
5784                 if (ret != 3) {
5785                         pr_warn("failed to read kallsyms entry: %d\n", ret);
5786                         err = -EINVAL;
5787                         goto out;
5788                 }
5789
5790                 ext = find_extern_by_name(obj, sym_name);
5791                 if (!ext || ext->type != EXT_KSYM)
5792                         continue;
5793
5794                 if (ext->is_set && ext->ksym.addr != sym_addr) {
5795                         pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
5796                                 sym_name, ext->ksym.addr, sym_addr);
5797                         err = -EINVAL;
5798                         goto out;
5799                 }
5800                 if (!ext->is_set) {
5801                         ext->is_set = true;
5802                         ext->ksym.addr = sym_addr;
5803                         pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
5804                 }
5805         }
5806
5807 out:
5808         fclose(f);
5809         return err;
5810 }
5811
5812 static int bpf_object__resolve_externs(struct bpf_object *obj,
5813                                        const char *extra_kconfig)
5814 {
5815         bool need_config = false, need_kallsyms = false;
5816         struct extern_desc *ext;
5817         void *kcfg_data = NULL;
5818         int err, i;
5819
5820         if (obj->nr_extern == 0)
5821                 return 0;
5822
5823         if (obj->kconfig_map_idx >= 0)
5824                 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
5825
5826         for (i = 0; i < obj->nr_extern; i++) {
5827                 ext = &obj->externs[i];
5828
5829                 if (ext->type == EXT_KCFG &&
5830                     strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
5831                         void *ext_val = kcfg_data + ext->kcfg.data_off;
5832                         __u32 kver = get_kernel_version();
5833
5834                         if (!kver) {
5835                                 pr_warn("failed to get kernel version\n");
5836                                 return -EINVAL;
5837                         }
5838                         err = set_kcfg_value_num(ext, ext_val, kver);
5839                         if (err)
5840                                 return err;
5841                         pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
5842                 } else if (ext->type == EXT_KCFG &&
5843                            strncmp(ext->name, "CONFIG_", 7) == 0) {
5844                         need_config = true;
5845                 } else if (ext->type == EXT_KSYM) {
5846                         need_kallsyms = true;
5847                 } else {
5848                         pr_warn("unrecognized extern '%s'\n", ext->name);
5849                         return -EINVAL;
5850                 }
5851         }
5852         if (need_config && extra_kconfig) {
5853                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
5854                 if (err)
5855                         return -EINVAL;
5856                 need_config = false;
5857                 for (i = 0; i < obj->nr_extern; i++) {
5858                         ext = &obj->externs[i];
5859                         if (ext->type == EXT_KCFG && !ext->is_set) {
5860                                 need_config = true;
5861                                 break;
5862                         }
5863                 }
5864         }
5865         if (need_config) {
5866                 err = bpf_object__read_kconfig_file(obj, kcfg_data);
5867                 if (err)
5868                         return -EINVAL;
5869         }
5870         if (need_kallsyms) {
5871                 err = bpf_object__read_kallsyms_file(obj);
5872                 if (err)
5873                         return -EINVAL;
5874         }
5875         for (i = 0; i < obj->nr_extern; i++) {
5876                 ext = &obj->externs[i];
5877
5878                 if (!ext->is_set && !ext->is_weak) {
5879                         pr_warn("extern %s (strong) not resolved\n", ext->name);
5880                         return -ESRCH;
5881                 } else if (!ext->is_set) {
5882                         pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
5883                                  ext->name);
5884                 }
5885         }
5886
5887         return 0;
5888 }
5889
5890 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
5891 {
5892         struct bpf_object *obj;
5893         int err, i;
5894
5895         if (!attr)
5896                 return -EINVAL;
5897         obj = attr->obj;
5898         if (!obj)
5899                 return -EINVAL;
5900
5901         if (obj->loaded) {
5902                 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
5903                 return -EINVAL;
5904         }
5905
5906         err = bpf_object__probe_loading(obj);
5907         err = err ? : bpf_object__probe_caps(obj);
5908         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
5909         err = err ? : bpf_object__sanitize_and_load_btf(obj);
5910         err = err ? : bpf_object__sanitize_maps(obj);
5911         err = err ? : bpf_object__load_vmlinux_btf(obj);
5912         err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
5913         err = err ? : bpf_object__create_maps(obj);
5914         err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
5915         err = err ? : bpf_object__load_progs(obj, attr->log_level);
5916
5917         btf__free(obj->btf_vmlinux);
5918         obj->btf_vmlinux = NULL;
5919
5920         obj->loaded = true; /* doesn't matter if successfully or not */
5921
5922         if (err)
5923                 goto out;
5924
5925         return 0;
5926 out:
5927         /* unpin any maps that were auto-pinned during load */
5928         for (i = 0; i < obj->nr_maps; i++)
5929                 if (obj->maps[i].pinned && !obj->maps[i].reused)
5930                         bpf_map__unpin(&obj->maps[i], NULL);
5931
5932         bpf_object__unload(obj);
5933         pr_warn("failed to load object '%s'\n", obj->path);
5934         return err;
5935 }
5936
5937 int bpf_object__load(struct bpf_object *obj)
5938 {
5939         struct bpf_object_load_attr attr = {
5940                 .obj = obj,
5941         };
5942
5943         return bpf_object__load_xattr(&attr);
5944 }
5945
5946 static int make_parent_dir(const char *path)
5947 {
5948         char *cp, errmsg[STRERR_BUFSIZE];
5949         char *dname, *dir;
5950         int err = 0;
5951
5952         dname = strdup(path);
5953         if (dname == NULL)
5954                 return -ENOMEM;
5955
5956         dir = dirname(dname);
5957         if (mkdir(dir, 0700) && errno != EEXIST)
5958                 err = -errno;
5959
5960         free(dname);
5961         if (err) {
5962                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5963                 pr_warn("failed to mkdir %s: %s\n", path, cp);
5964         }
5965         return err;
5966 }
5967
5968 static int check_path(const char *path)
5969 {
5970         char *cp, errmsg[STRERR_BUFSIZE];
5971         struct statfs st_fs;
5972         char *dname, *dir;
5973         int err = 0;
5974
5975         if (path == NULL)
5976                 return -EINVAL;
5977
5978         dname = strdup(path);
5979         if (dname == NULL)
5980                 return -ENOMEM;
5981
5982         dir = dirname(dname);
5983         if (statfs(dir, &st_fs)) {
5984                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5985                 pr_warn("failed to statfs %s: %s\n", dir, cp);
5986                 err = -errno;
5987         }
5988         free(dname);
5989
5990         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
5991                 pr_warn("specified path %s is not on BPF FS\n", path);
5992                 err = -EINVAL;
5993         }
5994
5995         return err;
5996 }
5997
5998 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
5999                               int instance)
6000 {
6001         char *cp, errmsg[STRERR_BUFSIZE];
6002         int err;
6003
6004         err = make_parent_dir(path);
6005         if (err)
6006                 return err;
6007
6008         err = check_path(path);
6009         if (err)
6010                 return err;
6011
6012         if (prog == NULL) {
6013                 pr_warn("invalid program pointer\n");
6014                 return -EINVAL;
6015         }
6016
6017         if (instance < 0 || instance >= prog->instances.nr) {
6018                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
6019                         instance, prog->section_name, prog->instances.nr);
6020                 return -EINVAL;
6021         }
6022
6023         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
6024                 err = -errno;
6025                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
6026                 pr_warn("failed to pin program: %s\n", cp);
6027                 return err;
6028         }
6029         pr_debug("pinned program '%s'\n", path);
6030
6031         return 0;
6032 }
6033
6034 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
6035                                 int instance)
6036 {
6037         int err;
6038
6039         err = check_path(path);
6040         if (err)
6041                 return err;
6042
6043         if (prog == NULL) {
6044                 pr_warn("invalid program pointer\n");
6045                 return -EINVAL;
6046         }
6047
6048         if (instance < 0 || instance >= prog->instances.nr) {
6049                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
6050                         instance, prog->section_name, prog->instances.nr);
6051                 return -EINVAL;
6052         }
6053
6054         err = unlink(path);
6055         if (err != 0)
6056                 return -errno;
6057         pr_debug("unpinned program '%s'\n", path);
6058
6059         return 0;
6060 }
6061
6062 int bpf_program__pin(struct bpf_program *prog, const char *path)
6063 {
6064         int i, err;
6065
6066         err = make_parent_dir(path);
6067         if (err)
6068                 return err;
6069
6070         err = check_path(path);
6071         if (err)
6072                 return err;
6073
6074         if (prog == NULL) {
6075                 pr_warn("invalid program pointer\n");
6076                 return -EINVAL;
6077         }
6078
6079         if (prog->instances.nr <= 0) {
6080                 pr_warn("no instances of prog %s to pin\n",
6081                            prog->section_name);
6082                 return -EINVAL;
6083         }
6084
6085         if (prog->instances.nr == 1) {
6086                 /* don't create subdirs when pinning single instance */
6087                 return bpf_program__pin_instance(prog, path, 0);
6088         }
6089
6090         for (i = 0; i < prog->instances.nr; i++) {
6091                 char buf[PATH_MAX];
6092                 int len;
6093
6094                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6095                 if (len < 0) {
6096                         err = -EINVAL;
6097                         goto err_unpin;
6098                 } else if (len >= PATH_MAX) {
6099                         err = -ENAMETOOLONG;
6100                         goto err_unpin;
6101                 }
6102
6103                 err = bpf_program__pin_instance(prog, buf, i);
6104                 if (err)
6105                         goto err_unpin;
6106         }
6107
6108         return 0;
6109
6110 err_unpin:
6111         for (i = i - 1; i >= 0; i--) {
6112                 char buf[PATH_MAX];
6113                 int len;
6114
6115                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6116                 if (len < 0)
6117                         continue;
6118                 else if (len >= PATH_MAX)
6119                         continue;
6120
6121                 bpf_program__unpin_instance(prog, buf, i);
6122         }
6123
6124         rmdir(path);
6125
6126         return err;
6127 }
6128
6129 int bpf_program__unpin(struct bpf_program *prog, const char *path)
6130 {
6131         int i, err;
6132
6133         err = check_path(path);
6134         if (err)
6135                 return err;
6136
6137         if (prog == NULL) {
6138                 pr_warn("invalid program pointer\n");
6139                 return -EINVAL;
6140         }
6141
6142         if (prog->instances.nr <= 0) {
6143                 pr_warn("no instances of prog %s to pin\n",
6144                            prog->section_name);
6145                 return -EINVAL;
6146         }
6147
6148         if (prog->instances.nr == 1) {
6149                 /* don't create subdirs when pinning single instance */
6150                 return bpf_program__unpin_instance(prog, path, 0);
6151         }
6152
6153         for (i = 0; i < prog->instances.nr; i++) {
6154                 char buf[PATH_MAX];
6155                 int len;
6156
6157                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6158                 if (len < 0)
6159                         return -EINVAL;
6160                 else if (len >= PATH_MAX)
6161                         return -ENAMETOOLONG;
6162
6163                 err = bpf_program__unpin_instance(prog, buf, i);
6164                 if (err)
6165                         return err;
6166         }
6167
6168         err = rmdir(path);
6169         if (err)
6170                 return -errno;
6171
6172         return 0;
6173 }
6174
6175 int bpf_map__pin(struct bpf_map *map, const char *path)
6176 {
6177         char *cp, errmsg[STRERR_BUFSIZE];
6178         int err;
6179
6180         if (map == NULL) {
6181                 pr_warn("invalid map pointer\n");
6182                 return -EINVAL;
6183         }
6184
6185         if (map->pin_path) {
6186                 if (path && strcmp(path, map->pin_path)) {
6187                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
6188                                 bpf_map__name(map), map->pin_path, path);
6189                         return -EINVAL;
6190                 } else if (map->pinned) {
6191                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
6192                                  bpf_map__name(map), map->pin_path);
6193                         return 0;
6194                 }
6195         } else {
6196                 if (!path) {
6197                         pr_warn("missing a path to pin map '%s' at\n",
6198                                 bpf_map__name(map));
6199                         return -EINVAL;
6200                 } else if (map->pinned) {
6201                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
6202                         return -EEXIST;
6203                 }
6204
6205                 map->pin_path = strdup(path);
6206                 if (!map->pin_path) {
6207                         err = -errno;
6208                         goto out_err;
6209                 }
6210         }
6211
6212         err = make_parent_dir(map->pin_path);
6213         if (err)
6214                 return err;
6215
6216         err = check_path(map->pin_path);
6217         if (err)
6218                 return err;
6219
6220         if (bpf_obj_pin(map->fd, map->pin_path)) {
6221                 err = -errno;
6222                 goto out_err;
6223         }
6224
6225         map->pinned = true;
6226         pr_debug("pinned map '%s'\n", map->pin_path);
6227
6228         return 0;
6229
6230 out_err:
6231         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
6232         pr_warn("failed to pin map: %s\n", cp);
6233         return err;
6234 }
6235
6236 int bpf_map__unpin(struct bpf_map *map, const char *path)
6237 {
6238         int err;
6239
6240         if (map == NULL) {
6241                 pr_warn("invalid map pointer\n");
6242                 return -EINVAL;
6243         }
6244
6245         if (map->pin_path) {
6246                 if (path && strcmp(path, map->pin_path)) {
6247                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
6248                                 bpf_map__name(map), map->pin_path, path);
6249                         return -EINVAL;
6250                 }
6251                 path = map->pin_path;
6252         } else if (!path) {
6253                 pr_warn("no path to unpin map '%s' from\n",
6254                         bpf_map__name(map));
6255                 return -EINVAL;
6256         }
6257
6258         err = check_path(path);
6259         if (err)
6260                 return err;
6261
6262         err = unlink(path);
6263         if (err != 0)
6264                 return -errno;
6265
6266         map->pinned = false;
6267         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
6268
6269         return 0;
6270 }
6271
6272 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
6273 {
6274         char *new = NULL;
6275
6276         if (path) {
6277                 new = strdup(path);
6278                 if (!new)
6279                         return -errno;
6280         }
6281
6282         free(map->pin_path);
6283         map->pin_path = new;
6284         return 0;
6285 }
6286
6287 const char *bpf_map__get_pin_path(const struct bpf_map *map)
6288 {
6289         return map->pin_path;
6290 }
6291
6292 bool bpf_map__is_pinned(const struct bpf_map *map)
6293 {
6294         return map->pinned;
6295 }
6296
6297 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
6298 {
6299         struct bpf_map *map;
6300         int err;
6301
6302         if (!obj)
6303                 return -ENOENT;
6304
6305         if (!obj->loaded) {
6306                 pr_warn("object not yet loaded; load it first\n");
6307                 return -ENOENT;
6308         }
6309
6310         bpf_object__for_each_map(map, obj) {
6311                 char *pin_path = NULL;
6312                 char buf[PATH_MAX];
6313
