tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2
   3 /*
   4  * Common eBPF ELF object loading operations.
   5  *
   6  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
   7  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
   8  * Copyright (C) 2015 Huawei Inc.
   9  * Copyright (C) 2017 Nicira, Inc.
  10  * Copyright (C) 2019 Isovalent, Inc.
  11  */
  12
  13 #ifndef _GNU_SOURCE
  14 #define _GNU_SOURCE
  15 #endif
  16 #include <stdlib.h>
  17 #include <stdio.h>
  18 #include <stdarg.h>
  19 #include <libgen.h>
  20 #include <inttypes.h>
  21 #include <limits.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <endian.h>
  25 #include <fcntl.h>
  26 #include <errno.h>
  27 #include <ctype.h>
  28 #include <asm/unistd.h>
  29 #include <linux/err.h>
  30 #include <linux/kernel.h>
  31 #include <linux/bpf.h>
  32 #include <linux/btf.h>
  33 #include <linux/filter.h>
  34 #include <linux/list.h>
  35 #include <linux/limits.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/ring_buffer.h>
  38 #include <linux/version.h>
  39 #include <sys/epoll.h>
  40 #include <sys/ioctl.h>
  41 #include <sys/mman.h>
  42 #include <sys/stat.h>
  43 #include <sys/types.h>
  44 #include <sys/vfs.h>
  45 #include <sys/utsname.h>
  46 #include <sys/resource.h>
  47 #include <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
  72 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
  73 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
  74
  75 static 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
  84 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  85
  86 libbpf_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)
  95 void 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
 107 static 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
 149 static 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 */
 155 enum libbpf_strict_mode libbpf_mode = LIBBPF_STRICT_NONE;
 156
 157 int 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 enum kern_feature_id {
 172         /* v4.14: kernel support for program & map names. */
 173         FEAT_PROG_NAME,
 174         /* v5.2: kernel support for global data sections. */
 175         FEAT_GLOBAL_DATA,
 176         /* BTF support */
 177         FEAT_BTF,
 178         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 179         FEAT_BTF_FUNC,
 180         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 181         FEAT_BTF_DATASEC,
 182         /* BTF_FUNC_GLOBAL is supported */
 183         FEAT_BTF_GLOBAL_FUNC,
 184         /* BPF_F_MMAPABLE is supported for arrays */
 185         FEAT_ARRAY_MMAP,
 186         /* kernel support for expected_attach_type in BPF_PROG_LOAD */
 187         FEAT_EXP_ATTACH_TYPE,
 188         /* bpf_probe_read_{kernel,user}[_str] helpers */
 189         FEAT_PROBE_READ_KERN,
 190         /* BPF_PROG_BIND_MAP is supported */
 191         FEAT_PROG_BIND_MAP,
 192         /* Kernel support for module BTFs */
 193         FEAT_MODULE_BTF,
 194         /* BTF_KIND_FLOAT support */
 195         FEAT_BTF_FLOAT,
 196         /* BPF perf link support */
 197         FEAT_PERF_LINK,
 198         __FEAT_CNT,
 199 };
 200
 201 static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id);
 202
 203 enum reloc_type {
 204         RELO_LD64,
 205         RELO_CALL,
 206         RELO_DATA,
 207         RELO_EXTERN_VAR,
 208         RELO_EXTERN_FUNC,
 209         RELO_SUBPROG_ADDR,
 210 };
 211
 212 struct reloc_desc {
 213         enum reloc_type type;
 214         int insn_idx;
 215         int map_idx;
 216         int sym_off;
 217 };
 218
 219 struct bpf_sec_def;
 220
 221 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
 222                                         struct bpf_program *prog);
 223
 224 struct bpf_sec_def {
 225         const char *sec;
 226         size_t len;
 227         enum bpf_prog_type prog_type;
 228         enum bpf_attach_type expected_attach_type;
 229         bool is_exp_attach_type_optional;
 230         bool is_attachable;
 231         bool is_attach_btf;
 232         bool is_sleepable;
 233         attach_fn_t attach_fn;
 234 };
 235
 236 /*
 237  * bpf_prog should be a better name but it has been used in
 238  * linux/filter.h.
 239  */
 240 struct bpf_program {
 241         const struct bpf_sec_def *sec_def;
 242         char *sec_name;
 243         size_t sec_idx;
 244         /* this program's instruction offset (in number of instructions)
 245          * within its containing ELF section
 246          */
 247         size_t sec_insn_off;
 248         /* number of original instructions in ELF section belonging to this
 249          * program, not taking into account subprogram instructions possible
 250          * appended later during relocation
 251          */
 252         size_t sec_insn_cnt;
 253         /* Offset (in number of instructions) of the start of instruction
 254          * belonging to this BPF program  within its containing main BPF
 255          * program. For the entry-point (main) BPF program, this is always
 256          * zero. For a sub-program, this gets reset before each of main BPF
 257          * programs are processed and relocated and is used to determined
 258          * whether sub-program was already appended to the main program, and
 259          * if yes, at which instruction offset.
 260          */
 261         size_t sub_insn_off;
 262
 263         char *name;
 264         /* sec_name with / replaced by _; makes recursive pinning
 265          * in bpf_object__pin_programs easier
 266          */
 267         char *pin_name;
 268
 269         /* instructions that belong to BPF program; insns[0] is located at
 270          * sec_insn_off instruction within its ELF section in ELF file, so
 271          * when mapping ELF file instruction index to the local instruction,
 272          * one needs to subtract sec_insn_off; and vice versa.
 273          */
 274         struct bpf_insn *insns;
 275         /* actual number of instruction in this BPF program's image; for
 276          * entry-point BPF programs this includes the size of main program
 277          * itself plus all the used sub-programs, appended at the end
 278          */
 279         size_t insns_cnt;
 280
 281         struct reloc_desc *reloc_desc;
 282         int nr_reloc;
 283         int log_level;
 284
 285         struct {
 286                 int nr;
 287                 int *fds;
 288         } instances;
 289         bpf_program_prep_t preprocessor;
 290
 291         struct bpf_object *obj;
 292         void *priv;
 293         bpf_program_clear_priv_t clear_priv;
 294
 295         bool load;
 296         bool mark_btf_static;
 297         enum bpf_prog_type type;
 298         enum bpf_attach_type expected_attach_type;
 299         int prog_ifindex;
 300         __u32 attach_btf_obj_fd;
 301         __u32 attach_btf_id;
 302         __u32 attach_prog_fd;
 303         void *func_info;
 304         __u32 func_info_rec_size;
 305         __u32 func_info_cnt;
 306
 307         void *line_info;
 308         __u32 line_info_rec_size;
 309         __u32 line_info_cnt;
 310         __u32 prog_flags;
 311 };
 312
 313 struct bpf_struct_ops {
 314         const char *tname;
 315         const struct btf_type *type;
 316         struct bpf_program **progs;
 317         __u32 *kern_func_off;
 318         /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
 319         void *data;
 320         /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
 321          *      btf_vmlinux's format.
 322          * struct bpf_struct_ops_tcp_congestion_ops {
 323          *      [... some other kernel fields ...]
 324          *      struct tcp_congestion_ops data;
 325          * }
 326          * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
 327          * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
 328          * from "data".
 329          */
 330         void *kern_vdata;
 331         __u32 type_id;
 332 };
 333
 334 #define DATA_SEC ".data"
 335 #define BSS_SEC ".bss"
 336 #define RODATA_SEC ".rodata"
 337 #define KCONFIG_SEC ".kconfig"
 338 #define KSYMS_SEC ".ksyms"
 339 #define STRUCT_OPS_SEC ".struct_ops"
 340
 341 enum libbpf_map_type {
 342         LIBBPF_MAP_UNSPEC,
 343         LIBBPF_MAP_DATA,
 344         LIBBPF_MAP_BSS,
 345         LIBBPF_MAP_RODATA,
 346         LIBBPF_MAP_KCONFIG,
 347 };
 348
 349 static const char * const libbpf_type_to_btf_name[] = {
 350         [LIBBPF_MAP_DATA]       = DATA_SEC,
 351         [LIBBPF_MAP_BSS]        = BSS_SEC,
 352         [LIBBPF_MAP_RODATA]     = RODATA_SEC,
 353         [LIBBPF_MAP_KCONFIG]    = KCONFIG_SEC,
 354 };
 355
 356 struct bpf_map {
 357         char *name;
 358         int fd;
 359         int sec_idx;
 360         size_t sec_offset;
 361         int map_ifindex;
 362         int inner_map_fd;
 363         struct bpf_map_def def;
 364         __u32 numa_node;
 365         __u32 btf_var_idx;
 366         __u32 btf_key_type_id;
 367         __u32 btf_value_type_id;
 368         __u32 btf_vmlinux_value_type_id;
 369         void *priv;
 370         bpf_map_clear_priv_t clear_priv;
 371         enum libbpf_map_type libbpf_type;
 372         void *mmaped;
 373         struct bpf_struct_ops *st_ops;
 374         struct bpf_map *inner_map;
 375         void **init_slots;
 376         int init_slots_sz;
 377         char *pin_path;
 378         bool pinned;
 379         bool reused;
 380 };
 381
 382 enum extern_type {
 383         EXT_UNKNOWN,
 384         EXT_KCFG,
 385         EXT_KSYM,
 386 };
 387
 388 enum kcfg_type {
 389         KCFG_UNKNOWN,
 390         KCFG_CHAR,
 391         KCFG_BOOL,
 392         KCFG_INT,
 393         KCFG_TRISTATE,
 394         KCFG_CHAR_ARR,
 395 };
 396
 397 struct extern_desc {
 398         enum extern_type type;
 399         int sym_idx;
 400         int btf_id;
 401         int sec_btf_id;
 402         const char *name;
 403         bool is_set;
 404         bool is_weak;
 405         union {
 406                 struct {
 407                         enum kcfg_type type;
 408                         int sz;
 409                         int align;
 410                         int data_off;
 411                         bool is_signed;
 412                 } kcfg;
 413                 struct {
 414                         unsigned long long addr;
 415
 416                         /* target btf_id of the corresponding kernel var. */
 417                         int kernel_btf_obj_fd;
 418                         int kernel_btf_id;
 419
 420                         /* local btf_id of the ksym extern's type. */
 421                         __u32 type_id;
 422                 } ksym;
 423         };
 424 };
 425
 426 static LIST_HEAD(bpf_objects_list);
 427
 428 struct module_btf {
 429         struct btf *btf;
 430         char *name;
 431         __u32 id;
 432         int fd;
 433 };
 434
 435 struct bpf_object {
 436         char name[BPF_OBJ_NAME_LEN];
 437         char license[64];
 438         __u32 kern_version;
 439
 440         struct bpf_program *programs;
 441         size_t nr_programs;
 442         struct bpf_map *maps;
 443         size_t nr_maps;
 444         size_t maps_cap;
 445
 446         char *kconfig;
 447         struct extern_desc *externs;
 448         int nr_extern;
 449         int kconfig_map_idx;
 450         int rodata_map_idx;
 451
 452         bool loaded;
 453         bool has_subcalls;
 454
 455         struct bpf_gen *gen_loader;
 456
 457         /*
 458          * Information when doing elf related work. Only valid if fd
 459          * is valid.
 460          */
 461         struct {
 462                 int fd;
 463                 const void *obj_buf;
 464                 size_t obj_buf_sz;
 465                 Elf *elf;
 466                 GElf_Ehdr ehdr;
 467                 Elf_Data *symbols;
 468                 Elf_Data *data;
 469                 Elf_Data *rodata;
 470                 Elf_Data *bss;
 471                 Elf_Data *st_ops_data;
 472                 size_t shstrndx; /* section index for section name strings */
 473                 size_t strtabidx;
 474                 struct {
 475                         GElf_Shdr shdr;
 476                         Elf_Data *data;
 477                 } *reloc_sects;
 478                 int nr_reloc_sects;
 479                 int maps_shndx;
 480                 int btf_maps_shndx;
 481                 __u32 btf_maps_sec_btf_id;
 482                 int text_shndx;
 483                 int symbols_shndx;
 484                 int data_shndx;
 485                 int rodata_shndx;
 486                 int bss_shndx;
 487                 int st_ops_shndx;
 488         } efile;
 489         /*
 490          * All loaded bpf_object is linked in a list, which is
 491          * hidden to caller. bpf_objects__<func> handlers deal with
 492          * all objects.
 493          */
 494         struct list_head list;
 495
 496         struct btf *btf;
 497         struct btf_ext *btf_ext;
 498
 499         /* Parse and load BTF vmlinux if any of the programs in the object need
 500          * it at load time.
 501          */
 502         struct btf *btf_vmlinux;
 503         /* Path to the custom BTF to be used for BPF CO-RE relocations as an
 504          * override for vmlinux BTF.
