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