6314                 if (path) {
6315                         int len;
6316
6317                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
6318                                        bpf_map__name(map));
6319                         if (len < 0) {
6320                                 err = -EINVAL;
6321                                 goto err_unpin_maps;
6322                         } else if (len >= PATH_MAX) {
6323                                 err = -ENAMETOOLONG;
6324                                 goto err_unpin_maps;
6325                         }
6326                         pin_path = buf;
6327                 } else if (!map->pin_path) {
6328                         continue;
6329                 }
6330
6331                 err = bpf_map__pin(map, pin_path);
6332                 if (err)
6333                         goto err_unpin_maps;
6334         }
6335
6336         return 0;
6337
6338 err_unpin_maps:
6339         while ((map = bpf_map__prev(map, obj))) {
6340                 if (!map->pin_path)
6341                         continue;
6342
6343                 bpf_map__unpin(map, NULL);
6344         }
6345
6346         return err;
6347 }
6348
6349 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
6350 {
6351         struct bpf_map *map;
6352         int err;
6353
6354         if (!obj)
6355                 return -ENOENT;
6356
6357         bpf_object__for_each_map(map, obj) {
6358                 char *pin_path = NULL;
6359                 char buf[PATH_MAX];
6360
6361                 if (path) {
6362                         int len;
6363
6364                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
6365                                        bpf_map__name(map));
6366                         if (len < 0)
6367                                 return -EINVAL;
6368                         else if (len >= PATH_MAX)
6369                                 return -ENAMETOOLONG;
6370                         pin_path = buf;
6371                 } else if (!map->pin_path) {
6372                         continue;
6373                 }
6374
6375                 err = bpf_map__unpin(map, pin_path);
6376                 if (err)
6377                         return err;
6378         }
6379
6380         return 0;
6381 }
6382
6383 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
6384 {
6385         struct bpf_program *prog;
6386         int err;
6387
6388         if (!obj)
6389                 return -ENOENT;
6390
6391         if (!obj->loaded) {
6392                 pr_warn("object not yet loaded; load it first\n");
6393                 return -ENOENT;
6394         }
6395
6396         bpf_object__for_each_program(prog, obj) {
6397                 char buf[PATH_MAX];
6398                 int len;
6399
6400                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6401                                prog->pin_name);
6402                 if (len < 0) {
6403                         err = -EINVAL;
6404                         goto err_unpin_programs;
6405                 } else if (len >= PATH_MAX) {
6406                         err = -ENAMETOOLONG;
6407                         goto err_unpin_programs;
6408                 }
6409
6410                 err = bpf_program__pin(prog, buf);
6411                 if (err)
6412                         goto err_unpin_programs;
6413         }
6414
6415         return 0;
6416
6417 err_unpin_programs:
6418         while ((prog = bpf_program__prev(prog, obj))) {
6419                 char buf[PATH_MAX];
6420                 int len;
6421
6422                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6423                                prog->pin_name);
6424                 if (len < 0)
6425                         continue;
6426                 else if (len >= PATH_MAX)
6427                         continue;
6428
6429                 bpf_program__unpin(prog, buf);
6430         }
6431
6432         return err;
6433 }
6434
6435 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
6436 {
6437         struct bpf_program *prog;
6438         int err;
6439
6440         if (!obj)
6441                 return -ENOENT;
6442
6443         bpf_object__for_each_program(prog, obj) {
6444                 char buf[PATH_MAX];
6445                 int len;
6446
6447                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6448                                prog->pin_name);
6449                 if (len < 0)
6450                         return -EINVAL;
6451                 else if (len >= PATH_MAX)
6452                         return -ENAMETOOLONG;
6453
6454                 err = bpf_program__unpin(prog, buf);
6455                 if (err)
6456                         return err;
6457         }
6458
6459         return 0;
6460 }
6461
6462 int bpf_object__pin(struct bpf_object *obj, const char *path)
6463 {
6464         int err;
6465
6466         err = bpf_object__pin_maps(obj, path);
6467         if (err)
6468                 return err;
6469
6470         err = bpf_object__pin_programs(obj, path);
6471         if (err) {
6472                 bpf_object__unpin_maps(obj, path);
6473                 return err;
6474         }
6475
6476         return 0;
6477 }
6478
6479 static void bpf_map__destroy(struct bpf_map *map)
6480 {
6481         if (map->clear_priv)
6482                 map->clear_priv(map, map->priv);
6483         map->priv = NULL;
6484         map->clear_priv = NULL;
6485
6486         if (map->inner_map) {
6487                 bpf_map__destroy(map->inner_map);
6488                 zfree(&map->inner_map);
6489         }
6490
6491         zfree(&map->init_slots);
6492         map->init_slots_sz = 0;
6493
6494         if (map->mmaped) {
6495                 munmap(map->mmaped, bpf_map_mmap_sz(map));
6496                 map->mmaped = NULL;
6497         }
6498
6499         if (map->st_ops) {
6500                 zfree(&map->st_ops->data);
6501                 zfree(&map->st_ops->progs);
6502                 zfree(&map->st_ops->kern_func_off);
6503                 zfree(&map->st_ops);
6504         }
6505
6506         zfree(&map->name);
6507         zfree(&map->pin_path);
6508
6509         if (map->fd >= 0)
6510                 zclose(map->fd);
6511 }
6512
6513 void bpf_object__close(struct bpf_object *obj)
6514 {
6515         size_t i;
6516
6517         if (IS_ERR_OR_NULL(obj))
6518                 return;
6519
6520         if (obj->clear_priv)
6521                 obj->clear_priv(obj, obj->priv);
6522
6523         bpf_object__elf_finish(obj);
6524         bpf_object__unload(obj);
6525         btf__free(obj->btf);
6526         btf_ext__free(obj->btf_ext);
6527
6528         for (i = 0; i < obj->nr_maps; i++)
6529                 bpf_map__destroy(&obj->maps[i]);
6530
6531         zfree(&obj->kconfig);
6532         zfree(&obj->externs);
6533         obj->nr_extern = 0;
6534
6535         zfree(&obj->maps);
6536         obj->nr_maps = 0;
6537
6538         if (obj->programs && obj->nr_programs) {
6539                 for (i = 0; i < obj->nr_programs; i++)
6540                         bpf_program__exit(&obj->programs[i]);
6541         }
6542         zfree(&obj->programs);
6543
6544         list_del(&obj->list);
6545         free(obj);
6546 }
6547
6548 struct bpf_object *
6549 bpf_object__next(struct bpf_object *prev)
6550 {
6551         struct bpf_object *next;
6552
6553         if (!prev)
6554                 next = list_first_entry(&bpf_objects_list,
6555                                         struct bpf_object,
6556                                         list);
6557         else
6558                 next = list_next_entry(prev, list);
6559
6560         /* Empty list is noticed here so don't need checking on entry. */
6561         if (&next->list == &bpf_objects_list)
6562                 return NULL;
6563
6564         return next;
6565 }
6566
6567 const char *bpf_object__name(const struct bpf_object *obj)
6568 {
6569         return obj ? obj->name : ERR_PTR(-EINVAL);
6570 }
6571
6572 unsigned int bpf_object__kversion(const struct bpf_object *obj)
6573 {
6574         return obj ? obj->kern_version : 0;
6575 }
6576
6577 struct btf *bpf_object__btf(const struct bpf_object *obj)
6578 {
6579         return obj ? obj->btf : NULL;
6580 }
6581
6582 int bpf_object__btf_fd(const struct bpf_object *obj)
6583 {
6584         return obj->btf ? btf__fd(obj->btf) : -1;
6585 }
6586
6587 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
6588                          bpf_object_clear_priv_t clear_priv)
6589 {
6590         if (obj->priv && obj->clear_priv)
6591                 obj->clear_priv(obj, obj->priv);
6592
6593         obj->priv = priv;
6594         obj->clear_priv = clear_priv;
6595         return 0;
6596 }
6597
6598 void *bpf_object__priv(const struct bpf_object *obj)
6599 {
6600         return obj ? obj->priv : ERR_PTR(-EINVAL);
6601 }
6602
6603 static struct bpf_program *
6604 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
6605                     bool forward)
6606 {
6607         size_t nr_programs = obj->nr_programs;
6608         ssize_t idx;
6609
6610         if (!nr_programs)
6611                 return NULL;
6612
6613         if (!p)
6614                 /* Iter from the beginning */
6615                 return forward ? &obj->programs[0] :
6616                         &obj->programs[nr_programs - 1];
6617
6618         if (p->obj != obj) {
6619                 pr_warn("error: program handler doesn't match object\n");
6620                 return NULL;
6621         }
6622
6623         idx = (p - obj->programs) + (forward ? 1 : -1);
6624         if (idx >= obj->nr_programs || idx < 0)
6625                 return NULL;
6626         return &obj->programs[idx];
6627 }
6628
6629 struct bpf_program *
6630 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
6631 {
6632         struct bpf_program *prog = prev;
6633
6634         do {
6635                 prog = __bpf_program__iter(prog, obj, true);
6636         } while (prog && bpf_program__is_function_storage(prog, obj));
6637
6638         return prog;
6639 }
6640
6641 struct bpf_program *
6642 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
6643 {
6644         struct bpf_program *prog = next;
6645
6646         do {
6647                 prog = __bpf_program__iter(prog, obj, false);
6648         } while (prog && bpf_program__is_function_storage(prog, obj));
6649
6650         return prog;
6651 }
6652
6653 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
6654                           bpf_program_clear_priv_t clear_priv)
6655 {
6656         if (prog->priv && prog->clear_priv)
6657                 prog->clear_priv(prog, prog->priv);
6658
6659         prog->priv = priv;
6660         prog->clear_priv = clear_priv;
6661         return 0;
6662 }
6663
6664 void *bpf_program__priv(const struct bpf_program *prog)
6665 {
6666         return prog ? prog->priv : ERR_PTR(-EINVAL);
6667 }
6668
6669 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
6670 {
6671         prog->prog_ifindex = ifindex;
6672 }
6673
6674 const char *bpf_program__name(const struct bpf_program *prog)
6675 {
6676         return prog->name;
6677 }
6678
6679 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
6680 {
6681         const char *title;
6682
6683         title = prog->section_name;
6684         if (needs_copy) {
6685                 title = strdup(title);
6686                 if (!title) {
6687                         pr_warn("failed to strdup program title\n");
6688                         return ERR_PTR(-ENOMEM);
6689                 }
6690         }
6691
6692         return title;
6693 }
6694
6695 bool bpf_program__autoload(const struct bpf_program *prog)
6696 {
6697         return prog->load;
6698 }
6699
6700 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
6701 {
6702         if (prog->obj->loaded)
6703                 return -EINVAL;
6704
6705         prog->load = autoload;
6706         return 0;
6707 }
6708
6709 int bpf_program__fd(const struct bpf_program *prog)
6710 {
6711         return bpf_program__nth_fd(prog, 0);
6712 }
6713
6714 size_t bpf_program__size(const struct bpf_program *prog)
6715 {
6716         return prog->insns_cnt * sizeof(struct bpf_insn);
6717 }
6718
6719 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
6720                           bpf_program_prep_t prep)
6721 {
6722         int *instances_fds;
6723
6724         if (nr_instances <= 0 || !prep)
6725                 return -EINVAL;
6726
6727         if (prog->instances.nr > 0 || prog->instances.fds) {
6728                 pr_warn("Can't set pre-processor after loading\n");
6729                 return -EINVAL;
6730         }
6731
6732         instances_fds = malloc(sizeof(int) * nr_instances);
6733         if (!instances_fds) {
6734                 pr_warn("alloc memory failed for fds\n");
6735                 return -ENOMEM;
6736         }
6737
6738         /* fill all fd with -1 */
6739         memset(instances_fds, -1, sizeof(int) * nr_instances);
6740
6741         prog->instances.nr = nr_instances;
6742         prog->instances.fds = instances_fds;
6743         prog->preprocessor = prep;
6744         return 0;
6745 }
6746
6747 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
6748 {
6749         int fd;
6750
6751         if (!prog)
6752                 return -EINVAL;
6753
6754         if (n >= prog->instances.nr || n < 0) {
6755                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
6756                         n, prog->section_name, prog->instances.nr);
6757                 return -EINVAL;
6758         }
6759
6760         fd = prog->instances.fds[n];
6761         if (fd < 0) {
6762                 pr_warn("%dth instance of program '%s' is invalid\n",
6763                         n, prog->section_name);
6764                 return -ENOENT;
6765         }
6766
6767         return fd;
6768 }
6769
6770 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
6771 {
6772         return prog->type;
6773 }
6774
6775 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
6776 {
6777         prog->type = type;
6778 }
6779
6780 static bool bpf_program__is_type(const struct bpf_program *prog,
6781                                  enum bpf_prog_type type)
6782 {
6783         return prog ? (prog->type == type) : false;
6784 }
6785
6786 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
6787 int bpf_program__set_##NAME(struct bpf_program *prog)           \
6788 {                                                               \
6789         if (!prog)                                              \
6790                 return -EINVAL;                                 \
6791         bpf_program__set_type(prog, TYPE);                      \
6792         return 0;                                               \
6793 }                                                               \
6794                                                                 \
6795 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
6796 {                                                               \
6797         return bpf_program__is_type(prog, TYPE);                \
6798 }                                                               \
6799
6800 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
6801 BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
6802 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
6803 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
6804 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
6805 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
6806 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
6807 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
6808 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
6809 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
6810 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
6811 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
6812 BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
6813
6814 enum bpf_attach_type
6815 bpf_program__get_expected_attach_type(struct bpf_program *prog)
6816 {
6817         return prog->expected_attach_type;
6818 }
6819
6820 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
6821                                            enum bpf_attach_type type)
6822 {
6823         prog->expected_attach_type = type;
6824 }
6825
6826 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype_optional,           \
6827                           attachable, attach_btf)                           \
6828         {                                                                   \
6829                 .sec = string,                                              \
6830                 .len = sizeof(string) - 1,                                  \
6831                 .prog_type = ptype,                                         \
6832                 .expected_attach_type = eatype,                             \
6833                 .is_exp_attach_type_optional = eatype_optional,             \
6834                 .is_attachable = attachable,                                \
6835                 .is_attach_btf = attach_btf,                                \
6836         }
6837
6838 /* Programs that can NOT be attached. */
6839 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
6840
6841 /* Programs that can be attached. */
6842 #define BPF_APROG_SEC(string, ptype, atype) \
6843         BPF_PROG_SEC_IMPL(string, ptype, atype, true, 1, 0)
6844
6845 /* Programs that must specify expected attach type at load time. */
6846 #define BPF_EAPROG_SEC(string, ptype, eatype) \
6847         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 1, 0)
6848
6849 /* Programs that use BTF to identify attach point */
6850 #define BPF_PROG_BTF(string, ptype, eatype) \
6851         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 0, 1)
6852
6853 /* Programs that can be attached but attach type can't be identified by section
6854  * name. Kept for backward compatibility.