 505          */
 506         char *btf_custom_path;
 507         /* vmlinux BTF override for CO-RE relocations */
 508         struct btf *btf_vmlinux_override;
 509         /* Lazily initialized kernel module BTFs */
 510         struct module_btf *btf_modules;
 511         bool btf_modules_loaded;
 512         size_t btf_module_cnt;
 513         size_t btf_module_cap;
 514
 515         void *priv;
 516         bpf_object_clear_priv_t clear_priv;
 517
 518         char path[];
 519 };
 520 #define obj_elf_valid(o)        ((o)->efile.elf)
 521
 522 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
 523 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
 524 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
 525 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
 526 static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr);
 527 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
 528 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
 529
 530 void bpf_program__unload(struct bpf_program *prog)
 531 {
 532         int i;
 533
 534         if (!prog)
 535                 return;
 536
 537         /*
 538          * If the object is opened but the program was never loaded,
 539          * it is possible that prog->instances.nr == -1.
 540          */
 541         if (prog->instances.nr > 0) {
 542                 for (i = 0; i < prog->instances.nr; i++)
 543                         zclose(prog->instances.fds[i]);
 544         } else if (prog->instances.nr != -1) {
 545                 pr_warn("Internal error: instances.nr is %d\n",
 546                         prog->instances.nr);
 547         }
 548
 549         prog->instances.nr = -1;
 550         zfree(&prog->instances.fds);
 551
 552         zfree(&prog->func_info);
 553         zfree(&prog->line_info);
 554 }
 555
 556 static void bpf_program__exit(struct bpf_program *prog)
 557 {
 558         if (!prog)
 559                 return;
 560
 561         if (prog->clear_priv)
 562                 prog->clear_priv(prog, prog->priv);
 563
 564         prog->priv = NULL;
 565         prog->clear_priv = NULL;
 566
 567         bpf_program__unload(prog);
 568         zfree(&prog->name);
 569         zfree(&prog->sec_name);
 570         zfree(&prog->pin_name);
 571         zfree(&prog->insns);
 572         zfree(&prog->reloc_desc);
 573
 574         prog->nr_reloc = 0;
 575         prog->insns_cnt = 0;
 576         prog->sec_idx = -1;
 577 }
 578
 579 static char *__bpf_program__pin_name(struct bpf_program *prog)
 580 {
 581         char *name, *p;
 582
 583         name = p = strdup(prog->sec_name);
 584         while ((p = strchr(p, '/')))
 585                 *p = '_';
 586
 587         return name;
 588 }
 589
 590 static bool insn_is_subprog_call(const struct bpf_insn *insn)
 591 {
 592         return BPF_CLASS(insn->code) == BPF_JMP &&
 593                BPF_OP(insn->code) == BPF_CALL &&
 594                BPF_SRC(insn->code) == BPF_K &&
 595                insn->src_reg == BPF_PSEUDO_CALL &&
 596                insn->dst_reg == 0 &&
 597                insn->off == 0;
 598 }
 599
 600 static bool is_call_insn(const struct bpf_insn *insn)
 601 {
 602         return insn->code == (BPF_JMP | BPF_CALL);
 603 }
 604
 605 static bool insn_is_pseudo_func(struct bpf_insn *insn)
 606 {
 607         return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
 608 }
 609
 610 static int
 611 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
 612                       const char *name, size_t sec_idx, const char *sec_name,
 613                       size_t sec_off, void *insn_data, size_t insn_data_sz)
 614 {
 615         if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
 616                 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
 617                         sec_name, name, sec_off, insn_data_sz);
 618                 return -EINVAL;
 619         }
 620
 621         memset(prog, 0, sizeof(*prog));
 622         prog->obj = obj;
 623
 624         prog->sec_idx = sec_idx;
 625         prog->sec_insn_off = sec_off / BPF_INSN_SZ;
 626         prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
 627         /* insns_cnt can later be increased by appending used subprograms */
 628         prog->insns_cnt = prog->sec_insn_cnt;
 629
 630         prog->type = BPF_PROG_TYPE_UNSPEC;
 631         prog->load = true;
 632
 633         prog->instances.fds = NULL;
 634         prog->instances.nr = -1;
 635
 636         prog->sec_name = strdup(sec_name);
 637         if (!prog->sec_name)
 638                 goto errout;
 639
 640         prog->name = strdup(name);
 641         if (!prog->name)
 642                 goto errout;
 643
 644         prog->pin_name = __bpf_program__pin_name(prog);
 645         if (!prog->pin_name)
 646                 goto errout;
 647
 648         prog->insns = malloc(insn_data_sz);
 649         if (!prog->insns)
 650                 goto errout;
 651         memcpy(prog->insns, insn_data, insn_data_sz);
 652
 653         return 0;
 654 errout:
 655         pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
 656         bpf_program__exit(prog);
 657         return -ENOMEM;
 658 }
 659
 660 static int
 661 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
 662                          const char *sec_name, int sec_idx)
 663 {
 664         Elf_Data *symbols = obj->efile.symbols;
 665         struct bpf_program *prog, *progs;
 666         void *data = sec_data->d_buf;
 667         size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
 668         int nr_progs, err, i;
 669         const char *name;
 670         GElf_Sym sym;
 671
 672         progs = obj->programs;
 673         nr_progs = obj->nr_programs;
 674         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 675         sec_off = 0;
 676
 677         for (i = 0; i < nr_syms; i++) {
 678                 if (!gelf_getsym(symbols, i, &sym))
 679                         continue;
 680                 if (sym.st_shndx != sec_idx)
 681                         continue;
 682                 if (GELF_ST_TYPE(sym.st_info) != STT_FUNC)
 683                         continue;
 684
 685                 prog_sz = sym.st_size;
 686                 sec_off = sym.st_value;
 687
 688                 name = elf_sym_str(obj, sym.st_name);
 689                 if (!name) {
 690                         pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
 691                                 sec_name, sec_off);
 692                         return -LIBBPF_ERRNO__FORMAT;
 693                 }
 694
 695                 if (sec_off + prog_sz > sec_sz) {
 696                         pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
 697                                 sec_name, sec_off);
 698                         return -LIBBPF_ERRNO__FORMAT;
 699                 }
 700
 701                 if (sec_idx != obj->efile.text_shndx && GELF_ST_BIND(sym.st_info) == STB_LOCAL) {
 702                         pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
 703                         return -ENOTSUP;
 704                 }
 705
 706                 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
 707                          sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
 708
 709                 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
 710                 if (!progs) {
 711                         /*
 712                          * In this case the original obj->programs
 713                          * is still valid, so don't need special treat for
 714                          * bpf_close_object().
 715                          */
 716                         pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
 717                                 sec_name, name);
 718                         return -ENOMEM;
 719                 }
 720                 obj->programs = progs;
 721
 722                 prog = &progs[nr_progs];
 723
 724                 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
 725                                             sec_off, data + sec_off, prog_sz);
 726                 if (err)
 727                         return err;
 728
 729                 /* if function is a global/weak symbol, but has restricted
 730                  * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
 731                  * as static to enable more permissive BPF verification mode
 732                  * with more outside context available to BPF verifier
 733                  */
 734                 if (GELF_ST_BIND(sym.st_info) != STB_LOCAL
 735                     && (GELF_ST_VISIBILITY(sym.st_other) == STV_HIDDEN
 736                         || GELF_ST_VISIBILITY(sym.st_other) == STV_INTERNAL))
 737                         prog->mark_btf_static = true;
 738
 739                 nr_progs++;
 740                 obj->nr_programs = nr_progs;
 741         }
 742
 743         return 0;
 744 }
 745
 746 static __u32 get_kernel_version(void)
 747 {
 748         __u32 major, minor, patch;
 749         struct utsname info;
 750
 751         uname(&info);
 752         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 753                 return 0;
 754         return KERNEL_VERSION(major, minor, patch);
 755 }
 756
 757 static const struct btf_member *
 758 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
 759 {
 760         struct btf_member *m;
 761         int i;
 762
 763         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 764                 if (btf_member_bit_offset(t, i) == bit_offset)
 765                         return m;
 766         }
 767
 768         return NULL;
 769 }
 770
 771 static const struct btf_member *
 772 find_member_by_name(const struct btf *btf, const struct btf_type *t,
 773                     const char *name)
 774 {
 775         struct btf_member *m;
 776         int i;
 777
 778         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 779                 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
 780                         return m;
 781         }
 782
 783         return NULL;
 784 }
 785
 786 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
 787 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
 788                                    const char *name, __u32 kind);
 789
 790 static int
 791 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
 792                            const struct btf_type **type, __u32 *type_id,
 793                            const struct btf_type **vtype, __u32 *vtype_id,
 794                            const struct btf_member **data_member)
 795 {
 796         const struct btf_type *kern_type, *kern_vtype;
 797         const struct btf_member *kern_data_member;
 798         __s32 kern_vtype_id, kern_type_id;
 799         __u32 i;
 800
 801         kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
 802         if (kern_type_id < 0) {
 803                 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
 804                         tname);
 805                 return kern_type_id;
 806         }
 807         kern_type = btf__type_by_id(btf, kern_type_id);
 808
 809         /* Find the corresponding "map_value" type that will be used
 810          * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 811          * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 812          * btf_vmlinux.
 813          */
 814         kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
 815                                                 tname, BTF_KIND_STRUCT);
 816         if (kern_vtype_id < 0) {
 817                 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
 818                         STRUCT_OPS_VALUE_PREFIX, tname);
 819                 return kern_vtype_id;
 820         }
 821         kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 822
 823         /* Find "struct tcp_congestion_ops" from
 824          * struct bpf_struct_ops_tcp_congestion_ops {
 825          *      [ ... ]
 826          *      struct tcp_congestion_ops data;
 827          * }
 828          */
 829         kern_data_member = btf_members(kern_vtype);
 830         for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
 831                 if (kern_data_member->type == kern_type_id)
 832                         break;
 833         }
 834         if (i == btf_vlen(kern_vtype)) {
 835                 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
 836                         tname, STRUCT_OPS_VALUE_PREFIX, tname);
 837                 return -EINVAL;
 838         }
 839
 840         *type = kern_type;
 841         *type_id = kern_type_id;
 842         *vtype = kern_vtype;
 843         *vtype_id = kern_vtype_id;
 844         *data_member = kern_data_member;
 845
 846         return 0;
 847 }
 848
 849 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
 850 {
 851         return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
 852 }
 853
 854 /* Init the map's fields that depend on kern_btf */
 855 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
 856                                          const struct btf *btf,
 857                                          const struct btf *kern_btf)
 858 {
 859         const struct btf_member *member, *kern_member, *kern_data_member;
 860         const struct btf_type *type, *kern_type, *kern_vtype;
 861         __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
 862         struct bpf_struct_ops *st_ops;
 863         void *data, *kern_data;
 864         const char *tname;
 865         int err;
 866
 867         st_ops = map->st_ops;
 868         type = st_ops->type;
 869         tname = st_ops->tname;
 870         err = find_struct_ops_kern_types(kern_btf, tname,
 871                                          &kern_type, &kern_type_id,
 872                                          &kern_vtype, &kern_vtype_id,
 873                                          &kern_data_member);
 874         if (err)
 875                 return err;
 876
 877         pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
 878                  map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
 879
 880         map->def.value_size = kern_vtype->size;
 881         map->btf_vmlinux_value_type_id = kern_vtype_id;
 882
 883         st_ops->kern_vdata = calloc(1, kern_vtype->size);
 884         if (!st_ops->kern_vdata)
 885                 return -ENOMEM;
 886
 887         data = st_ops->data;
 888         kern_data_off = kern_data_member->offset / 8;
 889         kern_data = st_ops->kern_vdata + kern_data_off;
 890
 891         member = btf_members(type);
 892         for (i = 0; i < btf_vlen(type); i++, member++) {
 893                 const struct btf_type *mtype, *kern_mtype;
 894                 __u32 mtype_id, kern_mtype_id;
 895                 void *mdata, *kern_mdata;
 896                 __s64 msize, kern_msize;
 897                 __u32 moff, kern_moff;
 898                 __u32 kern_member_idx;
 899                 const char *mname;
 900
 901                 mname = btf__name_by_offset(btf, member->name_off);
 902                 kern_member = find_member_by_name(kern_btf, kern_type, mname);
 903                 if (!kern_member) {
 904                         pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
 905                                 map->name, mname);
 906                         return -ENOTSUP;
 907                 }
 908
 909                 kern_member_idx = kern_member - btf_members(kern_type);
 910                 if (btf_member_bitfield_size(type, i) ||
 911                     btf_member_bitfield_size(kern_type, kern_member_idx)) {
 912                         pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
 913                                 map->name, mname);
 914                         return -ENOTSUP;
 915                 }
 916
 917                 moff = member->offset / 8;
 918                 kern_moff = kern_member->offset / 8;
 919
 920                 mdata = data + moff;
 921                 kern_mdata = kern_data + kern_moff;
 922
 923                 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
 924                 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
 925                                                     &kern_mtype_id);
 926                 if (BTF_INFO_KIND(mtype->info) !=
 927                     BTF_INFO_KIND(kern_mtype->info)) {
 928                         pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
 929                                 map->name, mname, BTF_INFO_KIND(mtype->info),
 930                                 BTF_INFO_KIND(kern_mtype->info));
 931                         return -ENOTSUP;
 932                 }
 933
 934                 if (btf_is_ptr(mtype)) {
 935                         struct bpf_program *prog;
 936
 937                         prog = st_ops->progs[i];
 938                         if (!prog)
 939                                 continue;
 940
 941                         kern_mtype = skip_mods_and_typedefs(kern_btf,
 942                                                             kern_mtype->type,
 943                                                             &kern_mtype_id);
 944
 945                         /* mtype->type must be a func_proto which was
 946                          * guaranteed in bpf_object__collect_st_ops_relos(),
 947                          * so only check kern_mtype for func_proto here.
