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 <limits.h>
  22#include <string.h>
  23#include <unistd.h>
  24#include <endian.h>
  25#include <fcntl.h>
  26#include <errno.h>
  27#include <ctype.h>
  28#include <asm/unistd.h>
  29#include <linux/err.h>
  30#include <linux/kernel.h>
  31#include <linux/bpf.h>
  32#include <linux/btf.h>
  33#include <linux/filter.h>
  34#include <linux/list.h>
  35#include <linux/limits.h>
  36#include <linux/perf_event.h>
  37#include <linux/ring_buffer.h>
  38#include <linux/version.h>
  39#include <sys/epoll.h>
  40#include <sys/ioctl.h>
  41#include <sys/mman.h>
  42#include <sys/stat.h>
  43#include <sys/types.h>
  44#include <sys/vfs.h>
  45#include <sys/utsname.h>
  46#include <sys/resource.h>
  47#include <libelf.h>
  48#include <gelf.h>
  49#include <zlib.h>
  50
  51#include "libbpf.h"
  52#include "bpf.h"
  53#include "btf.h"
  54#include "str_error.h"
  55#include "libbpf_internal.h"
  56#include "hashmap.h"
  57#include "bpf_gen_internal.h"
  58
  59#ifndef BPF_FS_MAGIC
  60#define BPF_FS_MAGIC            0xcafe4a11
  61#endif
  62
  63#define BPF_INSN_SZ (sizeof(struct bpf_insn))
  64
  65/* vsprintf() in __base_pr() uses nonliteral format string. It may break
  66 * compilation if user enables corresponding warning. Disable it explicitly.
  67 */
  68#pragma GCC diagnostic ignored "-Wformat-nonliteral"
  69
  70#define __printf(a, b)  __attribute__((format(printf, a, b)))
  71
  72static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
  73static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
  74
  75static int __base_pr(enum libbpf_print_level level, const char *format,
  76                     va_list args)
  77{
  78        if (level == LIBBPF_DEBUG)
  79                return 0;
  80
  81        return vfprintf(stderr, format, args);
  82}
  83
  84static libbpf_print_fn_t __libbpf_pr = __base_pr;
  85
  86libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  87{
  88        libbpf_print_fn_t old_print_fn = __libbpf_pr;
  89
  90        __libbpf_pr = fn;
  91        return old_print_fn;
  92}
  93
  94__printf(2, 3)
  95void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  96{
  97        va_list args;
  98
  99        if (!__libbpf_pr)
 100                return;
 101
 102        va_start(args, format);
 103        __libbpf_pr(level, format, args);
 104        va_end(args);
 105}
 106
 107static void pr_perm_msg(int err)
 108{
 109        struct rlimit limit;
 110        char buf[100];
 111
 112        if (err != -EPERM || geteuid() != 0)
 113                return;
 114
 115        err = getrlimit(RLIMIT_MEMLOCK, &limit);
 116        if (err)
 117                return;
 118
 119        if (limit.rlim_cur == RLIM_INFINITY)
 120                return;
 121
 122        if (limit.rlim_cur < 1024)
 123                snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
 124        else if (limit.rlim_cur < 1024*1024)
 125                snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
 126        else
 127                snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
 128
 129        pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
 130                buf);
 131}
 132
 133#define STRERR_BUFSIZE  128
 134
 135/* Copied from tools/perf/util/util.h */
 136#ifndef zfree
 137# define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 138#endif
 139
 140#ifndef zclose
 141# define zclose(fd) ({                  \
 142        int ___err = 0;                 \
 143        if ((fd) >= 0)                  \
 144                ___err = close((fd));   \
 145        fd = -1;                        \
 146        ___err; })
 147#endif
 148
 149static inline __u64 ptr_to_u64(const void *ptr)
 150{
 151        return (__u64) (unsigned long) ptr;
 152}
 153
 154/* this goes away in libbpf 1.0 */
 155enum libbpf_strict_mode libbpf_mode = LIBBPF_STRICT_NONE;
 156
 157int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
 158{
 159        libbpf_mode = mode;
 160        return 0;
 161}
 162
 163__u32 libbpf_major_version(void)
 164{
 165        return LIBBPF_MAJOR_VERSION;
 166}
 167
 168__u32 libbpf_minor_version(void)
 169{
 170        return LIBBPF_MINOR_VERSION;
 171}
 172
 173const char *libbpf_version_string(void)
 174{
 175#define __S(X) #X
 176#define _S(X) __S(X)
 177        return  "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
 178#undef _S
 179#undef __S
 180}
 181
 182enum reloc_type {
 183        RELO_LD64,
 184        RELO_CALL,
 185        RELO_DATA,
 186        RELO_EXTERN_VAR,
 187        RELO_EXTERN_FUNC,
 188        RELO_SUBPROG_ADDR,
 189        RELO_CORE,
 190};
 191
 192struct reloc_desc {
 193        enum reloc_type type;
 194        int insn_idx;
 195        union {
 196                const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
 197                struct {
 198                        int map_idx;
 199                        int sym_off;
 200                };
 201        };
 202};
 203
 204/* stored as sec_def->cookie for all libbpf-supported SEC()s */
 205enum sec_def_flags {
 206        SEC_NONE = 0,
 207        /* expected_attach_type is optional, if kernel doesn't support that */
 208        SEC_EXP_ATTACH_OPT = 1,
 209        /* legacy, only used by libbpf_get_type_names() and
 210         * libbpf_attach_type_by_name(), not used by libbpf itself at all.
 211         * This used to be associated with cgroup (and few other) BPF programs
 212         * that were attachable through BPF_PROG_ATTACH command. Pretty
 213         * meaningless nowadays, though.
 214         */
 215        SEC_ATTACHABLE = 2,
 216        SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
 217        /* attachment target is specified through BTF ID in either kernel or
 218         * other BPF program's BTF object */
 219        SEC_ATTACH_BTF = 4,
 220        /* BPF program type allows sleeping/blocking in kernel */
 221        SEC_SLEEPABLE = 8,
 222        /* allow non-strict prefix matching */
 223        SEC_SLOPPY_PFX = 16,
 224        /* BPF program support non-linear XDP buffer */
 225        SEC_XDP_FRAGS = 32,
 226        /* deprecated sec definitions not supposed to be used */
 227        SEC_DEPRECATED = 64,
 228};
 229
 230struct bpf_sec_def {
 231        char *sec;
 232        enum bpf_prog_type prog_type;
 233        enum bpf_attach_type expected_attach_type;
 234        long cookie;
 235        int handler_id;
 236
 237        libbpf_prog_setup_fn_t prog_setup_fn;
 238        libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
 239        libbpf_prog_attach_fn_t prog_attach_fn;
 240};
 241
 242/*
 243 * bpf_prog should be a better name but it has been used in
 244 * linux/filter.h.
 245 */
 246struct bpf_program {
 247        const struct bpf_sec_def *sec_def;
 248        char *sec_name;
 249        size_t sec_idx;
 250        /* this program's instruction offset (in number of instructions)
 251         * within its containing ELF section
 252         */
 253        size_t sec_insn_off;
 254        /* number of original instructions in ELF section belonging to this
 255         * program, not taking into account subprogram instructions possible
 256         * appended later during relocation
 257         */
 258        size_t sec_insn_cnt;
 259        /* Offset (in number of instructions) of the start of instruction
 260         * belonging to this BPF program  within its containing main BPF
 261         * program. For the entry-point (main) BPF program, this is always
 262         * zero. For a sub-program, this gets reset before each of main BPF
 263         * programs are processed and relocated and is used to determined
 264         * whether sub-program was already appended to the main program, and
 265         * if yes, at which instruction offset.
 266         */
 267        size_t sub_insn_off;
 268
 269        char *name;
 270        /* name with / replaced by _; makes recursive pinning
 271         * in bpf_object__pin_programs easier
 272         */
 273        char *pin_name;
 274
 275        /* instructions that belong to BPF program; insns[0] is located at
 276         * sec_insn_off instruction within its ELF section in ELF file, so
 277         * when mapping ELF file instruction index to the local instruction,
 278         * one needs to subtract sec_insn_off; and vice versa.
 279         */
 280        struct bpf_insn *insns;
 281        /* actual number of instruction in this BPF program's image; for
 282         * entry-point BPF programs this includes the size of main program
 283         * itself plus all the used sub-programs, appended at the end
 284         */
 285        size_t insns_cnt;
 286
 287        struct reloc_desc *reloc_desc;
 288        int nr_reloc;
 289
 290        /* BPF verifier log settings */
 291        char *log_buf;
 292        size_t log_size;
 293        __u32 log_level;
 294
 295        struct {
 296                int nr;
 297                int *fds;
 298        } instances;
 299        bpf_program_prep_t preprocessor;
 300
 301        struct bpf_object *obj;
 302        void *priv;
 303        bpf_program_clear_priv_t clear_priv;
 304
 305        bool load;
 306        bool mark_btf_static;
 307        enum bpf_prog_type type;
 308        enum bpf_attach_type expected_attach_type;
 309        int prog_ifindex;
 310        __u32 attach_btf_obj_fd;
 311        __u32 attach_btf_id;
 312        __u32 attach_prog_fd;
 313        void *func_info;
 314        __u32 func_info_rec_size;
 315        __u32 func_info_cnt;
 316
 317        void *line_info;
 318        __u32 line_info_rec_size;
 319        __u32 line_info_cnt;
 320        __u32 prog_flags;
 321};
 322
 323struct bpf_struct_ops {
 324        const char *tname;
 325        const struct btf_type *type;
 326        struct bpf_program **progs;
 327        __u32 *kern_func_off;
 328        /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
 329        void *data;
 330        /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
 331         *      btf_vmlinux's format.
 332         * struct bpf_struct_ops_tcp_congestion_ops {
 333         *      [... some other kernel fields ...]
 334         *      struct tcp_congestion_ops data;
 335         * }
 336         * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
 337         * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
 338         * from "data".
 339         */
 340        void *kern_vdata;
 341        __u32 type_id;
 342};
 343
 344#define DATA_SEC ".data"
 345#define BSS_SEC ".bss"
 346#define RODATA_SEC ".rodata"
 347#define KCONFIG_SEC ".kconfig"
 348#define KSYMS_SEC ".ksyms"
 349#define STRUCT_OPS_SEC ".struct_ops"
 350
 351enum libbpf_map_type {
 352        LIBBPF_MAP_UNSPEC,
 353        LIBBPF_MAP_DATA,
 354        LIBBPF_MAP_BSS,
 355        LIBBPF_MAP_RODATA,
 356        LIBBPF_MAP_KCONFIG,
 357};
 358
 359struct bpf_map {
 360        char *name;
 361        /* real_name is defined for special internal maps (.rodata*,
 362         * .data*, .bss, .kconfig) and preserves their original ELF section
 363         * name. This is important to be be able to find corresponding BTF
 364         * DATASEC information.
 365         */
 366        char *real_name;
 367        int fd;
 368        int sec_idx;
 369        size_t sec_offset;
 370        int map_ifindex;
 371        int inner_map_fd;
 372        struct bpf_map_def def;
 373        __u32 numa_node;
 374        __u32 btf_var_idx;
 375        __u32 btf_key_type_id;
 376        __u32 btf_value_type_id;
 377        __u32 btf_vmlinux_value_type_id;
 378        void *priv;
 379        bpf_map_clear_priv_t clear_priv;
 380        enum libbpf_map_type libbpf_type;
 381        void *mmaped;
 382        struct bpf_struct_ops *st_ops;
 383        struct bpf_map *inner_map;
 384        void **init_slots;
 385        int init_slots_sz;
 386        char *pin_path;
 387        bool pinned;
 388        bool reused;
 389        bool skipped;
 390        __u64 map_extra;
 391};
 392
 393enum extern_type {
 394        EXT_UNKNOWN,
 395        EXT_KCFG,
 396        EXT_KSYM,
 397};
 398
 399enum kcfg_type {
 400        KCFG_UNKNOWN,
 401        KCFG_CHAR,
 402        KCFG_BOOL,
 403        KCFG_INT,
 404        KCFG_TRISTATE,
 405        KCFG_CHAR_ARR,
 406};
 407
 408struct extern_desc {
 409        enum extern_type type;
 410        int sym_idx;
 411        int btf_id;
 412        int sec_btf_id;
 413        const char *name;
 414        bool is_set;
 415        bool is_weak;
 416        union {
 417                struct {
 418                        enum kcfg_type type;
 419                        int sz;
 420                        int align;
 421                        int data_off;
 422                        bool is_signed;
 423                } kcfg;
 424                struct {
 425                        unsigned long long addr;
 426
 427                        /* target btf_id of the corresponding kernel var. */
 428                        int kernel_btf_obj_fd;
 429                        int kernel_btf_id;
 430
 431                        /* local btf_id of the ksym extern's type. */
 432                        __u32 type_id;
 433                        /* BTF fd index to be patched in for insn->off, this is
 434                         * 0 for vmlinux BTF, index in obj->fd_array for module
 435                         * BTF
 436                         */
 437                        __s16 btf_fd_idx;
 438                } ksym;
 439        };
 440};
 441
 442static LIST_HEAD(bpf_objects_list);
 443
 444struct module_btf {
 445        struct btf *btf;
 446        char *name;
 447        __u32 id;
 448        int fd;
 449        int fd_array_idx;
 450};
 451
 452enum sec_type {
 453        SEC_UNUSED = 0,
 454        SEC_RELO,
 455        SEC_BSS,
 456        SEC_DATA,
 457        SEC_RODATA,
 458};
 459
 460struct elf_sec_desc {
 461        enum sec_type sec_type;
 462        Elf64_Shdr *shdr;
 463        Elf_Data *data;
 464};
 465
 466struct elf_state {
 467        int fd;
 468        const void *obj_buf;
 469        size_t obj_buf_sz;
 470        Elf *elf;
 471        Elf64_Ehdr *ehdr;
 472        Elf_Data *symbols;
 473        Elf_Data *st_ops_data;
 474        size_t shstrndx; /* section index for section name strings */
 475        size_t strtabidx;
 476        struct elf_sec_desc *secs;
 477        int sec_cnt;
 478        int maps_shndx;
 479        int btf_maps_shndx;
 480        __u32 btf_maps_sec_btf_id;
 481        int text_shndx;
 482        int symbols_shndx;
 483        int st_ops_shndx;
 484};
 485
 486struct bpf_object {
 487        char name[BPF_OBJ_NAME_LEN];
 488        char license[64];
 489        __u32 kern_version;
 490
 491        struct bpf_program *programs;
 492        size_t nr_programs;
 493        struct bpf_map *maps;
 494        size_t nr_maps;
 495        size_t maps_cap;
 496
 497        char *kconfig;
 498        struct extern_desc *externs;
 499        int nr_extern;
 500        int kconfig_map_idx;
 501
 502        bool loaded;
 503        bool has_subcalls;
 504        bool has_rodata;
 505
 506        struct bpf_gen *gen_loader;
 507
 508        /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
 509        struct elf_state efile;
 510        /*
 511         * All loaded bpf_object are linked in a list, which is
 512         * hidden to caller. bpf_objects__<func> handlers deal with
 513         * all objects.
 514         */
 515        struct list_head list;
 516
 517        struct btf *btf;
 518        struct btf_ext *btf_ext;
 519
 520        /* Parse and load BTF vmlinux if any of the programs in the object need
 521         * it at load time.
 522         */
 523        struct btf *btf_vmlinux;
 524        /* Path to the custom BTF to be used for BPF CO-RE relocations as an
 525         * override for vmlinux BTF.
 526         */
 527        char *btf_custom_path;
 528        /* vmlinux BTF override for CO-RE relocations */
 529        struct btf *btf_vmlinux_override;
 530        /* Lazily initialized kernel module BTFs */
 531        struct module_btf *btf_modules;
 532        bool btf_modules_loaded;
 533        size_t btf_module_cnt;
 534        size_t btf_module_cap;
 535
 536        /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
 537        char *log_buf;
 538        size_t log_size;
 539        __u32 log_level;
 540
 541        void *priv;
 542        bpf_object_clear_priv_t clear_priv;
 543
 544        int *fd_array;
 545        size_t fd_array_cap;
 546        size_t fd_array_cnt;
 547
 548        char path[];
 549};
 550
 551static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
 552static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
 553static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
 554static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
 555static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
 556static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
 557static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
 558static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
 559static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
 560
 561void bpf_program__unload(struct bpf_program *prog)
 562{
 563        int i;
 564
 565        if (!prog)
 566                return;
 567
 568        /*
 569         * If the object is opened but the program was never loaded,
 570         * it is possible that prog->instances.nr == -1.
 571         */
 572        if (prog->instances.nr > 0) {
 573                for (i = 0; i < prog->instances.nr; i++)
 574                        zclose(prog->instances.fds[i]);
 575        } else if (prog->instances.nr != -1) {
 576                pr_warn("Internal error: instances.nr is %d\n",
 577                        prog->instances.nr);
 578        }
 579
 580        prog->instances.nr = -1;
 581        zfree(&prog->instances.fds);
 582
 583        zfree(&prog->func_info);
 584        zfree(&prog->line_info);
 585}
 586
 587static void bpf_program__exit(struct bpf_program *prog)
 588{
 589        if (!prog)
 590                return;
 591
 592        if (prog->clear_priv)
 593                prog->clear_priv(prog, prog->priv);
 594
 595        prog->priv = NULL;
 596        prog->clear_priv = NULL;
 597
 598        bpf_program__unload(prog);
 599        zfree(&prog->name);
 600        zfree(&prog->sec_name);
 601        zfree(&prog->pin_name);
 602        zfree(&prog->insns);
 603        zfree(&prog->reloc_desc);
 604
 605        prog->nr_reloc = 0;
 606        prog->insns_cnt = 0;
 607        prog->sec_idx = -1;
 608}
 609
 610static char *__bpf_program__pin_name(struct bpf_program *prog)
 611{
 612        char *name, *p;
 613
 614        if (libbpf_mode & LIBBPF_STRICT_SEC_NAME)
 615                name = strdup(prog->name);
 616        else
 617                name = strdup(prog->sec_name);
 618
 619        if (!name)
 620                return NULL;
 621
 622        p = name;
 623
 624        while ((p = strchr(p, '/')))
 625                *p = '_';
 626
 627        return name;
 628}
 629
 630static bool insn_is_subprog_call(const struct bpf_insn *insn)
 631{
 632        return BPF_CLASS(insn->code) == BPF_JMP &&
 633               BPF_OP(insn->code) == BPF_CALL &&
 634               BPF_SRC(insn->code) == BPF_K &&
 635               insn->src_reg == BPF_PSEUDO_CALL &&
 636               insn->dst_reg == 0 &&
 637               insn->off == 0;
 638}
 639
 640static bool is_call_insn(const struct bpf_insn *insn)
 641{
 642        return insn->code == (BPF_JMP | BPF_CALL);
 643}
 644
 645static bool insn_is_pseudo_func(struct bpf_insn *insn)
 646{
 647        return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
 648}
 649
 650static int
 651bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
 652                      const char *name, size_t sec_idx, const char *sec_name,
 653                      size_t sec_off, void *insn_data, size_t insn_data_sz)
 654{
 655        if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
 656                pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
 657                        sec_name, name, sec_off, insn_data_sz);
 658                return -EINVAL;
 659        }
 660
 661        memset(prog, 0, sizeof(*prog));
 662        prog->obj = obj;
 663
 664        prog->sec_idx = sec_idx;
 665        prog->sec_insn_off = sec_off / BPF_INSN_SZ;
 666        prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
 667        /* insns_cnt can later be increased by appending used subprograms */
 668        prog->insns_cnt = prog->sec_insn_cnt;
 669
 670        prog->type = BPF_PROG_TYPE_UNSPEC;
 671        prog->load = true;
 672
 673        prog->instances.fds = NULL;
 674        prog->instances.nr = -1;
 675
 676        /* inherit object's log_level */
 677        prog->log_level = obj->log_level;
 678
 679        prog->sec_name = strdup(sec_name);
 680        if (!prog->sec_name)
 681                goto errout;
 682
 683        prog->name = strdup(name);
 684        if (!prog->name)
 685                goto errout;
 686
 687        prog->pin_name = __bpf_program__pin_name(prog);
 688        if (!prog->pin_name)
 689                goto errout;
 690
 691        prog->insns = malloc(insn_data_sz);
 692        if (!prog->insns)
 693                goto errout;
 694        memcpy(prog->insns, insn_data, insn_data_sz);
 695
 696        return 0;
 697errout:
 698        pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
 699        bpf_program__exit(prog);
 700        return -ENOMEM;
 701}
 702
 703static int
 704bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
 705                         const char *sec_name, int sec_idx)
 706{
 707        Elf_Data *symbols = obj->efile.symbols;
 708        struct bpf_program *prog, *progs;
 709        void *data = sec_data->d_buf;
 710        size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
 711        int nr_progs, err, i;
 712        const char *name;
 713        Elf64_Sym *sym;
 714
 715        progs = obj->programs;
 716        nr_progs = obj->nr_programs;
 717        nr_syms = symbols->d_size / sizeof(Elf64_Sym);
 718        sec_off = 0;
 719
 720        for (i = 0; i < nr_syms; i++) {
 721                sym = elf_sym_by_idx(obj, i);
 722
 723                if (sym->st_shndx != sec_idx)
 724                        continue;
 725                if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
 726                        continue;
 727
 728                prog_sz = sym->st_size;
 729                sec_off = sym->st_value;
 730
 731                name = elf_sym_str(obj, sym->st_name);
 732                if (!name) {
 733                        pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
 734                                sec_name, sec_off);
 735                        return -LIBBPF_ERRNO__FORMAT;
 736                }
 737
 738                if (sec_off + prog_sz > sec_sz) {
 739                        pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
 740                                sec_name, sec_off);
 741                        return -LIBBPF_ERRNO__FORMAT;
 742                }
 743
 744                if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
 745                        pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
 746                        return -ENOTSUP;
 747                }
 748
 749                pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
 750                         sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
 751
 752                progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
 753                if (!progs) {
 754                        /*
 755                         * In this case the original obj->programs
 756                         * is still valid, so don't need special treat for
 757                         * bpf_close_object().
 758                         */
 759                        pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
 760                                sec_name, name);
 761                        return -ENOMEM;
 762                }
 763                obj->programs = progs;
 764
 765                prog = &progs[nr_progs];
 766
 767                err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
 768                                            sec_off, data + sec_off, prog_sz);
 769                if (err)
 770                        return err;
 771
 772                /* if function is a global/weak symbol, but has restricted
 773                 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
 774                 * as static to enable more permissive BPF verification mode
 775                 * with more outside context available to BPF verifier
 776                 */
 777                if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL
 778                    && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
 779                        || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
 780                        prog->mark_btf_static = true;
 781
 782                nr_progs++;
 783                obj->nr_programs = nr_progs;
 784        }
 785
 786        return 0;
 787}
 788
 789__u32 get_kernel_version(void)
 790{
 791        /* On Ubuntu LINUX_VERSION_CODE doesn't correspond to info.release,
 792         * but Ubuntu provides /proc/version_signature file, as described at
 793         * https://ubuntu.com/kernel, with an example contents below, which we
 794         * can use to get a proper LINUX_VERSION_CODE.
 795         *
 796         *   Ubuntu 5.4.0-12.15-generic 5.4.8
 797         *
 798         * In the above, 5.4.8 is what kernel is actually expecting, while
 799         * uname() call will return 5.4.0 in info.release.
 800         */
 801        const char *ubuntu_kver_file = "/proc/version_signature";
 802        __u32 major, minor, patch;
 803        struct utsname info;
 804
 805        if (access(ubuntu_kver_file, R_OK) == 0) {
 806                FILE *f;
 807
 808                f = fopen(ubuntu_kver_file, "r");
 809                if (f) {
 810                        if (fscanf(f, "%*s %*s %d.%d.%d\n", &major, &minor, &patch) == 3) {
 811                                fclose(f);
 812                                return KERNEL_VERSION(major, minor, patch);
 813                        }
 814                        fclose(f);
 815                }
 816                /* something went wrong, fall back to uname() approach */
 817        }
 818
 819        uname(&info);
 820        if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 821                return 0;
 822        return KERNEL_VERSION(major, minor, patch);
 823}
 824
 825static const struct btf_member *
 826find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
 827{
 828        struct btf_member *m;
 829        int i;
 830
 831        for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 832                if (btf_member_bit_offset(t, i) == bit_offset)
 833                        return m;
 834        }
 835
 836        return NULL;
 837}
 838
 839static const struct btf_member *
 840find_member_by_name(const struct btf *btf, const struct btf_type *t,
 841                    const char *name)
 842{
 843        struct btf_member *m;
 844        int i;
 845
 846        for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 847                if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
 848                        return m;
 849        }
 850
 851        return NULL;
 852}
 853
 854#define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
 855static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
 856                                   const char *name, __u32 kind);
 857
 858static int
 859find_struct_ops_kern_types(const struct btf *btf, const char *tname,
 860                           const struct btf_type **type, __u32 *type_id,
 861                           const struct btf_type **vtype, __u32 *vtype_id,
 862                           const struct btf_member **data_member)
 863{
 864        const struct btf_type *kern_type, *kern_vtype;
 865        const struct btf_member *kern_data_member;
 866        __s32 kern_vtype_id, kern_type_id;
 867        __u32 i;
 868
 869        kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
 870        if (kern_type_id < 0) {
 871                pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
 872                        tname);
 873                return kern_type_id;
 874        }
 875        kern_type = btf__type_by_id(btf, kern_type_id);
 876
 877        /* Find the corresponding "map_value" type that will be used
 878         * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 879         * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 880         * btf_vmlinux.
 881         */
 882        kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
 883                                                tname, BTF_KIND_STRUCT);
 884        if (kern_vtype_id < 0) {
 885                pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
 886                        STRUCT_OPS_VALUE_PREFIX, tname);
 887                return kern_vtype_id;
 888        }
 889        kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 890
 891        /* Find "struct tcp_congestion_ops" from
 892         * struct bpf_struct_ops_tcp_congestion_ops {
 893         *      [ ... ]
 894         *      struct tcp_congestion_ops data;
 895         * }
 896         */
 897        kern_data_member = btf_members(kern_vtype);
 898        for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
 899                if (kern_data_member->type == kern_type_id)
 900                        break;
 901        }
 902        if (i == btf_vlen(kern_vtype)) {
 903                pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
 904                        tname, STRUCT_OPS_VALUE_PREFIX, tname);
 905                return -EINVAL;
 906        }
 907
 908        *type = kern_type;
 909        *type_id = kern_type_id;
 910        *vtype = kern_vtype;
 911        *vtype_id = kern_vtype_id;
 912        *data_member = kern_data_member;
 913
 914        return 0;
 915}
 916
 917static bool bpf_map__is_struct_ops(const struct bpf_map *map)
 918{
 919        return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
 920}
 921
 922/* Init the map's fields that depend on kern_btf */
 923static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
 924                                         const struct btf *btf,
 925                                         const struct btf *kern_btf)
 926{
 927        const struct btf_member *member, *kern_member, *kern_data_member;
 928        const struct btf_type *type, *kern_type, *kern_vtype;
 929        __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
 930        struct bpf_struct_ops *st_ops;
 931        void *data, *kern_data;
 932        const char *tname;
 933        int err;
 934
 935        st_ops = map->st_ops;
 936        type = st_ops->type;
 937        tname = st_ops->tname;
 938        err = find_struct_ops_kern_types(kern_btf, tname,
 939                                         &kern_type, &kern_type_id,
 940                                         &kern_vtype, &kern_vtype_id,
 941                                         &kern_data_member);
 942        if (err)
 943                return err;
 944
 945        pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
 946                 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
 947
 948        map->def.value_size = kern_vtype->size;
 949        map->btf_vmlinux_value_type_id = kern_vtype_id;
 950
 951        st_ops->kern_vdata = calloc(1, kern_vtype->size);
 952        if (!st_ops->kern_vdata)
 953                return -ENOMEM;
 954
 955        data = st_ops->data;
 956        kern_data_off = kern_data_member->offset / 8;
 957        kern_data = st_ops->kern_vdata + kern_data_off;
 958
 959        member = btf_members(type);
 960        for (i = 0; i < btf_vlen(type); i++, member++) {
 961                const struct btf_type *mtype, *kern_mtype;
 962                __u32 mtype_id, kern_mtype_id;
 963                void *mdata, *kern_mdata;
 964                __s64 msize, kern_msize;
 965                __u32 moff, kern_moff;
 966                __u32 kern_member_idx;
 967                const char *mname;
 968
 969                mname = btf__name_by_offset(btf, member->name_off);
 970                kern_member = find_member_by_name(kern_btf, kern_type, mname);
 971                if (!kern_member) {
 972                        pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
 973                                map->name, mname);
 974                        return -ENOTSUP;
 975                }
 976
 977                kern_member_idx = kern_member - btf_members(kern_type);
 978                if (btf_member_bitfield_size(type, i) ||
 979                    btf_member_bitfield_size(kern_type, kern_member_idx)) {
 980                        pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
 981                                map->name, mname);
 982                        return -ENOTSUP;
 983                }
 984
 985                moff = member->offset / 8;
 986                kern_moff = kern_member->offset / 8;
 987
 988                mdata = data + moff;
 989                kern_mdata = kern_data + kern_moff;
 990
 991                mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
 992                kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
 993                                                    &kern_mtype_id);
 994                if (BTF_INFO_KIND(mtype->info) !=
 995                    BTF_INFO_KIND(kern_mtype->info)) {
 996                        pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
 997                                map->name, mname, BTF_INFO_KIND(mtype->info),
 998                                BTF_INFO_KIND(kern_mtype->info));
 999                        return -ENOTSUP;
1000                }
1001
1002                if (btf_is_ptr(mtype)) {
1003                        struct bpf_program *prog;
1004
1005                        prog = st_ops->progs[i];
1006                        if (!prog)
1007                                continue;
1008
1009                        kern_mtype = skip_mods_and_typedefs(kern_btf,
1010                                                            kern_mtype->type,
1011                                                            &kern_mtype_id);
1012
1013                        /* mtype->type must be a func_proto which was
1014                         * guaranteed in bpf_object__collect_st_ops_relos(),
1015                         * so only check kern_mtype for func_proto here.
1016                         */
1017                        if (!btf_is_func_proto(kern_mtype)) {
1018                                pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1019                                        map->name, mname);
1020                                return -ENOTSUP;
1021                        }
1022
1023                        prog->attach_btf_id = kern_type_id;
1024                        prog->expected_attach_type = kern_member_idx;
1025
1026                        st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1027
1028                        pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1029                                 map->name, mname, prog->name, moff,
1030                                 kern_moff);
1031
1032                        continue;
1033                }
1034
1035                msize = btf__resolve_size(btf, mtype_id);
1036                kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1037                if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
1038                        pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1039                                map->name, mname, (ssize_t)msize,
1040                                (ssize_t)kern_msize);
1041                        return -ENOTSUP;
1042                }
1043
1044                pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1045                         map->name, mname, (unsigned int)msize,
1046                         moff, kern_moff);
1047                memcpy(kern_mdata, mdata, msize);
1048        }
1049
1050        return 0;
1051}
1052
1053static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1054{
1055        struct bpf_map *map;
1056        size_t i;
1057        int err;
1058
1059        for (i = 0; i < obj->nr_maps; i++) {
1060                map = &obj->maps[i];
1061
1062                if (!bpf_map__is_struct_ops(map))
1063                        continue;
1064
1065                err = bpf_map__init_kern_struct_ops(map, obj->btf,
1066                                                    obj->btf_vmlinux);
1067                if (err)
1068                        return err;
1069        }
1070
1071        return 0;
1072}
1073
1074static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
1075{
1076        const struct btf_type *type, *datasec;
1077        const struct btf_var_secinfo *vsi;
1078        struct bpf_struct_ops *st_ops;
1079        const char *tname, *var_name;
1080        __s32 type_id, datasec_id;
1081        const struct btf *btf;
1082        struct bpf_map *map;
1083        __u32 i;
1084
1085        if (obj->efile.st_ops_shndx == -1)
1086                return 0;
1087
1088        btf = obj->btf;
1089        datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
1090                                            BTF_KIND_DATASEC);
1091        if (datasec_id < 0) {
1092                pr_warn("struct_ops init: DATASEC %s not found\n",
1093                        STRUCT_OPS_SEC);
1094                return -EINVAL;
1095        }
1096
1097        datasec = btf__type_by_id(btf, datasec_id);
1098        vsi = btf_var_secinfos(datasec);
1099        for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1100                type = btf__type_by_id(obj->btf, vsi->type);
1101                var_name = btf__name_by_offset(obj->btf, type->name_off);
1102
1103                type_id = btf__resolve_type(obj->btf, vsi->type);
1104                if (type_id < 0) {
1105                        pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1106                                vsi->type, STRUCT_OPS_SEC);
1107                        return -EINVAL;
1108                }
1109
1110                type = btf__type_by_id(obj->btf, type_id);
1111                tname = btf__name_by_offset(obj->btf, type->name_off);
1112                if (!tname[0]) {
1113                        pr_warn("struct_ops init: anonymous type is not supported\n");
1114                        return -ENOTSUP;
1115                }
1116                if (!btf_is_struct(type)) {
1117                        pr_warn("struct_ops init: %s is not a struct\n", tname);
1118                        return -EINVAL;
1119                }
1120
1121                map = bpf_object__add_map(obj);
1122                if (IS_ERR(map))
1123                        return PTR_ERR(map);
1124
1125                map->sec_idx = obj->efile.st_ops_shndx;
1126                map->sec_offset = vsi->offset;
1127                map->name = strdup(var_name);
1128                if (!map->name)
1129                        return -ENOMEM;
1130
1131                map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1132                map->def.key_size = sizeof(int);
1133                map->def.value_size = type->size;
1134                map->def.max_entries = 1;
1135
1136                map->st_ops = calloc(1, sizeof(*map->st_ops));
1137                if (!map->st_ops)
1138                        return -ENOMEM;
1139                st_ops = map->st_ops;
1140                st_ops->data = malloc(type->size);
1141                st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1142                st_ops->kern_func_off = malloc(btf_vlen(type) *
1143                                               sizeof(*st_ops->kern_func_off));
1144                if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1145                        return -ENOMEM;
1146
1147                if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
1148                        pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1149                                var_name, STRUCT_OPS_SEC);
1150                        return -EINVAL;
1151                }
1152
1153                memcpy(st_ops->data,
1154                       obj->efile.st_ops_data->d_buf + vsi->offset,
1155                       type->size);
1156                st_ops->tname = tname;
1157                st_ops->type = type;
1158                st_ops->type_id = type_id;
1159
1160                pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1161                         tname, type_id, var_name, vsi->offset);
1162        }
1163
1164        return 0;
1165}
1166
1167static struct bpf_object *bpf_object__new(const char *path,
1168                                          const void *obj_buf,
1169                                          size_t obj_buf_sz,
1170                                          const char *obj_name)
1171{
1172        bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST);
1173        struct bpf_object *obj;
1174        char *end;
1175
1176        obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1177        if (!obj) {
1178                pr_warn("alloc memory failed for %s\n", path);
1179                return ERR_PTR(-ENOMEM);
1180        }
1181
1182        strcpy(obj->path, path);
1183        if (obj_name) {
1184                libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1185        } else {
1186                /* Using basename() GNU version which doesn't modify arg. */
1187                libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1188                end = strchr(obj->name, '.');
1189                if (end)
1190                        *end = 0;
1191        }
1192
1193        obj->efile.fd = -1;
1194        /*
1195         * Caller of this function should also call
1196         * bpf_object__elf_finish() after data collection to return
1197         * obj_buf to user. If not, we should duplicate the buffer to
1198         * avoid user freeing them before elf finish.
1199         */
1200        obj->efile.obj_buf = obj_buf;
1201        obj->efile.obj_buf_sz = obj_buf_sz;
1202        obj->efile.maps_shndx = -1;
1203        obj->efile.btf_maps_shndx = -1;
1204        obj->efile.st_ops_shndx = -1;
1205        obj->kconfig_map_idx = -1;
1206
1207        obj->kern_version = get_kernel_version();
1208        obj->loaded = false;
1209
1210        INIT_LIST_HEAD(&obj->list);
1211        if (!strict)
1212                list_add(&obj->list, &bpf_objects_list);
1213        return obj;
1214}
1215
1216static void bpf_object__elf_finish(struct bpf_object *obj)
1217{
1218        if (!obj->efile.elf)
1219                return;
1220
1221        if (obj->efile.elf) {
1222                elf_end(obj->efile.elf);
1223                obj->efile.elf = NULL;
1224        }
1225        obj->efile.symbols = NULL;
1226        obj->efile.st_ops_data = NULL;
1227
1228        zfree(&obj->efile.secs);
1229        obj->efile.sec_cnt = 0;
1230        zclose(obj->efile.fd);
1231        obj->efile.obj_buf = NULL;
1232        obj->efile.obj_buf_sz = 0;
1233}
1234
1235static int bpf_object__elf_init(struct bpf_object *obj)
1236{
1237        Elf64_Ehdr *ehdr;
1238        int err = 0;
1239        Elf *elf;
1240
1241        if (obj->efile.elf) {
1242                pr_warn("elf: init internal error\n");
1243                return -LIBBPF_ERRNO__LIBELF;
1244        }
1245
1246        if (obj->efile.obj_buf_sz > 0) {
1247                /*
1248                 * obj_buf should have been validated by
1249                 * bpf_object__open_buffer().
1250                 */
1251                elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1252        } else {
1253                obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1254                if (obj->efile.fd < 0) {
1255                        char errmsg[STRERR_BUFSIZE], *cp;
1256
1257                        err = -errno;
1258                        cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1259                        pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1260                        return err;
1261                }
1262
1263                elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1264        }
1265
1266        if (!elf) {
1267                pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1268                err = -LIBBPF_ERRNO__LIBELF;
1269                goto errout;
1270        }
1271
1272        obj->efile.elf = elf;
1273
1274        if (elf_kind(elf) != ELF_K_ELF) {
1275                err = -LIBBPF_ERRNO__FORMAT;
1276                pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1277                goto errout;
1278        }
1279
1280        if (gelf_getclass(elf) != ELFCLASS64) {
1281                err = -LIBBPF_ERRNO__FORMAT;
1282                pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1283                goto errout;
1284        }
1285
1286        obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1287        if (!obj->efile.ehdr) {
1288                pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1289                err = -LIBBPF_ERRNO__FORMAT;
1290                goto errout;
1291        }
1292
1293        if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1294                pr_warn("elf: failed to get section names section index for %s: %s\n",
1295                        obj->path, elf_errmsg(-1));
1296                err = -LIBBPF_ERRNO__FORMAT;
1297                goto errout;
1298        }
1299
1300        /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1301        if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1302                pr_warn("elf: failed to get section names strings from %s: %s\n",
1303                        obj->path, elf_errmsg(-1));
1304                err = -LIBBPF_ERRNO__FORMAT;
1305                goto errout;
1306        }
1307
1308        /* Old LLVM set e_machine to EM_NONE */
1309        if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1310                pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1311                err = -LIBBPF_ERRNO__FORMAT;
1312                goto errout;
1313        }
1314
1315        return 0;
1316errout:
1317        bpf_object__elf_finish(obj);
1318        return err;
1319}
1320
1321static int bpf_object__check_endianness(struct bpf_object *obj)
1322{
1323#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1324        if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB)
1325                return 0;
1326#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1327        if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB)
1328                return 0;
1329#else
1330# error "Unrecognized __BYTE_ORDER__"
1331#endif
1332        pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1333        return -LIBBPF_ERRNO__ENDIAN;
1334}
1335
1336static int
1337bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1338{
1339        /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1340         * go over allowed ELF data section buffer
1341         */
1342        libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1343        pr_debug("license of %s is %s\n", obj->path, obj->license);
1344        return 0;
1345}
1346
1347static int
1348bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1349{
1350        __u32 kver;
1351
1352        if (size != sizeof(kver)) {
1353                pr_warn("invalid kver section in %s\n", obj->path);
1354                return -LIBBPF_ERRNO__FORMAT;
1355        }
1356        memcpy(&kver, data, sizeof(kver));
1357        obj->kern_version = kver;
1358        pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1359        return 0;
1360}
1361
1362static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1363{
1364        if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1365            type == BPF_MAP_TYPE_HASH_OF_MAPS)
1366                return true;
1367        return false;
1368}
1369
1370static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1371{
1372        Elf_Data *data;
1373        Elf_Scn *scn;
1374
1375        if (!name)
1376                return -EINVAL;
1377
1378        scn = elf_sec_by_name(obj, name);
1379        data = elf_sec_data(obj, scn);
1380        if (data) {
1381                *size = data->d_size;
1382                return 0; /* found it */
1383        }
1384
1385        return -ENOENT;
1386}
1387
1388static int find_elf_var_offset(const struct bpf_object *obj, const char *name, __u32 *off)
1389{
1390        Elf_Data *symbols = obj->efile.symbols;
1391        const char *sname;
1392        size_t si;
1393
1394        if (!name || !off)
1395                return -EINVAL;
1396
1397        for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1398                Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1399
1400                if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL ||
1401                    ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1402                        continue;
1403
1404                sname = elf_sym_str(obj, sym->st_name);
1405                if (!sname) {
1406                        pr_warn("failed to get sym name string for var %s\n", name);
1407                        return -EIO;
1408                }
1409                if (strcmp(name, sname) == 0) {
1410                        *off = sym->st_value;
1411                        return 0;
1412                }
1413        }
1414
1415        return -ENOENT;
1416}
1417
1418static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1419{
1420        struct bpf_map *new_maps;
1421        size_t new_cap;
1422        int i;
1423
1424        if (obj->nr_maps < obj->maps_cap)
1425                return &obj->maps[obj->nr_maps++];
1426
1427        new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1428        new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps));
1429        if (!new_maps) {
1430                pr_warn("alloc maps for object failed\n");
1431                return ERR_PTR(-ENOMEM);
1432        }
1433
1434        obj->maps_cap = new_cap;
1435        obj->maps = new_maps;
1436
1437        /* zero out new maps */
1438        memset(obj->maps + obj->nr_maps, 0,
1439               (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1440        /*
1441         * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1442         * when failure (zclose won't close negative fd)).
1443         */
1444        for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1445                obj->maps[i].fd = -1;
1446                obj->maps[i].inner_map_fd = -1;
1447        }
1448
1449        return &obj->maps[obj->nr_maps++];
1450}
1451
1452static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1453{
1454        long page_sz = sysconf(_SC_PAGE_SIZE);
1455        size_t map_sz;
1456
1457        map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1458        map_sz = roundup(map_sz, page_sz);
1459        return map_sz;
1460}
1461
1462static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1463{
1464        char map_name[BPF_OBJ_NAME_LEN], *p;
1465        int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1466
1467        /* This is one of the more confusing parts of libbpf for various
1468         * reasons, some of which are historical. The original idea for naming
1469         * internal names was to include as much of BPF object name prefix as
1470         * possible, so that it can be distinguished from similar internal
1471         * maps of a different BPF object.
1472         * As an example, let's say we have bpf_object named 'my_object_name'
1473         * and internal map corresponding to '.rodata' ELF section. The final
1474         * map name advertised to user and to the kernel will be
1475         * 'my_objec.rodata', taking first 8 characters of object name and
1476         * entire 7 characters of '.rodata'.
1477         * Somewhat confusingly, if internal map ELF section name is shorter
1478         * than 7 characters, e.g., '.bss', we still reserve 7 characters
1479         * for the suffix, even though we only have 4 actual characters, and
1480         * resulting map will be called 'my_objec.bss', not even using all 15
1481         * characters allowed by the kernel. Oh well, at least the truncated
1482         * object name is somewhat consistent in this case. But if the map
1483         * name is '.kconfig', we'll still have entirety of '.kconfig' added
1484         * (8 chars) and thus will be left with only first 7 characters of the
1485         * object name ('my_obje'). Happy guessing, user, that the final map
1486         * name will be "my_obje.kconfig".
1487         * Now, with libbpf starting to support arbitrarily named .rodata.*
1488         * and .data.* data sections, it's possible that ELF section name is
1489         * longer than allowed 15 chars, so we now need to be careful to take
1490         * only up to 15 first characters of ELF name, taking no BPF object
1491         * name characters at all. So '.rodata.abracadabra' will result in
1492         * '.rodata.abracad' kernel and user-visible name.
1493         * We need to keep this convoluted logic intact for .data, .bss and
1494         * .rodata maps, but for new custom .data.custom and .rodata.custom
1495         * maps we use their ELF names as is, not prepending bpf_object name
1496         * in front. We still need to truncate them to 15 characters for the
1497         * kernel. Full name can be recovered for such maps by using DATASEC
1498         * BTF type associated with such map's value type, though.
1499         */
1500        if (sfx_len >= BPF_OBJ_NAME_LEN)
1501                sfx_len = BPF_OBJ_NAME_LEN - 1;
1502
1503        /* if there are two or more dots in map name, it's a custom dot map */
1504        if (strchr(real_name + 1, '.') != NULL)
1505                pfx_len = 0;
1506        else
1507                pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1508
1509        snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1510                 sfx_len, real_name);
1511
1512        /* sanitise map name to characters allowed by kernel */
1513        for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1514                if (!isalnum(*p) && *p != '_' && *p != '.')
1515                        *p = '_';
1516
1517        return strdup(map_name);
1518}
1519
1520static int
1521bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map);
1522
1523static int
1524bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1525                              const char *real_name, int sec_idx, void *data, size_t data_sz)
1526{
1527        struct bpf_map_def *def;
1528        struct bpf_map *map;
1529        int err;
1530
1531        map = bpf_object__add_map(obj);
1532        if (IS_ERR(map))
1533                return PTR_ERR(map);
1534
1535        map->libbpf_type = type;
1536        map->sec_idx = sec_idx;
1537        map->sec_offset = 0;
1538        map->real_name = strdup(real_name);
1539        map->name = internal_map_name(obj, real_name);
1540        if (!map->real_name || !map->name) {
1541                zfree(&map->real_name);
1542                zfree(&map->name);
1543                return -ENOMEM;
1544        }
1545
1546        def = &map->def;
1547        def->type = BPF_MAP_TYPE_ARRAY;
1548        def->key_size = sizeof(int);
1549        def->value_size = data_sz;
1550        def->max_entries = 1;
1551        def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1552                         ? BPF_F_RDONLY_PROG : 0;
1553        def->map_flags |= BPF_F_MMAPABLE;
1554
1555        pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1556                 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1557
1558        map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1559                           MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1560        if (map->mmaped == MAP_FAILED) {
1561                err = -errno;
1562                map->mmaped = NULL;
1563                pr_warn("failed to alloc map '%s' content buffer: %d\n",
1564                        map->name, err);
1565                zfree(&map->real_name);
1566                zfree(&map->name);
1567                return err;
1568        }
1569
1570        /* failures are fine because of maps like .rodata.str1.1 */
1571        (void) bpf_map_find_btf_info(obj, map);
1572
1573        if (data)
1574                memcpy(map->mmaped, data, data_sz);
1575
1576        pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1577        return 0;
1578}
1579
1580static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1581{
1582        struct elf_sec_desc *sec_desc;
1583        const char *sec_name;
1584        int err = 0, sec_idx;
1585
1586        /*
1587         * Populate obj->maps with libbpf internal maps.