6855  */
6856 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
6857
6858 #define SEC_DEF(sec_pfx, ptype, ...) {                                      \
6859         .sec = sec_pfx,                                                     \
6860         .len = sizeof(sec_pfx) - 1,                                         \
6861         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
6862         __VA_ARGS__                                                         \
6863 }
6864
6865 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
6866                                       struct bpf_program *prog);
6867 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
6868                                   struct bpf_program *prog);
6869 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
6870                                       struct bpf_program *prog);
6871 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
6872                                      struct bpf_program *prog);
6873 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
6874                                    struct bpf_program *prog);
6875 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
6876                                     struct bpf_program *prog);
6877
6878 static const struct bpf_sec_def section_defs[] = {
6879         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
6880         BPF_PROG_SEC("sk_reuseport",            BPF_PROG_TYPE_SK_REUSEPORT),
6881         SEC_DEF("kprobe/", KPROBE,
6882                 .attach_fn = attach_kprobe),
6883         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
6884         SEC_DEF("kretprobe/", KPROBE,
6885                 .attach_fn = attach_kprobe),
6886         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
6887         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
6888         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
6889         SEC_DEF("tracepoint/", TRACEPOINT,
6890                 .attach_fn = attach_tp),
6891         SEC_DEF("tp/", TRACEPOINT,
6892                 .attach_fn = attach_tp),
6893         SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
6894                 .attach_fn = attach_raw_tp),
6895         SEC_DEF("raw_tp/", RAW_TRACEPOINT,
6896                 .attach_fn = attach_raw_tp),
6897         SEC_DEF("tp_btf/", TRACING,
6898                 .expected_attach_type = BPF_TRACE_RAW_TP,
6899                 .is_attach_btf = true,
6900                 .attach_fn = attach_trace),
6901         SEC_DEF("fentry/", TRACING,
6902                 .expected_attach_type = BPF_TRACE_FENTRY,
6903                 .is_attach_btf = true,
6904                 .attach_fn = attach_trace),
6905         SEC_DEF("fmod_ret/", TRACING,
6906                 .expected_attach_type = BPF_MODIFY_RETURN,
6907                 .is_attach_btf = true,
6908                 .attach_fn = attach_trace),
6909         SEC_DEF("fexit/", TRACING,
6910                 .expected_attach_type = BPF_TRACE_FEXIT,
6911                 .is_attach_btf = true,
6912                 .attach_fn = attach_trace),
6913         SEC_DEF("freplace/", EXT,
6914                 .is_attach_btf = true,
6915                 .attach_fn = attach_trace),
6916         SEC_DEF("lsm/", LSM,
6917                 .is_attach_btf = true,
6918                 .expected_attach_type = BPF_LSM_MAC,
6919                 .attach_fn = attach_lsm),
6920         SEC_DEF("iter/", TRACING,
6921                 .expected_attach_type = BPF_TRACE_ITER,
6922                 .is_attach_btf = true,
6923                 .attach_fn = attach_iter),
6924         BPF_EAPROG_SEC("xdp_devmap/",           BPF_PROG_TYPE_XDP,
6925                                                 BPF_XDP_DEVMAP),
6926         BPF_EAPROG_SEC("xdp_cpumap/",           BPF_PROG_TYPE_XDP,
6927                                                 BPF_XDP_CPUMAP),
6928         BPF_EAPROG_SEC("xdp",                   BPF_PROG_TYPE_XDP,
6929                                                 BPF_XDP),
6930         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
6931         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
6932         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
6933         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
6934         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
6935         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
6936                                                 BPF_CGROUP_INET_INGRESS),
6937         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
6938                                                 BPF_CGROUP_INET_EGRESS),
6939         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
6940         BPF_EAPROG_SEC("cgroup/sock_create",    BPF_PROG_TYPE_CGROUP_SOCK,
6941                                                 BPF_CGROUP_INET_SOCK_CREATE),
6942         BPF_EAPROG_SEC("cgroup/sock_release",   BPF_PROG_TYPE_CGROUP_SOCK,
6943                                                 BPF_CGROUP_INET_SOCK_RELEASE),
6944         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
6945                                                 BPF_CGROUP_INET_SOCK_CREATE),
6946         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
6947                                                 BPF_CGROUP_INET4_POST_BIND),
6948         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
6949                                                 BPF_CGROUP_INET6_POST_BIND),
6950         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
6951                                                 BPF_CGROUP_DEVICE),
6952         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
6953                                                 BPF_CGROUP_SOCK_OPS),
6954         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
6955                                                 BPF_SK_SKB_STREAM_PARSER),
6956         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
6957                                                 BPF_SK_SKB_STREAM_VERDICT),
6958         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
6959         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
6960                                                 BPF_SK_MSG_VERDICT),
6961         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
6962                                                 BPF_LIRC_MODE2),
6963         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
6964                                                 BPF_FLOW_DISSECTOR),
6965         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6966                                                 BPF_CGROUP_INET4_BIND),
6967         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6968                                                 BPF_CGROUP_INET6_BIND),
6969         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6970                                                 BPF_CGROUP_INET4_CONNECT),
6971         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6972                                                 BPF_CGROUP_INET6_CONNECT),
6973         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6974                                                 BPF_CGROUP_UDP4_SENDMSG),
6975         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6976                                                 BPF_CGROUP_UDP6_SENDMSG),
6977         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6978                                                 BPF_CGROUP_UDP4_RECVMSG),
6979         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6980                                                 BPF_CGROUP_UDP6_RECVMSG),
6981         BPF_EAPROG_SEC("cgroup/getpeername4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6982                                                 BPF_CGROUP_INET4_GETPEERNAME),
6983         BPF_EAPROG_SEC("cgroup/getpeername6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6984                                                 BPF_CGROUP_INET6_GETPEERNAME),
6985         BPF_EAPROG_SEC("cgroup/getsockname4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6986                                                 BPF_CGROUP_INET4_GETSOCKNAME),
6987         BPF_EAPROG_SEC("cgroup/getsockname6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6988                                                 BPF_CGROUP_INET6_GETSOCKNAME),
6989         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
6990                                                 BPF_CGROUP_SYSCTL),
6991         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
6992                                                 BPF_CGROUP_GETSOCKOPT),
6993         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
6994                                                 BPF_CGROUP_SETSOCKOPT),
6995         BPF_PROG_SEC("struct_ops",              BPF_PROG_TYPE_STRUCT_OPS),
6996         BPF_EAPROG_SEC("sk_lookup/",            BPF_PROG_TYPE_SK_LOOKUP,
6997                                                 BPF_SK_LOOKUP),
6998 };
6999
7000 #undef BPF_PROG_SEC_IMPL
7001 #undef BPF_PROG_SEC
7002 #undef BPF_APROG_SEC
7003 #undef BPF_EAPROG_SEC
7004 #undef BPF_APROG_COMPAT
7005 #undef SEC_DEF
7006
7007 #define MAX_TYPE_NAME_SIZE 32
7008
7009 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
7010 {
7011         int i, n = ARRAY_SIZE(section_defs);
7012
7013         for (i = 0; i < n; i++) {
7014                 if (strncmp(sec_name,
7015                             section_defs[i].sec, section_defs[i].len))
7016                         continue;
7017                 return &section_defs[i];
7018         }
7019         return NULL;
7020 }
7021
7022 static char *libbpf_get_type_names(bool attach_type)
7023 {
7024         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
7025         char *buf;
7026
7027         buf = malloc(len);
7028         if (!buf)
7029                 return NULL;
7030
7031         buf[0] = '\0';
7032         /* Forge string buf with all available names */
7033         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7034                 if (attach_type && !section_defs[i].is_attachable)
7035                         continue;
7036
7037                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
7038                         free(buf);
7039                         return NULL;
7040                 }
7041                 strcat(buf, " ");
7042                 strcat(buf, section_defs[i].sec);
7043         }
7044
7045         return buf;
7046 }
7047
7048 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
7049                              enum bpf_attach_type *expected_attach_type)
7050 {
7051         const struct bpf_sec_def *sec_def;
7052         char *type_names;
7053
7054         if (!name)
7055                 return -EINVAL;
7056
7057         sec_def = find_sec_def(name);
7058         if (sec_def) {
7059                 *prog_type = sec_def->prog_type;
7060                 *expected_attach_type = sec_def->expected_attach_type;
7061                 return 0;
7062         }
7063
7064         pr_debug("failed to guess program type from ELF section '%s'\n", name);
7065         type_names = libbpf_get_type_names(false);
7066         if (type_names != NULL) {
7067                 pr_debug("supported section(type) names are:%s\n", type_names);
7068                 free(type_names);
7069         }
7070
7071         return -ESRCH;
7072 }
7073
7074 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
7075                                                      size_t offset)
7076 {
7077         struct bpf_map *map;
7078         size_t i;
7079
7080         for (i = 0; i < obj->nr_maps; i++) {
7081                 map = &obj->maps[i];
7082                 if (!bpf_map__is_struct_ops(map))
7083                         continue;
7084                 if (map->sec_offset <= offset &&
7085                     offset - map->sec_offset < map->def.value_size)
7086                         return map;
7087         }
7088
7089         return NULL;
7090 }
7091
7092 /* Collect the reloc from ELF and populate the st_ops->progs[] */
7093 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
7094                                             GElf_Shdr *shdr, Elf_Data *data)
7095 {
7096         const struct btf_member *member;
7097         struct bpf_struct_ops *st_ops;
7098         struct bpf_program *prog;
7099         unsigned int shdr_idx;
7100         const struct btf *btf;
7101         struct bpf_map *map;
7102         Elf_Data *symbols;
7103         unsigned int moff;
7104         const char *name;
7105         __u32 member_idx;
7106         GElf_Sym sym;
7107         GElf_Rel rel;
7108         int i, nrels;
7109
7110         symbols = obj->efile.symbols;
7111         btf = obj->btf;
7112         nrels = shdr->sh_size / shdr->sh_entsize;
7113         for (i = 0; i < nrels; i++) {
7114                 if (!gelf_getrel(data, i, &rel)) {
7115                         pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
7116                         return -LIBBPF_ERRNO__FORMAT;
7117                 }
7118
7119                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
7120                         pr_warn("struct_ops reloc: symbol %zx not found\n",
7121                                 (size_t)GELF_R_SYM(rel.r_info));
7122                         return -LIBBPF_ERRNO__FORMAT;
7123                 }
7124
7125                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
7126                                   sym.st_name) ? : "<?>";
7127                 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
7128                 if (!map) {
7129                         pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
7130                                 (size_t)rel.r_offset);
7131                         return -EINVAL;
7132                 }
7133
7134                 moff = rel.r_offset - map->sec_offset;
7135                 shdr_idx = sym.st_shndx;
7136                 st_ops = map->st_ops;
7137                 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel.r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
7138                          map->name,
7139                          (long long)(rel.r_info >> 32),
7140                          (long long)sym.st_value,
7141                          shdr_idx, (size_t)rel.r_offset,
7142                          map->sec_offset, sym.st_name, name);
7143
7144                 if (shdr_idx >= SHN_LORESERVE) {
7145                         pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
7146                                 map->name, (size_t)rel.r_offset, shdr_idx);
7147                         return -LIBBPF_ERRNO__RELOC;
7148                 }
7149
7150                 member = find_member_by_offset(st_ops->type, moff * 8);
7151                 if (!member) {
7152                         pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
7153                                 map->name, moff);
7154                         return -EINVAL;
7155                 }
7156                 member_idx = member - btf_members(st_ops->type);
7157                 name = btf__name_by_offset(btf, member->name_off);
7158
7159                 if (!resolve_func_ptr(btf, member->type, NULL)) {
7160                         pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
7161                                 map->name, name);
7162                         return -EINVAL;
7163                 }
7164
7165                 prog = bpf_object__find_prog_by_idx(obj, shdr_idx);
7166                 if (!prog) {
7167                         pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
7168                                 map->name, shdr_idx, name);
7169                         return -EINVAL;
7170                 }
7171
7172                 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
7173                         const struct bpf_sec_def *sec_def;
7174
7175                         sec_def = find_sec_def(prog->section_name);
7176                         if (sec_def &&
7177                             sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
7178                                 /* for pr_warn */
7179                                 prog->type = sec_def->prog_type;
7180                                 goto invalid_prog;
7181                         }
7182
7183                         prog->type = BPF_PROG_TYPE_STRUCT_OPS;
7184                         prog->attach_btf_id = st_ops->type_id;
7185                         prog->expected_attach_type = member_idx;
7186                 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
7187                            prog->attach_btf_id != st_ops->type_id ||
7188                            prog->expected_attach_type != member_idx) {
7189                         goto invalid_prog;
7190                 }
7191                 st_ops->progs[member_idx] = prog;
7192         }
7193
7194         return 0;
7195
7196 invalid_prog:
7197         pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
7198                 map->name, prog->name, prog->section_name, prog->type,
7199                 prog->attach_btf_id, prog->expected_attach_type, name);
7200         return -EINVAL;
7201 }
7202
7203 #define BTF_TRACE_PREFIX "btf_trace_"
7204 #define BTF_LSM_PREFIX "bpf_lsm_"
7205 #define BTF_ITER_PREFIX "bpf_iter_"
7206 #define BTF_MAX_NAME_SIZE 128
7207
7208 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
7209                                    const char *name, __u32 kind)
7210 {
7211         char btf_type_name[BTF_MAX_NAME_SIZE];
7212         int ret;
7213
7214         ret = snprintf(btf_type_name, sizeof(btf_type_name),
7215                        "%s%s", prefix, name);
7216         /* snprintf returns the number of characters written excluding the
7217          * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
7218          * indicates truncation.