 948                          */
 949                         if (!btf_is_func_proto(kern_mtype)) {
 950                                 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
 951                                         map->name, mname);
 952                                 return -ENOTSUP;
 953                         }
 954
 955                         prog->attach_btf_id = kern_type_id;
 956                         prog->expected_attach_type = kern_member_idx;
 957
 958                         st_ops->kern_func_off[i] = kern_data_off + kern_moff;
 959
 960                         pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
 961                                  map->name, mname, prog->name, moff,
 962                                  kern_moff);
 963
 964                         continue;
 965                 }
 966
 967                 msize = btf__resolve_size(btf, mtype_id);
 968                 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
 969                 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
 970                         pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
 971                                 map->name, mname, (ssize_t)msize,
 972                                 (ssize_t)kern_msize);
 973                         return -ENOTSUP;
 974                 }
 975
 976                 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
 977                          map->name, mname, (unsigned int)msize,
 978                          moff, kern_moff);
 979                 memcpy(kern_mdata, mdata, msize);
 980         }
 981
 982         return 0;
 983 }
 984
 985 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
 986 {
 987         struct bpf_map *map;
 988         size_t i;
 989         int err;
 990
 991         for (i = 0; i < obj->nr_maps; i++) {
 992                 map = &obj->maps[i];
 993
 994                 if (!bpf_map__is_struct_ops(map))
 995                         continue;
 996
 997                 err = bpf_map__init_kern_struct_ops(map, obj->btf,
 998                                                     obj->btf_vmlinux);
 999                 if (err)
1000                         return err;
1001         }
1002
1003         return 0;
1004 }
1005
1006 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
1007 {
1008         const struct btf_type *type, *datasec;
1009         const struct btf_var_secinfo *vsi;
1010         struct bpf_struct_ops *st_ops;
1011         const char *tname, *var_name;
1012         __s32 type_id, datasec_id;
1013         const struct btf *btf;
1014         struct bpf_map *map;
1015         __u32 i;
1016
1017         if (obj->efile.st_ops_shndx == -1)
1018                 return 0;
1019
1020         btf = obj->btf;
1021         datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
1022                                             BTF_KIND_DATASEC);
1023         if (datasec_id < 0) {
1024                 pr_warn("struct_ops init: DATASEC %s not found\n",
1025                         STRUCT_OPS_SEC);
1026                 return -EINVAL;
1027         }
1028
1029         datasec = btf__type_by_id(btf, datasec_id);
1030         vsi = btf_var_secinfos(datasec);
1031         for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1032                 type = btf__type_by_id(obj->btf, vsi->type);
1033                 var_name = btf__name_by_offset(obj->btf, type->name_off);
1034
1035                 type_id = btf__resolve_type(obj->btf, vsi->type);
1036                 if (type_id < 0) {
1037                         pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1038                                 vsi->type, STRUCT_OPS_SEC);
1039                         return -EINVAL;
1040                 }
1041
1042                 type = btf__type_by_id(obj->btf, type_id);
1043                 tname = btf__name_by_offset(obj->btf, type->name_off);
1044                 if (!tname[0]) {
1045                         pr_warn("struct_ops init: anonymous type is not supported\n");
1046                         return -ENOTSUP;
1047                 }
1048                 if (!btf_is_struct(type)) {
1049                         pr_warn("struct_ops init: %s is not a struct\n", tname);
1050                         return -EINVAL;
1051                 }
1052
1053                 map = bpf_object__add_map(obj);
1054                 if (IS_ERR(map))
1055                         return PTR_ERR(map);
1056
1057                 map->sec_idx = obj->efile.st_ops_shndx;
1058                 map->sec_offset = vsi->offset;
1059                 map->name = strdup(var_name);
1060                 if (!map->name)
1061                         return -ENOMEM;
1062
1063                 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1064                 map->def.key_size = sizeof(int);
1065                 map->def.value_size = type->size;
1066                 map->def.max_entries = 1;
1067
1068                 map->st_ops = calloc(1, sizeof(*map->st_ops));
1069                 if (!map->st_ops)
1070                         return -ENOMEM;
1071                 st_ops = map->st_ops;
1072                 st_ops->data = malloc(type->size);
1073                 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1074                 st_ops->kern_func_off = malloc(btf_vlen(type) *
1075                                                sizeof(*st_ops->kern_func_off));
1076                 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1077                         return -ENOMEM;
1078
1079                 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
1080                         pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1081                                 var_name, STRUCT_OPS_SEC);
1082                         return -EINVAL;
1083                 }
1084
1085                 memcpy(st_ops->data,
1086                        obj->efile.st_ops_data->d_buf + vsi->offset,
1087                        type->size);
1088                 st_ops->tname = tname;
1089                 st_ops->type = type;
1090                 st_ops->type_id = type_id;
1091
1092                 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1093                          tname, type_id, var_name, vsi->offset);
1094         }
1095
1096         return 0;
1097 }
1098
1099 static struct bpf_object *bpf_object__new(const char *path,
1100                                           const void *obj_buf,
1101                                           size_t obj_buf_sz,
1102                                           const char *obj_name)
1103 {
1104         struct bpf_object *obj;
1105         char *end;
1106
1107         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1108         if (!obj) {
1109                 pr_warn("alloc memory failed for %s\n", path);
1110                 return ERR_PTR(-ENOMEM);
1111         }
1112
1113         strcpy(obj->path, path);
1114         if (obj_name) {
1115                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
1116                 obj->name[sizeof(obj->name) - 1] = 0;
1117         } else {
1118                 /* Using basename() GNU version which doesn't modify arg. */
1119                 strncpy(obj->name, basename((void *)path),
1120                         sizeof(obj->name) - 1);
1121                 end = strchr(obj->name, '.');
1122                 if (end)
1123                         *end = 0;
1124         }
1125
1126         obj->efile.fd = -1;
1127         /*
1128          * Caller of this function should also call
1129          * bpf_object__elf_finish() after data collection to return
1130          * obj_buf to user. If not, we should duplicate the buffer to
1131          * avoid user freeing them before elf finish.
1132          */
1133         obj->efile.obj_buf = obj_buf;
1134         obj->efile.obj_buf_sz = obj_buf_sz;
1135         obj->efile.maps_shndx = -1;
1136         obj->efile.btf_maps_shndx = -1;
1137         obj->efile.data_shndx = -1;
1138         obj->efile.rodata_shndx = -1;
1139         obj->efile.bss_shndx = -1;
1140         obj->efile.st_ops_shndx = -1;
1141         obj->kconfig_map_idx = -1;
1142         obj->rodata_map_idx = -1;
1143
1144         obj->kern_version = get_kernel_version();
1145         obj->loaded = false;
1146
1147         INIT_LIST_HEAD(&obj->list);
1148         list_add(&obj->list, &bpf_objects_list);
1149         return obj;
1150 }
1151
1152 static void bpf_object__elf_finish(struct bpf_object *obj)
1153 {
1154         if (!obj_elf_valid(obj))
1155                 return;
1156
1157         if (obj->efile.elf) {
1158                 elf_end(obj->efile.elf);
1159                 obj->efile.elf = NULL;
1160         }
1161         obj->efile.symbols = NULL;
1162         obj->efile.data = NULL;
1163         obj->efile.rodata = NULL;
1164         obj->efile.bss = NULL;
1165         obj->efile.st_ops_data = NULL;
1166
1167         zfree(&obj->efile.reloc_sects);
1168         obj->efile.nr_reloc_sects = 0;
1169         zclose(obj->efile.fd);
1170         obj->efile.obj_buf = NULL;
1171         obj->efile.obj_buf_sz = 0;
1172 }
1173
1174 static int bpf_object__elf_init(struct bpf_object *obj)
1175 {
1176         int err = 0;
1177         GElf_Ehdr *ep;
1178
1179         if (obj_elf_valid(obj)) {
1180                 pr_warn("elf: init internal error\n");
1181                 return -LIBBPF_ERRNO__LIBELF;
1182         }
1183
1184         if (obj->efile.obj_buf_sz > 0) {
1185                 /*
1186                  * obj_buf should have been validated by
1187                  * bpf_object__open_buffer().
1188                  */
1189                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
1190                                             obj->efile.obj_buf_sz);
1191         } else {
1192                 obj->efile.fd = open(obj->path, O_RDONLY);
1193                 if (obj->efile.fd < 0) {
1194                         char errmsg[STRERR_BUFSIZE], *cp;
1195
1196                         err = -errno;
1197                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1198                         pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1199                         return err;
1200                 }
1201
1202                 obj->efile.elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1203         }
1204
1205         if (!obj->efile.elf) {
1206                 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1207                 err = -LIBBPF_ERRNO__LIBELF;
1208                 goto errout;
1209         }
1210
1211         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
1212                 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1213                 err = -LIBBPF_ERRNO__FORMAT;
1214                 goto errout;
1215         }
1216         ep = &obj->efile.ehdr;
1217
1218         if (elf_getshdrstrndx(obj->efile.elf, &obj->efile.shstrndx)) {
1219                 pr_warn("elf: failed to get section names section index for %s: %s\n",
1220                         obj->path, elf_errmsg(-1));
1221                 err = -LIBBPF_ERRNO__FORMAT;
1222                 goto errout;
1223         }
1224
1225         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1226         if (!elf_rawdata(elf_getscn(obj->efile.elf, obj->efile.shstrndx), NULL)) {
1227                 pr_warn("elf: failed to get section names strings from %s: %s\n",
1228                         obj->path, elf_errmsg(-1));
1229                 err = -LIBBPF_ERRNO__FORMAT;
1230                 goto errout;
1231         }
1232
1233         /* Old LLVM set e_machine to EM_NONE */
1234         if (ep->e_type != ET_REL ||
1235             (ep->e_machine && ep->e_machine != EM_BPF)) {
1236                 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1237                 err = -LIBBPF_ERRNO__FORMAT;
1238                 goto errout;
1239         }
1240
1241         return 0;
1242 errout:
1243         bpf_object__elf_finish(obj);
1244         return err;
1245 }
1246
1247 static int bpf_object__check_endianness(struct bpf_object *obj)
1248 {
1249 #if __BYTE_ORDER == __LITTLE_ENDIAN
1250         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
1251                 return 0;
1252 #elif __BYTE_ORDER == __BIG_ENDIAN
1253         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
1254                 return 0;
1255 #else
1256 # error "Unrecognized __BYTE_ORDER__"
1257 #endif
1258         pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1259         return -LIBBPF_ERRNO__ENDIAN;
1260 }
1261
1262 static int
1263 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1264 {
1265         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
1266         pr_debug("license of %s is %s\n", obj->path, obj->license);
1267         return 0;
1268 }
1269
1270 static int
1271 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1272 {
1273         __u32 kver;
1274
1275         if (size != sizeof(kver)) {
1276                 pr_warn("invalid kver section in %s\n", obj->path);
1277                 return -LIBBPF_ERRNO__FORMAT;
1278         }
1279         memcpy(&kver, data, sizeof(kver));
1280         obj->kern_version = kver;
1281         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1282         return 0;
1283 }
1284
1285 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1286 {
1287         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1288             type == BPF_MAP_TYPE_HASH_OF_MAPS)
1289                 return true;
1290         return false;
1291 }
1292
1293 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
1294                              __u32 *size)
1295 {
1296         int ret = -ENOENT;
1297
1298         *size = 0;
1299         if (!name) {
1300                 return -EINVAL;
1301         } else if (!strcmp(name, DATA_SEC)) {
1302                 if (obj->efile.data)
1303                         *size = obj->efile.data->d_size;
1304         } else if (!strcmp(name, BSS_SEC)) {
1305                 if (obj->efile.bss)
1306                         *size = obj->efile.bss->d_size;
1307         } else if (!strcmp(name, RODATA_SEC)) {
1308                 if (obj->efile.rodata)
1309                         *size = obj->efile.rodata->d_size;
1310         } else if (!strcmp(name, STRUCT_OPS_SEC)) {
1311                 if (obj->efile.st_ops_data)
1312                         *size = obj->efile.st_ops_data->d_size;
1313         } else {
1314                 Elf_Scn *scn = elf_sec_by_name(obj, name);
1315                 Elf_Data *data = elf_sec_data(obj, scn);
1316
1317                 if (data) {
1318                         ret = 0; /* found it */
1319                         *size = data->d_size;
1320                 }
1321         }
1322
1323         return *size ? 0 : ret;
1324 }
1325
1326 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
1327                                 __u32 *off)
1328 {
1329         Elf_Data *symbols = obj->efile.symbols;
1330         const char *sname;
1331         size_t si;
1332
1333         if (!name || !off)
1334                 return -EINVAL;
1335
1336         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
1337                 GElf_Sym sym;
1338
1339                 if (!gelf_getsym(symbols, si, &sym))
1340                         continue;
1341                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1342                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
1343                         continue;
1344
1345                 sname = elf_sym_str(obj, sym.st_name);
1346                 if (!sname) {
1347                         pr_warn("failed to get sym name string for var %s\n",
1348                                 name);
1349                         return -EIO;
1350                 }
1351                 if (strcmp(name, sname) == 0) {
1352                         *off = sym.st_value;
1353                         return 0;
1354                 }
1355         }
1356
1357         return -ENOENT;
1358 }
1359
1360 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1361 {
1362         struct bpf_map *new_maps;
1363         size_t new_cap;
1364         int i;
1365
1366         if (obj->nr_maps < obj->maps_cap)
1367                 return &obj->maps[obj->nr_maps++];
1368
1369         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1370         new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps));
1371         if (!new_maps) {
1372                 pr_warn("alloc maps for object failed\n");
1373                 return ERR_PTR(-ENOMEM);
1374         }
1375
1376         obj->maps_cap = new_cap;
1377         obj->maps = new_maps;
1378
1379         /* zero out new maps */
1380         memset(obj->maps + obj->nr_maps, 0,
1381                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1382         /*
1383          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1384          * when failure (zclose won't close negative fd)).
