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.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                 err = 0;
2438         }
2439         if (btf_ext_data) {
2440                 if (!obj->btf) {
2441                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2442                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2443                         goto out;
2444                 }
2445                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
2446                                             btf_ext_data->d_size);
2447                 if (IS_ERR(obj->btf_ext)) {
2448                         pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
2449                                 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
2450                         obj->btf_ext = NULL;
2451                         goto out;
2452                 }
2453         }
2454 out:
2455         if (err && libbpf_needs_btf(obj)) {
2456                 pr_warn("BTF is required, but is missing or corrupted.\n");
2457                 return err;
2458         }
2459         return 0;
2460 }
2461
2462 static int bpf_object__finalize_btf(struct bpf_object *obj)
2463 {
2464         int err;
2465
2466         if (!obj->btf)
2467                 return 0;
2468
2469         err = btf__finalize_data(obj, obj->btf);
2470         if (err) {
2471                 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2472                 return err;
2473         }
2474
2475         return 0;
2476 }
2477
2478 static inline bool libbpf_prog_needs_vmlinux_btf(struct bpf_program *prog)
2479 {
2480         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2481             prog->type == BPF_PROG_TYPE_LSM)
2482                 return true;
2483
2484         /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2485          * also need vmlinux BTF
2486          */
2487         if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2488                 return true;
2489
2490         return false;
2491 }
2492
2493 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj)
2494 {
2495         bool need_vmlinux_btf = false;
2496         struct bpf_program *prog;
2497         int err;
2498
2499         /* CO-RE relocations need kernel BTF */
2500         if (obj->btf_ext && obj->btf_ext->field_reloc_info.len)
2501                 need_vmlinux_btf = true;
2502
2503         bpf_object__for_each_program(prog, obj) {
2504                 if (!prog->load)
2505                         continue;
2506                 if (libbpf_prog_needs_vmlinux_btf(prog)) {
2507                         need_vmlinux_btf = true;
2508                         break;
2509                 }
2510         }
2511
2512         if (!need_vmlinux_btf)
2513                 return 0;
2514
2515         obj->btf_vmlinux = libbpf_find_kernel_btf();
2516         if (IS_ERR(obj->btf_vmlinux)) {
2517                 err = PTR_ERR(obj->btf_vmlinux);
2518                 pr_warn("Error loading vmlinux BTF: %d\n", err);
2519                 obj->btf_vmlinux = NULL;
2520                 return err;
2521         }
2522         return 0;
2523 }
2524
2525 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2526 {
2527         struct btf *kern_btf = obj->btf;
2528         bool btf_mandatory, sanitize;
2529         int err = 0;
2530
2531         if (!obj->btf)
2532                 return 0;
2533
2534         sanitize = btf_needs_sanitization(obj);
2535         if (sanitize) {
2536                 const void *raw_data;
2537                 __u32 sz;
2538
2539                 /* clone BTF to sanitize a copy and leave the original intact */
2540                 raw_data = btf__get_raw_data(obj->btf, &sz);
2541                 kern_btf = btf__new(raw_data, sz);
2542                 if (IS_ERR(kern_btf))
2543                         return PTR_ERR(kern_btf);
2544
2545                 bpf_object__sanitize_btf(obj, kern_btf);
2546         }
2547
2548         err = btf__load(kern_btf);
2549         if (sanitize) {
2550                 if (!err) {
2551                         /* move fd to libbpf's BTF */
2552                         btf__set_fd(obj->btf, btf__fd(kern_btf));
2553                         btf__set_fd(kern_btf, -1);
2554                 }
2555                 btf__free(kern_btf);
2556         }
2557         if (err) {
2558                 btf_mandatory = kernel_needs_btf(obj);
2559                 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
2560                         btf_mandatory ? "BTF is mandatory, can't proceed."
2561                                       : "BTF is optional, ignoring.");
2562                 if (!btf_mandatory)
2563                         err = 0;
2564         }
2565         return err;
2566 }
2567
2568 static int bpf_object__elf_collect(struct bpf_object *obj)
2569 {
2570         Elf *elf = obj->efile.elf;
2571         GElf_Ehdr *ep = &obj->efile.ehdr;
2572         Elf_Data *btf_ext_data = NULL;
2573         Elf_Data *btf_data = NULL;
2574         Elf_Scn *scn = NULL;
2575         int idx = 0, err = 0;
2576
2577         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
2578         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
2579                 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
2580                 return -LIBBPF_ERRNO__FORMAT;
2581         }
2582
2583         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2584                 char *name;
2585                 GElf_Shdr sh;
2586                 Elf_Data *data;
2587
2588                 idx++;
2589                 if (gelf_getshdr(scn, &sh) != &sh) {
2590                         pr_warn("failed to get section(%d) header from %s\n",
2591                                 idx, obj->path);
2592                         return -LIBBPF_ERRNO__FORMAT;
2593                 }
2594
2595                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
2596                 if (!name) {
2597                         pr_warn("failed to get section(%d) name from %s\n",
2598                                 idx, obj->path);
2599                         return -LIBBPF_ERRNO__FORMAT;
2600                 }
2601
2602                 data = elf_getdata(scn, 0);
2603                 if (!data) {
2604                         pr_warn("failed to get section(%d) data from %s(%s)\n",
2605                                 idx, name, obj->path);
2606                         return -LIBBPF_ERRNO__FORMAT;
2607                 }
2608                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2609                          idx, name, (unsigned long)data->d_size,
2610                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
2611                          (int)sh.sh_type);
2612
2613                 if (strcmp(name, "license") == 0) {
2614                         err = bpf_object__init_license(obj,
2615                                                        data->d_buf,
2616                                                        data->d_size);
2617                         if (err)
2618                                 return err;
2619                 } else if (strcmp(name, "version") == 0) {
2620                         err = bpf_object__init_kversion(obj,
2621                                                         data->d_buf,
2622                                                         data->d_size);
2623                         if (err)
2624                                 return err;
2625                 } else if (strcmp(name, "maps") == 0) {
2626                         obj->efile.maps_shndx = idx;
2627                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2628                         obj->efile.btf_maps_shndx = idx;
2629                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2630                         btf_data = data;
2631                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
2632                         btf_ext_data = data;
2633                 } else if (sh.sh_type == SHT_SYMTAB) {
2634                         if (obj->efile.symbols) {
2635                                 pr_warn("bpf: multiple SYMTAB in %s\n",
2636                                         obj->path);
2637                                 return -LIBBPF_ERRNO__FORMAT;
2638                         }
2639                         obj->efile.symbols = data;
2640                         obj->efile.symbols_shndx = idx;
2641                         obj->efile.strtabidx = sh.sh_link;
2642                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2643                         if (sh.sh_flags & SHF_EXECINSTR) {
2644                                 if (strcmp(name, ".text") == 0)
2645                                         obj->efile.text_shndx = idx;
2646                                 err = bpf_object__add_program(obj, data->d_buf,
2647                                                               data->d_size,
2648                                                               name, idx);
2649                                 if (err) {
2650                                         char errmsg[STRERR_BUFSIZE];
2651                                         char *cp;
2652
2653                                         cp = libbpf_strerror_r(-err, errmsg,
2654                                                                sizeof(errmsg));
2655                                         pr_warn("failed to alloc program %s (%s): %s",
2656                                                 name, obj->path, cp);
2657                                         return err;
2658                                 }
2659                         } else if (strcmp(name, DATA_SEC) == 0) {
2660                                 obj->efile.data = data;
2661                                 obj->efile.data_shndx = idx;
2662                         } else if (strcmp(name, RODATA_SEC) == 0) {
2663                                 obj->efile.rodata = data;
2664                                 obj->efile.rodata_shndx = idx;
2665                         } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
2666                                 obj->efile.st_ops_data = data;
2667                                 obj->efile.st_ops_shndx = idx;
2668                         } else {
2669                                 pr_debug("skip section(%d) %s\n", idx, name);
2670                         }
2671                 } else if (sh.sh_type == SHT_REL) {
2672                         int nr_sects = obj->efile.nr_reloc_sects;
2673                         void *sects = obj->efile.reloc_sects;
2674                         int sec = sh.sh_info; /* points to other section */
2675
2676                         /* Only do relo for section with exec instructions */
2677                         if (!section_have_execinstr(obj, sec) &&
2678                             strcmp(name, ".rel" STRUCT_OPS_SEC) &&
2679                             strcmp(name, ".rel" MAPS_ELF_SEC)) {
2680                                 pr_debug("skip relo %s(%d) for section(%d)\n",
2681                                          name, idx, sec);
2682                                 continue;
2683                         }
2684
2685                         sects = reallocarray(sects, nr_sects + 1,
2686                                              sizeof(*obj->efile.reloc_sects));
2687                         if (!sects) {
2688                                 pr_warn("reloc_sects realloc failed\n");
2689                                 return -ENOMEM;
2690                         }
2691
2692                         obj->efile.reloc_sects = sects;
2693                         obj->efile.nr_reloc_sects++;
2694
2695                         obj->efile.reloc_sects[nr_sects].shdr = sh;
2696                         obj->efile.reloc_sects[nr_sects].data = data;
2697                 } else if (sh.sh_type == SHT_NOBITS &&
2698                            strcmp(name, BSS_SEC) == 0) {
2699                         obj->efile.bss = data;
2700                         obj->efile.bss_shndx = idx;
2701                 } else {
2702                         pr_debug("skip section(%d) %s\n", idx, name);
2703                 }
2704         }
2705
2706         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
2707                 pr_warn("Corrupted ELF file: index of strtab invalid\n");
2708                 return -LIBBPF_ERRNO__FORMAT;
2709         }
2710         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
2711 }
2712
2713 static bool sym_is_extern(const GElf_Sym *sym)
2714 {
2715         int bind = GELF_ST_BIND(sym->st_info);
2716         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2717         return sym->st_shndx == SHN_UNDEF &&
2718                (bind == STB_GLOBAL || bind == STB_WEAK) &&
2719                GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2720 }
2721
2722 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2723 {
2724         const struct btf_type *t;
2725         const char *var_name;
2726         int i, n;
2727
2728         if (!btf)
2729                 return -ESRCH;
2730
2731         n = btf__get_nr_types(btf);
2732         for (i = 1; i <= n; i++) {
2733                 t = btf__type_by_id(btf, i);
2734
2735                 if (!btf_is_var(t))
2736                         continue;
2737
2738                 var_name = btf__name_by_offset(btf, t->name_off);
2739                 if (strcmp(var_name, ext_name))
2740                         continue;
2741
2742                 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2743                         return -EINVAL;
2744
2745                 return i;
2746         }
2747
2748         return -ENOENT;
2749 }
2750
2751 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
2752         const struct btf_var_secinfo *vs;
2753         const struct btf_type *t;
2754         int i, j, n;
2755
2756         if (!btf)
2757                 return -ESRCH;
2758
2759         n = btf__get_nr_types(btf);
2760         for (i = 1; i <= n; i++) {
2761                 t = btf__type_by_id(btf, i);
2762
2763                 if (!btf_is_datasec(t))
2764                         continue;
2765
2766                 vs = btf_var_secinfos(t);
2767                 for (j = 0; j < btf_vlen(t); j++, vs++) {
2768                         if (vs->type == ext_btf_id)
2769                                 return i;
2770                 }
2771         }
2772
2773         return -ENOENT;
2774 }
2775
2776 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
2777                                      bool *is_signed)
2778 {
2779         const struct btf_type *t;
2780         const char *name;
2781
2782         t = skip_mods_and_typedefs(btf, id, NULL);
2783         name = btf__name_by_offset(btf, t->name_off);
2784
2785         if (is_signed)
2786                 *is_signed = false;
2787         switch (btf_kind(t)) {
2788         case BTF_KIND_INT: {
2789                 int enc = btf_int_encoding(t);
2790
2791                 if (enc & BTF_INT_BOOL)
2792                         return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
2793                 if (is_signed)
2794                         *is_signed = enc & BTF_INT_SIGNED;
2795                 if (t->size == 1)
2796                         return KCFG_CHAR;
2797                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
2798                         return KCFG_UNKNOWN;
2799                 return KCFG_INT;
2800         }
2801         case BTF_KIND_ENUM:
2802                 if (t->size != 4)
2803                         return KCFG_UNKNOWN;
2804                 if (strcmp(name, "libbpf_tristate"))
2805                         return KCFG_UNKNOWN;
2806                 return KCFG_TRISTATE;
2807         case BTF_KIND_ARRAY:
2808                 if (btf_array(t)->nelems == 0)
2809                         return KCFG_UNKNOWN;
2810                 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
2811                         return KCFG_UNKNOWN;
2812                 return KCFG_CHAR_ARR;
2813         default:
2814                 return KCFG_UNKNOWN;
2815         }
2816 }
2817
2818 static int cmp_externs(const void *_a, const void *_b)
2819 {
2820         const struct extern_desc *a = _a;
2821         const struct extern_desc *b = _b;
2822
2823         if (a->type != b->type)
2824                 return a->type < b->type ? -1 : 1;
2825
2826         if (a->type == EXT_KCFG) {
2827                 /* descending order by alignment requirements */
2828                 if (a->kcfg.align != b->kcfg.align)
2829                         return a->kcfg.align > b->kcfg.align ? -1 : 1;
2830                 /* ascending order by size, within same alignment class */
2831                 if (a->kcfg.sz != b->kcfg.sz)
2832                         return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
2833         }
2834
2835         /* resolve ties by name */
2836         return strcmp(a->name, b->name);
2837 }
2838
2839 static int find_int_btf_id(const struct btf *btf)
2840 {
2841         const struct btf_type *t;
2842         int i, n;
2843
2844         n = btf__get_nr_types(btf);
2845         for (i = 1; i <= n; i++) {
2846                 t = btf__type_by_id(btf, i);
2847
2848                 if (btf_is_int(t) && btf_int_bits(t) == 32)
2849                         return i;
2850         }
2851
2852         return 0;
2853 }
2854
2855 static int bpf_object__collect_externs(struct bpf_object *obj)
2856 {
2857         struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
2858         const struct btf_type *t;
2859         struct extern_desc *ext;
2860         int i, n, off;
2861         const char *ext_name, *sec_name;
2862         Elf_Scn *scn;
2863         GElf_Shdr sh;
2864
2865         if (!obj->efile.symbols)
2866                 return 0;
2867
2868         scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
2869         if (!scn)
2870                 return -LIBBPF_ERRNO__FORMAT;
2871         if (gelf_getshdr(scn, &sh) != &sh)
2872                 return -LIBBPF_ERRNO__FORMAT;
2873         n = sh.sh_size / sh.sh_entsize;
2874
2875         pr_debug("looking for externs among %d symbols...\n", n);
2876         for (i = 0; i < n; i++) {
2877                 GElf_Sym sym;
2878
2879                 if (!gelf_getsym(obj->efile.symbols, i, &sym))
2880                         return -LIBBPF_ERRNO__FORMAT;
2881                 if (!sym_is_extern(&sym))
2882                         continue;
2883                 ext_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2884                                       sym.st_name);
2885                 if (!ext_name || !ext_name[0])
2886                         continue;
2887
2888                 ext = obj->externs;
2889                 ext = reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
2890                 if (!ext)
2891                         return -ENOMEM;
2892                 obj->externs = ext;
2893                 ext = &ext[obj->nr_extern];
2894                 memset(ext, 0, sizeof(*ext));
2895                 obj->nr_extern++;
2896
2897                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
2898                 if (ext->btf_id <= 0) {
2899                         pr_warn("failed to find BTF for extern '%s': %d\n",
2900                                 ext_name, ext->btf_id);
2901                         return ext->btf_id;
2902                 }
2903                 t = btf__type_by_id(obj->btf, ext->btf_id);
2904                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
2905                 ext->sym_idx = i;
2906                 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
2907
2908                 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
2909                 if (ext->sec_btf_id <= 0) {
2910                         pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
2911                                 ext_name, ext->btf_id, ext->sec_btf_id);
2912                         return ext->sec_btf_id;
2913                 }
2914                 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
2915                 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
2916
2917                 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
2918                         kcfg_sec = sec;
2919                         ext->type = EXT_KCFG;
2920                         ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
2921                         if (ext->kcfg.sz <= 0) {
2922                                 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
2923                                         ext_name, ext->kcfg.sz);
2924                                 return ext->kcfg.sz;
2925                         }
2926                         ext->kcfg.align = btf__align_of(obj->btf, t->type);
2927                         if (ext->kcfg.align <= 0) {
2928                                 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
2929                                         ext_name, ext->kcfg.align);
2930                                 return -EINVAL;
2931                         }
2932                         ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
2933                                                         &ext->kcfg.is_signed);
2934                         if (ext->kcfg.type == KCFG_UNKNOWN) {
2935                                 pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
2936                                 return -ENOTSUP;
2937                         }
2938                 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
2939                         const struct btf_type *vt;
2940
2941                         ksym_sec = sec;
2942                         ext->type = EXT_KSYM;
2943
2944                         vt = skip_mods_and_typedefs(obj->btf, t->type, NULL);
2945                         if (!btf_is_void(vt)) {
2946                                 pr_warn("extern (ksym) '%s' is not typeless (void)\n", ext_name);
2947                                 return -ENOTSUP;
2948                         }
2949                 } else {
2950                         pr_warn("unrecognized extern section '%s'\n", sec_name);
2951                         return -ENOTSUP;
2952                 }
2953         }
2954         pr_debug("collected %d externs total\n", obj->nr_extern);
2955
2956         if (!obj->nr_extern)
2957                 return 0;
2958
2959         /* sort externs by type, for kcfg ones also by (align, size, name) */
2960         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
2961
2962         /* for .ksyms section, we need to turn all externs into allocated
2963          * variables in BTF to pass kernel verification; we do this by
2964          * pretending that each extern is a 8-byte variable
2965          */
2966         if (ksym_sec) {
2967                 /* find existing 4-byte integer type in BTF to use for fake
2968                  * extern variables in DATASEC
2969                  */
2970                 int int_btf_id = find_int_btf_id(obj->btf);
2971
2972                 for (i = 0; i < obj->nr_extern; i++) {
2973                         ext = &obj->externs[i];
2974                         if (ext->type != EXT_KSYM)
2975                                 continue;
2976                         pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
2977                                  i, ext->sym_idx, ext->name);
2978                 }
2979
2980                 sec = ksym_sec;
2981                 n = btf_vlen(sec);
2982                 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
2983                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
2984                         struct btf_type *vt;
2985
2986                         vt = (void *)btf__type_by_id(obj->btf, vs->type);
2987                         ext_name = btf__name_by_offset(obj->btf, vt->name_off);
2988                         ext = find_extern_by_name(obj, ext_name);
2989                         if (!ext) {
2990                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
2991                                         ext_name);
2992                                 return -ESRCH;
2993                         }
2994                         btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
2995                         vt->type = int_btf_id;
2996                         vs->offset = off;
2997                         vs->size = sizeof(int);
2998                 }
2999                 sec->size = off;
3000         }
3001
3002         if (kcfg_sec) {
3003                 sec = kcfg_sec;
3004                 /* for kcfg externs calculate their offsets within a .kconfig map */
3005                 off = 0;
3006                 for (i = 0; i < obj->nr_extern; i++) {
3007                         ext = &obj->externs[i];
3008                         if (ext->type != EXT_KCFG)
3009                                 continue;
3010
3011                         ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3012                         off = ext->kcfg.data_off + ext->kcfg.sz;
3013                         pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3014                                  i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3015                 }
3016                 sec->size = off;
3017                 n = btf_vlen(sec);
3018                 for (i = 0; i < n; i++) {
3019                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3020
3021                         t = btf__type_by_id(obj->btf, vs->type);
3022                         ext_name = btf__name_by_offset(obj->btf, t->name_off);
3023                         ext = find_extern_by_name(obj, ext_name);
3024                         if (!ext) {
3025                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
3026                                         ext_name);
3027                                 return -ESRCH;
3028                         }
3029                         btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3030                         vs->offset = ext->kcfg.data_off;
3031                 }
3032         }
3033         return 0;
3034 }
3035
3036 static struct bpf_program *
3037 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
3038 {
3039         struct bpf_program *prog;
3040         size_t i;
3041
3042         for (i = 0; i < obj->nr_programs; i++) {
3043                 prog = &obj->programs[i];
3044                 if (prog->idx == idx)
3045                         return prog;
3046         }
3047         return NULL;
3048 }
3049
3050 struct bpf_program *
3051 bpf_object__find_program_by_title(const struct bpf_object *obj,
3052                                   const char *title)
3053 {
3054         struct bpf_program *pos;
3055
3056         bpf_object__for_each_program(pos, obj) {
3057                 if (pos->section_name && !strcmp(pos->section_name, title))
3058                         return pos;
3059         }
3060         return NULL;
3061 }
3062
3063 struct bpf_program *
3064 bpf_object__find_program_by_name(const struct bpf_object *obj,
3065                                  const char *name)
3066 {
3067         struct bpf_program *prog;
3068
3069         bpf_object__for_each_program(prog, obj) {
3070                 if (!strcmp(prog->name, name))
3071                         return prog;
3072         }
3073         return NULL;
3074 }
3075
3076 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3077                                       int shndx)
3078 {
3079         return shndx == obj->efile.data_shndx ||
3080                shndx == obj->efile.bss_shndx ||
3081                shndx == obj->efile.rodata_shndx;
3082 }
3083
3084 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3085                                       int shndx)
3086 {
3087         return shndx == obj->efile.maps_shndx ||
3088                shndx == obj->efile.btf_maps_shndx;
3089 }
3090
3091 static enum libbpf_map_type
3092 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3093 {
3094         if (shndx == obj->efile.data_shndx)
3095                 return LIBBPF_MAP_DATA;
3096         else if (shndx == obj->efile.bss_shndx)
3097                 return LIBBPF_MAP_BSS;
3098         else if (shndx == obj->efile.rodata_shndx)
3099                 return LIBBPF_MAP_RODATA;
3100         else if (shndx == obj->efile.symbols_shndx)
3101                 return LIBBPF_MAP_KCONFIG;
3102         else
3103                 return LIBBPF_MAP_UNSPEC;
3104 }
3105
3106 static int bpf_program__record_reloc(struct bpf_program *prog,
3107                                      struct reloc_desc *reloc_desc,
3108                                      __u32 insn_idx, const char *name,
3109                                      const GElf_Sym *sym, const GElf_Rel *rel)
3110 {
3111         struct bpf_insn *insn = &prog->insns[insn_idx];
3112         size_t map_idx, nr_maps = prog->obj->nr_maps;
3113         struct bpf_object *obj = prog->obj;
3114         __u32 shdr_idx = sym->st_shndx;
3115         enum libbpf_map_type type;
3116         struct bpf_map *map;
3117
3118         /* sub-program call relocation */
3119         if (insn->code == (BPF_JMP | BPF_CALL)) {
3120                 if (insn->src_reg != BPF_PSEUDO_CALL) {
3121                         pr_warn("incorrect bpf_call opcode\n");
3122                         return -LIBBPF_ERRNO__RELOC;
3123                 }
3124                 /* text_shndx can be 0, if no default "main" program exists */
3125                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3126                         pr_warn("bad call relo against section %u\n", shdr_idx);
3127                         return -LIBBPF_ERRNO__RELOC;
3128                 }
3129                 if (sym->st_value % 8) {
3130                         pr_warn("bad call relo offset: %zu\n",
3131                                 (size_t)sym->st_value);
3132                         return -LIBBPF_ERRNO__RELOC;
3133                 }
3134                 reloc_desc->type = RELO_CALL;
3135                 reloc_desc->insn_idx = insn_idx;
3136                 reloc_desc->sym_off = sym->st_value;
3137                 obj->has_pseudo_calls = true;
3138                 return 0;
3139         }
3140
3141         if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
3142                 pr_warn("invalid relo for insns[%d].code 0x%x\n",
3143                         insn_idx, insn->code);
3144                 return -LIBBPF_ERRNO__RELOC;
3145         }
3146
3147         if (sym_is_extern(sym)) {
3148                 int sym_idx = GELF_R_SYM(rel->r_info);
3149                 int i, n = obj->nr_extern;
3150                 struct extern_desc *ext;
3151
3152                 for (i = 0; i < n; i++) {
3153                         ext = &obj->externs[i];
3154                         if (ext->sym_idx == sym_idx)
3155                                 break;
3156                 }
3157                 if (i >= n) {
3158                         pr_warn("extern relo failed to find extern for sym %d\n",
3159                                 sym_idx);
3160                         return -LIBBPF_ERRNO__RELOC;
3161                 }
3162                 pr_debug("found extern #%d '%s' (sym %d) for insn %u\n",
3163                          i, ext->name, ext->sym_idx, insn_idx);
3164                 reloc_desc->type = RELO_EXTERN;
3165                 reloc_desc->insn_idx = insn_idx;
3166                 reloc_desc->sym_off = i; /* sym_off stores extern index */
3167                 return 0;
3168         }
3169
3170         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3171                 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
3172                         name, shdr_idx);
3173                 return -LIBBPF_ERRNO__RELOC;
3174         }
3175
3176         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
3177
3178         /* generic map reference relocation */
3179         if (type == LIBBPF_MAP_UNSPEC) {
3180                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
3181                         pr_warn("bad map relo against section %u\n",
3182                                 shdr_idx);
3183                         return -LIBBPF_ERRNO__RELOC;
3184                 }
3185                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3186                         map = &obj->maps[map_idx];
3187                         if (map->libbpf_type != type ||
3188                             map->sec_idx != sym->st_shndx ||
3189                             map->sec_offset != sym->st_value)
3190                                 continue;
3191                         pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
3192                                  map_idx, map->name, map->sec_idx,
3193                                  map->sec_offset, insn_idx);
3194                         break;
3195                 }
3196                 if (map_idx >= nr_maps) {
3197                         pr_warn("map relo failed to find map for sec %u, off %zu\n",
3198                                 shdr_idx, (size_t)sym->st_value);
3199                         return -LIBBPF_ERRNO__RELOC;
3200                 }
3201                 reloc_desc->type = RELO_LD64;
3202                 reloc_desc->insn_idx = insn_idx;
3203                 reloc_desc->map_idx = map_idx;
3204                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
3205                 return 0;
3206         }
3207
3208         /* global data map relocation */
3209         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
3210                 pr_warn("bad data relo against section %u\n", shdr_idx);
3211                 return -LIBBPF_ERRNO__RELOC;
3212         }
3213         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3214                 map = &obj->maps[map_idx];
3215                 if (map->libbpf_type != type)
3216                         continue;
3217                 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
3218                          map_idx, map->name, map->sec_idx, map->sec_offset,
3219                          insn_idx);
3220                 break;
3221         }
3222         if (map_idx >= nr_maps) {
3223                 pr_warn("data relo failed to find map for sec %u\n",
3224                         shdr_idx);
3225                 return -LIBBPF_ERRNO__RELOC;
3226         }
3227
3228         reloc_desc->type = RELO_DATA;
3229         reloc_desc->insn_idx = insn_idx;
3230         reloc_desc->map_idx = map_idx;
3231         reloc_desc->sym_off = sym->st_value;
3232         return 0;
3233 }
3234
3235 static int
3236 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
3237                            Elf_Data *data, struct bpf_object *obj)
3238 {
3239         Elf_Data *symbols = obj->efile.symbols;
3240         int err, i, nrels;
3241
3242         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
3243         nrels = shdr->sh_size / shdr->sh_entsize;
3244
3245         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
3246         if (!prog->reloc_desc) {
3247                 pr_warn("failed to alloc memory in relocation\n");
3248                 return -ENOMEM;
3249         }
3250         prog->nr_reloc = nrels;
3251
3252         for (i = 0; i < nrels; i++) {
3253                 const char *name;
3254                 __u32 insn_idx;
3255                 GElf_Sym sym;
3256                 GElf_Rel rel;
3257
3258                 if (!gelf_getrel(data, i, &rel)) {
3259                         pr_warn("relocation: failed to get %d reloc\n", i);
3260                         return -LIBBPF_ERRNO__FORMAT;
3261                 }
3262                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
3263                         pr_warn("relocation: symbol %"PRIx64" not found\n",
3264                                 GELF_R_SYM(rel.r_info));
3265                         return -LIBBPF_ERRNO__FORMAT;
3266                 }
3267                 if (rel.r_offset % sizeof(struct bpf_insn))
3268                         return -LIBBPF_ERRNO__FORMAT;
3269
3270                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
3271                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
3272                                   sym.st_name) ? : "<?>";
3273
3274                 pr_debug("relo for shdr %u, symb %zu, value %zu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
3275                          (__u32)sym.st_shndx, (size_t)GELF_R_SYM(rel.r_info),
3276                          (size_t)sym.st_value, GELF_ST_TYPE(sym.st_info),
3277                          GELF_ST_BIND(sym.st_info), sym.st_name, name,
3278                          insn_idx);
3279
3280                 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
3281                                                 insn_idx, name, &sym, &rel);
3282                 if (err)
3283                         return err;
3284         }
3285         return 0;
3286 }
3287
3288 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
3289 {
3290         struct bpf_map_def *def = &map->def;
3291         __u32 key_type_id = 0, value_type_id = 0;
3292         int ret;
3293
3294         /* if it's BTF-defined map, we don't need to search for type IDs.