7219          */
7220         if (ret < 0 || ret >= sizeof(btf_type_name))
7221                 return -ENAMETOOLONG;
7222         return btf__find_by_name_kind(btf, btf_type_name, kind);
7223 }
7224
7225 static inline int __find_vmlinux_btf_id(struct btf *btf, const char *name,
7226                                         enum bpf_attach_type attach_type)
7227 {
7228         int err;
7229
7230         if (attach_type == BPF_TRACE_RAW_TP)
7231                 err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name,
7232                                               BTF_KIND_TYPEDEF);
7233         else if (attach_type == BPF_LSM_MAC)
7234                 err = find_btf_by_prefix_kind(btf, BTF_LSM_PREFIX, name,
7235                                               BTF_KIND_FUNC);
7236         else if (attach_type == BPF_TRACE_ITER)
7237                 err = find_btf_by_prefix_kind(btf, BTF_ITER_PREFIX, name,
7238                                               BTF_KIND_FUNC);
7239         else
7240                 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
7241
7242         if (err <= 0)
7243                 pr_warn("%s is not found in vmlinux BTF\n", name);
7244
7245         return err;
7246 }
7247
7248 int libbpf_find_vmlinux_btf_id(const char *name,
7249                                enum bpf_attach_type attach_type)
7250 {
7251         struct btf *btf;
7252         int err;
7253
7254         btf = libbpf_find_kernel_btf();
7255         if (IS_ERR(btf)) {
7256                 pr_warn("vmlinux BTF is not found\n");
7257                 return -EINVAL;
7258         }
7259
7260         err = __find_vmlinux_btf_id(btf, name, attach_type);
7261         btf__free(btf);
7262         return err;
7263 }
7264
7265 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
7266 {
7267         struct bpf_prog_info_linear *info_linear;
7268         struct bpf_prog_info *info;
7269         struct btf *btf = NULL;
7270         int err = -EINVAL;
7271
7272         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
7273         if (IS_ERR_OR_NULL(info_linear)) {
7274                 pr_warn("failed get_prog_info_linear for FD %d\n",
7275                         attach_prog_fd);
7276                 return -EINVAL;
7277         }
7278         info = &info_linear->info;
7279         if (!info->btf_id) {
7280                 pr_warn("The target program doesn't have BTF\n");
7281                 goto out;
7282         }
7283         if (btf__get_from_id(info->btf_id, &btf)) {
7284                 pr_warn("Failed to get BTF of the program\n");
7285                 goto out;
7286         }
7287         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
7288         btf__free(btf);
7289         if (err <= 0) {
7290                 pr_warn("%s is not found in prog's BTF\n", name);
7291                 goto out;
7292         }
7293 out:
7294         free(info_linear);
7295         return err;
7296 }
7297
7298 static int libbpf_find_attach_btf_id(struct bpf_program *prog)
7299 {
7300         enum bpf_attach_type attach_type = prog->expected_attach_type;
7301         __u32 attach_prog_fd = prog->attach_prog_fd;
7302         const char *name = prog->section_name;
7303         int i, err;
7304
7305         if (!name)
7306                 return -EINVAL;
7307
7308         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7309                 if (!section_defs[i].is_attach_btf)
7310                         continue;
7311                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
7312                         continue;
7313                 if (attach_prog_fd)
7314                         err = libbpf_find_prog_btf_id(name + section_defs[i].len,
7315                                                       attach_prog_fd);
7316                 else
7317                         err = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
7318                                                     name + section_defs[i].len,
7319                                                     attach_type);
7320                 return err;
7321         }
7322         pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
7323         return -ESRCH;
7324 }
7325
7326 int libbpf_attach_type_by_name(const char *name,
7327                                enum bpf_attach_type *attach_type)
7328 {
7329         char *type_names;
7330         int i;
7331
7332         if (!name)
7333                 return -EINVAL;
7334
7335         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7336                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
7337                         continue;
7338                 if (!section_defs[i].is_attachable)
7339                         return -EINVAL;
7340                 *attach_type = section_defs[i].expected_attach_type;
7341                 return 0;
7342         }
7343         pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
7344         type_names = libbpf_get_type_names(true);
7345         if (type_names != NULL) {
7346                 pr_debug("attachable section(type) names are:%s\n", type_names);
7347                 free(type_names);
7348         }
7349
7350         return -EINVAL;
7351 }
7352
7353 int bpf_map__fd(const struct bpf_map *map)
7354 {
7355         return map ? map->fd : -EINVAL;
7356 }
7357
7358 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
7359 {
7360         return map ? &map->def : ERR_PTR(-EINVAL);
7361 }
7362
7363 const char *bpf_map__name(const struct bpf_map *map)
7364 {
7365         return map ? map->name : NULL;
7366 }
7367
7368 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
7369 {
7370         return map->def.type;
7371 }
7372
7373 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
7374 {
7375         if (map->fd >= 0)
7376                 return -EBUSY;
7377         map->def.type = type;
7378         return 0;
7379 }
7380
7381 __u32 bpf_map__map_flags(const struct bpf_map *map)
7382 {
7383         return map->def.map_flags;
7384 }
7385
7386 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
7387 {
7388         if (map->fd >= 0)
7389                 return -EBUSY;
7390         map->def.map_flags = flags;
7391         return 0;
7392 }
7393
7394 __u32 bpf_map__numa_node(const struct bpf_map *map)
7395 {
7396         return map->numa_node;
7397 }
7398
7399 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
7400 {
7401         if (map->fd >= 0)
7402                 return -EBUSY;
7403         map->numa_node = numa_node;
7404         return 0;
7405 }
7406
7407 __u32 bpf_map__key_size(const struct bpf_map *map)
7408 {
7409         return map->def.key_size;
7410 }
7411
7412 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
7413 {
7414         if (map->fd >= 0)
7415                 return -EBUSY;
7416         map->def.key_size = size;
7417         return 0;
7418 }
7419
7420 __u32 bpf_map__value_size(const struct bpf_map *map)
7421 {
7422         return map->def.value_size;
7423 }
7424
7425 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
7426 {
7427         if (map->fd >= 0)
7428                 return -EBUSY;
7429         map->def.value_size = size;
7430         return 0;
7431 }
7432
7433 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
7434 {
7435         return map ? map->btf_key_type_id : 0;
7436 }
7437
7438 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
7439 {
7440         return map ? map->btf_value_type_id : 0;
7441 }
7442
7443 int bpf_map__set_priv(struct bpf_map *map, void *priv,
7444                      bpf_map_clear_priv_t clear_priv)
7445 {
7446         if (!map)
7447                 return -EINVAL;
7448
7449         if (map->priv) {
7450                 if (map->clear_priv)
7451                         map->clear_priv(map, map->priv);
7452         }
7453
7454         map->priv = priv;
7455         map->clear_priv = clear_priv;
7456         return 0;
7457 }
7458
7459 void *bpf_map__priv(const struct bpf_map *map)
7460 {
7461         return map ? map->priv : ERR_PTR(-EINVAL);
7462 }
7463
7464 int bpf_map__set_initial_value(struct bpf_map *map,
7465                                const void *data, size_t size)
7466 {
7467         if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
7468             size != map->def.value_size || map->fd >= 0)
7469                 return -EINVAL;
7470
7471         memcpy(map->mmaped, data, size);
7472         return 0;
7473 }
7474
7475 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
7476 {
7477         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
7478 }
7479
7480 bool bpf_map__is_internal(const struct bpf_map *map)
7481 {
7482         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
7483 }
7484
7485 __u32 bpf_map__ifindex(const struct bpf_map *map)
7486 {
7487         return map->map_ifindex;
7488 }
7489
7490 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
7491 {
7492         if (map->fd >= 0)
7493                 return -EBUSY;
7494         map->map_ifindex = ifindex;
7495         return 0;
7496 }
7497
7498 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
7499 {
7500         if (!bpf_map_type__is_map_in_map(map->def.type)) {
7501                 pr_warn("error: unsupported map type\n");
7502                 return -EINVAL;
7503         }
7504         if (map->inner_map_fd != -1) {
7505                 pr_warn("error: inner_map_fd already specified\n");
7506                 return -EINVAL;
7507         }
7508         map->inner_map_fd = fd;
7509         return 0;
7510 }
7511
7512 static struct bpf_map *
7513 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
7514 {
7515         ssize_t idx;
7516         struct bpf_map *s, *e;
7517
7518         if (!obj || !obj->maps)
7519                 return NULL;
7520
7521         s = obj->maps;
7522         e = obj->maps + obj->nr_maps;
7523
7524         if ((m < s) || (m >= e)) {
7525                 pr_warn("error in %s: map handler doesn't belong to object\n",
7526                          __func__);
7527                 return NULL;
7528         }
7529
7530         idx = (m - obj->maps) + i;
7531         if (idx >= obj->nr_maps || idx < 0)
7532                 return NULL;
7533         return &obj->maps[idx];
7534 }
7535
7536 struct bpf_map *
7537 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
7538 {
7539         if (prev == NULL)
7540                 return obj->maps;
7541
7542         return __bpf_map__iter(prev, obj, 1);
7543 }
7544
7545 struct bpf_map *
7546 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
7547 {
7548         if (next == NULL) {
7549                 if (!obj->nr_maps)
7550                         return NULL;
7551                 return obj->maps + obj->nr_maps - 1;
7552         }
7553
7554         return __bpf_map__iter(next, obj, -1);
7555 }
7556
7557 struct bpf_map *
7558 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
7559 {
7560         struct bpf_map *pos;
7561
7562         bpf_object__for_each_map(pos, obj) {
7563                 if (pos->name && !strcmp(pos->name, name))
7564                         return pos;
7565         }
7566         return NULL;
7567 }
7568
7569 int
7570 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
7571 {
7572         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
7573 }
7574
7575 struct bpf_map *
7576 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
7577 {
7578         return ERR_PTR(-ENOTSUP);
7579 }
7580
7581 long libbpf_get_error(const void *ptr)
7582 {
7583         return PTR_ERR_OR_ZERO(ptr);
7584 }
7585
7586 int bpf_prog_load(const char *file, enum bpf_prog_type type,
7587                   struct bpf_object **pobj, int *prog_fd)
7588 {
7589         struct bpf_prog_load_attr attr;
7590
7591         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
7592         attr.file = file;
7593         attr.prog_type = type;
7594         attr.expected_attach_type = 0;
7595
7596         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
7597 }
7598
7599 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
7600                         struct bpf_object **pobj, int *prog_fd)
7601 {
7602         struct bpf_object_open_attr open_attr = {};
7603         struct bpf_program *prog, *first_prog = NULL;
7604         struct bpf_object *obj;
7605         struct bpf_map *map;
7606         int err;
7607
7608         if (!attr)
7609                 return -EINVAL;
7610         if (!attr->file)
7611                 return -EINVAL;
7612
7613         open_attr.file = attr->file;
7614         open_attr.prog_type = attr->prog_type;
7615
7616         obj = bpf_object__open_xattr(&open_attr);
7617         if (IS_ERR_OR_NULL(obj))
7618                 return -ENOENT;
7619
7620         bpf_object__for_each_program(prog, obj) {
7621                 enum bpf_attach_type attach_type = attr->expected_attach_type;
7622                 /*
7623                  * to preserve backwards compatibility, bpf_prog_load treats
7624                  * attr->prog_type, if specified, as an override to whatever
7625                  * bpf_object__open guessed
7626                  */
7627                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
7628                         bpf_program__set_type(prog, attr->prog_type);
7629                         bpf_program__set_expected_attach_type(prog,
7630                                                               attach_type);
7631                 }
7632                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
7633                         /*
7634                          * we haven't guessed from section name and user
7635                          * didn't provide a fallback type, too bad...