1588         */
1589        for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1590                sec_desc = &obj->efile.secs[sec_idx];
1591
1592                switch (sec_desc->sec_type) {
1593                case SEC_DATA:
1594                        sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1595                        err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1596                                                            sec_name, sec_idx,
1597                                                            sec_desc->data->d_buf,
1598                                                            sec_desc->data->d_size);
1599                        break;
1600                case SEC_RODATA:
1601                        obj->has_rodata = true;
1602                        sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1603                        err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1604                                                            sec_name, sec_idx,
1605                                                            sec_desc->data->d_buf,
1606                                                            sec_desc->data->d_size);
1607                        break;
1608                case SEC_BSS:
1609                        sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1610                        err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1611                                                            sec_name, sec_idx,
1612                                                            NULL,
1613                                                            sec_desc->data->d_size);
1614                        break;
1615                default:
1616                        /* skip */
1617                        break;
1618                }
1619                if (err)
1620                        return err;
1621        }
1622        return 0;
1623}
1624
1625
1626static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1627                                               const void *name)
1628{
1629        int i;
1630
1631        for (i = 0; i < obj->nr_extern; i++) {
1632                if (strcmp(obj->externs[i].name, name) == 0)
1633                        return &obj->externs[i];
1634        }
1635        return NULL;
1636}
1637
1638static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1639                              char value)
1640{
1641        switch (ext->kcfg.type) {
1642        case KCFG_BOOL:
1643                if (value == 'm') {
1644                        pr_warn("extern (kcfg) %s=%c should be tristate or char\n",
1645                                ext->name, value);
1646                        return -EINVAL;
1647                }
1648                *(bool *)ext_val = value == 'y' ? true : false;
1649                break;
1650        case KCFG_TRISTATE:
1651                if (value == 'y')
1652                        *(enum libbpf_tristate *)ext_val = TRI_YES;
1653                else if (value == 'm')
1654                        *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1655                else /* value == 'n' */
1656                        *(enum libbpf_tristate *)ext_val = TRI_NO;
1657                break;
1658        case KCFG_CHAR:
1659                *(char *)ext_val = value;
1660                break;
1661        case KCFG_UNKNOWN:
1662        case KCFG_INT:
1663        case KCFG_CHAR_ARR:
1664        default:
1665                pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n",
1666                        ext->name, value);
1667                return -EINVAL;
1668        }
1669        ext->is_set = true;
1670        return 0;
1671}
1672
1673static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1674                              const char *value)
1675{
1676        size_t len;
1677
1678        if (ext->kcfg.type != KCFG_CHAR_ARR) {
1679                pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value);
1680                return -EINVAL;
1681        }
1682
1683        len = strlen(value);
1684        if (value[len - 1] != '"') {
1685                pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1686                        ext->name, value);
1687                return -EINVAL;
1688        }
1689
1690        /* strip quotes */
1691        len -= 2;
1692        if (len >= ext->kcfg.sz) {
1693                pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1694                        ext->name, value, len, ext->kcfg.sz - 1);
1695                len = ext->kcfg.sz - 1;
1696        }
1697        memcpy(ext_val, value + 1, len);
1698        ext_val[len] = '\0';
1699        ext->is_set = true;
1700        return 0;
1701}
1702
1703static int parse_u64(const char *value, __u64 *res)
1704{
1705        char *value_end;
1706        int err;
1707
1708        errno = 0;
1709        *res = strtoull(value, &value_end, 0);
1710        if (errno) {
1711                err = -errno;
1712                pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1713                return err;
1714        }
1715        if (*value_end) {
1716                pr_warn("failed to parse '%s' as integer completely\n", value);
1717                return -EINVAL;
1718        }
1719        return 0;
1720}
1721
1722static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1723{
1724        int bit_sz = ext->kcfg.sz * 8;
1725
1726        if (ext->kcfg.sz == 8)
1727                return true;
1728
1729        /* Validate that value stored in u64 fits in integer of `ext->sz`
1730         * bytes size without any loss of information. If the target integer
1731         * is signed, we rely on the following limits of integer type of
1732         * Y bits and subsequent transformation:
1733         *
1734         *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1735         *            0 <= X + 2^(Y-1) <= 2^Y - 1
1736         *            0 <= X + 2^(Y-1) <  2^Y
1737         *
1738         *  For unsigned target integer, check that all the (64 - Y) bits are
1739         *  zero.
1740         */
1741        if (ext->kcfg.is_signed)
1742                return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1743        else
1744                return (v >> bit_sz) == 0;
1745}
1746
1747static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1748                              __u64 value)
1749{
1750        if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1751                pr_warn("extern (kcfg) %s=%llu should be integer\n",
1752                        ext->name, (unsigned long long)value);
1753                return -EINVAL;
1754        }
1755        if (!is_kcfg_value_in_range(ext, value)) {
1756                pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n",
1757                        ext->name, (unsigned long long)value, ext->kcfg.sz);
1758                return -ERANGE;
1759        }
1760        switch (ext->kcfg.sz) {
1761                case 1: *(__u8 *)ext_val = value; break;
1762                case 2: *(__u16 *)ext_val = value; break;
1763                case 4: *(__u32 *)ext_val = value; break;
1764                case 8: *(__u64 *)ext_val = value; break;
1765                default:
1766                        return -EINVAL;
1767        }
1768        ext->is_set = true;
1769        return 0;
1770}
1771
1772static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1773                                            char *buf, void *data)
1774{
1775        struct extern_desc *ext;
1776        char *sep, *value;
1777        int len, err = 0;
1778        void *ext_val;
1779        __u64 num;
1780
1781        if (!str_has_pfx(buf, "CONFIG_"))
1782                return 0;
1783
1784        sep = strchr(buf, '=');
1785        if (!sep) {
1786                pr_warn("failed to parse '%s': no separator\n", buf);
1787                return -EINVAL;
1788        }
1789
1790        /* Trim ending '\n' */
1791        len = strlen(buf);
1792        if (buf[len - 1] == '\n')
1793                buf[len - 1] = '\0';
1794        /* Split on '=' and ensure that a value is present. */
1795        *sep = '\0';
1796        if (!sep[1]) {
1797                *sep = '=';
1798                pr_warn("failed to parse '%s': no value\n", buf);
1799                return -EINVAL;
1800        }
1801
1802        ext = find_extern_by_name(obj, buf);
1803        if (!ext || ext->is_set)
1804                return 0;
1805
1806        ext_val = data + ext->kcfg.data_off;
1807        value = sep + 1;
1808
1809        switch (*value) {
1810        case 'y': case 'n': case 'm':
1811                err = set_kcfg_value_tri(ext, ext_val, *value);
1812                break;
1813        case '"':
1814                err = set_kcfg_value_str(ext, ext_val, value);
1815                break;
1816        default:
1817                /* assume integer */
1818                err = parse_u64(value, &num);
1819                if (err) {
1820                        pr_warn("extern (kcfg) %s=%s should be integer\n",
1821                                ext->name, value);
1822                        return err;
1823                }
1824                err = set_kcfg_value_num(ext, ext_val, num);
1825                break;
1826        }
1827        if (err)
1828                return err;
1829        pr_debug("extern (kcfg) %s=%s\n", ext->name, value);
1830        return 0;
1831}
1832
1833static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1834{
1835        char buf[PATH_MAX];
1836        struct utsname uts;
1837        int len, err = 0;
1838        gzFile file;
1839
1840        uname(&uts);
1841        len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1842        if (len < 0)
1843                return -EINVAL;
1844        else if (len >= PATH_MAX)
1845                return -ENAMETOOLONG;
1846
1847        /* gzopen also accepts uncompressed files. */
1848        file = gzopen(buf, "r");
1849        if (!file)
1850                file = gzopen("/proc/config.gz", "r");
1851
1852        if (!file) {
1853                pr_warn("failed to open system Kconfig\n");
1854                return -ENOENT;
1855        }
1856
1857        while (gzgets(file, buf, sizeof(buf))) {
1858                err = bpf_object__process_kconfig_line(obj, buf, data);
1859                if (err) {
1860                        pr_warn("error parsing system Kconfig line '%s': %d\n",
1861                                buf, err);
1862                        goto out;
1863                }
1864        }
1865
1866out:
1867        gzclose(file);
1868        return err;
1869}
1870
1871static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1872                                        const char *config, void *data)
1873{
1874        char buf[PATH_MAX];
1875        int err = 0;
1876        FILE *file;
1877
1878        file = fmemopen((void *)config, strlen(config), "r");
1879        if (!file) {
1880                err = -errno;
1881                pr_warn("failed to open in-memory Kconfig: %d\n", err);
1882                return err;
1883        }
1884
1885        while (fgets(buf, sizeof(buf), file)) {
1886                err = bpf_object__process_kconfig_line(obj, buf, data);
1887                if (err) {
1888                        pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1889                                buf, err);
1890                        break;
1891                }
1892        }
1893
1894        fclose(file);
1895        return err;
1896}
1897
1898static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1899{
1900        struct extern_desc *last_ext = NULL, *ext;
1901        size_t map_sz;
1902        int i, err;
1903
1904        for (i = 0; i < obj->nr_extern; i++) {
1905                ext = &obj->externs[i];
1906                if (ext->type == EXT_KCFG)
1907                        last_ext = ext;
1908        }
1909
1910        if (!last_ext)
1911                return 0;
1912
1913        map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
1914        err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1915                                            ".kconfig", obj->efile.symbols_shndx,
1916                                            NULL, map_sz);
1917        if (err)
1918                return err;
1919
1920        obj->kconfig_map_idx = obj->nr_maps - 1;
1921
1922        return 0;
1923}
1924
1925static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1926{
1927        Elf_Data *symbols = obj->efile.symbols;
1928        int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1929        Elf_Data *data = NULL;
1930        Elf_Scn *scn;
1931
1932        if (obj->efile.maps_shndx < 0)
1933                return 0;
1934
1935        if (libbpf_mode & LIBBPF_STRICT_MAP_DEFINITIONS) {
1936                pr_warn("legacy map definitions in SEC(\"maps\") are not supported\n");
1937                return -EOPNOTSUPP;
1938        }
1939
1940        if (!symbols)
1941                return -EINVAL;
1942
1943        scn = elf_sec_by_idx(obj, obj->efile.maps_shndx);
1944        data = elf_sec_data(obj, scn);
1945        if (!scn || !data) {
1946                pr_warn("elf: failed to get legacy map definitions for %s\n",
1947                        obj->path);
1948                return -EINVAL;
1949        }
1950
1951        /*
1952         * Count number of maps. Each map has a name.
1953         * Array of maps is not supported: only the first element is
1954         * considered.
1955         *
1956         * TODO: Detect array of map and report error.
1957         */
1958        nr_syms = symbols->d_size / sizeof(Elf64_Sym);
1959        for (i = 0; i < nr_syms; i++) {
1960                Elf64_Sym *sym = elf_sym_by_idx(obj, i);
1961
1962                if (sym->st_shndx != obj->efile.maps_shndx)
1963                        continue;
1964                if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION)
1965                        continue;
1966                nr_maps++;
1967        }
1968        /* Assume equally sized map definitions */
1969        pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n",
1970                 nr_maps, data->d_size, obj->path);
1971
1972        if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1973                pr_warn("elf: unable to determine legacy map definition size in %s\n",
1974                        obj->path);
1975                return -EINVAL;
1976        }
1977        map_def_sz = data->d_size / nr_maps;
1978
1979        /* Fill obj->maps using data in "maps" section.  */
1980        for (i = 0; i < nr_syms; i++) {
1981                Elf64_Sym *sym = elf_sym_by_idx(obj, i);
1982                const char *map_name;
1983                struct bpf_map_def *def;
1984                struct bpf_map *map;
1985
1986                if (sym->st_shndx != obj->efile.maps_shndx)
1987                        continue;
1988                if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION)
1989                        continue;
1990
1991                map = bpf_object__add_map(obj);
1992                if (IS_ERR(map))
1993                        return PTR_ERR(map);
1994
1995                map_name = elf_sym_str(obj, sym->st_name);
1996                if (!map_name) {
1997                        pr_warn("failed to get map #%d name sym string for obj %s\n",
1998                                i, obj->path);
1999                        return -LIBBPF_ERRNO__FORMAT;
2000                }
2001
2002                pr_warn("map '%s' (legacy): legacy map definitions are deprecated, use BTF-defined maps instead\n", map_name);
2003
2004                if (ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
2005                        pr_warn("map '%s' (legacy): static maps are not supported\n", map_name);
2006                        return -ENOTSUP;
2007                }
2008
2009                map->libbpf_type = LIBBPF_MAP_UNSPEC;
2010                map->sec_idx = sym->st_shndx;
2011                map->sec_offset = sym->st_value;
2012                pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
2013                         map_name, map->sec_idx, map->sec_offset);
2014                if (sym->st_value + map_def_sz > data->d_size) {
2015                        pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
2016                                obj->path, map_name);
2017                        return -EINVAL;
2018                }
2019
2020                map->name = strdup(map_name);
2021                if (!map->name) {
2022                        pr_warn("map '%s': failed to alloc map name\n", map_name);
2023                        return -ENOMEM;
2024                }
2025                pr_debug("map %d is \"%s\"\n", i, map->name);
2026                def = (struct bpf_map_def *)(data->d_buf + sym->st_value);
2027                /*
2028                 * If the definition of the map in the object file fits in
2029                 * bpf_map_def, copy it.  Any extra fields in our version
2030                 * of bpf_map_def will default to zero as a result of the
2031                 * calloc above.
2032                 */
2033                if (map_def_sz <= sizeof(struct bpf_map_def)) {
2034                        memcpy(&map->def, def, map_def_sz);
2035                } else {
2036                        /*
2037                         * Here the map structure being read is bigger than what
2038                         * we expect, truncate if the excess bits are all zero.
2039                         * If they are not zero, reject this map as
2040                         * incompatible.
2041                         */
2042                        char *b;
2043
2044                        for (b = ((char *)def) + sizeof(struct bpf_map_def);
2045                             b < ((char *)def) + map_def_sz; b++) {
2046                                if (*b != 0) {
2047                                        pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
2048                                                obj->path, map_name);
2049                                        if (strict)
2050                                                return -EINVAL;
2051                                }
2052                        }
2053                        memcpy(&map->def, def, sizeof(struct bpf_map_def));
2054                }
2055
2056                /* btf info may not exist but fill it in if it does exist */
2057                (void) bpf_map_find_btf_info(obj, map);
2058        }
2059        return 0;
2060}
2061
2062const struct btf_type *
2063skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2064{
2065        const struct btf_type *t = btf__type_by_id(btf, id);
2066
2067        if (res_id)
2068                *res_id = id;
2069
2070        while (btf_is_mod(t) || btf_is_typedef(t)) {
2071                if (res_id)
2072                        *res_id = t->type;
2073                t = btf__type_by_id(btf, t->type);
2074        }
2075
2076        return t;
2077}
2078
2079static const struct btf_type *
2080resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2081{
2082        const struct btf_type *t;
2083
2084        t = skip_mods_and_typedefs(btf, id, NULL);
2085        if (!btf_is_ptr(t))
2086                return NULL;
2087
2088        t = skip_mods_and_typedefs(btf, t->type, res_id);
2089
2090        return btf_is_func_proto(t) ? t : NULL;
2091}
2092
2093static const char *__btf_kind_str(__u16 kind)
2094{
2095        switch (kind) {
2096        case BTF_KIND_UNKN: return "void";
2097        case BTF_KIND_INT: return "int";
2098        case BTF_KIND_PTR: return "ptr";
2099        case BTF_KIND_ARRAY: return "array";
2100        case BTF_KIND_STRUCT: return "struct";
2101        case BTF_KIND_UNION: return "union";
2102        case BTF_KIND_ENUM: return "enum";
2103        case BTF_KIND_FWD: return "fwd";
2104        case BTF_KIND_TYPEDEF: return "typedef";
2105        case BTF_KIND_VOLATILE: return "volatile";
2106        case BTF_KIND_CONST: return "const";
2107        case BTF_KIND_RESTRICT: return "restrict";
2108        case BTF_KIND_FUNC: return "func";
2109        case BTF_KIND_FUNC_PROTO: return "func_proto";
2110        case BTF_KIND_VAR: return "var";
2111        case BTF_KIND_DATASEC: return "datasec";
2112        case BTF_KIND_FLOAT: return "float";
2113        case BTF_KIND_DECL_TAG: return "decl_tag";
2114        case BTF_KIND_TYPE_TAG: return "type_tag";
2115        default: return "unknown";
2116        }
2117}
2118
2119const char *btf_kind_str(const struct btf_type *t)
2120{
2121        return __btf_kind_str(btf_kind(t));
2122}
2123
2124/*
2125 * Fetch integer attribute of BTF map definition. Such attributes are
2126 * represented using a pointer to an array, in which dimensionality of array
2127 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2128 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2129 * type definition, while using only sizeof(void *) space in ELF data section.
2130 */
2131static bool get_map_field_int(const char *map_name, const struct btf *btf,
2132                              const struct btf_member *m, __u32 *res)
2133{
2134        const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2135        const char *name = btf__name_by_offset(btf, m->name_off);
2136        const struct btf_array *arr_info;
2137        const struct btf_type *arr_t;
2138
2139        if (!btf_is_ptr(t)) {
2140                pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2141                        map_name, name, btf_kind_str(t));
2142                return false;
2143        }
2144
2145        arr_t = btf__type_by_id(btf, t->type);
2146        if (!arr_t) {
2147                pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2148                        map_name, name, t->type);
2149                return false;
2150        }
2151        if (!btf_is_array(arr_t)) {
2152                pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2153                        map_name, name, btf_kind_str(arr_t));
2154                return false;
2155        }
2156        arr_info = btf_array(arr_t);
2157        *res = arr_info->nelems;
2158        return true;
2159}
2160
2161static int build_map_pin_path(struct bpf_map *map, const char *path)
2162{
2163        char buf[PATH_MAX];
2164        int len;
2165
2166        if (!path)
2167                path = "/sys/fs/bpf";
2168
2169        len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
2170        if (len < 0)
2171                return -EINVAL;
2172        else if (len >= PATH_MAX)
2173                return -ENAMETOOLONG;
2174
2175        return bpf_map__set_pin_path(map, buf);
2176}
2177
2178int parse_btf_map_def(const char *map_name, struct btf *btf,
2179                      const struct btf_type *def_t, bool strict,
2180                      struct btf_map_def *map_def, struct btf_map_def *inner_def)
2181{
2182        const struct btf_type *t;
2183        const struct btf_member *m;
2184        bool is_inner = inner_def == NULL;
2185        int vlen, i;
2186
2187        vlen = btf_vlen(def_t);
2188        m = btf_members(def_t);
2189        for (i = 0; i < vlen; i++, m++) {
2190                const char *name = btf__name_by_offset(btf, m->name_off);
2191
2192                if (!name) {
2193                        pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2194                        return -EINVAL;
2195                }
2196                if (strcmp(name, "type") == 0) {
2197                        if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2198                                return -EINVAL;
2199                        map_def->parts |= MAP_DEF_MAP_TYPE;
2200                } else if (strcmp(name, "max_entries") == 0) {
2201                        if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2202                                return -EINVAL;
2203                        map_def->parts |= MAP_DEF_MAX_ENTRIES;
2204                } else if (strcmp(name, "map_flags") == 0) {
2205                        if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2206                                return -EINVAL;
2207                        map_def->parts |= MAP_DEF_MAP_FLAGS;
2208                } else if (strcmp(name, "numa_node") == 0) {
2209                        if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2210                                return -EINVAL;
2211                        map_def->parts |= MAP_DEF_NUMA_NODE;
2212                } else if (strcmp(name, "key_size") == 0) {
2213                        __u32 sz;
2214
2215                        if (!get_map_field_int(map_name, btf, m, &sz))
2216                                return -EINVAL;
2217                        if (map_def->key_size && map_def->key_size != sz) {
2218                                pr_warn("map '%s': conflicting key size %u != %u.\n",
2219                                        map_name, map_def->key_size, sz);
2220                                return -EINVAL;
2221                        }
2222                        map_def->key_size = sz;
2223                        map_def->parts |= MAP_DEF_KEY_SIZE;
2224                } else if (strcmp(name, "key") == 0) {
2225                        __s64 sz;
2226
2227                        t = btf__type_by_id(btf, m->type);
2228                        if (!t) {
2229                                pr_warn("map '%s': key type [%d] not found.\n",
2230                                        map_name, m->type);
2231                                return -EINVAL;
2232                        }
2233                        if (!btf_is_ptr(t)) {
2234                                pr_warn("map '%s': key spec is not PTR: %s.\n",
2235                                        map_name, btf_kind_str(t));
2236                                return -EINVAL;
2237                        }
2238                        sz = btf__resolve_size(btf, t->type);
2239                        if (sz < 0) {
2240                                pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2241                                        map_name, t->type, (ssize_t)sz);
2242                                return sz;
2243                        }
2244                        if (map_def->key_size && map_def->key_size != sz) {
2245                                pr_warn("map '%s': conflicting key size %u != %zd.\n",
2246                                        map_name, map_def->key_size, (ssize_t)sz);
2247                                return -EINVAL;
2248                        }
2249                        map_def->key_size = sz;
2250                        map_def->key_type_id = t->type;
2251                        map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2252                } else if (strcmp(name, "value_size") == 0) {
2253                        __u32 sz;
2254
2255                        if (!get_map_field_int(map_name, btf, m, &sz))
2256                                return -EINVAL;
2257                        if (map_def->value_size && map_def->value_size != sz) {
2258                                pr_warn("map '%s': conflicting value size %u != %u.\n",
2259                                        map_name, map_def->value_size, sz);
2260                                return -EINVAL;
2261                        }
2262                        map_def->value_size = sz;
2263                        map_def->parts |= MAP_DEF_VALUE_SIZE;
2264                } else if (strcmp(name, "value") == 0) {
2265                        __s64 sz;
2266
2267                        t = btf__type_by_id(btf, m->type);
2268                        if (!t) {
2269                                pr_warn("map '%s': value type [%d] not found.\n",
2270                                        map_name, m->type);
2271                                return -EINVAL;
2272                        }
2273                        if (!btf_is_ptr(t)) {
2274                                pr_warn("map '%s': value spec is not PTR: %s.\n",
2275                                        map_name, btf_kind_str(t));
2276                                return -EINVAL;
2277                        }
2278                        sz = btf__resolve_size(btf, t->type);
2279                        if (sz < 0) {
2280                                pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2281                                        map_name, t->type, (ssize_t)sz);
2282                                return sz;
2283                        }
2284                        if (map_def->value_size && map_def->value_size != sz) {
2285                                pr_warn("map '%s': conflicting value size %u != %zd.\n",
2286                                        map_name, map_def->value_size, (ssize_t)sz);
2287                                return -EINVAL;
2288                        }
2289                        map_def->value_size = sz;
2290                        map_def->value_type_id = t->type;
2291                        map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2292                }
2293                else if (strcmp(name, "values") == 0) {
2294                        bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2295                        bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2296                        const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2297                        char inner_map_name[128];
2298                        int err;
2299
2300                        if (is_inner) {
2301                                pr_warn("map '%s': multi-level inner maps not supported.\n",
2302                                        map_name);
2303                                return -ENOTSUP;
2304                        }
2305                        if (i != vlen - 1) {
2306                                pr_warn("map '%s': '%s' member should be last.\n",
2307                                        map_name, name);
2308                                return -EINVAL;
2309                        }
2310                        if (!is_map_in_map && !is_prog_array) {
2311                                pr_warn("map '%s': should be map-in-map or prog-array.\n",
2312                                        map_name);
2313                                return -ENOTSUP;
2314                        }
2315                        if (map_def->value_size && map_def->value_size != 4) {
2316                                pr_warn("map '%s': conflicting value size %u != 4.\n",
2317                                        map_name, map_def->value_size);
2318                                return -EINVAL;
2319                        }
2320                        map_def->value_size = 4;
2321                        t = btf__type_by_id(btf, m->type);
2322                        if (!t) {
2323                                pr_warn("map '%s': %s type [%d] not found.\n",
2324                                        map_name, desc, m->type);
2325                                return -EINVAL;
2326                        }
2327                        if (!btf_is_array(t) || btf_array(t)->nelems) {
2328                                pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2329                                        map_name, desc);
2330                                return -EINVAL;
2331                        }
2332                        t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2333                        if (!btf_is_ptr(t)) {
2334                                pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2335                                        map_name, desc, btf_kind_str(t));
2336                                return -EINVAL;
2337                        }
2338                        t = skip_mods_and_typedefs(btf, t->type, NULL);
2339                        if (is_prog_array) {
2340                                if (!btf_is_func_proto(t)) {
2341                                        pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2342                                                map_name, btf_kind_str(t));
2343                                        return -EINVAL;
2344                                }
2345                                continue;
2346                        }
2347                        if (!btf_is_struct(t)) {
2348                                pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2349                                        map_name, btf_kind_str(t));
2350                                return -EINVAL;
2351                        }
2352
2353                        snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2354                        err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2355                        if (err)
2356                                return err;
2357
2358                        map_def->parts |= MAP_DEF_INNER_MAP;
2359                } else if (strcmp(name, "pinning") == 0) {
2360                        __u32 val;
2361
2362                        if (is_inner) {
2363                                pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2364                                return -EINVAL;
2365                        }
2366                        if (!get_map_field_int(map_name, btf, m, &val))
2367                                return -EINVAL;
2368                        if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2369                                pr_warn("map '%s': invalid pinning value %u.\n",
2370                                        map_name, val);
2371                                return -EINVAL;
2372                        }
2373                        map_def->pinning = val;
2374                        map_def->parts |= MAP_DEF_PINNING;
2375                } else if (strcmp(name, "map_extra") == 0) {
2376                        __u32 map_extra;
2377
2378                        if (!get_map_field_int(map_name, btf, m, &map_extra))
2379                                return -EINVAL;
2380                        map_def->map_extra = map_extra;
2381                        map_def->parts |= MAP_DEF_MAP_EXTRA;
2382                } else {
2383                        if (strict) {
2384                                pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2385                                return -ENOTSUP;
2386                        }
2387                        pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2388                }
2389        }
2390
2391        if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2392                pr_warn("map '%s': map type isn't specified.\n", map_name);
2393                return -EINVAL;
2394        }
2395
2396        return 0;
2397}
2398
2399static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2400{
2401        map->def.type = def->map_type;
2402        map->def.key_size = def->key_size;
2403        map->def.value_size = def->value_size;
2404        map->def.max_entries = def->max_entries;
2405        map->def.map_flags = def->map_flags;
2406        map->map_extra = def->map_extra;
2407
2408        map->numa_node = def->numa_node;
2409        map->btf_key_type_id = def->key_type_id;
2410        map->btf_value_type_id = def->value_type_id;
2411
2412        if (def->parts & MAP_DEF_MAP_TYPE)
2413                pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2414
2415        if (def->parts & MAP_DEF_KEY_TYPE)
2416                pr_debug("map '%s': found key [%u], sz = %u.\n",
2417                         map->name, def->key_type_id, def->key_size);
2418        else if (def->parts & MAP_DEF_KEY_SIZE)
2419                pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2420
2421        if (def->parts & MAP_DEF_VALUE_TYPE)
2422                pr_debug("map '%s': found value [%u], sz = %u.\n",
2423                         map->name, def->value_type_id, def->value_size);
2424        else if (def->parts & MAP_DEF_VALUE_SIZE)
2425                pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2426
2427        if (def->parts & MAP_DEF_MAX_ENTRIES)
2428                pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2429        if (def->parts & MAP_DEF_MAP_FLAGS)
2430                pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2431        if (def->parts & MAP_DEF_MAP_EXTRA)
2432                pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2433                         (unsigned long long)def->map_extra);
2434        if (def->parts & MAP_DEF_PINNING)
2435                pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2436        if (def->parts & MAP_DEF_NUMA_NODE)
2437                pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2438
2439        if (def->parts & MAP_DEF_INNER_MAP)
2440                pr_debug("map '%s': found inner map definition.\n", map->name);
2441}
2442
2443static const char *btf_var_linkage_str(__u32 linkage)
2444{
2445        switch (linkage) {
2446        case BTF_VAR_STATIC: return "static";
2447        case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2448        case BTF_VAR_GLOBAL_EXTERN: return "extern";
2449        default: return "unknown";
2450        }
2451}
2452
2453static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2454                                         const struct btf_type *sec,
2455                                         int var_idx, int sec_idx,
2456                                         const Elf_Data *data, bool strict,
2457                                         const char *pin_root_path)
2458{
2459        struct btf_map_def map_def = {}, inner_def = {};
2460        const struct btf_type *var, *def;
2461        const struct btf_var_secinfo *vi;
2462        const struct btf_var *var_extra;
2463        const char *map_name;
2464        struct bpf_map *map;
2465        int err;
2466
2467        vi = btf_var_secinfos(sec) + var_idx;
2468        var = btf__type_by_id(obj->btf, vi->type);
2469        var_extra = btf_var(var);
2470        map_name = btf__name_by_offset(obj->btf, var->name_off);
2471
2472        if (map_name == NULL || map_name[0] == '\0') {
2473                pr_warn("map #%d: empty name.\n", var_idx);
2474                return -EINVAL;
2475        }
2476        if ((__u64)vi->offset + vi->size > data->d_size) {
2477                pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2478                return -EINVAL;
2479        }
2480        if (!btf_is_var(var)) {
2481                pr_warn("map '%s': unexpected var kind %s.\n",
2482                        map_name, btf_kind_str(var));
2483                return -EINVAL;
2484        }
2485        if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2486                pr_warn("map '%s': unsupported map linkage %s.\n",
2487                        map_name, btf_var_linkage_str(var_extra->linkage));
2488                return -EOPNOTSUPP;
2489        }
2490
2491        def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2492        if (!btf_is_struct(def)) {
2493                pr_warn("map '%s': unexpected def kind %s.\n",
2494                        map_name, btf_kind_str(var));
2495                return -EINVAL;
2496        }
2497        if (def->size > vi->size) {
2498                pr_warn("map '%s': invalid def size.\n", map_name);
2499                return -EINVAL;
2500        }
2501
2502        map = bpf_object__add_map(obj);
2503        if (IS_ERR(map))
2504                return PTR_ERR(map);
2505        map->name = strdup(map_name);
2506        if (!map->name) {
2507                pr_warn("map '%s': failed to alloc map name.\n", map_name);
2508                return -ENOMEM;
2509        }
2510        map->libbpf_type = LIBBPF_MAP_UNSPEC;
2511        map->def.type = BPF_MAP_TYPE_UNSPEC;
2512        map->sec_idx = sec_idx;
2513        map->sec_offset = vi->offset;
2514        map->btf_var_idx = var_idx;
2515        pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2516                 map_name, map->sec_idx, map->sec_offset);
2517
2518        err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2519        if (err)
2520                return err;
2521
2522        fill_map_from_def(map, &map_def);
2523
2524        if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2525                err = build_map_pin_path(map, pin_root_path);
2526                if (err) {
2527                        pr_warn("map '%s': couldn't build pin path.\n", map->name);
2528                        return err;
2529                }
2530        }
2531
2532        if (map_def.parts & MAP_DEF_INNER_MAP) {
2533                map->inner_map = calloc(1, sizeof(*map->inner_map));
2534                if (!map->inner_map)
2535                        return -ENOMEM;
2536                map->inner_map->fd = -1;
2537                map->inner_map->sec_idx = sec_idx;
2538                map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2539                if (!map->inner_map->name)
2540                        return -ENOMEM;
2541                sprintf(map->inner_map->name, "%s.inner", map_name);
2542
2543                fill_map_from_def(map->inner_map, &inner_def);
2544        }
2545
2546        err = bpf_map_find_btf_info(obj, map);
2547        if (err)
2548                return err;
2549
2550        return 0;
2551}
2552
2553static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2554                                          const char *pin_root_path)
2555{
2556        const struct btf_type *sec = NULL;
2557        int nr_types, i, vlen, err;
2558        const struct btf_type *t;
2559        const char *name;
2560        Elf_Data *data;
2561        Elf_Scn *scn;
2562
2563        if (obj->efile.btf_maps_shndx < 0)
2564                return 0;
2565
2566        scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2567        data = elf_sec_data(obj, scn);
2568        if (!scn || !data) {
2569                pr_warn("elf: failed to get %s map definitions for %s\n",
2570                        MAPS_ELF_SEC, obj->path);
2571                return -EINVAL;
2572        }
2573
2574        nr_types = btf__type_cnt(obj->btf);
2575        for (i = 1; i < nr_types; i++) {
2576                t = btf__type_by_id(obj->btf, i);
2577                if (!btf_is_datasec(t))
2578                        continue;
2579                name = btf__name_by_offset(obj->btf, t->name_off);
2580                if (strcmp(name, MAPS_ELF_SEC) == 0) {
2581                        sec = t;
2582                        obj->efile.btf_maps_sec_btf_id = i;
2583                        break;
2584                }
2585        }
2586
2587        if (!sec) {
2588                pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2589                return -ENOENT;
2590        }
2591
2592        vlen = btf_vlen(sec);
2593        for (i = 0; i < vlen; i++) {
2594                err = bpf_object__init_user_btf_map(obj, sec, i,
2595                                                    obj->efile.btf_maps_shndx,
2596                                                    data, strict,
2597                                                    pin_root_path);
2598                if (err)
2599                        return err;
2600        }
2601
2602        return 0;
2603}
2604
2605static int bpf_object__init_maps(struct bpf_object *obj,
2606                                 const struct bpf_object_open_opts *opts)
2607{
2608        const char *pin_root_path;
2609        bool strict;
2610        int err;
2611
2612        strict = !OPTS_GET(opts, relaxed_maps, false);
2613        pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2614
2615        err = bpf_object__init_user_maps(obj, strict);
2616        err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2617        err = err ?: bpf_object__init_global_data_maps(obj);
2618        err = err ?: bpf_object__init_kconfig_map(obj);
2619        err = err ?: bpf_object__init_struct_ops_maps(obj);
2620
2621        return err;
2622}
2623
2624static bool section_have_execinstr(struct bpf_object *obj, int idx)
2625{
2626        Elf64_Shdr *sh;
2627
2628        sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
2629        if (!sh)
2630                return false;
2631
2632        return sh->sh_flags & SHF_EXECINSTR;
2633}
2634
2635static bool btf_needs_sanitization(struct bpf_object *obj)
2636{
2637        bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2638        bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2639        bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2640        bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2641        bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2642        bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2643
2644        return !has_func || !has_datasec || !has_func_global || !has_float ||
2645               !has_decl_tag || !has_type_tag;
2646}
2647
2648static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2649{
2650        bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2651        bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2652        bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2653        bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2654        bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2655        bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2656        struct btf_type *t;
2657        int i, j, vlen;
2658
2659        for (i = 1; i < btf__type_cnt(btf); i++) {
2660                t = (struct btf_type *)btf__type_by_id(btf, i);
2661
2662                if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
2663                        /* replace VAR/DECL_TAG with INT */
2664                        t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2665                        /*
2666                         * using size = 1 is the safest choice, 4 will be too
2667                         * big and cause kernel BTF validation failure if
2668                         * original variable took less than 4 bytes
2669                         */
2670                        t->size = 1;
2671                        *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2672                } else if (!has_datasec && btf_is_datasec(t)) {
2673                        /* replace DATASEC with STRUCT */
2674                        const struct btf_var_secinfo *v = btf_var_secinfos(t);
2675                        struct btf_member *m = btf_members(t);
2676                        struct btf_type *vt;
2677                        char *name;
2678
2679                        name = (char *)btf__name_by_offset(btf, t->name_off);
2680                        while (*name) {
2681                                if (*name == '.')
2682                                        *name = '_';
2683                                name++;
2684                        }
2685
2686                        vlen = btf_vlen(t);
2687                        t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2688                        for (j = 0; j < vlen; j++, v++, m++) {
2689                                /* order of field assignments is important */
2690                                m->offset = v->offset * 8;
2691                                m->type = v->type;
2692                                /* preserve variable name as member name */
2693                                vt = (void *)btf__type_by_id(btf, v->type);
2694                                m->name_off = vt->name_off;
2695                        }
2696                } else if (!has_func && btf_is_func_proto(t)) {
2697                        /* replace FUNC_PROTO with ENUM */
2698                        vlen = btf_vlen(t);
2699                        t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2700                        t->size = sizeof(__u32); /* kernel enforced */
2701                } else if (!has_func && btf_is_func(t)) {
2702                        /* replace FUNC with TYPEDEF */
2703                        t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2704                } else if (!has_func_global && btf_is_func(t)) {
2705                        /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2706                        t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2707                } else if (!has_float && btf_is_float(t)) {
2708                        /* replace FLOAT with an equally-sized empty STRUCT;
2709                         * since C compilers do not accept e.g. "float" as a
2710                         * valid struct name, make it anonymous
2711                         */
2712                        t->name_off = 0;
2713                        t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2714                } else if (!has_type_tag && btf_is_type_tag(t)) {
2715                        /* replace TYPE_TAG with a CONST */
2716                        t->name_off = 0;
2717                        t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
2718                }
2719        }
2720}
2721
2722static bool libbpf_needs_btf(const struct bpf_object *obj)
2723{
2724        return obj->efile.btf_maps_shndx >= 0 ||
2725               obj->efile.st_ops_shndx >= 0 ||
2726               obj->nr_extern > 0;
2727}
2728
2729static bool kernel_needs_btf(const struct bpf_object *obj)
2730{
2731        return obj->efile.st_ops_shndx >= 0;
2732}
2733
2734static int bpf_object__init_btf(struct bpf_object *obj,
2735                                Elf_Data *btf_data,
2736                                Elf_Data *btf_ext_data)
2737{
2738        int err = -ENOENT;
2739
2740        if (btf_data) {
2741                obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2742                err = libbpf_get_error(obj->btf);
2743                if (err) {
2744                        obj->btf = NULL;
2745                        pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2746                        goto out;
2747                }
2748                /* enforce 8-byte pointers for BPF-targeted BTFs */
2749                btf__set_pointer_size(obj->btf, 8);
2750        }
2751        if (btf_ext_data) {
2752                if (!obj->btf) {
2753                        pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2754                                 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2755                        goto out;
2756                }
2757                obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2758                err = libbpf_get_error(obj->btf_ext);
2759                if (err) {
2760                        pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2761                                BTF_EXT_ELF_SEC, err);
2762                        obj->btf_ext = NULL;
2763                        goto out;
2764                }
2765        }
2766out:
2767        if (err && libbpf_needs_btf(obj)) {
2768                pr_warn("BTF is required, but is missing or corrupted.\n");
2769                return err;
2770        }
2771        return 0;
2772}
2773
2774static int compare_vsi_off(const void *_a, const void *_b)
2775{
2776        const struct btf_var_secinfo *a = _a;
2777        const struct btf_var_secinfo *b = _b;
2778
2779        return a->offset - b->offset;
2780}
2781
2782static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
2783                             struct btf_type *t)
2784{
2785        __u32 size = 0, off = 0, i, vars = btf_vlen(t);
2786        const char *name = btf__name_by_offset(btf, t->name_off);
2787        const struct btf_type *t_var;
2788        struct btf_var_secinfo *vsi;
2789        const struct btf_var *var;
2790        int ret;
2791
2792        if (!name) {
2793                pr_debug("No name found in string section for DATASEC kind.\n");
2794                return -ENOENT;
2795        }
2796
2797        /* .extern datasec size and var offsets were set correctly during
2798         * extern collection step, so just skip straight to sorting variables
2799         */
2800        if (t->size)
2801                goto sort_vars;
2802
2803        ret = find_elf_sec_sz(obj, name, &size);
2804        if (ret || !size) {
2805                pr_debug("Invalid size for section %s: %u bytes\n", name, size);
2806                return -ENOENT;
2807        }
2808
2809        t->size = size;
2810
2811        for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
2812                t_var = btf__type_by_id(btf, vsi->type);
2813                if (!t_var || !btf_is_var(t_var)) {
2814                        pr_debug("Non-VAR type seen in section %s\n", name);
2815                        return -EINVAL;
2816                }
2817
2818                var = btf_var(t_var);
2819                if (var->linkage == BTF_VAR_STATIC)
2820                        continue;
2821
2822                name = btf__name_by_offset(btf, t_var->name_off);
2823                if (!name) {
2824                        pr_debug("No name found in string section for VAR kind\n");
2825                        return -ENOENT;
2826                }
2827
2828                ret = find_elf_var_offset(obj, name, &off);
2829                if (ret) {
2830                        pr_debug("No offset found in symbol table for VAR %s\n",
2831                                 name);
2832                        return -ENOENT;
2833                }
2834
2835                vsi->offset = off;
2836        }
2837
2838sort_vars:
2839        qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
2840        return 0;
2841}
2842
2843static int btf_finalize_data(struct bpf_object *obj, struct btf *btf)
2844{
2845        int err = 0;
2846        __u32 i, n = btf__type_cnt(btf);
2847
2848        for (i = 1; i < n; i++) {
2849                struct btf_type *t = btf_type_by_id(btf, i);
2850
2851                /* Loader needs to fix up some of the things compiler
2852                 * couldn't get its hands on while emitting BTF. This
2853                 * is section size and global variable offset. We use
2854                 * the info from the ELF itself for this purpose.
2855                 */
2856                if (btf_is_datasec(t)) {
2857                        err = btf_fixup_datasec(obj, btf, t);
2858                        if (err)
2859                                break;
2860                }
2861        }
2862
2863        return libbpf_err(err);
2864}
2865
2866int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
2867{
2868        return btf_finalize_data(obj, btf);
2869}
2870
2871static int bpf_object__finalize_btf(struct bpf_object *obj)
2872{
2873        int err;
2874
2875        if (!obj->btf)
2876                return 0;
2877
2878        err = btf_finalize_data(obj, obj->btf);
2879        if (err) {
2880                pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2881                return err;
2882        }
2883
2884        return 0;
2885}
2886
2887static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
2888{
2889        if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2890            prog->type == BPF_PROG_TYPE_LSM)
2891                return true;
2892
2893        /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2894         * also need vmlinux BTF
2895         */
2896        if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2897                return true;
2898
2899        return false;
2900}
2901
2902static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
2903{
2904        struct bpf_program *prog;
2905        int i;
2906
2907        /* CO-RE relocations need kernel BTF, only when btf_custom_path
2908         * is not specified
2909         */
2910        if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
2911                return true;
2912
2913        /* Support for typed ksyms needs kernel BTF */
2914        for (i = 0; i < obj->nr_extern; i++) {
2915                const struct extern_desc *ext;
2916
2917                ext = &obj->externs[i];
2918                if (ext->type == EXT_KSYM && ext->ksym.type_id)
2919                        return true;
2920        }
2921
2922        bpf_object__for_each_program(prog, obj) {
2923                if (!prog->load)
2924                        continue;
2925                if (prog_needs_vmlinux_btf(prog))
2926                        return true;
2927        }
2928
2929        return false;
2930}
2931
2932static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
2933{
2934        int err;
2935
2936        /* btf_vmlinux could be loaded earlier */
2937        if (obj->btf_vmlinux || obj->gen_loader)
2938                return 0;
2939
2940        if (!force && !obj_needs_vmlinux_btf(obj))
2941                return 0;
2942
2943        obj->btf_vmlinux = btf__load_vmlinux_btf();
2944        err = libbpf_get_error(obj->btf_vmlinux);
2945        if (err) {
2946                pr_warn("Error loading vmlinux BTF: %d\n", err);
2947                obj->btf_vmlinux = NULL;
2948                return err;
2949        }
2950        return 0;
2951}
2952
2953static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2954{
2955        struct btf *kern_btf = obj->btf;
2956        bool btf_mandatory, sanitize;
2957        int i, err = 0;
2958
2959        if (!obj->btf)
2960                return 0;
2961
2962        if (!kernel_supports(obj, FEAT_BTF)) {
2963                if (kernel_needs_btf(obj)) {
2964                        err = -EOPNOTSUPP;
2965                        goto report;
2966                }
2967                pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
2968                return 0;
2969        }
2970
2971        /* Even though some subprogs are global/weak, user might prefer more
2972         * permissive BPF verification process that BPF verifier performs for
2973         * static functions, taking into account more context from the caller
2974         * functions. In such case, they need to mark such subprogs with
2975         * __attribute__((visibility("hidden"))) and libbpf will adjust
2976         * corresponding FUNC BTF type to be marked as static and trigger more
2977         * involved BPF verification process.
2978         */
2979        for (i = 0; i < obj->nr_programs; i++) {
2980                struct bpf_program *prog = &obj->programs[i];
2981                struct btf_type *t;
2982                const char *name;
2983                int j, n;
2984
2985                if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
2986                        continue;
2987
2988                n = btf__type_cnt(obj->btf);
2989                for (j = 1; j < n; j++) {
2990                        t = btf_type_by_id(obj->btf, j);
2991                        if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
2992                                continue;
2993
2994                        name = btf__str_by_offset(obj->btf, t->name_off);
2995                        if (strcmp(name, prog->name) != 0)
2996                                continue;
2997
2998                        t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
2999                        break;
3000                }
3001        }
3002
3003        sanitize = btf_needs_sanitization(obj);
3004        if (sanitize) {
3005                const void *raw_data;
3006                __u32 sz;
3007
3008                /* clone BTF to sanitize a copy and leave the original intact */
3009                raw_data = btf__raw_data(obj->btf, &sz);
3010                kern_btf = btf__new(raw_data, sz);
3011                err = libbpf_get_error(kern_btf);
3012                if (err)
3013                        return err;
3014
3015                /* enforce 8-byte pointers for BPF-targeted BTFs */
3016                btf__set_pointer_size(obj->btf, 8);
3017                bpf_object__sanitize_btf(obj, kern_btf);
3018        }
3019
3020        if (obj->gen_loader) {
3021                __u32 raw_size = 0;
3022                const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3023
3024                if (!raw_data)
3025                        return -ENOMEM;
3026                bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3027                /* Pretend to have valid FD to pass various fd >= 0 checks.
3028                 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3029                 */
3030                btf__set_fd(kern_btf, 0);
3031        } else {
3032                /* currently BPF_BTF_LOAD only supports log_level 1 */
3033                err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3034                                           obj->log_level ? 1 : 0);
3035        }
3036        if (sanitize) {
3037                if (!err) {
3038                        /* move fd to libbpf's BTF */
3039                        btf__set_fd(obj->btf, btf__fd(kern_btf));
3040                        btf__set_fd(kern_btf, -1);
3041                }
3042                btf__free(kern_btf);
3043        }
3044report:
3045        if (err) {
3046                btf_mandatory = kernel_needs_btf(obj);
3047                pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
3048                        btf_mandatory ? "BTF is mandatory, can't proceed."