1385          */
1386         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1387                 obj->maps[i].fd = -1;
1388                 obj->maps[i].inner_map_fd = -1;
1389         }
1390
1391         return &obj->maps[obj->nr_maps++];
1392 }
1393
1394 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1395 {
1396         long page_sz = sysconf(_SC_PAGE_SIZE);
1397         size_t map_sz;
1398
1399         map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1400         map_sz = roundup(map_sz, page_sz);
1401         return map_sz;
1402 }
1403
1404 static char *internal_map_name(struct bpf_object *obj,
1405                                enum libbpf_map_type type)
1406 {
1407         char map_name[BPF_OBJ_NAME_LEN], *p;
1408         const char *sfx = libbpf_type_to_btf_name[type];
1409         int sfx_len = max((size_t)7, strlen(sfx));
1410         int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
1411                           strlen(obj->name));
1412
1413         snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1414                  sfx_len, libbpf_type_to_btf_name[type]);
1415
1416         /* sanitise map name to characters allowed by kernel */
1417         for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1418                 if (!isalnum(*p) && *p != '_' && *p != '.')
1419                         *p = '_';
1420
1421         return strdup(map_name);
1422 }
1423
1424 static int
1425 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1426                               int sec_idx, void *data, size_t data_sz)
1427 {
1428         struct bpf_map_def *def;
1429         struct bpf_map *map;
1430         int err;
1431
1432         map = bpf_object__add_map(obj);
1433         if (IS_ERR(map))
1434                 return PTR_ERR(map);
1435
1436         map->libbpf_type = type;
1437         map->sec_idx = sec_idx;
1438         map->sec_offset = 0;
1439         map->name = internal_map_name(obj, type);
1440         if (!map->name) {
1441                 pr_warn("failed to alloc map name\n");
1442                 return -ENOMEM;
1443         }
1444
1445         def = &map->def;
1446         def->type = BPF_MAP_TYPE_ARRAY;
1447         def->key_size = sizeof(int);
1448         def->value_size = data_sz;
1449         def->max_entries = 1;
1450         def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1451                          ? BPF_F_RDONLY_PROG : 0;
1452         def->map_flags |= BPF_F_MMAPABLE;
1453
1454         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1455                  map->name, map->sec_idx, map->sec_offset, def->map_flags);
1456
1457         map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1458                            MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1459         if (map->mmaped == MAP_FAILED) {
1460                 err = -errno;
1461                 map->mmaped = NULL;
1462                 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1463                         map->name, err);
1464                 zfree(&map->name);
1465                 return err;
1466         }
1467
1468         if (data)
1469                 memcpy(map->mmaped, data, data_sz);
1470
1471         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1472         return 0;
1473 }
1474
1475 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1476 {
1477         int err;
1478
1479         /*
1480          * Populate obj->maps with libbpf internal maps.
1481          */
1482         if (obj->efile.data_shndx >= 0) {
1483                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1484                                                     obj->efile.data_shndx,
1485                                                     obj->efile.data->d_buf,
1486                                                     obj->efile.data->d_size);
1487                 if (err)
1488                         return err;
1489         }
1490         if (obj->efile.rodata_shndx >= 0) {
1491                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1492                                                     obj->efile.rodata_shndx,
1493                                                     obj->efile.rodata->d_buf,
1494                                                     obj->efile.rodata->d_size);
1495                 if (err)
1496                         return err;
1497
1498                 obj->rodata_map_idx = obj->nr_maps - 1;
1499         }
1500         if (obj->efile.bss_shndx >= 0) {
1501                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1502                                                     obj->efile.bss_shndx,
1503                                                     NULL,
1504                                                     obj->efile.bss->d_size);
1505                 if (err)
1506                         return err;
1507         }
1508         return 0;
1509 }
1510
1511
1512 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1513                                                const void *name)
1514 {
1515         int i;
1516
1517         for (i = 0; i < obj->nr_extern; i++) {
1518                 if (strcmp(obj->externs[i].name, name) == 0)
1519                         return &obj->externs[i];
1520         }
1521         return NULL;
1522 }
1523
1524 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1525                               char value)
1526 {
1527         switch (ext->kcfg.type) {
1528         case KCFG_BOOL:
1529                 if (value == 'm') {
1530                         pr_warn("extern (kcfg) %s=%c should be tristate or char\n",
1531                                 ext->name, value);
1532                         return -EINVAL;
1533                 }
1534                 *(bool *)ext_val = value == 'y' ? true : false;
1535                 break;
1536         case KCFG_TRISTATE:
1537                 if (value == 'y')
1538                         *(enum libbpf_tristate *)ext_val = TRI_YES;
1539                 else if (value == 'm')
1540                         *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1541                 else /* value == 'n' */
1542                         *(enum libbpf_tristate *)ext_val = TRI_NO;
1543                 break;
1544         case KCFG_CHAR:
1545                 *(char *)ext_val = value;
1546                 break;
1547         case KCFG_UNKNOWN:
1548         case KCFG_INT:
1549         case KCFG_CHAR_ARR:
1550         default:
1551                 pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n",
1552                         ext->name, value);
1553                 return -EINVAL;
1554         }
1555         ext->is_set = true;
1556         return 0;
1557 }
1558
1559 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1560                               const char *value)
1561 {
1562         size_t len;
1563
1564         if (ext->kcfg.type != KCFG_CHAR_ARR) {
1565                 pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value);
1566                 return -EINVAL;
1567         }
1568
1569         len = strlen(value);
1570         if (value[len - 1] != '"') {
1571                 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1572                         ext->name, value);
1573                 return -EINVAL;
1574         }
1575
1576         /* strip quotes */
1577         len -= 2;
1578         if (len >= ext->kcfg.sz) {
1579                 pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1580                         ext->name, value, len, ext->kcfg.sz - 1);
1581                 len = ext->kcfg.sz - 1;
1582         }
1583         memcpy(ext_val, value + 1, len);
1584         ext_val[len] = '\0';
1585         ext->is_set = true;
1586         return 0;
1587 }
1588
1589 static int parse_u64(const char *value, __u64 *res)
1590 {
1591         char *value_end;
1592         int err;
1593
1594         errno = 0;
1595         *res = strtoull(value, &value_end, 0);
1596         if (errno) {
1597                 err = -errno;
1598                 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1599                 return err;
1600         }
1601         if (*value_end) {
1602                 pr_warn("failed to parse '%s' as integer completely\n", value);
1603                 return -EINVAL;
1604         }
1605         return 0;
1606 }
1607
1608 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1609 {
1610         int bit_sz = ext->kcfg.sz * 8;
1611
1612         if (ext->kcfg.sz == 8)
1613                 return true;
1614
1615         /* Validate that value stored in u64 fits in integer of `ext->sz`
1616          * bytes size without any loss of information. If the target integer
1617          * is signed, we rely on the following limits of integer type of
1618          * Y bits and subsequent transformation:
1619          *
1620          *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1621          *            0 <= X + 2^(Y-1) <= 2^Y - 1
1622          *            0 <= X + 2^(Y-1) <  2^Y
1623          *
1624          *  For unsigned target integer, check that all the (64 - Y) bits are
1625          *  zero.
1626          */
1627         if (ext->kcfg.is_signed)
1628                 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1629         else
1630                 return (v >> bit_sz) == 0;
1631 }
1632
1633 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1634                               __u64 value)
1635 {
1636         if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1637                 pr_warn("extern (kcfg) %s=%llu should be integer\n",
1638                         ext->name, (unsigned long long)value);
1639                 return -EINVAL;
1640         }
1641         if (!is_kcfg_value_in_range(ext, value)) {
1642                 pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n",
1643                         ext->name, (unsigned long long)value, ext->kcfg.sz);
1644                 return -ERANGE;
1645         }
1646         switch (ext->kcfg.sz) {
1647                 case 1: *(__u8 *)ext_val = value; break;
1648                 case 2: *(__u16 *)ext_val = value; break;
1649                 case 4: *(__u32 *)ext_val = value; break;
1650                 case 8: *(__u64 *)ext_val = value; break;
1651                 default:
1652                         return -EINVAL;
1653         }
1654         ext->is_set = true;
1655         return 0;
1656 }
1657
1658 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1659                                             char *buf, void *data)
1660 {
1661         struct extern_desc *ext;
1662         char *sep, *value;
1663         int len, err = 0;
1664         void *ext_val;
1665         __u64 num;
1666
1667         if (strncmp(buf, "CONFIG_", 7))
1668                 return 0;
1669
1670         sep = strchr(buf, '=');
1671         if (!sep) {
1672                 pr_warn("failed to parse '%s': no separator\n", buf);
1673                 return -EINVAL;
1674         }
1675
1676         /* Trim ending '\n' */
1677         len = strlen(buf);
1678         if (buf[len - 1] == '\n')
1679                 buf[len - 1] = '\0';
1680         /* Split on '=' and ensure that a value is present. */
1681         *sep = '\0';
1682         if (!sep[1]) {
1683                 *sep = '=';
1684                 pr_warn("failed to parse '%s': no value\n", buf);
1685                 return -EINVAL;
1686         }
1687
1688         ext = find_extern_by_name(obj, buf);
1689         if (!ext || ext->is_set)
1690                 return 0;
1691
1692         ext_val = data + ext->kcfg.data_off;
1693         value = sep + 1;
1694
1695         switch (*value) {
1696         case 'y': case 'n': case 'm':
1697                 err = set_kcfg_value_tri(ext, ext_val, *value);
1698                 break;
1699         case '"':
1700                 err = set_kcfg_value_str(ext, ext_val, value);
1701                 break;
1702         default:
1703                 /* assume integer */
1704                 err = parse_u64(value, &num);
1705                 if (err) {
1706                         pr_warn("extern (kcfg) %s=%s should be integer\n",
1707                                 ext->name, value);
1708                         return err;
1709                 }
1710                 err = set_kcfg_value_num(ext, ext_val, num);
1711                 break;
1712         }
1713         if (err)
1714                 return err;
1715         pr_debug("extern (kcfg) %s=%s\n", ext->name, value);
1716         return 0;
1717 }
1718
1719 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1720 {
1721         char buf[PATH_MAX];
1722         struct utsname uts;
1723         int len, err = 0;
1724         gzFile file;
1725
1726         uname(&uts);
1727         len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1728         if (len < 0)
1729                 return -EINVAL;
1730         else if (len >= PATH_MAX)
1731                 return -ENAMETOOLONG;
1732
1733         /* gzopen also accepts uncompressed files. */
1734         file = gzopen(buf, "r");
1735         if (!file)
1736                 file = gzopen("/proc/config.gz", "r");
1737
1738         if (!file) {
1739                 pr_warn("failed to open system Kconfig\n");
1740                 return -ENOENT;
1741         }
1742
1743         while (gzgets(file, buf, sizeof(buf))) {
1744                 err = bpf_object__process_kconfig_line(obj, buf, data);
1745                 if (err) {
1746                         pr_warn("error parsing system Kconfig line '%s': %d\n",
1747                                 buf, err);
1748                         goto out;
1749                 }
1750         }
1751
1752 out:
1753         gzclose(file);
1754         return err;
1755 }
1756
1757 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1758                                         const char *config, void *data)
1759 {
1760         char buf[PATH_MAX];
1761         int err = 0;
1762         FILE *file;
1763
1764         file = fmemopen((void *)config, strlen(config), "r");
1765         if (!file) {
1766                 err = -errno;
1767                 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1768                 return err;
1769         }
1770
1771         while (fgets(buf, sizeof(buf), file)) {
1772                 err = bpf_object__process_kconfig_line(obj, buf, data);
1773                 if (err) {
1774                         pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1775                                 buf, err);
1776                         break;
1777                 }
1778         }
1779
1780         fclose(file);
1781         return err;
1782 }
1783
1784 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1785 {
1786         struct extern_desc *last_ext = NULL, *ext;
1787         size_t map_sz;
1788         int i, err;
1789
1790         for (i = 0; i < obj->nr_extern; i++) {
1791                 ext = &obj->externs[i];
1792                 if (ext->type == EXT_KCFG)
1793                         last_ext = ext;
1794         }
1795
1796         if (!last_ext)
1797                 return 0;
1798
1799         map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
1800         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1801                                             obj->efile.symbols_shndx,
1802                                             NULL, map_sz);
1803         if (err)
1804                 return err;
1805
1806         obj->kconfig_map_idx = obj->nr_maps - 1;
1807
1808         return 0;
1809 }
1810
1811 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1812 {
1813         Elf_Data *symbols = obj->efile.symbols;
1814         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1815         Elf_Data *data = NULL;
1816         Elf_Scn *scn;
1817
1818         if (obj->efile.maps_shndx < 0)
1819                 return 0;
1820
1821         if (!symbols)
1822                 return -EINVAL;
1823
1824         scn = elf_sec_by_idx(obj, obj->efile.maps_shndx);
1825         data = elf_sec_data(obj, scn);
1826         if (!scn || !data) {
1827                 pr_warn("elf: failed to get legacy map definitions for %s\n",
1828                         obj->path);
1829                 return -EINVAL;
1830         }
1831
1832         /*
1833          * Count number of maps. Each map has a name.