7636                          */
7637                         bpf_object__close(obj);
7638                         return -EINVAL;
7639                 }
7640
7641                 prog->prog_ifindex = attr->ifindex;
7642                 prog->log_level = attr->log_level;
7643                 prog->prog_flags = attr->prog_flags;
7644                 if (!first_prog)
7645                         first_prog = prog;
7646         }
7647
7648         bpf_object__for_each_map(map, obj) {
7649                 if (!bpf_map__is_offload_neutral(map))
7650                         map->map_ifindex = attr->ifindex;
7651         }
7652
7653         if (!first_prog) {
7654                 pr_warn("object file doesn't contain bpf program\n");
7655                 bpf_object__close(obj);
7656                 return -ENOENT;
7657         }
7658
7659         err = bpf_object__load(obj);
7660         if (err) {
7661                 bpf_object__close(obj);
7662                 return err;
7663         }
7664
7665         *pobj = obj;
7666         *prog_fd = bpf_program__fd(first_prog);
7667         return 0;
7668 }
7669
7670 struct bpf_link {
7671         int (*detach)(struct bpf_link *link);
7672         int (*destroy)(struct bpf_link *link);
7673         char *pin_path;         /* NULL, if not pinned */
7674         int fd;                 /* hook FD, -1 if not applicable */
7675         bool disconnected;
7676 };
7677
7678 /* Replace link's underlying BPF program with the new one */
7679 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
7680 {
7681         return bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
7682 }
7683
7684 /* Release "ownership" of underlying BPF resource (typically, BPF program
7685  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
7686  * link, when destructed through bpf_link__destroy() call won't attempt to
7687  * detach/unregisted that BPF resource. This is useful in situations where,
7688  * say, attached BPF program has to outlive userspace program that attached it
7689  * in the system. Depending on type of BPF program, though, there might be
7690  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
7691  * exit of userspace program doesn't trigger automatic detachment and clean up
7692  * inside the kernel.
7693  */
7694 void bpf_link__disconnect(struct bpf_link *link)
7695 {
7696         link->disconnected = true;
7697 }
7698
7699 int bpf_link__destroy(struct bpf_link *link)
7700 {
7701         int err = 0;
7702
7703         if (IS_ERR_OR_NULL(link))
7704                 return 0;
7705
7706         if (!link->disconnected && link->detach)
7707                 err = link->detach(link);
7708         if (link->destroy)
7709                 link->destroy(link);
7710         if (link->pin_path)
7711                 free(link->pin_path);
7712         free(link);
7713
7714         return err;
7715 }
7716
7717 int bpf_link__fd(const struct bpf_link *link)
7718 {
7719         return link->fd;
7720 }
7721
7722 const char *bpf_link__pin_path(const struct bpf_link *link)
7723 {
7724         return link->pin_path;
7725 }
7726
7727 static int bpf_link__detach_fd(struct bpf_link *link)
7728 {
7729         return close(link->fd);
7730 }
7731
7732 struct bpf_link *bpf_link__open(const char *path)
7733 {
7734         struct bpf_link *link;
7735         int fd;
7736
7737         fd = bpf_obj_get(path);
7738         if (fd < 0) {
7739                 fd = -errno;
7740                 pr_warn("failed to open link at %s: %d\n", path, fd);
7741                 return ERR_PTR(fd);
7742         }
7743
7744         link = calloc(1, sizeof(*link));
7745         if (!link) {
7746                 close(fd);
7747                 return ERR_PTR(-ENOMEM);
7748         }
7749         link->detach = &bpf_link__detach_fd;
7750         link->fd = fd;
7751
7752         link->pin_path = strdup(path);
7753         if (!link->pin_path) {
7754                 bpf_link__destroy(link);
7755                 return ERR_PTR(-ENOMEM);
7756         }
7757
7758         return link;
7759 }
7760
7761 int bpf_link__detach(struct bpf_link *link)
7762 {
7763         return bpf_link_detach(link->fd) ? -errno : 0;
7764 }
7765
7766 int bpf_link__pin(struct bpf_link *link, const char *path)
7767 {
7768         int err;
7769
7770         if (link->pin_path)
7771                 return -EBUSY;
7772         err = make_parent_dir(path);
7773         if (err)
7774                 return err;
7775         err = check_path(path);
7776         if (err)
7777                 return err;
7778
7779         link->pin_path = strdup(path);
7780         if (!link->pin_path)
7781                 return -ENOMEM;
7782
7783         if (bpf_obj_pin(link->fd, link->pin_path)) {
7784                 err = -errno;
7785                 zfree(&link->pin_path);
7786                 return err;
7787         }
7788
7789         pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
7790         return 0;
7791 }
7792
7793 int bpf_link__unpin(struct bpf_link *link)
7794 {
7795         int err;
7796
7797         if (!link->pin_path)
7798                 return -EINVAL;
7799
7800         err = unlink(link->pin_path);
7801         if (err != 0)
7802                 return -errno;
7803
7804         pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
7805         zfree(&link->pin_path);
7806         return 0;
7807 }
7808
7809 static int bpf_link__detach_perf_event(struct bpf_link *link)
7810 {
7811         int err;
7812
7813         err = ioctl(link->fd, PERF_EVENT_IOC_DISABLE, 0);
7814         if (err)
7815                 err = -errno;
7816
7817         close(link->fd);
7818         return err;
7819 }
7820
7821 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
7822                                                 int pfd)
7823 {
7824         char errmsg[STRERR_BUFSIZE];
7825         struct bpf_link *link;
7826         int prog_fd, err;
7827
7828         if (pfd < 0) {
7829                 pr_warn("program '%s': invalid perf event FD %d\n",
7830                         bpf_program__title(prog, false), pfd);
7831                 return ERR_PTR(-EINVAL);
7832         }
7833         prog_fd = bpf_program__fd(prog);
7834         if (prog_fd < 0) {
7835                 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
7836                         bpf_program__title(prog, false));
7837                 return ERR_PTR(-EINVAL);
7838         }
7839
7840         link = calloc(1, sizeof(*link));
7841         if (!link)
7842                 return ERR_PTR(-ENOMEM);
7843         link->detach = &bpf_link__detach_perf_event;
7844         link->fd = pfd;
7845
7846         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
7847                 err = -errno;
7848                 free(link);
7849                 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
7850                         bpf_program__title(prog, false), pfd,
7851                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7852                 if (err == -EPROTO)
7853                         pr_warn("program '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
7854                                 bpf_program__title(prog, false), pfd);
7855                 return ERR_PTR(err);
7856         }
7857         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7858                 err = -errno;
7859                 free(link);
7860                 pr_warn("program '%s': failed to enable pfd %d: %s\n",
7861                         bpf_program__title(prog, false), pfd,
7862                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7863                 return ERR_PTR(err);
7864         }
7865         return link;
7866 }
7867
7868 /*
7869  * this function is expected to parse integer in the range of [0, 2^31-1] from
7870  * given file using scanf format string fmt. If actual parsed value is
7871  * negative, the result might be indistinguishable from error
7872  */
7873 static int parse_uint_from_file(const char *file, const char *fmt)
7874 {
7875         char buf[STRERR_BUFSIZE];
7876         int err, ret;
7877         FILE *f;
7878
7879         f = fopen(file, "r");
7880         if (!f) {
7881                 err = -errno;
7882                 pr_debug("failed to open '%s': %s\n", file,
7883                          libbpf_strerror_r(err, buf, sizeof(buf)));
7884                 return err;
7885         }
7886         err = fscanf(f, fmt, &ret);
7887         if (err != 1) {
7888                 err = err == EOF ? -EIO : -errno;
7889                 pr_debug("failed to parse '%s': %s\n", file,
7890                         libbpf_strerror_r(err, buf, sizeof(buf)));
7891                 fclose(f);
7892                 return err;
7893         }
7894         fclose(f);
7895         return ret;
7896 }
7897
7898 static int determine_kprobe_perf_type(void)
7899 {
7900         const char *file = "/sys/bus/event_source/devices/kprobe/type";
7901
7902         return parse_uint_from_file(file, "%d\n");
7903 }
7904
7905 static int determine_uprobe_perf_type(void)
7906 {
7907         const char *file = "/sys/bus/event_source/devices/uprobe/type";
7908
7909         return parse_uint_from_file(file, "%d\n");
7910 }
7911
7912 static int determine_kprobe_retprobe_bit(void)
7913 {
7914         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
7915
7916         return parse_uint_from_file(file, "config:%d\n");
7917 }
7918
7919 static int determine_uprobe_retprobe_bit(void)
7920 {
7921         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
7922
7923         return parse_uint_from_file(file, "config:%d\n");
7924 }
7925
7926 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
7927                                  uint64_t offset, int pid)
7928 {
7929         struct perf_event_attr attr = {};
7930         char errmsg[STRERR_BUFSIZE];
7931         int type, pfd, err;
7932
7933         type = uprobe ? determine_uprobe_perf_type()
7934                       : determine_kprobe_perf_type();
7935         if (type < 0) {
7936                 pr_warn("failed to determine %s perf type: %s\n",
7937                         uprobe ? "uprobe" : "kprobe",
7938                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
7939                 return type;
7940         }
7941         if (retprobe) {
7942                 int bit = uprobe ? determine_uprobe_retprobe_bit()
7943                                  : determine_kprobe_retprobe_bit();
7944
7945                 if (bit < 0) {
7946                         pr_warn("failed to determine %s retprobe bit: %s\n",
7947                                 uprobe ? "uprobe" : "kprobe",
7948                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
7949                         return bit;
7950                 }
7951                 attr.config |= 1 << bit;
7952         }
7953         attr.size = sizeof(attr);
7954         attr.type = type;
7955         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
7956         attr.config2 = offset;           /* kprobe_addr or probe_offset */
7957
7958         /* pid filter is meaningful only for uprobes */
7959         pfd = syscall(__NR_perf_event_open, &attr,
7960                       pid < 0 ? -1 : pid /* pid */,
7961                       pid == -1 ? 0 : -1 /* cpu */,
7962                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7963         if (pfd < 0) {
7964                 err = -errno;
7965                 pr_warn("%s perf_event_open() failed: %s\n",
7966                         uprobe ? "uprobe" : "kprobe",
7967                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7968                 return err;
7969         }
7970         return pfd;
7971 }
7972
7973 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
7974                                             bool retprobe,
7975                                             const char *func_name)
7976 {
7977         char errmsg[STRERR_BUFSIZE];
7978         struct bpf_link *link;
7979         int pfd, err;
7980
7981         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
7982                                     0 /* offset */, -1 /* pid */);
7983         if (pfd < 0) {
7984                 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
7985                         bpf_program__title(prog, false),
7986                         retprobe ? "kretprobe" : "kprobe", func_name,
7987                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7988                 return ERR_PTR(pfd);
7989         }
7990         link = bpf_program__attach_perf_event(prog, pfd);
7991         if (IS_ERR(link)) {
7992                 close(pfd);
7993                 err = PTR_ERR(link);
7994                 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
7995                         bpf_program__title(prog, false),
7996                         retprobe ? "kretprobe" : "kprobe", func_name,
7997                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7998                 return link;
7999         }
8000         return link;
8001 }
8002
8003 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
8004                                       struct bpf_program *prog)
8005 {
8006         const char *func_name;
8007         bool retprobe;
8008
8009         func_name = bpf_program__title(prog, false) + sec->len;
8010         retprobe = strcmp(sec->sec, "kretprobe/") == 0;
8011
8012         return bpf_program__attach_kprobe(prog, retprobe, func_name);
8013 }
8014
8015 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
8016                                             bool retprobe, pid_t pid,
8017                                             const char *binary_path,
8018                                             size_t func_offset)
8019 {
8020         char errmsg[STRERR_BUFSIZE];
8021         struct bpf_link *link;
8022         int pfd, err;
8023
8024         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
8025                                     binary_path, func_offset, pid);
8026         if (pfd < 0) {
8027                 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
8028                         bpf_program__title(prog, false),
8029                         retprobe ? "uretprobe" : "uprobe",
8030                         binary_path, func_offset,
8031                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8032                 return ERR_PTR(pfd);
8033         }
8034         link = bpf_program__attach_perf_event(prog, pfd);
8035         if (IS_ERR(link)) {
8036                 close(pfd);
8037                 err = PTR_ERR(link);
8038                 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
8039                         bpf_program__title(prog, false),
8040                         retprobe ? "uretprobe" : "uprobe",
8041                         binary_path, func_offset,
8042                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8043                 return link;
8044         }
8045         return link;
8046 }
8047
8048 static int determine_tracepoint_id(const char *tp_category,
8049                                    const char *tp_name)
8050 {
8051         char file[PATH_MAX];
8052         int ret;
8053
8054         ret = snprintf(file, sizeof(file),
8055                        "/sys/kernel/debug/tracing/events/%s/%s/id",
8056                        tp_category, tp_name);
8057         if (ret < 0)
8058                 return -errno;
8059         if (ret >= sizeof(file)) {
8060                 pr_debug("tracepoint %s/%s path is too long\n",
8061                          tp_category, tp_name);
8062                 return -E2BIG;
8063         }
8064         return parse_uint_from_file(file, "%d\n");
8065 }
8066
8067 static int perf_event_open_tracepoint(const char *tp_category,
8068                                       const char *tp_name)
8069 {
8070         struct perf_event_attr attr = {};
8071         char errmsg[STRERR_BUFSIZE];
8072         int tp_id, pfd, err;
8073
8074         tp_id = determine_tracepoint_id(tp_category, tp_name);
8075         if (tp_id < 0) {
8076                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
8077                         tp_category, tp_name,
8078                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
8079                 return tp_id;
8080         }
8081
8082         attr.type = PERF_TYPE_TRACEPOINT;
8083         attr.size = sizeof(attr);
8084         attr.config = tp_id;
8085
8086         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
8087                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
8088         if (pfd < 0) {
8089                 err = -errno;
8090                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
8091                         tp_category, tp_name,