3049                                      : "BTF is optional, ignoring.");
3050                if (!btf_mandatory)
3051                        err = 0;
3052        }
3053        return err;
3054}
3055
3056static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3057{
3058        const char *name;
3059
3060        name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3061        if (!name) {
3062                pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3063                        off, obj->path, elf_errmsg(-1));
3064                return NULL;
3065        }
3066
3067        return name;
3068}
3069
3070static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3071{
3072        const char *name;
3073
3074        name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3075        if (!name) {
3076                pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3077                        off, obj->path, elf_errmsg(-1));
3078                return NULL;
3079        }
3080
3081        return name;
3082}
3083
3084static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3085{
3086        Elf_Scn *scn;
3087
3088        scn = elf_getscn(obj->efile.elf, idx);
3089        if (!scn) {
3090                pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3091                        idx, obj->path, elf_errmsg(-1));
3092                return NULL;
3093        }
3094        return scn;
3095}
3096
3097static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3098{
3099        Elf_Scn *scn = NULL;
3100        Elf *elf = obj->efile.elf;
3101        const char *sec_name;
3102
3103        while ((scn = elf_nextscn(elf, scn)) != NULL) {
3104                sec_name = elf_sec_name(obj, scn);
3105                if (!sec_name)
3106                        return NULL;
3107
3108                if (strcmp(sec_name, name) != 0)
3109                        continue;
3110
3111                return scn;
3112        }
3113        return NULL;
3114}
3115
3116static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3117{
3118        Elf64_Shdr *shdr;
3119
3120        if (!scn)
3121                return NULL;
3122
3123        shdr = elf64_getshdr(scn);
3124        if (!shdr) {
3125                pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3126                        elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3127                return NULL;
3128        }
3129
3130        return shdr;
3131}
3132
3133static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3134{
3135        const char *name;
3136        Elf64_Shdr *sh;
3137
3138        if (!scn)
3139                return NULL;
3140
3141        sh = elf_sec_hdr(obj, scn);
3142        if (!sh)
3143                return NULL;
3144
3145        name = elf_sec_str(obj, sh->sh_name);
3146        if (!name) {
3147                pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3148                        elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3149                return NULL;
3150        }
3151
3152        return name;
3153}
3154
3155static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3156{
3157        Elf_Data *data;
3158
3159        if (!scn)
3160                return NULL;
3161
3162        data = elf_getdata(scn, 0);
3163        if (!data) {
3164                pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3165                        elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3166                        obj->path, elf_errmsg(-1));
3167                return NULL;
3168        }
3169
3170        return data;
3171}
3172
3173static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3174{
3175        if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3176                return NULL;
3177
3178        return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3179}
3180
3181static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3182{
3183        if (idx >= data->d_size / sizeof(Elf64_Rel))
3184                return NULL;
3185
3186        return (Elf64_Rel *)data->d_buf + idx;
3187}
3188
3189static bool is_sec_name_dwarf(const char *name)
3190{
3191        /* approximation, but the actual list is too long */
3192        return str_has_pfx(name, ".debug_");
3193}
3194
3195static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3196{
3197        /* no special handling of .strtab */
3198        if (hdr->sh_type == SHT_STRTAB)
3199                return true;
3200
3201        /* ignore .llvm_addrsig section as well */
3202        if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3203                return true;
3204
3205        /* no subprograms will lead to an empty .text section, ignore it */
3206        if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3207            strcmp(name, ".text") == 0)
3208                return true;
3209
3210        /* DWARF sections */
3211        if (is_sec_name_dwarf(name))
3212                return true;
3213
3214        if (str_has_pfx(name, ".rel")) {
3215                name += sizeof(".rel") - 1;
3216                /* DWARF section relocations */
3217                if (is_sec_name_dwarf(name))
3218                        return true;
3219
3220                /* .BTF and .BTF.ext don't need relocations */
3221                if (strcmp(name, BTF_ELF_SEC) == 0 ||
3222                    strcmp(name, BTF_EXT_ELF_SEC) == 0)
3223                        return true;
3224        }
3225
3226        return false;
3227}
3228
3229static int cmp_progs(const void *_a, const void *_b)
3230{
3231        const struct bpf_program *a = _a;
3232        const struct bpf_program *b = _b;
3233
3234        if (a->sec_idx != b->sec_idx)
3235                return a->sec_idx < b->sec_idx ? -1 : 1;
3236
3237        /* sec_insn_off can't be the same within the section */
3238        return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3239}
3240
3241static int bpf_object__elf_collect(struct bpf_object *obj)
3242{
3243        struct elf_sec_desc *sec_desc;
3244        Elf *elf = obj->efile.elf;
3245        Elf_Data *btf_ext_data = NULL;
3246        Elf_Data *btf_data = NULL;
3247        int idx = 0, err = 0;
3248        const char *name;
3249        Elf_Data *data;
3250        Elf_Scn *scn;
3251        Elf64_Shdr *sh;
3252
3253        /* ELF section indices are 0-based, but sec #0 is special "invalid"
3254         * section. e_shnum does include sec #0, so e_shnum is the necessary
3255         * size of an array to keep all the sections.
3256         */
3257        obj->efile.sec_cnt = obj->efile.ehdr->e_shnum;
3258        obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3259        if (!obj->efile.secs)
3260                return -ENOMEM;
3261
3262        /* a bunch of ELF parsing functionality depends on processing symbols,
3263         * so do the first pass and find the symbol table
3264         */
3265        scn = NULL;
3266        while ((scn = elf_nextscn(elf, scn)) != NULL) {
3267                sh = elf_sec_hdr(obj, scn);
3268                if (!sh)
3269                        return -LIBBPF_ERRNO__FORMAT;
3270
3271                if (sh->sh_type == SHT_SYMTAB) {
3272                        if (obj->efile.symbols) {
3273                                pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3274                                return -LIBBPF_ERRNO__FORMAT;
3275                        }
3276
3277                        data = elf_sec_data(obj, scn);
3278                        if (!data)
3279                                return -LIBBPF_ERRNO__FORMAT;
3280
3281                        idx = elf_ndxscn(scn);
3282
3283                        obj->efile.symbols = data;
3284                        obj->efile.symbols_shndx = idx;
3285                        obj->efile.strtabidx = sh->sh_link;
3286                }
3287        }
3288
3289        if (!obj->efile.symbols) {
3290                pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3291                        obj->path);
3292                return -ENOENT;
3293        }
3294
3295        scn = NULL;
3296        while ((scn = elf_nextscn(elf, scn)) != NULL) {
3297                idx = elf_ndxscn(scn);
3298                sec_desc = &obj->efile.secs[idx];
3299
3300                sh = elf_sec_hdr(obj, scn);
3301                if (!sh)
3302                        return -LIBBPF_ERRNO__FORMAT;
3303
3304                name = elf_sec_str(obj, sh->sh_name);
3305                if (!name)
3306                        return -LIBBPF_ERRNO__FORMAT;
3307
3308                if (ignore_elf_section(sh, name))
3309                        continue;
3310
3311                data = elf_sec_data(obj, scn);
3312                if (!data)
3313                        return -LIBBPF_ERRNO__FORMAT;
3314
3315                pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3316                         idx, name, (unsigned long)data->d_size,
3317                         (int)sh->sh_link, (unsigned long)sh->sh_flags,
3318                         (int)sh->sh_type);
3319
3320                if (strcmp(name, "license") == 0) {
3321                        err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3322                        if (err)
3323                                return err;
3324                } else if (strcmp(name, "version") == 0) {
3325                        err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3326                        if (err)
3327                                return err;
3328                } else if (strcmp(name, "maps") == 0) {
3329                        obj->efile.maps_shndx = idx;
3330                } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3331                        obj->efile.btf_maps_shndx = idx;
3332                } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3333                        if (sh->sh_type != SHT_PROGBITS)
3334                                return -LIBBPF_ERRNO__FORMAT;
3335                        btf_data = data;
3336                } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3337                        if (sh->sh_type != SHT_PROGBITS)
3338                                return -LIBBPF_ERRNO__FORMAT;
3339                        btf_ext_data = data;
3340                } else if (sh->sh_type == SHT_SYMTAB) {
3341                        /* already processed during the first pass above */
3342                } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3343                        if (sh->sh_flags & SHF_EXECINSTR) {
3344                                if (strcmp(name, ".text") == 0)
3345                                        obj->efile.text_shndx = idx;
3346                                err = bpf_object__add_programs(obj, data, name, idx);
3347                                if (err)
3348                                        return err;
3349                        } else if (strcmp(name, DATA_SEC) == 0 ||
3350                                   str_has_pfx(name, DATA_SEC ".")) {
3351                                sec_desc->sec_type = SEC_DATA;
3352                                sec_desc->shdr = sh;
3353                                sec_desc->data = data;
3354                        } else if (strcmp(name, RODATA_SEC) == 0 ||
3355                                   str_has_pfx(name, RODATA_SEC ".")) {
3356                                sec_desc->sec_type = SEC_RODATA;
3357                                sec_desc->shdr = sh;
3358                                sec_desc->data = data;
3359                        } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3360                                obj->efile.st_ops_data = data;
3361                                obj->efile.st_ops_shndx = idx;
3362                        } else {
3363                                pr_info("elf: skipping unrecognized data section(%d) %s\n",
3364                                        idx, name);
3365                        }
3366                } else if (sh->sh_type == SHT_REL) {
3367                        int targ_sec_idx = sh->sh_info; /* points to other section */
3368
3369                        if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3370                            targ_sec_idx >= obj->efile.sec_cnt)
3371                                return -LIBBPF_ERRNO__FORMAT;
3372
3373                        /* Only do relo for section with exec instructions */
3374                        if (!section_have_execinstr(obj, targ_sec_idx) &&
3375                            strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3376                            strcmp(name, ".rel" MAPS_ELF_SEC)) {
3377                                pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3378                                        idx, name, targ_sec_idx,
3379                                        elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3380                                continue;
3381                        }
3382
3383                        sec_desc->sec_type = SEC_RELO;
3384                        sec_desc->shdr = sh;
3385                        sec_desc->data = data;
3386                } else if (sh->sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) {
3387                        sec_desc->sec_type = SEC_BSS;
3388                        sec_desc->shdr = sh;
3389                        sec_desc->data = data;
3390                } else {
3391                        pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3392                                (size_t)sh->sh_size);
3393                }
3394        }
3395
3396        if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3397                pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3398                return -LIBBPF_ERRNO__FORMAT;
3399        }
3400
3401        /* sort BPF programs by section name and in-section instruction offset
3402         * for faster search */
3403        if (obj->nr_programs)
3404                qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3405
3406        return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3407}
3408
3409static bool sym_is_extern(const Elf64_Sym *sym)
3410{
3411        int bind = ELF64_ST_BIND(sym->st_info);
3412        /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3413        return sym->st_shndx == SHN_UNDEF &&
3414               (bind == STB_GLOBAL || bind == STB_WEAK) &&
3415               ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
3416}
3417
3418static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
3419{
3420        int bind = ELF64_ST_BIND(sym->st_info);
3421        int type = ELF64_ST_TYPE(sym->st_info);
3422
3423        /* in .text section */
3424        if (sym->st_shndx != text_shndx)
3425                return false;
3426
3427        /* local function */
3428        if (bind == STB_LOCAL && type == STT_SECTION)
3429                return true;
3430
3431        /* global function */
3432        return bind == STB_GLOBAL && type == STT_FUNC;
3433}
3434
3435static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3436{
3437        const struct btf_type *t;
3438        const char *tname;
3439        int i, n;
3440
3441        if (!btf)
3442                return -ESRCH;
3443
3444        n = btf__type_cnt(btf);
3445        for (i = 1; i < n; i++) {
3446                t = btf__type_by_id(btf, i);
3447
3448                if (!btf_is_var(t) && !btf_is_func(t))
3449                        continue;
3450
3451                tname = btf__name_by_offset(btf, t->name_off);
3452                if (strcmp(tname, ext_name))
3453                        continue;
3454
3455                if (btf_is_var(t) &&
3456                    btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3457                        return -EINVAL;
3458
3459                if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3460                        return -EINVAL;
3461
3462                return i;
3463        }
3464
3465        return -ENOENT;
3466}
3467
3468static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3469        const struct btf_var_secinfo *vs;
3470        const struct btf_type *t;
3471        int i, j, n;
3472
3473        if (!btf)
3474                return -ESRCH;
3475
3476        n = btf__type_cnt(btf);
3477        for (i = 1; i < n; i++) {
3478                t = btf__type_by_id(btf, i);
3479
3480                if (!btf_is_datasec(t))
3481                        continue;
3482
3483                vs = btf_var_secinfos(t);
3484                for (j = 0; j < btf_vlen(t); j++, vs++) {
3485                        if (vs->type == ext_btf_id)
3486                                return i;
3487                }
3488        }
3489
3490        return -ENOENT;
3491}
3492
3493static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3494                                     bool *is_signed)
3495{
3496        const struct btf_type *t;
3497        const char *name;
3498
3499        t = skip_mods_and_typedefs(btf, id, NULL);
3500        name = btf__name_by_offset(btf, t->name_off);
3501
3502        if (is_signed)
3503                *is_signed = false;
3504        switch (btf_kind(t)) {
3505        case BTF_KIND_INT: {
3506                int enc = btf_int_encoding(t);
3507
3508                if (enc & BTF_INT_BOOL)
3509                        return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3510                if (is_signed)
3511                        *is_signed = enc & BTF_INT_SIGNED;
3512                if (t->size == 1)
3513                        return KCFG_CHAR;
3514                if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3515                        return KCFG_UNKNOWN;
3516                return KCFG_INT;
3517        }
3518        case BTF_KIND_ENUM:
3519                if (t->size != 4)
3520                        return KCFG_UNKNOWN;
3521                if (strcmp(name, "libbpf_tristate"))
3522                        return KCFG_UNKNOWN;
3523                return KCFG_TRISTATE;
3524        case BTF_KIND_ARRAY:
3525                if (btf_array(t)->nelems == 0)
3526                        return KCFG_UNKNOWN;
3527                if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3528                        return KCFG_UNKNOWN;
3529                return KCFG_CHAR_ARR;
3530        default:
3531                return KCFG_UNKNOWN;
3532        }
3533}
3534
3535static int cmp_externs(const void *_a, const void *_b)
3536{
3537        const struct extern_desc *a = _a;
3538        const struct extern_desc *b = _b;
3539
3540        if (a->type != b->type)
3541                return a->type < b->type ? -1 : 1;
3542
3543        if (a->type == EXT_KCFG) {
3544                /* descending order by alignment requirements */
3545                if (a->kcfg.align != b->kcfg.align)
3546                        return a->kcfg.align > b->kcfg.align ? -1 : 1;
3547                /* ascending order by size, within same alignment class */
3548                if (a->kcfg.sz != b->kcfg.sz)
3549                        return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3550        }
3551
3552        /* resolve ties by name */
3553        return strcmp(a->name, b->name);
3554}
3555
3556static int find_int_btf_id(const struct btf *btf)
3557{
3558        const struct btf_type *t;
3559        int i, n;
3560
3561        n = btf__type_cnt(btf);
3562        for (i = 1; i < n; i++) {
3563                t = btf__type_by_id(btf, i);
3564
3565                if (btf_is_int(t) && btf_int_bits(t) == 32)
3566                        return i;
3567        }
3568
3569        return 0;
3570}
3571
3572static int add_dummy_ksym_var(struct btf *btf)
3573{
3574        int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
3575        const struct btf_var_secinfo *vs;
3576        const struct btf_type *sec;
3577
3578        if (!btf)
3579                return 0;
3580
3581        sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
3582                                            BTF_KIND_DATASEC);
3583        if (sec_btf_id < 0)
3584                return 0;
3585
3586        sec = btf__type_by_id(btf, sec_btf_id);
3587        vs = btf_var_secinfos(sec);
3588        for (i = 0; i < btf_vlen(sec); i++, vs++) {
3589                const struct btf_type *vt;
3590
3591                vt = btf__type_by_id(btf, vs->type);
3592                if (btf_is_func(vt))
3593                        break;
3594        }
3595
3596        /* No func in ksyms sec.  No need to add dummy var. */
3597        if (i == btf_vlen(sec))
3598                return 0;
3599
3600        int_btf_id = find_int_btf_id(btf);
3601        dummy_var_btf_id = btf__add_var(btf,
3602                                        "dummy_ksym",
3603                                        BTF_VAR_GLOBAL_ALLOCATED,
3604                                        int_btf_id);
3605        if (dummy_var_btf_id < 0)
3606                pr_warn("cannot create a dummy_ksym var\n");
3607
3608        return dummy_var_btf_id;
3609}
3610
3611static int bpf_object__collect_externs(struct bpf_object *obj)
3612{
3613        struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3614        const struct btf_type *t;
3615        struct extern_desc *ext;
3616        int i, n, off, dummy_var_btf_id;
3617        const char *ext_name, *sec_name;
3618        Elf_Scn *scn;
3619        Elf64_Shdr *sh;
3620
3621        if (!obj->efile.symbols)
3622                return 0;
3623
3624        scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3625        sh = elf_sec_hdr(obj, scn);
3626        if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
3627                return -LIBBPF_ERRNO__FORMAT;
3628
3629        dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
3630        if (dummy_var_btf_id < 0)
3631                return dummy_var_btf_id;
3632
3633        n = sh->sh_size / sh->sh_entsize;
3634        pr_debug("looking for externs among %d symbols...\n", n);
3635
3636        for (i = 0; i < n; i++) {
3637                Elf64_Sym *sym = elf_sym_by_idx(obj, i);
3638
3639                if (!sym)
3640                        return -LIBBPF_ERRNO__FORMAT;
3641                if (!sym_is_extern(sym))
3642                        continue;
3643                ext_name = elf_sym_str(obj, sym->st_name);
3644                if (!ext_name || !ext_name[0])
3645                        continue;
3646
3647                ext = obj->externs;
3648                ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3649                if (!ext)
3650                        return -ENOMEM;
3651                obj->externs = ext;
3652                ext = &ext[obj->nr_extern];
3653                memset(ext, 0, sizeof(*ext));
3654                obj->nr_extern++;
3655
3656                ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3657                if (ext->btf_id <= 0) {
3658                        pr_warn("failed to find BTF for extern '%s': %d\n",
3659                                ext_name, ext->btf_id);
3660                        return ext->btf_id;
3661                }
3662                t = btf__type_by_id(obj->btf, ext->btf_id);
3663                ext->name = btf__name_by_offset(obj->btf, t->name_off);
3664                ext->sym_idx = i;
3665                ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
3666
3667                ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3668                if (ext->sec_btf_id <= 0) {
3669                        pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3670                                ext_name, ext->btf_id, ext->sec_btf_id);
3671                        return ext->sec_btf_id;
3672                }
3673                sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3674                sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3675
3676                if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3677                        if (btf_is_func(t)) {
3678                                pr_warn("extern function %s is unsupported under %s section\n",
3679                                        ext->name, KCONFIG_SEC);
3680                                return -ENOTSUP;
3681                        }
3682                        kcfg_sec = sec;
3683                        ext->type = EXT_KCFG;
3684                        ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3685                        if (ext->kcfg.sz <= 0) {
3686                                pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3687                                        ext_name, ext->kcfg.sz);
3688                                return ext->kcfg.sz;
3689                        }
3690                        ext->kcfg.align = btf__align_of(obj->btf, t->type);
3691                        if (ext->kcfg.align <= 0) {
3692                                pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3693                                        ext_name, ext->kcfg.align);
3694                                return -EINVAL;
3695                        }
3696                        ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3697                                                        &ext->kcfg.is_signed);
3698                        if (ext->kcfg.type == KCFG_UNKNOWN) {
3699                                pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
3700                                return -ENOTSUP;
3701                        }
3702                } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3703                        ksym_sec = sec;
3704                        ext->type = EXT_KSYM;
3705                        skip_mods_and_typedefs(obj->btf, t->type,
3706                                               &ext->ksym.type_id);
3707                } else {
3708                        pr_warn("unrecognized extern section '%s'\n", sec_name);
3709                        return -ENOTSUP;
3710                }
3711        }
3712        pr_debug("collected %d externs total\n", obj->nr_extern);
3713
3714        if (!obj->nr_extern)
3715                return 0;
3716
3717        /* sort externs by type, for kcfg ones also by (align, size, name) */
3718        qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3719
3720        /* for .ksyms section, we need to turn all externs into allocated
3721         * variables in BTF to pass kernel verification; we do this by
3722         * pretending that each extern is a 8-byte variable
3723         */
3724        if (ksym_sec) {
3725                /* find existing 4-byte integer type in BTF to use for fake
3726                 * extern variables in DATASEC
3727                 */
3728                int int_btf_id = find_int_btf_id(obj->btf);
3729                /* For extern function, a dummy_var added earlier
3730                 * will be used to replace the vs->type and
3731                 * its name string will be used to refill
3732                 * the missing param's name.
3733                 */
3734                const struct btf_type *dummy_var;
3735
3736                dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
3737                for (i = 0; i < obj->nr_extern; i++) {
3738                        ext = &obj->externs[i];
3739                        if (ext->type != EXT_KSYM)
3740                                continue;
3741                        pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
3742                                 i, ext->sym_idx, ext->name);
3743                }
3744
3745                sec = ksym_sec;
3746                n = btf_vlen(sec);
3747                for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
3748                        struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3749                        struct btf_type *vt;
3750
3751                        vt = (void *)btf__type_by_id(obj->btf, vs->type);
3752                        ext_name = btf__name_by_offset(obj->btf, vt->name_off);
3753                        ext = find_extern_by_name(obj, ext_name);
3754                        if (!ext) {
3755                                pr_warn("failed to find extern definition for BTF %s '%s'\n",
3756                                        btf_kind_str(vt), ext_name);
3757                                return -ESRCH;
3758                        }
3759                        if (btf_is_func(vt)) {
3760                                const struct btf_type *func_proto;
3761                                struct btf_param *param;
3762                                int j;
3763
3764                                func_proto = btf__type_by_id(obj->btf,
3765                                                             vt->type);
3766                                param = btf_params(func_proto);
3767                                /* Reuse the dummy_var string if the
3768                                 * func proto does not have param name.
3769                                 */
3770                                for (j = 0; j < btf_vlen(func_proto); j++)
3771                                        if (param[j].type && !param[j].name_off)
3772                                                param[j].name_off =
3773                                                        dummy_var->name_off;
3774                                vs->type = dummy_var_btf_id;
3775                                vt->info &= ~0xffff;
3776                                vt->info |= BTF_FUNC_GLOBAL;
3777                        } else {
3778                                btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3779                                vt->type = int_btf_id;
3780                        }
3781                        vs->offset = off;
3782                        vs->size = sizeof(int);
3783                }
3784                sec->size = off;
3785        }
3786
3787        if (kcfg_sec) {
3788                sec = kcfg_sec;
3789                /* for kcfg externs calculate their offsets within a .kconfig map */
3790                off = 0;
3791                for (i = 0; i < obj->nr_extern; i++) {
3792                        ext = &obj->externs[i];
3793                        if (ext->type != EXT_KCFG)
3794                                continue;
3795
3796                        ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3797                        off = ext->kcfg.data_off + ext->kcfg.sz;
3798                        pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3799                                 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3800                }
3801                sec->size = off;
3802                n = btf_vlen(sec);
3803                for (i = 0; i < n; i++) {
3804                        struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3805
3806                        t = btf__type_by_id(obj->btf, vs->type);
3807                        ext_name = btf__name_by_offset(obj->btf, t->name_off);
3808                        ext = find_extern_by_name(obj, ext_name);
3809                        if (!ext) {
3810                                pr_warn("failed to find extern definition for BTF var '%s'\n",
3811                                        ext_name);
3812                                return -ESRCH;
3813                        }
3814                        btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3815                        vs->offset = ext->kcfg.data_off;
3816                }
3817        }
3818        return 0;
3819}
3820
3821struct bpf_program *
3822bpf_object__find_program_by_title(const struct bpf_object *obj,
3823                                  const char *title)
3824{
3825        struct bpf_program *pos;
3826
3827        bpf_object__for_each_program(pos, obj) {
3828                if (pos->sec_name && !strcmp(pos->sec_name, title))
3829                        return pos;
3830        }
3831        return errno = ENOENT, NULL;
3832}
3833
3834static bool prog_is_subprog(const struct bpf_object *obj,
3835                            const struct bpf_program *prog)
3836{
3837        /* For legacy reasons, libbpf supports an entry-point BPF programs
3838         * without SEC() attribute, i.e., those in the .text section. But if
3839         * there are 2 or more such programs in the .text section, they all
3840         * must be subprograms called from entry-point BPF programs in
3841         * designated SEC()'tions, otherwise there is no way to distinguish
3842         * which of those programs should be loaded vs which are a subprogram.
3843         * Similarly, if there is a function/program in .text and at least one
3844         * other BPF program with custom SEC() attribute, then we just assume
3845         * .text programs are subprograms (even if they are not called from
3846         * other programs), because libbpf never explicitly supported mixing
3847         * SEC()-designated BPF programs and .text entry-point BPF programs.
3848         *
3849         * In libbpf 1.0 strict mode, we always consider .text
3850         * programs to be subprograms.
3851         */
3852
3853        if (libbpf_mode & LIBBPF_STRICT_SEC_NAME)
3854                return prog->sec_idx == obj->efile.text_shndx;
3855
3856        return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
3857}
3858
3859struct bpf_program *
3860bpf_object__find_program_by_name(const struct bpf_object *obj,
3861                                 const char *name)
3862{
3863        struct bpf_program *prog;
3864
3865        bpf_object__for_each_program(prog, obj) {
3866                if (prog_is_subprog(obj, prog))
3867                        continue;
3868                if (!strcmp(prog->name, name))
3869                        return prog;
3870        }
3871        return errno = ENOENT, NULL;
3872}
3873
3874static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3875                                      int shndx)
3876{
3877        switch (obj->efile.secs[shndx].sec_type) {
3878        case SEC_BSS:
3879        case SEC_DATA:
3880        case SEC_RODATA:
3881                return true;
3882        default:
3883                return false;
3884        }
3885}
3886
3887static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3888                                      int shndx)
3889{
3890        return shndx == obj->efile.maps_shndx ||
3891               shndx == obj->efile.btf_maps_shndx;
3892}
3893
3894static enum libbpf_map_type
3895bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3896{
3897        if (shndx == obj->efile.symbols_shndx)
3898                return LIBBPF_MAP_KCONFIG;
3899
3900        switch (obj->efile.secs[shndx].sec_type) {
3901        case SEC_BSS:
3902                return LIBBPF_MAP_BSS;
3903        case SEC_DATA:
3904                return LIBBPF_MAP_DATA;
3905        case SEC_RODATA:
3906                return LIBBPF_MAP_RODATA;
3907        default:
3908                return LIBBPF_MAP_UNSPEC;
3909        }
3910}
3911
3912static int bpf_program__record_reloc(struct bpf_program *prog,
3913                                     struct reloc_desc *reloc_desc,
3914                                     __u32 insn_idx, const char *sym_name,
3915                                     const Elf64_Sym *sym, const Elf64_Rel *rel)
3916{
3917        struct bpf_insn *insn = &prog->insns[insn_idx];
3918        size_t map_idx, nr_maps = prog->obj->nr_maps;
3919        struct bpf_object *obj = prog->obj;
3920        __u32 shdr_idx = sym->st_shndx;
3921        enum libbpf_map_type type;
3922        const char *sym_sec_name;
3923        struct bpf_map *map;
3924
3925        if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
3926                pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
3927                        prog->name, sym_name, insn_idx, insn->code);
3928                return -LIBBPF_ERRNO__RELOC;
3929        }
3930
3931        if (sym_is_extern(sym)) {
3932                int sym_idx = ELF64_R_SYM(rel->r_info);
3933                int i, n = obj->nr_extern;
3934                struct extern_desc *ext;
3935
3936                for (i = 0; i < n; i++) {
3937                        ext = &obj->externs[i];
3938                        if (ext->sym_idx == sym_idx)
3939                                break;
3940                }
3941                if (i >= n) {
3942                        pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
3943                                prog->name, sym_name, sym_idx);
3944                        return -LIBBPF_ERRNO__RELOC;
3945                }
3946                pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
3947                         prog->name, i, ext->name, ext->sym_idx, insn_idx);
3948                if (insn->code == (BPF_JMP | BPF_CALL))
3949                        reloc_desc->type = RELO_EXTERN_FUNC;
3950                else
3951                        reloc_desc->type = RELO_EXTERN_VAR;
3952                reloc_desc->insn_idx = insn_idx;
3953                reloc_desc->sym_off = i; /* sym_off stores extern index */
3954                return 0;
3955        }
3956
3957        /* sub-program call relocation */
3958        if (is_call_insn(insn)) {
3959                if (insn->src_reg != BPF_PSEUDO_CALL) {
3960                        pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
3961                        return -LIBBPF_ERRNO__RELOC;
3962                }
3963                /* text_shndx can be 0, if no default "main" program exists */
3964                if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3965                        sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3966                        pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
3967                                prog->name, sym_name, sym_sec_name);
3968                        return -LIBBPF_ERRNO__RELOC;
3969                }
3970                if (sym->st_value % BPF_INSN_SZ) {
3971                        pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
3972                                prog->name, sym_name, (size_t)sym->st_value);
3973                        return -LIBBPF_ERRNO__RELOC;
3974                }
3975                reloc_desc->type = RELO_CALL;
3976                reloc_desc->insn_idx = insn_idx;
3977                reloc_desc->sym_off = sym->st_value;
3978                return 0;
3979        }
3980
3981        if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3982                pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
3983                        prog->name, sym_name, shdr_idx);
3984                return -LIBBPF_ERRNO__RELOC;
3985        }
3986
3987        /* loading subprog addresses */
3988        if (sym_is_subprog(sym, obj->efile.text_shndx)) {
3989                /* global_func: sym->st_value = offset in the section, insn->imm = 0.
3990                 * local_func: sym->st_value = 0, insn->imm = offset in the section.
3991                 */
3992                if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
3993                        pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
3994                                prog->name, sym_name, (size_t)sym->st_value, insn->imm);
3995                        return -LIBBPF_ERRNO__RELOC;
3996                }
3997
3998                reloc_desc->type = RELO_SUBPROG_ADDR;
3999                reloc_desc->insn_idx = insn_idx;
4000                reloc_desc->sym_off = sym->st_value;
4001                return 0;
4002        }
4003
4004        type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4005        sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4006
4007        /* generic map reference relocation */
4008        if (type == LIBBPF_MAP_UNSPEC) {
4009                if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4010                        pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4011                                prog->name, sym_name, sym_sec_name);
4012                        return -LIBBPF_ERRNO__RELOC;
4013                }
4014                for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4015                        map = &obj->maps[map_idx];
4016                        if (map->libbpf_type != type ||
4017                            map->sec_idx != sym->st_shndx ||
4018                            map->sec_offset != sym->st_value)
4019                                continue;
4020                        pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4021                                 prog->name, map_idx, map->name, map->sec_idx,
4022                                 map->sec_offset, insn_idx);
4023                        break;
4024                }
4025                if (map_idx >= nr_maps) {
4026                        pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4027                                prog->name, sym_sec_name, (size_t)sym->st_value);
4028                        return -LIBBPF_ERRNO__RELOC;
4029                }
4030                reloc_desc->type = RELO_LD64;
4031                reloc_desc->insn_idx = insn_idx;
4032                reloc_desc->map_idx = map_idx;
4033                reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4034                return 0;
4035        }
4036
4037        /* global data map relocation */
4038        if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4039                pr_warn("prog '%s': bad data relo against section '%s'\n",
4040                        prog->name, sym_sec_name);
4041                return -LIBBPF_ERRNO__RELOC;
4042        }
4043        for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4044                map = &obj->maps[map_idx];
4045                if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4046                        continue;
4047                pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4048                         prog->name, map_idx, map->name, map->sec_idx,
4049                         map->sec_offset, insn_idx);
4050                break;
4051        }
4052        if (map_idx >= nr_maps) {
4053                pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4054                        prog->name, sym_sec_name);
4055                return -LIBBPF_ERRNO__RELOC;
4056        }
4057
4058        reloc_desc->type = RELO_DATA;
4059        reloc_desc->insn_idx = insn_idx;
4060        reloc_desc->map_idx = map_idx;
4061        reloc_desc->sym_off = sym->st_value;
4062        return 0;
4063}
4064
4065static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4066{
4067        return insn_idx >= prog->sec_insn_off &&
4068               insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4069}
4070
4071static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4072                                                 size_t sec_idx, size_t insn_idx)
4073{
4074        int l = 0, r = obj->nr_programs - 1, m;
4075        struct bpf_program *prog;
4076
4077        while (l < r) {
4078                m = l + (r - l + 1) / 2;
4079                prog = &obj->programs[m];
4080
4081                if (prog->sec_idx < sec_idx ||
4082                    (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4083                        l = m;
4084                else
4085                        r = m - 1;
4086        }
4087        /* matching program could be at index l, but it still might be the
4088         * wrong one, so we need to double check conditions for the last time
4089         */
4090        prog = &obj->programs[l];
4091        if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4092                return prog;
4093        return NULL;
4094}
4095
4096static int
4097bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4098{
4099        const char *relo_sec_name, *sec_name;
4100        size_t sec_idx = shdr->sh_info, sym_idx;
4101        struct bpf_program *prog;
4102        struct reloc_desc *relos;
4103        int err, i, nrels;
4104        const char *sym_name;
4105        __u32 insn_idx;
4106        Elf_Scn *scn;
4107        Elf_Data *scn_data;
4108        Elf64_Sym *sym;
4109        Elf64_Rel *rel;
4110
4111        if (sec_idx >= obj->efile.sec_cnt)
4112                return -EINVAL;
4113
4114        scn = elf_sec_by_idx(obj, sec_idx);
4115        scn_data = elf_sec_data(obj, scn);
4116
4117        relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4118        sec_name = elf_sec_name(obj, scn);
4119        if (!relo_sec_name || !sec_name)
4120                return -EINVAL;
4121
4122        pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4123                 relo_sec_name, sec_idx, sec_name);
4124        nrels = shdr->sh_size / shdr->sh_entsize;
4125
4126        for (i = 0; i < nrels; i++) {
4127                rel = elf_rel_by_idx(data, i);
4128                if (!rel) {
4129                        pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4130                        return -LIBBPF_ERRNO__FORMAT;
4131                }
4132
4133                sym_idx = ELF64_R_SYM(rel->r_info);
4134                sym = elf_sym_by_idx(obj, sym_idx);
4135                if (!sym) {
4136                        pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4137                                relo_sec_name, sym_idx, i);
4138                        return -LIBBPF_ERRNO__FORMAT;
4139                }
4140
4141                if (sym->st_shndx >= obj->efile.sec_cnt) {
4142                        pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4143                                relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4144                        return -LIBBPF_ERRNO__FORMAT;
4145                }
4146
4147                if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4148                        pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4149                                relo_sec_name, (size_t)rel->r_offset, i);
4150                        return -LIBBPF_ERRNO__FORMAT;
4151                }
4152
4153                insn_idx = rel->r_offset / BPF_INSN_SZ;
4154                /* relocations against static functions are recorded as
4155                 * relocations against the section that contains a function;
4156                 * in such case, symbol will be STT_SECTION and sym.st_name
4157                 * will point to empty string (0), so fetch section name
4158                 * instead
4159                 */
4160                if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4161                        sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4162                else
4163                        sym_name = elf_sym_str(obj, sym->st_name);
4164                sym_name = sym_name ?: "<?";
4165
4166                pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4167                         relo_sec_name, i, insn_idx, sym_name);
4168
4169                prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4170                if (!prog) {
4171                        pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4172                                relo_sec_name, i, sec_name, insn_idx);
4173                        continue;
4174                }
4175
4176                relos = libbpf_reallocarray(prog->reloc_desc,
4177                                            prog->nr_reloc + 1, sizeof(*relos));
4178                if (!relos)
4179                        return -ENOMEM;
4180                prog->reloc_desc = relos;
4181
4182                /* adjust insn_idx to local BPF program frame of reference */
4183                insn_idx -= prog->sec_insn_off;
4184                err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4185                                                insn_idx, sym_name, sym, rel);
4186                if (err)
4187                        return err;
4188
4189                prog->nr_reloc++;
4190        }
4191        return 0;
4192}
4193
4194static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
4195{
4196        struct bpf_map_def *def = &map->def;
4197        __u32 key_type_id = 0, value_type_id = 0;
4198        int ret;
4199
4200        if (!obj->btf)
4201                return -ENOENT;
4202
4203        /* if it's BTF-defined map, we don't need to search for type IDs.
4204         * For struct_ops map, it does not need btf_key_type_id and
4205         * btf_value_type_id.
4206         */
4207        if (map->sec_idx == obj->efile.btf_maps_shndx ||
4208            bpf_map__is_struct_ops(map))
4209                return 0;
4210
4211        if (!bpf_map__is_internal(map)) {
4212                pr_warn("Use of BPF_ANNOTATE_KV_PAIR is deprecated, use BTF-defined maps in .maps section instead\n");
4213#pragma GCC diagnostic push
4214#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
4215                ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
4216                                           def->value_size, &key_type_id,
4217                                           &value_type_id);
4218#pragma GCC diagnostic pop
4219        } else {
4220                /*
4221                 * LLVM annotates global data differently in BTF, that is,
4222                 * only as '.data', '.bss' or '.rodata'.
4223                 */
4224                ret = btf__find_by_name(obj->btf, map->real_name);
4225        }
4226        if (ret < 0)
4227                return ret;
4228
4229        map->btf_key_type_id = key_type_id;
4230        map->btf_value_type_id = bpf_map__is_internal(map) ?
4231                                 ret : value_type_id;
4232        return 0;
4233}
4234
4235static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4236{
4237        char file[PATH_MAX], buff[4096];
4238        FILE *fp;
4239        __u32 val;
4240        int err;
4241
4242        snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4243        memset(info, 0, sizeof(*info));
4244
4245        fp = fopen(file, "r");
4246        if (!fp) {
4247                err = -errno;
4248                pr_warn("failed to open %s: %d. No procfs support?\n", file,
4249                        err);
4250                return err;
4251        }
4252
4253        while (fgets(buff, sizeof(buff), fp)) {
4254                if (sscanf(buff, "map_type:\t%u", &val) == 1)
4255                        info->type = val;
4256                else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4257                        info->key_size = val;
4258                else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4259                        info->value_size = val;
4260                else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4261                        info->max_entries = val;
4262                else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4263                        info->map_flags = val;
4264        }
4265
4266        fclose(fp);
4267
4268        return 0;
4269}
4270
4271int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4272{
4273        struct bpf_map_info info = {};
4274        __u32 len = sizeof(info);
4275        int new_fd, err;
4276        char *new_name;
4277
4278        err = bpf_obj_get_info_by_fd(fd, &info, &len);
4279        if (err && errno == EINVAL)
4280                err = bpf_get_map_info_from_fdinfo(fd, &info);
4281        if (err)
4282                return libbpf_err(err);
4283
4284        new_name = strdup(info.name);
4285        if (!new_name)
4286                return libbpf_err(-errno);
4287
4288        new_fd = open("/", O_RDONLY | O_CLOEXEC);
4289        if (new_fd < 0) {
4290                err = -errno;
4291                goto err_free_new_name;
4292        }
4293
4294        new_fd = dup3(fd, new_fd, O_CLOEXEC);
4295        if (new_fd < 0) {
4296                err = -errno;
4297                goto err_close_new_fd;
4298        }
4299
4300        err = zclose(map->fd);
4301        if (err) {
4302                err = -errno;
4303                goto err_close_new_fd;
4304        }
4305        free(map->name);
4306
4307        map->fd = new_fd;
4308        map->name = new_name;
4309        map->def.type = info.type;
4310        map->def.key_size = info.key_size;
4311        map->def.value_size = info.value_size;
4312        map->def.max_entries = info.max_entries;
4313        map->def.map_flags = info.map_flags;
4314        map->btf_key_type_id = info.btf_key_type_id;
4315        map->btf_value_type_id = info.btf_value_type_id;
4316        map->reused = true;
4317        map->map_extra = info.map_extra;
4318
4319        return 0;
4320
4321err_close_new_fd:
4322        close(new_fd);
4323err_free_new_name:
4324        free(new_name);
4325        return libbpf_err(err);
4326}
4327
4328__u32 bpf_map__max_entries(const struct bpf_map *map)
4329{
4330        return map->def.max_entries;
4331}
4332
4333struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4334{
4335        if (!bpf_map_type__is_map_in_map(map->def.type))
4336                return errno = EINVAL, NULL;
4337
4338        return map->inner_map;
4339}
4340
4341int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4342{
4343        if (map->fd >= 0)
4344                return libbpf_err(-EBUSY);
4345        map->def.max_entries = max_entries;
4346        return 0;
4347}
4348
4349int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
4350{
4351        if (!map || !max_entries)
4352                return libbpf_err(-EINVAL);
4353
4354        return bpf_map__set_max_entries(map, max_entries);
4355}
4356
4357static int
4358bpf_object__probe_loading(struct bpf_object *obj)
4359{
4360        char *cp, errmsg[STRERR_BUFSIZE];
4361        struct bpf_insn insns[] = {
4362                BPF_MOV64_IMM(BPF_REG_0, 0),
4363                BPF_EXIT_INSN(),
4364        };
4365        int ret, insn_cnt = ARRAY_SIZE(insns);
4366
4367        if (obj->gen_loader)
4368                return 0;
4369
4370        ret = bump_rlimit_memlock();
4371        if (ret)
4372                pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
4373
4374        /* make sure basic loading works */
4375        ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4376        if (ret < 0)
4377                ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4378        if (ret < 0) {
4379                ret = errno;
4380                cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4381                pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4382                        "program. Make sure your kernel supports BPF "
4383                        "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4384                        "set to big enough value.\n", __func__, cp, ret);
4385                return -ret;
4386        }
4387        close(ret);
4388
4389        return 0;
4390}
4391
4392static int probe_fd(int fd)
4393{
4394        if (fd >= 0)
4395                close(fd);
4396        return fd >= 0;
4397}
4398
4399static int probe_kern_prog_name(void)
4400{
4401        struct bpf_insn insns[] = {
4402                BPF_MOV64_IMM(BPF_REG_0, 0),
4403                BPF_EXIT_INSN(),
4404        };
4405        int ret, insn_cnt = ARRAY_SIZE(insns);
4406
4407        /* make sure loading with name works */
4408        ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "test", "GPL", insns, insn_cnt, NULL);
4409        return probe_fd(ret);
4410}
4411
4412static int probe_kern_global_data(void)
4413{
4414        char *cp, errmsg[STRERR_BUFSIZE];
4415        struct bpf_insn insns[] = {
4416                BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4417                BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4418                BPF_MOV64_IMM(BPF_REG_0, 0),
4419                BPF_EXIT_INSN(),
4420        };
4421        int ret, map, insn_cnt = ARRAY_SIZE(insns);
4422
4423        map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL);
4424        if (map < 0) {
4425                ret = -errno;
4426                cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4427                pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4428                        __func__, cp, -ret);
4429                return ret;
4430        }
4431
4432        insns[0].imm = map;
4433
4434        ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4435        close(map);
4436        return probe_fd(ret);
4437}
4438
4439static int probe_kern_btf(void)
4440{
4441        static const char strs[] = "\0int";
4442        __u32 types[] = {
4443                /* int */
4444                BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4445        };
4446
4447        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4448                                             strs, sizeof(strs)));
4449}
4450
4451static int probe_kern_btf_func(void)
4452{
4453        static const char strs[] = "\0int\0x\0a";
4454        /* void x(int a) {} */
4455        __u32 types[] = {
4456                /* int */
4457                BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4458                /* FUNC_PROTO */                                /* [2] */
4459                BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4460                BTF_PARAM_ENC(7, 1),
4461                /* FUNC x */                                    /* [3] */
4462                BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4463        };
4464
4465        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4466                                             strs, sizeof(strs)));
4467}
4468
4469static int probe_kern_btf_func_global(void)
4470{
4471        static const char strs[] = "\0int\0x\0a";
4472        /* static void x(int a) {} */
4473        __u32 types[] = {
4474                /* int */
4475                BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4476                /* FUNC_PROTO */                                /* [2] */
4477                BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4478                BTF_PARAM_ENC(7, 1),
4479                /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
4480                BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4481        };
4482
4483        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4484                                             strs, sizeof(strs)));
4485}
4486
4487static int probe_kern_btf_datasec(void)
4488{
4489        static const char strs[] = "\0x\0.data";
4490        /* static int a; */
4491        __u32 types[] = {
4492                /* int */
4493                BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4494                /* VAR x */                                     /* [2] */
4495                BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4496                BTF_VAR_STATIC,
4497                /* DATASEC val */                               /* [3] */
4498                BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4499                BTF_VAR_SECINFO_ENC(2, 0, 4),
4500        };
4501
4502        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4503                                             strs, sizeof(strs)));
4504}
4505
4506static int probe_kern_btf_float(void)
4507{
4508        static const char strs[] = "\0float";
4509        __u32 types[] = {
4510                /* float */
4511                BTF_TYPE_FLOAT_ENC(1, 4),
4512        };
4513
4514        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4515                                             strs, sizeof(strs)));
4516}
4517
4518static int probe_kern_btf_decl_tag(void)
4519{
4520        static const char strs[] = "\0tag";
4521        __u32 types[] = {
4522                /* int */
4523                BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4524                /* VAR x */                                     /* [2] */
4525                BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4526                BTF_VAR_STATIC,
4527                /* attr */
4528                BTF_TYPE_DECL_TAG_ENC(1, 2, -1),
4529        };
4530
4531        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4532                                             strs, sizeof(strs)));
4533}
4534
4535static int probe_kern_btf_type_tag(void)
4536{
4537        static const char strs[] = "\0tag";
4538        __u32 types[] = {
4539                /* int */
4540                BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),          /* [1] */
4541                /* attr */
4542                BTF_TYPE_TYPE_TAG_ENC(1, 1),                            /* [2] */
4543                /* ptr */
4544                BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2),   /* [3] */
4545        };
4546
4547        return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4548                                             strs, sizeof(strs)));
4549}
4550
4551static int probe_kern_array_mmap(void)
4552{
4553        LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
4554        int fd;
4555
4556        fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), sizeof(int), 1, &opts);
4557        return probe_fd(fd);
4558}
4559
4560static int probe_kern_exp_attach_type(void)
4561{
4562        LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE);
4563        struct bpf_insn insns[] = {
4564                BPF_MOV64_IMM(BPF_REG_0, 0),
4565                BPF_EXIT_INSN(),
4566        };
4567        int fd, insn_cnt = ARRAY_SIZE(insns);
4568
4569        /* use any valid combination of program type and (optional)
4570         * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
4571         * to see if kernel supports expected_attach_type field for
4572         * BPF_PROG_LOAD command
4573         */
4574        fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts);
4575        return probe_fd(fd);
4576}
4577
4578static int probe_kern_probe_read_kernel(void)
4579{
4580        struct bpf_insn insns[] = {
4581                BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),   /* r1 = r10 (fp) */
4582                BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8),  /* r1 += -8 */
4583                BPF_MOV64_IMM(BPF_REG_2, 8),            /* r2 = 8 */
4584                BPF_MOV64_IMM(BPF_REG_3, 0),            /* r3 = 0 */
4585                BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
4586                BPF_EXIT_INSN(),
4587        };
4588        int fd, insn_cnt = ARRAY_SIZE(insns);
4589
4590        fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL);
4591        return probe_fd(fd);
4592}
4593
4594static int probe_prog_bind_map(void)
4595{
4596        char *cp, errmsg[STRERR_BUFSIZE];
4597        struct bpf_insn insns[] = {
4598                BPF_MOV64_IMM(BPF_REG_0, 0),
4599                BPF_EXIT_INSN(),
4600        };
4601        int ret, map, prog, insn_cnt = ARRAY_SIZE(insns);
4602
4603        map = bpf_map_create(BPF_MAP_TYPE_ARRAY, NULL, sizeof(int), 32, 1, NULL);
4604        if (map < 0) {
4605                ret = -errno;
4606                cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4607                pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4608                        __func__, cp, -ret);
4609                return ret;
4610        }
4611
4612        prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4613        if (prog < 0) {
4614                close(map);
4615                return 0;
4616        }
4617
4618        ret = bpf_prog_bind_map(prog, map, NULL);
4619
4620        close(map);
4621        close(prog);
4622
4623        return ret >= 0;
4624}
4625
4626static int probe_module_btf(void)
4627{
4628        static const char strs[] = "\0int";
4629        __u32 types[] = {
4630                /* int */
4631                BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4632        };
4633        struct bpf_btf_info info;
4634        __u32 len = sizeof(info);
4635        char name[16];
4636        int fd, err;
4637
4638        fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4639        if (fd < 0)
4640                return 0; /* BTF not supported at all */
4641
4642        memset(&info, 0, sizeof(info));
4643        info.name = ptr_to_u64(name);
4644        info.name_len = sizeof(name);
4645
4646        /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4647         * kernel's module BTF support coincides with support for
4648         * name/name_len fields in struct bpf_btf_info.