3295          * For struct_ops map, it does not need btf_key_type_id and
3296          * btf_value_type_id.
3297          */
3298         if (map->sec_idx == obj->efile.btf_maps_shndx ||
3299             bpf_map__is_struct_ops(map))
3300                 return 0;
3301
3302         if (!bpf_map__is_internal(map)) {
3303                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
3304                                            def->value_size, &key_type_id,
3305                                            &value_type_id);
3306         } else {
3307                 /*
3308                  * LLVM annotates global data differently in BTF, that is,
3309                  * only as '.data', '.bss' or '.rodata'.
3310                  */
3311                 ret = btf__find_by_name(obj->btf,
3312                                 libbpf_type_to_btf_name[map->libbpf_type]);
3313         }
3314         if (ret < 0)
3315                 return ret;
3316
3317         map->btf_key_type_id = key_type_id;
3318         map->btf_value_type_id = bpf_map__is_internal(map) ?
3319                                  ret : value_type_id;
3320         return 0;
3321 }
3322
3323 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3324 {
3325         struct bpf_map_info info = {};
3326         __u32 len = sizeof(info);
3327         int new_fd, err;
3328         char *new_name;
3329
3330         err = bpf_obj_get_info_by_fd(fd, &info, &len);
3331         if (err)
3332                 return err;
3333
3334         new_name = strdup(info.name);
3335         if (!new_name)
3336                 return -errno;
3337
3338         new_fd = open("/", O_RDONLY | O_CLOEXEC);
3339         if (new_fd < 0) {
3340                 err = -errno;
3341                 goto err_free_new_name;
3342         }
3343
3344         new_fd = dup3(fd, new_fd, O_CLOEXEC);
3345         if (new_fd < 0) {
3346                 err = -errno;
3347                 goto err_close_new_fd;
3348         }
3349
3350         err = zclose(map->fd);
3351         if (err) {
3352                 err = -errno;
3353                 goto err_close_new_fd;
3354         }
3355         free(map->name);
3356
3357         map->fd = new_fd;
3358         map->name = new_name;
3359         map->def.type = info.type;
3360         map->def.key_size = info.key_size;
3361         map->def.value_size = info.value_size;
3362         map->def.max_entries = info.max_entries;
3363         map->def.map_flags = info.map_flags;
3364         map->btf_key_type_id = info.btf_key_type_id;
3365         map->btf_value_type_id = info.btf_value_type_id;
3366         map->reused = true;
3367
3368         return 0;
3369
3370 err_close_new_fd:
3371         close(new_fd);
3372 err_free_new_name:
3373         free(new_name);
3374         return err;
3375 }
3376
3377 __u32 bpf_map__max_entries(const struct bpf_map *map)
3378 {
3379         return map->def.max_entries;
3380 }
3381
3382 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
3383 {
3384         if (map->fd >= 0)
3385                 return -EBUSY;
3386         map->def.max_entries = max_entries;
3387         return 0;
3388 }
3389
3390 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
3391 {
3392         if (!map || !max_entries)
3393                 return -EINVAL;
3394
3395         return bpf_map__set_max_entries(map, max_entries);
3396 }
3397
3398 static int
3399 bpf_object__probe_loading(struct bpf_object *obj)
3400 {
3401         struct bpf_load_program_attr attr;
3402         char *cp, errmsg[STRERR_BUFSIZE];
3403         struct bpf_insn insns[] = {
3404                 BPF_MOV64_IMM(BPF_REG_0, 0),
3405                 BPF_EXIT_INSN(),
3406         };
3407         int ret;
3408
3409         /* make sure basic loading works */
3410
3411         memset(&attr, 0, sizeof(attr));
3412         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3413         attr.insns = insns;
3414         attr.insns_cnt = ARRAY_SIZE(insns);
3415         attr.license = "GPL";
3416
3417         ret = bpf_load_program_xattr(&attr, NULL, 0);
3418         if (ret < 0) {
3419                 ret = errno;
3420                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3421                 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
3422                         "program. Make sure your kernel supports BPF "
3423                         "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
3424                         "set to big enough value.\n", __func__, cp, ret);
3425                 return -ret;
3426         }
3427         close(ret);
3428
3429         return 0;
3430 }
3431
3432 static int
3433 bpf_object__probe_name(struct bpf_object *obj)
3434 {
3435         struct bpf_load_program_attr attr;
3436         struct bpf_insn insns[] = {
3437                 BPF_MOV64_IMM(BPF_REG_0, 0),
3438                 BPF_EXIT_INSN(),
3439         };
3440         int ret;
3441
3442         /* make sure loading with name works */
3443
3444         memset(&attr, 0, sizeof(attr));
3445         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3446         attr.insns = insns;
3447         attr.insns_cnt = ARRAY_SIZE(insns);
3448         attr.license = "GPL";
3449         attr.name = "test";
3450         ret = bpf_load_program_xattr(&attr, NULL, 0);
3451         if (ret >= 0) {
3452                 obj->caps.name = 1;
3453                 close(ret);
3454         }
3455
3456         return 0;
3457 }
3458
3459 static int
3460 bpf_object__probe_global_data(struct bpf_object *obj)
3461 {
3462         struct bpf_load_program_attr prg_attr;
3463         struct bpf_create_map_attr map_attr;
3464         char *cp, errmsg[STRERR_BUFSIZE];
3465         struct bpf_insn insns[] = {
3466                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
3467                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
3468                 BPF_MOV64_IMM(BPF_REG_0, 0),
3469                 BPF_EXIT_INSN(),
3470         };
3471         int ret, map;
3472
3473         memset(&map_attr, 0, sizeof(map_attr));
3474         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3475         map_attr.key_size = sizeof(int);
3476         map_attr.value_size = 32;
3477         map_attr.max_entries = 1;
3478
3479         map = bpf_create_map_xattr(&map_attr);
3480         if (map < 0) {
3481                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3482                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3483                         __func__, cp, errno);
3484                 return -errno;
3485         }
3486
3487         insns[0].imm = map;
3488
3489         memset(&prg_attr, 0, sizeof(prg_attr));
3490         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3491         prg_attr.insns = insns;
3492         prg_attr.insns_cnt = ARRAY_SIZE(insns);
3493         prg_attr.license = "GPL";
3494
3495         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
3496         if (ret >= 0) {
3497                 obj->caps.global_data = 1;
3498                 close(ret);
3499         }
3500
3501         close(map);
3502         return 0;
3503 }
3504
3505 static int bpf_object__probe_btf_func(struct bpf_object *obj)
3506 {
3507         static const char strs[] = "\0int\0x\0a";
3508         /* void x(int a) {} */
3509         __u32 types[] = {
3510                 /* int */
3511                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3512                 /* FUNC_PROTO */                                /* [2] */
3513                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3514                 BTF_PARAM_ENC(7, 1),
3515                 /* FUNC x */                                    /* [3] */
3516                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
3517         };
3518         int btf_fd;
3519
3520         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3521                                       strs, sizeof(strs));
3522         if (btf_fd >= 0) {
3523                 obj->caps.btf_func = 1;
3524                 close(btf_fd);
3525                 return 1;
3526         }
3527
3528         return 0;
3529 }
3530
3531 static int bpf_object__probe_btf_func_global(struct bpf_object *obj)
3532 {
3533         static const char strs[] = "\0int\0x\0a";
3534         /* static void x(int a) {} */
3535         __u32 types[] = {
3536                 /* int */
3537                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3538                 /* FUNC_PROTO */                                /* [2] */
3539                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3540                 BTF_PARAM_ENC(7, 1),
3541                 /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
3542                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
3543         };
3544         int btf_fd;
3545
3546         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3547                                       strs, sizeof(strs));
3548         if (btf_fd >= 0) {
3549                 obj->caps.btf_func_global = 1;
3550                 close(btf_fd);
3551                 return 1;
3552         }
3553
3554         return 0;
3555 }
3556
3557 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
3558 {
3559         static const char strs[] = "\0x\0.data";
3560         /* static int a; */
3561         __u32 types[] = {
3562                 /* int */
3563                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3564                 /* VAR x */                                     /* [2] */
3565                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
3566                 BTF_VAR_STATIC,
3567                 /* DATASEC val */                               /* [3] */
3568                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
3569                 BTF_VAR_SECINFO_ENC(2, 0, 4),
3570         };
3571         int btf_fd;
3572
3573         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
3574                                       strs, sizeof(strs));
3575         if (btf_fd >= 0) {
3576                 obj->caps.btf_datasec = 1;
3577                 close(btf_fd);
3578                 return 1;
3579         }
3580
3581         return 0;
3582 }
3583
3584 static int bpf_object__probe_array_mmap(struct bpf_object *obj)
3585 {
3586         struct bpf_create_map_attr attr = {
3587                 .map_type = BPF_MAP_TYPE_ARRAY,
3588                 .map_flags = BPF_F_MMAPABLE,
3589                 .key_size = sizeof(int),
3590                 .value_size = sizeof(int),
3591                 .max_entries = 1,
3592         };
3593         int fd;
3594
3595         fd = bpf_create_map_xattr(&attr);
3596         if (fd >= 0) {
3597                 obj->caps.array_mmap = 1;
3598                 close(fd);
3599                 return 1;
3600         }
3601
3602         return 0;
3603 }
3604
3605 static int
3606 bpf_object__probe_exp_attach_type(struct bpf_object *obj)
3607 {
3608         struct bpf_load_program_attr attr;
3609         struct bpf_insn insns[] = {
3610                 BPF_MOV64_IMM(BPF_REG_0, 0),
3611                 BPF_EXIT_INSN(),
3612         };
3613         int fd;
3614
3615         memset(&attr, 0, sizeof(attr));
3616         /* use any valid combination of program type and (optional)
3617          * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
3618          * to see if kernel supports expected_attach_type field for
3619          * BPF_PROG_LOAD command
3620          */
3621         attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
3622         attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE;
3623         attr.insns = insns;
3624         attr.insns_cnt = ARRAY_SIZE(insns);
3625         attr.license = "GPL";
3626
3627         fd = bpf_load_program_xattr(&attr, NULL, 0);
3628         if (fd >= 0) {
3629                 obj->caps.exp_attach_type = 1;
3630                 close(fd);
3631                 return 1;
3632         }
3633         return 0;
3634 }
3635
3636 static int
3637 bpf_object__probe_caps(struct bpf_object *obj)
3638 {
3639         int (*probe_fn[])(struct bpf_object *obj) = {
3640                 bpf_object__probe_name,
3641                 bpf_object__probe_global_data,
3642                 bpf_object__probe_btf_func,
3643                 bpf_object__probe_btf_func_global,
3644                 bpf_object__probe_btf_datasec,
3645                 bpf_object__probe_array_mmap,
3646                 bpf_object__probe_exp_attach_type,
3647         };
3648         int i, ret;
3649
3650         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
3651                 ret = probe_fn[i](obj);
3652                 if (ret < 0)
3653                         pr_debug("Probe #%d failed with %d.\n", i, ret);
3654         }
3655
3656         return 0;
3657 }
3658
3659 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
3660 {
3661         struct bpf_map_info map_info = {};
3662         char msg[STRERR_BUFSIZE];
3663         __u32 map_info_len;
3664
3665         map_info_len = sizeof(map_info);
3666
3667         if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
3668                 pr_warn("failed to get map info for map FD %d: %s\n",
3669                         map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
3670                 return false;
3671         }
3672
3673         return (map_info.type == map->def.type &&
3674                 map_info.key_size == map->def.key_size &&
3675                 map_info.value_size == map->def.value_size &&
3676                 map_info.max_entries == map->def.max_entries &&
3677                 map_info.map_flags == map->def.map_flags);
3678 }
3679
3680 static int
3681 bpf_object__reuse_map(struct bpf_map *map)
3682 {
3683         char *cp, errmsg[STRERR_BUFSIZE];
3684         int err, pin_fd;
3685
3686         pin_fd = bpf_obj_get(map->pin_path);
3687         if (pin_fd < 0) {
3688                 err = -errno;
3689                 if (err == -ENOENT) {
3690                         pr_debug("found no pinned map to reuse at '%s'\n",
3691                                  map->pin_path);
3692                         return 0;
3693                 }
3694
3695                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3696                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
3697                         map->pin_path, cp);
3698                 return err;
3699         }
3700
3701         if (!map_is_reuse_compat(map, pin_fd)) {
3702                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
3703                         map->pin_path);
3704                 close(pin_fd);
3705                 return -EINVAL;
3706         }
3707
3708         err = bpf_map__reuse_fd(map, pin_fd);
3709         if (err) {
3710                 close(pin_fd);
3711                 return err;
3712         }
3713         map->pinned = true;
3714         pr_debug("reused pinned map at '%s'\n", map->pin_path);
3715
3716         return 0;
3717 }
3718
3719 static int
3720 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
3721 {
3722         enum libbpf_map_type map_type = map->libbpf_type;
3723         char *cp, errmsg[STRERR_BUFSIZE];
3724         int err, zero = 0;
3725
3726         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
3727         if (err) {
3728                 err = -errno;
3729                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3730                 pr_warn("Error setting initial map(%s) contents: %s\n",
3731                         map->name, cp);
3732                 return err;
3733         }
3734
3735         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
3736         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
3737                 err = bpf_map_freeze(map->fd);
3738                 if (err) {
3739                         err = -errno;
3740                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3741                         pr_warn("Error freezing map(%s) as read-only: %s\n",
3742                                 map->name, cp);
3743                         return err;
3744                 }
3745         }
3746         return 0;
3747 }
3748
3749 static void bpf_map__destroy(struct bpf_map *map);
3750
3751 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map)
3752 {
3753         struct bpf_create_map_attr create_attr;
3754         struct bpf_map_def *def = &map->def;
3755
3756         memset(&create_attr, 0, sizeof(create_attr));
3757
3758         if (obj->caps.name)
3759                 create_attr.name = map->name;
3760         create_attr.map_ifindex = map->map_ifindex;
3761         create_attr.map_type = def->type;
3762         create_attr.map_flags = def->map_flags;
3763         create_attr.key_size = def->key_size;
3764         create_attr.value_size = def->value_size;
3765         create_attr.numa_node = map->numa_node;
3766
3767         if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) {
3768                 int nr_cpus;
3769
3770                 nr_cpus = libbpf_num_possible_cpus();
3771                 if (nr_cpus < 0) {
3772                         pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
3773                                 map->name, nr_cpus);
3774                         return nr_cpus;
3775                 }
3776                 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
3777                 create_attr.max_entries = nr_cpus;
3778         } else {
3779                 create_attr.max_entries = def->max_entries;
3780         }
3781
3782         if (bpf_map__is_struct_ops(map))
3783                 create_attr.btf_vmlinux_value_type_id =
3784                         map->btf_vmlinux_value_type_id;
3785
3786         create_attr.btf_fd = 0;
3787         create_attr.btf_key_type_id = 0;
3788         create_attr.btf_value_type_id = 0;
3789         if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) {
3790                 create_attr.btf_fd = btf__fd(obj->btf);
3791                 create_attr.btf_key_type_id = map->btf_key_type_id;
3792                 create_attr.btf_value_type_id = map->btf_value_type_id;
3793         }
3794
3795         if (bpf_map_type__is_map_in_map(def->type)) {
3796                 if (map->inner_map) {
3797                         int err;
3798
3799                         err = bpf_object__create_map(obj, map->inner_map);
3800                         if (err) {
3801                                 pr_warn("map '%s': failed to create inner map: %d\n",
3802                                         map->name, err);
3803                                 return err;
3804                         }
3805                         map->inner_map_fd = bpf_map__fd(map->inner_map);
3806                 }
3807                 if (map->inner_map_fd >= 0)
3808                         create_attr.inner_map_fd = map->inner_map_fd;
3809         }
3810
3811         map->fd = bpf_create_map_xattr(&create_attr);
3812         if (map->fd < 0 && (create_attr.btf_key_type_id ||
3813                             create_attr.btf_value_type_id)) {
3814                 char *cp, errmsg[STRERR_BUFSIZE];
3815                 int err = -errno;
3816
3817                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3818                 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
3819                         map->name, cp, err);
3820                 create_attr.btf_fd = 0;
3821                 create_attr.btf_key_type_id = 0;
3822                 create_attr.btf_value_type_id = 0;
3823                 map->btf_key_type_id = 0;
3824                 map->btf_value_type_id = 0;
3825                 map->fd = bpf_create_map_xattr(&create_attr);
3826         }
3827
3828         if (map->fd < 0)
3829                 return -errno;
3830
3831         if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
3832                 bpf_map__destroy(map->inner_map);
3833                 zfree(&map->inner_map);
3834         }
3835
3836         return 0;
3837 }
3838
3839 static int
3840 bpf_object__create_maps(struct bpf_object *obj)
3841 {
3842         struct bpf_map *map;
3843         char *cp, errmsg[STRERR_BUFSIZE];
3844         unsigned int i, j;
3845         int err;
3846
3847         for (i = 0; i < obj->nr_maps; i++) {
3848                 map = &obj->maps[i];
3849
3850                 if (map->pin_path) {
3851                         err = bpf_object__reuse_map(map);
3852                         if (err) {
3853                                 pr_warn("map '%s': error reusing pinned map\n",
3854                                         map->name);
3855                                 goto err_out;
3856                         }
3857                 }
3858
3859                 if (map->fd >= 0) {
3860                         pr_debug("map '%s': skipping creation (preset fd=%d)\n",
3861                                  map->name, map->fd);
3862                         continue;
3863                 }
3864
3865                 err = bpf_object__create_map(obj, map);
3866                 if (err)
3867                         goto err_out;
3868
3869                 pr_debug("map '%s': created successfully, fd=%d\n", map->name,
3870                          map->fd);
3871
3872                 if (bpf_map__is_internal(map)) {
3873                         err = bpf_object__populate_internal_map(obj, map);
3874                         if (err < 0) {
3875                                 zclose(map->fd);
3876                                 goto err_out;
3877                         }
3878                 }
3879
3880                 if (map->init_slots_sz) {
3881                         for (j = 0; j < map->init_slots_sz; j++) {
3882                                 const struct bpf_map *targ_map;
3883                                 int fd;
3884
3885                                 if (!map->init_slots[j])
3886                                         continue;
3887
3888                                 targ_map = map->init_slots[j];
3889                                 fd = bpf_map__fd(targ_map);
3890                                 err = bpf_map_update_elem(map->fd, &j, &fd, 0);
3891                                 if (err) {
3892                                         err = -errno;
3893                                         pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
3894                                                 map->name, j, targ_map->name,
3895                                                 fd, err);
3896                                         goto err_out;
3897                                 }
3898                                 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
3899                                          map->name, j, targ_map->name, fd);
3900                         }
3901                         zfree(&map->init_slots);
3902                         map->init_slots_sz = 0;
3903                 }
3904
3905                 if (map->pin_path && !map->pinned) {
3906                         err = bpf_map__pin(map, NULL);
3907                         if (err) {
3908                                 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
3909                                         map->name, map->pin_path, err);
3910                                 zclose(map->fd);
3911                                 goto err_out;
3912                         }
3913                 }
3914         }
3915
3916         return 0;
3917
3918 err_out:
3919         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3920         pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
3921         pr_perm_msg(err);
3922         for (j = 0; j < i; j++)
3923                 zclose(obj->maps[j].fd);
3924         return err;
3925 }
3926
3927 static int
3928 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
3929                         void *btf_prog_info, const char *info_name)
3930 {
3931         if (err != -ENOENT) {
3932                 pr_warn("Error in loading %s for sec %s.\n",
3933                         info_name, prog->section_name);
3934                 return err;
3935         }
3936
3937         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
3938
3939         if (btf_prog_info) {
3940                 /*
3941                  * Some info has already been found but has problem
3942                  * in the last btf_ext reloc. Must have to error out.
3943                  */
3944                 pr_warn("Error in relocating %s for sec %s.\n",
3945                         info_name, prog->section_name);
3946                 return err;
3947         }
3948
3949         /* Have problem loading the very first info. Ignore the rest. */
3950         pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
3951                 info_name, prog->section_name, info_name);
3952         return 0;
3953 }
3954
3955 static int
3956 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
3957                           const char *section_name,  __u32 insn_offset)
3958 {
3959         int err;
3960
3961         if (!insn_offset || prog->func_info) {
3962                 /*
3963                  * !insn_offset => main program
3964                  *
3965                  * For sub prog, the main program's func_info has to
3966                  * be loaded first (i.e. prog->func_info != NULL)
3967                  */
3968                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
3969                                                section_name, insn_offset,
3970                                                &prog->func_info,
3971                                                &prog->func_info_cnt);
3972                 if (err)
3973                         return check_btf_ext_reloc_err(prog, err,
3974                                                        prog->func_info,
3975                                                        "bpf_func_info");
3976
3977                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
3978         }
3979
3980         if (!insn_offset || prog->line_info) {
3981                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
3982                                                section_name, insn_offset,
3983                                                &prog->line_info,
3984                                                &prog->line_info_cnt);
3985                 if (err)
3986                         return check_btf_ext_reloc_err(prog, err,
3987                                                        prog->line_info,
3988                                                        "bpf_line_info");
3989
3990                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
3991         }
3992
3993         return 0;
3994 }
3995
3996 #define BPF_CORE_SPEC_MAX_LEN 64
3997
3998 /* represents BPF CO-RE field or array element accessor */
3999 struct bpf_core_accessor {
4000         __u32 type_id;          /* struct/union type or array element type */
4001         __u32 idx;              /* field index or array index */
4002         const char *name;       /* field name or NULL for array accessor */
4003 };
4004
4005 struct bpf_core_spec {
4006         const struct btf *btf;
4007         /* high-level spec: named fields and array indices only */
4008         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
4009         /* high-level spec length */
4010         int len;
4011         /* raw, low-level spec: 1-to-1 with accessor spec string */
4012         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
4013         /* raw spec length */
4014         int raw_len;
4015         /* field bit offset represented by spec */
4016         __u32 bit_offset;
4017 };
4018
4019 static bool str_is_empty(const char *s)
4020 {
4021         return !s || !s[0];
4022 }
4023
4024 static bool is_flex_arr(const struct btf *btf,
4025                         const struct bpf_core_accessor *acc,
4026                         const struct btf_array *arr)
4027 {
4028         const struct btf_type *t;
4029
4030         /* not a flexible array, if not inside a struct or has non-zero size */
4031         if (!acc->name || arr->nelems > 0)
4032                 return false;
4033
4034         /* has to be the last member of enclosing struct */
4035         t = btf__type_by_id(btf, acc->type_id);
4036         return acc->idx == btf_vlen(t) - 1;
4037 }
4038
4039 /*
4040  * Turn bpf_field_reloc into a low- and high-level spec representation,
4041  * validating correctness along the way, as well as calculating resulting
4042  * field bit offset, specified by accessor string. Low-level spec captures
4043  * every single level of nestedness, including traversing anonymous
4044  * struct/union members. High-level one only captures semantically meaningful
4045  * "turning points": named fields and array indicies.
4046  * E.g., for this case:
4047  *
4048  *   struct sample {
4049  *       int __unimportant;
4050  *       struct {
4051  *           int __1;
4052  *           int __2;
4053  *           int a[7];
4054  *       };
4055  *   };
4056  *
4057  *   struct sample *s = ...;
4058  *
4059  *   int x = &s->a[3]; // access string = '0:1:2:3'
4060  *
4061  * Low-level spec has 1:1 mapping with each element of access string (it's
4062  * just a parsed access string representation): [0, 1, 2, 3].
4063  *
4064  * High-level spec will capture only 3 points:
4065  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
4066  *   - field 'a' access (corresponds to '2' in low-level spec);
4067  *   - array element #3 access (corresponds to '3' in low-level spec).
4068  *
4069  */
4070 static int bpf_core_spec_parse(const struct btf *btf,
4071                                __u32 type_id,
4072                                const char *spec_str,
4073                                struct bpf_core_spec *spec)
4074 {
4075         int access_idx, parsed_len, i;
4076         struct bpf_core_accessor *acc;
4077         const struct btf_type *t;
4078         const char *name;
4079         __u32 id;
4080         __s64 sz;
4081
4082         if (str_is_empty(spec_str) || *spec_str == ':')
4083                 return -EINVAL;
4084
4085         memset(spec, 0, sizeof(*spec));
4086         spec->btf = btf;
4087
4088         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
4089         while (*spec_str) {
4090                 if (*spec_str == ':')
4091                         ++spec_str;
4092                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
4093                         return -EINVAL;
4094                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4095                         return -E2BIG;
4096                 spec_str += parsed_len;
4097                 spec->raw_spec[spec->raw_len++] = access_idx;
4098         }
4099
4100         if (spec->raw_len == 0)
4101                 return -EINVAL;
4102
4103         /* first spec value is always reloc type array index */
4104         t = skip_mods_and_typedefs(btf, type_id, &id);
4105         if (!t)
4106                 return -EINVAL;
4107
4108         access_idx = spec->raw_spec[0];
4109         spec->spec[0].type_id = id;
4110         spec->spec[0].idx = access_idx;
4111         spec->len++;
4112
4113         sz = btf__resolve_size(btf, id);
4114         if (sz < 0)
4115                 return sz;
4116         spec->bit_offset = access_idx * sz * 8;
4117
4118         for (i = 1; i < spec->raw_len; i++) {
4119                 t = skip_mods_and_typedefs(btf, id, &id);
4120                 if (!t)
4121                         return -EINVAL;
4122
4123                 access_idx = spec->raw_spec[i];
4124                 acc = &spec->spec[spec->len];
4125
4126                 if (btf_is_composite(t)) {
4127                         const struct btf_member *m;
4128                         __u32 bit_offset;
4129
4130                         if (access_idx >= btf_vlen(t))
4131                                 return -EINVAL;
4132
4133                         bit_offset = btf_member_bit_offset(t, access_idx);
4134                         spec->bit_offset += bit_offset;
4135
4136                         m = btf_members(t) + access_idx;
4137                         if (m->name_off) {
4138                                 name = btf__name_by_offset(btf, m->name_off);
4139                                 if (str_is_empty(name))
4140                                         return -EINVAL;
4141
4142                                 acc->type_id = id;
4143                                 acc->idx = access_idx;
4144                                 acc->name = name;
4145                                 spec->len++;
4146                         }
4147
4148                         id = m->type;
4149                 } else if (btf_is_array(t)) {
4150                         const struct btf_array *a = btf_array(t);
4151                         bool flex;
4152
4153                         t = skip_mods_and_typedefs(btf, a->type, &id);
4154                         if (!t)
4155                                 return -EINVAL;
4156
4157                         flex = is_flex_arr(btf, acc - 1, a);
4158                         if (!flex && access_idx >= a->nelems)
4159                                 return -EINVAL;
4160
4161                         spec->spec[spec->len].type_id = id;
4162                         spec->spec[spec->len].idx = access_idx;
4163                         spec->len++;
4164
4165                         sz = btf__resolve_size(btf, id);
4166                         if (sz < 0)
4167                                 return sz;
4168                         spec->bit_offset += access_idx * sz * 8;
4169                 } else {
4170                         pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
4171                                 type_id, spec_str, i, id, btf_kind(t));
4172                         return -EINVAL;
4173                 }
4174         }
4175
4176         return 0;
4177 }
4178
4179 static bool bpf_core_is_flavor_sep(const char *s)
4180 {
4181         /* check X___Y name pattern, where X and Y are not underscores */
4182         return s[0] != '_' &&                                 /* X */
4183                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
4184                s[4] != '_';                                   /* Y */
4185 }
4186
4187 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
4188  * before last triple underscore. Struct name part after last triple
4189  * underscore is ignored by BPF CO-RE relocation during relocation matching.