1834          * Array of maps is not supported: only the first element is
1835          * considered.
1836          *
1837          * TODO: Detect array of map and report error.
1838          */
1839         nr_syms = symbols->d_size / sizeof(GElf_Sym);
1840         for (i = 0; i < nr_syms; i++) {
1841                 GElf_Sym sym;
1842
1843                 if (!gelf_getsym(symbols, i, &sym))
1844                         continue;
1845                 if (sym.st_shndx != obj->efile.maps_shndx)
1846                         continue;
1847                 nr_maps++;
1848         }
1849         /* Assume equally sized map definitions */
1850         pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n",
1851                  nr_maps, data->d_size, obj->path);
1852
1853         if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1854                 pr_warn("elf: unable to determine legacy map definition size in %s\n",
1855                         obj->path);
1856                 return -EINVAL;
1857         }
1858         map_def_sz = data->d_size / nr_maps;
1859
1860         /* Fill obj->maps using data in "maps" section.  */
1861         for (i = 0; i < nr_syms; i++) {
1862                 GElf_Sym sym;
1863                 const char *map_name;
1864                 struct bpf_map_def *def;
1865                 struct bpf_map *map;
1866
1867                 if (!gelf_getsym(symbols, i, &sym))
1868                         continue;
1869                 if (sym.st_shndx != obj->efile.maps_shndx)
1870                         continue;
1871
1872                 map = bpf_object__add_map(obj);
1873                 if (IS_ERR(map))
1874                         return PTR_ERR(map);
1875
1876                 map_name = elf_sym_str(obj, sym.st_name);
1877                 if (!map_name) {
1878                         pr_warn("failed to get map #%d name sym string for obj %s\n",
1879                                 i, obj->path);
1880                         return -LIBBPF_ERRNO__FORMAT;
1881                 }
1882
1883                 if (GELF_ST_TYPE(sym.st_info) == STT_SECTION
1884                     || GELF_ST_BIND(sym.st_info) == STB_LOCAL) {
1885                         pr_warn("map '%s' (legacy): static maps are not supported\n", map_name);
1886                         return -ENOTSUP;
1887                 }
1888
1889                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1890                 map->sec_idx = sym.st_shndx;
1891                 map->sec_offset = sym.st_value;
1892                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1893                          map_name, map->sec_idx, map->sec_offset);
1894                 if (sym.st_value + map_def_sz > data->d_size) {
1895                         pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1896                                 obj->path, map_name);
1897                         return -EINVAL;
1898                 }
1899
1900                 map->name = strdup(map_name);
1901                 if (!map->name) {
1902                         pr_warn("failed to alloc map name\n");
1903                         return -ENOMEM;
1904                 }
1905                 pr_debug("map %d is \"%s\"\n", i, map->name);
1906                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1907                 /*
1908                  * If the definition of the map in the object file fits in
1909                  * bpf_map_def, copy it.  Any extra fields in our version
1910                  * of bpf_map_def will default to zero as a result of the
1911                  * calloc above.
1912                  */
1913                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1914                         memcpy(&map->def, def, map_def_sz);
1915                 } else {
1916                         /*
1917                          * Here the map structure being read is bigger than what
1918                          * we expect, truncate if the excess bits are all zero.
1919                          * If they are not zero, reject this map as
1920                          * incompatible.
1921                          */
1922                         char *b;
1923
1924                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1925                              b < ((char *)def) + map_def_sz; b++) {
1926                                 if (*b != 0) {
1927                                         pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1928                                                 obj->path, map_name);
1929                                         if (strict)
1930                                                 return -EINVAL;
1931                                 }
1932                         }
1933                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1934                 }
1935         }
1936         return 0;
1937 }
1938
1939 const struct btf_type *
1940 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1941 {
1942         const struct btf_type *t = btf__type_by_id(btf, id);
1943
1944         if (res_id)
1945                 *res_id = id;
1946
1947         while (btf_is_mod(t) || btf_is_typedef(t)) {
1948                 if (res_id)
1949                         *res_id = t->type;
1950                 t = btf__type_by_id(btf, t->type);
1951         }
1952
1953         return t;
1954 }
1955
1956 static const struct btf_type *
1957 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
1958 {
1959         const struct btf_type *t;
1960
1961         t = skip_mods_and_typedefs(btf, id, NULL);
1962         if (!btf_is_ptr(t))
1963                 return NULL;
1964
1965         t = skip_mods_and_typedefs(btf, t->type, res_id);
1966
1967         return btf_is_func_proto(t) ? t : NULL;
1968 }
1969
1970 static const char *__btf_kind_str(__u16 kind)
1971 {
1972         switch (kind) {
1973         case BTF_KIND_UNKN: return "void";
1974         case BTF_KIND_INT: return "int";
1975         case BTF_KIND_PTR: return "ptr";
1976         case BTF_KIND_ARRAY: return "array";
1977         case BTF_KIND_STRUCT: return "struct";
1978         case BTF_KIND_UNION: return "union";
1979         case BTF_KIND_ENUM: return "enum";
1980         case BTF_KIND_FWD: return "fwd";
1981         case BTF_KIND_TYPEDEF: return "typedef";
1982         case BTF_KIND_VOLATILE: return "volatile";
1983         case BTF_KIND_CONST: return "const";
1984         case BTF_KIND_RESTRICT: return "restrict";
1985         case BTF_KIND_FUNC: return "func";
1986         case BTF_KIND_FUNC_PROTO: return "func_proto";
1987         case BTF_KIND_VAR: return "var";
1988         case BTF_KIND_DATASEC: return "datasec";
1989         case BTF_KIND_FLOAT: return "float";
1990         default: return "unknown";
1991         }
1992 }
1993
1994 const char *btf_kind_str(const struct btf_type *t)
1995 {
1996         return __btf_kind_str(btf_kind(t));
1997 }
1998
1999 /*
2000  * Fetch integer attribute of BTF map definition. Such attributes are
2001  * represented using a pointer to an array, in which dimensionality of array
2002  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2003  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2004  * type definition, while using only sizeof(void *) space in ELF data section.
2005  */
2006 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2007                               const struct btf_member *m, __u32 *res)
2008 {
2009         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2010         const char *name = btf__name_by_offset(btf, m->name_off);
2011         const struct btf_array *arr_info;
2012         const struct btf_type *arr_t;
2013
2014         if (!btf_is_ptr(t)) {
2015                 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2016                         map_name, name, btf_kind_str(t));
2017                 return false;
2018         }
2019
2020         arr_t = btf__type_by_id(btf, t->type);
2021         if (!arr_t) {
2022                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2023                         map_name, name, t->type);
2024                 return false;
2025         }
2026         if (!btf_is_array(arr_t)) {
2027                 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2028                         map_name, name, btf_kind_str(arr_t));
2029                 return false;
2030         }
2031         arr_info = btf_array(arr_t);
2032         *res = arr_info->nelems;
2033         return true;
2034 }
2035
2036 static int build_map_pin_path(struct bpf_map *map, const char *path)
2037 {
2038         char buf[PATH_MAX];
2039         int len;
2040
2041         if (!path)
2042                 path = "/sys/fs/bpf";
2043
2044         len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
2045         if (len < 0)
2046                 return -EINVAL;
2047         else if (len >= PATH_MAX)
2048                 return -ENAMETOOLONG;
2049
2050         return bpf_map__set_pin_path(map, buf);
2051 }
2052
2053 int parse_btf_map_def(const char *map_name, struct btf *btf,
2054                       const struct btf_type *def_t, bool strict,
2055                       struct btf_map_def *map_def, struct btf_map_def *inner_def)
2056 {
2057         const struct btf_type *t;
2058         const struct btf_member *m;
2059         bool is_inner = inner_def == NULL;
2060         int vlen, i;
2061
2062         vlen = btf_vlen(def_t);
2063         m = btf_members(def_t);
2064         for (i = 0; i < vlen; i++, m++) {
2065                 const char *name = btf__name_by_offset(btf, m->name_off);
2066
2067                 if (!name) {
2068                         pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2069                         return -EINVAL;
2070                 }
2071                 if (strcmp(name, "type") == 0) {
2072                         if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2073                                 return -EINVAL;
2074                         map_def->parts |= MAP_DEF_MAP_TYPE;
2075                 } else if (strcmp(name, "max_entries") == 0) {
2076                         if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2077                                 return -EINVAL;
2078                         map_def->parts |= MAP_DEF_MAX_ENTRIES;
2079                 } else if (strcmp(name, "map_flags") == 0) {
2080                         if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2081                                 return -EINVAL;
2082                         map_def->parts |= MAP_DEF_MAP_FLAGS;
2083                 } else if (strcmp(name, "numa_node") == 0) {
2084                         if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2085                                 return -EINVAL;
2086                         map_def->parts |= MAP_DEF_NUMA_NODE;
2087                 } else if (strcmp(name, "key_size") == 0) {
2088                         __u32 sz;
2089
2090                         if (!get_map_field_int(map_name, btf, m, &sz))
2091                                 return -EINVAL;
2092                         if (map_def->key_size && map_def->key_size != sz) {
2093                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
2094                                         map_name, map_def->key_size, sz);
2095                                 return -EINVAL;
2096                         }
2097                         map_def->key_size = sz;
2098                         map_def->parts |= MAP_DEF_KEY_SIZE;
2099                 } else if (strcmp(name, "key") == 0) {
2100                         __s64 sz;
2101
2102                         t = btf__type_by_id(btf, m->type);
2103                         if (!t) {
2104                                 pr_warn("map '%s': key type [%d] not found.\n",
2105                                         map_name, m->type);
2106                                 return -EINVAL;
2107                         }
2108                         if (!btf_is_ptr(t)) {
2109                                 pr_warn("map '%s': key spec is not PTR: %s.\n",
2110                                         map_name, btf_kind_str(t));
2111                                 return -EINVAL;
2112                         }
2113                         sz = btf__resolve_size(btf, t->type);
2114                         if (sz < 0) {
2115                                 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2116                                         map_name, t->type, (ssize_t)sz);
2117                                 return sz;
2118                         }
2119                         if (map_def->key_size && map_def->key_size != sz) {
2120                                 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2121                                         map_name, map_def->key_size, (ssize_t)sz);
2122                                 return -EINVAL;
2123                         }
2124                         map_def->key_size = sz;
2125                         map_def->key_type_id = t->type;
2126                         map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2127                 } else if (strcmp(name, "value_size") == 0) {
2128                         __u32 sz;
2129
2130                         if (!get_map_field_int(map_name, btf, m, &sz))
2131                                 return -EINVAL;
2132                         if (map_def->value_size && map_def->value_size != sz) {
2133                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
2134                                         map_name, map_def->value_size, sz);
2135                                 return -EINVAL;
2136                         }
2137                         map_def->value_size = sz;
2138                         map_def->parts |= MAP_DEF_VALUE_SIZE;
2139                 } else if (strcmp(name, "value") == 0) {
2140                         __s64 sz;
2141
2142                         t = btf__type_by_id(btf, m->type);
2143                         if (!t) {
2144                                 pr_warn("map '%s': value type [%d] not found.\n",
2145                                         map_name, m->type);
2146                                 return -EINVAL;
2147                         }
2148                         if (!btf_is_ptr(t)) {
2149                                 pr_warn("map '%s': value spec is not PTR: %s.\n",
2150                                         map_name, btf_kind_str(t));
2151                                 return -EINVAL;
2152                         }
2153                         sz = btf__resolve_size(btf, t->type);
2154                         if (sz < 0) {
2155                                 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2156                                         map_name, t->type, (ssize_t)sz);
2157                                 return sz;
2158                         }
2159                         if (map_def->value_size && map_def->value_size != sz) {
2160                                 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2161                                         map_name, map_def->value_size, (ssize_t)sz);
2162                                 return -EINVAL;
2163                         }
2164                         map_def->value_size = sz;
2165                         map_def->value_type_id = t->type;
2166                         map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2167                 }
2168                 else if (strcmp(name, "values") == 0) {
2169                         char inner_map_name[128];
2170                         int err;
2171
2172                         if (is_inner) {
2173                                 pr_warn("map '%s': multi-level inner maps not supported.\n",