4649         */
4650        err = bpf_obj_get_info_by_fd(fd, &info, &len);
4651        close(fd);
4652        return !err;
4653}
4654
4655static int probe_perf_link(void)
4656{
4657        struct bpf_insn insns[] = {
4658                BPF_MOV64_IMM(BPF_REG_0, 0),
4659                BPF_EXIT_INSN(),
4660        };
4661        int prog_fd, link_fd, err;
4662
4663        prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL",
4664                                insns, ARRAY_SIZE(insns), NULL);
4665        if (prog_fd < 0)
4666                return -errno;
4667
4668        /* use invalid perf_event FD to get EBADF, if link is supported;
4669         * otherwise EINVAL should be returned
4670         */
4671        link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
4672        err = -errno; /* close() can clobber errno */
4673
4674        if (link_fd >= 0)
4675                close(link_fd);
4676        close(prog_fd);
4677
4678        return link_fd < 0 && err == -EBADF;
4679}
4680
4681enum kern_feature_result {
4682        FEAT_UNKNOWN = 0,
4683        FEAT_SUPPORTED = 1,
4684        FEAT_MISSING = 2,
4685};
4686
4687typedef int (*feature_probe_fn)(void);
4688
4689static struct kern_feature_desc {
4690        const char *desc;
4691        feature_probe_fn probe;
4692        enum kern_feature_result res;
4693} feature_probes[__FEAT_CNT] = {
4694        [FEAT_PROG_NAME] = {
4695                "BPF program name", probe_kern_prog_name,
4696        },
4697        [FEAT_GLOBAL_DATA] = {
4698                "global variables", probe_kern_global_data,
4699        },
4700        [FEAT_BTF] = {
4701                "minimal BTF", probe_kern_btf,
4702        },
4703        [FEAT_BTF_FUNC] = {
4704                "BTF functions", probe_kern_btf_func,
4705        },
4706        [FEAT_BTF_GLOBAL_FUNC] = {
4707                "BTF global function", probe_kern_btf_func_global,
4708        },
4709        [FEAT_BTF_DATASEC] = {
4710                "BTF data section and variable", probe_kern_btf_datasec,
4711        },
4712        [FEAT_ARRAY_MMAP] = {
4713                "ARRAY map mmap()", probe_kern_array_mmap,
4714        },
4715        [FEAT_EXP_ATTACH_TYPE] = {
4716                "BPF_PROG_LOAD expected_attach_type attribute",
4717                probe_kern_exp_attach_type,
4718        },
4719        [FEAT_PROBE_READ_KERN] = {
4720                "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
4721        },
4722        [FEAT_PROG_BIND_MAP] = {
4723                "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
4724        },
4725        [FEAT_MODULE_BTF] = {
4726                "module BTF support", probe_module_btf,
4727        },
4728        [FEAT_BTF_FLOAT] = {
4729                "BTF_KIND_FLOAT support", probe_kern_btf_float,
4730        },
4731        [FEAT_PERF_LINK] = {
4732                "BPF perf link support", probe_perf_link,
4733        },
4734        [FEAT_BTF_DECL_TAG] = {
4735                "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag,
4736        },
4737        [FEAT_BTF_TYPE_TAG] = {
4738                "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
4739        },
4740        [FEAT_MEMCG_ACCOUNT] = {
4741                "memcg-based memory accounting", probe_memcg_account,
4742        },
4743};
4744
4745bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
4746{
4747        struct kern_feature_desc *feat = &feature_probes[feat_id];
4748        int ret;
4749
4750        if (obj && obj->gen_loader)
4751                /* To generate loader program assume the latest kernel
4752                 * to avoid doing extra prog_load, map_create syscalls.
4753                 */
4754                return true;
4755
4756        if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
4757                ret = feat->probe();
4758                if (ret > 0) {
4759                        WRITE_ONCE(feat->res, FEAT_SUPPORTED);
4760                } else if (ret == 0) {
4761                        WRITE_ONCE(feat->res, FEAT_MISSING);
4762                } else {
4763                        pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
4764                        WRITE_ONCE(feat->res, FEAT_MISSING);
4765                }
4766        }
4767
4768        return READ_ONCE(feat->res) == FEAT_SUPPORTED;
4769}
4770
4771static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
4772{
4773        struct bpf_map_info map_info = {};
4774        char msg[STRERR_BUFSIZE];
4775        __u32 map_info_len;
4776        int err;
4777
4778        map_info_len = sizeof(map_info);
4779
4780        err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len);
4781        if (err && errno == EINVAL)
4782                err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
4783        if (err) {
4784                pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
4785                        libbpf_strerror_r(errno, msg, sizeof(msg)));
4786                return false;
4787        }
4788
4789        return (map_info.type == map->def.type &&
4790                map_info.key_size == map->def.key_size &&
4791                map_info.value_size == map->def.value_size &&
4792                map_info.max_entries == map->def.max_entries &&
4793                map_info.map_flags == map->def.map_flags &&
4794                map_info.map_extra == map->map_extra);
4795}
4796
4797static int
4798bpf_object__reuse_map(struct bpf_map *map)
4799{
4800        char *cp, errmsg[STRERR_BUFSIZE];
4801        int err, pin_fd;
4802
4803        pin_fd = bpf_obj_get(map->pin_path);
4804        if (pin_fd < 0) {
4805                err = -errno;
4806                if (err == -ENOENT) {
4807                        pr_debug("found no pinned map to reuse at '%s'\n",
4808                                 map->pin_path);
4809                        return 0;
4810                }
4811
4812                cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4813                pr_warn("couldn't retrieve pinned map '%s': %s\n",
4814                        map->pin_path, cp);
4815                return err;
4816        }
4817
4818        if (!map_is_reuse_compat(map, pin_fd)) {
4819                pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
4820                        map->pin_path);
4821                close(pin_fd);
4822                return -EINVAL;
4823        }
4824
4825        err = bpf_map__reuse_fd(map, pin_fd);
4826        close(pin_fd);
4827        if (err) {
4828                return err;
4829        }
4830        map->pinned = true;
4831        pr_debug("reused pinned map at '%s'\n", map->pin_path);
4832
4833        return 0;
4834}
4835
4836static int
4837bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
4838{
4839        enum libbpf_map_type map_type = map->libbpf_type;
4840        char *cp, errmsg[STRERR_BUFSIZE];
4841        int err, zero = 0;
4842
4843        if (obj->gen_loader) {
4844                bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
4845                                         map->mmaped, map->def.value_size);
4846                if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
4847                        bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
4848                return 0;
4849        }
4850        err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
4851        if (err) {
4852                err = -errno;
4853                cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4854                pr_warn("Error setting initial map(%s) contents: %s\n",
4855                        map->name, cp);
4856                return err;
4857        }
4858
4859        /* Freeze .rodata and .kconfig map as read-only from syscall side. */
4860        if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
4861                err = bpf_map_freeze(map->fd);
4862                if (err) {
4863                        err = -errno;
4864                        cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4865                        pr_warn("Error freezing map(%s) as read-only: %s\n",
4866                                map->name, cp);
4867                        return err;
4868                }
4869        }
4870        return 0;
4871}
4872
4873static void bpf_map__destroy(struct bpf_map *map);
4874
4875static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
4876{
4877        LIBBPF_OPTS(bpf_map_create_opts, create_attr);
4878        struct bpf_map_def *def = &map->def;
4879        const char *map_name = NULL;
4880        int err = 0;
4881
4882        if (kernel_supports(obj, FEAT_PROG_NAME))
4883                map_name = map->name;
4884        create_attr.map_ifindex = map->map_ifindex;
4885        create_attr.map_flags = def->map_flags;
4886        create_attr.numa_node = map->numa_node;
4887        create_attr.map_extra = map->map_extra;
4888
4889        if (bpf_map__is_struct_ops(map))
4890                create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4891
4892        if (obj->btf && btf__fd(obj->btf) >= 0) {
4893                create_attr.btf_fd = btf__fd(obj->btf);
4894                create_attr.btf_key_type_id = map->btf_key_type_id;
4895                create_attr.btf_value_type_id = map->btf_value_type_id;
4896        }
4897
4898        if (bpf_map_type__is_map_in_map(def->type)) {
4899                if (map->inner_map) {
4900                        err = bpf_object__create_map(obj, map->inner_map, true);
4901                        if (err) {
4902                                pr_warn("map '%s': failed to create inner map: %d\n",
4903                                        map->name, err);
4904                                return err;
4905                        }
4906                        map->inner_map_fd = bpf_map__fd(map->inner_map);
4907                }
4908                if (map->inner_map_fd >= 0)
4909                        create_attr.inner_map_fd = map->inner_map_fd;
4910        }
4911
4912        switch (def->type) {
4913        case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
4914        case BPF_MAP_TYPE_CGROUP_ARRAY:
4915        case BPF_MAP_TYPE_STACK_TRACE:
4916        case BPF_MAP_TYPE_ARRAY_OF_MAPS:
4917        case BPF_MAP_TYPE_HASH_OF_MAPS:
4918        case BPF_MAP_TYPE_DEVMAP:
4919        case BPF_MAP_TYPE_DEVMAP_HASH:
4920        case BPF_MAP_TYPE_CPUMAP:
4921        case BPF_MAP_TYPE_XSKMAP:
4922        case BPF_MAP_TYPE_SOCKMAP:
4923        case BPF_MAP_TYPE_SOCKHASH:
4924        case BPF_MAP_TYPE_QUEUE:
4925        case BPF_MAP_TYPE_STACK:
4926        case BPF_MAP_TYPE_RINGBUF:
4927                create_attr.btf_fd = 0;
4928                create_attr.btf_key_type_id = 0;
4929                create_attr.btf_value_type_id = 0;
4930                map->btf_key_type_id = 0;
4931                map->btf_value_type_id = 0;
4932        default:
4933                break;
4934        }
4935
4936        if (obj->gen_loader) {
4937                bpf_gen__map_create(obj->gen_loader, def->type, map_name,
4938                                    def->key_size, def->value_size, def->max_entries,
4939                                    &create_attr, is_inner ? -1 : map - obj->maps);
4940                /* Pretend to have valid FD to pass various fd >= 0 checks.
4941                 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
4942                 */
4943                map->fd = 0;
4944        } else {
4945                map->fd = bpf_map_create(def->type, map_name,
4946                                         def->key_size, def->value_size,
4947                                         def->max_entries, &create_attr);
4948        }
4949        if (map->fd < 0 && (create_attr.btf_key_type_id ||
4950                            create_attr.btf_value_type_id)) {
4951                char *cp, errmsg[STRERR_BUFSIZE];
4952
4953                err = -errno;
4954                cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4955                pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
4956                        map->name, cp, err);
4957                create_attr.btf_fd = 0;
4958                create_attr.btf_key_type_id = 0;
4959                create_attr.btf_value_type_id = 0;
4960                map->btf_key_type_id = 0;
4961                map->btf_value_type_id = 0;
4962                map->fd = bpf_map_create(def->type, map_name,
4963                                         def->key_size, def->value_size,
4964                                         def->max_entries, &create_attr);
4965        }
4966
4967        err = map->fd < 0 ? -errno : 0;
4968
4969        if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
4970                if (obj->gen_loader)
4971                        map->inner_map->fd = -1;
4972                bpf_map__destroy(map->inner_map);
4973                zfree(&map->inner_map);
4974        }
4975
4976        return err;
4977}
4978
4979static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
4980{
4981        const struct bpf_map *targ_map;
4982        unsigned int i;
4983        int fd, err = 0;
4984
4985        for (i = 0; i < map->init_slots_sz; i++) {
4986                if (!map->init_slots[i])
4987                        continue;
4988
4989                targ_map = map->init_slots[i];
4990                fd = bpf_map__fd(targ_map);
4991
4992                if (obj->gen_loader) {
4993                        bpf_gen__populate_outer_map(obj->gen_loader,
4994                                                    map - obj->maps, i,
4995                                                    targ_map - obj->maps);
4996                } else {
4997                        err = bpf_map_update_elem(map->fd, &i, &fd, 0);
4998                }
4999                if (err) {
5000                        err = -errno;
5001                        pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
5002                                map->name, i, targ_map->name, fd, err);
5003                        return err;
5004                }
5005                pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5006                         map->name, i, targ_map->name, fd);
5007        }
5008
5009        zfree(&map->init_slots);
5010        map->init_slots_sz = 0;
5011
5012        return 0;
5013}
5014
5015static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5016{
5017        const struct bpf_program *targ_prog;
5018        unsigned int i;
5019        int fd, err;
5020
5021        if (obj->gen_loader)
5022                return -ENOTSUP;
5023
5024        for (i = 0; i < map->init_slots_sz; i++) {
5025                if (!map->init_slots[i])
5026                        continue;
5027
5028                targ_prog = map->init_slots[i];
5029                fd = bpf_program__fd(targ_prog);
5030
5031                err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5032                if (err) {
5033                        err = -errno;
5034                        pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n",
5035                                map->name, i, targ_prog->name, fd, err);
5036                        return err;
5037                }
5038                pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5039                         map->name, i, targ_prog->name, fd);
5040        }
5041
5042        zfree(&map->init_slots);
5043        map->init_slots_sz = 0;
5044
5045        return 0;
5046}
5047
5048static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5049{
5050        struct bpf_map *map;
5051        int i, err;
5052
5053        for (i = 0; i < obj->nr_maps; i++) {
5054                map = &obj->maps[i];
5055
5056                if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5057                        continue;
5058
5059                err = init_prog_array_slots(obj, map);
5060                if (err < 0) {
5061                        zclose(map->fd);
5062                        return err;
5063                }
5064        }
5065        return 0;
5066}
5067
5068static int map_set_def_max_entries(struct bpf_map *map)
5069{
5070        if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5071                int nr_cpus;
5072
5073                nr_cpus = libbpf_num_possible_cpus();
5074                if (nr_cpus < 0) {
5075                        pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5076                                map->name, nr_cpus);
5077                        return nr_cpus;
5078                }
5079                pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5080                map->def.max_entries = nr_cpus;
5081        }
5082
5083        return 0;
5084}
5085
5086static int
5087bpf_object__create_maps(struct bpf_object *obj)
5088{
5089        struct bpf_map *map;
5090        char *cp, errmsg[STRERR_BUFSIZE];
5091        unsigned int i, j;
5092        int err;
5093        bool retried;
5094
5095        for (i = 0; i < obj->nr_maps; i++) {
5096                map = &obj->maps[i];
5097
5098                /* To support old kernels, we skip creating global data maps
5099                 * (.rodata, .data, .kconfig, etc); later on, during program
5100                 * loading, if we detect that at least one of the to-be-loaded
5101                 * programs is referencing any global data map, we'll error
5102                 * out with program name and relocation index logged.
5103                 * This approach allows to accommodate Clang emitting
5104                 * unnecessary .rodata.str1.1 sections for string literals,
5105                 * but also it allows to have CO-RE applications that use
5106                 * global variables in some of BPF programs, but not others.
5107                 * If those global variable-using programs are not loaded at
5108                 * runtime due to bpf_program__set_autoload(prog, false),
5109                 * bpf_object loading will succeed just fine even on old
5110                 * kernels.
5111                 */
5112                if (bpf_map__is_internal(map) &&
5113                    !kernel_supports(obj, FEAT_GLOBAL_DATA)) {
5114                        map->skipped = true;
5115                        continue;
5116                }
5117
5118                err = map_set_def_max_entries(map);
5119                if (err)
5120                        goto err_out;
5121
5122                retried = false;
5123retry:
5124                if (map->pin_path) {
5125                        err = bpf_object__reuse_map(map);
5126                        if (err) {
5127                                pr_warn("map '%s': error reusing pinned map\n",
5128                                        map->name);
5129                                goto err_out;
5130                        }
5131                        if (retried && map->fd < 0) {
5132                                pr_warn("map '%s': cannot find pinned map\n",
5133                                        map->name);
5134                                err = -ENOENT;
5135                                goto err_out;
5136                        }
5137                }
5138
5139                if (map->fd >= 0) {
5140                        pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5141                                 map->name, map->fd);
5142                } else {
5143                        err = bpf_object__create_map(obj, map, false);
5144                        if (err)
5145                                goto err_out;
5146
5147                        pr_debug("map '%s': created successfully, fd=%d\n",
5148                                 map->name, map->fd);
5149
5150                        if (bpf_map__is_internal(map)) {
5151                                err = bpf_object__populate_internal_map(obj, map);
5152                                if (err < 0) {
5153                                        zclose(map->fd);
5154                                        goto err_out;
5155                                }
5156                        }
5157
5158                        if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5159                                err = init_map_in_map_slots(obj, map);
5160                                if (err < 0) {
5161                                        zclose(map->fd);
5162                                        goto err_out;
5163                                }
5164                        }
5165                }
5166
5167                if (map->pin_path && !map->pinned) {
5168                        err = bpf_map__pin(map, NULL);
5169                        if (err) {
5170                                zclose(map->fd);
5171                                if (!retried && err == -EEXIST) {
5172                                        retried = true;
5173                                        goto retry;
5174                                }
5175                                pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
5176                                        map->name, map->pin_path, err);
5177                                goto err_out;
5178                        }
5179                }
5180        }
5181
5182        return 0;
5183
5184err_out:
5185        cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5186        pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
5187        pr_perm_msg(err);
5188        for (j = 0; j < i; j++)
5189                zclose(obj->maps[j].fd);
5190        return err;
5191}
5192
5193static bool bpf_core_is_flavor_sep(const char *s)
5194{
5195        /* check X___Y name pattern, where X and Y are not underscores */
5196        return s[0] != '_' &&                                 /* X */
5197               s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
5198               s[4] != '_';                                   /* Y */
5199}
5200
5201/* Given 'some_struct_name___with_flavor' return the length of a name prefix
5202 * before last triple underscore. Struct name part after last triple
5203 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5204 */
5205size_t bpf_core_essential_name_len(const char *name)
5206{
5207        size_t n = strlen(name);
5208        int i;
5209
5210        for (i = n - 5; i >= 0; i--) {
5211                if (bpf_core_is_flavor_sep(name + i))
5212                        return i + 1;
5213        }
5214        return n;
5215}
5216
5217void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5218{
5219        if (!cands)
5220                return;
5221
5222        free(cands->cands);
5223        free(cands);
5224}
5225
5226int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5227                       size_t local_essent_len,
5228                       const struct btf *targ_btf,
5229                       const char *targ_btf_name,
5230                       int targ_start_id,
5231                       struct bpf_core_cand_list *cands)
5232{
5233        struct bpf_core_cand *new_cands, *cand;
5234        const struct btf_type *t, *local_t;
5235        const char *targ_name, *local_name;
5236        size_t targ_essent_len;
5237        int n, i;
5238
5239        local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5240        local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5241
5242        n = btf__type_cnt(targ_btf);
5243        for (i = targ_start_id; i < n; i++) {
5244                t = btf__type_by_id(targ_btf, i);
5245                if (btf_kind(t) != btf_kind(local_t))
5246                        continue;
5247
5248                targ_name = btf__name_by_offset(targ_btf, t->name_off);
5249                if (str_is_empty(targ_name))
5250                        continue;
5251
5252                targ_essent_len = bpf_core_essential_name_len(targ_name);
5253                if (targ_essent_len != local_essent_len)
5254                        continue;
5255
5256                if (strncmp(local_name, targ_name, local_essent_len) != 0)
5257                        continue;
5258
5259                pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5260                         local_cand->id, btf_kind_str(local_t),
5261                         local_name, i, btf_kind_str(t), targ_name,
5262                         targ_btf_name);
5263                new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5264                                              sizeof(*cands->cands));
5265                if (!new_cands)
5266                        return -ENOMEM;
5267
5268                cand = &new_cands[cands->len];
5269                cand->btf = targ_btf;
5270                cand->id = i;
5271
5272                cands->cands = new_cands;
5273                cands->len++;
5274        }
5275        return 0;
5276}
5277
5278static int load_module_btfs(struct bpf_object *obj)
5279{
5280        struct bpf_btf_info info;
5281        struct module_btf *mod_btf;
5282        struct btf *btf;
5283        char name[64];
5284        __u32 id = 0, len;
5285        int err, fd;
5286
5287        if (obj->btf_modules_loaded)
5288                return 0;
5289
5290        if (obj->gen_loader)
5291                return 0;
5292
5293        /* don't do this again, even if we find no module BTFs */
5294        obj->btf_modules_loaded = true;
5295
5296        /* kernel too old to support module BTFs */
5297        if (!kernel_supports(obj, FEAT_MODULE_BTF))
5298                return 0;
5299
5300        while (true) {
5301                err = bpf_btf_get_next_id(id, &id);
5302                if (err && errno == ENOENT)
5303                        return 0;
5304                if (err) {
5305                        err = -errno;
5306                        pr_warn("failed to iterate BTF objects: %d\n", err);
5307                        return err;
5308                }
5309
5310                fd = bpf_btf_get_fd_by_id(id);
5311                if (fd < 0) {
5312                        if (errno == ENOENT)
5313                                continue; /* expected race: BTF was unloaded */
5314                        err = -errno;
5315                        pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
5316                        return err;
5317                }
5318
5319                len = sizeof(info);
5320                memset(&info, 0, sizeof(info));
5321                info.name = ptr_to_u64(name);
5322                info.name_len = sizeof(name);
5323
5324                err = bpf_obj_get_info_by_fd(fd, &info, &len);
5325                if (err) {
5326                        err = -errno;
5327                        pr_warn("failed to get BTF object #%d info: %d\n", id, err);
5328                        goto err_out;
5329                }
5330
5331                /* ignore non-module BTFs */
5332                if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5333                        close(fd);
5334                        continue;
5335                }
5336
5337                btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5338                err = libbpf_get_error(btf);
5339                if (err) {
5340                        pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
5341                                name, id, err);
5342                        goto err_out;
5343                }
5344
5345                err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5346                                        sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5347                if (err)
5348                        goto err_out;
5349
5350                mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5351
5352                mod_btf->btf = btf;
5353                mod_btf->id = id;
5354                mod_btf->fd = fd;
5355                mod_btf->name = strdup(name);
5356                if (!mod_btf->name) {
5357                        err = -ENOMEM;
5358                        goto err_out;
5359                }
5360                continue;
5361
5362err_out:
5363                close(fd);
5364                return err;
5365        }
5366
5367        return 0;
5368}
5369
5370static struct bpf_core_cand_list *
5371bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5372{
5373        struct bpf_core_cand local_cand = {};
5374        struct bpf_core_cand_list *cands;
5375        const struct btf *main_btf;
5376        const struct btf_type *local_t;
5377        const char *local_name;
5378        size_t local_essent_len;
5379        int err, i;
5380
5381        local_cand.btf = local_btf;
5382        local_cand.id = local_type_id;
5383        local_t = btf__type_by_id(local_btf, local_type_id);
5384        if (!local_t)
5385                return ERR_PTR(-EINVAL);
5386
5387        local_name = btf__name_by_offset(local_btf, local_t->name_off);
5388        if (str_is_empty(local_name))
5389                return ERR_PTR(-EINVAL);
5390        local_essent_len = bpf_core_essential_name_len(local_name);
5391
5392        cands = calloc(1, sizeof(*cands));
5393        if (!cands)
5394                return ERR_PTR(-ENOMEM);
5395
5396        /* Attempt to find target candidates in vmlinux BTF first */
5397        main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5398        err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5399        if (err)
5400                goto err_out;
5401
5402        /* if vmlinux BTF has any candidate, don't got for module BTFs */
5403        if (cands->len)
5404                return cands;
5405
5406        /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5407        if (obj->btf_vmlinux_override)
5408                return cands;
5409
5410        /* now look through module BTFs, trying to still find candidates */
5411        err = load_module_btfs(obj);
5412        if (err)
5413                goto err_out;
5414
5415        for (i = 0; i < obj->btf_module_cnt; i++) {
5416                err = bpf_core_add_cands(&local_cand, local_essent_len,
5417                                         obj->btf_modules[i].btf,
5418                                         obj->btf_modules[i].name,
5419                                         btf__type_cnt(obj->btf_vmlinux),
5420                                         cands);
5421                if (err)
5422                        goto err_out;
5423        }
5424
5425        return cands;
5426err_out:
5427        bpf_core_free_cands(cands);
5428        return ERR_PTR(err);
5429}
5430
5431/* Check local and target types for compatibility. This check is used for
5432 * type-based CO-RE relocations and follow slightly different rules than
5433 * field-based relocations. This function assumes that root types were already
5434 * checked for name match. Beyond that initial root-level name check, names
5435 * are completely ignored. Compatibility rules are as follows:
5436 *   - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5437 *     kind should match for local and target types (i.e., STRUCT is not
5438 *     compatible with UNION);
5439 *   - for ENUMs, the size is ignored;
5440 *   - for INT, size and signedness are ignored;
5441 *   - for ARRAY, dimensionality is ignored, element types are checked for
5442 *     compatibility recursively;
5443 *   - CONST/VOLATILE/RESTRICT modifiers are ignored;
5444 *   - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5445 *   - FUNC_PROTOs are compatible if they have compatible signature: same
5446 *     number of input args and compatible return and argument types.
5447 * These rules are not set in stone and probably will be adjusted as we get
5448 * more experience with using BPF CO-RE relocations.
5449 */
5450int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5451                              const struct btf *targ_btf, __u32 targ_id)
5452{
5453        const struct btf_type *local_type, *targ_type;
5454        int depth = 32; /* max recursion depth */
5455
5456        /* caller made sure that names match (ignoring flavor suffix) */
5457        local_type = btf__type_by_id(local_btf, local_id);
5458        targ_type = btf__type_by_id(targ_btf, targ_id);
5459        if (btf_kind(local_type) != btf_kind(targ_type))
5460                return 0;
5461
5462recur:
5463        depth--;
5464        if (depth < 0)
5465                return -EINVAL;
5466
5467        local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
5468        targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
5469        if (!local_type || !targ_type)
5470                return -EINVAL;
5471
5472        if (btf_kind(local_type) != btf_kind(targ_type))
5473                return 0;
5474
5475        switch (btf_kind(local_type)) {
5476        case BTF_KIND_UNKN:
5477        case BTF_KIND_STRUCT:
5478        case BTF_KIND_UNION:
5479        case BTF_KIND_ENUM:
5480        case BTF_KIND_FWD:
5481                return 1;
5482        case BTF_KIND_INT:
5483                /* just reject deprecated bitfield-like integers; all other
5484                 * integers are by default compatible between each other
5485                 */
5486                return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0;
5487        case BTF_KIND_PTR:
5488                local_id = local_type->type;
5489                targ_id = targ_type->type;
5490                goto recur;
5491        case BTF_KIND_ARRAY:
5492                local_id = btf_array(local_type)->type;
5493                targ_id = btf_array(targ_type)->type;
5494                goto recur;
5495        case BTF_KIND_FUNC_PROTO: {
5496                struct btf_param *local_p = btf_params(local_type);
5497                struct btf_param *targ_p = btf_params(targ_type);
5498                __u16 local_vlen = btf_vlen(local_type);
5499                __u16 targ_vlen = btf_vlen(targ_type);
5500                int i, err;
5501
5502                if (local_vlen != targ_vlen)
5503                        return 0;
5504
5505                for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
5506                        skip_mods_and_typedefs(local_btf, local_p->type, &local_id);
5507                        skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id);
5508                        err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id);
5509                        if (err <= 0)
5510                                return err;
5511                }
5512
5513                /* tail recurse for return type check */
5514                skip_mods_and_typedefs(local_btf, local_type->type, &local_id);
5515                skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id);
5516                goto recur;
5517        }
5518        default:
5519                pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n",
5520                        btf_kind_str(local_type), local_id, targ_id);
5521                return 0;
5522        }
5523}
5524
5525static size_t bpf_core_hash_fn(const void *key, void *ctx)
5526{
5527        return (size_t)key;
5528}
5529
5530static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
5531{
5532        return k1 == k2;
5533}
5534
5535static void *u32_as_hash_key(__u32 x)
5536{
5537        return (void *)(uintptr_t)x;
5538}
5539
5540static int record_relo_core(struct bpf_program *prog,
5541                            const struct bpf_core_relo *core_relo, int insn_idx)
5542{
5543        struct reloc_desc *relos, *relo;
5544
5545        relos = libbpf_reallocarray(prog->reloc_desc,
5546                                    prog->nr_reloc + 1, sizeof(*relos));
5547        if (!relos)
5548                return -ENOMEM;
5549        relo = &relos[prog->nr_reloc];
5550        relo->type = RELO_CORE;
5551        relo->insn_idx = insn_idx;
5552        relo->core_relo = core_relo;
5553        prog->reloc_desc = relos;
5554        prog->nr_reloc++;
5555        return 0;
5556}
5557
5558static int bpf_core_resolve_relo(struct bpf_program *prog,
5559                                 const struct bpf_core_relo *relo,
5560                                 int relo_idx,
5561                                 const struct btf *local_btf,
5562                                 struct hashmap *cand_cache,
5563                                 struct bpf_core_relo_res *targ_res)
5564{
5565        struct bpf_core_spec specs_scratch[3] = {};
5566        const void *type_key = u32_as_hash_key(relo->type_id);
5567        struct bpf_core_cand_list *cands = NULL;
5568        const char *prog_name = prog->name;
5569        const struct btf_type *local_type;
5570        const char *local_name;
5571        __u32 local_id = relo->type_id;
5572        int err;
5573
5574        local_type = btf__type_by_id(local_btf, local_id);
5575        if (!local_type)
5576                return -EINVAL;
5577
5578        local_name = btf__name_by_offset(local_btf, local_type->name_off);
5579        if (!local_name)
5580                return -EINVAL;
5581
5582        if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
5583            !hashmap__find(cand_cache, type_key, (void **)&cands)) {
5584                cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5585                if (IS_ERR(cands)) {
5586                        pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5587                                prog_name, relo_idx, local_id, btf_kind_str(local_type),
5588                                local_name, PTR_ERR(cands));
5589                        return PTR_ERR(cands);
5590                }
5591                err = hashmap__set(cand_cache, type_key, cands, NULL, NULL);
5592                if (err) {
5593                        bpf_core_free_cands(cands);
5594                        return err;
5595                }
5596        }
5597
5598        return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
5599                                       targ_res);
5600}
5601
5602static int
5603bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5604{
5605        const struct btf_ext_info_sec *sec;
5606        struct bpf_core_relo_res targ_res;
5607        const struct bpf_core_relo *rec;
5608        const struct btf_ext_info *seg;
5609        struct hashmap_entry *entry;
5610        struct hashmap *cand_cache = NULL;
5611        struct bpf_program *prog;
5612        struct bpf_insn *insn;
5613        const char *sec_name;
5614        int i, err = 0, insn_idx, sec_idx;
5615
5616        if (obj->btf_ext->core_relo_info.len == 0)
5617                return 0;
5618
5619        if (targ_btf_path) {
5620                obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5621                err = libbpf_get_error(obj->btf_vmlinux_override);
5622                if (err) {
5623                        pr_warn("failed to parse target BTF: %d\n", err);
5624                        return err;
5625                }
5626        }
5627
5628        cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5629        if (IS_ERR(cand_cache)) {
5630                err = PTR_ERR(cand_cache);
5631                goto out;
5632        }
5633
5634        seg = &obj->btf_ext->core_relo_info;
5635        for_each_btf_ext_sec(seg, sec) {
5636                sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5637                if (str_is_empty(sec_name)) {
5638                        err = -EINVAL;
5639                        goto out;
5640                }
5641                /* bpf_object's ELF is gone by now so it's not easy to find
5642                 * section index by section name, but we can find *any*
5643                 * bpf_program within desired section name and use it's
5644                 * prog->sec_idx to do a proper search by section index and
5645                 * instruction offset
5646                 */
5647                prog = NULL;
5648                for (i = 0; i < obj->nr_programs; i++) {
5649                        prog = &obj->programs[i];
5650                        if (strcmp(prog->sec_name, sec_name) == 0)
5651                                break;
5652                }
5653                if (!prog) {
5654                        pr_warn("sec '%s': failed to find a BPF program\n", sec_name);
5655                        return -ENOENT;
5656                }
5657                sec_idx = prog->sec_idx;
5658
5659                pr_debug("sec '%s': found %d CO-RE relocations\n",
5660                         sec_name, sec->num_info);
5661
5662                for_each_btf_ext_rec(seg, sec, i, rec) {
5663                        if (rec->insn_off % BPF_INSN_SZ)
5664                                return -EINVAL;
5665                        insn_idx = rec->insn_off / BPF_INSN_SZ;
5666                        prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5667                        if (!prog) {
5668                                pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n",
5669                                        sec_name, insn_idx, i);
5670                                err = -EINVAL;
5671                                goto out;
5672                        }
5673                        /* no need to apply CO-RE relocation if the program is
5674                         * not going to be loaded
5675                         */
5676                        if (!prog->load)
5677                                continue;
5678
5679                        /* adjust insn_idx from section frame of reference to the local
5680                         * program's frame of reference; (sub-)program code is not yet
5681                         * relocated, so it's enough to just subtract in-section offset
5682                         */
5683                        insn_idx = insn_idx - prog->sec_insn_off;
5684                        if (insn_idx >= prog->insns_cnt)
5685                                return -EINVAL;
5686                        insn = &prog->insns[insn_idx];
5687
5688                        if (prog->obj->gen_loader) {
5689                                err = record_relo_core(prog, rec, insn_idx);
5690                                if (err) {
5691                                        pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n",
5692                                                prog->name, i, err);
5693                                        goto out;
5694                                }
5695                                continue;
5696                        }
5697
5698                        err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
5699                        if (err) {
5700                                pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5701                                        prog->name, i, err);
5702                                goto out;
5703                        }
5704
5705                        err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
5706                        if (err) {
5707                                pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n",
5708                                        prog->name, i, insn_idx, err);
5709                                goto out;
5710                        }
5711                }
5712        }
5713
5714out:
5715        /* obj->btf_vmlinux and module BTFs are freed after object load */
5716        btf__free(obj->btf_vmlinux_override);
5717        obj->btf_vmlinux_override = NULL;
5718
5719        if (!IS_ERR_OR_NULL(cand_cache)) {
5720                hashmap__for_each_entry(cand_cache, entry, i) {
5721                        bpf_core_free_cands(entry->value);
5722                }
5723                hashmap__free(cand_cache);
5724        }
5725        return err;
5726}
5727
5728/* Relocate data references within program code:
5729 *  - map references;
5730 *  - global variable references;
5731 *  - extern references.
5732 */
5733static int
5734bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
5735{
5736        int i;
5737
5738        for (i = 0; i < prog->nr_reloc; i++) {
5739                struct reloc_desc *relo = &prog->reloc_desc[i];
5740                struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5741                struct extern_desc *ext;
5742
5743                switch (relo->type) {
5744                case RELO_LD64:
5745                        if (obj->gen_loader) {
5746                                insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
5747                                insn[0].imm = relo->map_idx;
5748                        } else {
5749                                insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5750                                insn[0].imm = obj->maps[relo->map_idx].fd;
5751                        }
5752                        break;
5753                case RELO_DATA:
5754                        insn[1].imm = insn[0].imm + relo->sym_off;
5755                        if (obj->gen_loader) {
5756                                insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5757                                insn[0].imm = relo->map_idx;
5758                        } else {
5759                                const struct bpf_map *map = &obj->maps[relo->map_idx];
5760
5761                                if (map->skipped) {
5762                                        pr_warn("prog '%s': relo #%d: kernel doesn't support global data\n",
5763                                                prog->name, i);
5764                                        return -ENOTSUP;
5765                                }
5766                                insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5767                                insn[0].imm = obj->maps[relo->map_idx].fd;
5768                        }
5769                        break;
5770                case RELO_EXTERN_VAR:
5771                        ext = &obj->externs[relo->sym_off];
5772                        if (ext->type == EXT_KCFG) {
5773                                if (obj->gen_loader) {
5774                                        insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5775                                        insn[0].imm = obj->kconfig_map_idx;
5776                                } else {
5777                                        insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5778                                        insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5779                                }
5780                                insn[1].imm = ext->kcfg.data_off;
5781                        } else /* EXT_KSYM */ {
5782                                if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
5783                                        insn[0].src_reg = BPF_PSEUDO_BTF_ID;
5784                                        insn[0].imm = ext->ksym.kernel_btf_id;
5785                                        insn[1].imm = ext->ksym.kernel_btf_obj_fd;
5786                                } else { /* typeless ksyms or unresolved typed ksyms */
5787                                        insn[0].imm = (__u32)ext->ksym.addr;
5788                                        insn[1].imm = ext->ksym.addr >> 32;
5789                                }
5790                        }
5791                        break;
5792                case RELO_EXTERN_FUNC:
5793                        ext = &obj->externs[relo->sym_off];
5794                        insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
5795                        if (ext->is_set) {
5796                                insn[0].imm = ext->ksym.kernel_btf_id;
5797                                insn[0].off = ext->ksym.btf_fd_idx;
5798                        } else { /* unresolved weak kfunc */
5799                                insn[0].imm = 0;
5800                                insn[0].off = 0;
5801                        }
5802                        break;
5803                case RELO_SUBPROG_ADDR:
5804                        if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
5805                                pr_warn("prog '%s': relo #%d: bad insn\n",
5806                                        prog->name, i);
5807                                return -EINVAL;
5808                        }
5809                        /* handled already */
5810                        break;
5811                case RELO_CALL:
5812                        /* handled already */
5813                        break;
5814                case RELO_CORE:
5815                        /* will be handled by bpf_program_record_relos() */
5816                        break;
5817                default:
5818                        pr_warn("prog '%s': relo #%d: bad relo type %d\n",
5819                                prog->name, i, relo->type);
5820                        return -EINVAL;
5821                }
5822        }
5823
5824        return 0;
5825}
5826
5827static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
5828                                    const struct bpf_program *prog,
5829                                    const struct btf_ext_info *ext_info,
5830                                    void **prog_info, __u32 *prog_rec_cnt,
5831                                    __u32 *prog_rec_sz)
5832{
5833        void *copy_start = NULL, *copy_end = NULL;
5834        void *rec, *rec_end, *new_prog_info;
5835        const struct btf_ext_info_sec *sec;
5836        size_t old_sz, new_sz;
5837        const char *sec_name;
5838        int i, off_adj;
5839
5840        for_each_btf_ext_sec(ext_info, sec) {
5841                sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5842                if (!sec_name)
5843                        return -EINVAL;
5844                if (strcmp(sec_name, prog->sec_name) != 0)
5845                        continue;
5846
5847                for_each_btf_ext_rec(ext_info, sec, i, rec) {
5848                        __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
5849
5850                        if (insn_off < prog->sec_insn_off)
5851                                continue;
5852                        if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
5853                                break;
5854
5855                        if (!copy_start)
5856                                copy_start = rec;
5857                        copy_end = rec + ext_info->rec_size;
5858                }
5859
5860                if (!copy_start)
5861                        return -ENOENT;
5862
5863                /* append func/line info of a given (sub-)program to the main
5864                 * program func/line info
5865                 */
5866                old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
5867                new_sz = old_sz + (copy_end - copy_start);
5868                new_prog_info = realloc(*prog_info, new_sz);
5869                if (!new_prog_info)
5870                        return -ENOMEM;
5871                *prog_info = new_prog_info;
5872                *prog_rec_cnt = new_sz / ext_info->rec_size;
5873                memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
5874
5875                /* Kernel instruction offsets are in units of 8-byte
5876                 * instructions, while .BTF.ext instruction offsets generated
5877                 * by Clang are in units of bytes. So convert Clang offsets
5878                 * into kernel offsets and adjust offset according to program
5879                 * relocated position.
5880                 */
5881                off_adj = prog->sub_insn_off - prog->sec_insn_off;
5882                rec = new_prog_info + old_sz;
5883                rec_end = new_prog_info + new_sz;
5884                for (; rec < rec_end; rec += ext_info->rec_size) {
5885                        __u32 *insn_off = rec;
5886
5887                        *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
5888                }
5889                *prog_rec_sz = ext_info->rec_size;
5890                return 0;
5891        }
5892
5893        return -ENOENT;
5894}
5895
5896static int
5897reloc_prog_func_and_line_info(const struct bpf_object *obj,
5898                              struct bpf_program *main_prog,
5899                              const struct bpf_program *prog)
5900{
5901        int err;
5902
5903        /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
5904         * supprot func/line info
5905         */
5906        if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
5907                return 0;
5908
5909        /* only attempt func info relocation if main program's func_info
5910         * relocation was successful
5911         */
5912        if (main_prog != prog && !main_prog->func_info)
5913                goto line_info;
5914
5915        err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
5916                                       &main_prog->func_info,
5917                                       &main_prog->func_info_cnt,
5918                                       &main_prog->func_info_rec_size);
5919        if (err) {
5920                if (err != -ENOENT) {
5921                        pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
5922                                prog->name, err);
5923                        return err;
5924                }
5925                if (main_prog->func_info) {
5926                        /*
5927                         * Some info has already been found but has problem
5928                         * in the last btf_ext reloc. Must have to error out.
5929                         */
5930                        pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
5931                        return err;
5932                }
5933                /* Have problem loading the very first info. Ignore the rest. */
5934                pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
5935                        prog->name);
5936        }
5937
5938line_info:
5939        /* don't relocate line info if main program's relocation failed */
5940        if (main_prog != prog && !main_prog->line_info)
5941                return 0;
5942
5943        err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
5944                                       &main_prog->line_info,
5945                                       &main_prog->line_info_cnt,
5946                                       &main_prog->line_info_rec_size);
5947        if (err) {
5948                if (err != -ENOENT) {
5949                        pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
5950                                prog->name, err);
5951                        return err;
5952                }
5953                if (main_prog->line_info) {
5954                        /*
5955                         * Some info has already been found but has problem
5956                         * in the last btf_ext reloc. Must have to error out.
5957                         */
5958                        pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
5959                        return err;
5960                }
5961                /* Have problem loading the very first info. Ignore the rest. */
5962                pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
5963                        prog->name);
5964        }
5965        return 0;
5966}
5967
5968static int cmp_relo_by_insn_idx(const void *key, const void *elem)
5969{
5970        size_t insn_idx = *(const size_t *)key;
5971        const struct reloc_desc *relo = elem;
5972
5973        if (insn_idx == relo->insn_idx)
5974                return 0;
5975        return insn_idx < relo->insn_idx ? -1 : 1;
5976}
5977
5978static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
5979{
5980        if (!prog->nr_reloc)
5981                return NULL;
5982        return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
5983                       sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
5984}
5985
5986static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
5987{
5988        int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
5989        struct reloc_desc *relos;
5990        int i;
5991
5992        if (main_prog == subprog)
5993                return 0;
5994        relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
5995        if (!relos)
5996                return -ENOMEM;
5997        if (subprog->nr_reloc)
5998                memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
5999                       sizeof(*relos) * subprog->nr_reloc);
6000
6001        for (i = main_prog->nr_reloc; i < new_cnt; i++)
6002                relos[i].insn_idx += subprog->sub_insn_off;
6003        /* After insn_idx adjustment the 'relos' array is still sorted
6004         * by insn_idx and doesn't break bsearch.
6005         */
6006        main_prog->reloc_desc = relos;
6007        main_prog->nr_reloc = new_cnt;
6008        return 0;
6009}
6010
6011static int
6012bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6013                       struct bpf_program *prog)
6014{
6015        size_t sub_insn_idx, insn_idx, new_cnt;
6016        struct bpf_program *subprog;
6017        struct bpf_insn *insns, *insn;
6018        struct reloc_desc *relo;
6019        int err;
6020
6021        err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6022        if (err)
6023                return err;
6024
6025        for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6026                insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6027                if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6028                        continue;
6029
6030                relo = find_prog_insn_relo(prog, insn_idx);
6031                if (relo && relo->type == RELO_EXTERN_FUNC)
6032                        /* kfunc relocations will be handled later
6033                         * in bpf_object__relocate_data()
6034                         */
6035                        continue;
6036                if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6037                        pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6038                                prog->name, insn_idx, relo->type);
6039                        return -LIBBPF_ERRNO__RELOC;
6040                }
6041                if (relo) {
6042                        /* sub-program instruction index is a combination of
6043                         * an offset of a symbol pointed to by relocation and
6044                         * call instruction's imm field; for global functions,
6045                         * call always has imm = -1, but for static functions
6046                         * relocation is against STT_SECTION and insn->imm
6047                         * points to a start of a static function
6048                         *
6049                         * for subprog addr relocation, the relo->sym_off + insn->imm is
6050                         * the byte offset in the corresponding section.
6051                         */
6052                        if (relo->type == RELO_CALL)
6053                                sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6054                        else
6055                                sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6056                } else if (insn_is_pseudo_func(insn)) {
6057                        /*
6058                         * RELO_SUBPROG_ADDR relo is always emitted even if both
6059                         * functions are in the same section, so it shouldn't reach here.
6060                         */
6061                        pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6062                                prog->name, insn_idx);
6063                        return -LIBBPF_ERRNO__RELOC;
6064                } else {
6065                        /* if subprogram call is to a static function within
6066                         * the same ELF section, there won't be any relocation
6067                         * emitted, but it also means there is no additional
6068                         * offset necessary, insns->imm is relative to
6069                         * instruction's original position within the section
6070                         */
6071                        sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6072                }
6073
6074                /* we enforce that sub-programs should be in .text section */
6075                subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6076                if (!subprog) {
6077                        pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6078                                prog->name);
6079                        return -LIBBPF_ERRNO__RELOC;
6080                }
6081
6082                /* if it's the first call instruction calling into this
6083                 * subprogram (meaning this subprog hasn't been processed
6084                 * yet) within the context of current main program:
6085                 *   - append it at the end of main program's instructions blog;
6086                 *   - process is recursively, while current program is put on hold;
6087                 *   - if that subprogram calls some other not yet processes
6088                 *   subprogram, same thing will happen recursively until
6089                 *   there are no more unprocesses subprograms left to append
6090                 *   and relocate.