4190  */
4191 static size_t bpf_core_essential_name_len(const char *name)
4192 {
4193         size_t n = strlen(name);
4194         int i;
4195
4196         for (i = n - 5; i >= 0; i--) {
4197                 if (bpf_core_is_flavor_sep(name + i))
4198                         return i + 1;
4199         }
4200         return n;
4201 }
4202
4203 /* dynamically sized list of type IDs */
4204 struct ids_vec {
4205         __u32 *data;
4206         int len;
4207 };
4208
4209 static void bpf_core_free_cands(struct ids_vec *cand_ids)
4210 {
4211         free(cand_ids->data);
4212         free(cand_ids);
4213 }
4214
4215 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
4216                                            __u32 local_type_id,
4217                                            const struct btf *targ_btf)
4218 {
4219         size_t local_essent_len, targ_essent_len;
4220         const char *local_name, *targ_name;
4221         const struct btf_type *t;
4222         struct ids_vec *cand_ids;
4223         __u32 *new_ids;
4224         int i, err, n;
4225
4226         t = btf__type_by_id(local_btf, local_type_id);
4227         if (!t)
4228                 return ERR_PTR(-EINVAL);
4229
4230         local_name = btf__name_by_offset(local_btf, t->name_off);
4231         if (str_is_empty(local_name))
4232                 return ERR_PTR(-EINVAL);
4233         local_essent_len = bpf_core_essential_name_len(local_name);
4234
4235         cand_ids = calloc(1, sizeof(*cand_ids));
4236         if (!cand_ids)
4237                 return ERR_PTR(-ENOMEM);
4238
4239         n = btf__get_nr_types(targ_btf);
4240         for (i = 1; i <= n; i++) {
4241                 t = btf__type_by_id(targ_btf, i);
4242                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
4243                 if (str_is_empty(targ_name))
4244                         continue;
4245
4246                 t = skip_mods_and_typedefs(targ_btf, i, NULL);
4247                 if (!btf_is_composite(t) && !btf_is_array(t))
4248                         continue;
4249
4250                 targ_essent_len = bpf_core_essential_name_len(targ_name);
4251                 if (targ_essent_len != local_essent_len)
4252                         continue;
4253
4254                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
4255                         pr_debug("[%d] %s: found candidate [%d] %s\n",
4256                                  local_type_id, local_name, i, targ_name);
4257                         new_ids = reallocarray(cand_ids->data,
4258                                                cand_ids->len + 1,
4259                                                sizeof(*cand_ids->data));
4260                         if (!new_ids) {
4261                                 err = -ENOMEM;
4262                                 goto err_out;
4263                         }
4264                         cand_ids->data = new_ids;
4265                         cand_ids->data[cand_ids->len++] = i;
4266                 }
4267         }
4268         return cand_ids;
4269 err_out:
4270         bpf_core_free_cands(cand_ids);
4271         return ERR_PTR(err);
4272 }
4273
4274 /* Check two types for compatibility, skipping const/volatile/restrict and
4275  * typedefs, to ensure we are relocating compatible entities:
4276  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
4277  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
4278  *   - any two PTRs are always compatible;
4279  *   - for ENUMs, names should be the same (ignoring flavor suffix) or at
4280  *     least one of enums should be anonymous;
4281  *   - for ENUMs, check sizes, names are ignored;
4282  *   - for INT, size and signedness are ignored;
4283  *   - for ARRAY, dimensionality is ignored, element types are checked for
4284  *     compatibility recursively;
4285  *   - everything else shouldn't be ever a target of relocation.
4286  * These rules are not set in stone and probably will be adjusted as we get
4287  * more experience with using BPF CO-RE relocations.
4288  */
4289 static int bpf_core_fields_are_compat(const struct btf *local_btf,
4290                                       __u32 local_id,
4291                                       const struct btf *targ_btf,
4292                                       __u32 targ_id)
4293 {
4294         const struct btf_type *local_type, *targ_type;
4295
4296 recur:
4297         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
4298         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4299         if (!local_type || !targ_type)
4300                 return -EINVAL;
4301
4302         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
4303                 return 1;
4304         if (btf_kind(local_type) != btf_kind(targ_type))
4305                 return 0;
4306
4307         switch (btf_kind(local_type)) {
4308         case BTF_KIND_PTR:
4309                 return 1;
4310         case BTF_KIND_FWD:
4311         case BTF_KIND_ENUM: {
4312                 const char *local_name, *targ_name;
4313                 size_t local_len, targ_len;
4314
4315                 local_name = btf__name_by_offset(local_btf,
4316                                                  local_type->name_off);
4317                 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
4318                 local_len = bpf_core_essential_name_len(local_name);
4319                 targ_len = bpf_core_essential_name_len(targ_name);
4320                 /* one of them is anonymous or both w/ same flavor-less names */
4321                 return local_len == 0 || targ_len == 0 ||
4322                        (local_len == targ_len &&
4323                         strncmp(local_name, targ_name, local_len) == 0);
4324         }
4325         case BTF_KIND_INT:
4326                 /* just reject deprecated bitfield-like integers; all other
4327                  * integers are by default compatible between each other
4328                  */
4329                 return btf_int_offset(local_type) == 0 &&
4330                        btf_int_offset(targ_type) == 0;
4331         case BTF_KIND_ARRAY:
4332                 local_id = btf_array(local_type)->type;
4333                 targ_id = btf_array(targ_type)->type;
4334                 goto recur;
4335         default:
4336                 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
4337                         btf_kind(local_type), local_id, targ_id);
4338                 return 0;
4339         }
4340 }
4341
4342 /*
4343  * Given single high-level named field accessor in local type, find
4344  * corresponding high-level accessor for a target type. Along the way,
4345  * maintain low-level spec for target as well. Also keep updating target
4346  * bit offset.
4347  *
4348  * Searching is performed through recursive exhaustive enumeration of all
4349  * fields of a struct/union. If there are any anonymous (embedded)
4350  * structs/unions, they are recursively searched as well. If field with
4351  * desired name is found, check compatibility between local and target types,
4352  * before returning result.
4353  *
4354  * 1 is returned, if field is found.
4355  * 0 is returned if no compatible field is found.
4356  * <0 is returned on error.
4357  */
4358 static int bpf_core_match_member(const struct btf *local_btf,
4359                                  const struct bpf_core_accessor *local_acc,
4360                                  const struct btf *targ_btf,
4361                                  __u32 targ_id,
4362                                  struct bpf_core_spec *spec,
4363                                  __u32 *next_targ_id)
4364 {
4365         const struct btf_type *local_type, *targ_type;
4366         const struct btf_member *local_member, *m;
4367         const char *local_name, *targ_name;
4368         __u32 local_id;
4369         int i, n, found;
4370
4371         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4372         if (!targ_type)
4373                 return -EINVAL;
4374         if (!btf_is_composite(targ_type))
4375                 return 0;
4376
4377         local_id = local_acc->type_id;
4378         local_type = btf__type_by_id(local_btf, local_id);
4379         local_member = btf_members(local_type) + local_acc->idx;
4380         local_name = btf__name_by_offset(local_btf, local_member->name_off);
4381
4382         n = btf_vlen(targ_type);
4383         m = btf_members(targ_type);
4384         for (i = 0; i < n; i++, m++) {
4385                 __u32 bit_offset;
4386
4387                 bit_offset = btf_member_bit_offset(targ_type, i);
4388
4389                 /* too deep struct/union/array nesting */
4390                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4391                         return -E2BIG;
4392
4393                 /* speculate this member will be the good one */
4394                 spec->bit_offset += bit_offset;
4395                 spec->raw_spec[spec->raw_len++] = i;
4396
4397                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
4398                 if (str_is_empty(targ_name)) {
4399                         /* embedded struct/union, we need to go deeper */
4400                         found = bpf_core_match_member(local_btf, local_acc,
4401                                                       targ_btf, m->type,
4402                                                       spec, next_targ_id);
4403                         if (found) /* either found or error */
4404                                 return found;
4405                 } else if (strcmp(local_name, targ_name) == 0) {
4406                         /* matching named field */
4407                         struct bpf_core_accessor *targ_acc;
4408
4409                         targ_acc = &spec->spec[spec->len++];
4410                         targ_acc->type_id = targ_id;
4411                         targ_acc->idx = i;
4412                         targ_acc->name = targ_name;
4413
4414                         *next_targ_id = m->type;
4415                         found = bpf_core_fields_are_compat(local_btf,
4416                                                            local_member->type,
4417                                                            targ_btf, m->type);
4418                         if (!found)
4419                                 spec->len--; /* pop accessor */
4420                         return found;
4421                 }
4422                 /* member turned out not to be what we looked for */
4423                 spec->bit_offset -= bit_offset;
4424                 spec->raw_len--;
4425         }
4426
4427         return 0;
4428 }
4429
4430 /*
4431  * Try to match local spec to a target type and, if successful, produce full
4432  * target spec (high-level, low-level + bit offset).
4433  */
4434 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
4435                                const struct btf *targ_btf, __u32 targ_id,
4436                                struct bpf_core_spec *targ_spec)
4437 {
4438         const struct btf_type *targ_type;
4439         const struct bpf_core_accessor *local_acc;
4440         struct bpf_core_accessor *targ_acc;
4441         int i, sz, matched;
4442
4443         memset(targ_spec, 0, sizeof(*targ_spec));
4444         targ_spec->btf = targ_btf;
4445
4446         local_acc = &local_spec->spec[0];
4447         targ_acc = &targ_spec->spec[0];
4448
4449         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
4450                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
4451                                                    &targ_id);
4452                 if (!targ_type)
4453                         return -EINVAL;
4454
4455                 if (local_acc->name) {
4456                         matched = bpf_core_match_member(local_spec->btf,
4457                                                         local_acc,
4458                                                         targ_btf, targ_id,
4459                                                         targ_spec, &targ_id);
4460                         if (matched <= 0)
4461                                 return matched;
4462                 } else {
4463                         /* for i=0, targ_id is already treated as array element
4464                          * type (because it's the original struct), for others
4465                          * we should find array element type first
4466                          */
4467                         if (i > 0) {
4468                                 const struct btf_array *a;
4469                                 bool flex;
4470
4471                                 if (!btf_is_array(targ_type))
4472                                         return 0;
4473
4474                                 a = btf_array(targ_type);
4475                                 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
4476                                 if (!flex && local_acc->idx >= a->nelems)
4477                                         return 0;
4478                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
4479                                                             &targ_id))
4480                                         return -EINVAL;
4481                         }
4482
4483                         /* too deep struct/union/array nesting */
4484                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4485                                 return -E2BIG;
4486
4487                         targ_acc->type_id = targ_id;
4488                         targ_acc->idx = local_acc->idx;
4489                         targ_acc->name = NULL;
4490                         targ_spec->len++;
4491                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
4492                         targ_spec->raw_len++;
4493
4494                         sz = btf__resolve_size(targ_btf, targ_id);
4495                         if (sz < 0)
4496                                 return sz;
4497                         targ_spec->bit_offset += local_acc->idx * sz * 8;
4498                 }
4499         }
4500
4501         return 1;
4502 }
4503
4504 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
4505                                     const struct bpf_field_reloc *relo,
4506                                     const struct bpf_core_spec *spec,
4507                                     __u32 *val, bool *validate)
4508 {
4509         const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
4510         const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
4511         __u32 byte_off, byte_sz, bit_off, bit_sz;
4512         const struct btf_member *m;
4513         const struct btf_type *mt;
4514         bool bitfield;
4515         __s64 sz;
4516
4517         /* a[n] accessor needs special handling */
4518         if (!acc->name) {
4519                 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
4520                         *val = spec->bit_offset / 8;
4521                 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
4522                         sz = btf__resolve_size(spec->btf, acc->type_id);
4523                         if (sz < 0)
4524                                 return -EINVAL;
4525                         *val = sz;
4526                 } else {
4527                         pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
4528                                 bpf_program__title(prog, false),
4529                                 relo->kind, relo->insn_off / 8);
4530                         return -EINVAL;
4531                 }
4532                 if (validate)
4533                         *validate = true;
4534                 return 0;
4535         }
4536
4537         m = btf_members(t) + acc->idx;
4538         mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
4539         bit_off = spec->bit_offset;
4540         bit_sz = btf_member_bitfield_size(t, acc->idx);
4541
4542         bitfield = bit_sz > 0;
4543         if (bitfield) {
4544                 byte_sz = mt->size;
4545                 byte_off = bit_off / 8 / byte_sz * byte_sz;
4546                 /* figure out smallest int size necessary for bitfield load */
4547                 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
4548                         if (byte_sz >= 8) {
4549                                 /* bitfield can't be read with 64-bit read */
4550                                 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
4551                                         bpf_program__title(prog, false),
4552                                         relo->kind, relo->insn_off / 8);
4553                                 return -E2BIG;
4554                         }
4555                         byte_sz *= 2;
4556                         byte_off = bit_off / 8 / byte_sz * byte_sz;
4557                 }
4558         } else {
4559                 sz = btf__resolve_size(spec->btf, m->type);
4560                 if (sz < 0)
4561                         return -EINVAL;
4562                 byte_sz = sz;
4563                 byte_off = spec->bit_offset / 8;
4564                 bit_sz = byte_sz * 8;
4565         }
4566
4567         /* for bitfields, all the relocatable aspects are ambiguous and we
4568          * might disagree with compiler, so turn off validation of expected
4569          * value, except for signedness
4570          */
4571         if (validate)
4572                 *validate = !bitfield;
4573
4574         switch (relo->kind) {
4575         case BPF_FIELD_BYTE_OFFSET:
4576                 *val = byte_off;
4577                 break;
4578         case BPF_FIELD_BYTE_SIZE:
4579                 *val = byte_sz;
4580                 break;
4581         case BPF_FIELD_SIGNED:
4582                 /* enums will be assumed unsigned */
4583                 *val = btf_is_enum(mt) ||
4584                        (btf_int_encoding(mt) & BTF_INT_SIGNED);
4585                 if (validate)
4586                         *validate = true; /* signedness is never ambiguous */
4587                 break;
4588         case BPF_FIELD_LSHIFT_U64:
4589 #if __BYTE_ORDER == __LITTLE_ENDIAN
4590                 *val = 64 - (bit_off + bit_sz - byte_off  * 8);
4591 #else
4592                 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
4593 #endif
4594                 break;
4595         case BPF_FIELD_RSHIFT_U64:
4596                 *val = 64 - bit_sz;
4597                 if (validate)
4598                         *validate = true; /* right shift is never ambiguous */
4599                 break;
4600         case BPF_FIELD_EXISTS:
4601         default:
4602                 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
4603                         bpf_program__title(prog, false),
4604                         relo->kind, relo->insn_off / 8);
4605                 return -EINVAL;
4606         }
4607
4608         return 0;
4609 }
4610
4611 /*
4612  * Patch relocatable BPF instruction.
4613  *
4614  * Patched value is determined by relocation kind and target specification.
4615  * For field existence relocation target spec will be NULL if field is not
4616  * found.
4617  * Expected insn->imm value is determined using relocation kind and local
4618  * spec, and is checked before patching instruction. If actual insn->imm value
4619  * is wrong, bail out with error.
4620  *
4621  * Currently three kinds of BPF instructions are supported:
4622  * 1. rX = <imm> (assignment with immediate operand);
4623  * 2. rX += <imm> (arithmetic operations with immediate operand);
4624  */
4625 static int bpf_core_reloc_insn(struct bpf_program *prog,
4626                                const struct bpf_field_reloc *relo,
4627                                int relo_idx,
4628                                const struct bpf_core_spec *local_spec,
4629                                const struct bpf_core_spec *targ_spec)
4630 {
4631         __u32 orig_val, new_val;
4632         struct bpf_insn *insn;
4633         bool validate = true;
4634         int insn_idx, err;
4635         __u8 class;
4636
4637         if (relo->insn_off % sizeof(struct bpf_insn))
4638                 return -EINVAL;
4639         insn_idx = relo->insn_off / sizeof(struct bpf_insn);
4640         insn = &prog->insns[insn_idx];
4641         class = BPF_CLASS(insn->code);
4642
4643         if (relo->kind == BPF_FIELD_EXISTS) {
4644                 orig_val = 1; /* can't generate EXISTS relo w/o local field */
4645                 new_val = targ_spec ? 1 : 0;
4646         } else if (!targ_spec) {
4647                 pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n",
4648                          bpf_program__title(prog, false), relo_idx, insn_idx);
4649                 insn->code = BPF_JMP | BPF_CALL;
4650                 insn->dst_reg = 0;
4651                 insn->src_reg = 0;
4652                 insn->off = 0;
4653                 /* if this instruction is reachable (not a dead code),
4654                  * verifier will complain with the following message:
4655                  * invalid func unknown#195896080
4656                  */
4657                 insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */
4658                 return 0;
4659         } else {
4660                 err = bpf_core_calc_field_relo(prog, relo, local_spec,
4661                                                &orig_val, &validate);
4662                 if (err)
4663                         return err;
4664                 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
4665                                                &new_val, NULL);
4666                 if (err)
4667                         return err;
4668         }
4669
4670         switch (class) {
4671         case BPF_ALU:
4672         case BPF_ALU64:
4673                 if (BPF_SRC(insn->code) != BPF_K)
4674                         return -EINVAL;
4675                 if (validate && insn->imm != orig_val) {
4676                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
4677                                 bpf_program__title(prog, false), relo_idx,
4678                                 insn_idx, insn->imm, orig_val, new_val);
4679                         return -EINVAL;
4680                 }
4681                 orig_val = insn->imm;
4682                 insn->imm = new_val;
4683                 pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n",
4684                          bpf_program__title(prog, false), relo_idx, insn_idx,
4685                          orig_val, new_val);
4686                 break;
4687         case BPF_LDX:
4688         case BPF_ST:
4689         case BPF_STX:
4690                 if (validate && insn->off != orig_val) {
4691                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (LD/LDX/ST/STX) value: got %u, exp %u -> %u\n",
4692                                 bpf_program__title(prog, false), relo_idx,
4693                                 insn_idx, insn->off, orig_val, new_val);
4694                         return -EINVAL;
4695                 }
4696                 if (new_val > SHRT_MAX) {
4697                         pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n",
4698                                 bpf_program__title(prog, false), relo_idx,
4699                                 insn_idx, new_val);
4700                         return -ERANGE;
4701                 }
4702                 orig_val = insn->off;
4703                 insn->off = new_val;
4704                 pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n",
4705                          bpf_program__title(prog, false), relo_idx, insn_idx,
4706                          orig_val, new_val);
4707                 break;
4708         default:
4709                 pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
4710                         bpf_program__title(prog, false), relo_idx,
4711                         insn_idx, insn->code, insn->src_reg, insn->dst_reg,
4712                         insn->off, insn->imm);
4713                 return -EINVAL;
4714         }
4715
4716         return 0;
4717 }
4718
4719 /* Output spec definition in the format:
4720  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
4721  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
4722  */
4723 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
4724 {
4725         const struct btf_type *t;
4726         const char *s;
4727         __u32 type_id;
4728         int i;
4729
4730         type_id = spec->spec[0].type_id;
4731         t = btf__type_by_id(spec->btf, type_id);
4732         s = btf__name_by_offset(spec->btf, t->name_off);
4733         libbpf_print(level, "[%u] %s + ", type_id, s);
4734
4735         for (i = 0; i < spec->raw_len; i++)
4736                 libbpf_print(level, "%d%s", spec->raw_spec[i],
4737                              i == spec->raw_len - 1 ? " => " : ":");
4738
4739         libbpf_print(level, "%u.%u @ &x",
4740                      spec->bit_offset / 8, spec->bit_offset % 8);
4741
4742         for (i = 0; i < spec->len; i++) {
4743                 if (spec->spec[i].name)
4744                         libbpf_print(level, ".%s", spec->spec[i].name);
4745                 else
4746                         libbpf_print(level, "[%u]", spec->spec[i].idx);
4747         }
4748
4749 }
4750
4751 static size_t bpf_core_hash_fn(const void *key, void *ctx)
4752 {
4753         return (size_t)key;
4754 }
4755
4756 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
4757 {
4758         return k1 == k2;
4759 }
4760
4761 static void *u32_as_hash_key(__u32 x)
4762 {
4763         return (void *)(uintptr_t)x;
4764 }
4765
4766 /*
4767  * CO-RE relocate single instruction.
4768  *
4769  * The outline and important points of the algorithm:
4770  * 1. For given local type, find corresponding candidate target types.
4771  *    Candidate type is a type with the same "essential" name, ignoring
4772  *    everything after last triple underscore (___). E.g., `sample`,
4773  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
4774  *    for each other. Names with triple underscore are referred to as
4775  *    "flavors" and are useful, among other things, to allow to
4776  *    specify/support incompatible variations of the same kernel struct, which
4777  *    might differ between different kernel versions and/or build
4778  *    configurations.
4779  *
4780  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
4781  *    converter, when deduplicated BTF of a kernel still contains more than
4782  *    one different types with the same name. In that case, ___2, ___3, etc
4783  *    are appended starting from second name conflict. But start flavors are
4784  *    also useful to be defined "locally", in BPF program, to extract same
4785  *    data from incompatible changes between different kernel
4786  *    versions/configurations. For instance, to handle field renames between
4787  *    kernel versions, one can use two flavors of the struct name with the
4788  *    same common name and use conditional relocations to extract that field,
4789  *    depending on target kernel version.
4790  * 2. For each candidate type, try to match local specification to this
4791  *    candidate target type. Matching involves finding corresponding
4792  *    high-level spec accessors, meaning that all named fields should match,
4793  *    as well as all array accesses should be within the actual bounds. Also,
4794  *    types should be compatible (see bpf_core_fields_are_compat for details).
4795  * 3. It is supported and expected that there might be multiple flavors
4796  *    matching the spec. As long as all the specs resolve to the same set of
4797  *    offsets across all candidates, there is no error. If there is any
4798  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
4799  *    imprefection of BTF deduplication, which can cause slight duplication of
4800  *    the same BTF type, if some directly or indirectly referenced (by
4801  *    pointer) type gets resolved to different actual types in different
4802  *    object files. If such situation occurs, deduplicated BTF will end up
4803  *    with two (or more) structurally identical types, which differ only in
4804  *    types they refer to through pointer. This should be OK in most cases and
4805  *    is not an error.
4806  * 4. Candidate types search is performed by linearly scanning through all
4807  *    types in target BTF. It is anticipated that this is overall more
4808  *    efficient memory-wise and not significantly worse (if not better)
4809  *    CPU-wise compared to prebuilding a map from all local type names to
4810  *    a list of candidate type names. It's also sped up by caching resolved
4811  *    list of matching candidates per each local "root" type ID, that has at
4812  *    least one bpf_field_reloc associated with it. This list is shared
4813  *    between multiple relocations for the same type ID and is updated as some
4814  *    of the candidates are pruned due to structural incompatibility.
4815  */
4816 static int bpf_core_reloc_field(struct bpf_program *prog,
4817                                  const struct bpf_field_reloc *relo,
4818                                  int relo_idx,
4819                                  const struct btf *local_btf,
4820                                  const struct btf *targ_btf,
4821                                  struct hashmap *cand_cache)
4822 {
4823         const char *prog_name = bpf_program__title(prog, false);
4824         struct bpf_core_spec local_spec, cand_spec, targ_spec;
4825         const void *type_key = u32_as_hash_key(relo->type_id);
4826         const struct btf_type *local_type, *cand_type;
4827         const char *local_name, *cand_name;
4828         struct ids_vec *cand_ids;
4829         __u32 local_id, cand_id;
4830         const char *spec_str;
4831         int i, j, err;
4832
4833         local_id = relo->type_id;
4834         local_type = btf__type_by_id(local_btf, local_id);
4835         if (!local_type)
4836                 return -EINVAL;
4837
4838         local_name = btf__name_by_offset(local_btf, local_type->name_off);
4839         if (str_is_empty(local_name))
4840                 return -EINVAL;
4841
4842         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
4843         if (str_is_empty(spec_str))
4844                 return -EINVAL;
4845
4846         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
4847         if (err) {
4848                 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
4849                         prog_name, relo_idx, local_id, local_name, spec_str,
4850                         err);
4851                 return -EINVAL;
4852         }
4853
4854         pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
4855                  relo->kind);
4856         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
4857         libbpf_print(LIBBPF_DEBUG, "\n");
4858
4859         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
4860                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
4861                 if (IS_ERR(cand_ids)) {
4862                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
4863                                 prog_name, relo_idx, local_id, local_name,
4864                                 PTR_ERR(cand_ids));
4865                         return PTR_ERR(cand_ids);
4866                 }
4867                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
4868                 if (err) {
4869                         bpf_core_free_cands(cand_ids);
4870                         return err;
4871                 }
4872         }
4873
4874         for (i = 0, j = 0; i < cand_ids->len; i++) {
4875                 cand_id = cand_ids->data[i];
4876                 cand_type = btf__type_by_id(targ_btf, cand_id);
4877                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
4878
4879                 err = bpf_core_spec_match(&local_spec, targ_btf,
4880                                           cand_id, &cand_spec);
4881                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
4882                          prog_name, relo_idx, i, cand_name);
4883                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
4884                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
4885                 if (err < 0) {
4886                         pr_warn("prog '%s': relo #%d: matching error: %d\n",
4887                                 prog_name, relo_idx, err);
4888                         return err;
4889                 }
4890                 if (err == 0)
4891                         continue;
4892
4893                 if (j == 0) {
4894                         targ_spec = cand_spec;
4895                 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
4896                         /* if there are many candidates, they should all
4897                          * resolve to the same bit offset
4898                          */
4899                         pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
4900                                 prog_name, relo_idx, cand_spec.bit_offset,
4901                                 targ_spec.bit_offset);
4902                         return -EINVAL;
4903                 }
4904
4905                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
4906         }
4907
4908         /*
4909          * For BPF_FIELD_EXISTS relo or when used BPF program has field
4910          * existence checks or kernel version/config checks, it's expected
4911          * that we might not find any candidates. In this case, if field
4912          * wasn't found in any candidate, the list of candidates shouldn't
4913          * change at all, we'll just handle relocating appropriately,
4914          * depending on relo's kind.