8092                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8093                 return err;
8094         }
8095         return pfd;
8096 }
8097
8098 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
8099                                                 const char *tp_category,
8100                                                 const char *tp_name)
8101 {
8102         char errmsg[STRERR_BUFSIZE];
8103         struct bpf_link *link;
8104         int pfd, err;
8105
8106         pfd = perf_event_open_tracepoint(tp_category, tp_name);
8107         if (pfd < 0) {
8108                 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
8109                         bpf_program__title(prog, false),
8110                         tp_category, tp_name,
8111                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8112                 return ERR_PTR(pfd);
8113         }
8114         link = bpf_program__attach_perf_event(prog, pfd);
8115         if (IS_ERR(link)) {
8116                 close(pfd);
8117                 err = PTR_ERR(link);
8118                 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
8119                         bpf_program__title(prog, false),
8120                         tp_category, tp_name,
8121                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8122                 return link;
8123         }
8124         return link;
8125 }
8126
8127 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
8128                                   struct bpf_program *prog)
8129 {
8130         char *sec_name, *tp_cat, *tp_name;
8131         struct bpf_link *link;
8132
8133         sec_name = strdup(bpf_program__title(prog, false));
8134         if (!sec_name)
8135                 return ERR_PTR(-ENOMEM);
8136
8137         /* extract "tp/<category>/<name>" */
8138         tp_cat = sec_name + sec->len;
8139         tp_name = strchr(tp_cat, '/');
8140         if (!tp_name) {
8141                 link = ERR_PTR(-EINVAL);
8142                 goto out;
8143         }
8144         *tp_name = '\0';
8145         tp_name++;
8146
8147         link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
8148 out:
8149         free(sec_name);
8150         return link;
8151 }
8152
8153 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
8154                                                     const char *tp_name)
8155 {
8156         char errmsg[STRERR_BUFSIZE];
8157         struct bpf_link *link;
8158         int prog_fd, pfd;
8159
8160         prog_fd = bpf_program__fd(prog);
8161         if (prog_fd < 0) {
8162                 pr_warn("program '%s': can't attach before loaded\n",
8163                         bpf_program__title(prog, false));
8164                 return ERR_PTR(-EINVAL);
8165         }
8166
8167         link = calloc(1, sizeof(*link));
8168         if (!link)
8169                 return ERR_PTR(-ENOMEM);
8170         link->detach = &bpf_link__detach_fd;
8171
8172         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
8173         if (pfd < 0) {
8174                 pfd = -errno;
8175                 free(link);
8176                 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
8177                         bpf_program__title(prog, false), tp_name,
8178                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8179                 return ERR_PTR(pfd);
8180         }
8181         link->fd = pfd;
8182         return link;
8183 }
8184
8185 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
8186                                       struct bpf_program *prog)
8187 {
8188         const char *tp_name = bpf_program__title(prog, false) + sec->len;
8189
8190         return bpf_program__attach_raw_tracepoint(prog, tp_name);
8191 }
8192
8193 /* Common logic for all BPF program types that attach to a btf_id */
8194 static struct bpf_link *bpf_program__attach_btf_id(struct bpf_program *prog)
8195 {
8196         char errmsg[STRERR_BUFSIZE];
8197         struct bpf_link *link;
8198         int prog_fd, pfd;
8199
8200         prog_fd = bpf_program__fd(prog);
8201         if (prog_fd < 0) {
8202                 pr_warn("program '%s': can't attach before loaded\n",
8203                         bpf_program__title(prog, false));
8204                 return ERR_PTR(-EINVAL);
8205         }
8206
8207         link = calloc(1, sizeof(*link));
8208         if (!link)
8209                 return ERR_PTR(-ENOMEM);
8210         link->detach = &bpf_link__detach_fd;
8211
8212         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
8213         if (pfd < 0) {
8214                 pfd = -errno;
8215                 free(link);
8216                 pr_warn("program '%s': failed to attach: %s\n",
8217                         bpf_program__title(prog, false),
8218                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8219                 return ERR_PTR(pfd);
8220         }
8221         link->fd = pfd;
8222         return (struct bpf_link *)link;
8223 }
8224
8225 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
8226 {
8227         return bpf_program__attach_btf_id(prog);
8228 }
8229
8230 struct bpf_link *bpf_program__attach_lsm(struct bpf_program *prog)
8231 {
8232         return bpf_program__attach_btf_id(prog);
8233 }
8234
8235 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
8236                                      struct bpf_program *prog)
8237 {
8238         return bpf_program__attach_trace(prog);
8239 }
8240
8241 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
8242                                    struct bpf_program *prog)
8243 {
8244         return bpf_program__attach_lsm(prog);
8245 }
8246
8247 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
8248                                     struct bpf_program *prog)
8249 {
8250         return bpf_program__attach_iter(prog, NULL);
8251 }
8252
8253 static struct bpf_link *
8254 bpf_program__attach_fd(struct bpf_program *prog, int target_fd,
8255                        const char *target_name)
8256 {
8257         enum bpf_attach_type attach_type;
8258         char errmsg[STRERR_BUFSIZE];
8259         struct bpf_link *link;
8260         int prog_fd, link_fd;
8261
8262         prog_fd = bpf_program__fd(prog);
8263         if (prog_fd < 0) {
8264                 pr_warn("program '%s': can't attach before loaded\n",
8265                         bpf_program__title(prog, false));
8266                 return ERR_PTR(-EINVAL);
8267         }
8268
8269         link = calloc(1, sizeof(*link));
8270         if (!link)
8271                 return ERR_PTR(-ENOMEM);
8272         link->detach = &bpf_link__detach_fd;
8273
8274         attach_type = bpf_program__get_expected_attach_type(prog);
8275         link_fd = bpf_link_create(prog_fd, target_fd, attach_type, NULL);
8276         if (link_fd < 0) {
8277                 link_fd = -errno;
8278                 free(link);
8279                 pr_warn("program '%s': failed to attach to %s: %s\n",
8280                         bpf_program__title(prog, false), target_name,
8281                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
8282                 return ERR_PTR(link_fd);
8283         }
8284         link->fd = link_fd;
8285         return link;
8286 }
8287
8288 struct bpf_link *
8289 bpf_program__attach_cgroup(struct bpf_program *prog, int cgroup_fd)
8290 {
8291         return bpf_program__attach_fd(prog, cgroup_fd, "cgroup");
8292 }
8293
8294 struct bpf_link *
8295 bpf_program__attach_netns(struct bpf_program *prog, int netns_fd)
8296 {
8297         return bpf_program__attach_fd(prog, netns_fd, "netns");
8298 }
8299
8300 struct bpf_link *bpf_program__attach_xdp(struct bpf_program *prog, int ifindex)
8301 {
8302         /* target_fd/target_ifindex use the same field in LINK_CREATE */
8303         return bpf_program__attach_fd(prog, ifindex, "xdp");
8304 }
8305
8306 struct bpf_link *
8307 bpf_program__attach_iter(struct bpf_program *prog,
8308                          const struct bpf_iter_attach_opts *opts)
8309 {
8310         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
8311         char errmsg[STRERR_BUFSIZE];
8312         struct bpf_link *link;
8313         int prog_fd, link_fd;
8314         __u32 target_fd = 0;
8315
8316         if (!OPTS_VALID(opts, bpf_iter_attach_opts))
8317                 return ERR_PTR(-EINVAL);
8318
8319         link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
8320         link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
8321
8322         prog_fd = bpf_program__fd(prog);
8323         if (prog_fd < 0) {
8324                 pr_warn("program '%s': can't attach before loaded\n",
8325                         bpf_program__title(prog, false));
8326                 return ERR_PTR(-EINVAL);
8327         }
8328
8329         link = calloc(1, sizeof(*link));
8330         if (!link)
8331                 return ERR_PTR(-ENOMEM);
8332         link->detach = &bpf_link__detach_fd;
8333
8334         link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
8335                                   &link_create_opts);
8336         if (link_fd < 0) {
8337                 link_fd = -errno;
8338                 free(link);
8339                 pr_warn("program '%s': failed to attach to iterator: %s\n",
8340                         bpf_program__title(prog, false),
8341                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
8342                 return ERR_PTR(link_fd);
8343         }
8344         link->fd = link_fd;
8345         return link;
8346 }
8347
8348 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
8349 {
8350         const struct bpf_sec_def *sec_def;
8351
8352         sec_def = find_sec_def(bpf_program__title(prog, false));
8353         if (!sec_def || !sec_def->attach_fn)
8354                 return ERR_PTR(-ESRCH);
8355
8356         return sec_def->attach_fn(sec_def, prog);
8357 }
8358
8359 static int bpf_link__detach_struct_ops(struct bpf_link *link)
8360 {
8361         __u32 zero = 0;
8362
8363         if (bpf_map_delete_elem(link->fd, &zero))
8364                 return -errno;
8365
8366         return 0;
8367 }
8368
8369 struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
8370 {
8371         struct bpf_struct_ops *st_ops;
8372         struct bpf_link *link;
8373         __u32 i, zero = 0;
8374         int err;
8375
8376         if (!bpf_map__is_struct_ops(map) || map->fd == -1)
8377                 return ERR_PTR(-EINVAL);
8378
8379         link = calloc(1, sizeof(*link));
8380         if (!link)
8381                 return ERR_PTR(-EINVAL);
8382
8383         st_ops = map->st_ops;
8384         for (i = 0; i < btf_vlen(st_ops->type); i++) {
8385                 struct bpf_program *prog = st_ops->progs[i];
8386                 void *kern_data;
8387                 int prog_fd;
8388
8389                 if (!prog)
8390                         continue;
8391
8392                 prog_fd = bpf_program__fd(prog);
8393                 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
8394                 *(unsigned long *)kern_data = prog_fd;
8395         }
8396
8397         err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
8398         if (err) {
8399                 err = -errno;
8400                 free(link);
8401                 return ERR_PTR(err);
8402         }
8403
8404         link->detach = bpf_link__detach_struct_ops;
8405         link->fd = map->fd;
8406
8407         return link;
8408 }
8409
8410 enum bpf_perf_event_ret
8411 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
8412                            void **copy_mem, size_t *copy_size,
8413                            bpf_perf_event_print_t fn, void *private_data)
8414 {
8415         struct perf_event_mmap_page *header = mmap_mem;
8416         __u64 data_head = ring_buffer_read_head(header);
8417         __u64 data_tail = header->data_tail;
8418         void *base = ((__u8 *)header) + page_size;
8419         int ret = LIBBPF_PERF_EVENT_CONT;
8420         struct perf_event_header *ehdr;
8421         size_t ehdr_size;
8422
8423         while (data_head != data_tail) {
8424                 ehdr = base + (data_tail & (mmap_size - 1));
8425                 ehdr_size = ehdr->size;
8426
8427                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
8428                         void *copy_start = ehdr;
8429                         size_t len_first = base + mmap_size - copy_start;
8430                         size_t len_secnd = ehdr_size - len_first;
8431
8432                         if (*copy_size < ehdr_size) {
8433                                 free(*copy_mem);
8434                                 *copy_mem = malloc(ehdr_size);
8435                                 if (!*copy_mem) {
8436                                         *copy_size = 0;
8437                                         ret = LIBBPF_PERF_EVENT_ERROR;
8438                                         break;
8439                                 }
8440                                 *copy_size = ehdr_size;
8441                         }
8442
8443                         memcpy(*copy_mem, copy_start, len_first);
8444                         memcpy(*copy_mem + len_first, base, len_secnd);
8445                         ehdr = *copy_mem;
8446                 }
8447
8448                 ret = fn(ehdr, private_data);
8449                 data_tail += ehdr_size;
8450                 if (ret != LIBBPF_PERF_EVENT_CONT)
8451                         break;
8452         }
8453
8454         ring_buffer_write_tail(header, data_tail);
8455         return ret;
8456 }
8457
8458 struct perf_buffer;
8459
8460 struct perf_buffer_params {
8461         struct perf_event_attr *attr;
8462         /* if event_cb is specified, it takes precendence */
8463         perf_buffer_event_fn event_cb;
8464         /* sample_cb and lost_cb are higher-level common-case callbacks */
8465         perf_buffer_sample_fn sample_cb;
8466         perf_buffer_lost_fn lost_cb;
8467         void *ctx;
8468         int cpu_cnt;
8469         int *cpus;
8470         int *map_keys;
8471 };
8472
8473 struct perf_cpu_buf {
8474         struct perf_buffer *pb;
8475         void *base; /* mmap()'ed memory */
8476         void *buf; /* for reconstructing segmented data */
8477         size_t buf_size;
8478         int fd;
8479         int cpu;
8480         int map_key;
8481 };
8482
8483 struct perf_buffer {
8484         perf_buffer_event_fn event_cb;
8485         perf_buffer_sample_fn sample_cb;
8486         perf_buffer_lost_fn lost_cb;
8487         void *ctx; /* passed into callbacks */
8488
8489         size_t page_size;
8490         size_t mmap_size;
8491         struct perf_cpu_buf **cpu_bufs;
8492         struct epoll_event *events;
8493         int cpu_cnt; /* number of allocated CPU buffers */
8494         int epoll_fd; /* perf event FD */
8495         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
8496 };
8497
8498 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
8499                                       struct perf_cpu_buf *cpu_buf)
8500 {
8501         if (!cpu_buf)
8502                 return;
8503         if (cpu_buf->base &&
8504             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
8505                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
8506         if (cpu_buf->fd >= 0) {
8507                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
8508                 close(cpu_buf->fd);
8509         }
8510         free(cpu_buf->buf);
8511         free(cpu_buf);
8512 }
8513
8514 void perf_buffer__free(struct perf_buffer *pb)
8515 {
8516         int i;
8517
8518         if (IS_ERR_OR_NULL(pb))
8519                 return;
8520         if (pb->cpu_bufs) {
8521                 for (i = 0; i < pb->cpu_cnt; i++) {
8522                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
8523
8524                         if (!cpu_buf)
8525                                 continue;
8526
8527                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
8528                         perf_buffer__free_cpu_buf(pb, cpu_buf);
8529                 }
8530                 free(pb->cpu_bufs);
8531         }
8532         if (pb->epoll_fd >= 0)
8533                 close(pb->epoll_fd);
8534         free(pb->events);
8535         free(pb);
8536 }
8537
8538 static struct perf_cpu_buf *
8539 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
8540                           int cpu, int map_key)
8541 {
8542         struct perf_cpu_buf *cpu_buf;
8543         char msg[STRERR_BUFSIZE];
8544         int err;
8545
8546         cpu_buf = calloc(1, sizeof(*cpu_buf));
8547         if (!cpu_buf)
8548                 return ERR_PTR(-ENOMEM);
8549
8550         cpu_buf->pb = pb;
8551         cpu_buf->cpu = cpu;