6091                 */
6092                if (subprog->sub_insn_off == 0) {
6093                        subprog->sub_insn_off = main_prog->insns_cnt;
6094
6095                        new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6096                        insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6097                        if (!insns) {
6098                                pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6099                                return -ENOMEM;
6100                        }
6101                        main_prog->insns = insns;
6102                        main_prog->insns_cnt = new_cnt;
6103
6104                        memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6105                               subprog->insns_cnt * sizeof(*insns));
6106
6107                        pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6108                                 main_prog->name, subprog->insns_cnt, subprog->name);
6109
6110                        /* The subprog insns are now appended. Append its relos too. */
6111                        err = append_subprog_relos(main_prog, subprog);
6112                        if (err)
6113                                return err;
6114                        err = bpf_object__reloc_code(obj, main_prog, subprog);
6115                        if (err)
6116                                return err;
6117                }
6118
6119                /* main_prog->insns memory could have been re-allocated, so
6120                 * calculate pointer again
6121                 */
6122                insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6123                /* calculate correct instruction position within current main
6124                 * prog; each main prog can have a different set of
6125                 * subprograms appended (potentially in different order as
6126                 * well), so position of any subprog can be different for
6127                 * different main programs */
6128                insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6129
6130                pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6131                         prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6132        }
6133
6134        return 0;
6135}
6136
6137/*
6138 * Relocate sub-program calls.
6139 *
6140 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6141 * main prog) is processed separately. For each subprog (non-entry functions,
6142 * that can be called from either entry progs or other subprogs) gets their
6143 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6144 * hasn't been yet appended and relocated within current main prog. Once its
6145 * relocated, sub_insn_off will point at the position within current main prog
6146 * where given subprog was appended. This will further be used to relocate all
6147 * the call instructions jumping into this subprog.
6148 *
6149 * We start with main program and process all call instructions. If the call
6150 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6151 * is zero), subprog instructions are appended at the end of main program's
6152 * instruction array. Then main program is "put on hold" while we recursively
6153 * process newly appended subprogram. If that subprogram calls into another
6154 * subprogram that hasn't been appended, new subprogram is appended again to
6155 * the *main* prog's instructions (subprog's instructions are always left
6156 * untouched, as they need to be in unmodified state for subsequent main progs
6157 * and subprog instructions are always sent only as part of a main prog) and
6158 * the process continues recursively. Once all the subprogs called from a main
6159 * prog or any of its subprogs are appended (and relocated), all their
6160 * positions within finalized instructions array are known, so it's easy to
6161 * rewrite call instructions with correct relative offsets, corresponding to
6162 * desired target subprog.
6163 *
6164 * Its important to realize that some subprogs might not be called from some
6165 * main prog and any of its called/used subprogs. Those will keep their
6166 * subprog->sub_insn_off as zero at all times and won't be appended to current
6167 * main prog and won't be relocated within the context of current main prog.
6168 * They might still be used from other main progs later.
6169 *
6170 * Visually this process can be shown as below. Suppose we have two main
6171 * programs mainA and mainB and BPF object contains three subprogs: subA,
6172 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6173 * subC both call subB:
6174 *
6175 *        +--------+ +-------+
6176 *        |        v v       |
6177 *     +--+---+ +--+-+-+ +---+--+
6178 *     | subA | | subB | | subC |
6179 *     +--+---+ +------+ +---+--+
6180 *        ^                  ^
6181 *        |                  |
6182 *    +---+-------+   +------+----+
6183 *    |   mainA   |   |   mainB   |
6184 *    +-----------+   +-----------+
6185 *
6186 * We'll start relocating mainA, will find subA, append it and start
6187 * processing sub A recursively:
6188 *
6189 *    +-----------+------+
6190 *    |   mainA   | subA |
6191 *    +-----------+------+
6192 *
6193 * At this point we notice that subB is used from subA, so we append it and
6194 * relocate (there are no further subcalls from subB):
6195 *
6196 *    +-----------+------+------+
6197 *    |   mainA   | subA | subB |
6198 *    +-----------+------+------+
6199 *
6200 * At this point, we relocate subA calls, then go one level up and finish with
6201 * relocatin mainA calls. mainA is done.
6202 *
6203 * For mainB process is similar but results in different order. We start with
6204 * mainB and skip subA and subB, as mainB never calls them (at least
6205 * directly), but we see subC is needed, so we append and start processing it:
6206 *
6207 *    +-----------+------+
6208 *    |   mainB   | subC |
6209 *    +-----------+------+
6210 * Now we see subC needs subB, so we go back to it, append and relocate it:
6211 *
6212 *    +-----------+------+------+
6213 *    |   mainB   | subC | subB |
6214 *    +-----------+------+------+
6215 *
6216 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6217 */
6218static int
6219bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6220{
6221        struct bpf_program *subprog;
6222        int i, err;
6223
6224        /* mark all subprogs as not relocated (yet) within the context of
6225         * current main program
6226         */
6227        for (i = 0; i < obj->nr_programs; i++) {
6228                subprog = &obj->programs[i];
6229                if (!prog_is_subprog(obj, subprog))
6230                        continue;
6231
6232                subprog->sub_insn_off = 0;
6233        }
6234
6235        err = bpf_object__reloc_code(obj, prog, prog);
6236        if (err)
6237                return err;
6238
6239
6240        return 0;
6241}
6242
6243static void
6244bpf_object__free_relocs(struct bpf_object *obj)
6245{
6246        struct bpf_program *prog;
6247        int i;
6248
6249        /* free up relocation descriptors */
6250        for (i = 0; i < obj->nr_programs; i++) {
6251                prog = &obj->programs[i];
6252                zfree(&prog->reloc_desc);
6253                prog->nr_reloc = 0;
6254        }
6255}
6256
6257static int cmp_relocs(const void *_a, const void *_b)
6258{
6259        const struct reloc_desc *a = _a;
6260        const struct reloc_desc *b = _b;
6261
6262        if (a->insn_idx != b->insn_idx)
6263                return a->insn_idx < b->insn_idx ? -1 : 1;
6264
6265        /* no two relocations should have the same insn_idx, but ... */
6266        if (a->type != b->type)
6267                return a->type < b->type ? -1 : 1;
6268
6269        return 0;
6270}
6271
6272static void bpf_object__sort_relos(struct bpf_object *obj)
6273{
6274        int i;
6275
6276        for (i = 0; i < obj->nr_programs; i++) {
6277                struct bpf_program *p = &obj->programs[i];
6278
6279                if (!p->nr_reloc)
6280                        continue;
6281
6282                qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6283        }
6284}
6285
6286static int
6287bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6288{
6289        struct bpf_program *prog;
6290        size_t i, j;
6291        int err;
6292
6293        if (obj->btf_ext) {
6294                err = bpf_object__relocate_core(obj, targ_btf_path);
6295                if (err) {
6296                        pr_warn("failed to perform CO-RE relocations: %d\n",
6297                                err);
6298                        return err;
6299                }
6300                if (obj->gen_loader)
6301                        bpf_object__sort_relos(obj);
6302        }
6303
6304        /* Before relocating calls pre-process relocations and mark
6305         * few ld_imm64 instructions that points to subprogs.
6306         * Otherwise bpf_object__reloc_code() later would have to consider
6307         * all ld_imm64 insns as relocation candidates. That would
6308         * reduce relocation speed, since amount of find_prog_insn_relo()
6309         * would increase and most of them will fail to find a relo.
6310         */
6311        for (i = 0; i < obj->nr_programs; i++) {
6312                prog = &obj->programs[i];
6313                for (j = 0; j < prog->nr_reloc; j++) {
6314                        struct reloc_desc *relo = &prog->reloc_desc[j];
6315                        struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6316
6317                        /* mark the insn, so it's recognized by insn_is_pseudo_func() */
6318                        if (relo->type == RELO_SUBPROG_ADDR)
6319                                insn[0].src_reg = BPF_PSEUDO_FUNC;
6320                }
6321        }
6322
6323        /* relocate subprogram calls and append used subprograms to main
6324         * programs; each copy of subprogram code needs to be relocated
6325         * differently for each main program, because its code location might
6326         * have changed.
6327         * Append subprog relos to main programs to allow data relos to be
6328         * processed after text is completely relocated.
6329         */
6330        for (i = 0; i < obj->nr_programs; i++) {
6331                prog = &obj->programs[i];
6332                /* sub-program's sub-calls are relocated within the context of
6333                 * its main program only
6334                 */
6335                if (prog_is_subprog(obj, prog))
6336                        continue;
6337                if (!prog->load)
6338                        continue;
6339
6340                err = bpf_object__relocate_calls(obj, prog);
6341                if (err) {
6342                        pr_warn("prog '%s': failed to relocate calls: %d\n",
6343                                prog->name, err);
6344                        return err;
6345                }
6346        }
6347        /* Process data relos for main programs */
6348        for (i = 0; i < obj->nr_programs; i++) {
6349                prog = &obj->programs[i];
6350                if (prog_is_subprog(obj, prog))
6351                        continue;
6352                if (!prog->load)
6353                        continue;
6354                err = bpf_object__relocate_data(obj, prog);
6355                if (err) {
6356                        pr_warn("prog '%s': failed to relocate data references: %d\n",
6357                                prog->name, err);
6358                        return err;
6359                }
6360        }
6361        if (!obj->gen_loader)
6362                bpf_object__free_relocs(obj);
6363        return 0;
6364}
6365
6366static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6367                                            Elf64_Shdr *shdr, Elf_Data *data);
6368
6369static int bpf_object__collect_map_relos(struct bpf_object *obj,
6370                                         Elf64_Shdr *shdr, Elf_Data *data)
6371{
6372        const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6373        int i, j, nrels, new_sz;
6374        const struct btf_var_secinfo *vi = NULL;
6375        const struct btf_type *sec, *var, *def;
6376        struct bpf_map *map = NULL, *targ_map = NULL;
6377        struct bpf_program *targ_prog = NULL;
6378        bool is_prog_array, is_map_in_map;
6379        const struct btf_member *member;
6380        const char *name, *mname, *type;
6381        unsigned int moff;
6382        Elf64_Sym *sym;
6383        Elf64_Rel *rel;
6384        void *tmp;
6385
6386        if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6387                return -EINVAL;
6388        sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6389        if (!sec)
6390                return -EINVAL;
6391
6392        nrels = shdr->sh_size / shdr->sh_entsize;
6393        for (i = 0; i < nrels; i++) {
6394                rel = elf_rel_by_idx(data, i);
6395                if (!rel) {
6396                        pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6397                        return -LIBBPF_ERRNO__FORMAT;
6398                }
6399
6400                sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
6401                if (!sym) {
6402                        pr_warn(".maps relo #%d: symbol %zx not found\n",
6403                                i, (size_t)ELF64_R_SYM(rel->r_info));
6404                        return -LIBBPF_ERRNO__FORMAT;
6405                }
6406                name = elf_sym_str(obj, sym->st_name) ?: "<?>";
6407
6408                pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
6409                         i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
6410                         (size_t)rel->r_offset, sym->st_name, name);
6411
6412                for (j = 0; j < obj->nr_maps; j++) {
6413                        map = &obj->maps[j];
6414                        if (map->sec_idx != obj->efile.btf_maps_shndx)
6415                                continue;
6416
6417                        vi = btf_var_secinfos(sec) + map->btf_var_idx;
6418                        if (vi->offset <= rel->r_offset &&
6419                            rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6420                                break;
6421                }
6422                if (j == obj->nr_maps) {
6423                        pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
6424                                i, name, (size_t)rel->r_offset);
6425                        return -EINVAL;
6426                }
6427
6428                is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
6429                is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
6430                type = is_map_in_map ? "map" : "prog";
6431                if (is_map_in_map) {
6432                        if (sym->st_shndx != obj->efile.btf_maps_shndx) {
6433                                pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6434                                        i, name);
6435                                return -LIBBPF_ERRNO__RELOC;
6436                        }
6437                        if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6438                            map->def.key_size != sizeof(int)) {
6439                                pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6440                                        i, map->name, sizeof(int));
6441                                return -EINVAL;
6442                        }
6443                        targ_map = bpf_object__find_map_by_name(obj, name);
6444                        if (!targ_map) {
6445                                pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
6446                                        i, name);
6447                                return -ESRCH;
6448                        }
6449                } else if (is_prog_array) {
6450                        targ_prog = bpf_object__find_program_by_name(obj, name);
6451                        if (!targ_prog) {
6452                                pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
6453                                        i, name);
6454                                return -ESRCH;
6455                        }
6456                        if (targ_prog->sec_idx != sym->st_shndx ||
6457                            targ_prog->sec_insn_off * 8 != sym->st_value ||
6458                            prog_is_subprog(obj, targ_prog)) {
6459                                pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
6460                                        i, name);
6461                                return -LIBBPF_ERRNO__RELOC;
6462                        }
6463                } else {
6464                        return -EINVAL;
6465                }
6466
6467                var = btf__type_by_id(obj->btf, vi->type);
6468                def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6469                if (btf_vlen(def) == 0)
6470                        return -EINVAL;
6471                member = btf_members(def) + btf_vlen(def) - 1;
6472                mname = btf__name_by_offset(obj->btf, member->name_off);
6473                if (strcmp(mname, "values"))
6474                        return -EINVAL;
6475
6476                moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6477                if (rel->r_offset - vi->offset < moff)
6478                        return -EINVAL;
6479
6480                moff = rel->r_offset - vi->offset - moff;
6481                /* here we use BPF pointer size, which is always 64 bit, as we
6482                 * are parsing ELF that was built for BPF target
6483                 */
6484                if (moff % bpf_ptr_sz)
6485                        return -EINVAL;
6486                moff /= bpf_ptr_sz;
6487                if (moff >= map->init_slots_sz) {
6488                        new_sz = moff + 1;
6489                        tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6490                        if (!tmp)
6491                                return -ENOMEM;
6492                        map->init_slots = tmp;
6493                        memset(map->init_slots + map->init_slots_sz, 0,
6494                               (new_sz - map->init_slots_sz) * host_ptr_sz);
6495                        map->init_slots_sz = new_sz;
6496                }
6497                map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
6498
6499                pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
6500                         i, map->name, moff, type, name);
6501        }
6502
6503        return 0;
6504}
6505
6506static int bpf_object__collect_relos(struct bpf_object *obj)
6507{
6508        int i, err;
6509
6510        for (i = 0; i < obj->efile.sec_cnt; i++) {
6511                struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
6512                Elf64_Shdr *shdr;
6513                Elf_Data *data;
6514                int idx;
6515
6516                if (sec_desc->sec_type != SEC_RELO)
6517                        continue;
6518
6519                shdr = sec_desc->shdr;
6520                data = sec_desc->data;
6521                idx = shdr->sh_info;
6522
6523                if (shdr->sh_type != SHT_REL) {
6524                        pr_warn("internal error at %d\n", __LINE__);
6525                        return -LIBBPF_ERRNO__INTERNAL;
6526                }
6527
6528                if (idx == obj->efile.st_ops_shndx)
6529                        err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6530                else if (idx == obj->efile.btf_maps_shndx)
6531                        err = bpf_object__collect_map_relos(obj, shdr, data);
6532                else
6533                        err = bpf_object__collect_prog_relos(obj, shdr, data);
6534                if (err)
6535                        return err;
6536        }
6537
6538        bpf_object__sort_relos(obj);
6539        return 0;
6540}
6541
6542static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6543{
6544        if (BPF_CLASS(insn->code) == BPF_JMP &&
6545            BPF_OP(insn->code) == BPF_CALL &&
6546            BPF_SRC(insn->code) == BPF_K &&
6547            insn->src_reg == 0 &&
6548            insn->dst_reg == 0) {
6549                    *func_id = insn->imm;
6550                    return true;
6551        }
6552        return false;
6553}
6554
6555static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
6556{
6557        struct bpf_insn *insn = prog->insns;
6558        enum bpf_func_id func_id;
6559        int i;
6560
6561        if (obj->gen_loader)
6562                return 0;
6563
6564        for (i = 0; i < prog->insns_cnt; i++, insn++) {
6565                if (!insn_is_helper_call(insn, &func_id))
6566                        continue;
6567
6568                /* on kernels that don't yet support
6569                 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6570                 * to bpf_probe_read() which works well for old kernels
6571                 */
6572                switch (func_id) {
6573                case BPF_FUNC_probe_read_kernel:
6574                case BPF_FUNC_probe_read_user:
6575                        if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6576                                insn->imm = BPF_FUNC_probe_read;
6577                        break;
6578                case BPF_FUNC_probe_read_kernel_str:
6579                case BPF_FUNC_probe_read_user_str:
6580                        if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6581                                insn->imm = BPF_FUNC_probe_read_str;
6582                        break;
6583                default:
6584                        break;
6585                }
6586        }
6587        return 0;
6588}
6589
6590static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
6591                                     int *btf_obj_fd, int *btf_type_id);
6592
6593/* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
6594static int libbpf_prepare_prog_load(struct bpf_program *prog,
6595                                    struct bpf_prog_load_opts *opts, long cookie)
6596{
6597        enum sec_def_flags def = cookie;
6598
6599        /* old kernels might not support specifying expected_attach_type */
6600        if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
6601                opts->expected_attach_type = 0;
6602
6603        if (def & SEC_SLEEPABLE)
6604                opts->prog_flags |= BPF_F_SLEEPABLE;
6605
6606        if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
6607                opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
6608
6609        if (def & SEC_DEPRECATED)
6610                pr_warn("SEC(\"%s\") is deprecated, please see https://github.com/libbpf/libbpf/wiki/Libbpf-1.0-migration-guide#bpf-program-sec-annotation-deprecations for details\n",
6611                        prog->sec_name);
6612
6613        if ((prog->type == BPF_PROG_TYPE_TRACING ||
6614             prog->type == BPF_PROG_TYPE_LSM ||
6615             prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
6616                int btf_obj_fd = 0, btf_type_id = 0, err;
6617                const char *attach_name;
6618
6619                attach_name = strchr(prog->sec_name, '/') + 1;
6620                err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
6621                if (err)
6622                        return err;
6623
6624                /* cache resolved BTF FD and BTF type ID in the prog */
6625                prog->attach_btf_obj_fd = btf_obj_fd;
6626                prog->attach_btf_id = btf_type_id;
6627
6628                /* but by now libbpf common logic is not utilizing
6629                 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
6630                 * this callback is called after opts were populated by
6631                 * libbpf, so this callback has to update opts explicitly here
6632                 */
6633                opts->attach_btf_obj_fd = btf_obj_fd;
6634                opts->attach_btf_id = btf_type_id;
6635        }
6636        return 0;
6637}
6638
6639static int bpf_object_load_prog_instance(struct bpf_object *obj, struct bpf_program *prog,
6640                                         struct bpf_insn *insns, int insns_cnt,
6641                                         const char *license, __u32 kern_version,
6642                                         int *prog_fd)
6643{
6644        LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
6645        const char *prog_name = NULL;
6646        char *cp, errmsg[STRERR_BUFSIZE];
6647        size_t log_buf_size = 0;
6648        char *log_buf = NULL, *tmp;
6649        int btf_fd, ret, err;
6650        bool own_log_buf = true;
6651        __u32 log_level = prog->log_level;
6652
6653        if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6654                /*
6655                 * The program type must be set.  Most likely we couldn't find a proper
6656                 * section definition at load time, and thus we didn't infer the type.
6657                 */
6658                pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
6659                        prog->name, prog->sec_name);
6660                return -EINVAL;
6661        }
6662
6663        if (!insns || !insns_cnt)
6664                return -EINVAL;
6665
6666        load_attr.expected_attach_type = prog->expected_attach_type;
6667        if (kernel_supports(obj, FEAT_PROG_NAME))
6668                prog_name = prog->name;
6669        load_attr.attach_prog_fd = prog->attach_prog_fd;
6670        load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
6671        load_attr.attach_btf_id = prog->attach_btf_id;
6672        load_attr.kern_version = kern_version;
6673        load_attr.prog_ifindex = prog->prog_ifindex;
6674
6675        /* specify func_info/line_info only if kernel supports them */
6676        btf_fd = bpf_object__btf_fd(obj);
6677        if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
6678                load_attr.prog_btf_fd = btf_fd;
6679                load_attr.func_info = prog->func_info;
6680                load_attr.func_info_rec_size = prog->func_info_rec_size;
6681                load_attr.func_info_cnt = prog->func_info_cnt;
6682                load_attr.line_info = prog->line_info;
6683                load_attr.line_info_rec_size = prog->line_info_rec_size;
6684                load_attr.line_info_cnt = prog->line_info_cnt;
6685        }
6686        load_attr.log_level = log_level;
6687        load_attr.prog_flags = prog->prog_flags;
6688        load_attr.fd_array = obj->fd_array;
6689
6690        /* adjust load_attr if sec_def provides custom preload callback */
6691        if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
6692                err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
6693                if (err < 0) {
6694                        pr_warn("prog '%s': failed to prepare load attributes: %d\n",
6695                                prog->name, err);
6696                        return err;
6697                }
6698        }
6699
6700        if (obj->gen_loader) {
6701                bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
6702                                   license, insns, insns_cnt, &load_attr,
6703                                   prog - obj->programs);
6704                *prog_fd = -1;
6705                return 0;
6706        }
6707
6708retry_load:
6709        /* if log_level is zero, we don't request logs initiallly even if
6710         * custom log_buf is specified; if the program load fails, then we'll
6711         * bump log_level to 1 and use either custom log_buf or we'll allocate
6712         * our own and retry the load to get details on what failed
6713         */
6714        if (log_level) {
6715                if (prog->log_buf) {
6716                        log_buf = prog->log_buf;
6717                        log_buf_size = prog->log_size;
6718                        own_log_buf = false;
6719                } else if (obj->log_buf) {
6720                        log_buf = obj->log_buf;
6721                        log_buf_size = obj->log_size;
6722                        own_log_buf = false;
6723                } else {
6724                        log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
6725                        tmp = realloc(log_buf, log_buf_size);
6726                        if (!tmp) {
6727                                ret = -ENOMEM;
6728                                goto out;
6729                        }
6730                        log_buf = tmp;
6731                        log_buf[0] = '\0';
6732                        own_log_buf = true;
6733                }
6734        }
6735
6736        load_attr.log_buf = log_buf;
6737        load_attr.log_size = log_buf_size;
6738        load_attr.log_level = log_level;
6739
6740        ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
6741        if (ret >= 0) {
6742                if (log_level && own_log_buf) {
6743                        pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6744                                 prog->name, log_buf);
6745                }
6746
6747                if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
6748                        struct bpf_map *map;
6749                        int i;
6750
6751                        for (i = 0; i < obj->nr_maps; i++) {
6752                                map = &prog->obj->maps[i];
6753                                if (map->libbpf_type != LIBBPF_MAP_RODATA)
6754                                        continue;
6755
6756                                if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) {
6757                                        cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6758                                        pr_warn("prog '%s': failed to bind map '%s': %s\n",
6759                                                prog->name, map->real_name, cp);
6760                                        /* Don't fail hard if can't bind rodata. */
6761                                }
6762                        }
6763                }
6764
6765                *prog_fd = ret;
6766                ret = 0;
6767                goto out;
6768        }
6769
6770        if (log_level == 0) {
6771                log_level = 1;
6772                goto retry_load;
6773        }
6774        /* On ENOSPC, increase log buffer size and retry, unless custom
6775         * log_buf is specified.
6776         * Be careful to not overflow u32, though. Kernel's log buf size limit
6777         * isn't part of UAPI so it can always be bumped to full 4GB. So don't
6778         * multiply by 2 unless we are sure we'll fit within 32 bits.
6779         * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
6780         */
6781        if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
6782                goto retry_load;
6783
6784        ret = -errno;
6785        cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6786        pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp);
6787        pr_perm_msg(ret);
6788
6789        if (own_log_buf && log_buf && log_buf[0] != '\0') {
6790                pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
6791                        prog->name, log_buf);
6792        }
6793        if (insns_cnt >= BPF_MAXINSNS) {
6794                pr_warn("prog '%s': program too large (%d insns), at most %d insns\n",
6795                        prog->name, insns_cnt, BPF_MAXINSNS);
6796        }
6797
6798out:
6799        if (own_log_buf)
6800                free(log_buf);
6801        return ret;
6802}
6803
6804static int bpf_program_record_relos(struct bpf_program *prog)
6805{
6806        struct bpf_object *obj = prog->obj;
6807        int i;
6808
6809        for (i = 0; i < prog->nr_reloc; i++) {
6810                struct reloc_desc *relo = &prog->reloc_desc[i];
6811                struct extern_desc *ext = &obj->externs[relo->sym_off];
6812
6813                switch (relo->type) {
6814                case RELO_EXTERN_VAR:
6815                        if (ext->type != EXT_KSYM)
6816                                continue;
6817                        bpf_gen__record_extern(obj->gen_loader, ext->name,
6818                                               ext->is_weak, !ext->ksym.type_id,
6819                                               BTF_KIND_VAR, relo->insn_idx);
6820                        break;
6821                case RELO_EXTERN_FUNC:
6822                        bpf_gen__record_extern(obj->gen_loader, ext->name,
6823                                               ext->is_weak, false, BTF_KIND_FUNC,
6824                                               relo->insn_idx);
6825                        break;
6826                case RELO_CORE: {
6827                        struct bpf_core_relo cr = {
6828                                .insn_off = relo->insn_idx * 8,
6829                                .type_id = relo->core_relo->type_id,
6830                                .access_str_off = relo->core_relo->access_str_off,
6831                                .kind = relo->core_relo->kind,
6832                        };
6833
6834                        bpf_gen__record_relo_core(obj->gen_loader, &cr);
6835                        break;
6836                }
6837                default:
6838                        continue;
6839                }
6840        }
6841        return 0;
6842}
6843
6844static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
6845                                const char *license, __u32 kern_ver)
6846{
6847        int err = 0, fd, i;
6848
6849        if (obj->loaded) {
6850                pr_warn("prog '%s': can't load after object was loaded\n", prog->name);
6851                return libbpf_err(-EINVAL);
6852        }
6853
6854        if (prog->instances.nr < 0 || !prog->instances.fds) {
6855                if (prog->preprocessor) {
6856                        pr_warn("Internal error: can't load program '%s'\n",
6857                                prog->name);
6858                        return libbpf_err(-LIBBPF_ERRNO__INTERNAL);
6859                }
6860
6861                prog->instances.fds = malloc(sizeof(int));
6862                if (!prog->instances.fds) {
6863                        pr_warn("Not enough memory for BPF fds\n");
6864                        return libbpf_err(-ENOMEM);
6865                }
6866                prog->instances.nr = 1;
6867                prog->instances.fds[0] = -1;
6868        }
6869
6870        if (!prog->preprocessor) {
6871                if (prog->instances.nr != 1) {
6872                        pr_warn("prog '%s': inconsistent nr(%d) != 1\n",
6873                                prog->name, prog->instances.nr);
6874                }
6875                if (obj->gen_loader)
6876                        bpf_program_record_relos(prog);
6877                err = bpf_object_load_prog_instance(obj, prog,
6878                                                    prog->insns, prog->insns_cnt,
6879                                                    license, kern_ver, &fd);
6880                if (!err)
6881                        prog->instances.fds[0] = fd;
6882                goto out;
6883        }
6884
6885        for (i = 0; i < prog->instances.nr; i++) {
6886                struct bpf_prog_prep_result result;
6887                bpf_program_prep_t preprocessor = prog->preprocessor;
6888
6889                memset(&result, 0, sizeof(result));
6890                err = preprocessor(prog, i, prog->insns,
6891                                   prog->insns_cnt, &result);
6892                if (err) {
6893                        pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
6894                                i, prog->name);
6895                        goto out;
6896                }
6897
6898                if (!result.new_insn_ptr || !result.new_insn_cnt) {
6899                        pr_debug("Skip loading the %dth instance of program '%s'\n",
6900                                 i, prog->name);
6901                        prog->instances.fds[i] = -1;
6902                        if (result.pfd)
6903                                *result.pfd = -1;
6904                        continue;
6905                }
6906
6907                err = bpf_object_load_prog_instance(obj, prog,
6908                                                    result.new_insn_ptr, result.new_insn_cnt,
6909                                                    license, kern_ver, &fd);
6910                if (err) {
6911                        pr_warn("Loading the %dth instance of program '%s' failed\n",
6912                                i, prog->name);
6913                        goto out;
6914                }
6915
6916                if (result.pfd)
6917                        *result.pfd = fd;
6918                prog->instances.fds[i] = fd;
6919        }
6920out:
6921        if (err)
6922                pr_warn("failed to load program '%s'\n", prog->name);
6923        return libbpf_err(err);
6924}
6925
6926int bpf_program__load(struct bpf_program *prog, const char *license, __u32 kern_ver)
6927{
6928        return bpf_object_load_prog(prog->obj, prog, license, kern_ver);
6929}
6930
6931static int
6932bpf_object__load_progs(struct bpf_object *obj, int log_level)
6933{
6934        struct bpf_program *prog;
6935        size_t i;
6936        int err;
6937
6938        for (i = 0; i < obj->nr_programs; i++) {
6939                prog = &obj->programs[i];
6940                err = bpf_object__sanitize_prog(obj, prog);
6941                if (err)
6942                        return err;
6943        }
6944
6945        for (i = 0; i < obj->nr_programs; i++) {
6946                prog = &obj->programs[i];
6947                if (prog_is_subprog(obj, prog))
6948                        continue;
6949                if (!prog->load) {
6950                        pr_debug("prog '%s': skipped loading\n", prog->name);
6951                        continue;
6952                }
6953                prog->log_level |= log_level;
6954                err = bpf_object_load_prog(obj, prog, obj->license, obj->kern_version);
6955                if (err)
6956                        return err;
6957        }
6958        if (obj->gen_loader)
6959                bpf_object__free_relocs(obj);
6960        return 0;
6961}
6962
6963static const struct bpf_sec_def *find_sec_def(const char *sec_name);
6964
6965static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
6966{
6967        struct bpf_program *prog;
6968        int err;
6969
6970        bpf_object__for_each_program(prog, obj) {
6971                prog->sec_def = find_sec_def(prog->sec_name);
6972                if (!prog->sec_def) {
6973                        /* couldn't guess, but user might manually specify */
6974                        pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
6975                                prog->name, prog->sec_name);
6976                        continue;
6977                }
6978
6979                bpf_program__set_type(prog, prog->sec_def->prog_type);
6980                bpf_program__set_expected_attach_type(prog, prog->sec_def->expected_attach_type);
6981
6982#pragma GCC diagnostic push
6983#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
6984                if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
6985                    prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
6986                        prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
6987#pragma GCC diagnostic pop
6988
6989                /* sec_def can have custom callback which should be called
6990                 * after bpf_program is initialized to adjust its properties
6991                 */
6992                if (prog->sec_def->prog_setup_fn) {
6993                        err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
6994                        if (err < 0) {
6995                                pr_warn("prog '%s': failed to initialize: %d\n",
6996                                        prog->name, err);
6997                                return err;
6998                        }
6999                }
7000        }
7001
7002        return 0;
7003}
7004
7005static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7006                                          const struct bpf_object_open_opts *opts)
7007{
7008        const char *obj_name, *kconfig, *btf_tmp_path;
7009        struct bpf_object *obj;
7010        char tmp_name[64];
7011        int err;
7012        char *log_buf;
7013        size_t log_size;
7014        __u32 log_level;
7015
7016        if (elf_version(EV_CURRENT) == EV_NONE) {
7017                pr_warn("failed to init libelf for %s\n",
7018                        path ? : "(mem buf)");
7019                return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7020        }
7021
7022        if (!OPTS_VALID(opts, bpf_object_open_opts))
7023                return ERR_PTR(-EINVAL);
7024
7025        obj_name = OPTS_GET(opts, object_name, NULL);
7026        if (obj_buf) {
7027                if (!obj_name) {
7028                        snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
7029                                 (unsigned long)obj_buf,
7030                                 (unsigned long)obj_buf_sz);
7031                        obj_name = tmp_name;
7032                }
7033                path = obj_name;
7034                pr_debug("loading object '%s' from buffer\n", obj_name);
7035        }
7036
7037        log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7038        log_size = OPTS_GET(opts, kernel_log_size, 0);
7039        log_level = OPTS_GET(opts, kernel_log_level, 0);
7040        if (log_size > UINT_MAX)
7041                return ERR_PTR(-EINVAL);
7042        if (log_size && !log_buf)
7043                return ERR_PTR(-EINVAL);
7044
7045        obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7046        if (IS_ERR(obj))
7047                return obj;
7048
7049        obj->log_buf = log_buf;
7050        obj->log_size = log_size;
7051        obj->log_level = log_level;
7052
7053        btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
7054        if (btf_tmp_path) {
7055                if (strlen(btf_tmp_path) >= PATH_MAX) {
7056                        err = -ENAMETOOLONG;
7057                        goto out;
7058                }
7059                obj->btf_custom_path = strdup(btf_tmp_path);
7060                if (!obj->btf_custom_path) {
7061                        err = -ENOMEM;
7062                        goto out;
7063                }
7064        }
7065
7066        kconfig = OPTS_GET(opts, kconfig, NULL);
7067        if (kconfig) {
7068                obj->kconfig = strdup(kconfig);
7069                if (!obj->kconfig) {
7070                        err = -ENOMEM;
7071                        goto out;
7072                }
7073        }
7074
7075        err = bpf_object__elf_init(obj);
7076        err = err ? : bpf_object__check_endianness(obj);
7077        err = err ? : bpf_object__elf_collect(obj);
7078        err = err ? : bpf_object__collect_externs(obj);
7079        err = err ? : bpf_object__finalize_btf(obj);
7080        err = err ? : bpf_object__init_maps(obj, opts);
7081        err = err ? : bpf_object_init_progs(obj, opts);
7082        err = err ? : bpf_object__collect_relos(obj);
7083        if (err)
7084                goto out;
7085
7086        bpf_object__elf_finish(obj);
7087
7088        return obj;
7089out:
7090        bpf_object__close(obj);
7091        return ERR_PTR(err);
7092}
7093
7094static struct bpf_object *
7095__bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
7096{
7097        DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7098                .relaxed_maps = flags & MAPS_RELAX_COMPAT,
7099        );
7100
7101        /* param validation */
7102        if (!attr->file)
7103                return NULL;
7104
7105        pr_debug("loading %s\n", attr->file);
7106        return bpf_object_open(attr->file, NULL, 0, &opts);
7107}
7108
7109struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
7110{
7111        return libbpf_ptr(__bpf_object__open_xattr(attr, 0));
7112}
7113
7114struct bpf_object *bpf_object__open(const char *path)
7115{
7116        struct bpf_object_open_attr attr = {
7117                .file           = path,
7118                .prog_type      = BPF_PROG_TYPE_UNSPEC,
7119        };
7120
7121        return libbpf_ptr(__bpf_object__open_xattr(&attr, 0));
7122}
7123
7124struct bpf_object *
7125bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7126{
7127        if (!path)
7128                return libbpf_err_ptr(-EINVAL);
7129
7130        pr_debug("loading %s\n", path);
7131
7132        return libbpf_ptr(bpf_object_open(path, NULL, 0, opts));
7133}
7134
7135struct bpf_object *
7136bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7137                     const struct bpf_object_open_opts *opts)
7138{
7139        if (!obj_buf || obj_buf_sz == 0)
7140                return libbpf_err_ptr(-EINVAL);
7141
7142        return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts));
7143}
7144
7145struct bpf_object *
7146bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
7147                        const char *name)
7148{
7149        DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7150                .object_name = name,
7151                /* wrong default, but backwards-compatible */
7152                .relaxed_maps = true,
7153        );
7154
7155        /* returning NULL is wrong, but backwards-compatible */
7156        if (!obj_buf || obj_buf_sz == 0)
7157                return errno = EINVAL, NULL;
7158
7159        return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, &opts));
7160}
7161
7162static int bpf_object_unload(struct bpf_object *obj)
7163{
7164        size_t i;
7165
7166        if (!obj)
7167                return libbpf_err(-EINVAL);
7168
7169        for (i = 0; i < obj->nr_maps; i++) {
7170                zclose(obj->maps[i].fd);
7171                if (obj->maps[i].st_ops)
7172                        zfree(&obj->maps[i].st_ops->kern_vdata);
7173        }
7174
7175        for (i = 0; i < obj->nr_programs; i++)
7176                bpf_program__unload(&obj->programs[i]);
7177
7178        return 0;
7179}
7180
7181int bpf_object__unload(struct bpf_object *obj) __attribute__((alias("bpf_object_unload")));
7182
7183static int bpf_object__sanitize_maps(struct bpf_object *obj)
7184{
7185        struct bpf_map *m;
7186
7187        bpf_object__for_each_map(m, obj) {
7188                if (!bpf_map__is_internal(m))
7189                        continue;
7190                if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
7191                        m->def.map_flags ^= BPF_F_MMAPABLE;
7192        }
7193
7194        return 0;
7195}
7196
7197int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
7198{
7199        char sym_type, sym_name[500];
7200        unsigned long long sym_addr;
7201        int ret, err = 0;
7202        FILE *f;
7203
7204        f = fopen("/proc/kallsyms", "r");
7205        if (!f) {
7206                err = -errno;
7207                pr_warn("failed to open /proc/kallsyms: %d\n", err);
7208                return err;
7209        }
7210
7211        while (true) {
7212                ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7213                             &sym_addr, &sym_type, sym_name);
7214                if (ret == EOF && feof(f))
7215                        break;
7216                if (ret != 3) {
7217                        pr_warn("failed to read kallsyms entry: %d\n", ret);
7218                        err = -EINVAL;
7219                        break;
7220                }
7221
7222                err = cb(sym_addr, sym_type, sym_name, ctx);
7223                if (err)
7224                        break;
7225        }
7226
7227        fclose(f);
7228        return err;
7229}
7230
7231static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
7232                       const char *sym_name, void *ctx)
7233{
7234        struct bpf_object *obj = ctx;
7235        const struct btf_type *t;
7236        struct extern_desc *ext;
7237
7238        ext = find_extern_by_name(obj, sym_name);
7239        if (!ext || ext->type != EXT_KSYM)
7240                return 0;
7241
7242        t = btf__type_by_id(obj->btf, ext->btf_id);
7243        if (!btf_is_var(t))
7244                return 0;
7245
7246        if (ext->is_set && ext->ksym.addr != sym_addr) {
7247                pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
7248                        sym_name, ext->ksym.addr, sym_addr);
7249                return -EINVAL;
7250        }
7251        if (!ext->is_set) {
7252                ext->is_set = true;
7253                ext->ksym.addr = sym_addr;
7254                pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
7255        }
7256        return 0;
7257}
7258
7259static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7260{
7261        return libbpf_kallsyms_parse(kallsyms_cb, obj);
7262}
7263
7264static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
7265                            __u16 kind, struct btf **res_btf,
7266                            struct module_btf **res_mod_btf)
7267{
7268        struct module_btf *mod_btf;
7269        struct btf *btf;
7270        int i, id, err;
7271
7272        btf = obj->btf_vmlinux;
7273        mod_btf = NULL;
7274        id = btf__find_by_name_kind(btf, ksym_name, kind);
7275
7276        if (id == -ENOENT) {
7277                err = load_module_btfs(obj);
7278                if (err)
7279                        return err;
7280
7281                for (i = 0; i < obj->btf_module_cnt; i++) {
7282                        /* we assume module_btf's BTF FD is always >0 */
7283                        mod_btf = &obj->btf_modules[i];
7284                        btf = mod_btf->btf;
7285                        id = btf__find_by_name_kind_own(btf, ksym_name, kind);
7286                        if (id != -ENOENT)
7287                                break;
7288                }
7289        }
7290        if (id <= 0)
7291                return -ESRCH;
7292
7293        *res_btf = btf;
7294        *res_mod_btf = mod_btf;
7295        return id;
7296}
7297
7298static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
7299                                               struct extern_desc *ext)
7300{
7301        const struct btf_type *targ_var, *targ_type;
7302        __u32 targ_type_id, local_type_id;
7303        struct module_btf *mod_btf = NULL;
7304        const char *targ_var_name;
7305        struct btf *btf = NULL;
7306        int id, err;
7307
7308        id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
7309        if (id < 0) {
7310                if (id == -ESRCH && ext->is_weak)
7311                        return 0;
7312                pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
7313                        ext->name);
7314                return id;
7315        }
7316
7317        /* find local type_id */
7318        local_type_id = ext->ksym.type_id;
7319
7320        /* find target type_id */
7321        targ_var = btf__type_by_id(btf, id);
7322        targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7323        targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7324
7325        err = bpf_core_types_are_compat(obj->btf, local_type_id,
7326                                        btf, targ_type_id);
7327        if (err <= 0) {
7328                const struct btf_type *local_type;
7329                const char *targ_name, *local_name;
7330
7331                local_type = btf__type_by_id(obj->btf, local_type_id);
7332                local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7333                targ_name = btf__name_by_offset(btf, targ_type->name_off);
7334
7335                pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7336                        ext->name, local_type_id,
7337                        btf_kind_str(local_type), local_name, targ_type_id,
7338                        btf_kind_str(targ_type), targ_name);
7339                return -EINVAL;
7340        }
7341
7342        ext->is_set = true;
7343        ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7344        ext->ksym.kernel_btf_id = id;
7345        pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
7346                 ext->name, id, btf_kind_str(targ_var), targ_var_name);
7347
7348        return 0;
7349}
7350
7351static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
7352                                                struct extern_desc *ext)
7353{
7354        int local_func_proto_id, kfunc_proto_id, kfunc_id;
7355        struct module_btf *mod_btf = NULL;
7356        const struct btf_type *kern_func;
7357        struct btf *kern_btf = NULL;
7358        int ret;
7359
7360        local_func_proto_id = ext->ksym.type_id;
7361
7362        kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC, &kern_btf, &mod_btf);
7363        if (kfunc_id < 0) {
7364                if (kfunc_id == -ESRCH && ext->is_weak)
7365                        return 0;
7366                pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
7367                        ext->name);
7368                return kfunc_id;
7369        }
7370
7371        kern_func = btf__type_by_id(kern_btf, kfunc_id);
7372        kfunc_proto_id = kern_func->type;
7373
7374        ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
7375                                        kern_btf, kfunc_proto_id);
7376        if (ret <= 0) {
7377                pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n",
7378                        ext->name, local_func_proto_id, kfunc_proto_id);
7379                return -EINVAL;
7380        }
7381
7382        /* set index for module BTF fd in fd_array, if unset */
7383        if (mod_btf && !mod_btf->fd_array_idx) {
7384                /* insn->off is s16 */
7385                if (obj->fd_array_cnt == INT16_MAX) {
7386                        pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
7387                                ext->name, mod_btf->fd_array_idx);
7388                        return -E2BIG;
7389                }
7390                /* Cannot use index 0 for module BTF fd */
7391                if (!obj->fd_array_cnt)
7392                        obj->fd_array_cnt = 1;
7393
7394                ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
7395                                        obj->fd_array_cnt + 1);
7396                if (ret)
7397                        return ret;
7398                mod_btf->fd_array_idx = obj->fd_array_cnt;
7399                /* we assume module BTF FD is always >0 */
7400                obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
7401        }
7402
7403        ext->is_set = true;
7404        ext->ksym.kernel_btf_id = kfunc_id;
7405        ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
7406        pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n",
7407                 ext->name, kfunc_id);
7408
7409        return 0;
7410}
7411
7412static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7413{
7414        const struct btf_type *t;
7415        struct extern_desc *ext;
7416        int i, err;
7417
7418        for (i = 0; i < obj->nr_extern; i++) {
7419                ext = &obj->externs[i];
7420                if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7421                        continue;
7422
7423                if (obj->gen_loader) {
7424                        ext->is_set = true;
7425                        ext->ksym.kernel_btf_obj_fd = 0;
7426                        ext->ksym.kernel_btf_id = 0;
7427                        continue;
7428                }
7429                t = btf__type_by_id(obj->btf, ext->btf_id);
7430                if (btf_is_var(t))
7431                        err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
7432                else
7433                        err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
7434                if (err)
7435                        return err;
7436        }
7437        return 0;
7438}
7439
7440static int bpf_object__resolve_externs(struct bpf_object *obj,
7441                                       const char *extra_kconfig)
7442{
7443        bool need_config = false, need_kallsyms = false;
7444        bool need_vmlinux_btf = false;
7445        struct extern_desc *ext;
7446        void *kcfg_data = NULL;
7447        int err, i;
7448
7449        if (obj->nr_extern == 0)
7450                return 0;
7451
7452        if (obj->kconfig_map_idx >= 0)
7453                kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7454
7455        for (i = 0; i < obj->nr_extern; i++) {
7456                ext = &obj->externs[i];
7457
7458                if (ext->type == EXT_KCFG &&
7459                    strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7460                        void *ext_val = kcfg_data + ext->kcfg.data_off;
7461                        __u32 kver = get_kernel_version();
7462
7463                        if (!kver) {
7464                                pr_warn("failed to get kernel version\n");
7465                                return -EINVAL;
7466                        }
7467                        err = set_kcfg_value_num(ext, ext_val, kver);
7468                        if (err)
7469                                return err;
7470                        pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
7471                } else if (ext->type == EXT_KCFG && str_has_pfx(ext->name, "CONFIG_")) {
7472                        need_config = true;
7473                } else if (ext->type == EXT_KSYM) {
7474                        if (ext->ksym.type_id)
7475                                need_vmlinux_btf = true;
7476                        else
7477                                need_kallsyms = true;
7478                } else {
7479                        pr_warn("unrecognized extern '%s'\n", ext->name);
7480                        return -EINVAL;
7481                }
7482        }
7483        if (need_config && extra_kconfig) {
7484                err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
7485                if (err)
7486                        return -EINVAL;
7487                need_config = false;
7488                for (i = 0; i < obj->nr_extern; i++) {
7489                        ext = &obj->externs[i];
7490                        if (ext->type == EXT_KCFG && !ext->is_set) {
7491                                need_config = true;
7492                                break;
7493                        }
7494                }
7495        }
7496        if (need_config) {
7497                err = bpf_object__read_kconfig_file(obj, kcfg_data);
7498                if (err)
7499                        return -EINVAL;
7500        }
7501        if (need_kallsyms) {
7502                err = bpf_object__read_kallsyms_file(obj);
7503                if (err)
7504                        return -EINVAL;
7505        }
7506        if (need_vmlinux_btf) {
7507                err = bpf_object__resolve_ksyms_btf_id(obj);
7508                if (err)
7509                        return -EINVAL;
7510        }
7511        for (i = 0; i < obj->nr_extern; i++) {
7512                ext = &obj->externs[i];
7513
7514                if (!ext->is_set && !ext->is_weak) {
7515                        pr_warn("extern %s (strong) not resolved\n", ext->name);
7516                        return -ESRCH;
7517                } else if (!ext->is_set) {
7518                        pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
7519                                 ext->name);
7520                }
7521        }
7522
7523        return 0;
7524}
7525
7526static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
7527{
7528        int err, i;
7529
7530        if (!obj)
7531                return libbpf_err(-EINVAL);
7532
7533        if (obj->loaded) {
7534                pr_warn("object '%s': load can't be attempted twice\n", obj->name);
7535                return libbpf_err(-EINVAL);
7536        }
7537
7538        if (obj->gen_loader)
7539                bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
7540
7541        err = bpf_object__probe_loading(obj);
7542        err = err ? : bpf_object__load_vmlinux_btf(obj, false);
7543        err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
7544        err = err ? : bpf_object__sanitize_and_load_btf(obj);
7545        err = err ? : bpf_object__sanitize_maps(obj);
7546        err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
7547        err = err ? : bpf_object__create_maps(obj);
7548        err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
7549        err = err ? : bpf_object__load_progs(obj, extra_log_level);
7550        err = err ? : bpf_object_init_prog_arrays(obj);
7551
7552        if (obj->gen_loader) {
7553                /* reset FDs */
7554                if (obj->btf)
7555                        btf__set_fd(obj->btf, -1);
7556                for (i = 0; i < obj->nr_maps; i++)
7557                        obj->maps[i].fd = -1;
7558                if (!err)
7559                        err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
7560        }
7561
7562        /* clean up fd_array */
7563        zfree(&obj->fd_array);
7564
7565        /* clean up module BTFs */
7566        for (i = 0; i < obj->btf_module_cnt; i++) {
7567                close(obj->btf_modules[i].fd);
7568                btf__free(obj->btf_modules[i].btf);
7569                free(obj->btf_modules[i].name);
7570        }
7571        free(obj->btf_modules);
7572
7573        /* clean up vmlinux BTF */
7574        btf__free(obj->btf_vmlinux);
7575        obj->btf_vmlinux = NULL;
7576
7577        obj->loaded = true; /* doesn't matter if successfully or not */
7578
7579        if (err)
7580                goto out;
7581
7582        return 0;
7583out:
7584        /* unpin any maps that were auto-pinned during load */
7585        for (i = 0; i < obj->nr_maps; i++)
7586                if (obj->maps[i].pinned && !obj->maps[i].reused)
7587                        bpf_map__unpin(&obj->maps[i], NULL);
7588
7589        bpf_object_unload(obj);
7590        pr_warn("failed to load object '%s'\n", obj->path);
7591        return libbpf_err(err);
7592}
7593
7594int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
7595{
7596        return bpf_object_load(attr->obj, attr->log_level, attr->target_btf_path);
7597}
7598
7599int bpf_object__load(struct bpf_object *obj)
7600{
7601        return bpf_object_load(obj, 0, NULL);
7602}
7603
7604static int make_parent_dir(const char *path)
7605{
7606        char *cp, errmsg[STRERR_BUFSIZE];
7607        char *dname, *dir;
7608        int err = 0;
7609
7610        dname = strdup(path);
7611        if (dname == NULL)
7612                return -ENOMEM;
7613
7614        dir = dirname(dname);
7615        if (mkdir(dir, 0700) && errno != EEXIST)
7616                err = -errno;
7617
7618        free(dname);
7619        if (err) {
7620                cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7621                pr_warn("failed to mkdir %s: %s\n", path, cp);
7622        }
7623        return err;
7624}
7625
7626static int check_path(const char *path)
7627{
7628        char *cp, errmsg[STRERR_BUFSIZE];
7629        struct statfs st_fs;
7630        char *dname, *dir;
7631        int err = 0;
7632
7633        if (path == NULL)
7634                return -EINVAL;
7635
7636        dname = strdup(path);
7637        if (dname == NULL)
7638                return -ENOMEM;
7639
7640        dir = dirname(dname);
7641        if (statfs(dir, &st_fs)) {
7642                cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7643                pr_warn("failed to statfs %s: %s\n", dir, cp);
7644                err = -errno;
7645        }
7646        free(dname);
7647
7648        if (!err && st_fs.f_type != BPF_FS_MAGIC) {
7649                pr_warn("specified path %s is not on BPF FS\n", path);
7650                err = -EINVAL;
7651        }
7652
7653        return err;
7654}
7655
7656static int bpf_program_pin_instance(struct bpf_program *prog, const char *path, int instance)
7657{
7658        char *cp, errmsg[STRERR_BUFSIZE];
7659        int err;
7660
7661        err = make_parent_dir(path);
7662        if (err)
7663                return libbpf_err(err);
7664
7665        err = check_path(path);
7666        if (err)
7667                return libbpf_err(err);
7668
7669        if (prog == NULL) {
7670                pr_warn("invalid program pointer\n");
7671                return libbpf_err(-EINVAL);
7672        }
7673
7674        if (instance < 0 || instance >= prog->instances.nr) {
7675                pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7676                        instance, prog->name, prog->instances.nr);
7677                return libbpf_err(-EINVAL);
7678        }
7679
7680        if (bpf_obj_pin(prog->instances.fds[instance], path)) {
7681                err = -errno;
7682                cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
7683                pr_warn("failed to pin program: %s\n", cp);
7684                return libbpf_err(err);
7685        }
7686        pr_debug("pinned program '%s'\n", path);
7687
7688        return 0;
7689}
7690
7691static int bpf_program_unpin_instance(struct bpf_program *prog, const char *path, int instance)
7692{
7693        int err;
7694
7695        err = check_path(path);
7696        if (err)
7697                return libbpf_err(err);
7698
7699        if (prog == NULL) {
7700                pr_warn("invalid program pointer\n");
7701                return libbpf_err(-EINVAL);
7702        }
7703
7704        if (instance < 0 || instance >= prog->instances.nr) {
7705                pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7706                        instance, prog->name, prog->instances.nr);
7707                return libbpf_err(-EINVAL);
7708        }
7709
7710        err = unlink(path);
7711        if (err != 0)
7712                return libbpf_err(-errno);
7713
7714        pr_debug("unpinned program '%s'\n", path);
7715
7716        return 0;
7717}
7718
7719__attribute__((alias("bpf_program_pin_instance")))
7720int bpf_object__pin_instance(struct bpf_program *prog, const char *path, int instance);
7721
7722__attribute__((alias("bpf_program_unpin_instance")))
7723int bpf_program__unpin_instance(struct bpf_program *prog, const char *path, int instance);
7724
7725int bpf_program__pin(struct bpf_program *prog, const char *path)
7726{
7727        int i, err;
7728
7729        err = make_parent_dir(path);
7730        if (err)
7731                return libbpf_err(err);
7732
7733        err = check_path(path);
7734        if (err)
7735                return libbpf_err(err);
7736
7737        if (prog == NULL) {
7738                pr_warn("invalid program pointer\n");
7739                return libbpf_err(-EINVAL);
7740        }
7741
7742        if (prog->instances.nr <= 0) {
7743                pr_warn("no instances of prog %s to pin\n", prog->name);
7744                return libbpf_err(-EINVAL);
7745        }
7746
7747        if (prog->instances.nr == 1) {
7748                /* don't create subdirs when pinning single instance */
7749                return bpf_program_pin_instance(prog, path, 0);
7750        }
7751
7752        for (i = 0; i < prog->instances.nr; i++) {
7753                char buf[PATH_MAX];
7754                int len;
7755
7756                len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7757                if (len < 0) {
7758                        err = -EINVAL;
7759                        goto err_unpin;
7760                } else if (len >= PATH_MAX) {
7761                        err = -ENAMETOOLONG;
7762                        goto err_unpin;
7763                }
7764
7765                err = bpf_program_pin_instance(prog, buf, i);
7766                if (err)
7767                        goto err_unpin;
7768        }
7769
7770        return 0;
7771
7772err_unpin:
7773        for (i = i - 1; i >= 0; i--) {
7774                char buf[PATH_MAX];
7775                int len;
7776
7777                len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7778                if (len < 0)
7779                        continue;
7780                else if (len >= PATH_MAX)
7781                        continue;
7782
7783                bpf_program_unpin_instance(prog, buf, i);
7784        }
7785
7786        rmdir(path);
7787
7788        return libbpf_err(err);
7789}
7790
7791int bpf_program__unpin(struct bpf_program *prog, const char *path)
7792{
7793        int i, err;
7794
7795        err = check_path(path);
7796        if (err)
7797                return libbpf_err(err);
7798
7799        if (prog == NULL) {
7800                pr_warn("invalid program pointer\n");
7801                return libbpf_err(-EINVAL);
7802        }
7803
7804        if (prog->instances.nr <= 0) {
7805                pr_warn("no instances of prog %s to pin\n", prog->name);
7806                return libbpf_err(-EINVAL);
7807        }
7808
7809        if (prog->instances.nr == 1) {
7810                /* don't create subdirs when pinning single instance */
7811                return bpf_program_unpin_instance(prog, path, 0);
7812        }
7813
7814        for (i = 0; i < prog->instances.nr; i++) {
7815                char buf[PATH_MAX];
7816                int len;
7817
7818                len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7819                if (len < 0)
7820                        return libbpf_err(-EINVAL);
7821                else if (len >= PATH_MAX)
7822                        return libbpf_err(-ENAMETOOLONG);
7823
7824                err = bpf_program_unpin_instance(prog, buf, i);
7825                if (err)
7826                        return err;
7827        }
7828
7829        err = rmdir(path);
7830        if (err)
7831                return libbpf_err(-errno);
7832
7833        return 0;
7834}
7835
7836int bpf_map__pin(struct bpf_map *map, const char *path)
7837{
7838        char *cp, errmsg[STRERR_BUFSIZE];
7839        int err;
7840
7841        if (map == NULL) {
7842                pr_warn("invalid map pointer\n");
7843                return libbpf_err(-EINVAL);
7844        }
7845
7846        if (map->pin_path) {
7847                if (path && strcmp(path, map->pin_path)) {
7848                        pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7849                                bpf_map__name(map), map->pin_path, path);
7850                        return libbpf_err(-EINVAL);
7851                } else if (map->pinned) {
7852                        pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
7853                                 bpf_map__name(map), map->pin_path);
7854                        return 0;
7855                }
7856        } else {
7857                if (!path) {
7858                        pr_warn("missing a path to pin map '%s' at\n",
7859                                bpf_map__name(map));
7860                        return libbpf_err(-EINVAL);
7861                } else if (map->pinned) {
7862                        pr_warn("map '%s' already pinned\n", bpf_map__name(map));
7863                        return libbpf_err(-EEXIST);
7864                }
7865
7866                map->pin_path = strdup(path);
7867                if (!map->pin_path) {
7868                        err = -errno;
7869                        goto out_err;
7870                }
7871        }
7872
7873        err = make_parent_dir(map->pin_path);
7874        if (err)
7875                return libbpf_err(err);
7876
7877        err = check_path(map->pin_path);
7878        if (err)
7879                return libbpf_err(err);
7880
7881        if (bpf_obj_pin(map->fd, map->pin_path)) {
7882                err = -errno;
7883                goto out_err;
7884        }
7885
7886        map->pinned = true;
7887        pr_debug("pinned map '%s'\n", map->pin_path);
7888
7889        return 0;
7890
7891out_err:
7892        cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7893        pr_warn("failed to pin map: %s\n", cp);
7894        return libbpf_err(err);
7895}
7896
7897int bpf_map__unpin(struct bpf_map *map, const char *path)
7898{
7899        int err;
7900
7901        if (map == NULL) {
7902                pr_warn("invalid map pointer\n");
7903                return libbpf_err(-EINVAL);
7904        }
7905
7906        if (map->pin_path) {
7907                if (path && strcmp(path, map->pin_path)) {
7908                        pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7909                                bpf_map__name(map), map->pin_path, path);
7910                        return libbpf_err(-EINVAL);
7911                }
7912                path = map->pin_path;
7913        } else if (!path) {
7914                pr_warn("no path to unpin map '%s' from\n",
7915                        bpf_map__name(map));
7916                return libbpf_err(-EINVAL);
7917        }
7918
7919        err = check_path(path);
7920        if (err)
7921                return libbpf_err(err);
7922
7923        err = unlink(path);
7924        if (err != 0)
7925                return libbpf_err(-errno);
7926
7927        map->pinned = false;
7928        pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
7929
7930        return 0;
7931}
7932
7933int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
7934{
7935        char *new = NULL;
7936
7937        if (path) {
7938                new = strdup(path);
7939                if (!new)
7940                        return libbpf_err(-errno);
7941        }
7942
7943        free(map->pin_path);
7944        map->pin_path = new;
7945        return 0;
7946}
7947
7948__alias(bpf_map__pin_path)
7949const char *bpf_map__get_pin_path(const struct bpf_map *map);
7950
7951const char *bpf_map__pin_path(const struct bpf_map *map)
7952{
7953        return map->pin_path;
7954}
7955
7956bool bpf_map__is_pinned(const struct bpf_map *map)
7957{
7958        return map->pinned;
7959}
7960
7961static void sanitize_pin_path(char *s)
7962{
7963        /* bpffs disallows periods in path names */
7964        while (*s) {
7965                if (*s == '.')