4915          */
4916         if (j > 0)
4917                 cand_ids->len = j;
4918
4919         /*
4920          * If no candidates were found, it might be both a programmer error,
4921          * as well as expected case, depending whether instruction w/
4922          * relocation is guarded in some way that makes it unreachable (dead
4923          * code) if relocation can't be resolved. This is handled in
4924          * bpf_core_reloc_insn() uniformly by replacing that instruction with
4925          * BPF helper call insn (using invalid helper ID). If that instruction
4926          * is indeed unreachable, then it will be ignored and eliminated by
4927          * verifier. If it was an error, then verifier will complain and point
4928          * to a specific instruction number in its log.
4929          */
4930         if (j == 0)
4931                 pr_debug("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
4932                          prog_name, relo_idx, local_id, local_name, spec_str);
4933
4934         /* bpf_core_reloc_insn should know how to handle missing targ_spec */
4935         err = bpf_core_reloc_insn(prog, relo, relo_idx, &local_spec,
4936                                   j ? &targ_spec : NULL);
4937         if (err) {
4938                 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
4939                         prog_name, relo_idx, relo->insn_off, err);
4940                 return -EINVAL;
4941         }
4942
4943         return 0;
4944 }
4945
4946 static int
4947 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
4948 {
4949         const struct btf_ext_info_sec *sec;
4950         const struct bpf_field_reloc *rec;
4951         const struct btf_ext_info *seg;
4952         struct hashmap_entry *entry;
4953         struct hashmap *cand_cache = NULL;
4954         struct bpf_program *prog;
4955         struct btf *targ_btf;
4956         const char *sec_name;
4957         int i, err = 0;
4958
4959         if (targ_btf_path)
4960                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
4961         else
4962                 targ_btf = obj->btf_vmlinux;
4963         if (IS_ERR_OR_NULL(targ_btf)) {
4964                 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
4965                 return PTR_ERR(targ_btf);
4966         }
4967
4968         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
4969         if (IS_ERR(cand_cache)) {
4970                 err = PTR_ERR(cand_cache);
4971                 goto out;
4972         }
4973
4974         seg = &obj->btf_ext->field_reloc_info;
4975         for_each_btf_ext_sec(seg, sec) {
4976                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
4977                 if (str_is_empty(sec_name)) {
4978                         err = -EINVAL;
4979                         goto out;
4980                 }
4981                 prog = NULL;
4982                 for (i = 0; i < obj->nr_programs; i++) {
4983                         if (!strcmp(obj->programs[i].section_name, sec_name)) {
4984                                 prog = &obj->programs[i];
4985                                 break;
4986                         }
4987                 }
4988                 if (!prog) {
4989                         pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
4990                                 sec_name);
4991                         err = -EINVAL;
4992                         goto out;
4993                 }
4994
4995                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
4996                          sec_name, sec->num_info);
4997
4998                 for_each_btf_ext_rec(seg, sec, i, rec) {
4999                         err = bpf_core_reloc_field(prog, rec, i, obj->btf,
5000                                                    targ_btf, cand_cache);
5001                         if (err) {
5002                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5003                                         sec_name, i, err);
5004                                 goto out;
5005                         }
5006                 }
5007         }
5008
5009 out:
5010         /* obj->btf_vmlinux is freed at the end of object load phase */
5011         if (targ_btf != obj->btf_vmlinux)
5012                 btf__free(targ_btf);
5013         if (!IS_ERR_OR_NULL(cand_cache)) {
5014                 hashmap__for_each_entry(cand_cache, entry, i) {
5015                         bpf_core_free_cands(entry->value);
5016                 }
5017                 hashmap__free(cand_cache);
5018         }
5019         return err;
5020 }
5021
5022 static int
5023 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5024 {
5025         int err = 0;
5026
5027         if (obj->btf_ext->field_reloc_info.len)
5028                 err = bpf_core_reloc_fields(obj, targ_btf_path);
5029
5030         return err;
5031 }
5032
5033 static int
5034 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
5035                         struct reloc_desc *relo)
5036 {
5037         struct bpf_insn *insn, *new_insn;
5038         struct bpf_program *text;
5039         size_t new_cnt;
5040         int err;
5041
5042         if (prog->idx != obj->efile.text_shndx && prog->main_prog_cnt == 0) {
5043                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
5044                 if (!text) {
5045                         pr_warn("no .text section found yet relo into text exist\n");
5046                         return -LIBBPF_ERRNO__RELOC;
5047                 }
5048                 new_cnt = prog->insns_cnt + text->insns_cnt;
5049                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
5050                 if (!new_insn) {
5051                         pr_warn("oom in prog realloc\n");
5052                         return -ENOMEM;
5053                 }
5054                 prog->insns = new_insn;
5055
5056                 if (obj->btf_ext) {
5057                         err = bpf_program_reloc_btf_ext(prog, obj,
5058                                                         text->section_name,
5059                                                         prog->insns_cnt);
5060                         if (err)
5061                                 return err;
5062                 }
5063
5064                 memcpy(new_insn + prog->insns_cnt, text->insns,
5065                        text->insns_cnt * sizeof(*insn));
5066                 prog->main_prog_cnt = prog->insns_cnt;
5067                 prog->insns_cnt = new_cnt;
5068                 pr_debug("added %zd insn from %s to prog %s\n",
5069                          text->insns_cnt, text->section_name,
5070                          prog->section_name);
5071         }
5072
5073         insn = &prog->insns[relo->insn_idx];
5074         insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
5075         return 0;
5076 }
5077
5078 static int
5079 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
5080 {
5081         int i, err;
5082
5083         if (!prog)
5084                 return 0;
5085
5086         if (obj->btf_ext) {
5087                 err = bpf_program_reloc_btf_ext(prog, obj,
5088                                                 prog->section_name, 0);
5089                 if (err)
5090                         return err;
5091         }
5092
5093         if (!prog->reloc_desc)
5094                 return 0;
5095
5096         for (i = 0; i < prog->nr_reloc; i++) {
5097                 struct reloc_desc *relo = &prog->reloc_desc[i];
5098                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5099                 struct extern_desc *ext;
5100
5101                 if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
5102                         pr_warn("relocation out of range: '%s'\n",
5103                                 prog->section_name);
5104                         return -LIBBPF_ERRNO__RELOC;
5105                 }
5106
5107                 switch (relo->type) {
5108                 case RELO_LD64:
5109                         insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5110                         insn[0].imm = obj->maps[relo->map_idx].fd;
5111                         break;
5112                 case RELO_DATA:
5113                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5114                         insn[1].imm = insn[0].imm + relo->sym_off;
5115                         insn[0].imm = obj->maps[relo->map_idx].fd;
5116                         break;
5117                 case RELO_EXTERN:
5118                         ext = &obj->externs[relo->sym_off];
5119                         if (ext->type == EXT_KCFG) {
5120                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5121                                 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5122                                 insn[1].imm = ext->kcfg.data_off;
5123                         } else /* EXT_KSYM */ {
5124                                 insn[0].imm = (__u32)ext->ksym.addr;
5125                                 insn[1].imm = ext->ksym.addr >> 32;
5126                         }
5127                         break;
5128                 case RELO_CALL:
5129                         err = bpf_program__reloc_text(prog, obj, relo);
5130                         if (err)
5131                                 return err;
5132                         break;
5133                 default:
5134                         pr_warn("relo #%d: bad relo type %d\n", i, relo->type);
5135                         return -EINVAL;
5136                 }
5137         }
5138
5139         zfree(&prog->reloc_desc);
5140         prog->nr_reloc = 0;
5141         return 0;
5142 }
5143
5144 static int
5145 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
5146 {
5147         struct bpf_program *prog;
5148         size_t i;
5149         int err;
5150
5151         if (obj->btf_ext) {
5152                 err = bpf_object__relocate_core(obj, targ_btf_path);
5153                 if (err) {
5154                         pr_warn("failed to perform CO-RE relocations: %d\n",
5155                                 err);
5156                         return err;
5157                 }
5158         }
5159         /* ensure .text is relocated first, as it's going to be copied as-is
5160          * later for sub-program calls
5161          */
5162         for (i = 0; i < obj->nr_programs; i++) {
5163                 prog = &obj->programs[i];
5164                 if (prog->idx != obj->efile.text_shndx)
5165                         continue;
5166
5167                 err = bpf_program__relocate(prog, obj);
5168                 if (err) {
5169                         pr_warn("failed to relocate '%s'\n", prog->section_name);
5170                         return err;
5171                 }
5172                 break;
5173         }
5174         /* now relocate everything but .text, which by now is relocated
5175          * properly, so we can copy raw sub-program instructions as is safely
5176          */
5177         for (i = 0; i < obj->nr_programs; i++) {
5178                 prog = &obj->programs[i];
5179                 if (prog->idx == obj->efile.text_shndx)
5180                         continue;
5181
5182                 err = bpf_program__relocate(prog, obj);
5183                 if (err) {
5184                         pr_warn("failed to relocate '%s'\n", prog->section_name);
5185                         return err;
5186                 }
5187         }
5188         return 0;
5189 }
5190
5191 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
5192                                             GElf_Shdr *shdr, Elf_Data *data);
5193
5194 static int bpf_object__collect_map_relos(struct bpf_object *obj,
5195                                          GElf_Shdr *shdr, Elf_Data *data)
5196 {
5197         int i, j, nrels, new_sz, ptr_sz = sizeof(void *);
5198         const struct btf_var_secinfo *vi = NULL;
5199         const struct btf_type *sec, *var, *def;
5200         const struct btf_member *member;
5201         struct bpf_map *map, *targ_map;
5202         const char *name, *mname;
5203         Elf_Data *symbols;
5204         unsigned int moff;
5205         GElf_Sym sym;
5206         GElf_Rel rel;
5207         void *tmp;
5208
5209         if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
5210                 return -EINVAL;
5211         sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
5212         if (!sec)
5213                 return -EINVAL;
5214
5215         symbols = obj->efile.symbols;
5216         nrels = shdr->sh_size / shdr->sh_entsize;
5217         for (i = 0; i < nrels; i++) {
5218                 if (!gelf_getrel(data, i, &rel)) {
5219                         pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
5220                         return -LIBBPF_ERRNO__FORMAT;
5221                 }
5222                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
5223                         pr_warn(".maps relo #%d: symbol %zx not found\n",
5224                                 i, (size_t)GELF_R_SYM(rel.r_info));
5225                         return -LIBBPF_ERRNO__FORMAT;
5226                 }
5227                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
5228                                   sym.st_name) ? : "<?>";
5229                 if (sym.st_shndx != obj->efile.btf_maps_shndx) {
5230                         pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
5231                                 i, name);
5232                         return -LIBBPF_ERRNO__RELOC;
5233                 }
5234
5235                 pr_debug(".maps relo #%d: for %zd value %zd rel.r_offset %zu name %d ('%s')\n",
5236                          i, (ssize_t)(rel.r_info >> 32), (size_t)sym.st_value,
5237                          (size_t)rel.r_offset, sym.st_name, name);
5238
5239                 for (j = 0; j < obj->nr_maps; j++) {
5240                         map = &obj->maps[j];
5241                         if (map->sec_idx != obj->efile.btf_maps_shndx)
5242                                 continue;
5243
5244                         vi = btf_var_secinfos(sec) + map->btf_var_idx;
5245                         if (vi->offset <= rel.r_offset &&
5246                             rel.r_offset + sizeof(void *) <= vi->offset + vi->size)
5247                                 break;
5248                 }
5249                 if (j == obj->nr_maps) {
5250                         pr_warn(".maps relo #%d: cannot find map '%s' at rel.r_offset %zu\n",
5251                                 i, name, (size_t)rel.r_offset);
5252                         return -EINVAL;
5253                 }
5254
5255                 if (!bpf_map_type__is_map_in_map(map->def.type))
5256                         return -EINVAL;
5257                 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
5258                     map->def.key_size != sizeof(int)) {
5259                         pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
5260                                 i, map->name, sizeof(int));
5261                         return -EINVAL;
5262                 }
5263
5264                 targ_map = bpf_object__find_map_by_name(obj, name);
5265                 if (!targ_map)
5266                         return -ESRCH;
5267
5268                 var = btf__type_by_id(obj->btf, vi->type);
5269                 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
5270                 if (btf_vlen(def) == 0)
5271                         return -EINVAL;
5272                 member = btf_members(def) + btf_vlen(def) - 1;
5273                 mname = btf__name_by_offset(obj->btf, member->name_off);
5274                 if (strcmp(mname, "values"))
5275                         return -EINVAL;
5276
5277                 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
5278                 if (rel.r_offset - vi->offset < moff)
5279                         return -EINVAL;
5280
5281                 moff = rel.r_offset - vi->offset - moff;
5282                 if (moff % ptr_sz)
5283                         return -EINVAL;
5284                 moff /= ptr_sz;
5285                 if (moff >= map->init_slots_sz) {
5286                         new_sz = moff + 1;
5287                         tmp = realloc(map->init_slots, new_sz * ptr_sz);
5288                         if (!tmp)
5289                                 return -ENOMEM;
5290                         map->init_slots = tmp;
5291                         memset(map->init_slots + map->init_slots_sz, 0,
5292                                (new_sz - map->init_slots_sz) * ptr_sz);
5293                         map->init_slots_sz = new_sz;
5294                 }
5295                 map->init_slots[moff] = targ_map;
5296
5297                 pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n",
5298                          i, map->name, moff, name);
5299         }
5300
5301         return 0;
5302 }
5303
5304 static int bpf_object__collect_reloc(struct bpf_object *obj)
5305 {
5306         int i, err;
5307
5308         if (!obj_elf_valid(obj)) {
5309                 pr_warn("Internal error: elf object is closed\n");
5310                 return -LIBBPF_ERRNO__INTERNAL;
5311         }
5312
5313         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
5314                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
5315                 Elf_Data *data = obj->efile.reloc_sects[i].data;
5316                 int idx = shdr->sh_info;
5317                 struct bpf_program *prog;
5318
5319                 if (shdr->sh_type != SHT_REL) {
5320                         pr_warn("internal error at %d\n", __LINE__);
5321                         return -LIBBPF_ERRNO__INTERNAL;
5322                 }
5323
5324                 if (idx == obj->efile.st_ops_shndx) {
5325                         err = bpf_object__collect_st_ops_relos(obj, shdr, data);
5326                 } else if (idx == obj->efile.btf_maps_shndx) {
5327                         err = bpf_object__collect_map_relos(obj, shdr, data);
5328                 } else {
5329                         prog = bpf_object__find_prog_by_idx(obj, idx);
5330                         if (!prog) {
5331                                 pr_warn("relocation failed: no prog in section(%d)\n", idx);
5332                                 return -LIBBPF_ERRNO__RELOC;
5333                         }
5334                         err = bpf_program__collect_reloc(prog, shdr, data, obj);
5335                 }
5336                 if (err)
5337                         return err;
5338         }
5339         return 0;
5340 }
5341
5342 static int
5343 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
5344              char *license, __u32 kern_version, int *pfd)
5345 {
5346         struct bpf_load_program_attr load_attr;
5347         char *cp, errmsg[STRERR_BUFSIZE];
5348         size_t log_buf_size = 0;
5349         char *log_buf = NULL;
5350         int btf_fd, ret;
5351
5352         if (!insns || !insns_cnt)
5353                 return -EINVAL;
5354
5355         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
5356         load_attr.prog_type = prog->type;
5357         /* old kernels might not support specifying expected_attach_type */
5358         if (!prog->caps->exp_attach_type && prog->sec_def &&
5359             prog->sec_def->is_exp_attach_type_optional)
5360                 load_attr.expected_attach_type = 0;
5361         else
5362                 load_attr.expected_attach_type = prog->expected_attach_type;
5363         if (prog->caps->name)
5364                 load_attr.name = prog->name;
5365         load_attr.insns = insns;
5366         load_attr.insns_cnt = insns_cnt;
5367         load_attr.license = license;
5368         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
5369             prog->type == BPF_PROG_TYPE_LSM) {
5370                 load_attr.attach_btf_id = prog->attach_btf_id;
5371         } else if (prog->type == BPF_PROG_TYPE_TRACING ||
5372                    prog->type == BPF_PROG_TYPE_EXT) {
5373                 load_attr.attach_prog_fd = prog->attach_prog_fd;
5374                 load_attr.attach_btf_id = prog->attach_btf_id;
5375         } else {
5376                 load_attr.kern_version = kern_version;
5377                 load_attr.prog_ifindex = prog->prog_ifindex;
5378         }
5379         /* specify func_info/line_info only if kernel supports them */
5380         btf_fd = bpf_object__btf_fd(prog->obj);
5381         if (btf_fd >= 0 && prog->obj->caps.btf_func) {
5382                 load_attr.prog_btf_fd = btf_fd;
5383                 load_attr.func_info = prog->func_info;
5384                 load_attr.func_info_rec_size = prog->func_info_rec_size;
5385                 load_attr.func_info_cnt = prog->func_info_cnt;
5386                 load_attr.line_info = prog->line_info;
5387                 load_attr.line_info_rec_size = prog->line_info_rec_size;
5388                 load_attr.line_info_cnt = prog->line_info_cnt;
5389         }
5390         load_attr.log_level = prog->log_level;
5391         load_attr.prog_flags = prog->prog_flags;
5392
5393 retry_load:
5394         if (log_buf_size) {
5395                 log_buf = malloc(log_buf_size);
5396                 if (!log_buf)
5397                         return -ENOMEM;
5398
5399                 *log_buf = 0;
5400         }
5401
5402         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
5403
5404         if (ret >= 0) {
5405                 if (log_buf && load_attr.log_level)
5406                         pr_debug("verifier log:\n%s", log_buf);
5407                 *pfd = ret;
5408                 ret = 0;
5409                 goto out;
5410         }
5411
5412         if (!log_buf || errno == ENOSPC) {
5413                 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE,
5414                                    log_buf_size << 1);
5415
5416                 free(log_buf);
5417                 goto retry_load;
5418         }
5419         ret = -errno;
5420         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5421         pr_warn("load bpf program failed: %s\n", cp);
5422         pr_perm_msg(ret);
5423
5424         if (log_buf && log_buf[0] != '\0') {
5425                 ret = -LIBBPF_ERRNO__VERIFY;
5426                 pr_warn("-- BEGIN DUMP LOG ---\n");
5427                 pr_warn("\n%s\n", log_buf);
5428                 pr_warn("-- END LOG --\n");
5429         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
5430                 pr_warn("Program too large (%zu insns), at most %d insns\n",
5431                         load_attr.insns_cnt, BPF_MAXINSNS);
5432                 ret = -LIBBPF_ERRNO__PROG2BIG;
5433         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
5434                 /* Wrong program type? */
5435                 int fd;
5436
5437                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
5438                 load_attr.expected_attach_type = 0;
5439                 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
5440                 if (fd >= 0) {
5441                         close(fd);
5442                         ret = -LIBBPF_ERRNO__PROGTYPE;
5443                         goto out;
5444                 }
5445         }
5446
5447 out:
5448         free(log_buf);
5449         return ret;
5450 }
5451
5452 static int libbpf_find_attach_btf_id(struct bpf_program *prog);
5453
5454 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
5455 {
5456         int err = 0, fd, i, btf_id;
5457
5458         if (prog->obj->loaded) {
5459                 pr_warn("prog '%s'('%s'): can't load after object was loaded\n",
5460                         prog->name, prog->section_name);
5461                 return -EINVAL;
5462         }
5463
5464         if ((prog->type == BPF_PROG_TYPE_TRACING ||
5465              prog->type == BPF_PROG_TYPE_LSM ||
5466              prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
5467                 btf_id = libbpf_find_attach_btf_id(prog);
5468                 if (btf_id <= 0)
5469                         return btf_id;
5470                 prog->attach_btf_id = btf_id;
5471         }
5472
5473         if (prog->instances.nr < 0 || !prog->instances.fds) {
5474                 if (prog->preprocessor) {
5475                         pr_warn("Internal error: can't load program '%s'\n",
5476                                 prog->section_name);
5477                         return -LIBBPF_ERRNO__INTERNAL;
5478                 }
5479
5480                 prog->instances.fds = malloc(sizeof(int));
5481                 if (!prog->instances.fds) {
5482                         pr_warn("Not enough memory for BPF fds\n");
5483                         return -ENOMEM;
5484                 }
5485                 prog->instances.nr = 1;
5486                 prog->instances.fds[0] = -1;
5487         }
5488
5489         if (!prog->preprocessor) {
5490                 if (prog->instances.nr != 1) {
5491                         pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
5492                                 prog->section_name, prog->instances.nr);
5493                 }
5494                 err = load_program(prog, prog->insns, prog->insns_cnt,
5495                                    license, kern_ver, &fd);
5496                 if (!err)
5497                         prog->instances.fds[0] = fd;
5498                 goto out;
5499         }
5500
5501         for (i = 0; i < prog->instances.nr; i++) {
5502                 struct bpf_prog_prep_result result;
5503                 bpf_program_prep_t preprocessor = prog->preprocessor;
5504
5505                 memset(&result, 0, sizeof(result));
5506                 err = preprocessor(prog, i, prog->insns,
5507                                    prog->insns_cnt, &result);
5508                 if (err) {
5509                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
5510                                 i, prog->section_name);
5511                         goto out;
5512                 }
5513
5514                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
5515                         pr_debug("Skip loading the %dth instance of program '%s'\n",
5516                                  i, prog->section_name);
5517                         prog->instances.fds[i] = -1;
5518                         if (result.pfd)
5519                                 *result.pfd = -1;
5520                         continue;
5521                 }
5522
5523                 err = load_program(prog, result.new_insn_ptr,
5524                                    result.new_insn_cnt, license, kern_ver, &fd);
5525                 if (err) {
5526                         pr_warn("Loading the %dth instance of program '%s' failed\n",
5527                                 i, prog->section_name);
5528                         goto out;
5529                 }
5530
5531                 if (result.pfd)
5532                         *result.pfd = fd;
5533                 prog->instances.fds[i] = fd;
5534         }
5535 out:
5536         if (err)
5537                 pr_warn("failed to load program '%s'\n", prog->section_name);
5538         zfree(&prog->insns);
5539         prog->insns_cnt = 0;
5540         return err;
5541 }
5542
5543 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
5544                                              const struct bpf_object *obj)
5545 {
5546         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
5547 }
5548
5549 static int
5550 bpf_object__load_progs(struct bpf_object *obj, int log_level)
5551 {
5552         struct bpf_program *prog;
5553         size_t i;
5554         int err;
5555
5556         for (i = 0; i < obj->nr_programs; i++) {
5557                 prog = &obj->programs[i];
5558                 if (bpf_program__is_function_storage(prog, obj))
5559                         continue;
5560                 if (!prog->load) {
5561                         pr_debug("prog '%s'('%s'): skipped loading\n",
5562                                  prog->name, prog->section_name);
5563                         continue;
5564                 }
5565                 prog->log_level |= log_level;
5566                 err = bpf_program__load(prog, obj->license, obj->kern_version);
5567                 if (err)
5568                         return err;
5569         }
5570         return 0;
5571 }
5572
5573 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
5574
5575 static struct bpf_object *
5576 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
5577                    const struct bpf_object_open_opts *opts)
5578 {
5579         const char *obj_name, *kconfig;
5580         struct bpf_program *prog;
5581         struct bpf_object *obj;
5582         char tmp_name[64];
5583         int err;
5584
5585         if (elf_version(EV_CURRENT) == EV_NONE) {
5586                 pr_warn("failed to init libelf for %s\n",
5587                         path ? : "(mem buf)");
5588                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
5589         }
5590
5591         if (!OPTS_VALID(opts, bpf_object_open_opts))
5592                 return ERR_PTR(-EINVAL);
5593
5594         obj_name = OPTS_GET(opts, object_name, NULL);
5595         if (obj_buf) {
5596                 if (!obj_name) {
5597                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
5598                                  (unsigned long)obj_buf,
5599                                  (unsigned long)obj_buf_sz);
5600                         obj_name = tmp_name;
5601                 }
5602                 path = obj_name;
5603                 pr_debug("loading object '%s' from buffer\n", obj_name);
5604         }
5605
5606         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
5607         if (IS_ERR(obj))
5608                 return obj;
5609
5610         kconfig = OPTS_GET(opts, kconfig, NULL);
5611         if (kconfig) {
5612                 obj->kconfig = strdup(kconfig);
5613                 if (!obj->kconfig)
5614                         return ERR_PTR(-ENOMEM);
5615         }
5616
5617         err = bpf_object__elf_init(obj);
5618         err = err ? : bpf_object__check_endianness(obj);
5619         err = err ? : bpf_object__elf_collect(obj);
5620         err = err ? : bpf_object__collect_externs(obj);
5621         err = err ? : bpf_object__finalize_btf(obj);
5622         err = err ? : bpf_object__init_maps(obj, opts);
5623         err = err ? : bpf_object__init_prog_names(obj);
5624         err = err ? : bpf_object__collect_reloc(obj);
5625         if (err)
5626                 goto out;
5627         bpf_object__elf_finish(obj);
5628
5629         bpf_object__for_each_program(prog, obj) {
5630                 prog->sec_def = find_sec_def(prog->section_name);
5631                 if (!prog->sec_def)
5632                         /* couldn't guess, but user might manually specify */
5633                         continue;
5634
5635                 bpf_program__set_type(prog, prog->sec_def->prog_type);
5636                 bpf_program__set_expected_attach_type(prog,
5637                                 prog->sec_def->expected_attach_type);
5638
5639                 if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
5640                     prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
5641                         prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
5642         }
5643
5644         return obj;
5645 out:
5646         bpf_object__close(obj);
5647         return ERR_PTR(err);
5648 }
5649
5650 static struct bpf_object *
5651 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
5652 {
5653         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5654                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
5655         );
5656
5657         /* param validation */
5658         if (!attr->file)
5659                 return NULL;
5660
5661         pr_debug("loading %s\n", attr->file);
5662         return __bpf_object__open(attr->file, NULL, 0, &opts);
5663 }
5664
5665 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
5666 {
5667         return __bpf_object__open_xattr(attr, 0);
5668 }
5669
5670 struct bpf_object *bpf_object__open(const char *path)
5671 {
5672         struct bpf_object_open_attr attr = {
5673                 .file           = path,
5674                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
5675         };
5676
5677         return bpf_object__open_xattr(&attr);
5678 }
5679
5680 struct bpf_object *
5681 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
5682 {
5683         if (!path)
5684                 return ERR_PTR(-EINVAL);
5685
5686         pr_debug("loading %s\n", path);
5687
5688         return __bpf_object__open(path, NULL, 0, opts);
5689 }
5690
5691 struct bpf_object *
5692 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
5693                      const struct bpf_object_open_opts *opts)
5694 {
5695         if (!obj_buf || obj_buf_sz == 0)
5696                 return ERR_PTR(-EINVAL);
5697
5698         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
5699 }
5700
5701 struct bpf_object *
5702 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
5703                         const char *name)
5704 {
5705         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
5706                 .object_name = name,
5707                 /* wrong default, but backwards-compatible */
5708                 .relaxed_maps = true,
5709         );
5710
5711         /* returning NULL is wrong, but backwards-compatible */
5712         if (!obj_buf || obj_buf_sz == 0)
5713                 return NULL;
5714
5715         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
5716 }
5717
5718 int bpf_object__unload(struct bpf_object *obj)
5719 {
5720         size_t i;
5721
5722         if (!obj)
5723                 return -EINVAL;
5724
5725         for (i = 0; i < obj->nr_maps; i++) {
5726                 zclose(obj->maps[i].fd);
5727                 if (obj->maps[i].st_ops)
5728                         zfree(&obj->maps[i].st_ops->kern_vdata);
5729         }
5730
5731         for (i = 0; i < obj->nr_programs; i++)
5732                 bpf_program__unload(&obj->programs[i]);
5733
5734         return 0;
5735 }
5736
5737 static int bpf_object__sanitize_maps(struct bpf_object *obj)
5738 {
5739         struct bpf_map *m;
5740
5741         bpf_object__for_each_map(m, obj) {
5742                 if (!bpf_map__is_internal(m))
5743                         continue;
5744                 if (!obj->caps.global_data) {
5745                         pr_warn("kernel doesn't support global data\n");
5746                         return -ENOTSUP;
5747                 }
5748                 if (!obj->caps.array_mmap)
5749                         m->def.map_flags ^= BPF_F_MMAPABLE;
5750         }
5751
5752         return 0;
5753 }
5754
5755 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
5756 {
5757         char sym_type, sym_name[500];
5758         unsigned long long sym_addr;
5759         struct extern_desc *ext;
5760         int ret, err = 0;
5761         FILE *f;
5762
5763         f = fopen("/proc/kallsyms", "r");
5764         if (!f) {
5765                 err = -errno;
5766                 pr_warn("failed to open /proc/kallsyms: %d\n", err);
5767                 return err;
5768         }
5769
5770         while (true) {
5771                 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
5772                              &sym_addr, &sym_type, sym_name);
5773                 if (ret == EOF && feof(f))
5774                         break;
5775                 if (ret != 3) {
5776                         pr_warn("failed to read kallsyms entry: %d\n", ret);