8552         cpu_buf->map_key = map_key;
8553
8554         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
8555                               -1, PERF_FLAG_FD_CLOEXEC);
8556         if (cpu_buf->fd < 0) {
8557                 err = -errno;
8558                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
8559                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8560                 goto error;
8561         }
8562
8563         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
8564                              PROT_READ | PROT_WRITE, MAP_SHARED,
8565                              cpu_buf->fd, 0);
8566         if (cpu_buf->base == MAP_FAILED) {
8567                 cpu_buf->base = NULL;
8568                 err = -errno;
8569                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
8570                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8571                 goto error;
8572         }
8573
8574         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
8575                 err = -errno;
8576                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
8577                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8578                 goto error;
8579         }
8580
8581         return cpu_buf;
8582
8583 error:
8584         perf_buffer__free_cpu_buf(pb, cpu_buf);
8585         return (struct perf_cpu_buf *)ERR_PTR(err);
8586 }
8587
8588 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
8589                                               struct perf_buffer_params *p);
8590
8591 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
8592                                      const struct perf_buffer_opts *opts)
8593 {
8594         struct perf_buffer_params p = {};
8595         struct perf_event_attr attr = { 0, };
8596
8597         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
8598         attr.type = PERF_TYPE_SOFTWARE;
8599         attr.sample_type = PERF_SAMPLE_RAW;
8600         attr.sample_period = 1;
8601         attr.wakeup_events = 1;
8602
8603         p.attr = &attr;
8604         p.sample_cb = opts ? opts->sample_cb : NULL;
8605         p.lost_cb = opts ? opts->lost_cb : NULL;
8606         p.ctx = opts ? opts->ctx : NULL;
8607
8608         return __perf_buffer__new(map_fd, page_cnt, &p);
8609 }
8610
8611 struct perf_buffer *
8612 perf_buffer__new_raw(int map_fd, size_t page_cnt,
8613                      const struct perf_buffer_raw_opts *opts)
8614 {
8615         struct perf_buffer_params p = {};
8616
8617         p.attr = opts->attr;
8618         p.event_cb = opts->event_cb;
8619         p.ctx = opts->ctx;
8620         p.cpu_cnt = opts->cpu_cnt;
8621         p.cpus = opts->cpus;
8622         p.map_keys = opts->map_keys;
8623
8624         return __perf_buffer__new(map_fd, page_cnt, &p);
8625 }
8626
8627 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
8628                                               struct perf_buffer_params *p)
8629 {
8630         const char *online_cpus_file = "/sys/devices/system/cpu/online";
8631         struct bpf_map_info map;
8632         char msg[STRERR_BUFSIZE];
8633         struct perf_buffer *pb;
8634         bool *online = NULL;
8635         __u32 map_info_len;
8636         int err, i, j, n;
8637
8638         if (page_cnt & (page_cnt - 1)) {
8639                 pr_warn("page count should be power of two, but is %zu\n",
8640                         page_cnt);
8641                 return ERR_PTR(-EINVAL);
8642         }
8643
8644         /* best-effort sanity checks */
8645         memset(&map, 0, sizeof(map));
8646         map_info_len = sizeof(map);
8647         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
8648         if (err) {
8649                 err = -errno;
8650                 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
8651                  * -EBADFD, -EFAULT, or -E2BIG on real error
8652                  */
8653                 if (err != -EINVAL) {
8654                         pr_warn("failed to get map info for map FD %d: %s\n",
8655                                 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
8656                         return ERR_PTR(err);
8657                 }
8658                 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
8659                          map_fd);
8660         } else {
8661                 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
8662                         pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
8663                                 map.name);
8664                         return ERR_PTR(-EINVAL);
8665                 }
8666         }
8667
8668         pb = calloc(1, sizeof(*pb));
8669         if (!pb)
8670                 return ERR_PTR(-ENOMEM);
8671
8672         pb->event_cb = p->event_cb;
8673         pb->sample_cb = p->sample_cb;
8674         pb->lost_cb = p->lost_cb;
8675         pb->ctx = p->ctx;
8676
8677         pb->page_size = getpagesize();
8678         pb->mmap_size = pb->page_size * page_cnt;
8679         pb->map_fd = map_fd;
8680
8681         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
8682         if (pb->epoll_fd < 0) {
8683                 err = -errno;
8684                 pr_warn("failed to create epoll instance: %s\n",
8685                         libbpf_strerror_r(err, msg, sizeof(msg)));
8686                 goto error;
8687         }
8688
8689         if (p->cpu_cnt > 0) {
8690                 pb->cpu_cnt = p->cpu_cnt;
8691         } else {
8692                 pb->cpu_cnt = libbpf_num_possible_cpus();
8693                 if (pb->cpu_cnt < 0) {
8694                         err = pb->cpu_cnt;
8695                         goto error;
8696                 }
8697                 if (map.max_entries && map.max_entries < pb->cpu_cnt)
8698                         pb->cpu_cnt = map.max_entries;
8699         }
8700
8701         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
8702         if (!pb->events) {
8703                 err = -ENOMEM;
8704                 pr_warn("failed to allocate events: out of memory\n");
8705                 goto error;
8706         }
8707         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
8708         if (!pb->cpu_bufs) {
8709                 err = -ENOMEM;
8710                 pr_warn("failed to allocate buffers: out of memory\n");
8711                 goto error;
8712         }
8713
8714         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
8715         if (err) {
8716                 pr_warn("failed to get online CPU mask: %d\n", err);
8717                 goto error;
8718         }
8719
8720         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
8721                 struct perf_cpu_buf *cpu_buf;
8722                 int cpu, map_key;
8723
8724                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
8725                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
8726
8727                 /* in case user didn't explicitly requested particular CPUs to
8728                  * be attached to, skip offline/not present CPUs
8729                  */
8730                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
8731                         continue;
8732
8733                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
8734                 if (IS_ERR(cpu_buf)) {
8735                         err = PTR_ERR(cpu_buf);
8736                         goto error;
8737                 }
8738
8739                 pb->cpu_bufs[j] = cpu_buf;
8740
8741                 err = bpf_map_update_elem(pb->map_fd, &map_key,
8742                                           &cpu_buf->fd, 0);
8743                 if (err) {
8744                         err = -errno;
8745                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
8746                                 cpu, map_key, cpu_buf->fd,
8747                                 libbpf_strerror_r(err, msg, sizeof(msg)));
8748                         goto error;
8749                 }
8750
8751                 pb->events[j].events = EPOLLIN;
8752                 pb->events[j].data.ptr = cpu_buf;
8753                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
8754                               &pb->events[j]) < 0) {
8755                         err = -errno;
8756                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
8757                                 cpu, cpu_buf->fd,
8758                                 libbpf_strerror_r(err, msg, sizeof(msg)));
8759                         goto error;
8760                 }
8761                 j++;
8762         }
8763         pb->cpu_cnt = j;
8764         free(online);
8765
8766         return pb;
8767
8768 error:
8769         free(online);
8770         if (pb)
8771                 perf_buffer__free(pb);
8772         return ERR_PTR(err);
8773 }
8774
8775 struct perf_sample_raw {
8776         struct perf_event_header header;
8777         uint32_t size;
8778         char data[];
8779 };
8780
8781 struct perf_sample_lost {
8782         struct perf_event_header header;
8783         uint64_t id;
8784         uint64_t lost;
8785         uint64_t sample_id;
8786 };
8787
8788 static enum bpf_perf_event_ret
8789 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
8790 {
8791         struct perf_cpu_buf *cpu_buf = ctx;
8792         struct perf_buffer *pb = cpu_buf->pb;
8793         void *data = e;
8794
8795         /* user wants full control over parsing perf event */
8796         if (pb->event_cb)
8797                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
8798
8799         switch (e->type) {
8800         case PERF_RECORD_SAMPLE: {
8801                 struct perf_sample_raw *s = data;
8802
8803                 if (pb->sample_cb)
8804                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
8805                 break;
8806         }
8807         case PERF_RECORD_LOST: {
8808                 struct perf_sample_lost *s = data;
8809
8810                 if (pb->lost_cb)
8811                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
8812                 break;
8813         }
8814         default:
8815                 pr_warn("unknown perf sample type %d\n", e->type);
8816                 return LIBBPF_PERF_EVENT_ERROR;
8817         }
8818         return LIBBPF_PERF_EVENT_CONT;
8819 }
8820
8821 static int perf_buffer__process_records(struct perf_buffer *pb,
8822                                         struct perf_cpu_buf *cpu_buf)
8823 {
8824         enum bpf_perf_event_ret ret;
8825
8826         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
8827                                          pb->page_size, &cpu_buf->buf,
8828                                          &cpu_buf->buf_size,
8829                                          perf_buffer__process_record, cpu_buf);
8830         if (ret != LIBBPF_PERF_EVENT_CONT)
8831                 return ret;
8832         return 0;
8833 }
8834
8835 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
8836 {
8837         int i, cnt, err;
8838
8839         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
8840         for (i = 0; i < cnt; i++) {
8841                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
8842
8843                 err = perf_buffer__process_records(pb, cpu_buf);
8844                 if (err) {
8845                         pr_warn("error while processing records: %d\n", err);
8846                         return err;
8847                 }
8848         }
8849         return cnt < 0 ? -errno : cnt;
8850 }
8851
8852 int perf_buffer__consume(struct perf_buffer *pb)
8853 {
8854         int i, err;
8855
8856         for (i = 0; i < pb->cpu_cnt; i++) {
8857                 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
8858
8859                 if (!cpu_buf)
8860                         continue;
8861
8862                 err = perf_buffer__process_records(pb, cpu_buf);
8863                 if (err) {
8864                         pr_warn("error while processing records: %d\n", err);
8865                         return err;
8866                 }
8867         }
8868         return 0;
8869 }
8870
8871 struct bpf_prog_info_array_desc {
8872         int     array_offset;   /* e.g. offset of jited_prog_insns */
8873         int     count_offset;   /* e.g. offset of jited_prog_len */
8874         int     size_offset;    /* > 0: offset of rec size,
8875                                  * < 0: fix size of -size_offset
8876                                  */
8877 };
8878
8879 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
8880         [BPF_PROG_INFO_JITED_INSNS] = {
8881                 offsetof(struct bpf_prog_info, jited_prog_insns),
8882                 offsetof(struct bpf_prog_info, jited_prog_len),
8883                 -1,
8884         },
8885         [BPF_PROG_INFO_XLATED_INSNS] = {
8886                 offsetof(struct bpf_prog_info, xlated_prog_insns),
8887                 offsetof(struct bpf_prog_info, xlated_prog_len),
8888                 -1,
8889         },
8890         [BPF_PROG_INFO_MAP_IDS] = {
8891                 offsetof(struct bpf_prog_info, map_ids),
8892                 offsetof(struct bpf_prog_info, nr_map_ids),
8893                 -(int)sizeof(__u32),
8894         },
8895         [BPF_PROG_INFO_JITED_KSYMS] = {
8896                 offsetof(struct bpf_prog_info, jited_ksyms),
8897                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
8898                 -(int)sizeof(__u64),
8899         },
8900         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
8901                 offsetof(struct bpf_prog_info, jited_func_lens),
8902                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
8903                 -(int)sizeof(__u32),
8904         },
8905         [BPF_PROG_INFO_FUNC_INFO] = {
8906                 offsetof(struct bpf_prog_info, func_info),
8907                 offsetof(struct bpf_prog_info, nr_func_info),
8908                 offsetof(struct bpf_prog_info, func_info_rec_size),
8909         },
8910         [BPF_PROG_INFO_LINE_INFO] = {
8911                 offsetof(struct bpf_prog_info, line_info),
8912                 offsetof(struct bpf_prog_info, nr_line_info),
8913                 offsetof(struct bpf_prog_info, line_info_rec_size),
8914         },
8915         [BPF_PROG_INFO_JITED_LINE_INFO] = {
8916                 offsetof(struct bpf_prog_info, jited_line_info),
8917                 offsetof(struct bpf_prog_info, nr_jited_line_info),
8918                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
8919         },
8920         [BPF_PROG_INFO_PROG_TAGS] = {
8921                 offsetof(struct bpf_prog_info, prog_tags),
8922                 offsetof(struct bpf_prog_info, nr_prog_tags),
8923                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
8924         },
8925
8926 };
8927
8928 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
8929                                            int offset)
8930 {
8931         __u32 *array = (__u32 *)info;
8932
8933         if (offset >= 0)
8934                 return array[offset / sizeof(__u32)];
8935         return -(int)offset;
8936 }
8937
8938 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
8939                                            int offset)
8940 {
8941         __u64 *array = (__u64 *)info;
8942
8943         if (offset >= 0)
8944                 return array[offset / sizeof(__u64)];
8945         return -(int)offset;
8946 }
8947
8948 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
8949                                          __u32 val)
8950 {
8951         __u32 *array = (__u32 *)info;
8952
8953         if (offset >= 0)
8954                 array[offset / sizeof(__u32)] = val;
8955 }
8956
8957 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
8958                                          __u64 val)
8959 {
8960         __u64 *array = (__u64 *)info;
8961
8962         if (offset >= 0)
8963                 array[offset / sizeof(__u64)] = val;
8964 }
8965
8966 struct bpf_prog_info_linear *
8967 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
8968 {
8969         struct bpf_prog_info_linear *info_linear;
8970         struct bpf_prog_info info = {};
8971         __u32 info_len = sizeof(info);
8972         __u32 data_len = 0;
8973         int i, err;
8974         void *ptr;
8975
8976         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
8977                 return ERR_PTR(-EINVAL);
8978
8979         /* step 1: get array dimensions */
8980         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
8981         if (err) {
8982                 pr_debug("can't get prog info: %s", strerror(errno));
8983                 return ERR_PTR(-EFAULT);
8984         }
8985