7966                        *s = '_';
7967                s++;
7968        }
7969}
7970
7971int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
7972{
7973        struct bpf_map *map;
7974        int err;
7975
7976        if (!obj)
7977                return libbpf_err(-ENOENT);
7978
7979        if (!obj->loaded) {
7980                pr_warn("object not yet loaded; load it first\n");
7981                return libbpf_err(-ENOENT);
7982        }
7983
7984        bpf_object__for_each_map(map, obj) {
7985                char *pin_path = NULL;
7986                char buf[PATH_MAX];
7987
7988                if (map->skipped)
7989                        continue;
7990
7991                if (path) {
7992                        int len;
7993
7994                        len = snprintf(buf, PATH_MAX, "%s/%s", path,
7995                                       bpf_map__name(map));
7996                        if (len < 0) {
7997                                err = -EINVAL;
7998                                goto err_unpin_maps;
7999                        } else if (len >= PATH_MAX) {
8000                                err = -ENAMETOOLONG;
8001                                goto err_unpin_maps;
8002                        }
8003                        sanitize_pin_path(buf);
8004                        pin_path = buf;
8005                } else if (!map->pin_path) {
8006                        continue;
8007                }
8008
8009                err = bpf_map__pin(map, pin_path);
8010                if (err)
8011                        goto err_unpin_maps;
8012        }
8013
8014        return 0;
8015
8016err_unpin_maps:
8017        while ((map = bpf_object__prev_map(obj, map))) {
8018                if (!map->pin_path)
8019                        continue;
8020
8021                bpf_map__unpin(map, NULL);
8022        }
8023
8024        return libbpf_err(err);
8025}
8026
8027int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
8028{
8029        struct bpf_map *map;
8030        int err;
8031
8032        if (!obj)
8033                return libbpf_err(-ENOENT);
8034
8035        bpf_object__for_each_map(map, obj) {
8036                char *pin_path = NULL;
8037                char buf[PATH_MAX];
8038
8039                if (path) {
8040                        int len;
8041
8042                        len = snprintf(buf, PATH_MAX, "%s/%s", path,
8043                                       bpf_map__name(map));
8044                        if (len < 0)
8045                                return libbpf_err(-EINVAL);
8046                        else if (len >= PATH_MAX)
8047                                return libbpf_err(-ENAMETOOLONG);
8048                        sanitize_pin_path(buf);
8049                        pin_path = buf;
8050                } else if (!map->pin_path) {
8051                        continue;
8052                }
8053
8054                err = bpf_map__unpin(map, pin_path);
8055                if (err)
8056                        return libbpf_err(err);
8057        }
8058
8059        return 0;
8060}
8061
8062int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8063{
8064        struct bpf_program *prog;
8065        int err;
8066
8067        if (!obj)
8068                return libbpf_err(-ENOENT);
8069
8070        if (!obj->loaded) {
8071                pr_warn("object not yet loaded; load it first\n");
8072                return libbpf_err(-ENOENT);
8073        }
8074
8075        bpf_object__for_each_program(prog, obj) {
8076                char buf[PATH_MAX];
8077                int len;
8078
8079                len = snprintf(buf, PATH_MAX, "%s/%s", path,
8080                               prog->pin_name);
8081                if (len < 0) {
8082                        err = -EINVAL;
8083                        goto err_unpin_programs;
8084                } else if (len >= PATH_MAX) {
8085                        err = -ENAMETOOLONG;
8086                        goto err_unpin_programs;
8087                }
8088
8089                err = bpf_program__pin(prog, buf);
8090                if (err)
8091                        goto err_unpin_programs;
8092        }
8093
8094        return 0;
8095
8096err_unpin_programs:
8097        while ((prog = bpf_object__prev_program(obj, prog))) {
8098                char buf[PATH_MAX];
8099                int len;
8100
8101                len = snprintf(buf, PATH_MAX, "%s/%s", path,
8102                               prog->pin_name);
8103                if (len < 0)
8104                        continue;
8105                else if (len >= PATH_MAX)
8106                        continue;
8107
8108                bpf_program__unpin(prog, buf);
8109        }
8110
8111        return libbpf_err(err);
8112}
8113
8114int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8115{
8116        struct bpf_program *prog;
8117        int err;
8118
8119        if (!obj)
8120                return libbpf_err(-ENOENT);
8121
8122        bpf_object__for_each_program(prog, obj) {
8123                char buf[PATH_MAX];
8124                int len;
8125
8126                len = snprintf(buf, PATH_MAX, "%s/%s", path,
8127                               prog->pin_name);
8128                if (len < 0)
8129                        return libbpf_err(-EINVAL);
8130                else if (len >= PATH_MAX)
8131                        return libbpf_err(-ENAMETOOLONG);
8132
8133                err = bpf_program__unpin(prog, buf);
8134                if (err)
8135                        return libbpf_err(err);
8136        }
8137
8138        return 0;
8139}
8140
8141int bpf_object__pin(struct bpf_object *obj, const char *path)
8142{
8143        int err;
8144
8145        err = bpf_object__pin_maps(obj, path);
8146        if (err)
8147                return libbpf_err(err);
8148
8149        err = bpf_object__pin_programs(obj, path);
8150        if (err) {
8151                bpf_object__unpin_maps(obj, path);
8152                return libbpf_err(err);
8153        }
8154
8155        return 0;
8156}
8157
8158static void bpf_map__destroy(struct bpf_map *map)
8159{
8160        if (map->clear_priv)
8161                map->clear_priv(map, map->priv);
8162        map->priv = NULL;
8163        map->clear_priv = NULL;
8164
8165        if (map->inner_map) {
8166                bpf_map__destroy(map->inner_map);
8167                zfree(&map->inner_map);
8168        }
8169
8170        zfree(&map->init_slots);
8171        map->init_slots_sz = 0;
8172
8173        if (map->mmaped) {
8174                munmap(map->mmaped, bpf_map_mmap_sz(map));
8175                map->mmaped = NULL;
8176        }
8177
8178        if (map->st_ops) {
8179                zfree(&map->st_ops->data);
8180                zfree(&map->st_ops->progs);
8181                zfree(&map->st_ops->kern_func_off);
8182                zfree(&map->st_ops);
8183        }
8184
8185        zfree(&map->name);
8186        zfree(&map->real_name);
8187        zfree(&map->pin_path);
8188
8189        if (map->fd >= 0)
8190                zclose(map->fd);
8191}
8192
8193void bpf_object__close(struct bpf_object *obj)
8194{
8195        size_t i;
8196
8197        if (IS_ERR_OR_NULL(obj))
8198                return;
8199
8200        if (obj->clear_priv)
8201                obj->clear_priv(obj, obj->priv);
8202
8203        bpf_gen__free(obj->gen_loader);
8204        bpf_object__elf_finish(obj);
8205        bpf_object_unload(obj);
8206        btf__free(obj->btf);
8207        btf_ext__free(obj->btf_ext);
8208
8209        for (i = 0; i < obj->nr_maps; i++)
8210                bpf_map__destroy(&obj->maps[i]);
8211
8212        zfree(&obj->btf_custom_path);
8213        zfree(&obj->kconfig);
8214        zfree(&obj->externs);
8215        obj->nr_extern = 0;
8216
8217        zfree(&obj->maps);
8218        obj->nr_maps = 0;
8219
8220        if (obj->programs && obj->nr_programs) {
8221                for (i = 0; i < obj->nr_programs; i++)
8222                        bpf_program__exit(&obj->programs[i]);
8223        }
8224        zfree(&obj->programs);
8225
8226        list_del(&obj->list);
8227        free(obj);
8228}
8229
8230struct bpf_object *
8231bpf_object__next(struct bpf_object *prev)
8232{
8233        struct bpf_object *next;
8234        bool strict = (libbpf_mode & LIBBPF_STRICT_NO_OBJECT_LIST);
8235
8236        if (strict)
8237                return NULL;
8238
8239        if (!prev)
8240                next = list_first_entry(&bpf_objects_list,
8241                                        struct bpf_object,
8242                                        list);
8243        else
8244                next = list_next_entry(prev, list);
8245
8246        /* Empty list is noticed here so don't need checking on entry. */
8247        if (&next->list == &bpf_objects_list)
8248                return NULL;
8249
8250        return next;
8251}
8252
8253const char *bpf_object__name(const struct bpf_object *obj)
8254{
8255        return obj ? obj->name : libbpf_err_ptr(-EINVAL);
8256}
8257
8258unsigned int bpf_object__kversion(const struct bpf_object *obj)
8259{
8260        return obj ? obj->kern_version : 0;
8261}
8262
8263struct btf *bpf_object__btf(const struct bpf_object *obj)
8264{
8265        return obj ? obj->btf : NULL;
8266}
8267
8268int bpf_object__btf_fd(const struct bpf_object *obj)
8269{
8270        return obj->btf ? btf__fd(obj->btf) : -1;
8271}
8272
8273int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
8274{
8275        if (obj->loaded)
8276                return libbpf_err(-EINVAL);
8277
8278        obj->kern_version = kern_version;
8279
8280        return 0;
8281}
8282
8283int bpf_object__set_priv(struct bpf_object *obj, void *priv,
8284                         bpf_object_clear_priv_t clear_priv)
8285{
8286        if (obj->priv && obj->clear_priv)
8287                obj->clear_priv(obj, obj->priv);
8288
8289        obj->priv = priv;
8290        obj->clear_priv = clear_priv;
8291        return 0;
8292}
8293
8294void *bpf_object__priv(const struct bpf_object *obj)
8295{
8296        return obj ? obj->priv : libbpf_err_ptr(-EINVAL);
8297}
8298
8299int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
8300{
8301        struct bpf_gen *gen;
8302
8303        if (!opts)
8304                return -EFAULT;
8305        if (!OPTS_VALID(opts, gen_loader_opts))
8306                return -EINVAL;
8307        gen = calloc(sizeof(*gen), 1);
8308        if (!gen)
8309                return -ENOMEM;
8310        gen->opts = opts;
8311        obj->gen_loader = gen;
8312        return 0;
8313}
8314
8315static struct bpf_program *
8316__bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8317                    bool forward)
8318{
8319        size_t nr_programs = obj->nr_programs;
8320        ssize_t idx;
8321
8322        if (!nr_programs)
8323                return NULL;
8324
8325        if (!p)
8326                /* Iter from the beginning */
8327                return forward ? &obj->programs[0] :
8328                        &obj->programs[nr_programs - 1];
8329
8330        if (p->obj != obj) {
8331                pr_warn("error: program handler doesn't match object\n");
8332                return errno = EINVAL, NULL;
8333        }
8334
8335        idx = (p - obj->programs) + (forward ? 1 : -1);
8336        if (idx >= obj->nr_programs || idx < 0)
8337                return NULL;
8338        return &obj->programs[idx];
8339}
8340
8341struct bpf_program *
8342bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
8343{
8344        return bpf_object__next_program(obj, prev);
8345}
8346
8347struct bpf_program *
8348bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
8349{
8350        struct bpf_program *prog = prev;
8351
8352        do {
8353                prog = __bpf_program__iter(prog, obj, true);
8354        } while (prog && prog_is_subprog(obj, prog));
8355
8356        return prog;
8357}
8358
8359struct bpf_program *
8360bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
8361{
8362        return bpf_object__prev_program(obj, next);
8363}
8364
8365struct bpf_program *
8366bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
8367{
8368        struct bpf_program *prog = next;
8369
8370        do {
8371                prog = __bpf_program__iter(prog, obj, false);
8372        } while (prog && prog_is_subprog(obj, prog));
8373
8374        return prog;
8375}
8376
8377int bpf_program__set_priv(struct bpf_program *prog, void *priv,
8378                          bpf_program_clear_priv_t clear_priv)
8379{
8380        if (prog->priv && prog->clear_priv)
8381                prog->clear_priv(prog, prog->priv);
8382
8383        prog->priv = priv;
8384        prog->clear_priv = clear_priv;
8385        return 0;
8386}
8387
8388void *bpf_program__priv(const struct bpf_program *prog)
8389{
8390        return prog ? prog->priv : libbpf_err_ptr(-EINVAL);
8391}
8392
8393void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8394{
8395        prog->prog_ifindex = ifindex;
8396}
8397
8398const char *bpf_program__name(const struct bpf_program *prog)
8399{
8400        return prog->name;
8401}
8402
8403const char *bpf_program__section_name(const struct bpf_program *prog)
8404{
8405        return prog->sec_name;
8406}
8407
8408const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
8409{
8410        const char *title;
8411
8412        title = prog->sec_name;
8413        if (needs_copy) {
8414                title = strdup(title);
8415                if (!title) {
8416                        pr_warn("failed to strdup program title\n");
8417                        return libbpf_err_ptr(-ENOMEM);
8418                }
8419        }
8420
8421        return title;
8422}
8423
8424bool bpf_program__autoload(const struct bpf_program *prog)
8425{
8426        return prog->load;
8427}
8428
8429int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8430{
8431        if (prog->obj->loaded)
8432                return libbpf_err(-EINVAL);
8433
8434        prog->load = autoload;
8435        return 0;
8436}
8437
8438static int bpf_program_nth_fd(const struct bpf_program *prog, int n);
8439
8440int bpf_program__fd(const struct bpf_program *prog)
8441{
8442        return bpf_program_nth_fd(prog, 0);
8443}
8444
8445size_t bpf_program__size(const struct bpf_program *prog)
8446{
8447        return prog->insns_cnt * BPF_INSN_SZ;
8448}
8449
8450const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
8451{
8452        return prog->insns;
8453}
8454
8455size_t bpf_program__insn_cnt(const struct bpf_program *prog)
8456{
8457        return prog->insns_cnt;
8458}
8459
8460int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
8461                          bpf_program_prep_t prep)
8462{
8463        int *instances_fds;
8464
8465        if (nr_instances <= 0 || !prep)
8466                return libbpf_err(-EINVAL);
8467
8468        if (prog->instances.nr > 0 || prog->instances.fds) {
8469                pr_warn("Can't set pre-processor after loading\n");
8470                return libbpf_err(-EINVAL);
8471        }
8472
8473        instances_fds = malloc(sizeof(int) * nr_instances);
8474        if (!instances_fds) {
8475                pr_warn("alloc memory failed for fds\n");
8476                return libbpf_err(-ENOMEM);
8477        }
8478
8479        /* fill all fd with -1 */
8480        memset(instances_fds, -1, sizeof(int) * nr_instances);
8481
8482        prog->instances.nr = nr_instances;
8483        prog->instances.fds = instances_fds;
8484        prog->preprocessor = prep;
8485        return 0;
8486}
8487
8488__attribute__((alias("bpf_program_nth_fd")))
8489int bpf_program__nth_fd(const struct bpf_program *prog, int n);
8490
8491static int bpf_program_nth_fd(const struct bpf_program *prog, int n)
8492{
8493        int fd;
8494
8495        if (!prog)
8496                return libbpf_err(-EINVAL);
8497
8498        if (n >= prog->instances.nr || n < 0) {
8499                pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
8500                        n, prog->name, prog->instances.nr);
8501                return libbpf_err(-EINVAL);
8502        }
8503
8504        fd = prog->instances.fds[n];
8505        if (fd < 0) {
8506                pr_warn("%dth instance of program '%s' is invalid\n",
8507                        n, prog->name);
8508                return libbpf_err(-ENOENT);
8509        }
8510
8511        return fd;
8512}
8513
8514__alias(bpf_program__type)
8515enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
8516
8517enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
8518{
8519        return prog->type;
8520}
8521
8522void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8523{
8524        prog->type = type;
8525}
8526
8527static bool bpf_program__is_type(const struct bpf_program *prog,
8528                                 enum bpf_prog_type type)
8529{
8530        return prog ? (prog->type == type) : false;
8531}
8532
8533#define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
8534int bpf_program__set_##NAME(struct bpf_program *prog)           \
8535{                                                               \
8536        if (!prog)                                              \
8537                return libbpf_err(-EINVAL);                     \
8538        bpf_program__set_type(prog, TYPE);                      \
8539        return 0;                                               \
8540}                                                               \
8541                                                                \
8542bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
8543{                                                               \
8544        return bpf_program__is_type(prog, TYPE);                \
8545}                                                               \
8546
8547BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
8548BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
8549BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
8550BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
8551BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
8552BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
8553BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
8554BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
8555BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
8556BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
8557BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
8558BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
8559BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
8560
8561__alias(bpf_program__expected_attach_type)
8562enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
8563
8564enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
8565{
8566        return prog->expected_attach_type;
8567}
8568
8569void bpf_program__set_expected_attach_type(struct bpf_program *prog,
8570                                           enum bpf_attach_type type)
8571{
8572        prog->expected_attach_type = type;
8573}
8574
8575__u32 bpf_program__flags(const struct bpf_program *prog)
8576{
8577        return prog->prog_flags;
8578}
8579
8580int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
8581{
8582        if (prog->obj->loaded)
8583                return libbpf_err(-EBUSY);
8584
8585        prog->prog_flags = flags;
8586        return 0;
8587}
8588
8589__u32 bpf_program__log_level(const struct bpf_program *prog)
8590{
8591        return prog->log_level;
8592}
8593
8594int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
8595{
8596        if (prog->obj->loaded)
8597                return libbpf_err(-EBUSY);
8598
8599        prog->log_level = log_level;
8600        return 0;
8601}
8602
8603const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
8604{
8605        *log_size = prog->log_size;
8606        return prog->log_buf;
8607}
8608
8609int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
8610{
8611        if (log_size && !log_buf)
8612                return -EINVAL;
8613        if (prog->log_size > UINT_MAX)
8614                return -EINVAL;
8615        if (prog->obj->loaded)
8616                return -EBUSY;
8617
8618        prog->log_buf = log_buf;
8619        prog->log_size = log_size;
8620        return 0;
8621}
8622
8623#define SEC_DEF(sec_pfx, ptype, atype, flags, ...) {                        \
8624        .sec = (char *)sec_pfx,                                             \
8625        .prog_type = BPF_PROG_TYPE_##ptype,                                 \
8626        .expected_attach_type = atype,                                      \
8627        .cookie = (long)(flags),                                            \
8628        .prog_prepare_load_fn = libbpf_prepare_prog_load,                   \
8629        __VA_ARGS__                                                         \
8630}
8631
8632static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8633static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8634static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8635static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8636static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8637static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8638static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8639
8640static const struct bpf_sec_def section_defs[] = {
8641        SEC_DEF("socket",               SOCKET_FILTER, 0, SEC_NONE | SEC_SLOPPY_PFX),
8642        SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8643        SEC_DEF("sk_reuseport",         SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8644        SEC_DEF("kprobe/",              KPROBE, 0, SEC_NONE, attach_kprobe),
8645        SEC_DEF("uprobe/",              KPROBE, 0, SEC_NONE),
8646        SEC_DEF("kretprobe/",           KPROBE, 0, SEC_NONE, attach_kprobe),
8647        SEC_DEF("uretprobe/",           KPROBE, 0, SEC_NONE),
8648        SEC_DEF("kprobe.multi/",        KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8649        SEC_DEF("kretprobe.multi/",     KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8650        SEC_DEF("tc",                   SCHED_CLS, 0, SEC_NONE),
8651        SEC_DEF("classifier",           SCHED_CLS, 0, SEC_NONE | SEC_SLOPPY_PFX | SEC_DEPRECATED),
8652        SEC_DEF("action",               SCHED_ACT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8653        SEC_DEF("tracepoint/",          TRACEPOINT, 0, SEC_NONE, attach_tp),
8654        SEC_DEF("tp/",                  TRACEPOINT, 0, SEC_NONE, attach_tp),
8655        SEC_DEF("raw_tracepoint/",      RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8656        SEC_DEF("raw_tp/",              RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8657        SEC_DEF("raw_tracepoint.w/",    RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8658        SEC_DEF("raw_tp.w/",            RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8659        SEC_DEF("tp_btf/",              TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
8660        SEC_DEF("fentry/",              TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
8661        SEC_DEF("fmod_ret/",            TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
8662        SEC_DEF("fexit/",               TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
8663        SEC_DEF("fentry.s/",            TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8664        SEC_DEF("fmod_ret.s/",          TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8665        SEC_DEF("fexit.s/",             TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8666        SEC_DEF("freplace/",            EXT, 0, SEC_ATTACH_BTF, attach_trace),
8667        SEC_DEF("lsm/",                 LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
8668        SEC_DEF("lsm.s/",               LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
8669        SEC_DEF("iter/",                TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
8670        SEC_DEF("iter.s/",              TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
8671        SEC_DEF("syscall",              SYSCALL, 0, SEC_SLEEPABLE),
8672        SEC_DEF("xdp.frags/devmap",     XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
8673        SEC_DEF("xdp/devmap",           XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
8674        SEC_DEF("xdp_devmap/",          XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE | SEC_DEPRECATED),
8675        SEC_DEF("xdp.frags/cpumap",     XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
8676        SEC_DEF("xdp/cpumap",           XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
8677        SEC_DEF("xdp_cpumap/",          XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE | SEC_DEPRECATED),
8678        SEC_DEF("xdp.frags",            XDP, BPF_XDP, SEC_XDP_FRAGS),
8679        SEC_DEF("xdp",                  XDP, BPF_XDP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8680        SEC_DEF("perf_event",           PERF_EVENT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8681        SEC_DEF("lwt_in",               LWT_IN, 0, SEC_NONE | SEC_SLOPPY_PFX),
8682        SEC_DEF("lwt_out",              LWT_OUT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8683        SEC_DEF("lwt_xmit",             LWT_XMIT, 0, SEC_NONE | SEC_SLOPPY_PFX),
8684        SEC_DEF("lwt_seg6local",        LWT_SEG6LOCAL, 0, SEC_NONE | SEC_SLOPPY_PFX),
8685        SEC_DEF("cgroup_skb/ingress",   CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8686        SEC_DEF("cgroup_skb/egress",    CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8687        SEC_DEF("cgroup/skb",           CGROUP_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX),
8688        SEC_DEF("cgroup/sock_create",   CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8689        SEC_DEF("cgroup/sock_release",  CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8690        SEC_DEF("cgroup/sock",          CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8691        SEC_DEF("cgroup/post_bind4",    CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8692        SEC_DEF("cgroup/post_bind6",    CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8693        SEC_DEF("cgroup/dev",           CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8694        SEC_DEF("sockops",              SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8695        SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8696        SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8697        SEC_DEF("sk_skb",               SK_SKB, 0, SEC_NONE | SEC_SLOPPY_PFX),
8698        SEC_DEF("sk_msg",               SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8699        SEC_DEF("lirc_mode2",           LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8700        SEC_DEF("flow_dissector",       FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
8701        SEC_DEF("cgroup/bind4",         CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8702        SEC_DEF("cgroup/bind6",         CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8703        SEC_DEF("cgroup/connect4",      CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8704        SEC_DEF("cgroup/connect6",      CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8705        SEC_DEF("cgroup/sendmsg4",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8706        SEC_DEF("cgroup/sendmsg6",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8707        SEC_DEF("cgroup/recvmsg4",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8708        SEC_DEF("cgroup/recvmsg6",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8709        SEC_DEF("cgroup/getpeername4",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8710        SEC_DEF("cgroup/getpeername6",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8711        SEC_DEF("cgroup/getsockname4",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8712        SEC_DEF("cgroup/getsockname6",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8713        SEC_DEF("cgroup/sysctl",        CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8714        SEC_DEF("cgroup/getsockopt",    CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8715        SEC_DEF("cgroup/setsockopt",    CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8716        SEC_DEF("struct_ops+",          STRUCT_OPS, 0, SEC_NONE),
8717        SEC_DEF("sk_lookup",            SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
8718};
8719
8720static size_t custom_sec_def_cnt;
8721static struct bpf_sec_def *custom_sec_defs;
8722static struct bpf_sec_def custom_fallback_def;
8723static bool has_custom_fallback_def;
8724
8725static int last_custom_sec_def_handler_id;
8726
8727int libbpf_register_prog_handler(const char *sec,
8728                                 enum bpf_prog_type prog_type,
8729                                 enum bpf_attach_type exp_attach_type,
8730                                 const struct libbpf_prog_handler_opts *opts)
8731{
8732        struct bpf_sec_def *sec_def;
8733
8734        if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
8735                return libbpf_err(-EINVAL);
8736
8737        if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
8738                return libbpf_err(-E2BIG);
8739
8740        if (sec) {
8741                sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
8742                                              sizeof(*sec_def));
8743                if (!sec_def)
8744                        return libbpf_err(-ENOMEM);
8745
8746                custom_sec_defs = sec_def;
8747                sec_def = &custom_sec_defs[custom_sec_def_cnt];
8748        } else {
8749                if (has_custom_fallback_def)
8750                        return libbpf_err(-EBUSY);
8751
8752                sec_def = &custom_fallback_def;
8753        }
8754
8755        sec_def->sec = sec ? strdup(sec) : NULL;
8756        if (sec && !sec_def->sec)
8757                return libbpf_err(-ENOMEM);
8758
8759        sec_def->prog_type = prog_type;
8760        sec_def->expected_attach_type = exp_attach_type;
8761        sec_def->cookie = OPTS_GET(opts, cookie, 0);
8762
8763        sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
8764        sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
8765        sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
8766
8767        sec_def->handler_id = ++last_custom_sec_def_handler_id;
8768
8769        if (sec)
8770                custom_sec_def_cnt++;
8771        else
8772                has_custom_fallback_def = true;
8773
8774        return sec_def->handler_id;
8775}
8776
8777int libbpf_unregister_prog_handler(int handler_id)
8778{
8779        struct bpf_sec_def *sec_defs;
8780        int i;
8781
8782        if (handler_id <= 0)
8783                return libbpf_err(-EINVAL);
8784
8785        if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
8786                memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
8787                has_custom_fallback_def = false;
8788                return 0;
8789        }
8790
8791        for (i = 0; i < custom_sec_def_cnt; i++) {
8792                if (custom_sec_defs[i].handler_id == handler_id)
8793                        break;
8794        }
8795
8796        if (i == custom_sec_def_cnt)
8797                return libbpf_err(-ENOENT);
8798
8799        free(custom_sec_defs[i].sec);
8800        for (i = i + 1; i < custom_sec_def_cnt; i++)
8801                custom_sec_defs[i - 1] = custom_sec_defs[i];
8802        custom_sec_def_cnt--;
8803
8804        /* try to shrink the array, but it's ok if we couldn't */
8805        sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
8806        if (sec_defs)
8807                custom_sec_defs = sec_defs;
8808
8809        return 0;
8810}
8811
8812static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name,
8813                            bool allow_sloppy)
8814{
8815        size_t len = strlen(sec_def->sec);
8816
8817        /* "type/" always has to have proper SEC("type/extras") form */
8818        if (sec_def->sec[len - 1] == '/') {
8819                if (str_has_pfx(sec_name, sec_def->sec))
8820                        return true;
8821                return false;
8822        }
8823
8824        /* "type+" means it can be either exact SEC("type") or
8825         * well-formed SEC("type/extras") with proper '/' separator
8826         */
8827        if (sec_def->sec[len - 1] == '+') {
8828                len--;
8829                /* not even a prefix */
8830                if (strncmp(sec_name, sec_def->sec, len) != 0)
8831                        return false;
8832                /* exact match or has '/' separator */
8833                if (sec_name[len] == '\0' || sec_name[len] == '/')
8834                        return true;
8835                return false;
8836        }
8837
8838        /* SEC_SLOPPY_PFX definitions are allowed to be just prefix
8839         * matches, unless strict section name mode
8840         * (LIBBPF_STRICT_SEC_NAME) is enabled, in which case the
8841         * match has to be exact.
8842         */
8843        if (allow_sloppy && str_has_pfx(sec_name, sec_def->sec))
8844                return true;
8845
8846        /* Definitions not marked SEC_SLOPPY_PFX (e.g.,
8847         * SEC("syscall")) are exact matches in both modes.
8848         */
8849        return strcmp(sec_name, sec_def->sec) == 0;
8850}
8851
8852static const struct bpf_sec_def *find_sec_def(const char *sec_name)
8853{
8854        const struct bpf_sec_def *sec_def;
8855        int i, n;
8856        bool strict = libbpf_mode & LIBBPF_STRICT_SEC_NAME, allow_sloppy;
8857
8858        n = custom_sec_def_cnt;
8859        for (i = 0; i < n; i++) {
8860                sec_def = &custom_sec_defs[i];
8861                if (sec_def_matches(sec_def, sec_name, false))
8862                        return sec_def;
8863        }
8864
8865        n = ARRAY_SIZE(section_defs);
8866        for (i = 0; i < n; i++) {
8867                sec_def = &section_defs[i];
8868                allow_sloppy = (sec_def->cookie & SEC_SLOPPY_PFX) && !strict;
8869                if (sec_def_matches(sec_def, sec_name, allow_sloppy))
8870                        return sec_def;
8871        }
8872
8873        if (has_custom_fallback_def)
8874                return &custom_fallback_def;
8875
8876        return NULL;
8877}
8878
8879#define MAX_TYPE_NAME_SIZE 32
8880
8881static char *libbpf_get_type_names(bool attach_type)
8882{
8883        int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
8884        char *buf;
8885
8886        buf = malloc(len);
8887        if (!buf)
8888                return NULL;
8889
8890        buf[0] = '\0';
8891        /* Forge string buf with all available names */
8892        for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8893                const struct bpf_sec_def *sec_def = &section_defs[i];
8894
8895                if (attach_type) {
8896                        if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
8897                                continue;
8898
8899                        if (!(sec_def->cookie & SEC_ATTACHABLE))
8900                                continue;
8901                }
8902
8903                if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
8904                        free(buf);
8905                        return NULL;
8906                }
8907                strcat(buf, " ");
8908                strcat(buf, section_defs[i].sec);
8909        }
8910
8911        return buf;
8912}
8913
8914int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
8915                             enum bpf_attach_type *expected_attach_type)
8916{
8917        const struct bpf_sec_def *sec_def;
8918        char *type_names;
8919
8920        if (!name)
8921                return libbpf_err(-EINVAL);
8922
8923        sec_def = find_sec_def(name);
8924        if (sec_def) {
8925                *prog_type = sec_def->prog_type;
8926                *expected_attach_type = sec_def->expected_attach_type;
8927                return 0;
8928        }
8929
8930        pr_debug("failed to guess program type from ELF section '%s'\n", name);
8931        type_names = libbpf_get_type_names(false);
8932        if (type_names != NULL) {
8933                pr_debug("supported section(type) names are:%s\n", type_names);
8934                free(type_names);
8935        }
8936
8937        return libbpf_err(-ESRCH);
8938}
8939
8940static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
8941                                                     size_t offset)
8942{
8943        struct bpf_map *map;
8944        size_t i;
8945
8946        for (i = 0; i < obj->nr_maps; i++) {
8947                map = &obj->maps[i];
8948                if (!bpf_map__is_struct_ops(map))
8949                        continue;
8950                if (map->sec_offset <= offset &&
8951                    offset - map->sec_offset < map->def.value_size)
8952                        return map;
8953        }
8954
8955        return NULL;
8956}
8957
8958/* Collect the reloc from ELF and populate the st_ops->progs[] */
8959static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
8960                                            Elf64_Shdr *shdr, Elf_Data *data)
8961{
8962        const struct btf_member *member;
8963        struct bpf_struct_ops *st_ops;
8964        struct bpf_program *prog;
8965        unsigned int shdr_idx;
8966        const struct btf *btf;
8967        struct bpf_map *map;
8968        unsigned int moff, insn_idx;
8969        const char *name;
8970        __u32 member_idx;
8971        Elf64_Sym *sym;
8972        Elf64_Rel *rel;
8973        int i, nrels;
8974
8975        btf = obj->btf;
8976        nrels = shdr->sh_size / shdr->sh_entsize;
8977        for (i = 0; i < nrels; i++) {
8978                rel = elf_rel_by_idx(data, i);
8979                if (!rel) {
8980                        pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
8981                        return -LIBBPF_ERRNO__FORMAT;
8982                }
8983
8984                sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
8985                if (!sym) {
8986                        pr_warn("struct_ops reloc: symbol %zx not found\n",
8987                                (size_t)ELF64_R_SYM(rel->r_info));
8988                        return -LIBBPF_ERRNO__FORMAT;
8989                }
8990
8991                name = elf_sym_str(obj, sym->st_name) ?: "<?>";
8992                map = find_struct_ops_map_by_offset(obj, rel->r_offset);
8993                if (!map) {
8994                        pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
8995                                (size_t)rel->r_offset);
8996                        return -EINVAL;
8997                }
8998
8999                moff = rel->r_offset - map->sec_offset;
9000                shdr_idx = sym->st_shndx;
9001                st_ops = map->st_ops;
9002                pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
9003                         map->name,
9004                         (long long)(rel->r_info >> 32),
9005                         (long long)sym->st_value,
9006                         shdr_idx, (size_t)rel->r_offset,
9007                         map->sec_offset, sym->st_name, name);
9008
9009                if (shdr_idx >= SHN_LORESERVE) {
9010                        pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
9011                                map->name, (size_t)rel->r_offset, shdr_idx);
9012                        return -LIBBPF_ERRNO__RELOC;
9013                }
9014                if (sym->st_value % BPF_INSN_SZ) {
9015                        pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
9016                                map->name, (unsigned long long)sym->st_value);
9017                        return -LIBBPF_ERRNO__FORMAT;
9018                }
9019                insn_idx = sym->st_value / BPF_INSN_SZ;
9020
9021                member = find_member_by_offset(st_ops->type, moff * 8);
9022                if (!member) {
9023                        pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
9024                                map->name, moff);
9025                        return -EINVAL;
9026                }
9027                member_idx = member - btf_members(st_ops->type);
9028                name = btf__name_by_offset(btf, member->name_off);
9029
9030                if (!resolve_func_ptr(btf, member->type, NULL)) {
9031                        pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
9032                                map->name, name);
9033                        return -EINVAL;
9034                }
9035
9036                prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
9037                if (!prog) {
9038                        pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
9039                                map->name, shdr_idx, name);
9040                        return -EINVAL;
9041                }
9042
9043                /* prevent the use of BPF prog with invalid type */
9044                if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
9045                        pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
9046                                map->name, prog->name);
9047                        return -EINVAL;
9048                }
9049
9050                /* if we haven't yet processed this BPF program, record proper
9051                 * attach_btf_id and member_idx
9052                 */
9053                if (!prog->attach_btf_id) {
9054                        prog->attach_btf_id = st_ops->type_id;
9055                        prog->expected_attach_type = member_idx;
9056                }
9057
9058                /* struct_ops BPF prog can be re-used between multiple
9059                 * .struct_ops as long as it's the same struct_ops struct
9060                 * definition and the same function pointer field
9061                 */
9062                if (prog->attach_btf_id != st_ops->type_id ||
9063                    prog->expected_attach_type != member_idx) {
9064                        pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
9065                                map->name, prog->name, prog->sec_name, prog->type,
9066                                prog->attach_btf_id, prog->expected_attach_type, name);
9067                        return -EINVAL;
9068                }
9069
9070                st_ops->progs[member_idx] = prog;
9071        }
9072
9073        return 0;
9074}
9075
9076#define BTF_TRACE_PREFIX "btf_trace_"
9077#define BTF_LSM_PREFIX "bpf_lsm_"
9078#define BTF_ITER_PREFIX "bpf_iter_"
9079#define BTF_MAX_NAME_SIZE 128
9080
9081void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
9082                                const char **prefix, int *kind)
9083{
9084        switch (attach_type) {
9085        case BPF_TRACE_RAW_TP:
9086                *prefix = BTF_TRACE_PREFIX;
9087                *kind = BTF_KIND_TYPEDEF;
9088                break;
9089        case BPF_LSM_MAC:
9090                *prefix = BTF_LSM_PREFIX;
9091                *kind = BTF_KIND_FUNC;
9092                break;
9093        case BPF_TRACE_ITER:
9094                *prefix = BTF_ITER_PREFIX;
9095                *kind = BTF_KIND_FUNC;
9096                break;
9097        default:
9098                *prefix = "";
9099                *kind = BTF_KIND_FUNC;
9100        }
9101}
9102
9103static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
9104                                   const char *name, __u32 kind)
9105{
9106        char btf_type_name[BTF_MAX_NAME_SIZE];
9107        int ret;
9108
9109        ret = snprintf(btf_type_name, sizeof(btf_type_name),
9110                       "%s%s", prefix, name);
9111        /* snprintf returns the number of characters written excluding the
9112         * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
9113         * indicates truncation.