5777                         err = -EINVAL;
5778                         goto out;
5779                 }
5780
5781                 ext = find_extern_by_name(obj, sym_name);
5782                 if (!ext || ext->type != EXT_KSYM)
5783                         continue;
5784
5785                 if (ext->is_set && ext->ksym.addr != sym_addr) {
5786                         pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
5787                                 sym_name, ext->ksym.addr, sym_addr);
5788                         err = -EINVAL;
5789                         goto out;
5790                 }
5791                 if (!ext->is_set) {
5792                         ext->is_set = true;
5793                         ext->ksym.addr = sym_addr;
5794                         pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
5795                 }
5796         }
5797
5798 out:
5799         fclose(f);
5800         return err;
5801 }
5802
5803 static int bpf_object__resolve_externs(struct bpf_object *obj,
5804                                        const char *extra_kconfig)
5805 {
5806         bool need_config = false, need_kallsyms = false;
5807         struct extern_desc *ext;
5808         void *kcfg_data = NULL;
5809         int err, i;
5810
5811         if (obj->nr_extern == 0)
5812                 return 0;
5813
5814         if (obj->kconfig_map_idx >= 0)
5815                 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
5816
5817         for (i = 0; i < obj->nr_extern; i++) {
5818                 ext = &obj->externs[i];
5819
5820                 if (ext->type == EXT_KCFG &&
5821                     strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
5822                         void *ext_val = kcfg_data + ext->kcfg.data_off;
5823                         __u32 kver = get_kernel_version();
5824
5825                         if (!kver) {
5826                                 pr_warn("failed to get kernel version\n");
5827                                 return -EINVAL;
5828                         }
5829                         err = set_kcfg_value_num(ext, ext_val, kver);
5830                         if (err)
5831                                 return err;
5832                         pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
5833                 } else if (ext->type == EXT_KCFG &&
5834                            strncmp(ext->name, "CONFIG_", 7) == 0) {
5835                         need_config = true;
5836                 } else if (ext->type == EXT_KSYM) {
5837                         need_kallsyms = true;
5838                 } else {
5839                         pr_warn("unrecognized extern '%s'\n", ext->name);
5840                         return -EINVAL;
5841                 }
5842         }
5843         if (need_config && extra_kconfig) {
5844                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
5845                 if (err)
5846                         return -EINVAL;
5847                 need_config = false;
5848                 for (i = 0; i < obj->nr_extern; i++) {
5849                         ext = &obj->externs[i];
5850                         if (ext->type == EXT_KCFG && !ext->is_set) {
5851                                 need_config = true;
5852                                 break;
5853                         }
5854                 }
5855         }
5856         if (need_config) {
5857                 err = bpf_object__read_kconfig_file(obj, kcfg_data);
5858                 if (err)
5859                         return -EINVAL;
5860         }
5861         if (need_kallsyms) {
5862                 err = bpf_object__read_kallsyms_file(obj);
5863                 if (err)
5864                         return -EINVAL;
5865         }
5866         for (i = 0; i < obj->nr_extern; i++) {
5867                 ext = &obj->externs[i];
5868
5869                 if (!ext->is_set && !ext->is_weak) {
5870                         pr_warn("extern %s (strong) not resolved\n", ext->name);
5871                         return -ESRCH;
5872                 } else if (!ext->is_set) {
5873                         pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
5874                                  ext->name);
5875                 }
5876         }
5877
5878         return 0;
5879 }
5880
5881 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
5882 {
5883         struct bpf_object *obj;
5884         int err, i;
5885
5886         if (!attr)
5887                 return -EINVAL;
5888         obj = attr->obj;
5889         if (!obj)
5890                 return -EINVAL;
5891
5892         if (obj->loaded) {
5893                 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
5894                 return -EINVAL;
5895         }
5896
5897         err = bpf_object__probe_loading(obj);
5898         err = err ? : bpf_object__probe_caps(obj);
5899         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
5900         err = err ? : bpf_object__sanitize_and_load_btf(obj);
5901         err = err ? : bpf_object__sanitize_maps(obj);
5902         err = err ? : bpf_object__load_vmlinux_btf(obj);
5903         err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
5904         err = err ? : bpf_object__create_maps(obj);
5905         err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
5906         err = err ? : bpf_object__load_progs(obj, attr->log_level);
5907
5908         btf__free(obj->btf_vmlinux);
5909         obj->btf_vmlinux = NULL;
5910
5911         obj->loaded = true; /* doesn't matter if successfully or not */
5912
5913         if (err)
5914                 goto out;
5915
5916         return 0;
5917 out:
5918         /* unpin any maps that were auto-pinned during load */
5919         for (i = 0; i < obj->nr_maps; i++)
5920                 if (obj->maps[i].pinned && !obj->maps[i].reused)
5921                         bpf_map__unpin(&obj->maps[i], NULL);
5922
5923         bpf_object__unload(obj);
5924         pr_warn("failed to load object '%s'\n", obj->path);
5925         return err;
5926 }
5927
5928 int bpf_object__load(struct bpf_object *obj)
5929 {
5930         struct bpf_object_load_attr attr = {
5931                 .obj = obj,
5932         };
5933
5934         return bpf_object__load_xattr(&attr);
5935 }
5936
5937 static int make_parent_dir(const char *path)
5938 {
5939         char *cp, errmsg[STRERR_BUFSIZE];
5940         char *dname, *dir;
5941         int err = 0;
5942
5943         dname = strdup(path);
5944         if (dname == NULL)
5945                 return -ENOMEM;
5946
5947         dir = dirname(dname);
5948         if (mkdir(dir, 0700) && errno != EEXIST)
5949                 err = -errno;
5950
5951         free(dname);
5952         if (err) {
5953                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5954                 pr_warn("failed to mkdir %s: %s\n", path, cp);
5955         }
5956         return err;
5957 }
5958
5959 static int check_path(const char *path)
5960 {
5961         char *cp, errmsg[STRERR_BUFSIZE];
5962         struct statfs st_fs;
5963         char *dname, *dir;
5964         int err = 0;
5965
5966         if (path == NULL)
5967                 return -EINVAL;
5968
5969         dname = strdup(path);
5970         if (dname == NULL)
5971                 return -ENOMEM;
5972
5973         dir = dirname(dname);
5974         if (statfs(dir, &st_fs)) {
5975                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5976                 pr_warn("failed to statfs %s: %s\n", dir, cp);
5977                 err = -errno;
5978         }
5979         free(dname);
5980
5981         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
5982                 pr_warn("specified path %s is not on BPF FS\n", path);
5983                 err = -EINVAL;
5984         }
5985
5986         return err;
5987 }
5988
5989 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
5990                               int instance)
5991 {
5992         char *cp, errmsg[STRERR_BUFSIZE];
5993         int err;
5994
5995         err = make_parent_dir(path);
5996         if (err)
5997                 return err;
5998
5999         err = check_path(path);
6000         if (err)
6001                 return err;
6002
6003         if (prog == NULL) {
6004                 pr_warn("invalid program pointer\n");
6005                 return -EINVAL;
6006         }
6007
6008         if (instance < 0 || instance >= prog->instances.nr) {
6009                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
6010                         instance, prog->section_name, prog->instances.nr);
6011                 return -EINVAL;
6012         }
6013
6014         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
6015                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6016                 pr_warn("failed to pin program: %s\n", cp);
6017                 return -errno;
6018         }
6019         pr_debug("pinned program '%s'\n", path);
6020
6021         return 0;
6022 }
6023
6024 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
6025                                 int instance)
6026 {
6027         int err;
6028
6029         err = check_path(path);
6030         if (err)
6031                 return err;
6032
6033         if (prog == NULL) {
6034                 pr_warn("invalid program pointer\n");
6035                 return -EINVAL;
6036         }
6037
6038         if (instance < 0 || instance >= prog->instances.nr) {
6039                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
6040                         instance, prog->section_name, prog->instances.nr);
6041                 return -EINVAL;
6042         }
6043
6044         err = unlink(path);
6045         if (err != 0)
6046                 return -errno;
6047         pr_debug("unpinned program '%s'\n", path);
6048
6049         return 0;
6050 }
6051
6052 int bpf_program__pin(struct bpf_program *prog, const char *path)
6053 {
6054         int i, err;
6055
6056         err = make_parent_dir(path);
6057         if (err)
6058                 return err;
6059
6060         err = check_path(path);
6061         if (err)
6062                 return err;
6063
6064         if (prog == NULL) {
6065                 pr_warn("invalid program pointer\n");
6066                 return -EINVAL;
6067         }
6068
6069         if (prog->instances.nr <= 0) {
6070                 pr_warn("no instances of prog %s to pin\n",
6071                            prog->section_name);
6072                 return -EINVAL;
6073         }
6074
6075         if (prog->instances.nr == 1) {
6076                 /* don't create subdirs when pinning single instance */
6077                 return bpf_program__pin_instance(prog, path, 0);
6078         }
6079
6080         for (i = 0; i < prog->instances.nr; i++) {
6081                 char buf[PATH_MAX];
6082                 int len;
6083
6084                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6085                 if (len < 0) {
6086                         err = -EINVAL;
6087                         goto err_unpin;
6088                 } else if (len >= PATH_MAX) {
6089                         err = -ENAMETOOLONG;
6090                         goto err_unpin;
6091                 }
6092
6093                 err = bpf_program__pin_instance(prog, buf, i);
6094                 if (err)
6095                         goto err_unpin;
6096         }
6097
6098         return 0;
6099
6100 err_unpin:
6101         for (i = i - 1; i >= 0; i--) {
6102                 char buf[PATH_MAX];
6103                 int len;
6104
6105                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6106                 if (len < 0)
6107                         continue;
6108                 else if (len >= PATH_MAX)
6109                         continue;
6110
6111                 bpf_program__unpin_instance(prog, buf, i);
6112         }
6113
6114         rmdir(path);
6115
6116         return err;
6117 }
6118
6119 int bpf_program__unpin(struct bpf_program *prog, const char *path)
6120 {
6121         int i, err;
6122
6123         err = check_path(path);
6124         if (err)
6125                 return err;
6126
6127         if (prog == NULL) {
6128                 pr_warn("invalid program pointer\n");
6129                 return -EINVAL;
6130         }
6131
6132         if (prog->instances.nr <= 0) {
6133                 pr_warn("no instances of prog %s to pin\n",
6134                            prog->section_name);
6135                 return -EINVAL;
6136         }
6137
6138         if (prog->instances.nr == 1) {
6139                 /* don't create subdirs when pinning single instance */
6140                 return bpf_program__unpin_instance(prog, path, 0);
6141         }
6142
6143         for (i = 0; i < prog->instances.nr; i++) {
6144                 char buf[PATH_MAX];
6145                 int len;
6146
6147                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
6148                 if (len < 0)
6149                         return -EINVAL;
6150                 else if (len >= PATH_MAX)
6151                         return -ENAMETOOLONG;
6152
6153                 err = bpf_program__unpin_instance(prog, buf, i);
6154                 if (err)
6155                         return err;
6156         }
6157
6158         err = rmdir(path);
6159         if (err)
6160                 return -errno;
6161
6162         return 0;
6163 }
6164
6165 int bpf_map__pin(struct bpf_map *map, const char *path)
6166 {
6167         char *cp, errmsg[STRERR_BUFSIZE];
6168         int err;
6169
6170         if (map == NULL) {
6171                 pr_warn("invalid map pointer\n");
6172                 return -EINVAL;
6173         }
6174
6175         if (map->pin_path) {
6176                 if (path && strcmp(path, map->pin_path)) {
6177                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
6178                                 bpf_map__name(map), map->pin_path, path);
6179                         return -EINVAL;
6180                 } else if (map->pinned) {
6181                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
6182                                  bpf_map__name(map), map->pin_path);
6183                         return 0;
6184                 }
6185         } else {
6186                 if (!path) {
6187                         pr_warn("missing a path to pin map '%s' at\n",
6188                                 bpf_map__name(map));
6189                         return -EINVAL;
6190                 } else if (map->pinned) {
6191                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
6192                         return -EEXIST;
6193                 }
6194
6195                 map->pin_path = strdup(path);
6196                 if (!map->pin_path) {
6197                         err = -errno;
6198                         goto out_err;
6199                 }
6200         }
6201
6202         err = make_parent_dir(map->pin_path);
6203         if (err)
6204                 return err;
6205
6206         err = check_path(map->pin_path);
6207         if (err)
6208                 return err;
6209
6210         if (bpf_obj_pin(map->fd, map->pin_path)) {
6211                 err = -errno;
6212                 goto out_err;
6213         }
6214
6215         map->pinned = true;
6216         pr_debug("pinned map '%s'\n", map->pin_path);
6217
6218         return 0;
6219
6220 out_err:
6221         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
6222         pr_warn("failed to pin map: %s\n", cp);
6223         return err;
6224 }
6225
6226 int bpf_map__unpin(struct bpf_map *map, const char *path)
6227 {
6228         int err;
6229
6230         if (map == NULL) {
6231                 pr_warn("invalid map pointer\n");
6232                 return -EINVAL;
6233         }
6234
6235         if (map->pin_path) {
6236                 if (path && strcmp(path, map->pin_path)) {
6237                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
6238                                 bpf_map__name(map), map->pin_path, path);
6239                         return -EINVAL;
6240                 }
6241                 path = map->pin_path;
6242         } else if (!path) {
6243                 pr_warn("no path to unpin map '%s' from\n",
6244                         bpf_map__name(map));
6245                 return -EINVAL;
6246         }
6247
6248         err = check_path(path);
6249         if (err)
6250                 return err;
6251
6252         err = unlink(path);
6253         if (err != 0)
6254                 return -errno;
6255
6256         map->pinned = false;
6257         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
6258
6259         return 0;
6260 }
6261
6262 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
6263 {
6264         char *new = NULL;
6265
6266         if (path) {
6267                 new = strdup(path);
6268                 if (!new)
6269                         return -errno;
6270         }
6271
6272         free(map->pin_path);
6273         map->pin_path = new;
6274         return 0;
6275 }
6276
6277 const char *bpf_map__get_pin_path(const struct bpf_map *map)
6278 {
6279         return map->pin_path;
6280 }
6281
6282 bool bpf_map__is_pinned(const struct bpf_map *map)
6283 {
6284         return map->pinned;
6285 }
6286
6287 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
6288 {
6289         struct bpf_map *map;
6290         int err;
6291
6292         if (!obj)
6293                 return -ENOENT;
6294
6295         if (!obj->loaded) {
6296                 pr_warn("object not yet loaded; load it first\n");
6297                 return -ENOENT;
6298         }
6299
6300         bpf_object__for_each_map(map, obj) {
6301                 char *pin_path = NULL;
6302                 char buf[PATH_MAX];
6303
6304                 if (path) {
6305                         int len;
6306
6307                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
6308                                        bpf_map__name(map));
6309                         if (len < 0) {
6310                                 err = -EINVAL;
6311                                 goto err_unpin_maps;
6312                         } else if (len >= PATH_MAX) {
6313                                 err = -ENAMETOOLONG;
6314                                 goto err_unpin_maps;
6315                         }
6316                         pin_path = buf;
6317                 } else if (!map->pin_path) {
6318                         continue;
6319                 }
6320
6321                 err = bpf_map__pin(map, pin_path);
6322                 if (err)
6323                         goto err_unpin_maps;
6324         }
6325
6326         return 0;
6327
6328 err_unpin_maps:
6329         while ((map = bpf_map__prev(map, obj))) {
6330                 if (!map->pin_path)
6331                         continue;
6332
6333                 bpf_map__unpin(map, NULL);
6334         }
6335
6336         return err;
6337 }
6338
6339 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
6340 {
6341         struct bpf_map *map;
6342         int err;
6343
6344         if (!obj)
6345                 return -ENOENT;
6346
6347         bpf_object__for_each_map(map, obj) {
6348                 char *pin_path = NULL;
6349                 char buf[PATH_MAX];
6350
6351                 if (path) {
6352                         int len;
6353
6354                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
6355                                        bpf_map__name(map));
6356                         if (len < 0)
6357                                 return -EINVAL;
6358                         else if (len >= PATH_MAX)
6359                                 return -ENAMETOOLONG;
6360                         pin_path = buf;
6361                 } else if (!map->pin_path) {
6362                         continue;
6363                 }
6364
6365                 err = bpf_map__unpin(map, pin_path);
6366                 if (err)
6367                         return err;
6368         }
6369
6370         return 0;
6371 }
6372
6373 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
6374 {
6375         struct bpf_program *prog;
6376         int err;
6377
6378         if (!obj)
6379                 return -ENOENT;
6380
6381         if (!obj->loaded) {
6382                 pr_warn("object not yet loaded; load it first\n");
6383                 return -ENOENT;
6384         }
6385
6386         bpf_object__for_each_program(prog, obj) {
6387                 char buf[PATH_MAX];
6388                 int len;
6389
6390                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6391                                prog->pin_name);
6392                 if (len < 0) {
6393                         err = -EINVAL;
6394                         goto err_unpin_programs;
6395                 } else if (len >= PATH_MAX) {
6396                         err = -ENAMETOOLONG;
6397                         goto err_unpin_programs;
6398                 }
6399
6400                 err = bpf_program__pin(prog, buf);
6401                 if (err)
6402                         goto err_unpin_programs;
6403         }
6404
6405         return 0;
6406
6407 err_unpin_programs:
6408         while ((prog = bpf_program__prev(prog, obj))) {
6409                 char buf[PATH_MAX];
6410                 int len;
6411
6412                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6413                                prog->pin_name);
6414                 if (len < 0)
6415                         continue;
6416                 else if (len >= PATH_MAX)
6417                         continue;
6418
6419                 bpf_program__unpin(prog, buf);
6420         }
6421
6422         return err;
6423 }
6424
6425 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
6426 {
6427         struct bpf_program *prog;
6428         int err;
6429
6430         if (!obj)
6431                 return -ENOENT;
6432
6433         bpf_object__for_each_program(prog, obj) {
6434                 char buf[PATH_MAX];
6435                 int len;
6436
6437                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
6438                                prog->pin_name);
6439                 if (len < 0)
6440                         return -EINVAL;
6441                 else if (len >= PATH_MAX)
6442                         return -ENAMETOOLONG;
6443
6444                 err = bpf_program__unpin(prog, buf);
6445                 if (err)
6446                         return err;
6447         }
6448
6449         return 0;
6450 }
6451
6452 int bpf_object__pin(struct bpf_object *obj, const char *path)
6453 {
6454         int err;
6455
6456         err = bpf_object__pin_maps(obj, path);
6457         if (err)
6458                 return err;
6459
6460         err = bpf_object__pin_programs(obj, path);
6461         if (err) {
6462                 bpf_object__unpin_maps(obj, path);
6463                 return err;
6464         }
6465
6466         return 0;
6467 }
6468
6469 static void bpf_map__destroy(struct bpf_map *map)
6470 {
6471         if (map->clear_priv)
6472                 map->clear_priv(map, map->priv);
6473         map->priv = NULL;
6474         map->clear_priv = NULL;
6475
6476         if (map->inner_map) {
6477                 bpf_map__destroy(map->inner_map);
6478                 zfree(&map->inner_map);
6479         }
6480
6481         zfree(&map->init_slots);
6482         map->init_slots_sz = 0;
6483
6484         if (map->mmaped) {
6485                 munmap(map->mmaped, bpf_map_mmap_sz(map));
6486                 map->mmaped = NULL;
6487         }
6488
6489         if (map->st_ops) {
6490                 zfree(&map->st_ops->data);
6491                 zfree(&map->st_ops->progs);
6492                 zfree(&map->st_ops->kern_func_off);
6493                 zfree(&map->st_ops);
6494         }
6495
6496         zfree(&map->name);
6497         zfree(&map->pin_path);
6498
6499         if (map->fd >= 0)
6500                 zclose(map->fd);
6501 }
6502
6503 void bpf_object__close(struct bpf_object *obj)
6504 {
6505         size_t i;
6506
6507         if (!obj)
6508                 return;
6509
6510         if (obj->clear_priv)
6511                 obj->clear_priv(obj, obj->priv);
6512
6513         bpf_object__elf_finish(obj);
6514         bpf_object__unload(obj);
6515         btf__free(obj->btf);
6516         btf_ext__free(obj->btf_ext);
6517
6518         for (i = 0; i < obj->nr_maps; i++)
6519                 bpf_map__destroy(&obj->maps[i]);
6520
6521         zfree(&obj->kconfig);
6522         zfree(&obj->externs);
6523         obj->nr_extern = 0;
6524
6525         zfree(&obj->maps);
6526         obj->nr_maps = 0;
6527
6528         if (obj->programs && obj->nr_programs) {
6529                 for (i = 0; i < obj->nr_programs; i++)
6530                         bpf_program__exit(&obj->programs[i]);
6531         }
6532         zfree(&obj->programs);
6533
6534         list_del(&obj->list);
6535         free(obj);
6536 }
6537
6538 struct bpf_object *
6539 bpf_object__next(struct bpf_object *prev)
6540 {
6541         struct bpf_object *next;
6542
6543         if (!prev)
6544                 next = list_first_entry(&bpf_objects_list,
6545                                         struct bpf_object,
6546                                         list);
6547         else
6548                 next = list_next_entry(prev, list);
6549
6550         /* Empty list is noticed here so don't need checking on entry. */
6551         if (&next->list == &bpf_objects_list)
6552                 return NULL;
6553
6554         return next;
6555 }
6556
6557 const char *bpf_object__name(const struct bpf_object *obj)
6558 {
6559         return obj ? obj->name : ERR_PTR(-EINVAL);
6560 }
6561
6562 unsigned int bpf_object__kversion(const struct bpf_object *obj)
6563 {
6564         return obj ? obj->kern_version : 0;
6565 }
6566
6567 struct btf *bpf_object__btf(const struct bpf_object *obj)
6568 {
6569         return obj ? obj->btf : NULL;
6570 }
6571
6572 int bpf_object__btf_fd(const struct bpf_object *obj)
6573 {
6574         return obj->btf ? btf__fd(obj->btf) : -1;
6575 }
6576
6577 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
6578                          bpf_object_clear_priv_t clear_priv)
6579 {
6580         if (obj->priv && obj->clear_priv)
6581                 obj->clear_priv(obj, obj->priv);
6582
6583         obj->priv = priv;
6584         obj->clear_priv = clear_priv;
6585         return 0;
6586 }
6587
6588 void *bpf_object__priv(const struct bpf_object *obj)
6589 {
6590         return obj ? obj->priv : ERR_PTR(-EINVAL);
6591 }
6592
6593 static struct bpf_program *
6594 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
6595                     bool forward)
6596 {
6597         size_t nr_programs = obj->nr_programs;
6598         ssize_t idx;
6599
6600         if (!nr_programs)
6601                 return NULL;
6602
6603         if (!p)
6604                 /* Iter from the beginning */
6605                 return forward ? &obj->programs[0] :
6606                         &obj->programs[nr_programs - 1];
6607
6608         if (p->obj != obj) {
6609                 pr_warn("error: program handler doesn't match object\n");
6610                 return NULL;
6611         }
6612
6613         idx = (p - obj->programs) + (forward ? 1 : -1);
6614         if (idx >= obj->nr_programs || idx < 0)
6615                 return NULL;
6616         return &obj->programs[idx];
6617 }
6618
6619 struct bpf_program *
6620 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
6621 {
6622         struct bpf_program *prog = prev;
6623
6624         do {
6625                 prog = __bpf_program__iter(prog, obj, true);
6626         } while (prog && bpf_program__is_function_storage(prog, obj));
6627
6628         return prog;
6629 }
6630
6631 struct bpf_program *
6632 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
6633 {
6634         struct bpf_program *prog = next;
6635
6636         do {
6637                 prog = __bpf_program__iter(prog, obj, false);
6638         } while (prog && bpf_program__is_function_storage(prog, obj));
6639
6640         return prog;
6641 }
6642
6643 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
6644                           bpf_program_clear_priv_t clear_priv)
6645 {
6646         if (prog->priv && prog->clear_priv)
6647                 prog->clear_priv(prog, prog->priv);
6648
6649         prog->priv = priv;
6650         prog->clear_priv = clear_priv;
6651         return 0;
6652 }
6653
6654 void *bpf_program__priv(const struct bpf_program *prog)
6655 {
6656         return prog ? prog->priv : ERR_PTR(-EINVAL);
6657 }
6658
6659 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
6660 {
6661         prog->prog_ifindex = ifindex;
6662 }
6663
6664 const char *bpf_program__name(const struct bpf_program *prog)
6665 {
6666         return prog->name;
6667 }
6668
6669 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
6670 {
6671         const char *title;
6672
6673         title = prog->section_name;
6674         if (needs_copy) {
6675                 title = strdup(title);
6676                 if (!title) {
6677                         pr_warn("failed to strdup program title\n");
6678                         return ERR_PTR(-ENOMEM);
6679                 }
6680         }
6681
6682         return title;
6683 }
6684
6685 bool bpf_program__autoload(const struct bpf_program *prog)
6686 {
6687         return prog->load;
6688 }
6689
6690 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
6691 {
6692         if (prog->obj->loaded)
6693                 return -EINVAL;
6694
6695         prog->load = autoload;
6696         return 0;
6697 }
6698
6699 int bpf_program__fd(const struct bpf_program *prog)
6700 {
6701         return bpf_program__nth_fd(prog, 0);
6702 }
6703
6704 size_t bpf_program__size(const struct bpf_program *prog)
6705 {
6706         return prog->insns_cnt * sizeof(struct bpf_insn);
6707 }
6708
6709 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
6710                           bpf_program_prep_t prep)
6711 {
6712         int *instances_fds;
6713
6714         if (nr_instances <= 0 || !prep)
6715                 return -EINVAL;
6716
6717         if (prog->instances.nr > 0 || prog->instances.fds) {
6718                 pr_warn("Can't set pre-processor after loading\n");
6719                 return -EINVAL;
6720         }
6721
6722         instances_fds = malloc(sizeof(int) * nr_instances);
6723         if (!instances_fds) {
6724                 pr_warn("alloc memory failed for fds\n");
6725                 return -ENOMEM;
6726         }
6727
6728         /* fill all fd with -1 */
6729         memset(instances_fds, -1, sizeof(int) * nr_instances);
6730
6731         prog->instances.nr = nr_instances;
6732         prog->instances.fds = instances_fds;
6733         prog->preprocessor = prep;
6734         return 0;
6735 }
6736
6737 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
6738 {
6739         int fd;
6740
6741         if (!prog)
6742                 return -EINVAL;
6743
6744         if (n >= prog->instances.nr || n < 0) {
6745                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
6746                         n, prog->section_name, prog->instances.nr);
6747                 return -EINVAL;
6748         }
6749
6750         fd = prog->instances.fds[n];
6751         if (fd < 0) {
6752                 pr_warn("%dth instance of program '%s' is invalid\n",
6753                         n, prog->section_name);
6754                 return -ENOENT;
6755         }
6756
6757         return fd;
6758 }
6759
6760 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
6761 {
6762         return prog->type;
6763 }
6764
6765 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
6766 {
6767         prog->type = type;
6768 }
6769
6770 static bool bpf_program__is_type(const struct bpf_program *prog,
6771                                  enum bpf_prog_type type)
6772 {
6773         return prog ? (prog->type == type) : false;
6774 }
6775
6776 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
6777 int bpf_program__set_##NAME(struct bpf_program *prog)           \
6778 {                                                               \
6779         if (!prog)                                              \
6780                 return -EINVAL;                                 \
6781         bpf_program__set_type(prog, TYPE);                      \
6782         return 0;                                               \
6783 }                                                               \
6784                                                                 \
6785 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
6786 {                                                               \
6787         return bpf_program__is_type(prog, TYPE);                \
6788 }                                                               \
6789
6790 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
6791 BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
6792 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
6793 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
6794 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
6795 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
6796 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
6797 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
6798 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
6799 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
6800 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
6801 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
6802 BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
6803
6804 enum bpf_attach_type
6805 bpf_program__get_expected_attach_type(struct bpf_program *prog)
6806 {
6807         return prog->expected_attach_type;
6808 }
6809
6810 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
6811                                            enum bpf_attach_type type)
6812 {
6813         prog->expected_attach_type = type;
6814 }
6815
6816 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype_optional,           \
6817                           attachable, attach_btf)                           \
6818         {                                                                   \
6819                 .sec = string,                                              \
6820                 .len = sizeof(string) - 1,                                  \
6821                 .prog_type = ptype,                                         \
6822                 .expected_attach_type = eatype,                             \
6823                 .is_exp_attach_type_optional = eatype_optional,             \
6824                 .is_attachable = attachable,                                \
6825                 .is_attach_btf = attach_btf,                                \
6826         }
6827
6828 /* Programs that can NOT be attached. */
6829 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
6830
6831 /* Programs that can be attached. */
6832 #define BPF_APROG_SEC(string, ptype, atype) \
6833         BPF_PROG_SEC_IMPL(string, ptype, atype, true, 1, 0)
6834
6835 /* Programs that must specify expected attach type at load time. */
6836 #define BPF_EAPROG_SEC(string, ptype, eatype) \
6837         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 1, 0)
6838
6839 /* Programs that use BTF to identify attach point */
6840 #define BPF_PROG_BTF(string, ptype, eatype) \
6841         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 0, 1)
6842
6843 /* Programs that can be attached but attach type can't be identified by section
6844  * name. Kept for backward compatibility.