8986         /* step 2: calculate total size of all arrays */
8987         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8988                 bool include_array = (arrays & (1UL << i)) > 0;
8989                 struct bpf_prog_info_array_desc *desc;
8990                 __u32 count, size;
8991
8992                 desc = bpf_prog_info_array_desc + i;
8993
8994                 /* kernel is too old to support this field */
8995                 if (info_len < desc->array_offset + sizeof(__u32) ||
8996                     info_len < desc->count_offset + sizeof(__u32) ||
8997                     (desc->size_offset > 0 && info_len < desc->size_offset))
8998                         include_array = false;
8999
9000                 if (!include_array) {
9001                         arrays &= ~(1UL << i);  /* clear the bit */
9002                         continue;
9003                 }
9004
9005                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
9006                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
9007
9008                 data_len += count * size;
9009         }
9010
9011         /* step 3: allocate continuous memory */
9012         data_len = roundup(data_len, sizeof(__u64));
9013         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
9014         if (!info_linear)
9015                 return ERR_PTR(-ENOMEM);
9016
9017         /* step 4: fill data to info_linear->info */
9018         info_linear->arrays = arrays;
9019         memset(&info_linear->info, 0, sizeof(info));
9020         ptr = info_linear->data;
9021
9022         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9023                 struct bpf_prog_info_array_desc *desc;
9024                 __u32 count, size;
9025
9026                 if ((arrays & (1UL << i)) == 0)
9027                         continue;
9028
9029                 desc  = bpf_prog_info_array_desc + i;
9030                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
9031                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
9032                 bpf_prog_info_set_offset_u32(&info_linear->info,
9033                                              desc->count_offset, count);
9034                 bpf_prog_info_set_offset_u32(&info_linear->info,
9035                                              desc->size_offset, size);
9036                 bpf_prog_info_set_offset_u64(&info_linear->info,
9037                                              desc->array_offset,
9038                                              ptr_to_u64(ptr));
9039                 ptr += count * size;
9040         }
9041
9042         /* step 5: call syscall again to get required arrays */
9043         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
9044         if (err) {
9045                 pr_debug("can't get prog info: %s", strerror(errno));
9046                 free(info_linear);
9047                 return ERR_PTR(-EFAULT);
9048         }
9049
9050         /* step 6: verify the data */
9051         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9052                 struct bpf_prog_info_array_desc *desc;
9053                 __u32 v1, v2;
9054
9055                 if ((arrays & (1UL << i)) == 0)
9056                         continue;
9057
9058                 desc = bpf_prog_info_array_desc + i;
9059                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
9060                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
9061                                                    desc->count_offset);
9062                 if (v1 != v2)
9063                         pr_warn("%s: mismatch in element count\n", __func__);
9064
9065                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
9066                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
9067                                                    desc->size_offset);
9068                 if (v1 != v2)
9069                         pr_warn("%s: mismatch in rec size\n", __func__);
9070         }
9071
9072         /* step 7: update info_len and data_len */
9073         info_linear->info_len = sizeof(struct bpf_prog_info);
9074         info_linear->data_len = data_len;
9075
9076         return info_linear;
9077 }
9078
9079 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
9080 {
9081         int i;
9082
9083         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9084                 struct bpf_prog_info_array_desc *desc;
9085                 __u64 addr, offs;
9086
9087                 if ((info_linear->arrays & (1UL << i)) == 0)
9088                         continue;
9089
9090                 desc = bpf_prog_info_array_desc + i;
9091                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
9092                                                      desc->array_offset);
9093                 offs = addr - ptr_to_u64(info_linear->data);
9094                 bpf_prog_info_set_offset_u64(&info_linear->info,
9095                                              desc->array_offset, offs);
9096         }
9097 }
9098
9099 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
9100 {
9101         int i;
9102
9103         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9104                 struct bpf_prog_info_array_desc *desc;
9105                 __u64 addr, offs;
9106
9107                 if ((info_linear->arrays & (1UL << i)) == 0)
9108                         continue;
9109
9110                 desc = bpf_prog_info_array_desc + i;
9111                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
9112                                                      desc->array_offset);
9113                 addr = offs + ptr_to_u64(info_linear->data);
9114                 bpf_prog_info_set_offset_u64(&info_linear->info,
9115                                              desc->array_offset, addr);
9116         }
9117 }
9118
9119 int bpf_program__set_attach_target(struct bpf_program *prog,
9120                                    int attach_prog_fd,
9121                                    const char *attach_func_name)
9122 {
9123         int btf_id;
9124
9125         if (!prog || attach_prog_fd < 0 || !attach_func_name)
9126                 return -EINVAL;
9127
9128         if (attach_prog_fd)
9129                 btf_id = libbpf_find_prog_btf_id(attach_func_name,
9130                                                  attach_prog_fd);
9131         else
9132                 btf_id = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
9133                                                attach_func_name,
9134                                                prog->expected_attach_type);
9135
9136         if (btf_id < 0)
9137                 return btf_id;
9138
9139         prog->attach_btf_id = btf_id;
9140         prog->attach_prog_fd = attach_prog_fd;
9141         return 0;
9142 }
9143
9144 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
9145 {
9146         int err = 0, n, len, start, end = -1;
9147         bool *tmp;
9148
9149         *mask = NULL;
9150         *mask_sz = 0;
9151
9152         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
9153         while (*s) {
9154                 if (*s == ',' || *s == '\n') {
9155                         s++;
9156                         continue;
9157                 }
9158                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
9159                 if (n <= 0 || n > 2) {
9160                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
9161                         err = -EINVAL;
9162                         goto cleanup;
9163                 } else if (n == 1) {
9164                         end = start;
9165                 }
9166                 if (start < 0 || start > end) {
9167                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
9168                                 start, end, s);
9169                         err = -EINVAL;
9170                         goto cleanup;
9171                 }
9172                 tmp = realloc(*mask, end + 1);
9173                 if (!tmp) {
9174                         err = -ENOMEM;
9175                         goto cleanup;
9176                 }
9177                 *mask = tmp;
9178                 memset(tmp + *mask_sz, 0, start - *mask_sz);
9179                 memset(tmp + start, 1, end - start + 1);
9180                 *mask_sz = end + 1;
9181                 s += len;
9182         }
9183         if (!*mask_sz) {
9184                 pr_warn("Empty CPU range\n");
9185                 return -EINVAL;
9186         }
9187         return 0;
9188 cleanup:
9189         free(*mask);
9190         *mask = NULL;
9191         return err;
9192 }
9193
9194 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
9195 {
9196         int fd, err = 0, len;
9197         char buf[128];
9198
9199         fd = open(fcpu, O_RDONLY);
9200         if (fd < 0) {
9201                 err = -errno;
9202                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
9203                 return err;
9204         }
9205         len = read(fd, buf, sizeof(buf));
9206         close(fd);
9207         if (len <= 0) {
9208                 err = len ? -errno : -EINVAL;
9209                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
9210                 return err;
9211         }
9212         if (len >= sizeof(buf)) {
9213                 pr_warn("CPU mask is too big in file %s\n", fcpu);
9214                 return -E2BIG;
9215         }
9216         buf[len] = '\0';
9217
9218         return parse_cpu_mask_str(buf, mask, mask_sz);
9219 }
9220
9221 int libbpf_num_possible_cpus(void)
9222 {
9223         static const char *fcpu = "/sys/devices/system/cpu/possible";
9224         static int cpus;
9225         int err, n, i, tmp_cpus;
9226         bool *mask;
9227
9228         tmp_cpus = READ_ONCE(cpus);
9229         if (tmp_cpus > 0)
9230                 return tmp_cpus;
9231
9232         err = parse_cpu_mask_file(fcpu, &mask, &n);
9233         if (err)
9234                 return err;
9235
9236         tmp_cpus = 0;
9237         for (i = 0; i < n; i++) {
9238                 if (mask[i])
9239                         tmp_cpus++;
9240         }
9241         free(mask);
9242
9243         WRITE_ONCE(cpus, tmp_cpus);
9244         return tmp_cpus;
9245 }
9246
9247 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
9248                               const struct bpf_object_open_opts *opts)
9249 {
9250         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
9251                 .object_name = s->name,
9252         );
9253         struct bpf_object *obj;
9254         int i;
9255
9256         /* Attempt to preserve opts->object_name, unless overriden by user
9257          * explicitly. Overwriting object name for skeletons is discouraged,
9258          * as it breaks global data maps, because they contain object name
9259          * prefix as their own map name prefix. When skeleton is generated,
9260          * bpftool is making an assumption that this name will stay the same.
9261          */
9262         if (opts) {
9263                 memcpy(&skel_opts, opts, sizeof(*opts));
9264                 if (!opts->object_name)
9265                         skel_opts.object_name = s->name;
9266         }
9267
9268         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
9269         if (IS_ERR(obj)) {
9270                 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
9271                         s->name, PTR_ERR(obj));
9272                 return PTR_ERR(obj);
9273         }
9274
9275         *s->obj = obj;
9276
9277         for (i = 0; i < s->map_cnt; i++) {
9278                 struct bpf_map **map = s->maps[i].map;
9279                 const char *name = s->maps[i].name;
9280                 void **mmaped = s->maps[i].mmaped;
9281
9282                 *map = bpf_object__find_map_by_name(obj, name);
9283                 if (!*map) {
9284                         pr_warn("failed to find skeleton map '%s'\n", name);
9285                         return -ESRCH;
9286                 }
9287
9288                 /* externs shouldn't be pre-setup from user code */
9289                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
9290                         *mmaped = (*map)->mmaped;
9291         }
9292
9293         for (i = 0; i < s->prog_cnt; i++) {
9294                 struct bpf_program **prog = s->progs[i].prog;
9295                 const char *name = s->progs[i].name;
9296
9297                 *prog = bpf_object__find_program_by_name(obj, name);
9298                 if (!*prog) {
9299                         pr_warn("failed to find skeleton program '%s'\n", name);
9300                         return -ESRCH;
9301                 }
9302         }
9303
9304         return 0;
9305 }
9306
9307 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
9308 {
9309         int i, err;
9310
9311         err = bpf_object__load(*s->obj);
9312         if (err) {
9313                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
9314                 return err;
9315         }
9316
9317         for (i = 0; i < s->map_cnt; i++) {
9318                 struct bpf_map *map = *s->maps[i].map;
9319                 size_t mmap_sz = bpf_map_mmap_sz(map);
9320                 int prot, map_fd = bpf_map__fd(map);
9321                 void **mmaped = s->maps[i].mmaped;
9322
9323                 if (!mmaped)
9324                         continue;
9325
9326                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
9327                         *mmaped = NULL;
9328                         continue;
9329                 }
9330
9331                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
9332                         prot = PROT_READ;
9333                 else
9334                         prot = PROT_READ | PROT_WRITE;
9335
9336                 /* Remap anonymous mmap()-ed "map initialization image" as
9337                  * a BPF map-backed mmap()-ed memory, but preserving the same
9338                  * memory address. This will cause kernel to change process'
9339                  * page table to point to a different piece of kernel memory,
9340                  * but from userspace point of view memory address (and its
9341                  * contents, being identical at this point) will stay the
9342                  * same. This mapping will be released by bpf_object__close()
9343                  * as per normal clean up procedure, so we don't need to worry
9344                  * about it from skeleton's clean up perspective.
9345                  */
9346                 *mmaped = mmap(map->mmaped, mmap_sz, prot,
9347                                 MAP_SHARED | MAP_FIXED, map_fd, 0);
9348                 if (*mmaped == MAP_FAILED) {
9349                         err = -errno;
9350                         *mmaped = NULL;
9351                         pr_warn("failed to re-mmap() map '%s': %d\n",
9352                                  bpf_map__name(map), err);
9353                         return err;
9354                 }
9355         }
9356
9357         return 0;
9358 }
9359
9360 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
9361 {
9362         int i;
9363
9364         for (i = 0; i < s->prog_cnt; i++) {
9365                 struct bpf_program *prog = *s->progs[i].prog;
9366                 struct bpf_link **link = s->progs[i].link;
9367                 const struct bpf_sec_def *sec_def;
9368                 const char *sec_name = bpf_program__title(prog, false);
9369
9370                 if (!prog->load)
9371                         continue;
9372
9373                 sec_def = find_sec_def(sec_name);
9374                 if (!sec_def || !sec_def->attach_fn)
9375                         continue;
9376
9377                 *link = sec_def->attach_fn(sec_def, prog);
9378                 if (IS_ERR(*link)) {
9379                         pr_warn("failed to auto-attach program '%s': %ld\n",
9380                                 bpf_program__name(prog), PTR_ERR(*link));
9381                         return PTR_ERR(*link);
9382                 }
9383         }
9384
9385         return 0;
9386 }
9387
9388 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
9389 {
9390         int i;
9391
9392         for (i = 0; i < s->prog_cnt; i++) {
9393                 struct bpf_link **link = s->progs[i].link;
9394
9395                 bpf_link__destroy(*link);
9396                 *link = NULL;
9397         }
9398 }
9399
9400 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
9401 {
9402         if (s->progs)
9403                 bpf_object__detach_skeleton(s);
9404         if (s->obj)
9405                 bpf_object__close(*s->obj);
9406         free(s->maps);
9407         free(s->progs);
9408         free(s);
9409 }