9114         */
9115        if (ret < 0 || ret >= sizeof(btf_type_name))
9116                return -ENAMETOOLONG;
9117        return btf__find_by_name_kind(btf, btf_type_name, kind);
9118}
9119
9120static inline int find_attach_btf_id(struct btf *btf, const char *name,
9121                                     enum bpf_attach_type attach_type)
9122{
9123        const char *prefix;
9124        int kind;
9125
9126        btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
9127        return find_btf_by_prefix_kind(btf, prefix, name, kind);
9128}
9129
9130int libbpf_find_vmlinux_btf_id(const char *name,
9131                               enum bpf_attach_type attach_type)
9132{
9133        struct btf *btf;
9134        int err;
9135
9136        btf = btf__load_vmlinux_btf();
9137        err = libbpf_get_error(btf);
9138        if (err) {
9139                pr_warn("vmlinux BTF is not found\n");
9140                return libbpf_err(err);
9141        }
9142
9143        err = find_attach_btf_id(btf, name, attach_type);
9144        if (err <= 0)
9145                pr_warn("%s is not found in vmlinux BTF\n", name);
9146
9147        btf__free(btf);
9148        return libbpf_err(err);
9149}
9150
9151static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
9152{
9153        struct bpf_prog_info info = {};
9154        __u32 info_len = sizeof(info);
9155        struct btf *btf;
9156        int err;
9157
9158        err = bpf_obj_get_info_by_fd(attach_prog_fd, &info, &info_len);
9159        if (err) {
9160                pr_warn("failed bpf_obj_get_info_by_fd for FD %d: %d\n",
9161                        attach_prog_fd, err);
9162                return err;
9163        }
9164
9165        err = -EINVAL;
9166        if (!info.btf_id) {
9167                pr_warn("The target program doesn't have BTF\n");
9168                goto out;
9169        }
9170        btf = btf__load_from_kernel_by_id(info.btf_id);
9171        err = libbpf_get_error(btf);
9172        if (err) {
9173                pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err);
9174                goto out;
9175        }
9176        err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9177        btf__free(btf);
9178        if (err <= 0) {
9179                pr_warn("%s is not found in prog's BTF\n", name);
9180                goto out;
9181        }
9182out:
9183        return err;
9184}
9185
9186static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9187                              enum bpf_attach_type attach_type,
9188                              int *btf_obj_fd, int *btf_type_id)
9189{
9190        int ret, i;
9191
9192        ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
9193        if (ret > 0) {
9194                *btf_obj_fd = 0; /* vmlinux BTF */
9195                *btf_type_id = ret;
9196                return 0;
9197        }
9198        if (ret != -ENOENT)
9199                return ret;
9200
9201        ret = load_module_btfs(obj);
9202        if (ret)
9203                return ret;
9204
9205        for (i = 0; i < obj->btf_module_cnt; i++) {
9206                const struct module_btf *mod = &obj->btf_modules[i];
9207
9208                ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
9209                if (ret > 0) {
9210                        *btf_obj_fd = mod->fd;
9211                        *btf_type_id = ret;
9212                        return 0;
9213                }
9214                if (ret == -ENOENT)
9215                        continue;
9216
9217                return ret;
9218        }
9219
9220        return -ESRCH;
9221}
9222
9223static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
9224                                     int *btf_obj_fd, int *btf_type_id)
9225{
9226        enum bpf_attach_type attach_type = prog->expected_attach_type;
9227        __u32 attach_prog_fd = prog->attach_prog_fd;
9228        int err = 0;
9229
9230        /* BPF program's BTF ID */
9231        if (attach_prog_fd) {
9232                err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9233                if (err < 0) {
9234                        pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9235                                 attach_prog_fd, attach_name, err);
9236                        return err;
9237                }
9238                *btf_obj_fd = 0;
9239                *btf_type_id = err;
9240                return 0;
9241        }
9242
9243        /* kernel/module BTF ID */
9244        if (prog->obj->gen_loader) {
9245                bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
9246                *btf_obj_fd = 0;
9247                *btf_type_id = 1;
9248        } else {
9249                err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9250        }
9251        if (err) {
9252                pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err);
9253                return err;
9254        }
9255        return 0;
9256}
9257
9258int libbpf_attach_type_by_name(const char *name,
9259                               enum bpf_attach_type *attach_type)
9260{
9261        char *type_names;
9262        const struct bpf_sec_def *sec_def;
9263
9264        if (!name)
9265                return libbpf_err(-EINVAL);
9266
9267        sec_def = find_sec_def(name);
9268        if (!sec_def) {
9269                pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9270                type_names = libbpf_get_type_names(true);
9271                if (type_names != NULL) {
9272                        pr_debug("attachable section(type) names are:%s\n", type_names);
9273                        free(type_names);
9274                }
9275
9276                return libbpf_err(-EINVAL);
9277        }
9278
9279        if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9280                return libbpf_err(-EINVAL);
9281        if (!(sec_def->cookie & SEC_ATTACHABLE))
9282                return libbpf_err(-EINVAL);
9283
9284        *attach_type = sec_def->expected_attach_type;
9285        return 0;
9286}
9287
9288int bpf_map__fd(const struct bpf_map *map)
9289{
9290        return map ? map->fd : libbpf_err(-EINVAL);
9291}
9292
9293const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
9294{
9295        return map ? &map->def : libbpf_err_ptr(-EINVAL);
9296}
9297
9298static bool map_uses_real_name(const struct bpf_map *map)
9299{
9300        /* Since libbpf started to support custom .data.* and .rodata.* maps,
9301         * their user-visible name differs from kernel-visible name. Users see
9302         * such map's corresponding ELF section name as a map name.
9303         * This check distinguishes .data/.rodata from .data.* and .rodata.*
9304         * maps to know which name has to be returned to the user.
9305         */
9306        if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
9307                return true;
9308        if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
9309                return true;
9310        return false;
9311}
9312
9313const char *bpf_map__name(const struct bpf_map *map)
9314{
9315        if (!map)
9316                return NULL;
9317
9318        if (map_uses_real_name(map))
9319                return map->real_name;
9320
9321        return map->name;
9322}
9323
9324enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9325{
9326        return map->def.type;
9327}
9328
9329int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9330{
9331        if (map->fd >= 0)
9332                return libbpf_err(-EBUSY);
9333        map->def.type = type;
9334        return 0;
9335}
9336
9337__u32 bpf_map__map_flags(const struct bpf_map *map)
9338{
9339        return map->def.map_flags;
9340}
9341
9342int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9343{
9344        if (map->fd >= 0)
9345                return libbpf_err(-EBUSY);
9346        map->def.map_flags = flags;
9347        return 0;
9348}
9349
9350__u64 bpf_map__map_extra(const struct bpf_map *map)
9351{
9352        return map->map_extra;
9353}
9354
9355int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
9356{
9357        if (map->fd >= 0)
9358                return libbpf_err(-EBUSY);
9359        map->map_extra = map_extra;
9360        return 0;
9361}
9362
9363__u32 bpf_map__numa_node(const struct bpf_map *map)
9364{
9365        return map->numa_node;
9366}
9367
9368int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9369{
9370        if (map->fd >= 0)
9371                return libbpf_err(-EBUSY);
9372        map->numa_node = numa_node;
9373        return 0;
9374}
9375
9376__u32 bpf_map__key_size(const struct bpf_map *map)
9377{
9378        return map->def.key_size;
9379}
9380
9381int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9382{
9383        if (map->fd >= 0)
9384                return libbpf_err(-EBUSY);
9385        map->def.key_size = size;
9386        return 0;
9387}
9388
9389__u32 bpf_map__value_size(const struct bpf_map *map)
9390{
9391        return map->def.value_size;
9392}
9393
9394int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9395{
9396        if (map->fd >= 0)
9397                return libbpf_err(-EBUSY);
9398        map->def.value_size = size;
9399        return 0;
9400}
9401
9402__u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9403{
9404        return map ? map->btf_key_type_id : 0;
9405}
9406
9407__u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9408{
9409        return map ? map->btf_value_type_id : 0;
9410}
9411
9412int bpf_map__set_priv(struct bpf_map *map, void *priv,
9413                     bpf_map_clear_priv_t clear_priv)
9414{
9415        if (!map)
9416                return libbpf_err(-EINVAL);
9417
9418        if (map->priv) {
9419                if (map->clear_priv)
9420                        map->clear_priv(map, map->priv);
9421        }
9422
9423        map->priv = priv;
9424        map->clear_priv = clear_priv;
9425        return 0;
9426}
9427
9428void *bpf_map__priv(const struct bpf_map *map)
9429{
9430        return map ? map->priv : libbpf_err_ptr(-EINVAL);
9431}
9432
9433int bpf_map__set_initial_value(struct bpf_map *map,
9434                               const void *data, size_t size)
9435{
9436        if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9437            size != map->def.value_size || map->fd >= 0)
9438                return libbpf_err(-EINVAL);
9439
9440        memcpy(map->mmaped, data, size);
9441        return 0;
9442}
9443
9444const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
9445{
9446        if (!map->mmaped)
9447                return NULL;
9448        *psize = map->def.value_size;
9449        return map->mmaped;
9450}
9451
9452bool bpf_map__is_offload_neutral(const struct bpf_map *map)
9453{
9454        return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
9455}
9456
9457bool bpf_map__is_internal(const struct bpf_map *map)
9458{
9459        return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9460}
9461
9462__u32 bpf_map__ifindex(const struct bpf_map *map)
9463{
9464        return map->map_ifindex;
9465}
9466
9467int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9468{
9469        if (map->fd >= 0)
9470                return libbpf_err(-EBUSY);
9471        map->map_ifindex = ifindex;
9472        return 0;
9473}
9474
9475int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9476{
9477        if (!bpf_map_type__is_map_in_map(map->def.type)) {
9478                pr_warn("error: unsupported map type\n");
9479                return libbpf_err(-EINVAL);
9480        }
9481        if (map->inner_map_fd != -1) {
9482                pr_warn("error: inner_map_fd already specified\n");
9483                return libbpf_err(-EINVAL);
9484        }
9485        if (map->inner_map) {
9486                bpf_map__destroy(map->inner_map);
9487                zfree(&map->inner_map);
9488        }
9489        map->inner_map_fd = fd;
9490        return 0;
9491}
9492
9493static struct bpf_map *
9494__bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9495{
9496        ssize_t idx;
9497        struct bpf_map *s, *e;
9498
9499        if (!obj || !obj->maps)
9500                return errno = EINVAL, NULL;
9501
9502        s = obj->maps;
9503        e = obj->maps + obj->nr_maps;
9504
9505        if ((m < s) || (m >= e)) {
9506                pr_warn("error in %s: map handler doesn't belong to object\n",
9507                         __func__);
9508                return errno = EINVAL, NULL;
9509        }
9510
9511        idx = (m - obj->maps) + i;
9512        if (idx >= obj->nr_maps || idx < 0)
9513                return NULL;
9514        return &obj->maps[idx];
9515}
9516
9517struct bpf_map *
9518bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
9519{
9520        return bpf_object__next_map(obj, prev);
9521}
9522
9523struct bpf_map *
9524bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
9525{
9526        if (prev == NULL)
9527                return obj->maps;
9528
9529        return __bpf_map__iter(prev, obj, 1);
9530}
9531
9532struct bpf_map *
9533bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
9534{
9535        return bpf_object__prev_map(obj, next);
9536}
9537
9538struct bpf_map *
9539bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
9540{
9541        if (next == NULL) {
9542                if (!obj->nr_maps)
9543                        return NULL;
9544                return obj->maps + obj->nr_maps - 1;
9545        }
9546
9547        return __bpf_map__iter(next, obj, -1);
9548}
9549
9550struct bpf_map *
9551bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9552{
9553        struct bpf_map *pos;
9554
9555        bpf_object__for_each_map(pos, obj) {
9556                /* if it's a special internal map name (which always starts
9557                 * with dot) then check if that special name matches the
9558                 * real map name (ELF section name)
9559                 */
9560                if (name[0] == '.') {
9561                        if (pos->real_name && strcmp(pos->real_name, name) == 0)
9562                                return pos;
9563                        continue;
9564                }
9565                /* otherwise map name has to be an exact match */
9566                if (map_uses_real_name(pos)) {
9567                        if (strcmp(pos->real_name, name) == 0)
9568                                return pos;
9569                        continue;
9570                }
9571                if (strcmp(pos->name, name) == 0)
9572                        return pos;
9573        }
9574        return errno = ENOENT, NULL;
9575}
9576
9577int
9578bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9579{
9580        return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9581}
9582
9583struct bpf_map *
9584bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
9585{
9586        return libbpf_err_ptr(-ENOTSUP);
9587}
9588
9589long libbpf_get_error(const void *ptr)
9590{
9591        if (!IS_ERR_OR_NULL(ptr))
9592                return 0;
9593
9594        if (IS_ERR(ptr))
9595                errno = -PTR_ERR(ptr);
9596
9597        /* If ptr == NULL, then errno should be already set by the failing
9598         * API, because libbpf never returns NULL on success and it now always
9599         * sets errno on error. So no extra errno handling for ptr == NULL
9600         * case.
9601         */
9602        return -errno;
9603}
9604
9605__attribute__((alias("bpf_prog_load_xattr2")))
9606int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
9607                        struct bpf_object **pobj, int *prog_fd);
9608
9609static int bpf_prog_load_xattr2(const struct bpf_prog_load_attr *attr,
9610                                struct bpf_object **pobj, int *prog_fd)
9611{
9612        struct bpf_object_open_attr open_attr = {};
9613        struct bpf_program *prog, *first_prog = NULL;
9614        struct bpf_object *obj;
9615        struct bpf_map *map;
9616        int err;
9617
9618        if (!attr)
9619                return libbpf_err(-EINVAL);
9620        if (!attr->file)
9621                return libbpf_err(-EINVAL);
9622
9623        open_attr.file = attr->file;
9624        open_attr.prog_type = attr->prog_type;
9625
9626        obj = __bpf_object__open_xattr(&open_attr, 0);
9627        err = libbpf_get_error(obj);
9628        if (err)
9629                return libbpf_err(-ENOENT);
9630
9631        bpf_object__for_each_program(prog, obj) {
9632                enum bpf_attach_type attach_type = attr->expected_attach_type;
9633                /*
9634                 * to preserve backwards compatibility, bpf_prog_load treats
9635                 * attr->prog_type, if specified, as an override to whatever
9636                 * bpf_object__open guessed
9637                 */
9638                if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
9639                        bpf_program__set_type(prog, attr->prog_type);
9640                        bpf_program__set_expected_attach_type(prog,
9641                                                              attach_type);
9642                }
9643                if (bpf_program__type(prog) == BPF_PROG_TYPE_UNSPEC) {
9644                        /*
9645                         * we haven't guessed from section name and user
9646                         * didn't provide a fallback type, too bad...
9647                         */
9648                        bpf_object__close(obj);
9649                        return libbpf_err(-EINVAL);
9650                }
9651
9652                prog->prog_ifindex = attr->ifindex;
9653                prog->log_level = attr->log_level;
9654                prog->prog_flags |= attr->prog_flags;
9655                if (!first_prog)
9656                        first_prog = prog;
9657        }
9658
9659        bpf_object__for_each_map(map, obj) {
9660                if (map->def.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
9661                        map->map_ifindex = attr->ifindex;
9662        }
9663
9664        if (!first_prog) {
9665                pr_warn("object file doesn't contain bpf program\n");
9666                bpf_object__close(obj);
9667                return libbpf_err(-ENOENT);
9668        }
9669
9670        err = bpf_object__load(obj);
9671        if (err) {
9672                bpf_object__close(obj);
9673                return libbpf_err(err);
9674        }
9675
9676        *pobj = obj;
9677        *prog_fd = bpf_program__fd(first_prog);
9678        return 0;
9679}
9680
9681COMPAT_VERSION(bpf_prog_load_deprecated, bpf_prog_load, LIBBPF_0.0.1)
9682int bpf_prog_load_deprecated(const char *file, enum bpf_prog_type type,
9683                             struct bpf_object **pobj, int *prog_fd)
9684{
9685        struct bpf_prog_load_attr attr;
9686
9687        memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
9688        attr.file = file;
9689        attr.prog_type = type;
9690        attr.expected_attach_type = 0;
9691
9692        return bpf_prog_load_xattr2(&attr, pobj, prog_fd);
9693}
9694
9695struct bpf_link {
9696        int (*detach)(struct bpf_link *link);
9697        void (*dealloc)(struct bpf_link *link);
9698        char *pin_path;         /* NULL, if not pinned */
9699        int fd;                 /* hook FD, -1 if not applicable */
9700        bool disconnected;
9701};
9702
9703/* Replace link's underlying BPF program with the new one */
9704int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
9705{
9706        int ret;
9707
9708        ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
9709        return libbpf_err_errno(ret);
9710}
9711
9712/* Release "ownership" of underlying BPF resource (typically, BPF program
9713 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
9714 * link, when destructed through bpf_link__destroy() call won't attempt to
9715 * detach/unregisted that BPF resource. This is useful in situations where,
9716 * say, attached BPF program has to outlive userspace program that attached it
9717 * in the system. Depending on type of BPF program, though, there might be
9718 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
9719 * exit of userspace program doesn't trigger automatic detachment and clean up
9720 * inside the kernel.
9721 */
9722void bpf_link__disconnect(struct bpf_link *link)
9723{
9724        link->disconnected = true;
9725}
9726
9727int bpf_link__destroy(struct bpf_link *link)
9728{
9729        int err = 0;
9730
9731        if (IS_ERR_OR_NULL(link))
9732                return 0;
9733
9734        if (!link->disconnected && link->detach)
9735                err = link->detach(link);
9736        if (link->pin_path)
9737                free(link->pin_path);
9738        if (link->dealloc)
9739                link->dealloc(link);
9740        else
9741                free(link);
9742
9743        return libbpf_err(err);
9744}
9745
9746int bpf_link__fd(const struct bpf_link *link)
9747{
9748        return link->fd;
9749}
9750
9751const char *bpf_link__pin_path(const struct bpf_link *link)
9752{
9753        return link->pin_path;
9754}
9755
9756static int bpf_link__detach_fd(struct bpf_link *link)
9757{
9758        return libbpf_err_errno(close(link->fd));
9759}
9760
9761struct bpf_link *bpf_link__open(const char *path)
9762{
9763        struct bpf_link *link;
9764        int fd;
9765
9766        fd = bpf_obj_get(path);
9767        if (fd < 0) {
9768                fd = -errno;
9769                pr_warn("failed to open link at %s: %d\n", path, fd);
9770                return libbpf_err_ptr(fd);
9771        }
9772
9773        link = calloc(1, sizeof(*link));
9774        if (!link) {
9775                close(fd);
9776                return libbpf_err_ptr(-ENOMEM);
9777        }
9778        link->detach = &bpf_link__detach_fd;
9779        link->fd = fd;
9780
9781        link->pin_path = strdup(path);
9782        if (!link->pin_path) {
9783                bpf_link__destroy(link);
9784                return libbpf_err_ptr(-ENOMEM);
9785        }
9786
9787        return link;
9788}
9789
9790int bpf_link__detach(struct bpf_link *link)
9791{
9792        return bpf_link_detach(link->fd) ? -errno : 0;
9793}
9794
9795int bpf_link__pin(struct bpf_link *link, const char *path)
9796{
9797        int err;
9798
9799        if (link->pin_path)
9800                return libbpf_err(-EBUSY);
9801        err = make_parent_dir(path);
9802        if (err)
9803                return libbpf_err(err);
9804        err = check_path(path);
9805        if (err)
9806                return libbpf_err(err);
9807
9808        link->pin_path = strdup(path);
9809        if (!link->pin_path)
9810                return libbpf_err(-ENOMEM);
9811
9812        if (bpf_obj_pin(link->fd, link->pin_path)) {
9813                err = -errno;
9814                zfree(&link->pin_path);
9815                return libbpf_err(err);
9816        }
9817
9818        pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
9819        return 0;
9820}
9821
9822int bpf_link__unpin(struct bpf_link *link)
9823{
9824        int err;
9825
9826        if (!link->pin_path)
9827                return libbpf_err(-EINVAL);
9828
9829        err = unlink(link->pin_path);
9830        if (err != 0)
9831                return -errno;
9832
9833        pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
9834        zfree(&link->pin_path);
9835        return 0;
9836}
9837
9838struct bpf_link_perf {
9839        struct bpf_link link;
9840        int perf_event_fd;
9841        /* legacy kprobe support: keep track of probe identifier and type */
9842        char *legacy_probe_name;
9843        bool legacy_is_kprobe;
9844        bool legacy_is_retprobe;
9845};
9846
9847static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
9848static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
9849
9850static int bpf_link_perf_detach(struct bpf_link *link)
9851{
9852        struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9853        int err = 0;
9854
9855        if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
9856                err = -errno;
9857
9858        if (perf_link->perf_event_fd != link->fd)
9859                close(perf_link->perf_event_fd);
9860        close(link->fd);
9861
9862        /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
9863        if (perf_link->legacy_probe_name) {
9864                if (perf_link->legacy_is_kprobe) {
9865                        err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
9866                                                         perf_link->legacy_is_retprobe);
9867                } else {
9868                        err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
9869                                                         perf_link->legacy_is_retprobe);
9870                }
9871        }
9872
9873        return err;
9874}
9875
9876static void bpf_link_perf_dealloc(struct bpf_link *link)
9877{
9878        struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9879
9880        free(perf_link->legacy_probe_name);
9881        free(perf_link);
9882}
9883
9884struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
9885                                                     const struct bpf_perf_event_opts *opts)
9886{
9887        char errmsg[STRERR_BUFSIZE];
9888        struct bpf_link_perf *link;
9889        int prog_fd, link_fd = -1, err;
9890
9891        if (!OPTS_VALID(opts, bpf_perf_event_opts))
9892                return libbpf_err_ptr(-EINVAL);
9893
9894        if (pfd < 0) {
9895                pr_warn("prog '%s': invalid perf event FD %d\n",
9896                        prog->name, pfd);
9897                return libbpf_err_ptr(-EINVAL);
9898        }
9899        prog_fd = bpf_program__fd(prog);
9900        if (prog_fd < 0) {
9901                pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
9902                        prog->name);
9903                return libbpf_err_ptr(-EINVAL);
9904        }
9905
9906        link = calloc(1, sizeof(*link));
9907        if (!link)
9908                return libbpf_err_ptr(-ENOMEM);
9909        link->link.detach = &bpf_link_perf_detach;
9910        link->link.dealloc = &bpf_link_perf_dealloc;
9911        link->perf_event_fd = pfd;
9912
9913        if (kernel_supports(prog->obj, FEAT_PERF_LINK)) {
9914                DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
9915                        .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
9916
9917                link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
9918                if (link_fd < 0) {
9919                        err = -errno;
9920                        pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
9921                                prog->name, pfd,
9922                                err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9923                        goto err_out;
9924                }
9925                link->link.fd = link_fd;
9926        } else {
9927                if (OPTS_GET(opts, bpf_cookie, 0)) {
9928                        pr_warn("prog '%s': user context value is not supported\n", prog->name);
9929                        err = -EOPNOTSUPP;
9930                        goto err_out;
9931                }
9932
9933                if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
9934                        err = -errno;
9935                        pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
9936                                prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9937                        if (err == -EPROTO)
9938                                pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
9939                                        prog->name, pfd);
9940                        goto err_out;
9941                }
9942                link->link.fd = pfd;
9943        }
9944        if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9945                err = -errno;
9946                pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
9947                        prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9948                goto err_out;
9949        }
9950
9951        return &link->link;
9952err_out:
9953        if (link_fd >= 0)
9954                close(link_fd);
9955        free(link);
9956        return libbpf_err_ptr(err);
9957}
9958
9959struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
9960{
9961        return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
9962}
9963
9964/*
9965 * this function is expected to parse integer in the range of [0, 2^31-1] from
9966 * given file using scanf format string fmt. If actual parsed value is
9967 * negative, the result might be indistinguishable from error
9968 */
9969static int parse_uint_from_file(const char *file, const char *fmt)
9970{
9971        char buf[STRERR_BUFSIZE];
9972        int err, ret;
9973        FILE *f;
9974
9975        f = fopen(file, "r");
9976        if (!f) {
9977                err = -errno;
9978                pr_debug("failed to open '%s': %s\n", file,
9979                         libbpf_strerror_r(err, buf, sizeof(buf)));
9980                return err;
9981        }
9982        err = fscanf(f, fmt, &ret);
9983        if (err != 1) {
9984                err = err == EOF ? -EIO : -errno;
9985                pr_debug("failed to parse '%s': %s\n", file,
9986                        libbpf_strerror_r(err, buf, sizeof(buf)));
9987                fclose(f);
9988                return err;
9989        }
9990        fclose(f);
9991        return ret;
9992}
9993
9994static int determine_kprobe_perf_type(void)
9995{
9996        const char *file = "/sys/bus/event_source/devices/kprobe/type";
9997
9998        return parse_uint_from_file(file, "%d\n");
9999}
10000
10001static int determine_uprobe_perf_type(void)
10002{
10003        const char *file = "/sys/bus/event_source/devices/uprobe/type";
10004
10005        return parse_uint_from_file(file, "%d\n");
10006}
10007
10008static int determine_kprobe_retprobe_bit(void)
10009{
10010        const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
10011
10012        return parse_uint_from_file(file, "config:%d\n");
10013}
10014
10015static int determine_uprobe_retprobe_bit(void)
10016{
10017        const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
10018
10019        return parse_uint_from_file(file, "config:%d\n");
10020}
10021
10022#define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
10023#define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
10024
10025static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
10026                                 uint64_t offset, int pid, size_t ref_ctr_off)
10027{
10028        struct perf_event_attr attr = {};
10029        char errmsg[STRERR_BUFSIZE];
10030        int type, pfd, err;
10031
10032        if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
10033                return -EINVAL;
10034
10035        type = uprobe ? determine_uprobe_perf_type()
10036                      : determine_kprobe_perf_type();
10037        if (type < 0) {
10038                pr_warn("failed to determine %s perf type: %s\n",
10039                        uprobe ? "uprobe" : "kprobe",
10040                        libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
10041                return type;
10042        }
10043        if (retprobe) {
10044                int bit = uprobe ? determine_uprobe_retprobe_bit()
10045                                 : determine_kprobe_retprobe_bit();
10046
10047                if (bit < 0) {
10048                        pr_warn("failed to determine %s retprobe bit: %s\n",
10049                                uprobe ? "uprobe" : "kprobe",
10050                                libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
10051                        return bit;
10052                }
10053                attr.config |= 1 << bit;
10054        }
10055        attr.size = sizeof(attr);
10056        attr.type = type;
10057        attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
10058        attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
10059        attr.config2 = offset;           /* kprobe_addr or probe_offset */
10060
10061        /* pid filter is meaningful only for uprobes */
10062        pfd = syscall(__NR_perf_event_open, &attr,
10063                      pid < 0 ? -1 : pid /* pid */,
10064                      pid == -1 ? 0 : -1 /* cpu */,
10065                      -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10066        if (pfd < 0) {
10067                err = -errno;
10068                pr_warn("%s perf_event_open() failed: %s\n",
10069                        uprobe ? "uprobe" : "kprobe",
10070                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10071                return err;
10072        }
10073        return pfd;
10074}
10075
10076static int append_to_file(const char *file, const char *fmt, ...)
10077{
10078        int fd, n, err = 0;
10079        va_list ap;
10080
10081        fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
10082        if (fd < 0)
10083                return -errno;
10084
10085        va_start(ap, fmt);
10086        n = vdprintf(fd, fmt, ap);
10087        va_end(ap);
10088
10089        if (n < 0)
10090                err = -errno;
10091
10092        close(fd);
10093        return err;
10094}
10095
10096static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
10097                                         const char *kfunc_name, size_t offset)
10098{
10099        static int index = 0;
10100
10101        snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
10102                 __sync_fetch_and_add(&index, 1));
10103}
10104
10105static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
10106                                   const char *kfunc_name, size_t offset)
10107{
10108        const char *file = "/sys/kernel/debug/tracing/kprobe_events";
10109
10110        return append_to_file(file, "%c:%s/%s %s+0x%zx",
10111                              retprobe ? 'r' : 'p',
10112                              retprobe ? "kretprobes" : "kprobes",
10113                              probe_name, kfunc_name, offset);
10114}
10115
10116static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
10117{
10118        const char *file = "/sys/kernel/debug/tracing/kprobe_events";
10119
10120        return append_to_file(file, "-:%s/%s", retprobe ? "kretprobes" : "kprobes", probe_name);
10121}
10122
10123static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10124{
10125        char file[256];
10126
10127        snprintf(file, sizeof(file),
10128                 "/sys/kernel/debug/tracing/events/%s/%s/id",
10129                 retprobe ? "kretprobes" : "kprobes", probe_name);
10130
10131        return parse_uint_from_file(file, "%d\n");
10132}
10133
10134static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
10135                                         const char *kfunc_name, size_t offset, int pid)
10136{
10137        struct perf_event_attr attr = {};
10138        char errmsg[STRERR_BUFSIZE];
10139        int type, pfd, err;
10140
10141        err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
10142        if (err < 0) {
10143                pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
10144                        kfunc_name, offset,
10145                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10146                return err;
10147        }
10148        type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
10149        if (type < 0) {
10150                pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
10151                        kfunc_name, offset,
10152                        libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
10153                return type;
10154        }
10155        attr.size = sizeof(attr);
10156        attr.config = type;
10157        attr.type = PERF_TYPE_TRACEPOINT;
10158
10159        pfd = syscall(__NR_perf_event_open, &attr,
10160                      pid < 0 ? -1 : pid, /* pid */
10161                      pid == -1 ? 0 : -1, /* cpu */
10162                      -1 /* group_fd */,  PERF_FLAG_FD_CLOEXEC);
10163        if (pfd < 0) {
10164                err = -errno;
10165                pr_warn("legacy kprobe perf_event_open() failed: %s\n",
10166                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10167                return err;
10168        }
10169        return pfd;
10170}
10171
10172struct bpf_link *
10173bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
10174                                const char *func_name,
10175                                const struct bpf_kprobe_opts *opts)
10176{
10177        DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10178        char errmsg[STRERR_BUFSIZE];
10179        char *legacy_probe = NULL;
10180        struct bpf_link *link;
10181        size_t offset;
10182        bool retprobe, legacy;
10183        int pfd, err;
10184
10185        if (!OPTS_VALID(opts, bpf_kprobe_opts))
10186                return libbpf_err_ptr(-EINVAL);
10187
10188        retprobe = OPTS_GET(opts, retprobe, false);
10189        offset = OPTS_GET(opts, offset, 0);
10190        pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10191
10192        legacy = determine_kprobe_perf_type() < 0;
10193        if (!legacy) {
10194                pfd = perf_event_open_probe(false /* uprobe */, retprobe,
10195                                            func_name, offset,
10196                                            -1 /* pid */, 0 /* ref_ctr_off */);
10197        } else {
10198                char probe_name[256];
10199
10200                gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
10201                                             func_name, offset);
10202
10203                legacy_probe = strdup(probe_name);
10204                if (!legacy_probe)
10205                        return libbpf_err_ptr(-ENOMEM);
10206
10207                pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
10208                                                    offset, -1 /* pid */);
10209        }
10210        if (pfd < 0) {
10211                err = -errno;
10212                pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
10213                        prog->name, retprobe ? "kretprobe" : "kprobe",
10214                        func_name, offset,
10215                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10216                goto err_out;
10217        }
10218        link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10219        err = libbpf_get_error(link);
10220        if (err) {
10221                close(pfd);
10222                pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
10223                        prog->name, retprobe ? "kretprobe" : "kprobe",
10224                        func_name, offset,
10225                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10226                goto err_out;
10227        }
10228        if (legacy) {
10229                struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10230
10231                perf_link->legacy_probe_name = legacy_probe;
10232                perf_link->legacy_is_kprobe = true;
10233                perf_link->legacy_is_retprobe = retprobe;
10234        }
10235
10236        return link;
10237err_out:
10238        free(legacy_probe);
10239        return libbpf_err_ptr(err);
10240}
10241
10242struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
10243                                            bool retprobe,
10244                                            const char *func_name)
10245{
10246        DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
10247                .retprobe = retprobe,
10248        );
10249
10250        return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
10251}
10252
10253/* Adapted from perf/util/string.c */
10254static bool glob_match(const char *str, const char *pat)
10255{
10256        while (*str && *pat && *pat != '*') {
10257                if (*pat == '?') {      /* Matches any single character */
10258                        str++;
10259                        pat++;
10260                        continue;
10261                }
10262                if (*str != *pat)
10263                        return false;
10264                str++;
10265                pat++;
10266        }
10267        /* Check wild card */
10268        if (*pat == '*') {
10269                while (*pat == '*')
10270                        pat++;
10271                if (!*pat) /* Tail wild card matches all */
10272                        return true;
10273                while (*str)
10274                        if (glob_match(str++, pat))
10275                                return true;
10276        }
10277        return !*str && !*pat;
10278}
10279
10280struct kprobe_multi_resolve {
10281        const char *pattern;
10282        unsigned long *addrs;
10283        size_t cap;
10284        size_t cnt;
10285};
10286
10287static int
10288resolve_kprobe_multi_cb(unsigned long long sym_addr, char sym_type,
10289                        const char *sym_name, void *ctx)
10290{
10291        struct kprobe_multi_resolve *res = ctx;
10292        int err;
10293
10294        if (!glob_match(sym_name, res->pattern))
10295                return 0;
10296
10297        err = libbpf_ensure_mem((void **) &res->addrs, &res->cap, sizeof(unsigned long),
10298                                res->cnt + 1);
10299        if (err)
10300                return err;
10301
10302        res->addrs[res->cnt++] = (unsigned long) sym_addr;
10303        return 0;
10304}
10305
10306struct bpf_link *
10307bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
10308                                      const char *pattern,
10309                                      const struct bpf_kprobe_multi_opts *opts)
10310{
10311        LIBBPF_OPTS(bpf_link_create_opts, lopts);
10312        struct kprobe_multi_resolve res = {
10313                .pattern = pattern,
10314        };
10315        struct bpf_link *link = NULL;
10316        char errmsg[STRERR_BUFSIZE];
10317        const unsigned long *addrs;
10318        int err, link_fd, prog_fd;
10319        const __u64 *cookies;
10320        const char **syms;
10321        bool retprobe;
10322        size_t cnt;
10323
10324        if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
10325                return libbpf_err_ptr(-EINVAL);
10326
10327        syms    = OPTS_GET(opts, syms, false);
10328        addrs   = OPTS_GET(opts, addrs, false);
10329        cnt     = OPTS_GET(opts, cnt, false);
10330        cookies = OPTS_GET(opts, cookies, false);
10331
10332        if (!pattern && !addrs && !syms)
10333                return libbpf_err_ptr(-EINVAL);
10334        if (pattern && (addrs || syms || cookies || cnt))
10335                return libbpf_err_ptr(-EINVAL);
10336        if (!pattern && !cnt)
10337                return libbpf_err_ptr(-EINVAL);
10338        if (addrs && syms)
10339                return libbpf_err_ptr(-EINVAL);
10340
10341        if (pattern) {
10342                err = libbpf_kallsyms_parse(resolve_kprobe_multi_cb, &res);
10343                if (err)
10344                        goto error;
10345                if (!res.cnt) {
10346                        err = -ENOENT;
10347                        goto error;
10348                }
10349                addrs = res.addrs;
10350                cnt = res.cnt;
10351        }
10352
10353        retprobe = OPTS_GET(opts, retprobe, false);
10354
10355        lopts.kprobe_multi.syms = syms;
10356        lopts.kprobe_multi.addrs = addrs;
10357        lopts.kprobe_multi.cookies = cookies;
10358        lopts.kprobe_multi.cnt = cnt;
10359        lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
10360
10361        link = calloc(1, sizeof(*link));
10362        if (!link) {
10363                err = -ENOMEM;
10364                goto error;
10365        }
10366        link->detach = &bpf_link__detach_fd;
10367
10368        prog_fd = bpf_program__fd(prog);
10369        link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts);
10370        if (link_fd < 0) {
10371                err = -errno;
10372                pr_warn("prog '%s': failed to attach: %s\n",
10373                        prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10374                goto error;
10375        }
10376        link->fd = link_fd;
10377        free(res.addrs);
10378        return link;
10379
10380error:
10381        free(link);
10382        free(res.addrs);
10383        return libbpf_err_ptr(err);
10384}
10385
10386static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10387{
10388        DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
10389        unsigned long offset = 0;
10390        const char *func_name;
10391        char *func;
10392        int n;
10393
10394        opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
10395        if (opts.retprobe)
10396                func_name = prog->sec_name + sizeof("kretprobe/") - 1;
10397        else
10398                func_name = prog->sec_name + sizeof("kprobe/") - 1;
10399
10400        n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
10401        if (n < 1) {
10402                pr_warn("kprobe name is invalid: %s\n", func_name);
10403                return -EINVAL;
10404        }
10405        if (opts.retprobe && offset != 0) {
10406                free(func);
10407                pr_warn("kretprobes do not support offset specification\n");
10408                return -EINVAL;
10409        }
10410
10411        opts.offset = offset;
10412        *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
10413        free(func);
10414        return libbpf_get_error(*link);
10415}
10416
10417static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10418{
10419        LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
10420        const char *spec;
10421        char *pattern;
10422        int n;
10423
10424        opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
10425        if (opts.retprobe)
10426                spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
10427        else
10428                spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
10429
10430        n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
10431        if (n < 1) {
10432                pr_warn("kprobe multi pattern is invalid: %s\n", pattern);
10433                return -EINVAL;
10434        }
10435
10436        *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
10437        free(pattern);
10438        return libbpf_get_error(*link);
10439}
10440
10441static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
10442                                         const char *binary_path, uint64_t offset)
10443{
10444        int i;
10445
10446        snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
10447
10448        /* sanitize binary_path in the probe name */
10449        for (i = 0; buf[i]; i++) {
10450                if (!isalnum(buf[i]))
10451                        buf[i] = '_';
10452        }
10453}
10454
10455static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
10456                                          const char *binary_path, size_t offset)
10457{
10458        const char *file = "/sys/kernel/debug/tracing/uprobe_events";
10459
10460        return append_to_file(file, "%c:%s/%s %s:0x%zx",
10461                              retprobe ? 'r' : 'p',
10462                              retprobe ? "uretprobes" : "uprobes",
10463                              probe_name, binary_path, offset);
10464}
10465
10466static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
10467{
10468        const char *file = "/sys/kernel/debug/tracing/uprobe_events";
10469
10470        return append_to_file(file, "-:%s/%s", retprobe ? "uretprobes" : "uprobes", probe_name);
10471}
10472
10473static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10474{
10475        char file[512];
10476
10477        snprintf(file, sizeof(file),
10478                 "/sys/kernel/debug/tracing/events/%s/%s/id",
10479                 retprobe ? "uretprobes" : "uprobes", probe_name);
10480
10481        return parse_uint_from_file(file, "%d\n");
10482}
10483
10484static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
10485                                         const char *binary_path, size_t offset, int pid)
10486{
10487        struct perf_event_attr attr;
10488        int type, pfd, err;
10489
10490        err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
10491        if (err < 0) {
10492                pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n",
10493                        binary_path, (size_t)offset, err);
10494                return err;
10495        }
10496        type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
10497        if (type < 0) {
10498                pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n",
10499                        binary_path, offset, err);
10500                return type;
10501        }
10502
10503        memset(&attr, 0, sizeof(attr));
10504        attr.size = sizeof(attr);
10505        attr.config = type;
10506        attr.type = PERF_TYPE_TRACEPOINT;
10507
10508        pfd = syscall(__NR_perf_event_open, &attr,
10509                      pid < 0 ? -1 : pid, /* pid */
10510                      pid == -1 ? 0 : -1, /* cpu */
10511                      -1 /* group_fd */,  PERF_FLAG_FD_CLOEXEC);
10512        if (pfd < 0) {
10513                err = -errno;
10514                pr_warn("legacy uprobe perf_event_open() failed: %d\n", err);
10515                return err;
10516        }
10517        return pfd;
10518}
10519
10520LIBBPF_API struct bpf_link *
10521bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
10522                                const char *binary_path, size_t func_offset,
10523                                const struct bpf_uprobe_opts *opts)
10524{
10525        DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10526        char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL;
10527        struct bpf_link *link;
10528        size_t ref_ctr_off;
10529        int pfd, err;
10530        bool retprobe, legacy;
10531
10532        if (!OPTS_VALID(opts, bpf_uprobe_opts))
10533                return libbpf_err_ptr(-EINVAL);
10534
10535        retprobe = OPTS_GET(opts, retprobe, false);
10536        ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
10537        pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10538
10539        legacy = determine_uprobe_perf_type() < 0;
10540        if (!legacy) {
10541                pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
10542                                            func_offset, pid, ref_ctr_off);
10543        } else {
10544                char probe_name[512];
10545
10546                if (ref_ctr_off)
10547                        return libbpf_err_ptr(-EINVAL);
10548
10549                gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
10550                                             binary_path, func_offset);
10551
10552                legacy_probe = strdup(probe_name);
10553                if (!legacy_probe)
10554                        return libbpf_err_ptr(-ENOMEM);
10555
10556                pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
10557                                                    binary_path, func_offset, pid);
10558        }
10559        if (pfd < 0) {
10560                err = -errno;
10561                pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
10562                        prog->name, retprobe ? "uretprobe" : "uprobe",
10563                        binary_path, func_offset,
10564                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10565                goto err_out;
10566        }
10567
10568        link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10569        err = libbpf_get_error(link);
10570        if (err) {
10571                close(pfd);
10572                pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
10573                        prog->name, retprobe ? "uretprobe" : "uprobe",
10574                        binary_path, func_offset,
10575                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10576                goto err_out;
10577        }
10578        if (legacy) {
10579                struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10580
10581                perf_link->legacy_probe_name = legacy_probe;
10582                perf_link->legacy_is_kprobe = false;
10583                perf_link->legacy_is_retprobe = retprobe;
10584        }
10585        return link;
10586err_out:
10587        free(legacy_probe);
10588        return libbpf_err_ptr(err);
10589
10590}
10591
10592struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
10593                                            bool retprobe, pid_t pid,
10594                                            const char *binary_path,
10595                                            size_t func_offset)
10596{
10597        DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
10598
10599        return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
10600}
10601
10602static int determine_tracepoint_id(const char *tp_category,
10603                                   const char *tp_name)
10604{
10605        char file[PATH_MAX];
10606        int ret;
10607
10608        ret = snprintf(file, sizeof(file),
10609                       "/sys/kernel/debug/tracing/events/%s/%s/id",
10610                       tp_category, tp_name);
10611        if (ret < 0)
10612                return -errno;
10613        if (ret >= sizeof(file)) {
10614                pr_debug("tracepoint %s/%s path is too long\n",
10615                         tp_category, tp_name);
10616                return -E2BIG;
10617        }
10618        return parse_uint_from_file(file, "%d\n");
10619}
10620
10621static int perf_event_open_tracepoint(const char *tp_category,
10622                                      const char *tp_name)
10623{
10624        struct perf_event_attr attr = {};
10625        char errmsg[STRERR_BUFSIZE];
10626        int tp_id, pfd, err;
10627
10628        tp_id = determine_tracepoint_id(tp_category, tp_name);
10629        if (tp_id < 0) {
10630                pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
10631                        tp_category, tp_name,
10632                        libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
10633                return tp_id;
10634        }
10635
10636        attr.type = PERF_TYPE_TRACEPOINT;
10637        attr.size = sizeof(attr);
10638        attr.config = tp_id;
10639
10640        pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
10641                      -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10642        if (pfd < 0) {
10643                err = -errno;
10644                pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
10645                        tp_category, tp_name,
10646                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10647                return err;
10648        }
10649        return pfd;
10650}
10651
10652struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
10653                                                     const char *tp_category,
10654                                                     const char *tp_name,
10655                                                     const struct bpf_tracepoint_opts *opts)
10656{
10657        DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10658        char errmsg[STRERR_BUFSIZE];
10659        struct bpf_link *link;
10660        int pfd, err;
10661
10662        if (!OPTS_VALID(opts, bpf_tracepoint_opts))
10663                return libbpf_err_ptr(-EINVAL);
10664
10665        pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10666
10667        pfd = perf_event_open_tracepoint(tp_category, tp_name);
10668        if (pfd < 0) {
10669                pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
10670                        prog->name, tp_category, tp_name,
10671                        libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10672                return libbpf_err_ptr(pfd);
10673        }
10674        link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10675        err = libbpf_get_error(link);
10676        if (err) {
10677                close(pfd);
10678                pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
10679                        prog->name, tp_category, tp_name,
10680                        libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10681                return libbpf_err_ptr(err);
10682        }
10683        return link;
10684}
10685
10686struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
10687                                                const char *tp_category,
10688                                                const char *tp_name)
10689{
10690        return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
10691}
10692
10693static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10694{
10695        char *sec_name, *tp_cat, *tp_name;
10696
10697        sec_name = strdup(prog->sec_name);
10698        if (!sec_name)
10699                return -ENOMEM;
10700
10701        /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
10702        if (str_has_pfx(prog->sec_name, "tp/"))
10703                tp_cat = sec_name + sizeof("tp/") - 1;
10704        else
10705                tp_cat = sec_name + sizeof("tracepoint/") - 1;
10706        tp_name = strchr(tp_cat, '/');
10707        if (!tp_name) {
10708                free(sec_name);
10709                return -EINVAL;
10710        }
10711        *tp_name = '\0';
10712        tp_name++;
10713
10714        *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
10715        free(sec_name);
10716        return libbpf_get_error(*link);
10717}
10718
10719struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
10720                                                    const char *tp_name)
10721{
10722        char errmsg[STRERR_BUFSIZE];
10723        struct bpf_link *link;
10724        int prog_fd, pfd;
10725
10726        prog_fd = bpf_program__fd(prog);
10727        if (prog_fd < 0) {
10728                pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10729                return libbpf_err_ptr(-EINVAL);
10730        }
10731
10732        link = calloc(1, sizeof(*link));
10733        if (!link)
10734                return libbpf_err_ptr(-ENOMEM);
10735        link->detach = &bpf_link__detach_fd;
10736
10737        pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
10738        if (pfd < 0) {
10739                pfd = -errno;
10740                free(link);
10741                pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
10742                        prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10743                return libbpf_err_ptr(pfd);
10744        }
10745        link->fd = pfd;
10746        return link;
10747}
10748
10749static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10750{
10751        static const char *const prefixes[] = {
10752                "raw_tp/",
10753                "raw_tracepoint/",
10754                "raw_tp.w/",
10755                "raw_tracepoint.w/",
10756        };
10757        size_t i;
10758        const char *tp_name = NULL;
10759
10760        for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
10761                if (str_has_pfx(prog->sec_name, prefixes[i])) {
10762                        tp_name = prog->sec_name + strlen(prefixes[i]);
10763                        break;
10764                }
10765        }
10766        if (!tp_name) {
10767                pr_warn("prog '%s': invalid section name '%s'\n",
10768                        prog->name, prog->sec_name);
10769                return -EINVAL;
10770        }
10771
10772        *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
10773        return libbpf_get_error(link);
10774}
10775
10776/* Common logic for all BPF program types that attach to a btf_id */
10777static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog)
10778{
10779        char errmsg[STRERR_BUFSIZE];
10780        struct bpf_link *link;
10781        int prog_fd, pfd;
10782
10783        prog_fd = bpf_program__fd(prog);
10784        if (prog_fd < 0) {
10785                pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10786                return libbpf_err_ptr(-EINVAL);
10787        }
10788
10789        link = calloc(1, sizeof(*link));
10790        if (!link)
10791                return libbpf_err_ptr(-ENOMEM);
10792        link->detach = &bpf_link__detach_fd;
10793
10794        pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
10795        if (pfd < 0) {
10796                pfd = -errno;
10797                free(link);
10798                pr_warn("prog '%s': failed to attach: %s\n",
10799                        prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
10800                return libbpf_err_ptr(pfd);
10801        }
10802        link->fd = pfd;
10803        return (struct bpf_link *)link;
10804}
10805
10806struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
10807{
10808        return bpf_program__attach_btf_id(prog);
10809}
10810
10811struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
10812{
10813        return bpf_program__attach_btf_id(prog);
10814}
10815
10816static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10817{
10818        *link = bpf_program__attach_trace(prog);
10819        return libbpf_get_error(*link);
10820}
10821
10822static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10823{
10824        *link = bpf_program__attach_lsm(prog);
10825        return libbpf_get_error(*link);
10826}
10827
10828static struct bpf_link *
10829bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id,
10830                       const char *target_name)
10831{
10832        DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
10833                            .target_btf_id = btf_id);
10834        enum bpf_attach_type attach_type;
10835        char errmsg[STRERR_BUFSIZE];
10836        struct bpf_link *link;
10837        int prog_fd, link_fd;
10838
10839        prog_fd = bpf_program__fd(prog);
10840        if (prog_fd < 0) {
10841                pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10842                return libbpf_err_ptr(-EINVAL);
10843        }
10844
10845        link = calloc(1, sizeof(*link));
10846        if (!link)
10847                return libbpf_err_ptr(-ENOMEM);
10848        link->detach = &bpf_link__detach_fd;
10849
10850        attach_type = bpf_program__expected_attach_type(prog);
10851        link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
10852        if (link_fd < 0) {
10853                link_fd = -errno;
10854                free(link);
10855                pr_warn("prog '%s': failed to attach to %s: %s\n",
10856                        prog->name, target_name,
10857                        libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
10858                return libbpf_err_ptr(link_fd);
10859        }
10860        link->fd = link_fd;
10861        return link;
10862}
10863
10864struct bpf_link *
10865bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
10866{
10867        return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
10868}
10869
10870struct bpf_link *
10871bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
10872{
10873        return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
10874}
10875
10876struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
10877{
10878        /* target_fd/target_ifindex use the same field in LINK_CREATE */
10879        return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
10880}
10881
10882struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
10883                                              int target_fd,
10884                                              const char *attach_func_name)
10885{
10886        int btf_id;
10887
10888        if (!!target_fd != !!attach_func_name) {
10889                pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
10890                        prog->name);
10891                return libbpf_err_ptr(-EINVAL);
10892        }
10893
10894        if (prog->type != BPF_PROG_TYPE_EXT) {
10895                pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
10896                        prog->name);
10897                return libbpf_err_ptr(-EINVAL);
10898        }
10899
10900        if (target_fd) {
10901                btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
10902                if (btf_id < 0)
10903                        return libbpf_err_ptr(btf_id);
10904
10905                return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
10906        } else {
10907                /* no target, so use raw_tracepoint_open for compatibility
10908                 * with old kernels
10909                 */
10910                return bpf_program__attach_trace(prog);
10911        }
10912}
10913
10914struct bpf_link *
10915bpf_program__attach_iter(const struct bpf_program *prog,
10916                         const struct bpf_iter_attach_opts *opts)
10917{
10918        DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
10919        char errmsg[STRERR_BUFSIZE];
10920        struct bpf_link *link;
10921        int prog_fd, link_fd;
10922        __u32 target_fd = 0;
10923
10924        if (!OPTS_VALID(opts, bpf_iter_attach_opts))
10925                return libbpf_err_ptr(-EINVAL);
10926
10927        link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
10928        link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
10929
10930        prog_fd = bpf_program__fd(prog);
10931        if (prog_fd < 0) {
10932                pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10933                return libbpf_err_ptr(-EINVAL);
10934        }
10935
10936        link = calloc(1, sizeof(*link));
10937        if (!link)
10938                return libbpf_err_ptr(-ENOMEM);
10939        link->detach = &bpf_link__detach_fd;
10940
10941        link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
10942                                  &link_create_opts);
10943        if (link_fd < 0) {
10944                link_fd = -errno;
10945                free(link);
10946                pr_warn("prog '%s': failed to attach to iterator: %s\n",
10947                        prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
10948                return libbpf_err_ptr(link_fd);
10949        }
10950        link->fd = link_fd;
10951        return link;
10952}
10953
10954static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
10955{
10956        *link = bpf_program__attach_iter(prog, NULL);
10957        return libbpf_get_error(*link);
10958}
10959
10960struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
10961{
10962        struct bpf_link *link = NULL;
10963        int err;
10964
10965        if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
10966                return libbpf_err_ptr(-EOPNOTSUPP);
10967
10968        err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
10969        if (err)
10970                return libbpf_err_ptr(err);
10971
10972        /* When calling bpf_program__attach() explicitly, auto-attach support
10973         * is expected to work, so NULL returned link is considered an error.