6845  */
6846 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
6847
6848 #define SEC_DEF(sec_pfx, ptype, ...) {                                      \
6849         .sec = sec_pfx,                                                     \
6850         .len = sizeof(sec_pfx) - 1,                                         \
6851         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
6852         __VA_ARGS__                                                         \
6853 }
6854
6855 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
6856                                       struct bpf_program *prog);
6857 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
6858                                   struct bpf_program *prog);
6859 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
6860                                       struct bpf_program *prog);
6861 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
6862                                      struct bpf_program *prog);
6863 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
6864                                    struct bpf_program *prog);
6865 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
6866                                     struct bpf_program *prog);
6867
6868 static const struct bpf_sec_def section_defs[] = {
6869         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
6870         BPF_PROG_SEC("sk_reuseport",            BPF_PROG_TYPE_SK_REUSEPORT),
6871         SEC_DEF("kprobe/", KPROBE,
6872                 .attach_fn = attach_kprobe),
6873         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
6874         SEC_DEF("kretprobe/", KPROBE,
6875                 .attach_fn = attach_kprobe),
6876         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
6877         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
6878         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
6879         SEC_DEF("tracepoint/", TRACEPOINT,
6880                 .attach_fn = attach_tp),
6881         SEC_DEF("tp/", TRACEPOINT,
6882                 .attach_fn = attach_tp),
6883         SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
6884                 .attach_fn = attach_raw_tp),
6885         SEC_DEF("raw_tp/", RAW_TRACEPOINT,
6886                 .attach_fn = attach_raw_tp),
6887         SEC_DEF("tp_btf/", TRACING,
6888                 .expected_attach_type = BPF_TRACE_RAW_TP,
6889                 .is_attach_btf = true,
6890                 .attach_fn = attach_trace),
6891         SEC_DEF("fentry/", TRACING,
6892                 .expected_attach_type = BPF_TRACE_FENTRY,
6893                 .is_attach_btf = true,
6894                 .attach_fn = attach_trace),
6895         SEC_DEF("fmod_ret/", TRACING,
6896                 .expected_attach_type = BPF_MODIFY_RETURN,
6897                 .is_attach_btf = true,
6898                 .attach_fn = attach_trace),
6899         SEC_DEF("fexit/", TRACING,
6900                 .expected_attach_type = BPF_TRACE_FEXIT,
6901                 .is_attach_btf = true,
6902                 .attach_fn = attach_trace),
6903         SEC_DEF("freplace/", EXT,
6904                 .is_attach_btf = true,
6905                 .attach_fn = attach_trace),
6906         SEC_DEF("lsm/", LSM,
6907                 .is_attach_btf = true,
6908                 .expected_attach_type = BPF_LSM_MAC,
6909                 .attach_fn = attach_lsm),
6910         SEC_DEF("iter/", TRACING,
6911                 .expected_attach_type = BPF_TRACE_ITER,
6912                 .is_attach_btf = true,
6913                 .attach_fn = attach_iter),
6914         BPF_EAPROG_SEC("xdp_devmap/",           BPF_PROG_TYPE_XDP,
6915                                                 BPF_XDP_DEVMAP),
6916         BPF_EAPROG_SEC("xdp_cpumap/",           BPF_PROG_TYPE_XDP,
6917                                                 BPF_XDP_CPUMAP),
6918         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
6919         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
6920         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
6921         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
6922         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
6923         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
6924         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
6925                                                 BPF_CGROUP_INET_INGRESS),
6926         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
6927                                                 BPF_CGROUP_INET_EGRESS),
6928         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
6929         BPF_EAPROG_SEC("cgroup/sock_create",    BPF_PROG_TYPE_CGROUP_SOCK,
6930                                                 BPF_CGROUP_INET_SOCK_CREATE),
6931         BPF_EAPROG_SEC("cgroup/sock_release",   BPF_PROG_TYPE_CGROUP_SOCK,
6932                                                 BPF_CGROUP_INET_SOCK_RELEASE),
6933         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
6934                                                 BPF_CGROUP_INET_SOCK_CREATE),
6935         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
6936                                                 BPF_CGROUP_INET4_POST_BIND),
6937         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
6938                                                 BPF_CGROUP_INET6_POST_BIND),
6939         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
6940                                                 BPF_CGROUP_DEVICE),
6941         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
6942                                                 BPF_CGROUP_SOCK_OPS),
6943         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
6944                                                 BPF_SK_SKB_STREAM_PARSER),
6945         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
6946                                                 BPF_SK_SKB_STREAM_VERDICT),
6947         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
6948         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
6949                                                 BPF_SK_MSG_VERDICT),
6950         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
6951                                                 BPF_LIRC_MODE2),
6952         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
6953                                                 BPF_FLOW_DISSECTOR),
6954         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6955                                                 BPF_CGROUP_INET4_BIND),
6956         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6957                                                 BPF_CGROUP_INET6_BIND),
6958         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6959                                                 BPF_CGROUP_INET4_CONNECT),
6960         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6961                                                 BPF_CGROUP_INET6_CONNECT),
6962         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6963                                                 BPF_CGROUP_UDP4_SENDMSG),
6964         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6965                                                 BPF_CGROUP_UDP6_SENDMSG),
6966         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6967                                                 BPF_CGROUP_UDP4_RECVMSG),
6968         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6969                                                 BPF_CGROUP_UDP6_RECVMSG),
6970         BPF_EAPROG_SEC("cgroup/getpeername4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6971                                                 BPF_CGROUP_INET4_GETPEERNAME),
6972         BPF_EAPROG_SEC("cgroup/getpeername6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6973                                                 BPF_CGROUP_INET6_GETPEERNAME),
6974         BPF_EAPROG_SEC("cgroup/getsockname4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6975                                                 BPF_CGROUP_INET4_GETSOCKNAME),
6976         BPF_EAPROG_SEC("cgroup/getsockname6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
6977                                                 BPF_CGROUP_INET6_GETSOCKNAME),
6978         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
6979                                                 BPF_CGROUP_SYSCTL),
6980         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
6981                                                 BPF_CGROUP_GETSOCKOPT),
6982         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
6983                                                 BPF_CGROUP_SETSOCKOPT),
6984         BPF_PROG_SEC("struct_ops",              BPF_PROG_TYPE_STRUCT_OPS),
6985         BPF_EAPROG_SEC("sk_lookup/",            BPF_PROG_TYPE_SK_LOOKUP,
6986                                                 BPF_SK_LOOKUP),
6987 };
6988
6989 #undef BPF_PROG_SEC_IMPL
6990 #undef BPF_PROG_SEC
6991 #undef BPF_APROG_SEC
6992 #undef BPF_EAPROG_SEC
6993 #undef BPF_APROG_COMPAT
6994 #undef SEC_DEF
6995
6996 #define MAX_TYPE_NAME_SIZE 32
6997
6998 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
6999 {
7000         int i, n = ARRAY_SIZE(section_defs);
7001
7002         for (i = 0; i < n; i++) {
7003                 if (strncmp(sec_name,
7004                             section_defs[i].sec, section_defs[i].len))
7005                         continue;
7006                 return &section_defs[i];
7007         }
7008         return NULL;
7009 }
7010
7011 static char *libbpf_get_type_names(bool attach_type)
7012 {
7013         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
7014         char *buf;
7015
7016         buf = malloc(len);
7017         if (!buf)
7018                 return NULL;
7019
7020         buf[0] = '\0';
7021         /* Forge string buf with all available names */
7022         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7023                 if (attach_type && !section_defs[i].is_attachable)
7024                         continue;
7025
7026                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
7027                         free(buf);
7028                         return NULL;
7029                 }
7030                 strcat(buf, " ");
7031                 strcat(buf, section_defs[i].sec);
7032         }
7033
7034         return buf;
7035 }
7036
7037 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
7038                              enum bpf_attach_type *expected_attach_type)
7039 {
7040         const struct bpf_sec_def *sec_def;
7041         char *type_names;
7042
7043         if (!name)
7044                 return -EINVAL;
7045
7046         sec_def = find_sec_def(name);
7047         if (sec_def) {
7048                 *prog_type = sec_def->prog_type;
7049                 *expected_attach_type = sec_def->expected_attach_type;
7050                 return 0;
7051         }
7052
7053         pr_debug("failed to guess program type from ELF section '%s'\n", name);
7054         type_names = libbpf_get_type_names(false);
7055         if (type_names != NULL) {
7056                 pr_debug("supported section(type) names are:%s\n", type_names);
7057                 free(type_names);
7058         }
7059
7060         return -ESRCH;
7061 }
7062
7063 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
7064                                                      size_t offset)
7065 {
7066         struct bpf_map *map;
7067         size_t i;
7068
7069         for (i = 0; i < obj->nr_maps; i++) {
7070                 map = &obj->maps[i];
7071                 if (!bpf_map__is_struct_ops(map))
7072                         continue;
7073                 if (map->sec_offset <= offset &&
7074                     offset - map->sec_offset < map->def.value_size)
7075                         return map;
7076         }
7077
7078         return NULL;
7079 }
7080
7081 /* Collect the reloc from ELF and populate the st_ops->progs[] */
7082 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
7083                                             GElf_Shdr *shdr, Elf_Data *data)
7084 {
7085         const struct btf_member *member;
7086         struct bpf_struct_ops *st_ops;
7087         struct bpf_program *prog;
7088         unsigned int shdr_idx;
7089         const struct btf *btf;
7090         struct bpf_map *map;
7091         Elf_Data *symbols;
7092         unsigned int moff;
7093         const char *name;
7094         __u32 member_idx;
7095         GElf_Sym sym;
7096         GElf_Rel rel;
7097         int i, nrels;
7098
7099         symbols = obj->efile.symbols;
7100         btf = obj->btf;
7101         nrels = shdr->sh_size / shdr->sh_entsize;
7102         for (i = 0; i < nrels; i++) {
7103                 if (!gelf_getrel(data, i, &rel)) {
7104                         pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
7105                         return -LIBBPF_ERRNO__FORMAT;
7106                 }
7107
7108                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
7109                         pr_warn("struct_ops reloc: symbol %zx not found\n",
7110                                 (size_t)GELF_R_SYM(rel.r_info));
7111                         return -LIBBPF_ERRNO__FORMAT;
7112                 }
7113
7114                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
7115                                   sym.st_name) ? : "<?>";
7116                 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
7117                 if (!map) {
7118                         pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
7119                                 (size_t)rel.r_offset);
7120                         return -EINVAL;
7121                 }
7122
7123                 moff = rel.r_offset - map->sec_offset;
7124                 shdr_idx = sym.st_shndx;
7125                 st_ops = map->st_ops;
7126                 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",
7127                          map->name,
7128                          (long long)(rel.r_info >> 32),
7129                          (long long)sym.st_value,
7130                          shdr_idx, (size_t)rel.r_offset,
7131                          map->sec_offset, sym.st_name, name);
7132
7133                 if (shdr_idx >= SHN_LORESERVE) {
7134                         pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
7135                                 map->name, (size_t)rel.r_offset, shdr_idx);
7136                         return -LIBBPF_ERRNO__RELOC;
7137                 }
7138
7139                 member = find_member_by_offset(st_ops->type, moff * 8);
7140                 if (!member) {
7141                         pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
7142                                 map->name, moff);
7143                         return -EINVAL;
7144                 }
7145                 member_idx = member - btf_members(st_ops->type);
7146                 name = btf__name_by_offset(btf, member->name_off);
7147
7148                 if (!resolve_func_ptr(btf, member->type, NULL)) {
7149                         pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
7150                                 map->name, name);
7151                         return -EINVAL;
7152                 }
7153
7154                 prog = bpf_object__find_prog_by_idx(obj, shdr_idx);
7155                 if (!prog) {
7156                         pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
7157                                 map->name, shdr_idx, name);
7158                         return -EINVAL;
7159                 }
7160
7161                 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
7162                         const struct bpf_sec_def *sec_def;
7163
7164                         sec_def = find_sec_def(prog->section_name);
7165                         if (sec_def &&
7166                             sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
7167                                 /* for pr_warn */
7168                                 prog->type = sec_def->prog_type;
7169                                 goto invalid_prog;
7170                         }
7171
7172                         prog->type = BPF_PROG_TYPE_STRUCT_OPS;
7173                         prog->attach_btf_id = st_ops->type_id;
7174                         prog->expected_attach_type = member_idx;
7175                 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
7176                            prog->attach_btf_id != st_ops->type_id ||
7177                            prog->expected_attach_type != member_idx) {
7178                         goto invalid_prog;
7179                 }
7180                 st_ops->progs[member_idx] = prog;
7181         }
7182
7183         return 0;
7184
7185 invalid_prog:
7186         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",
7187                 map->name, prog->name, prog->section_name, prog->type,
7188                 prog->attach_btf_id, prog->expected_attach_type, name);
7189         return -EINVAL;
7190 }
7191
7192 #define BTF_TRACE_PREFIX "btf_trace_"
7193 #define BTF_LSM_PREFIX "bpf_lsm_"
7194 #define BTF_ITER_PREFIX "bpf_iter_"
7195 #define BTF_MAX_NAME_SIZE 128
7196
7197 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
7198                                    const char *name, __u32 kind)
7199 {
7200         char btf_type_name[BTF_MAX_NAME_SIZE];
7201         int ret;
7202
7203         ret = snprintf(btf_type_name, sizeof(btf_type_name),
7204                        "%s%s", prefix, name);
7205         /* snprintf returns the number of characters written excluding the
7206          * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
7207          * indicates truncation.
7208          */
7209         if (ret < 0 || ret >= sizeof(btf_type_name))
7210                 return -ENAMETOOLONG;
7211         return btf__find_by_name_kind(btf, btf_type_name, kind);
7212 }
7213
7214 static inline int __find_vmlinux_btf_id(struct btf *btf, const char *name,
7215                                         enum bpf_attach_type attach_type)
7216 {
7217         int err;
7218
7219         if (attach_type == BPF_TRACE_RAW_TP)
7220                 err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name,
7221                                               BTF_KIND_TYPEDEF);
7222         else if (attach_type == BPF_LSM_MAC)
7223                 err = find_btf_by_prefix_kind(btf, BTF_LSM_PREFIX, name,
7224                                               BTF_KIND_FUNC);
7225         else if (attach_type == BPF_TRACE_ITER)
7226                 err = find_btf_by_prefix_kind(btf, BTF_ITER_PREFIX, name,
7227                                               BTF_KIND_FUNC);
7228         else
7229                 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
7230
7231         if (err <= 0)
7232                 pr_warn("%s is not found in vmlinux BTF\n", name);
7233
7234         return err;
7235 }
7236
7237 int libbpf_find_vmlinux_btf_id(const char *name,
7238                                enum bpf_attach_type attach_type)
7239 {
7240         struct btf *btf;
7241         int err;
7242
7243         btf = libbpf_find_kernel_btf();
7244         if (IS_ERR(btf)) {
7245                 pr_warn("vmlinux BTF is not found\n");
7246                 return -EINVAL;
7247         }
7248
7249         err = __find_vmlinux_btf_id(btf, name, attach_type);
7250         btf__free(btf);
7251         return err;
7252 }
7253
7254 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
7255 {
7256         struct bpf_prog_info_linear *info_linear;
7257         struct bpf_prog_info *info;
7258         struct btf *btf = NULL;
7259         int err = -EINVAL;
7260
7261         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
7262         if (IS_ERR_OR_NULL(info_linear)) {
7263                 pr_warn("failed get_prog_info_linear for FD %d\n",
7264                         attach_prog_fd);
7265                 return -EINVAL;
7266         }
7267         info = &info_linear->info;
7268         if (!info->btf_id) {
7269                 pr_warn("The target program doesn't have BTF\n");
7270                 goto out;
7271         }
7272         if (btf__get_from_id(info->btf_id, &btf)) {
7273                 pr_warn("Failed to get BTF of the program\n");
7274                 goto out;
7275         }
7276         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
7277         btf__free(btf);
7278         if (err <= 0) {
7279                 pr_warn("%s is not found in prog's BTF\n", name);
7280                 goto out;
7281         }
7282 out:
7283         free(info_linear);
7284         return err;
7285 }
7286
7287 static int libbpf_find_attach_btf_id(struct bpf_program *prog)
7288 {
7289         enum bpf_attach_type attach_type = prog->expected_attach_type;
7290         __u32 attach_prog_fd = prog->attach_prog_fd;
7291         const char *name = prog->section_name;
7292         int i, err;
7293
7294         if (!name)
7295                 return -EINVAL;
7296
7297         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7298                 if (!section_defs[i].is_attach_btf)
7299                         continue;
7300                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
7301                         continue;
7302                 if (attach_prog_fd)
7303                         err = libbpf_find_prog_btf_id(name + section_defs[i].len,
7304                                                       attach_prog_fd);
7305                 else
7306                         err = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
7307                                                     name + section_defs[i].len,
7308                                                     attach_type);
7309                 return err;
7310         }
7311         pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
7312         return -ESRCH;
7313 }
7314
7315 int libbpf_attach_type_by_name(const char *name,
7316                                enum bpf_attach_type *attach_type)
7317 {
7318         char *type_names;
7319         int i;
7320
7321         if (!name)
7322                 return -EINVAL;
7323
7324         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
7325                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
7326                         continue;
7327                 if (!section_defs[i].is_attachable)
7328                         return -EINVAL;
7329                 *attach_type = section_defs[i].expected_attach_type;
7330                 return 0;
7331         }
7332         pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
7333         type_names = libbpf_get_type_names(true);
7334         if (type_names != NULL) {
7335                 pr_debug("attachable section(type) names are:%s\n", type_names);
7336                 free(type_names);
7337         }
7338
7339         return -EINVAL;
7340 }
7341
7342 int bpf_map__fd(const struct bpf_map *map)
7343 {
7344         return map ? map->fd : -EINVAL;
7345 }
7346
7347 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
7348 {
7349         return map ? &map->def : ERR_PTR(-EINVAL);
7350 }
7351
7352 const char *bpf_map__name(const struct bpf_map *map)
7353 {
7354         return map ? map->name : NULL;
7355 }
7356
7357 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
7358 {
7359         return map->def.type;
7360 }
7361
7362 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
7363 {
7364         if (map->fd >= 0)
7365                 return -EBUSY;
7366         map->def.type = type;
7367         return 0;
7368 }
7369
7370 __u32 bpf_map__map_flags(const struct bpf_map *map)
7371 {
7372         return map->def.map_flags;
7373 }
7374
7375 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
7376 {
7377         if (map->fd >= 0)
7378                 return -EBUSY;
7379         map->def.map_flags = flags;
7380         return 0;
7381 }
7382
7383 __u32 bpf_map__numa_node(const struct bpf_map *map)
7384 {
7385         return map->numa_node;
7386 }
7387
7388 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
7389 {
7390         if (map->fd >= 0)
7391                 return -EBUSY;
7392         map->numa_node = numa_node;
7393         return 0;
7394 }
7395
7396 __u32 bpf_map__key_size(const struct bpf_map *map)
7397 {
7398         return map->def.key_size;
7399 }
7400
7401 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
7402 {
7403         if (map->fd >= 0)
7404                 return -EBUSY;
7405         map->def.key_size = size;
7406         return 0;
7407 }
7408
7409 __u32 bpf_map__value_size(const struct bpf_map *map)
7410 {
7411         return map->def.value_size;
7412 }
7413
7414 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
7415 {
7416         if (map->fd >= 0)
7417                 return -EBUSY;
7418         map->def.value_size = size;
7419         return 0;
7420 }
7421
7422 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
7423 {
7424         return map ? map->btf_key_type_id : 0;
7425 }
7426
7427 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
7428 {
7429         return map ? map->btf_value_type_id : 0;
7430 }
7431
7432 int bpf_map__set_priv(struct bpf_map *map, void *priv,
7433                      bpf_map_clear_priv_t clear_priv)
7434 {
7435         if (!map)
7436                 return -EINVAL;
7437
7438         if (map->priv) {
7439                 if (map->clear_priv)
7440                         map->clear_priv(map, map->priv);
7441         }
7442
7443         map->priv = priv;
7444         map->clear_priv = clear_priv;
7445         return 0;
7446 }
7447
7448 void *bpf_map__priv(const struct bpf_map *map)
7449 {
7450         return map ? map->priv : ERR_PTR(-EINVAL);
7451 }
7452
7453 int bpf_map__set_initial_value(struct bpf_map *map,
7454                                const void *data, size_t size)
7455 {
7456         if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
7457             size != map->def.value_size || map->fd >= 0)
7458                 return -EINVAL;
7459
7460         memcpy(map->mmaped, data, size);
7461         return 0;
7462 }
7463
7464 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
7465 {
7466         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
7467 }
7468
7469 bool bpf_map__is_internal(const struct bpf_map *map)
7470 {
7471         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
7472 }
7473
7474 __u32 bpf_map__ifindex(const struct bpf_map *map)
7475 {
7476         return map->map_ifindex;
7477 }
7478
7479 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
7480 {
7481         if (map->fd >= 0)
7482                 return -EBUSY;
7483         map->map_ifindex = ifindex;
7484         return 0;
7485 }
7486
7487 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
7488 {
7489         if (!bpf_map_type__is_map_in_map(map->def.type)) {
7490                 pr_warn("error: unsupported map type\n");
7491                 return -EINVAL;
7492         }
7493         if (map->inner_map_fd != -1) {
7494                 pr_warn("error: inner_map_fd already specified\n");
7495                 return -EINVAL;
7496         }
7497         map->inner_map_fd = fd;
7498         return 0;
7499 }
7500
7501 static struct bpf_map *
7502 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
7503 {
7504         ssize_t idx;
7505         struct bpf_map *s, *e;
7506
7507         if (!obj || !obj->maps)
7508                 return NULL;
7509
7510         s = obj->maps;
7511         e = obj->maps + obj->nr_maps;
7512
7513         if ((m < s) || (m >= e)) {
7514                 pr_warn("error in %s: map handler doesn't belong to object\n",
7515                          __func__);
7516                 return NULL;
7517         }
7518
7519         idx = (m - obj->maps) + i;
7520         if (idx >= obj->nr_maps || idx < 0)
7521                 return NULL;
7522         return &obj->maps[idx];
7523 }
7524
7525 struct bpf_map *
7526 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
7527 {
7528         if (prev == NULL)
7529                 return obj->maps;
7530
7531         return __bpf_map__iter(prev, obj, 1);
7532 }
7533
7534 struct bpf_map *
7535 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
7536 {
7537         if (next == NULL) {
7538                 if (!obj->nr_maps)
7539                         return NULL;
7540                 return obj->maps + obj->nr_maps - 1;
7541         }
7542
7543         return __bpf_map__iter(next, obj, -1);
7544 }
7545
7546 struct bpf_map *
7547 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
7548 {
7549         struct bpf_map *pos;
7550
7551         bpf_object__for_each_map(pos, obj) {
7552                 if (pos->name && !strcmp(pos->name, name))
7553                         return pos;
7554         }
7555         return NULL;
7556 }
7557
7558 int
7559 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
7560 {
7561         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
7562 }
7563
7564 struct bpf_map *
7565 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
7566 {
7567         return ERR_PTR(-ENOTSUP);
7568 }
7569
7570 long libbpf_get_error(const void *ptr)
7571 {
7572         return PTR_ERR_OR_ZERO(ptr);
7573 }
7574
7575 int bpf_prog_load(const char *file, enum bpf_prog_type type,
7576                   struct bpf_object **pobj, int *prog_fd)
7577 {
7578         struct bpf_prog_load_attr attr;
7579
7580         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
7581         attr.file = file;
7582         attr.prog_type = type;
7583         attr.expected_attach_type = 0;
7584
7585         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
7586 }
7587
7588 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
7589                         struct bpf_object **pobj, int *prog_fd)
7590 {
7591         struct bpf_object_open_attr open_attr = {};
7592         struct bpf_program *prog, *first_prog = NULL;
7593         struct bpf_object *obj;
7594         struct bpf_map *map;
7595         int err;
7596
7597         if (!attr)
7598                 return -EINVAL;
7599         if (!attr->file)
7600                 return -EINVAL;
7601
7602         open_attr.file = attr->file;
7603         open_attr.prog_type = attr->prog_type;
7604
7605         obj = bpf_object__open_xattr(&open_attr);
7606         if (IS_ERR_OR_NULL(obj))
7607                 return -ENOENT;
7608
7609         bpf_object__for_each_program(prog, obj) {
7610                 enum bpf_attach_type attach_type = attr->expected_attach_type;
7611                 /*
7612                  * to preserve backwards compatibility, bpf_prog_load treats
7613                  * attr->prog_type, if specified, as an override to whatever
7614                  * bpf_object__open guessed
7615                  */
7616                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
7617                         bpf_program__set_type(prog, attr->prog_type);
7618                         bpf_program__set_expected_attach_type(prog,
7619                                                               attach_type);
7620                 }
7621                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
7622                         /*
7623                          * we haven't guessed from section name and user
7624                          * didn't provide a fallback type, too bad...