2174                                         map_name);
2175                                 return -ENOTSUP;
2176                         }
2177                         if (i != vlen - 1) {
2178                                 pr_warn("map '%s': '%s' member should be last.\n",
2179                                         map_name, name);
2180                                 return -EINVAL;
2181                         }
2182                         if (!bpf_map_type__is_map_in_map(map_def->map_type)) {
2183                                 pr_warn("map '%s': should be map-in-map.\n",
2184                                         map_name);
2185                                 return -ENOTSUP;
2186                         }
2187                         if (map_def->value_size && map_def->value_size != 4) {
2188                                 pr_warn("map '%s': conflicting value size %u != 4.\n",
2189                                         map_name, map_def->value_size);
2190                                 return -EINVAL;
2191                         }
2192                         map_def->value_size = 4;
2193                         t = btf__type_by_id(btf, m->type);
2194                         if (!t) {
2195                                 pr_warn("map '%s': map-in-map inner type [%d] not found.\n",
2196                                         map_name, m->type);
2197                                 return -EINVAL;
2198                         }
2199                         if (!btf_is_array(t) || btf_array(t)->nelems) {
2200                                 pr_warn("map '%s': map-in-map inner spec is not a zero-sized array.\n",
2201                                         map_name);
2202                                 return -EINVAL;
2203                         }
2204                         t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2205                         if (!btf_is_ptr(t)) {
2206                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2207                                         map_name, btf_kind_str(t));
2208                                 return -EINVAL;
2209                         }
2210                         t = skip_mods_and_typedefs(btf, t->type, NULL);
2211                         if (!btf_is_struct(t)) {
2212                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2213                                         map_name, btf_kind_str(t));
2214                                 return -EINVAL;
2215                         }
2216
2217                         snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2218                         err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2219                         if (err)
2220                                 return err;
2221
2222                         map_def->parts |= MAP_DEF_INNER_MAP;
2223                 } else if (strcmp(name, "pinning") == 0) {
2224                         __u32 val;
2225
2226                         if (is_inner) {
2227                                 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2228                                 return -EINVAL;
2229                         }
2230                         if (!get_map_field_int(map_name, btf, m, &val))
2231                                 return -EINVAL;
2232                         if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2233                                 pr_warn("map '%s': invalid pinning value %u.\n",
2234                                         map_name, val);
2235                                 return -EINVAL;
2236                         }
2237                         map_def->pinning = val;
2238                         map_def->parts |= MAP_DEF_PINNING;
2239                 } else {
2240                         if (strict) {
2241                                 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2242                                 return -ENOTSUP;
2243                         }
2244                         pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2245                 }
2246         }
2247
2248         if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2249                 pr_warn("map '%s': map type isn't specified.\n", map_name);
2250                 return -EINVAL;
2251         }
2252
2253         return 0;
2254 }
2255
2256 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2257 {
2258         map->def.type = def->map_type;
2259         map->def.key_size = def->key_size;
2260         map->def.value_size = def->value_size;
2261         map->def.max_entries = def->max_entries;
2262         map->def.map_flags = def->map_flags;
2263
2264         map->numa_node = def->numa_node;
2265         map->btf_key_type_id = def->key_type_id;
2266         map->btf_value_type_id = def->value_type_id;
2267
2268         if (def->parts & MAP_DEF_MAP_TYPE)
2269                 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2270
2271         if (def->parts & MAP_DEF_KEY_TYPE)
2272                 pr_debug("map '%s': found key [%u], sz = %u.\n",
2273                          map->name, def->key_type_id, def->key_size);
2274         else if (def->parts & MAP_DEF_KEY_SIZE)
2275                 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2276
2277         if (def->parts & MAP_DEF_VALUE_TYPE)
2278                 pr_debug("map '%s': found value [%u], sz = %u.\n",
2279                          map->name, def->value_type_id, def->value_size);
2280         else if (def->parts & MAP_DEF_VALUE_SIZE)
2281                 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2282
2283         if (def->parts & MAP_DEF_MAX_ENTRIES)
2284                 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2285         if (def->parts & MAP_DEF_MAP_FLAGS)
2286                 pr_debug("map '%s': found map_flags = %u.\n", map->name, def->map_flags);
2287         if (def->parts & MAP_DEF_PINNING)
2288                 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2289         if (def->parts & MAP_DEF_NUMA_NODE)
2290                 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2291
2292         if (def->parts & MAP_DEF_INNER_MAP)
2293                 pr_debug("map '%s': found inner map definition.\n", map->name);
2294 }
2295
2296 static const char *btf_var_linkage_str(__u32 linkage)
2297 {
2298         switch (linkage) {
2299         case BTF_VAR_STATIC: return "static";
2300         case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2301         case BTF_VAR_GLOBAL_EXTERN: return "extern";
2302         default: return "unknown";
2303         }
2304 }
2305
2306 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2307                                          const struct btf_type *sec,
2308                                          int var_idx, int sec_idx,
2309                                          const Elf_Data *data, bool strict,
2310                                          const char *pin_root_path)
2311 {
2312         struct btf_map_def map_def = {}, inner_def = {};
2313         const struct btf_type *var, *def;
2314         const struct btf_var_secinfo *vi;
2315         const struct btf_var *var_extra;
2316         const char *map_name;
2317         struct bpf_map *map;
2318         int err;
2319
2320         vi = btf_var_secinfos(sec) + var_idx;
2321         var = btf__type_by_id(obj->btf, vi->type);
2322         var_extra = btf_var(var);
2323         map_name = btf__name_by_offset(obj->btf, var->name_off);
2324
2325         if (map_name == NULL || map_name[0] == '\0') {
2326                 pr_warn("map #%d: empty name.\n", var_idx);
2327                 return -EINVAL;
2328         }
2329         if ((__u64)vi->offset + vi->size > data->d_size) {
2330                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2331                 return -EINVAL;
2332         }
2333         if (!btf_is_var(var)) {
2334                 pr_warn("map '%s': unexpected var kind %s.\n",
2335                         map_name, btf_kind_str(var));
2336                 return -EINVAL;
2337         }
2338         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2339                 pr_warn("map '%s': unsupported map linkage %s.\n",
2340                         map_name, btf_var_linkage_str(var_extra->linkage));
2341                 return -EOPNOTSUPP;
2342         }
2343
2344         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2345         if (!btf_is_struct(def)) {
2346                 pr_warn("map '%s': unexpected def kind %s.\n",
2347                         map_name, btf_kind_str(var));
2348                 return -EINVAL;
2349         }
2350         if (def->size > vi->size) {
2351                 pr_warn("map '%s': invalid def size.\n", map_name);
2352                 return -EINVAL;
2353         }
2354
2355         map = bpf_object__add_map(obj);
2356         if (IS_ERR(map))
2357                 return PTR_ERR(map);
2358         map->name = strdup(map_name);
2359         if (!map->name) {
2360                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2361                 return -ENOMEM;
2362         }
2363         map->libbpf_type = LIBBPF_MAP_UNSPEC;
2364         map->def.type = BPF_MAP_TYPE_UNSPEC;
2365         map->sec_idx = sec_idx;
2366         map->sec_offset = vi->offset;
2367         map->btf_var_idx = var_idx;
2368         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2369                  map_name, map->sec_idx, map->sec_offset);
2370
2371         err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2372         if (err)
2373                 return err;
2374
2375         fill_map_from_def(map, &map_def);
2376
2377         if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2378                 err = build_map_pin_path(map, pin_root_path);
2379                 if (err) {
2380                         pr_warn("map '%s': couldn't build pin path.\n", map->name);
2381                         return err;
2382                 }
2383         }
2384
2385         if (map_def.parts & MAP_DEF_INNER_MAP) {
2386                 map->inner_map = calloc(1, sizeof(*map->inner_map));
2387                 if (!map->inner_map)
2388                         return -ENOMEM;
2389                 map->inner_map->fd = -1;
2390                 map->inner_map->sec_idx = sec_idx;
2391                 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2392                 if (!map->inner_map->name)
2393                         return -ENOMEM;
2394                 sprintf(map->inner_map->name, "%s.inner", map_name);
2395
2396                 fill_map_from_def(map->inner_map, &inner_def);
2397         }
2398
2399         return 0;
2400 }
2401
2402 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2403                                           const char *pin_root_path)
2404 {
2405         const struct btf_type *sec = NULL;
2406         int nr_types, i, vlen, err;
2407         const struct btf_type *t;
2408         const char *name;
2409         Elf_Data *data;
2410         Elf_Scn *scn;
2411
2412         if (obj->efile.btf_maps_shndx < 0)
2413                 return 0;
2414
2415         scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2416         data = elf_sec_data(obj, scn);
2417         if (!scn || !data) {
2418                 pr_warn("elf: failed to get %s map definitions for %s\n",
2419                         MAPS_ELF_SEC, obj->path);
2420                 return -EINVAL;
2421         }
2422
2423         nr_types = btf__get_nr_types(obj->btf);
2424         for (i = 1; i <= nr_types; i++) {
2425                 t = btf__type_by_id(obj->btf, i);
2426                 if (!btf_is_datasec(t))
2427                         continue;
2428                 name = btf__name_by_offset(obj->btf, t->name_off);
2429                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2430                         sec = t;
2431                         obj->efile.btf_maps_sec_btf_id = i;
2432                         break;
2433                 }
2434         }
2435
2436         if (!sec) {
2437                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2438                 return -ENOENT;
2439         }
2440
2441         vlen = btf_vlen(sec);
2442         for (i = 0; i < vlen; i++) {
2443                 err = bpf_object__init_user_btf_map(obj, sec, i,
2444                                                     obj->efile.btf_maps_shndx,
2445                                                     data, strict,
2446                                                     pin_root_path);
2447                 if (err)
2448                         return err;
2449         }
2450
2451         return 0;
2452 }
2453
2454 static int bpf_object__init_maps(struct bpf_object *obj,
2455                                  const struct bpf_object_open_opts *opts)
2456 {
2457         const char *pin_root_path;
2458         bool strict;
2459         int err;
2460
2461         strict = !OPTS_GET(opts, relaxed_maps, false);
2462         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2463
2464         err = bpf_object__init_user_maps(obj, strict);
2465         err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2466         err = err ?: bpf_object__init_global_data_maps(obj);
2467         err = err ?: bpf_object__init_kconfig_map(obj);
2468         err = err ?: bpf_object__init_struct_ops_maps(obj);
2469
2470         return err;
2471 }
2472
2473 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2474 {
2475         GElf_Shdr sh;
2476
2477         if (elf_sec_hdr(obj, elf_sec_by_idx(obj, idx), &sh))
2478                 return false;
2479
2480         return sh.sh_flags & SHF_EXECINSTR;
2481 }
2482
2483 static bool btf_needs_sanitization(struct bpf_object *obj)
2484 {
2485         bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2486         bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2487         bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2488         bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2489
2490         return !has_func || !has_datasec || !has_func_global || !has_float;
2491 }
2492
2493 static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2494 {
2495         bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2496         bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2497         bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2498         bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2499         struct btf_type *t;
2500         int i, j, vlen;
2501
2502         for (i = 1; i <= btf__get_nr_types(btf); i++) {
2503                 t = (struct btf_type *)btf__type_by_id(btf, i);
2504
2505                 if (!has_datasec && btf_is_var(t)) {
2506                         /* replace VAR with INT */
2507                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2508                         /*
2509                          * using size = 1 is the safest choice, 4 will be too
2510                          * big and cause kernel BTF validation failure if
2511                          * original variable took less than 4 bytes
2512                          */
2513                         t->size = 1;
2514                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2515                 } else if (!has_datasec && btf_is_datasec(t)) {
2516                         /* replace DATASEC with STRUCT */
2517                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
2518                         struct btf_member *m = btf_members(t);
2519                         struct btf_type *vt;
2520                         char *name;
2521
2522                         name = (char *)btf__name_by_offset(btf, t->name_off);
2523                         while (*name) {
2524                                 if (*name == '.')