10974         * This is different for skeleton's attach, see comment in
10975         * bpf_object__attach_skeleton().
10976         */
10977        if (!link)
10978                return libbpf_err_ptr(-EOPNOTSUPP);
10979
10980        return link;
10981}
10982
10983static int bpf_link__detach_struct_ops(struct bpf_link *link)
10984{
10985        __u32 zero = 0;
10986
10987        if (bpf_map_delete_elem(link->fd, &zero))
10988                return -errno;
10989
10990        return 0;
10991}
10992
10993struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
10994{
10995        struct bpf_struct_ops *st_ops;
10996        struct bpf_link *link;
10997        __u32 i, zero = 0;
10998        int err;
10999
11000        if (!bpf_map__is_struct_ops(map) || map->fd == -1)
11001                return libbpf_err_ptr(-EINVAL);
11002
11003        link = calloc(1, sizeof(*link));
11004        if (!link)
11005                return libbpf_err_ptr(-EINVAL);
11006
11007        st_ops = map->st_ops;
11008        for (i = 0; i < btf_vlen(st_ops->type); i++) {
11009                struct bpf_program *prog = st_ops->progs[i];
11010                void *kern_data;
11011                int prog_fd;
11012
11013                if (!prog)
11014                        continue;
11015
11016                prog_fd = bpf_program__fd(prog);
11017                kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
11018                *(unsigned long *)kern_data = prog_fd;
11019        }
11020
11021        err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
11022        if (err) {
11023                err = -errno;
11024                free(link);
11025                return libbpf_err_ptr(err);
11026        }
11027
11028        link->detach = bpf_link__detach_struct_ops;
11029        link->fd = map->fd;
11030
11031        return link;
11032}
11033
11034static enum bpf_perf_event_ret
11035perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
11036                       void **copy_mem, size_t *copy_size,
11037                       bpf_perf_event_print_t fn, void *private_data)
11038{
11039        struct perf_event_mmap_page *header = mmap_mem;
11040        __u64 data_head = ring_buffer_read_head(header);
11041        __u64 data_tail = header->data_tail;
11042        void *base = ((__u8 *)header) + page_size;
11043        int ret = LIBBPF_PERF_EVENT_CONT;
11044        struct perf_event_header *ehdr;
11045        size_t ehdr_size;
11046
11047        while (data_head != data_tail) {
11048                ehdr = base + (data_tail & (mmap_size - 1));
11049                ehdr_size = ehdr->size;
11050
11051                if (((void *)ehdr) + ehdr_size > base + mmap_size) {
11052                        void *copy_start = ehdr;
11053                        size_t len_first = base + mmap_size - copy_start;
11054                        size_t len_secnd = ehdr_size - len_first;
11055
11056                        if (*copy_size < ehdr_size) {
11057                                free(*copy_mem);
11058                                *copy_mem = malloc(ehdr_size);
11059                                if (!*copy_mem) {
11060                                        *copy_size = 0;
11061                                        ret = LIBBPF_PERF_EVENT_ERROR;
11062                                        break;
11063                                }
11064                                *copy_size = ehdr_size;
11065                        }
11066
11067                        memcpy(*copy_mem, copy_start, len_first);
11068                        memcpy(*copy_mem + len_first, base, len_secnd);
11069                        ehdr = *copy_mem;
11070                }
11071
11072                ret = fn(ehdr, private_data);
11073                data_tail += ehdr_size;
11074                if (ret != LIBBPF_PERF_EVENT_CONT)
11075                        break;
11076        }
11077
11078        ring_buffer_write_tail(header, data_tail);
11079        return libbpf_err(ret);
11080}
11081
11082__attribute__((alias("perf_event_read_simple")))
11083enum bpf_perf_event_ret
11084bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
11085                           void **copy_mem, size_t *copy_size,
11086                           bpf_perf_event_print_t fn, void *private_data);
11087
11088struct perf_buffer;
11089
11090struct perf_buffer_params {
11091        struct perf_event_attr *attr;
11092        /* if event_cb is specified, it takes precendence */
11093        perf_buffer_event_fn event_cb;
11094        /* sample_cb and lost_cb are higher-level common-case callbacks */
11095        perf_buffer_sample_fn sample_cb;
11096        perf_buffer_lost_fn lost_cb;
11097        void *ctx;
11098        int cpu_cnt;
11099        int *cpus;
11100        int *map_keys;
11101};
11102
11103struct perf_cpu_buf {
11104        struct perf_buffer *pb;
11105        void *base; /* mmap()'ed memory */
11106        void *buf; /* for reconstructing segmented data */
11107        size_t buf_size;
11108        int fd;
11109        int cpu;
11110        int map_key;
11111};
11112
11113struct perf_buffer {
11114        perf_buffer_event_fn event_cb;
11115        perf_buffer_sample_fn sample_cb;
11116        perf_buffer_lost_fn lost_cb;
11117        void *ctx; /* passed into callbacks */
11118
11119        size_t page_size;
11120        size_t mmap_size;
11121        struct perf_cpu_buf **cpu_bufs;
11122        struct epoll_event *events;
11123        int cpu_cnt; /* number of allocated CPU buffers */
11124        int epoll_fd; /* perf event FD */
11125        int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
11126};
11127
11128static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
11129                                      struct perf_cpu_buf *cpu_buf)
11130{
11131        if (!cpu_buf)
11132                return;
11133        if (cpu_buf->base &&
11134            munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
11135                pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
11136        if (cpu_buf->fd >= 0) {
11137                ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
11138                close(cpu_buf->fd);
11139        }
11140        free(cpu_buf->buf);
11141        free(cpu_buf);
11142}
11143
11144void perf_buffer__free(struct perf_buffer *pb)
11145{
11146        int i;
11147
11148        if (IS_ERR_OR_NULL(pb))
11149                return;
11150        if (pb->cpu_bufs) {
11151                for (i = 0; i < pb->cpu_cnt; i++) {
11152                        struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
11153
11154                        if (!cpu_buf)
11155                                continue;
11156
11157                        bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
11158                        perf_buffer__free_cpu_buf(pb, cpu_buf);
11159                }
11160                free(pb->cpu_bufs);
11161        }
11162        if (pb->epoll_fd >= 0)
11163                close(pb->epoll_fd);
11164        free(pb->events);
11165        free(pb);
11166}
11167
11168static struct perf_cpu_buf *
11169perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
11170                          int cpu, int map_key)
11171{
11172        struct perf_cpu_buf *cpu_buf;
11173        char msg[STRERR_BUFSIZE];
11174        int err;
11175
11176        cpu_buf = calloc(1, sizeof(*cpu_buf));
11177        if (!cpu_buf)
11178                return ERR_PTR(-ENOMEM);
11179
11180        cpu_buf->pb = pb;
11181        cpu_buf->cpu = cpu;
11182        cpu_buf->map_key = map_key;
11183
11184        cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
11185                              -1, PERF_FLAG_FD_CLOEXEC);
11186        if (cpu_buf->fd < 0) {
11187                err = -errno;
11188                pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
11189                        cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11190                goto error;
11191        }
11192
11193        cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
11194                             PROT_READ | PROT_WRITE, MAP_SHARED,
11195                             cpu_buf->fd, 0);
11196        if (cpu_buf->base == MAP_FAILED) {
11197                cpu_buf->base = NULL;
11198                err = -errno;
11199                pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
11200                        cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11201                goto error;
11202        }
11203
11204        if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
11205                err = -errno;
11206                pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
11207                        cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
11208                goto error;
11209        }
11210
11211        return cpu_buf;
11212
11213error:
11214        perf_buffer__free_cpu_buf(pb, cpu_buf);
11215        return (struct perf_cpu_buf *)ERR_PTR(err);
11216}
11217
11218static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
11219                                              struct perf_buffer_params *p);
11220
11221DEFAULT_VERSION(perf_buffer__new_v0_6_0, perf_buffer__new, LIBBPF_0.6.0)
11222struct perf_buffer *perf_buffer__new_v0_6_0(int map_fd, size_t page_cnt,
11223                                            perf_buffer_sample_fn sample_cb,
11224                                            perf_buffer_lost_fn lost_cb,
11225                                            void *ctx,
11226                                            const struct perf_buffer_opts *opts)
11227{
11228        struct perf_buffer_params p = {};
11229        struct perf_event_attr attr = {};
11230
11231        if (!OPTS_VALID(opts, perf_buffer_opts))
11232                return libbpf_err_ptr(-EINVAL);
11233
11234        attr.config = PERF_COUNT_SW_BPF_OUTPUT;
11235        attr.type = PERF_TYPE_SOFTWARE;
11236        attr.sample_type = PERF_SAMPLE_RAW;
11237        attr.sample_period = 1;
11238        attr.wakeup_events = 1;
11239
11240        p.attr = &attr;
11241        p.sample_cb = sample_cb;
11242        p.lost_cb = lost_cb;
11243        p.ctx = ctx;
11244
11245        return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
11246}
11247
11248COMPAT_VERSION(perf_buffer__new_deprecated, perf_buffer__new, LIBBPF_0.0.4)
11249struct perf_buffer *perf_buffer__new_deprecated(int map_fd, size_t page_cnt,
11250                                                const struct perf_buffer_opts *opts)
11251{
11252        return perf_buffer__new_v0_6_0(map_fd, page_cnt,
11253                                       opts ? opts->sample_cb : NULL,
11254                                       opts ? opts->lost_cb : NULL,
11255                                       opts ? opts->ctx : NULL,
11256                                       NULL);
11257}
11258
11259DEFAULT_VERSION(perf_buffer__new_raw_v0_6_0, perf_buffer__new_raw, LIBBPF_0.6.0)
11260struct perf_buffer *perf_buffer__new_raw_v0_6_0(int map_fd, size_t page_cnt,
11261                                                struct perf_event_attr *attr,
11262                                                perf_buffer_event_fn event_cb, void *ctx,
11263                                                const struct perf_buffer_raw_opts *opts)
11264{
11265        struct perf_buffer_params p = {};
11266
11267        if (!attr)
11268                return libbpf_err_ptr(-EINVAL);
11269
11270        if (!OPTS_VALID(opts, perf_buffer_raw_opts))
11271                return libbpf_err_ptr(-EINVAL);
11272
11273        p.attr = attr;
11274        p.event_cb = event_cb;
11275        p.ctx = ctx;
11276        p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
11277        p.cpus = OPTS_GET(opts, cpus, NULL);
11278        p.map_keys = OPTS_GET(opts, map_keys, NULL);
11279
11280        return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
11281}
11282
11283COMPAT_VERSION(perf_buffer__new_raw_deprecated, perf_buffer__new_raw, LIBBPF_0.0.4)
11284struct perf_buffer *perf_buffer__new_raw_deprecated(int map_fd, size_t page_cnt,
11285                                                    const struct perf_buffer_raw_opts *opts)
11286{
11287        LIBBPF_OPTS(perf_buffer_raw_opts, inner_opts,
11288                .cpu_cnt = opts->cpu_cnt,
11289                .cpus = opts->cpus,
11290                .map_keys = opts->map_keys,
11291        );
11292
11293        return perf_buffer__new_raw_v0_6_0(map_fd, page_cnt, opts->attr,
11294                                           opts->event_cb, opts->ctx, &inner_opts);
11295}
11296
11297static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
11298                                              struct perf_buffer_params *p)
11299{
11300        const char *online_cpus_file = "/sys/devices/system/cpu/online";
11301        struct bpf_map_info map;
11302        char msg[STRERR_BUFSIZE];
11303        struct perf_buffer *pb;
11304        bool *online = NULL;
11305        __u32 map_info_len;
11306        int err, i, j, n;
11307
11308        if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
11309                pr_warn("page count should be power of two, but is %zu\n",
11310                        page_cnt);
11311                return ERR_PTR(-EINVAL);
11312        }
11313
11314        /* best-effort sanity checks */
11315        memset(&map, 0, sizeof(map));
11316        map_info_len = sizeof(map);
11317        err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
11318        if (err) {
11319                err = -errno;
11320                /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
11321                 * -EBADFD, -EFAULT, or -E2BIG on real error
11322                 */
11323                if (err != -EINVAL) {
11324                        pr_warn("failed to get map info for map FD %d: %s\n",
11325                                map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
11326                        return ERR_PTR(err);
11327                }
11328                pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
11329                         map_fd);
11330        } else {
11331                if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
11332                        pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
11333                                map.name);
11334                        return ERR_PTR(-EINVAL);
11335                }
11336        }
11337
11338        pb = calloc(1, sizeof(*pb));
11339        if (!pb)
11340                return ERR_PTR(-ENOMEM);
11341
11342        pb->event_cb = p->event_cb;
11343        pb->sample_cb = p->sample_cb;
11344        pb->lost_cb = p->lost_cb;
11345        pb->ctx = p->ctx;
11346
11347        pb->page_size = getpagesize();
11348        pb->mmap_size = pb->page_size * page_cnt;
11349        pb->map_fd = map_fd;
11350
11351        pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
11352        if (pb->epoll_fd < 0) {
11353                err = -errno;
11354                pr_warn("failed to create epoll instance: %s\n",
11355                        libbpf_strerror_r(err, msg, sizeof(msg)));
11356                goto error;
11357        }
11358
11359        if (p->cpu_cnt > 0) {
11360                pb->cpu_cnt = p->cpu_cnt;
11361        } else {
11362                pb->cpu_cnt = libbpf_num_possible_cpus();
11363                if (pb->cpu_cnt < 0) {
11364                        err = pb->cpu_cnt;
11365                        goto error;
11366                }
11367                if (map.max_entries && map.max_entries < pb->cpu_cnt)
11368                        pb->cpu_cnt = map.max_entries;
11369        }
11370
11371        pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
11372        if (!pb->events) {
11373                err = -ENOMEM;
11374                pr_warn("failed to allocate events: out of memory\n");
11375                goto error;
11376        }
11377        pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
11378        if (!pb->cpu_bufs) {
11379                err = -ENOMEM;
11380                pr_warn("failed to allocate buffers: out of memory\n");
11381                goto error;
11382        }
11383
11384        err = parse_cpu_mask_file(online_cpus_file, &online, &n);
11385        if (err) {
11386                pr_warn("failed to get online CPU mask: %d\n", err);
11387                goto error;
11388        }
11389
11390        for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
11391                struct perf_cpu_buf *cpu_buf;
11392                int cpu, map_key;
11393
11394                cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
11395                map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
11396
11397                /* in case user didn't explicitly requested particular CPUs to
11398                 * be attached to, skip offline/not present CPUs
11399                 */
11400                if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
11401                        continue;
11402
11403                cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
11404                if (IS_ERR(cpu_buf)) {
11405                        err = PTR_ERR(cpu_buf);
11406                        goto error;
11407                }
11408
11409                pb->cpu_bufs[j] = cpu_buf;
11410
11411                err = bpf_map_update_elem(pb->map_fd, &map_key,
11412                                          &cpu_buf->fd, 0);
11413                if (err) {
11414                        err = -errno;
11415                        pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
11416                                cpu, map_key, cpu_buf->fd,
11417                                libbpf_strerror_r(err, msg, sizeof(msg)));
11418                        goto error;
11419                }
11420
11421                pb->events[j].events = EPOLLIN;
11422                pb->events[j].data.ptr = cpu_buf;
11423                if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
11424                              &pb->events[j]) < 0) {
11425                        err = -errno;
11426                        pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
11427                                cpu, cpu_buf->fd,
11428                                libbpf_strerror_r(err, msg, sizeof(msg)));
11429                        goto error;
11430                }
11431                j++;
11432        }
11433        pb->cpu_cnt = j;
11434        free(online);
11435
11436        return pb;
11437
11438error:
11439        free(online);
11440        if (pb)
11441                perf_buffer__free(pb);
11442        return ERR_PTR(err);
11443}
11444
11445struct perf_sample_raw {
11446        struct perf_event_header header;
11447        uint32_t size;
11448        char data[];
11449};
11450
11451struct perf_sample_lost {
11452        struct perf_event_header header;
11453        uint64_t id;
11454        uint64_t lost;
11455        uint64_t sample_id;
11456};
11457
11458static enum bpf_perf_event_ret
11459perf_buffer__process_record(struct perf_event_header *e, void *ctx)
11460{
11461        struct perf_cpu_buf *cpu_buf = ctx;
11462        struct perf_buffer *pb = cpu_buf->pb;
11463        void *data = e;
11464
11465        /* user wants full control over parsing perf event */
11466        if (pb->event_cb)
11467                return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
11468
11469        switch (e->type) {
11470        case PERF_RECORD_SAMPLE: {
11471                struct perf_sample_raw *s = data;
11472
11473                if (pb->sample_cb)
11474                        pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
11475                break;
11476        }
11477        case PERF_RECORD_LOST: {
11478                struct perf_sample_lost *s = data;
11479
11480                if (pb->lost_cb)
11481                        pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
11482                break;
11483        }
11484        default:
11485                pr_warn("unknown perf sample type %d\n", e->type);
11486                return LIBBPF_PERF_EVENT_ERROR;
11487        }
11488        return LIBBPF_PERF_EVENT_CONT;
11489}
11490
11491static int perf_buffer__process_records(struct perf_buffer *pb,
11492                                        struct perf_cpu_buf *cpu_buf)
11493{
11494        enum bpf_perf_event_ret ret;
11495
11496        ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
11497                                     pb->page_size, &cpu_buf->buf,
11498                                     &cpu_buf->buf_size,
11499                                     perf_buffer__process_record, cpu_buf);
11500        if (ret != LIBBPF_PERF_EVENT_CONT)
11501                return ret;
11502        return 0;
11503}
11504
11505int perf_buffer__epoll_fd(const struct perf_buffer *pb)
11506{
11507        return pb->epoll_fd;
11508}
11509
11510int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
11511{
11512        int i, cnt, err;
11513
11514        cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
11515        if (cnt < 0)
11516                return -errno;
11517
11518        for (i = 0; i < cnt; i++) {
11519                struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
11520
11521                err = perf_buffer__process_records(pb, cpu_buf);
11522                if (err) {
11523                        pr_warn("error while processing records: %d\n", err);
11524                        return libbpf_err(err);
11525                }
11526        }
11527        return cnt;
11528}
11529
11530/* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
11531 * manager.
11532 */
11533size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
11534{
11535        return pb->cpu_cnt;
11536}
11537
11538/*
11539 * Return perf_event FD of a ring buffer in *buf_idx* slot of
11540 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
11541 * select()/poll()/epoll() Linux syscalls.
11542 */
11543int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
11544{
11545        struct perf_cpu_buf *cpu_buf;
11546
11547        if (buf_idx >= pb->cpu_cnt)
11548                return libbpf_err(-EINVAL);
11549
11550        cpu_buf = pb->cpu_bufs[buf_idx];
11551        if (!cpu_buf)
11552                return libbpf_err(-ENOENT);
11553
11554        return cpu_buf->fd;
11555}
11556
11557/*
11558 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
11559 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
11560 * consume, do nothing and return success.
11561 * Returns:
11562 *   - 0 on success;
11563 *   - <0 on failure.
11564 */
11565int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
11566{
11567        struct perf_cpu_buf *cpu_buf;
11568
11569        if (buf_idx >= pb->cpu_cnt)
11570                return libbpf_err(-EINVAL);
11571
11572        cpu_buf = pb->cpu_bufs[buf_idx];
11573        if (!cpu_buf)
11574                return libbpf_err(-ENOENT);
11575
11576        return perf_buffer__process_records(pb, cpu_buf);
11577}
11578
11579int perf_buffer__consume(struct perf_buffer *pb)
11580{
11581        int i, err;
11582
11583        for (i = 0; i < pb->cpu_cnt; i++) {
11584                struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
11585
11586                if (!cpu_buf)
11587                        continue;
11588
11589                err = perf_buffer__process_records(pb, cpu_buf);
11590                if (err) {
11591                        pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
11592                        return libbpf_err(err);
11593                }
11594        }
11595        return 0;
11596}
11597
11598struct bpf_prog_info_array_desc {
11599        int     array_offset;   /* e.g. offset of jited_prog_insns */
11600        int     count_offset;   /* e.g. offset of jited_prog_len */
11601        int     size_offset;    /* > 0: offset of rec size,
11602                                 * < 0: fix size of -size_offset
11603                                 */
11604};
11605
11606static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
11607        [BPF_PROG_INFO_JITED_INSNS] = {
11608                offsetof(struct bpf_prog_info, jited_prog_insns),
11609                offsetof(struct bpf_prog_info, jited_prog_len),
11610                -1,
11611        },
11612        [BPF_PROG_INFO_XLATED_INSNS] = {
11613                offsetof(struct bpf_prog_info, xlated_prog_insns),
11614                offsetof(struct bpf_prog_info, xlated_prog_len),
11615                -1,
11616        },
11617        [BPF_PROG_INFO_MAP_IDS] = {
11618                offsetof(struct bpf_prog_info, map_ids),
11619                offsetof(struct bpf_prog_info, nr_map_ids),
11620                -(int)sizeof(__u32),
11621        },
11622        [BPF_PROG_INFO_JITED_KSYMS] = {
11623                offsetof(struct bpf_prog_info, jited_ksyms),
11624                offsetof(struct bpf_prog_info, nr_jited_ksyms),
11625                -(int)sizeof(__u64),
11626        },
11627        [BPF_PROG_INFO_JITED_FUNC_LENS] = {
11628                offsetof(struct bpf_prog_info, jited_func_lens),
11629                offsetof(struct bpf_prog_info, nr_jited_func_lens),
11630                -(int)sizeof(__u32),
11631        },
11632        [BPF_PROG_INFO_FUNC_INFO] = {
11633                offsetof(struct bpf_prog_info, func_info),
11634                offsetof(struct bpf_prog_info, nr_func_info),
11635                offsetof(struct bpf_prog_info, func_info_rec_size),
11636        },
11637        [BPF_PROG_INFO_LINE_INFO] = {
11638                offsetof(struct bpf_prog_info, line_info),
11639                offsetof(struct bpf_prog_info, nr_line_info),
11640                offsetof(struct bpf_prog_info, line_info_rec_size),
11641        },
11642        [BPF_PROG_INFO_JITED_LINE_INFO] = {
11643                offsetof(struct bpf_prog_info, jited_line_info),
11644                offsetof(struct bpf_prog_info, nr_jited_line_info),
11645                offsetof(struct bpf_prog_info, jited_line_info_rec_size),
11646        },
11647        [BPF_PROG_INFO_PROG_TAGS] = {
11648                offsetof(struct bpf_prog_info, prog_tags),
11649                offsetof(struct bpf_prog_info, nr_prog_tags),
11650                -(int)sizeof(__u8) * BPF_TAG_SIZE,
11651        },
11652
11653};
11654
11655static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
11656                                           int offset)
11657{
11658        __u32 *array = (__u32 *)info;
11659
11660        if (offset >= 0)
11661                return array[offset / sizeof(__u32)];
11662        return -(int)offset;
11663}
11664
11665static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
11666                                           int offset)
11667{
11668        __u64 *array = (__u64 *)info;
11669
11670        if (offset >= 0)
11671                return array[offset / sizeof(__u64)];
11672        return -(int)offset;
11673}
11674
11675static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
11676                                         __u32 val)
11677{
11678        __u32 *array = (__u32 *)info;
11679
11680        if (offset >= 0)
11681                array[offset / sizeof(__u32)] = val;
11682}
11683
11684static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
11685                                         __u64 val)
11686{
11687        __u64 *array = (__u64 *)info;
11688
11689        if (offset >= 0)
11690                array[offset / sizeof(__u64)] = val;
11691}
11692
11693struct bpf_prog_info_linear *
11694bpf_program__get_prog_info_linear(int fd, __u64 arrays)
11695{
11696        struct bpf_prog_info_linear *info_linear;
11697        struct bpf_prog_info info = {};
11698        __u32 info_len = sizeof(info);
11699        __u32 data_len = 0;
11700        int i, err;
11701        void *ptr;
11702
11703        if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
11704                return libbpf_err_ptr(-EINVAL);
11705
11706        /* step 1: get array dimensions */
11707        err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
11708        if (err) {
11709                pr_debug("can't get prog info: %s", strerror(errno));
11710                return libbpf_err_ptr(-EFAULT);
11711        }
11712
11713        /* step 2: calculate total size of all arrays */
11714        for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11715                bool include_array = (arrays & (1UL << i)) > 0;
11716                struct bpf_prog_info_array_desc *desc;
11717                __u32 count, size;
11718
11719                desc = bpf_prog_info_array_desc + i;
11720
11721                /* kernel is too old to support this field */
11722                if (info_len < desc->array_offset + sizeof(__u32) ||
11723                    info_len < desc->count_offset + sizeof(__u32) ||
11724                    (desc->size_offset > 0 && info_len < desc->size_offset))
11725                        include_array = false;
11726
11727                if (!include_array) {
11728                        arrays &= ~(1UL << i);  /* clear the bit */
11729                        continue;
11730                }
11731
11732                count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11733                size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11734
11735                data_len += count * size;
11736        }
11737
11738        /* step 3: allocate continuous memory */
11739        data_len = roundup(data_len, sizeof(__u64));
11740        info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
11741        if (!info_linear)
11742                return libbpf_err_ptr(-ENOMEM);
11743
11744        /* step 4: fill data to info_linear->info */
11745        info_linear->arrays = arrays;
11746        memset(&info_linear->info, 0, sizeof(info));
11747        ptr = info_linear->data;
11748
11749        for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11750                struct bpf_prog_info_array_desc *desc;
11751                __u32 count, size;
11752
11753                if ((arrays & (1UL << i)) == 0)
11754                        continue;
11755
11756                desc  = bpf_prog_info_array_desc + i;
11757                count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11758                size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11759                bpf_prog_info_set_offset_u32(&info_linear->info,
11760                                             desc->count_offset, count);
11761                bpf_prog_info_set_offset_u32(&info_linear->info,
11762                                             desc->size_offset, size);
11763                bpf_prog_info_set_offset_u64(&info_linear->info,
11764                                             desc->array_offset,
11765                                             ptr_to_u64(ptr));
11766                ptr += count * size;
11767        }
11768
11769        /* step 5: call syscall again to get required arrays */
11770        err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
11771        if (err) {
11772                pr_debug("can't get prog info: %s", strerror(errno));
11773                free(info_linear);
11774                return libbpf_err_ptr(-EFAULT);
11775        }
11776
11777        /* step 6: verify the data */
11778        for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11779                struct bpf_prog_info_array_desc *desc;
11780                __u32 v1, v2;
11781
11782                if ((arrays & (1UL << i)) == 0)
11783                        continue;
11784
11785                desc = bpf_prog_info_array_desc + i;
11786                v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
11787                v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
11788                                                   desc->count_offset);
11789                if (v1 != v2)
11790                        pr_warn("%s: mismatch in element count\n", __func__);
11791
11792                v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
11793                v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
11794                                                   desc->size_offset);
11795                if (v1 != v2)
11796                        pr_warn("%s: mismatch in rec size\n", __func__);
11797        }
11798
11799        /* step 7: update info_len and data_len */
11800        info_linear->info_len = sizeof(struct bpf_prog_info);
11801        info_linear->data_len = data_len;
11802
11803        return info_linear;
11804}
11805
11806void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
11807{
11808        int i;
11809
11810        for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11811                struct bpf_prog_info_array_desc *desc;
11812                __u64 addr, offs;
11813
11814                if ((info_linear->arrays & (1UL << i)) == 0)
11815                        continue;
11816
11817                desc = bpf_prog_info_array_desc + i;
11818                addr = bpf_prog_info_read_offset_u64(&info_linear->info,
11819                                                     desc->array_offset);
11820                offs = addr - ptr_to_u64(info_linear->data);
11821                bpf_prog_info_set_offset_u64(&info_linear->info,
11822                                             desc->array_offset, offs);
11823        }
11824}
11825
11826void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
11827{
11828        int i;
11829
11830        for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
11831                struct bpf_prog_info_array_desc *desc;
11832                __u64 addr, offs;
11833
11834                if ((info_linear->arrays & (1UL << i)) == 0)
11835                        continue;
11836
11837                desc = bpf_prog_info_array_desc + i;
11838                offs = bpf_prog_info_read_offset_u64(&info_linear->info,
11839                                                     desc->array_offset);
11840                addr = offs + ptr_to_u64(info_linear->data);
11841                bpf_prog_info_set_offset_u64(&info_linear->info,
11842                                             desc->array_offset, addr);
11843        }
11844}
11845
11846int bpf_program__set_attach_target(struct bpf_program *prog,
11847                                   int attach_prog_fd,
11848                                   const char *attach_func_name)
11849{
11850        int btf_obj_fd = 0, btf_id = 0, err;
11851
11852        if (!prog || attach_prog_fd < 0)
11853                return libbpf_err(-EINVAL);
11854
11855        if (prog->obj->loaded)
11856                return libbpf_err(-EINVAL);
11857
11858        if (attach_prog_fd && !attach_func_name) {
11859                /* remember attach_prog_fd and let bpf_program__load() find
11860                 * BTF ID during the program load
11861                 */
11862                prog->attach_prog_fd = attach_prog_fd;
11863                return 0;
11864        }
11865
11866        if (attach_prog_fd) {
11867                btf_id = libbpf_find_prog_btf_id(attach_func_name,
11868                                                 attach_prog_fd);
11869                if (btf_id < 0)
11870                        return libbpf_err(btf_id);
11871        } else {
11872                if (!attach_func_name)
11873                        return libbpf_err(-EINVAL);
11874
11875                /* load btf_vmlinux, if not yet */
11876                err = bpf_object__load_vmlinux_btf(prog->obj, true);
11877                if (err)
11878                        return libbpf_err(err);
11879                err = find_kernel_btf_id(prog->obj, attach_func_name,
11880                                         prog->expected_attach_type,
11881                                         &btf_obj_fd, &btf_id);
11882                if (err)
11883                        return libbpf_err(err);
11884        }
11885
11886        prog->attach_btf_id = btf_id;
11887        prog->attach_btf_obj_fd = btf_obj_fd;
11888        prog->attach_prog_fd = attach_prog_fd;
11889        return 0;
11890}
11891
11892int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
11893{
11894        int err = 0, n, len, start, end = -1;
11895        bool *tmp;
11896
11897        *mask = NULL;
11898        *mask_sz = 0;
11899
11900        /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
11901        while (*s) {
11902                if (*s == ',' || *s == '\n') {
11903                        s++;
11904                        continue;
11905                }
11906                n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
11907                if (n <= 0 || n > 2) {
11908                        pr_warn("Failed to get CPU range %s: %d\n", s, n);
11909                        err = -EINVAL;
11910                        goto cleanup;
11911                } else if (n == 1) {
11912                        end = start;
11913                }
11914                if (start < 0 || start > end) {
11915                        pr_warn("Invalid CPU range [%d,%d] in %s\n",
11916                                start, end, s);
11917                        err = -EINVAL;
11918                        goto cleanup;
11919                }
11920                tmp = realloc(*mask, end + 1);
11921                if (!tmp) {
11922                        err = -ENOMEM;
11923                        goto cleanup;
11924                }
11925                *mask = tmp;
11926                memset(tmp + *mask_sz, 0, start - *mask_sz);
11927                memset(tmp + start, 1, end - start + 1);
11928                *mask_sz = end + 1;
11929                s += len;
11930        }
11931        if (!*mask_sz) {
11932                pr_warn("Empty CPU range\n");
11933                return -EINVAL;
11934        }
11935        return 0;
11936cleanup:
11937        free(*mask);
11938        *mask = NULL;
11939        return err;
11940}
11941
11942int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
11943{
11944        int fd, err = 0, len;
11945        char buf[128];
11946
11947        fd = open(fcpu, O_RDONLY | O_CLOEXEC);
11948        if (fd < 0) {
11949                err = -errno;
11950                pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
11951                return err;
11952        }
11953        len = read(fd, buf, sizeof(buf));
11954        close(fd);
11955        if (len <= 0) {
11956                err = len ? -errno : -EINVAL;
11957                pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
11958                return err;
11959        }
11960        if (len >= sizeof(buf)) {
11961                pr_warn("CPU mask is too big in file %s\n", fcpu);
11962                return -E2BIG;
11963        }
11964        buf[len] = '\0';
11965
11966        return parse_cpu_mask_str(buf, mask, mask_sz);
11967}
11968
11969int libbpf_num_possible_cpus(void)
11970{
11971        static const char *fcpu = "/sys/devices/system/cpu/possible";
11972        static int cpus;
11973        int err, n, i, tmp_cpus;
11974        bool *mask;
11975
11976        tmp_cpus = READ_ONCE(cpus);
11977        if (tmp_cpus > 0)
11978                return tmp_cpus;
11979
11980        err = parse_cpu_mask_file(fcpu, &mask, &n);
11981        if (err)
11982                return libbpf_err(err);
11983
11984        tmp_cpus = 0;
11985        for (i = 0; i < n; i++) {
11986                if (mask[i])
11987                        tmp_cpus++;
11988        }
11989        free(mask);
11990
11991        WRITE_ONCE(cpus, tmp_cpus);
11992        return tmp_cpus;
11993}
11994
11995static int populate_skeleton_maps(const struct bpf_object *obj,
11996                                  struct bpf_map_skeleton *maps,
11997                                  size_t map_cnt)
11998{
11999        int i;
12000
12001        for (i = 0; i < map_cnt; i++) {
12002                struct bpf_map **map = maps[i].map;
12003                const char *name = maps[i].name;
12004                void **mmaped = maps[i].mmaped;
12005
12006                *map = bpf_object__find_map_by_name(obj, name);
12007                if (!*map) {
12008                        pr_warn("failed to find skeleton map '%s'\n", name);
12009                        return -ESRCH;
12010                }
12011
12012                /* externs shouldn't be pre-setup from user code */
12013                if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
12014                        *mmaped = (*map)->mmaped;
12015        }
12016        return 0;
12017}
12018
12019static int populate_skeleton_progs(const struct bpf_object *obj,
12020                                   struct bpf_prog_skeleton *progs,
12021                                   size_t prog_cnt)
12022{
12023        int i;
12024
12025        for (i = 0; i < prog_cnt; i++) {
12026                struct bpf_program **prog = progs[i].prog;
12027                const char *name = progs[i].name;
12028
12029                *prog = bpf_object__find_program_by_name(obj, name);
12030                if (!*prog) {
12031                        pr_warn("failed to find skeleton program '%s'\n", name);
12032                        return -ESRCH;
12033                }
12034        }
12035        return 0;
12036}
12037
12038int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
12039                              const struct bpf_object_open_opts *opts)
12040{
12041        DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
12042                .object_name = s->name,
12043        );
12044        struct bpf_object *obj;
12045        int err;
12046
12047        /* Attempt to preserve opts->object_name, unless overriden by user
12048         * explicitly. Overwriting object name for skeletons is discouraged,
12049         * as it breaks global data maps, because they contain object name
12050         * prefix as their own map name prefix. When skeleton is generated,
12051         * bpftool is making an assumption that this name will stay the same.
12052         */
12053        if (opts) {
12054                memcpy(&skel_opts, opts, sizeof(*opts));
12055                if (!opts->object_name)
12056                        skel_opts.object_name = s->name;
12057        }
12058
12059        obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
12060        err = libbpf_get_error(obj);
12061        if (err) {
12062                pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
12063                        s->name, err);
12064                return libbpf_err(err);
12065        }
12066
12067        *s->obj = obj;
12068        err = populate_skeleton_maps(obj, s->maps, s->map_cnt);
12069        if (err) {
12070                pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err);
12071                return libbpf_err(err);
12072        }
12073
12074        err = populate_skeleton_progs(obj, s->progs, s->prog_cnt);
12075        if (err) {
12076                pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err);
12077                return libbpf_err(err);
12078        }
12079
12080        return 0;
12081}
12082
12083int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
12084{
12085        int err, len, var_idx, i;
12086        const char *var_name;
12087        const struct bpf_map *map;
12088        struct btf *btf;
12089        __u32 map_type_id;
12090        const struct btf_type *map_type, *var_type;
12091        const struct bpf_var_skeleton *var_skel;
12092        struct btf_var_secinfo *var;
12093
12094        if (!s->obj)
12095                return libbpf_err(-EINVAL);
12096
12097        btf = bpf_object__btf(s->obj);
12098        if (!btf) {
12099                pr_warn("subskeletons require BTF at runtime (object %s)\n",
12100                        bpf_object__name(s->obj));
12101                return libbpf_err(-errno);
12102        }
12103
12104        err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt);
12105        if (err) {
12106                pr_warn("failed to populate subskeleton maps: %d\n", err);
12107                return libbpf_err(err);
12108        }
12109
12110        err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt);
12111        if (err) {
12112                pr_warn("failed to populate subskeleton maps: %d\n", err);
12113                return libbpf_err(err);
12114        }
12115
12116        for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
12117                var_skel = &s->vars[var_idx];
12118                map = *var_skel->map;
12119                map_type_id = bpf_map__btf_value_type_id(map);
12120                map_type = btf__type_by_id(btf, map_type_id);
12121
12122                if (!btf_is_datasec(map_type)) {
12123                        pr_warn("type for map '%1$s' is not a datasec: %2$s",
12124                                bpf_map__name(map),
12125                                __btf_kind_str(btf_kind(map_type)));
12126                        return libbpf_err(-EINVAL);
12127                }
12128
12129                len = btf_vlen(map_type);
12130                var = btf_var_secinfos(map_type);
12131                for (i = 0; i < len; i++, var++) {
12132                        var_type = btf__type_by_id(btf, var->type);
12133                        var_name = btf__name_by_offset(btf, var_type->name_off);
12134                        if (strcmp(var_name, var_skel->name) == 0) {
12135                                *var_skel->addr = map->mmaped + var->offset;
12136                                break;
12137                        }
12138                }
12139        }
12140        return 0;
12141}
12142
12143void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
12144{
12145        if (!s)
12146                return;
12147        free(s->maps);
12148        free(s->progs);
12149        free(s->vars);
12150        free(s);
12151}
12152
12153int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
12154{
12155        int i, err;
12156
12157        err = bpf_object__load(*s->obj);
12158        if (err) {
12159                pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
12160                return libbpf_err(err);
12161        }
12162
12163        for (i = 0; i < s->map_cnt; i++) {
12164                struct bpf_map *map = *s->maps[i].map;
12165                size_t mmap_sz = bpf_map_mmap_sz(map);
12166                int prot, map_fd = bpf_map__fd(map);
12167                void **mmaped = s->maps[i].mmaped;
12168
12169                if (!mmaped)
12170                        continue;
12171
12172                if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
12173                        *mmaped = NULL;
12174                        continue;
12175                }
12176
12177                if (map->def.map_flags & BPF_F_RDONLY_PROG)
12178                        prot = PROT_READ;
12179                else
12180                        prot = PROT_READ | PROT_WRITE;
12181
12182                /* Remap anonymous mmap()-ed "map initialization image" as
12183                 * a BPF map-backed mmap()-ed memory, but preserving the same
12184                 * memory address. This will cause kernel to change process'
12185                 * page table to point to a different piece of kernel memory,
12186                 * but from userspace point of view memory address (and its
12187                 * contents, being identical at this point) will stay the
12188                 * same. This mapping will be released by bpf_object__close()
12189                 * as per normal clean up procedure, so we don't need to worry
12190                 * about it from skeleton's clean up perspective.
12191                 */
12192                *mmaped = mmap(map->mmaped, mmap_sz, prot,
12193                                MAP_SHARED | MAP_FIXED, map_fd, 0);
12194                if (*mmaped == MAP_FAILED) {
12195                        err = -errno;
12196                        *mmaped = NULL;
12197                        pr_warn("failed to re-mmap() map '%s': %d\n",
12198                                 bpf_map__name(map), err);
12199                        return libbpf_err(err);
12200                }
12201        }
12202
12203        return 0;
12204}
12205
12206int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
12207{
12208        int i, err;
12209
12210        for (i = 0; i < s->prog_cnt; i++) {
12211                struct bpf_program *prog = *s->progs[i].prog;
12212                struct bpf_link **link = s->progs[i].link;
12213
12214                if (!prog->load)
12215                        continue;
12216
12217                /* auto-attaching not supported for this program */
12218                if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
12219                        continue;
12220
12221                /* if user already set the link manually, don't attempt auto-attach */
12222                if (*link)
12223                        continue;
12224
12225                err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
12226                if (err) {
12227                        pr_warn("prog '%s': failed to auto-attach: %d\n",
12228                                bpf_program__name(prog), err);
12229                        return libbpf_err(err);
12230                }
12231
12232                /* It's possible that for some SEC() definitions auto-attach
12233                 * is supported in some cases (e.g., if definition completely
12234                 * specifies target information), but is not in other cases.
12235                 * SEC("uprobe") is one such case. If user specified target
12236                 * binary and function name, such BPF program can be
12237                 * auto-attached. But if not, it shouldn't trigger skeleton's
12238                 * attach to fail. It should just be skipped.
12239                 * attach_fn signals such case with returning 0 (no error) and
12240                 * setting link to NULL.
12241                 */
12242        }
12243
12244        return 0;
12245}
12246
12247void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
12248{
12249        int i;
12250
12251        for (i = 0; i < s->prog_cnt; i++) {
12252                struct bpf_link **link = s->progs[i].link;
12253
12254                bpf_link__destroy(*link);
12255                *link = NULL;
12256        }
12257}
12258
12259void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
12260{
12261        if (!s)
12262                return;
12263
12264        if (s->progs)
12265                bpf_object__detach_skeleton(s);
12266        if (s->obj)
12267                bpf_object__close(*s->obj);
12268        free(s->maps);
12269        free(s->progs);
12270        free(s);
12271}
12272