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