2525                                         *name = '_';
2526                                 name++;
2527                         }
2528
2529                         vlen = btf_vlen(t);
2530                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2531                         for (j = 0; j < vlen; j++, v++, m++) {
2532                                 /* order of field assignments is important */
2533                                 m->offset = v->offset * 8;
2534                                 m->type = v->type;
2535                                 /* preserve variable name as member name */
2536                                 vt = (void *)btf__type_by_id(btf, v->type);
2537                                 m->name_off = vt->name_off;
2538                         }
2539                 } else if (!has_func && btf_is_func_proto(t)) {
2540                         /* replace FUNC_PROTO with ENUM */
2541                         vlen = btf_vlen(t);
2542                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2543                         t->size = sizeof(__u32); /* kernel enforced */
2544                 } else if (!has_func && btf_is_func(t)) {
2545                         /* replace FUNC with TYPEDEF */
2546                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2547                 } else if (!has_func_global && btf_is_func(t)) {
2548                         /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2549                         t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2550                 } else if (!has_float && btf_is_float(t)) {
2551                         /* replace FLOAT with an equally-sized empty STRUCT;
2552                          * since C compilers do not accept e.g. "float" as a
2553                          * valid struct name, make it anonymous
2554                          */
2555                         t->name_off = 0;
2556                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2557                 }
2558         }
2559 }
2560
2561 static bool libbpf_needs_btf(const struct bpf_object *obj)
2562 {
2563         return obj->efile.btf_maps_shndx >= 0 ||
2564                obj->efile.st_ops_shndx >= 0 ||
2565                obj->nr_extern > 0;
2566 }
2567
2568 static bool kernel_needs_btf(const struct bpf_object *obj)
2569 {
2570         return obj->efile.st_ops_shndx >= 0;
2571 }
2572
2573 static int bpf_object__init_btf(struct bpf_object *obj,
2574                                 Elf_Data *btf_data,
2575                                 Elf_Data *btf_ext_data)
2576 {
2577         int err = -ENOENT;
2578
2579         if (btf_data) {
2580                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2581                 err = libbpf_get_error(obj->btf);
2582                 if (err) {
2583                         obj->btf = NULL;
2584                         pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2585                         goto out;
2586                 }
2587                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2588                 btf__set_pointer_size(obj->btf, 8);
2589         }
2590         if (btf_ext_data) {
2591                 if (!obj->btf) {
2592                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2593                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2594                         goto out;
2595                 }
2596                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2597                 err = libbpf_get_error(obj->btf_ext);
2598                 if (err) {
2599                         pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2600                                 BTF_EXT_ELF_SEC, err);
2601                         obj->btf_ext = NULL;
2602                         goto out;
2603                 }
2604         }
2605 out:
2606         if (err && libbpf_needs_btf(obj)) {
2607                 pr_warn("BTF is required, but is missing or corrupted.\n");
2608                 return err;
2609         }
2610         return 0;
2611 }
2612
2613 static int bpf_object__finalize_btf(struct bpf_object *obj)
2614 {
2615         int err;
2616
2617         if (!obj->btf)
2618                 return 0;
2619
2620         err = btf__finalize_data(obj, obj->btf);
2621         if (err) {
2622                 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2623                 return err;
2624         }
2625
2626         return 0;
2627 }
2628
2629 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
2630 {
2631         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2632             prog->type == BPF_PROG_TYPE_LSM)
2633                 return true;
2634
2635         /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2636          * also need vmlinux BTF
2637          */
2638         if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2639                 return true;
2640
2641         return false;
2642 }
2643
2644 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
2645 {
2646         struct bpf_program *prog;
2647         int i;
2648
2649         /* CO-RE relocations need kernel BTF, only when btf_custom_path
2650          * is not specified
2651          */
2652         if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
2653                 return true;
2654
2655         /* Support for typed ksyms needs kernel BTF */
2656         for (i = 0; i < obj->nr_extern; i++) {
2657                 const struct extern_desc *ext;
2658
2659                 ext = &obj->externs[i];
2660                 if (ext->type == EXT_KSYM && ext->ksym.type_id)
2661                         return true;
2662         }
2663
2664         bpf_object__for_each_program(prog, obj) {
2665                 if (!prog->load)
2666                         continue;
2667                 if (prog_needs_vmlinux_btf(prog))
2668                         return true;
2669         }
2670
2671         return false;
2672 }
2673
2674 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
2675 {
2676         int err;
2677
2678         /* btf_vmlinux could be loaded earlier */
2679         if (obj->btf_vmlinux || obj->gen_loader)
2680                 return 0;
2681
2682         if (!force && !obj_needs_vmlinux_btf(obj))
2683                 return 0;
2684
2685         obj->btf_vmlinux = btf__load_vmlinux_btf();
2686         err = libbpf_get_error(obj->btf_vmlinux);
2687         if (err) {
2688                 pr_warn("Error loading vmlinux BTF: %d\n", err);
2689                 obj->btf_vmlinux = NULL;
2690                 return err;
2691         }
2692         return 0;
2693 }
2694
2695 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2696 {
2697         struct btf *kern_btf = obj->btf;
2698         bool btf_mandatory, sanitize;
2699         int i, err = 0;
2700
2701         if (!obj->btf)
2702                 return 0;
2703
2704         if (!kernel_supports(obj, FEAT_BTF)) {
2705                 if (kernel_needs_btf(obj)) {
2706                         err = -EOPNOTSUPP;
2707                         goto report;
2708                 }
2709                 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
2710                 return 0;
2711         }
2712
2713         /* Even though some subprogs are global/weak, user might prefer more
2714          * permissive BPF verification process that BPF verifier performs for
2715          * static functions, taking into account more context from the caller
2716          * functions. In such case, they need to mark such subprogs with
2717          * __attribute__((visibility("hidden"))) and libbpf will adjust
2718          * corresponding FUNC BTF type to be marked as static and trigger more
2719          * involved BPF verification process.
2720          */
2721         for (i = 0; i < obj->nr_programs; i++) {
2722                 struct bpf_program *prog = &obj->programs[i];
2723                 struct btf_type *t;
2724                 const char *name;
2725                 int j, n;
2726
2727                 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
2728                         continue;
2729
2730                 n = btf__get_nr_types(obj->btf);
2731                 for (j = 1; j <= n; j++) {
2732                         t = btf_type_by_id(obj->btf, j);
2733                         if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
2734                                 continue;
2735
2736                         name = btf__str_by_offset(obj->btf, t->name_off);
2737                         if (strcmp(name, prog->name) != 0)
2738                                 continue;
2739
2740                         t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
2741                         break;
2742                 }
2743         }
2744
2745         sanitize = btf_needs_sanitization(obj);
2746         if (sanitize) {
2747                 const void *raw_data;
2748                 __u32 sz;
2749
2750                 /* clone BTF to sanitize a copy and leave the original intact */
2751                 raw_data = btf__get_raw_data(obj->btf, &sz);
2752                 kern_btf = btf__new(raw_data, sz);
2753                 err = libbpf_get_error(kern_btf);
2754                 if (err)
2755                         return err;
2756
2757                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2758                 btf__set_pointer_size(obj->btf, 8);
2759                 bpf_object__sanitize_btf(obj, kern_btf);
2760         }
2761
2762         if (obj->gen_loader) {
2763                 __u32 raw_size = 0;
2764                 const void *raw_data = btf__get_raw_data(kern_btf, &raw_size);
2765
2766                 if (!raw_data)
2767                         return -ENOMEM;
2768                 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
2769                 /* Pretend to have valid FD to pass various fd >= 0 checks.
2770                  * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
2771                  */
2772                 btf__set_fd(kern_btf, 0);
2773         } else {
2774                 err = btf__load_into_kernel(kern_btf);
2775         }
2776         if (sanitize) {
2777                 if (!err) {
2778                         /* move fd to libbpf's BTF */
2779                         btf__set_fd(obj->btf, btf__fd(kern_btf));
2780                         btf__set_fd(kern_btf, -1);
2781                 }
2782                 btf__free(kern_btf);
2783         }
2784 report:
2785         if (err) {
2786                 btf_mandatory = kernel_needs_btf(obj);
2787                 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
2788                         btf_mandatory ? "BTF is mandatory, can't proceed."
2789                                       : "BTF is optional, ignoring.");
2790                 if (!btf_mandatory)
2791                         err = 0;
2792         }
2793         return err;
2794 }
2795
2796 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
2797 {
2798         const char *name;
2799
2800         name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
2801         if (!name) {
2802                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
2803                         off, obj->path, elf_errmsg(-1));
2804                 return NULL;
2805         }
2806
2807         return name;
2808 }
2809
2810 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
2811 {
2812         const char *name;
2813
2814         name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
2815         if (!name) {
2816                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
2817                         off, obj->path, elf_errmsg(-1));
2818                 return NULL;
2819         }
2820
2821         return name;
2822 }
2823
2824 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
2825 {
2826         Elf_Scn *scn;
2827
2828         scn = elf_getscn(obj->efile.elf, idx);
2829         if (!scn) {
2830                 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
2831                         idx, obj->path, elf_errmsg(-1));
2832                 return NULL;
2833         }
2834         return scn;
2835 }
2836
2837 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
2838 {
2839         Elf_Scn *scn = NULL;
2840         Elf *elf = obj->efile.elf;
2841         const char *sec_name;
2842
2843         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2844                 sec_name = elf_sec_name(obj, scn);
2845                 if (!sec_name)
2846                         return NULL;
2847
2848                 if (strcmp(sec_name, name) != 0)
2849                         continue;
2850
2851                 return scn;
2852         }
2853         return NULL;
2854 }
2855
2856 static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr)
2857 {
2858         if (!scn)
2859                 return -EINVAL;
2860
2861         if (gelf_getshdr(scn, hdr) != hdr) {
2862                 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
2863                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
2864                 return -EINVAL;
2865         }
2866
2867         return 0;
2868 }
2869
2870 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
2871 {
2872         const char *name;
2873         GElf_Shdr sh;
2874
2875         if (!scn)
2876                 return NULL;
2877
2878         if (elf_sec_hdr(obj, scn, &sh))
2879                 return NULL;
2880
2881         name = elf_sec_str(obj, sh.sh_name);
2882         if (!name) {
2883                 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
2884                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
2885                 return NULL;
2886         }
2887
2888         return name;
2889 }
2890
2891 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
2892 {
2893         Elf_Data *data;
2894
2895         if (!scn)
2896                 return NULL;
2897
2898         data = elf_getdata(scn, 0);
2899         if (!data) {
2900                 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
2901                         elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
2902                         obj->path, elf_errmsg(-1));
2903                 return NULL;
2904         }
2905
2906         return data;
2907 }
2908
2909 static bool is_sec_name_dwarf(const char *name)
2910 {
2911         /* approximation, but the actual list is too long */
2912         return strncmp(name, ".debug_", sizeof(".debug_") - 1) == 0;
2913 }
2914
2915 static bool ignore_elf_section(GElf_Shdr *hdr, const char *name)
2916 {
2917         /* no special handling of .strtab */
2918         if (hdr->sh_type == SHT_STRTAB)
2919                 return true;
2920
2921         /* ignore .llvm_addrsig section as well */
2922         if (hdr->sh_type == SHT_LLVM_ADDRSIG)
2923                 return true;
2924
2925         /* no subprograms will lead to an empty .text section, ignore it */
2926         if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
2927             strcmp(name, ".text") == 0)
2928                 return true;
2929
2930         /* DWARF sections */
2931         if (is_sec_name_dwarf(name))
2932                 return true;
2933
2934         if (strncmp(name, ".rel", sizeof(".rel") - 1) == 0) {
2935                 name += sizeof(".rel") - 1;
2936                 /* DWARF section relocations */
2937                 if (is_sec_name_dwarf(name))
2938                         return true;
2939
2940                 /* .BTF and .BTF.ext don't need relocations */
2941                 if (strcmp(name, BTF_ELF_SEC) == 0 ||
2942                     strcmp(name, BTF_EXT_ELF_SEC) == 0)
2943                         return true;
2944         }
2945
2946         return false;
2947 }
2948
2949 static int cmp_progs(const void *_a, const void *_b)
2950 {
2951         const struct bpf_program *a = _a;
2952         const struct bpf_program *b = _b;
2953
2954         if (a->sec_idx != b->sec_idx)
2955                 return a->sec_idx < b->sec_idx ? -1 : 1;
2956
2957         /* sec_insn_off can't be the same within the section */
2958         return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
2959 }
2960
2961 static int bpf_object__elf_collect(struct bpf_object *obj)
2962 {
2963         Elf *elf = obj->efile.elf;
2964         Elf_Data *btf_ext_data = NULL;
2965         Elf_Data *btf_data = NULL;
2966         int idx = 0, err = 0;
2967         const char *name;
2968         Elf_Data *data;
2969         Elf_Scn *scn;
2970         GElf_Shdr sh;
2971
2972         /* a bunch of ELF parsing functionality depends on processing symbols,
2973          * so do the first pass and find the symbol table
2974          */
2975         scn = NULL;