7625                          */
7626                         bpf_object__close(obj);
7627                         return -EINVAL;
7628                 }
7629
7630                 prog->prog_ifindex = attr->ifindex;
7631                 prog->log_level = attr->log_level;
7632                 prog->prog_flags = attr->prog_flags;
7633                 if (!first_prog)
7634                         first_prog = prog;
7635         }
7636
7637         bpf_object__for_each_map(map, obj) {
7638                 if (!bpf_map__is_offload_neutral(map))
7639                         map->map_ifindex = attr->ifindex;
7640         }
7641
7642         if (!first_prog) {
7643                 pr_warn("object file doesn't contain bpf program\n");
7644                 bpf_object__close(obj);
7645                 return -ENOENT;
7646         }
7647
7648         err = bpf_object__load(obj);
7649         if (err) {
7650                 bpf_object__close(obj);
7651                 return err;
7652         }
7653
7654         *pobj = obj;
7655         *prog_fd = bpf_program__fd(first_prog);
7656         return 0;
7657 }
7658
7659 struct bpf_link {
7660         int (*detach)(struct bpf_link *link);
7661         int (*destroy)(struct bpf_link *link);
7662         char *pin_path;         /* NULL, if not pinned */
7663         int fd;                 /* hook FD, -1 if not applicable */
7664         bool disconnected;
7665 };
7666
7667 /* Replace link's underlying BPF program with the new one */
7668 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
7669 {
7670         return bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
7671 }
7672
7673 /* Release "ownership" of underlying BPF resource (typically, BPF program
7674  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
7675  * link, when destructed through bpf_link__destroy() call won't attempt to
7676  * detach/unregisted that BPF resource. This is useful in situations where,
7677  * say, attached BPF program has to outlive userspace program that attached it
7678  * in the system. Depending on type of BPF program, though, there might be
7679  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
7680  * exit of userspace program doesn't trigger automatic detachment and clean up
7681  * inside the kernel.
7682  */
7683 void bpf_link__disconnect(struct bpf_link *link)
7684 {
7685         link->disconnected = true;
7686 }
7687
7688 int bpf_link__destroy(struct bpf_link *link)
7689 {
7690         int err = 0;
7691
7692         if (!link)
7693                 return 0;
7694
7695         if (!link->disconnected && link->detach)
7696                 err = link->detach(link);
7697         if (link->destroy)
7698                 link->destroy(link);
7699         if (link->pin_path)
7700                 free(link->pin_path);
7701         free(link);
7702
7703         return err;
7704 }
7705
7706 int bpf_link__fd(const struct bpf_link *link)
7707 {
7708         return link->fd;
7709 }
7710
7711 const char *bpf_link__pin_path(const struct bpf_link *link)
7712 {
7713         return link->pin_path;
7714 }
7715
7716 static int bpf_link__detach_fd(struct bpf_link *link)
7717 {
7718         return close(link->fd);
7719 }
7720
7721 struct bpf_link *bpf_link__open(const char *path)
7722 {
7723         struct bpf_link *link;
7724         int fd;
7725
7726         fd = bpf_obj_get(path);
7727         if (fd < 0) {
7728                 fd = -errno;
7729                 pr_warn("failed to open link at %s: %d\n", path, fd);
7730                 return ERR_PTR(fd);
7731         }
7732
7733         link = calloc(1, sizeof(*link));
7734         if (!link) {
7735                 close(fd);
7736                 return ERR_PTR(-ENOMEM);
7737         }
7738         link->detach = &bpf_link__detach_fd;
7739         link->fd = fd;
7740
7741         link->pin_path = strdup(path);
7742         if (!link->pin_path) {
7743                 bpf_link__destroy(link);
7744                 return ERR_PTR(-ENOMEM);
7745         }
7746
7747         return link;
7748 }
7749
7750 int bpf_link__pin(struct bpf_link *link, const char *path)
7751 {
7752         int err;
7753
7754         if (link->pin_path)
7755                 return -EBUSY;
7756         err = make_parent_dir(path);
7757         if (err)
7758                 return err;
7759         err = check_path(path);
7760         if (err)
7761                 return err;
7762
7763         link->pin_path = strdup(path);
7764         if (!link->pin_path)
7765                 return -ENOMEM;
7766
7767         if (bpf_obj_pin(link->fd, link->pin_path)) {
7768                 err = -errno;
7769                 zfree(&link->pin_path);
7770                 return err;
7771         }
7772
7773         pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
7774         return 0;
7775 }
7776
7777 int bpf_link__unpin(struct bpf_link *link)
7778 {
7779         int err;
7780
7781         if (!link->pin_path)
7782                 return -EINVAL;
7783
7784         err = unlink(link->pin_path);
7785         if (err != 0)
7786                 return -errno;
7787
7788         pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
7789         zfree(&link->pin_path);
7790         return 0;
7791 }
7792
7793 static int bpf_link__detach_perf_event(struct bpf_link *link)
7794 {
7795         int err;
7796
7797         err = ioctl(link->fd, PERF_EVENT_IOC_DISABLE, 0);
7798         if (err)
7799                 err = -errno;
7800
7801         close(link->fd);
7802         return err;
7803 }
7804
7805 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
7806                                                 int pfd)
7807 {
7808         char errmsg[STRERR_BUFSIZE];
7809         struct bpf_link *link;
7810         int prog_fd, err;
7811
7812         if (pfd < 0) {
7813                 pr_warn("program '%s': invalid perf event FD %d\n",
7814                         bpf_program__title(prog, false), pfd);
7815                 return ERR_PTR(-EINVAL);
7816         }
7817         prog_fd = bpf_program__fd(prog);
7818         if (prog_fd < 0) {
7819                 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
7820                         bpf_program__title(prog, false));
7821                 return ERR_PTR(-EINVAL);
7822         }
7823
7824         link = calloc(1, sizeof(*link));
7825         if (!link)
7826                 return ERR_PTR(-ENOMEM);
7827         link->detach = &bpf_link__detach_perf_event;
7828         link->fd = pfd;
7829
7830         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
7831                 err = -errno;
7832                 free(link);
7833                 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
7834                         bpf_program__title(prog, false), pfd,
7835                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7836                 if (err == -EPROTO)
7837                         pr_warn("program '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
7838                                 bpf_program__title(prog, false), pfd);
7839                 return ERR_PTR(err);
7840         }
7841         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
7842                 err = -errno;
7843                 free(link);
7844                 pr_warn("program '%s': failed to enable pfd %d: %s\n",
7845                         bpf_program__title(prog, false), pfd,
7846                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7847                 return ERR_PTR(err);
7848         }
7849         return link;
7850 }
7851
7852 /*
7853  * this function is expected to parse integer in the range of [0, 2^31-1] from
7854  * given file using scanf format string fmt. If actual parsed value is
7855  * negative, the result might be indistinguishable from error
7856  */
7857 static int parse_uint_from_file(const char *file, const char *fmt)
7858 {
7859         char buf[STRERR_BUFSIZE];
7860         int err, ret;
7861         FILE *f;
7862
7863         f = fopen(file, "r");
7864         if (!f) {
7865                 err = -errno;
7866                 pr_debug("failed to open '%s': %s\n", file,
7867                          libbpf_strerror_r(err, buf, sizeof(buf)));
7868                 return err;
7869         }
7870         err = fscanf(f, fmt, &ret);
7871         if (err != 1) {
7872                 err = err == EOF ? -EIO : -errno;
7873                 pr_debug("failed to parse '%s': %s\n", file,
7874                         libbpf_strerror_r(err, buf, sizeof(buf)));
7875                 fclose(f);
7876                 return err;
7877         }
7878         fclose(f);
7879         return ret;
7880 }
7881
7882 static int determine_kprobe_perf_type(void)
7883 {
7884         const char *file = "/sys/bus/event_source/devices/kprobe/type";
7885
7886         return parse_uint_from_file(file, "%d\n");
7887 }
7888
7889 static int determine_uprobe_perf_type(void)
7890 {
7891         const char *file = "/sys/bus/event_source/devices/uprobe/type";
7892
7893         return parse_uint_from_file(file, "%d\n");
7894 }
7895
7896 static int determine_kprobe_retprobe_bit(void)
7897 {
7898         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
7899
7900         return parse_uint_from_file(file, "config:%d\n");
7901 }
7902
7903 static int determine_uprobe_retprobe_bit(void)
7904 {
7905         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
7906
7907         return parse_uint_from_file(file, "config:%d\n");
7908 }
7909
7910 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
7911                                  uint64_t offset, int pid)
7912 {
7913         struct perf_event_attr attr = {};
7914         char errmsg[STRERR_BUFSIZE];
7915         int type, pfd, err;
7916
7917         type = uprobe ? determine_uprobe_perf_type()
7918                       : determine_kprobe_perf_type();
7919         if (type < 0) {
7920                 pr_warn("failed to determine %s perf type: %s\n",
7921                         uprobe ? "uprobe" : "kprobe",
7922                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
7923                 return type;
7924         }
7925         if (retprobe) {
7926                 int bit = uprobe ? determine_uprobe_retprobe_bit()
7927                                  : determine_kprobe_retprobe_bit();
7928
7929                 if (bit < 0) {
7930                         pr_warn("failed to determine %s retprobe bit: %s\n",
7931                                 uprobe ? "uprobe" : "kprobe",
7932                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
7933                         return bit;
7934                 }
7935                 attr.config |= 1 << bit;
7936         }
7937         attr.size = sizeof(attr);
7938         attr.type = type;
7939         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
7940         attr.config2 = offset;           /* kprobe_addr or probe_offset */
7941
7942         /* pid filter is meaningful only for uprobes */
7943         pfd = syscall(__NR_perf_event_open, &attr,
7944                       pid < 0 ? -1 : pid /* pid */,
7945                       pid == -1 ? 0 : -1 /* cpu */,
7946                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
7947         if (pfd < 0) {
7948                 err = -errno;
7949                 pr_warn("%s perf_event_open() failed: %s\n",
7950                         uprobe ? "uprobe" : "kprobe",
7951                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7952                 return err;
7953         }
7954         return pfd;
7955 }
7956
7957 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
7958                                             bool retprobe,
7959                                             const char *func_name)
7960 {
7961         char errmsg[STRERR_BUFSIZE];
7962         struct bpf_link *link;
7963         int pfd, err;
7964
7965         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
7966                                     0 /* offset */, -1 /* pid */);
7967         if (pfd < 0) {
7968                 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
7969                         bpf_program__title(prog, false),
7970                         retprobe ? "kretprobe" : "kprobe", func_name,
7971                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
7972                 return ERR_PTR(pfd);
7973         }
7974         link = bpf_program__attach_perf_event(prog, pfd);
7975         if (IS_ERR(link)) {
7976                 close(pfd);
7977                 err = PTR_ERR(link);
7978                 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
7979                         bpf_program__title(prog, false),
7980                         retprobe ? "kretprobe" : "kprobe", func_name,
7981                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
7982                 return link;
7983         }
7984         return link;
7985 }
7986
7987 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
7988                                       struct bpf_program *prog)
7989 {
7990         const char *func_name;
7991         bool retprobe;
7992
7993         func_name = bpf_program__title(prog, false) + sec->len;
7994         retprobe = strcmp(sec->sec, "kretprobe/") == 0;
7995
7996         return bpf_program__attach_kprobe(prog, retprobe, func_name);
7997 }
7998
7999 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
8000                                             bool retprobe, pid_t pid,
8001                                             const char *binary_path,
8002                                             size_t func_offset)
8003 {
8004         char errmsg[STRERR_BUFSIZE];
8005         struct bpf_link *link;
8006         int pfd, err;
8007
8008         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
8009                                     binary_path, func_offset, pid);
8010         if (pfd < 0) {
8011                 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
8012                         bpf_program__title(prog, false),
8013                         retprobe ? "uretprobe" : "uprobe",
8014                         binary_path, func_offset,
8015                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8016                 return ERR_PTR(pfd);
8017         }
8018         link = bpf_program__attach_perf_event(prog, pfd);
8019         if (IS_ERR(link)) {
8020                 close(pfd);
8021                 err = PTR_ERR(link);
8022                 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
8023                         bpf_program__title(prog, false),
8024                         retprobe ? "uretprobe" : "uprobe",
8025                         binary_path, func_offset,
8026                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8027                 return link;
8028         }
8029         return link;
8030 }
8031
8032 static int determine_tracepoint_id(const char *tp_category,
8033                                    const char *tp_name)
8034 {
8035         char file[PATH_MAX];
8036         int ret;
8037
8038         ret = snprintf(file, sizeof(file),
8039                        "/sys/kernel/debug/tracing/events/%s/%s/id",
8040                        tp_category, tp_name);
8041         if (ret < 0)
8042                 return -errno;
8043         if (ret >= sizeof(file)) {
8044                 pr_debug("tracepoint %s/%s path is too long\n",
8045                          tp_category, tp_name);
8046                 return -E2BIG;
8047         }
8048         return parse_uint_from_file(file, "%d\n");
8049 }
8050
8051 static int perf_event_open_tracepoint(const char *tp_category,
8052                                       const char *tp_name)
8053 {
8054         struct perf_event_attr attr = {};
8055         char errmsg[STRERR_BUFSIZE];
8056         int tp_id, pfd, err;
8057
8058         tp_id = determine_tracepoint_id(tp_category, tp_name);
8059         if (tp_id < 0) {
8060                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
8061                         tp_category, tp_name,
8062                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
8063                 return tp_id;
8064         }
8065
8066         attr.type = PERF_TYPE_TRACEPOINT;
8067         attr.size = sizeof(attr);
8068         attr.config = tp_id;
8069
8070         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
8071                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
8072         if (pfd < 0) {
8073                 err = -errno;
8074                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
8075                         tp_category, tp_name,
8076                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8077                 return err;
8078         }
8079         return pfd;
8080 }
8081
8082 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
8083                                                 const char *tp_category,
8084                                                 const char *tp_name)
8085 {
8086         char errmsg[STRERR_BUFSIZE];
8087         struct bpf_link *link;
8088         int pfd, err;
8089
8090         pfd = perf_event_open_tracepoint(tp_category, tp_name);
8091         if (pfd < 0) {
8092                 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
8093                         bpf_program__title(prog, false),
8094                         tp_category, tp_name,
8095                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8096                 return ERR_PTR(pfd);
8097         }
8098         link = bpf_program__attach_perf_event(prog, pfd);
8099         if (IS_ERR(link)) {
8100                 close(pfd);
8101                 err = PTR_ERR(link);
8102                 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
8103                         bpf_program__title(prog, false),
8104                         tp_category, tp_name,
8105                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
8106                 return link;
8107         }
8108         return link;
8109 }
8110
8111 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
8112                                   struct bpf_program *prog)
8113 {
8114         char *sec_name, *tp_cat, *tp_name;
8115         struct bpf_link *link;
8116
8117         sec_name = strdup(bpf_program__title(prog, false));
8118         if (!sec_name)
8119                 return ERR_PTR(-ENOMEM);
8120
8121         /* extract "tp/<category>/<name>" */
8122         tp_cat = sec_name + sec->len;
8123         tp_name = strchr(tp_cat, '/');
8124         if (!tp_name) {
8125                 link = ERR_PTR(-EINVAL);
8126                 goto out;
8127         }
8128         *tp_name = '\0';
8129         tp_name++;
8130
8131         link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
8132 out:
8133         free(sec_name);
8134         return link;
8135 }
8136
8137 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
8138                                                     const char *tp_name)
8139 {
8140         char errmsg[STRERR_BUFSIZE];
8141         struct bpf_link *link;
8142         int prog_fd, pfd;
8143
8144         prog_fd = bpf_program__fd(prog);
8145         if (prog_fd < 0) {
8146                 pr_warn("program '%s': can't attach before loaded\n",
8147                         bpf_program__title(prog, false));
8148                 return ERR_PTR(-EINVAL);
8149         }
8150
8151         link = calloc(1, sizeof(*link));
8152         if (!link)
8153                 return ERR_PTR(-ENOMEM);
8154         link->detach = &bpf_link__detach_fd;
8155
8156         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
8157         if (pfd < 0) {
8158                 pfd = -errno;
8159                 free(link);
8160                 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
8161                         bpf_program__title(prog, false), tp_name,
8162                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8163                 return ERR_PTR(pfd);
8164         }
8165         link->fd = pfd;
8166         return link;
8167 }
8168
8169 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
8170                                       struct bpf_program *prog)
8171 {
8172         const char *tp_name = bpf_program__title(prog, false) + sec->len;
8173
8174         return bpf_program__attach_raw_tracepoint(prog, tp_name);
8175 }
8176
8177 /* Common logic for all BPF program types that attach to a btf_id */
8178 static struct bpf_link *bpf_program__attach_btf_id(struct bpf_program *prog)
8179 {
8180         char errmsg[STRERR_BUFSIZE];
8181         struct bpf_link *link;
8182         int prog_fd, pfd;
8183
8184         prog_fd = bpf_program__fd(prog);
8185         if (prog_fd < 0) {
8186                 pr_warn("program '%s': can't attach before loaded\n",
8187                         bpf_program__title(prog, false));
8188                 return ERR_PTR(-EINVAL);
8189         }
8190
8191         link = calloc(1, sizeof(*link));
8192         if (!link)
8193                 return ERR_PTR(-ENOMEM);
8194         link->detach = &bpf_link__detach_fd;
8195
8196         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
8197         if (pfd < 0) {
8198                 pfd = -errno;
8199                 free(link);
8200                 pr_warn("program '%s': failed to attach: %s\n",
8201                         bpf_program__title(prog, false),
8202                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
8203                 return ERR_PTR(pfd);
8204         }
8205         link->fd = pfd;
8206         return (struct bpf_link *)link;
8207 }
8208
8209 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
8210 {
8211         return bpf_program__attach_btf_id(prog);
8212 }
8213
8214 struct bpf_link *bpf_program__attach_lsm(struct bpf_program *prog)
8215 {
8216         return bpf_program__attach_btf_id(prog);
8217 }
8218
8219 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
8220                                      struct bpf_program *prog)
8221 {
8222         return bpf_program__attach_trace(prog);
8223 }
8224
8225 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
8226                                    struct bpf_program *prog)
8227 {
8228         return bpf_program__attach_lsm(prog);
8229 }
8230
8231 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
8232                                     struct bpf_program *prog)
8233 {
8234         return bpf_program__attach_iter(prog, NULL);
8235 }
8236
8237 static struct bpf_link *
8238 bpf_program__attach_fd(struct bpf_program *prog, int target_fd,
8239                        const char *target_name)
8240 {
8241         enum bpf_attach_type attach_type;
8242         char errmsg[STRERR_BUFSIZE];
8243         struct bpf_link *link;
8244         int prog_fd, link_fd;
8245
8246         prog_fd = bpf_program__fd(prog);
8247         if (prog_fd < 0) {
8248                 pr_warn("program '%s': can't attach before loaded\n",
8249                         bpf_program__title(prog, false));
8250                 return ERR_PTR(-EINVAL);
8251         }
8252
8253         link = calloc(1, sizeof(*link));
8254         if (!link)
8255                 return ERR_PTR(-ENOMEM);
8256         link->detach = &bpf_link__detach_fd;
8257
8258         attach_type = bpf_program__get_expected_attach_type(prog);
8259         link_fd = bpf_link_create(prog_fd, target_fd, attach_type, NULL);
8260         if (link_fd < 0) {
8261                 link_fd = -errno;
8262                 free(link);
8263                 pr_warn("program '%s': failed to attach to %s: %s\n",
8264                         bpf_program__title(prog, false), target_name,
8265                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
8266                 return ERR_PTR(link_fd);
8267         }
8268         link->fd = link_fd;
8269         return link;
8270 }
8271
8272 struct bpf_link *
8273 bpf_program__attach_cgroup(struct bpf_program *prog, int cgroup_fd)
8274 {
8275         return bpf_program__attach_fd(prog, cgroup_fd, "cgroup");
8276 }
8277
8278 struct bpf_link *
8279 bpf_program__attach_netns(struct bpf_program *prog, int netns_fd)
8280 {
8281         return bpf_program__attach_fd(prog, netns_fd, "netns");
8282 }
8283
8284 struct bpf_link *
8285 bpf_program__attach_iter(struct bpf_program *prog,
8286                          const struct bpf_iter_attach_opts *opts)
8287 {
8288         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
8289         char errmsg[STRERR_BUFSIZE];
8290         struct bpf_link *link;
8291         int prog_fd, link_fd;
8292         __u32 target_fd = 0;
8293
8294         if (!OPTS_VALID(opts, bpf_iter_attach_opts))
8295                 return ERR_PTR(-EINVAL);
8296
8297         if (OPTS_HAS(opts, map_fd)) {
8298                 target_fd = opts->map_fd;
8299                 link_create_opts.flags = BPF_ITER_LINK_MAP_FD;
8300         }
8301
8302         prog_fd = bpf_program__fd(prog);
8303         if (prog_fd < 0) {
8304                 pr_warn("program '%s': can't attach before loaded\n",
8305                         bpf_program__title(prog, false));
8306                 return ERR_PTR(-EINVAL);
8307         }
8308
8309         link = calloc(1, sizeof(*link));
8310         if (!link)
8311                 return ERR_PTR(-ENOMEM);
8312         link->detach = &bpf_link__detach_fd;
8313
8314         link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
8315                                   &link_create_opts);
8316         if (link_fd < 0) {
8317                 link_fd = -errno;
8318                 free(link);
8319                 pr_warn("program '%s': failed to attach to iterator: %s\n",
8320                         bpf_program__title(prog, false),
8321                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
8322                 return ERR_PTR(link_fd);
8323         }
8324         link->fd = link_fd;
8325         return link;
8326 }
8327
8328 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
8329 {
8330         const struct bpf_sec_def *sec_def;
8331
8332         sec_def = find_sec_def(bpf_program__title(prog, false));
8333         if (!sec_def || !sec_def->attach_fn)
8334                 return ERR_PTR(-ESRCH);
8335
8336         return sec_def->attach_fn(sec_def, prog);
8337 }
8338
8339 static int bpf_link__detach_struct_ops(struct bpf_link *link)
8340 {
8341         __u32 zero = 0;
8342
8343         if (bpf_map_delete_elem(link->fd, &zero))
8344                 return -errno;
8345
8346         return 0;
8347 }
8348
8349 struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
8350 {
8351         struct bpf_struct_ops *st_ops;
8352         struct bpf_link *link;
8353         __u32 i, zero = 0;
8354         int err;
8355
8356         if (!bpf_map__is_struct_ops(map) || map->fd == -1)
8357                 return ERR_PTR(-EINVAL);
8358
8359         link = calloc(1, sizeof(*link));
8360         if (!link)
8361                 return ERR_PTR(-EINVAL);
8362
8363         st_ops = map->st_ops;
8364         for (i = 0; i < btf_vlen(st_ops->type); i++) {
8365                 struct bpf_program *prog = st_ops->progs[i];
8366                 void *kern_data;
8367                 int prog_fd;
8368
8369                 if (!prog)
8370                         continue;
8371
8372                 prog_fd = bpf_program__fd(prog);
8373                 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
8374                 *(unsigned long *)kern_data = prog_fd;
8375         }
8376
8377         err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
8378         if (err) {
8379                 err = -errno;
8380                 free(link);
8381                 return ERR_PTR(err);
8382         }
8383
8384         link->detach = bpf_link__detach_struct_ops;
8385         link->fd = map->fd;
8386
8387         return link;
8388 }
8389
8390 enum bpf_perf_event_ret
8391 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
8392                            void **copy_mem, size_t *copy_size,
8393                            bpf_perf_event_print_t fn, void *private_data)
8394 {
8395         struct perf_event_mmap_page *header = mmap_mem;
8396         __u64 data_head = ring_buffer_read_head(header);
8397         __u64 data_tail = header->data_tail;
8398         void *base = ((__u8 *)header) + page_size;
8399         int ret = LIBBPF_PERF_EVENT_CONT;
8400         struct perf_event_header *ehdr;
8401         size_t ehdr_size;
8402
8403         while (data_head != data_tail) {
8404                 ehdr = base + (data_tail & (mmap_size - 1));
8405                 ehdr_size = ehdr->size;
8406
8407                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
8408                         void *copy_start = ehdr;
8409                         size_t len_first = base + mmap_size - copy_start;
8410                         size_t len_secnd = ehdr_size - len_first;
8411
8412                         if (*copy_size < ehdr_size) {
8413                                 free(*copy_mem);
8414                                 *copy_mem = malloc(ehdr_size);
8415                                 if (!*copy_mem) {
8416                                         *copy_size = 0;
8417                                         ret = LIBBPF_PERF_EVENT_ERROR;
8418                                         break;
8419                                 }
8420                                 *copy_size = ehdr_size;
8421                         }
8422
8423                         memcpy(*copy_mem, copy_start, len_first);
8424                         memcpy(*copy_mem + len_first, base, len_secnd);
8425                         ehdr = *copy_mem;
8426                 }
8427
8428                 ret = fn(ehdr, private_data);
8429                 data_tail += ehdr_size;
8430                 if (ret != LIBBPF_PERF_EVENT_CONT)
8431                         break;
8432         }
8433
8434         ring_buffer_write_tail(header, data_tail);
8435         return ret;
8436 }
8437
8438 struct perf_buffer;
8439
8440 struct perf_buffer_params {
8441         struct perf_event_attr *attr;
8442         /* if event_cb is specified, it takes precendence */
8443         perf_buffer_event_fn event_cb;
8444         /* sample_cb and lost_cb are higher-level common-case callbacks */
8445         perf_buffer_sample_fn sample_cb;
8446         perf_buffer_lost_fn lost_cb;
8447         void *ctx;
8448         int cpu_cnt;
8449         int *cpus;
8450         int *map_keys;
8451 };
8452
8453 struct perf_cpu_buf {
8454         struct perf_buffer *pb;
8455         void *base; /* mmap()'ed memory */
8456         void *buf; /* for reconstructing segmented data */
8457         size_t buf_size;
8458         int fd;
8459         int cpu;
8460         int map_key;
8461 };
8462
8463 struct perf_buffer {
8464         perf_buffer_event_fn event_cb;
8465         perf_buffer_sample_fn sample_cb;
8466         perf_buffer_lost_fn lost_cb;
8467         void *ctx; /* passed into callbacks */
8468
8469         size_t page_size;
8470         size_t mmap_size;
8471         struct perf_cpu_buf **cpu_bufs;
8472         struct epoll_event *events;
8473         int cpu_cnt; /* number of allocated CPU buffers */
8474         int epoll_fd; /* perf event FD */
8475         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
8476 };
8477
8478 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
8479                                       struct perf_cpu_buf *cpu_buf)
8480 {
8481         if (!cpu_buf)
8482                 return;
8483         if (cpu_buf->base &&
8484             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
8485                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
8486         if (cpu_buf->fd >= 0) {
8487                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
8488                 close(cpu_buf->fd);
8489         }
8490         free(cpu_buf->buf);
8491         free(cpu_buf);
8492 }
8493
8494 void perf_buffer__free(struct perf_buffer *pb)
8495 {
8496         int i;
8497
8498         if (!pb)
8499                 return;
8500         if (pb->cpu_bufs) {
8501                 for (i = 0; i < pb->cpu_cnt; i++) {
8502                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
8503
8504                         if (!cpu_buf)
8505                                 continue;
8506
8507                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
8508                         perf_buffer__free_cpu_buf(pb, cpu_buf);
8509                 }
8510                 free(pb->cpu_bufs);
8511         }
8512         if (pb->epoll_fd >= 0)
8513                 close(pb->epoll_fd);
8514         free(pb->events);
8515         free(pb);
8516 }
8517
8518 static struct perf_cpu_buf *
8519 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
8520                           int cpu, int map_key)
8521 {
8522         struct perf_cpu_buf *cpu_buf;
8523         char msg[STRERR_BUFSIZE];
8524         int err;
8525
8526         cpu_buf = calloc(1, sizeof(*cpu_buf));
8527         if (!cpu_buf)
8528                 return ERR_PTR(-ENOMEM);
8529
8530         cpu_buf->pb = pb;
8531         cpu_buf->cpu = cpu;
8532         cpu_buf->map_key = map_key;
8533
