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