8534         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
8535                               -1, PERF_FLAG_FD_CLOEXEC);
8536         if (cpu_buf->fd < 0) {
8537                 err = -errno;
8538                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
8539                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8540                 goto error;
8541         }
8542
8543         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
8544                              PROT_READ | PROT_WRITE, MAP_SHARED,
8545                              cpu_buf->fd, 0);
8546         if (cpu_buf->base == MAP_FAILED) {
8547                 cpu_buf->base = NULL;
8548                 err = -errno;
8549                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
8550                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8551                 goto error;
8552         }
8553
8554         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
8555                 err = -errno;
8556                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
8557                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
8558                 goto error;
8559         }
8560
8561         return cpu_buf;
8562
8563 error:
8564         perf_buffer__free_cpu_buf(pb, cpu_buf);
8565         return (struct perf_cpu_buf *)ERR_PTR(err);
8566 }
8567
8568 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
8569                                               struct perf_buffer_params *p);
8570
8571 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
8572                                      const struct perf_buffer_opts *opts)
8573 {
8574         struct perf_buffer_params p = {};
8575         struct perf_event_attr attr = { 0, };
8576
8577         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
8578         attr.type = PERF_TYPE_SOFTWARE;
8579         attr.sample_type = PERF_SAMPLE_RAW;
8580         attr.sample_period = 1;
8581         attr.wakeup_events = 1;
8582
8583         p.attr = &attr;
8584         p.sample_cb = opts ? opts->sample_cb : NULL;
8585         p.lost_cb = opts ? opts->lost_cb : NULL;
8586         p.ctx = opts ? opts->ctx : NULL;
8587
8588         return __perf_buffer__new(map_fd, page_cnt, &p);
8589 }
8590
8591 struct perf_buffer *
8592 perf_buffer__new_raw(int map_fd, size_t page_cnt,
8593                      const struct perf_buffer_raw_opts *opts)
8594 {
8595         struct perf_buffer_params p = {};
8596
8597         p.attr = opts->attr;
8598         p.event_cb = opts->event_cb;
8599         p.ctx = opts->ctx;
8600         p.cpu_cnt = opts->cpu_cnt;
8601         p.cpus = opts->cpus;
8602         p.map_keys = opts->map_keys;
8603
8604         return __perf_buffer__new(map_fd, page_cnt, &p);
8605 }
8606
8607 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
8608                                               struct perf_buffer_params *p)
8609 {
8610         const char *online_cpus_file = "/sys/devices/system/cpu/online";
8611         struct bpf_map_info map;
8612         char msg[STRERR_BUFSIZE];
8613         struct perf_buffer *pb;
8614         bool *online = NULL;
8615         __u32 map_info_len;
8616         int err, i, j, n;
8617
8618         if (page_cnt & (page_cnt - 1)) {
8619                 pr_warn("page count should be power of two, but is %zu\n",
8620                         page_cnt);
8621                 return ERR_PTR(-EINVAL);
8622         }
8623
8624         /* best-effort sanity checks */
8625         memset(&map, 0, sizeof(map));
8626         map_info_len = sizeof(map);
8627         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
8628         if (err) {
8629                 err = -errno;
8630                 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
8631                  * -EBADFD, -EFAULT, or -E2BIG on real error
8632                  */
8633                 if (err != -EINVAL) {
8634                         pr_warn("failed to get map info for map FD %d: %s\n",
8635                                 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
8636                         return ERR_PTR(err);
8637                 }
8638                 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
8639                          map_fd);
8640         } else {
8641                 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
8642                         pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
8643                                 map.name);
8644                         return ERR_PTR(-EINVAL);
8645                 }
8646         }
8647
8648         pb = calloc(1, sizeof(*pb));
8649         if (!pb)
8650                 return ERR_PTR(-ENOMEM);
8651
8652         pb->event_cb = p->event_cb;
8653         pb->sample_cb = p->sample_cb;
8654         pb->lost_cb = p->lost_cb;
8655         pb->ctx = p->ctx;
8656
8657         pb->page_size = getpagesize();
8658         pb->mmap_size = pb->page_size * page_cnt;
8659         pb->map_fd = map_fd;
8660
8661         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
8662         if (pb->epoll_fd < 0) {
8663                 err = -errno;
8664                 pr_warn("failed to create epoll instance: %s\n",
8665                         libbpf_strerror_r(err, msg, sizeof(msg)));
8666                 goto error;
8667         }
8668
8669         if (p->cpu_cnt > 0) {
8670                 pb->cpu_cnt = p->cpu_cnt;
8671         } else {
8672                 pb->cpu_cnt = libbpf_num_possible_cpus();
8673                 if (pb->cpu_cnt < 0) {
8674                         err = pb->cpu_cnt;
8675                         goto error;
8676                 }
8677                 if (map.max_entries && map.max_entries < pb->cpu_cnt)
8678                         pb->cpu_cnt = map.max_entries;
8679         }
8680
8681         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
8682         if (!pb->events) {
8683                 err = -ENOMEM;
8684                 pr_warn("failed to allocate events: out of memory\n");
8685                 goto error;
8686         }
8687         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
8688         if (!pb->cpu_bufs) {
8689                 err = -ENOMEM;
8690                 pr_warn("failed to allocate buffers: out of memory\n");
8691                 goto error;
8692         }
8693
8694         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
8695         if (err) {
8696                 pr_warn("failed to get online CPU mask: %d\n", err);
8697                 goto error;
8698         }
8699
8700         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
8701                 struct perf_cpu_buf *cpu_buf;
8702                 int cpu, map_key;
8703
8704                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
8705                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
8706
8707                 /* in case user didn't explicitly requested particular CPUs to
8708                  * be attached to, skip offline/not present CPUs
8709                  */
8710                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
8711                         continue;
8712
8713                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
8714                 if (IS_ERR(cpu_buf)) {
8715                         err = PTR_ERR(cpu_buf);
8716                         goto error;
8717                 }
8718
8719                 pb->cpu_bufs[j] = cpu_buf;
8720
8721                 err = bpf_map_update_elem(pb->map_fd, &map_key,
8722                                           &cpu_buf->fd, 0);
8723                 if (err) {
8724                         err = -errno;
8725                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
8726                                 cpu, map_key, cpu_buf->fd,
8727                                 libbpf_strerror_r(err, msg, sizeof(msg)));
8728                         goto error;
8729                 }
8730
8731                 pb->events[j].events = EPOLLIN;
8732                 pb->events[j].data.ptr = cpu_buf;
8733                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
8734                               &pb->events[j]) < 0) {
8735                         err = -errno;
8736                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
8737                                 cpu, cpu_buf->fd,
8738                                 libbpf_strerror_r(err, msg, sizeof(msg)));
8739                         goto error;
8740                 }
8741                 j++;
8742         }
8743         pb->cpu_cnt = j;
8744         free(online);
8745
8746         return pb;
8747
8748 error:
8749         free(online);
8750         if (pb)
8751                 perf_buffer__free(pb);
8752         return ERR_PTR(err);
8753 }
8754
8755 struct perf_sample_raw {
8756         struct perf_event_header header;
8757         uint32_t size;
8758         char data[];
8759 };
8760
8761 struct perf_sample_lost {
8762         struct perf_event_header header;
8763         uint64_t id;
8764         uint64_t lost;
8765         uint64_t sample_id;
8766 };
8767
8768 static enum bpf_perf_event_ret
8769 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
8770 {
8771         struct perf_cpu_buf *cpu_buf = ctx;
8772         struct perf_buffer *pb = cpu_buf->pb;
8773         void *data = e;
8774
8775         /* user wants full control over parsing perf event */
8776         if (pb->event_cb)
8777                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
8778
8779         switch (e->type) {
8780         case PERF_RECORD_SAMPLE: {
8781                 struct perf_sample_raw *s = data;
8782
8783                 if (pb->sample_cb)
8784                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
8785                 break;
8786         }
8787         case PERF_RECORD_LOST: {
8788                 struct perf_sample_lost *s = data;
8789
8790                 if (pb->lost_cb)
8791                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
8792                 break;
8793         }
8794         default:
8795                 pr_warn("unknown perf sample type %d\n", e->type);
8796                 return LIBBPF_PERF_EVENT_ERROR;
8797         }
8798         return LIBBPF_PERF_EVENT_CONT;
8799 }
8800
8801 static int perf_buffer__process_records(struct perf_buffer *pb,
8802                                         struct perf_cpu_buf *cpu_buf)
8803 {
8804         enum bpf_perf_event_ret ret;
8805
8806         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
8807                                          pb->page_size, &cpu_buf->buf,
8808                                          &cpu_buf->buf_size,
8809                                          perf_buffer__process_record, cpu_buf);
8810         if (ret != LIBBPF_PERF_EVENT_CONT)
8811                 return ret;
8812         return 0;
8813 }
8814
8815 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
8816 {
8817         int i, cnt, err;
8818
8819         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
8820         for (i = 0; i < cnt; i++) {
8821                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
8822
8823                 err = perf_buffer__process_records(pb, cpu_buf);
8824                 if (err) {
8825                         pr_warn("error while processing records: %d\n", err);
8826                         return err;
8827                 }
8828         }
8829         return cnt < 0 ? -errno : cnt;
8830 }
8831
8832 int perf_buffer__consume(struct perf_buffer *pb)
8833 {
8834         int i, err;
8835
8836         for (i = 0; i < pb->cpu_cnt; i++) {
8837                 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
8838
8839                 if (!cpu_buf)
8840                         continue;
8841
8842                 err = perf_buffer__process_records(pb, cpu_buf);
8843                 if (err) {
8844                         pr_warn("error while processing records: %d\n", err);
8845                         return err;
8846                 }
8847         }
8848         return 0;
8849 }
8850
8851 struct bpf_prog_info_array_desc {
8852         int     array_offset;   /* e.g. offset of jited_prog_insns */
8853         int     count_offset;   /* e.g. offset of jited_prog_len */
8854         int     size_offset;    /* > 0: offset of rec size,
8855                                  * < 0: fix size of -size_offset
8856                                  */
8857 };
8858
8859 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
8860         [BPF_PROG_INFO_JITED_INSNS] = {
8861                 offsetof(struct bpf_prog_info, jited_prog_insns),
8862                 offsetof(struct bpf_prog_info, jited_prog_len),
8863                 -1,
8864         },
8865         [BPF_PROG_INFO_XLATED_INSNS] = {
8866                 offsetof(struct bpf_prog_info, xlated_prog_insns),
8867                 offsetof(struct bpf_prog_info, xlated_prog_len),
8868                 -1,
8869         },
8870         [BPF_PROG_INFO_MAP_IDS] = {
8871                 offsetof(struct bpf_prog_info, map_ids),
8872                 offsetof(struct bpf_prog_info, nr_map_ids),
8873                 -(int)sizeof(__u32),
8874         },
8875         [BPF_PROG_INFO_JITED_KSYMS] = {
8876                 offsetof(struct bpf_prog_info, jited_ksyms),
8877                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
8878                 -(int)sizeof(__u64),
8879         },
8880         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
8881                 offsetof(struct bpf_prog_info, jited_func_lens),
8882                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
8883                 -(int)sizeof(__u32),
8884         },
8885         [BPF_PROG_INFO_FUNC_INFO] = {
8886                 offsetof(struct bpf_prog_info, func_info),
8887                 offsetof(struct bpf_prog_info, nr_func_info),
8888                 offsetof(struct bpf_prog_info, func_info_rec_size),
8889         },
8890         [BPF_PROG_INFO_LINE_INFO] = {
8891                 offsetof(struct bpf_prog_info, line_info),
8892                 offsetof(struct bpf_prog_info, nr_line_info),
8893                 offsetof(struct bpf_prog_info, line_info_rec_size),
8894         },
8895         [BPF_PROG_INFO_JITED_LINE_INFO] = {
8896                 offsetof(struct bpf_prog_info, jited_line_info),
8897                 offsetof(struct bpf_prog_info, nr_jited_line_info),
8898                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
8899         },
8900         [BPF_PROG_INFO_PROG_TAGS] = {
8901                 offsetof(struct bpf_prog_info, prog_tags),
8902                 offsetof(struct bpf_prog_info, nr_prog_tags),
8903                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
8904         },
8905
8906 };
8907
8908 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
8909                                            int offset)
8910 {
8911         __u32 *array = (__u32 *)info;
8912
8913         if (offset >= 0)
8914                 return array[offset / sizeof(__u32)];
8915         return -(int)offset;
8916 }
8917
8918 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
8919                                            int offset)
8920 {
8921         __u64 *array = (__u64 *)info;
8922
8923         if (offset >= 0)
8924                 return array[offset / sizeof(__u64)];
8925         return -(int)offset;
8926 }
8927
8928 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
8929                                          __u32 val)
8930 {
8931         __u32 *array = (__u32 *)info;
8932
8933         if (offset >= 0)
8934                 array[offset / sizeof(__u32)] = val;
8935 }
8936
8937 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
8938                                          __u64 val)
8939 {
8940         __u64 *array = (__u64 *)info;
8941
8942         if (offset >= 0)
8943                 array[offset / sizeof(__u64)] = val;
8944 }
8945
8946 struct bpf_prog_info_linear *
8947 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
8948 {
8949         struct bpf_prog_info_linear *info_linear;
8950         struct bpf_prog_info info = {};
8951         __u32 info_len = sizeof(info);
8952         __u32 data_len = 0;
8953         int i, err;
8954         void *ptr;
8955
8956         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
8957                 return ERR_PTR(-EINVAL);
8958
8959         /* step 1: get array dimensions */
8960         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
8961         if (err) {
8962                 pr_debug("can't get prog info: %s", strerror(errno));
8963                 return ERR_PTR(-EFAULT);
8964         }
8965
8966         /* step 2: calculate total size of all arrays */
8967         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
8968                 bool include_array = (arrays & (1UL << i)) > 0;
8969                 struct bpf_prog_info_array_desc *desc;
8970                 __u32 count, size;
8971
8972                 desc = bpf_prog_info_array_desc + i;
8973
8974                 /* kernel is too old to support this field */
8975                 if (info_len < desc->array_offset + sizeof(__u32) ||
8976                     info_len < desc->count_offset + sizeof(__u32) ||
8977                     (desc->size_offset > 0 && info_len < desc->size_offset))
8978                         include_array = false;
8979
8980                 if (!include_array) {
8981                         arrays &= ~(1UL << i);  /* clear the bit */
8982                         continue;
8983                 }
8984
8985                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
8986                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
8987
8988                 data_len += count * size;
8989         }
8990
8991         /* step 3: allocate continuous memory */
8992         data_len = roundup(data_len, sizeof(__u64));
8993         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
8994         if (!info_linear)
8995                 return ERR_PTR(-ENOMEM);
8996
8997         /* step 4: fill data to info_linear->info */
8998         info_linear->arrays = arrays;
8999         memset(&info_linear->info, 0, sizeof(info));
9000         ptr = info_linear->data;
9001
9002         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9003                 struct bpf_prog_info_array_desc *desc;
9004                 __u32 count, size;
9005
9006                 if ((arrays & (1UL << i)) == 0)
9007                         continue;
9008
9009                 desc  = bpf_prog_info_array_desc + i;
9010                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
9011                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
9012                 bpf_prog_info_set_offset_u32(&info_linear->info,
9013                                              desc->count_offset, count);
9014                 bpf_prog_info_set_offset_u32(&info_linear->info,
9015                                              desc->size_offset, size);
9016                 bpf_prog_info_set_offset_u64(&info_linear->info,
9017                                              desc->array_offset,
9018                                              ptr_to_u64(ptr));
9019                 ptr += count * size;
9020         }
9021
9022         /* step 5: call syscall again to get required arrays */
9023         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
9024         if (err) {
9025                 pr_debug("can't get prog info: %s", strerror(errno));
9026                 free(info_linear);
9027                 return ERR_PTR(-EFAULT);
9028         }
9029
9030         /* step 6: verify the data */
9031         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9032                 struct bpf_prog_info_array_desc *desc;
9033                 __u32 v1, v2;
9034
9035                 if ((arrays & (1UL << i)) == 0)
9036                         continue;
9037
9038                 desc = bpf_prog_info_array_desc + i;
9039                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
9040                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
9041                                                    desc->count_offset);
9042                 if (v1 != v2)
9043                         pr_warn("%s: mismatch in element count\n", __func__);
9044
9045                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
9046                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
9047                                                    desc->size_offset);
9048                 if (v1 != v2)
9049                         pr_warn("%s: mismatch in rec size\n", __func__);
9050         }
9051
9052         /* step 7: update info_len and data_len */
9053         info_linear->info_len = sizeof(struct bpf_prog_info);
9054         info_linear->data_len = data_len;
9055
9056         return info_linear;
9057 }
9058
9059 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
9060 {
9061         int i;
9062
9063         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
9064                 struct bpf_prog_info_array_desc *desc;
9065                 __u64 addr, offs;
9066
9067                 if ((info_linear->arrays & (1UL << i)) == 0)
9068                         continue;
9069
9070                 desc = bpf_prog_info_array_desc + i;
9071                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
9072                                                      desc->array_offset);
9073                 offs = addr - ptr_to_u64(info_linear->data);
9074                 bpf_prog_info_set_offset_u64(&info_linear->info,
9075                                              desc->array_offset, offs);
9076         }
9077 }
9078
9079 void bpf_program__bpil_offs_to_addr(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                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
9092                                                      desc->array_offset);
9093                 addr = offs + ptr_to_u64(info_linear->data);
9094                 bpf_prog_info_set_offset_u64(&info_linear->info,
9095                                              desc->array_offset, addr);
9096         }
9097 }
9098
9099 int bpf_program__set_attach_target(struct bpf_program *prog,
9100                                    int attach_prog_fd,
9101                                    const char *attach_func_name)
9102 {
9103         int btf_id;
9104
9105         if (!prog || attach_prog_fd < 0 || !attach_func_name)
9106                 return -EINVAL;
9107
9108         if (attach_prog_fd)
9109                 btf_id = libbpf_find_prog_btf_id(attach_func_name,
9110                                                  attach_prog_fd);
9111         else
9112                 btf_id = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
9113                                                attach_func_name,
9114                                                prog->expected_attach_type);
9115
9116         if (btf_id < 0)
9117                 return btf_id;
9118
9119         prog->attach_btf_id = btf_id;
9120         prog->attach_prog_fd = attach_prog_fd;
9121         return 0;
9122 }
9123
9124 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
9125 {
9126         int err = 0, n, len, start, end = -1;
9127         bool *tmp;
9128
9129         *mask = NULL;
9130         *mask_sz = 0;
9131
9132         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
9133         while (*s) {
9134                 if (*s == ',' || *s == '\n') {
9135                         s++;
9136                         continue;
9137                 }
9138                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
9139                 if (n <= 0 || n > 2) {
9140                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
9141                         err = -EINVAL;
9142                         goto cleanup;
9143                 } else if (n == 1) {
9144                         end = start;
9145                 }
9146                 if (start < 0 || start > end) {
9147                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
9148                                 start, end, s);
9149                         err = -EINVAL;
9150                         goto cleanup;
9151                 }
9152                 tmp = realloc(*mask, end + 1);
9153                 if (!tmp) {
9154                         err = -ENOMEM;
9155                         goto cleanup;
9156                 }
9157                 *mask = tmp;
9158                 memset(tmp + *mask_sz, 0, start - *mask_sz);
9159                 memset(tmp + start, 1, end - start + 1);
9160                 *mask_sz = end + 1;
9161                 s += len;
9162         }
9163         if (!*mask_sz) {
9164                 pr_warn("Empty CPU range\n");
9165                 return -EINVAL;
9166         }
9167         return 0;
9168 cleanup:
9169         free(*mask);
9170         *mask = NULL;
9171         return err;
9172 }
9173
9174 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
9175 {
9176         int fd, err = 0, len;
9177         char buf[128];
9178
9179         fd = open(fcpu, O_RDONLY);
9180         if (fd < 0) {
9181                 err = -errno;
9182                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
9183                 return err;
9184         }
9185         len = read(fd, buf, sizeof(buf));
9186         close(fd);
9187         if (len <= 0) {
9188                 err = len ? -errno : -EINVAL;
9189                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
9190                 return err;
9191         }
9192         if (len >= sizeof(buf)) {
9193                 pr_warn("CPU mask is too big in file %s\n", fcpu);
9194                 return -E2BIG;
9195         }
9196         buf[len] = '\0';
9197
9198         return parse_cpu_mask_str(buf, mask, mask_sz);
9199 }
9200
9201 int libbpf_num_possible_cpus(void)
9202 {
9203         static const char *fcpu = "/sys/devices/system/cpu/possible";
9204         static int cpus;
9205         int err, n, i, tmp_cpus;
9206         bool *mask;
9207
9208         tmp_cpus = READ_ONCE(cpus);
9209         if (tmp_cpus > 0)
9210                 return tmp_cpus;
9211
9212         err = parse_cpu_mask_file(fcpu, &mask, &n);
9213         if (err)
9214                 return err;
9215
9216         tmp_cpus = 0;
9217         for (i = 0; i < n; i++) {
9218                 if (mask[i])
9219                         tmp_cpus++;
9220         }
9221         free(mask);
9222
9223         WRITE_ONCE(cpus, tmp_cpus);
9224         return tmp_cpus;
9225 }
9226
9227 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
9228                               const struct bpf_object_open_opts *opts)
9229 {
9230         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
9231                 .object_name = s->name,
9232         );
9233         struct bpf_object *obj;
9234         int i;
9235
9236         /* Attempt to preserve opts->object_name, unless overriden by user
9237          * explicitly. Overwriting object name for skeletons is discouraged,
9238          * as it breaks global data maps, because they contain object name
9239          * prefix as their own map name prefix. When skeleton is generated,
9240          * bpftool is making an assumption that this name will stay the same.
9241          */
9242         if (opts) {
9243                 memcpy(&skel_opts, opts, sizeof(*opts));
9244                 if (!opts->object_name)
9245                         skel_opts.object_name = s->name;
9246         }
9247
9248         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
9249         if (IS_ERR(obj)) {
9250                 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
9251                         s->name, PTR_ERR(obj));
9252                 return PTR_ERR(obj);
9253         }
9254
9255         *s->obj = obj;
9256
9257         for (i = 0; i < s->map_cnt; i++) {
9258                 struct bpf_map **map = s->maps[i].map;
9259                 const char *name = s->maps[i].name;
9260                 void **mmaped = s->maps[i].mmaped;
9261
9262                 *map = bpf_object__find_map_by_name(obj, name);
9263                 if (!*map) {
9264                         pr_warn("failed to find skeleton map '%s'\n", name);
9265                         return -ESRCH;
9266                 }
9267
9268                 /* externs shouldn't be pre-setup from user code */
9269                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
9270                         *mmaped = (*map)->mmaped;
9271         }
9272
9273         for (i = 0; i < s->prog_cnt; i++) {
9274                 struct bpf_program **prog = s->progs[i].prog;
9275                 const char *name = s->progs[i].name;
9276
9277                 *prog = bpf_object__find_program_by_name(obj, name);
9278                 if (!*prog) {
9279                         pr_warn("failed to find skeleton program '%s'\n", name);
9280                         return -ESRCH;
9281                 }
9282         }
9283
9284         return 0;
9285 }
9286
9287 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
9288 {
9289         int i, err;
9290
9291         err = bpf_object__load(*s->obj);
9292         if (err) {
9293                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
9294                 return err;
9295         }
9296
9297         for (i = 0; i < s->map_cnt; i++) {
9298                 struct bpf_map *map = *s->maps[i].map;
9299                 size_t mmap_sz = bpf_map_mmap_sz(map);
9300                 int prot, map_fd = bpf_map__fd(map);
9301                 void **mmaped = s->maps[i].mmaped;
9302
9303                 if (!mmaped)
9304                         continue;
9305
9306                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
9307                         *mmaped = NULL;
9308                         continue;
9309                 }
9310
9311                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
9312                         prot = PROT_READ;
9313                 else
9314                         prot = PROT_READ | PROT_WRITE;
9315
9316                 /* Remap anonymous mmap()-ed "map initialization image" as
9317                  * a BPF map-backed mmap()-ed memory, but preserving the same
9318                  * memory address. This will cause kernel to change process'
9319                  * page table to point to a different piece of kernel memory,
9320                  * but from userspace point of view memory address (and its
9321                  * contents, being identical at this point) will stay the
9322                  * same. This mapping will be released by bpf_object__close()
9323                  * as per normal clean up procedure, so we don't need to worry
9324                  * about it from skeleton's clean up perspective.
9325                  */
9326                 *mmaped = mmap(map->mmaped, mmap_sz, prot,
9327                                 MAP_SHARED | MAP_FIXED, map_fd, 0);
9328                 if (*mmaped == MAP_FAILED) {
9329                         err = -errno;
9330                         *mmaped = NULL;
9331                         pr_warn("failed to re-mmap() map '%s': %d\n",
9332                                  bpf_map__name(map), err);
9333                         return err;
9334                 }
9335         }
9336
9337         return 0;
9338 }
9339
9340 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
9341 {
9342         int i;
9343
9344         for (i = 0; i < s->prog_cnt; i++) {
9345                 struct bpf_program *prog = *s->progs[i].prog;
9346                 struct bpf_link **link = s->progs[i].link;
9347                 const struct bpf_sec_def *sec_def;
9348                 const char *sec_name = bpf_program__title(prog, false);
9349
9350                 if (!prog->load)
9351                         continue;
9352
9353                 sec_def = find_sec_def(sec_name);
9354                 if (!sec_def || !sec_def->attach_fn)
9355                         continue;
9356
9357                 *link = sec_def->attach_fn(sec_def, prog);
9358                 if (IS_ERR(*link)) {
9359                         pr_warn("failed to auto-attach program '%s': %ld\n",
9360                                 bpf_program__name(prog), PTR_ERR(*link));
9361                         return PTR_ERR(*link);
9362                 }
9363         }
9364
9365         return 0;
9366 }
9367
9368 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
9369 {
9370         int i;
9371
9372         for (i = 0; i < s->prog_cnt; i++) {
9373                 struct bpf_link **link = s->progs[i].link;
9374
9375                 if (!IS_ERR_OR_NULL(*link))
9376                         bpf_link__destroy(*link);
9377                 *link = NULL;
9378         }
9379 }
9380
9381 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
9382 {
9383         if (s->progs)
9384                 bpf_object__detach_skeleton(s);
9385         if (s->obj)
9386                 bpf_object__close(*s->obj);
9387         free(s->maps);
9388         free(s->progs);
9389         free(s);
9390 }