2976         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2977                 if (elf_sec_hdr(obj, scn, &sh))
2978                         return -LIBBPF_ERRNO__FORMAT;
2979
2980                 if (sh.sh_type == SHT_SYMTAB) {
2981                         if (obj->efile.symbols) {
2982                                 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
2983                                 return -LIBBPF_ERRNO__FORMAT;
2984                         }
2985
2986                         data = elf_sec_data(obj, scn);
2987                         if (!data)
2988                                 return -LIBBPF_ERRNO__FORMAT;
2989
2990                         obj->efile.symbols = data;
2991                         obj->efile.symbols_shndx = elf_ndxscn(scn);
2992                         obj->efile.strtabidx = sh.sh_link;
2993                 }
2994         }
2995
2996         scn = NULL;
2997         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2998                 idx++;
2999
3000                 if (elf_sec_hdr(obj, scn, &sh))
3001                         return -LIBBPF_ERRNO__FORMAT;
3002
3003                 name = elf_sec_str(obj, sh.sh_name);
3004                 if (!name)
3005                         return -LIBBPF_ERRNO__FORMAT;
3006
3007                 if (ignore_elf_section(&sh, name))
3008                         continue;
3009
3010                 data = elf_sec_data(obj, scn);
3011                 if (!data)
3012                         return -LIBBPF_ERRNO__FORMAT;
3013
3014                 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3015                          idx, name, (unsigned long)data->d_size,
3016                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
3017                          (int)sh.sh_type);
3018
3019                 if (strcmp(name, "license") == 0) {
3020                         err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3021                         if (err)
3022                                 return err;
3023                 } else if (strcmp(name, "version") == 0) {
3024                         err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3025                         if (err)
3026                                 return err;
3027                 } else if (strcmp(name, "maps") == 0) {
3028                         obj->efile.maps_shndx = idx;
3029                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3030                         obj->efile.btf_maps_shndx = idx;
3031                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3032                         btf_data = data;
3033                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3034                         btf_ext_data = data;
3035                 } else if (sh.sh_type == SHT_SYMTAB) {
3036                         /* already processed during the first pass above */
3037                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
3038                         if (sh.sh_flags & SHF_EXECINSTR) {
3039                                 if (strcmp(name, ".text") == 0)
3040                                         obj->efile.text_shndx = idx;
3041                                 err = bpf_object__add_programs(obj, data, name, idx);
3042                                 if (err)
3043                                         return err;
3044                         } else if (strcmp(name, DATA_SEC) == 0) {
3045                                 obj->efile.data = data;
3046                                 obj->efile.data_shndx = idx;
3047                         } else if (strcmp(name, RODATA_SEC) == 0) {
3048                                 obj->efile.rodata = data;
3049                                 obj->efile.rodata_shndx = idx;
3050                         } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3051                                 obj->efile.st_ops_data = data;
3052                                 obj->efile.st_ops_shndx = idx;
3053                         } else {
3054                                 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3055                                         idx, name);
3056                         }
3057                 } else if (sh.sh_type == SHT_REL) {
3058                         int nr_sects = obj->efile.nr_reloc_sects;
3059                         void *sects = obj->efile.reloc_sects;
3060                         int sec = sh.sh_info; /* points to other section */
3061
3062                         /* Only do relo for section with exec instructions */
3063                         if (!section_have_execinstr(obj, sec) &&
3064                             strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3065                             strcmp(name, ".rel" MAPS_ELF_SEC)) {
3066                                 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3067                                         idx, name, sec,
3068                                         elf_sec_name(obj, elf_sec_by_idx(obj, sec)) ?: "<?>");
3069                                 continue;
3070                         }
3071
3072                         sects = libbpf_reallocarray(sects, nr_sects + 1,
3073                                                     sizeof(*obj->efile.reloc_sects));
3074                         if (!sects)
3075                                 return -ENOMEM;
3076
3077                         obj->efile.reloc_sects = sects;
3078                         obj->efile.nr_reloc_sects++;
3079
3080                         obj->efile.reloc_sects[nr_sects].shdr = sh;
3081                         obj->efile.reloc_sects[nr_sects].data = data;
3082                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) {
3083                         obj->efile.bss = data;
3084                         obj->efile.bss_shndx = idx;
3085                 } else {
3086                         pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3087                                 (size_t)sh.sh_size);
3088                 }
3089         }
3090
3091         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3092                 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3093                 return -LIBBPF_ERRNO__FORMAT;
3094         }
3095
3096         /* sort BPF programs by section name and in-section instruction offset
3097          * for faster search */
3098         qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3099
3100         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3101 }
3102
3103 static bool sym_is_extern(const GElf_Sym *sym)
3104 {
3105         int bind = GELF_ST_BIND(sym->st_info);
3106         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3107         return sym->st_shndx == SHN_UNDEF &&
3108                (bind == STB_GLOBAL || bind == STB_WEAK) &&
3109                GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
3110 }
3111
3112 static bool sym_is_subprog(const GElf_Sym *sym, int text_shndx)
3113 {
3114         int bind = GELF_ST_BIND(sym->st_info);
3115         int type = GELF_ST_TYPE(sym->st_info);
3116
3117         /* in .text section */
3118         if (sym->st_shndx != text_shndx)
3119                 return false;
3120
3121         /* local function */
3122         if (bind == STB_LOCAL && type == STT_SECTION)
3123                 return true;
3124
3125         /* global function */
3126         return bind == STB_GLOBAL && type == STT_FUNC;
3127 }
3128
3129 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3130 {
3131         const struct btf_type *t;
3132         const char *tname;
3133         int i, n;
3134
3135         if (!btf)
3136                 return -ESRCH;
3137
3138         n = btf__get_nr_types(btf);
3139         for (i = 1; i <= n; i++) {
3140                 t = btf__type_by_id(btf, i);
3141
3142                 if (!btf_is_var(t) && !btf_is_func(t))
3143                         continue;
3144
3145                 tname = btf__name_by_offset(btf, t->name_off);
3146                 if (strcmp(tname, ext_name))
3147                         continue;
3148
3149                 if (btf_is_var(t) &&
3150                     btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3151                         return -EINVAL;
3152
3153                 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3154                         return -EINVAL;
3155
3156                 return i;
3157         }
3158
3159         return -ENOENT;
3160 }
3161
3162 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3163         const struct btf_var_secinfo *vs;
3164         const struct btf_type *t;
3165         int i, j, n;
3166
3167         if (!btf)
3168                 return -ESRCH;
3169
3170         n = btf__get_nr_types(btf);
3171         for (i = 1; i <= n; i++) {
3172                 t = btf__type_by_id(btf, i);
3173
3174                 if (!btf_is_datasec(t))
3175                         continue;
3176
3177                 vs = btf_var_secinfos(t);
3178                 for (j = 0; j < btf_vlen(t); j++, vs++) {
3179                         if (vs->type == ext_btf_id)
3180                                 return i;
3181                 }
3182         }
3183
3184         return -ENOENT;
3185 }
3186
3187 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3188                                      bool *is_signed)
3189 {
3190         const struct btf_type *t;
3191         const char *name;
3192
3193         t = skip_mods_and_typedefs(btf, id, NULL);
3194         name = btf__name_by_offset(btf, t->name_off);
3195
3196         if (is_signed)
3197                 *is_signed = false;
3198         switch (btf_kind(t)) {
3199         case BTF_KIND_INT: {
3200                 int enc = btf_int_encoding(t);
3201
3202                 if (enc & BTF_INT_BOOL)
3203                         return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3204                 if (is_signed)
3205                         *is_signed = enc & BTF_INT_SIGNED;
3206                 if (t->size == 1)
3207                         return KCFG_CHAR;
3208                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3209                         return KCFG_UNKNOWN;
3210                 return KCFG_INT;
3211         }
3212         case BTF_KIND_ENUM:
3213                 if (t->size != 4)
3214                         return KCFG_UNKNOWN;
3215                 if (strcmp(name, "libbpf_tristate"))
3216                         return KCFG_UNKNOWN;
3217                 return KCFG_TRISTATE;
3218         case BTF_KIND_ARRAY:
3219                 if (btf_array(t)->nelems == 0)
3220                         return KCFG_UNKNOWN;
3221                 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3222                         return KCFG_UNKNOWN;
3223                 return KCFG_CHAR_ARR;
3224         default:
3225                 return KCFG_UNKNOWN;
3226         }
3227 }
3228
3229 static int cmp_externs(const void *_a, const void *_b)
3230 {
3231         const struct extern_desc *a = _a;
3232         const struct extern_desc *b = _b;
3233
3234         if (a->type != b->type)
3235                 return a->type < b->type ? -1 : 1;
3236
3237         if (a->type == EXT_KCFG) {
3238                 /* descending order by alignment requirements */
3239                 if (a->kcfg.align != b->kcfg.align)
3240                         return a->kcfg.align > b->kcfg.align ? -1 : 1;
3241                 /* ascending order by size, within same alignment class */
3242                 if (a->kcfg.sz != b->kcfg.sz)
3243                         return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3244         }
3245
3246         /* resolve ties by name */
3247         return strcmp(a->name, b->name);
3248 }
3249
3250 static int find_int_btf_id(const struct btf *btf)
3251 {
3252         const struct btf_type *t;
3253         int i, n;
3254
3255         n = btf__get_nr_types(btf);
3256         for (i = 1; i <= n; i++) {
3257                 t = btf__type_by_id(btf, i);
3258
3259                 if (btf_is_int(t) && btf_int_bits(t) == 32)
3260                         return i;
3261         }
3262
3263         return 0;
3264 }
3265
3266 static int add_dummy_ksym_var(struct btf *btf)
3267 {
3268         int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
3269         const struct btf_var_secinfo *vs;
3270         const struct btf_type *sec;
3271
3272         if (!btf)
3273                 return 0;
3274
3275         sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
3276                                             BTF_KIND_DATASEC);
3277         if (sec_btf_id < 0)
3278                 return 0;
3279
3280         sec = btf__type_by_id(btf, sec_btf_id);
3281         vs = btf_var_secinfos(sec);
3282         for (i = 0; i < btf_vlen(sec); i++, vs++) {
3283                 const struct btf_type *vt;
3284
3285                 vt = btf__type_by_id(btf, vs->type);
3286                 if (btf_is_func(vt))
3287                         break;
3288         }
3289
3290         /* No func in ksyms sec.  No need to add dummy var. */
3291         if (i == btf_vlen(sec))
3292                 return 0;
3293
3294         int_btf_id = find_int_btf_id(btf);
3295         dummy_var_btf_id = btf__add_var(btf,
3296                                         "dummy_ksym",
3297                                         BTF_VAR_GLOBAL_ALLOCATED,
3298                                         int_btf_id);
3299         if (dummy_var_btf_id < 0)
3300                 pr_warn("cannot create a dummy_ksym var\n");
3301
3302         return dummy_var_btf_id;
3303 }
3304
3305 static int bpf_object__collect_externs(struct bpf_object *obj)
3306 {
3307         struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3308         const struct btf_type *t;
3309         struct extern_desc *ext;
3310         int i, n, off, dummy_var_btf_id;
3311         const char *ext_name, *sec_name;
3312         Elf_Scn *scn;
3313         GElf_Shdr sh;
3314
3315         if (!obj->efile.symbols)
3316                 return 0;
3317
3318         scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3319         if (elf_sec_hdr(obj, scn, &sh))
3320                 return -LIBBPF_ERRNO__FORMAT;
3321
3322         dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
3323         if (dummy_var_btf_id < 0)
3324                 return dummy_var_btf_id;
3325
3326         n = sh.sh_size / sh.sh_entsize;
3327         pr_debug("looking for externs among %d symbols...\n", n);
3328
3329         for (i = 0; i < n; i++) {
3330                 GElf_Sym sym;
3331
3332                 if (!gelf_getsym(obj->efile.symbols, i, &sym))
3333                         return -LIBBPF_ERRNO__FORMAT;
3334                 if (!sym_is_extern(&sym))
3335                         continue;
3336                 ext_name = elf_sym_str(obj, sym.st_name);
3337                 if (!ext_name || !ext_name[0])
3338                         continue;
3339
3340                 ext = obj->externs;
3341                 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3342                 if (!ext)
3343                         return -ENOMEM;
3344                 obj->externs = ext;
3345                 ext = &ext[obj->nr_extern];
3346                 memset(ext, 0, sizeof(*ext));
3347                 obj->nr_extern++;
3348
3349                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3350                 if (ext->btf_id <= 0) {
3351                         pr_warn("failed to find BTF for extern '%s': %d\n",
3352                                 ext_name, ext->btf_id);
3353                         return ext->btf_id;
3354                 }
3355                 t = btf__type_by_id(obj->btf, ext->btf_id);
3356                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
3357                 ext->sym_idx = i;
3358                 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
3359
3360                 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3361                 if (ext->sec_btf_id <= 0) {
3362                         pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3363                                 ext_name, ext->btf_id, ext->sec_btf_id);
3364                         return ext->sec_btf_id;
3365                 }
3366                 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3367                 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3368
3369                 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3370                         if (btf_is_func(t)) {
3371                                 pr_warn("extern function %s is unsupported under %s section\n",
3372                                         ext->name, KCONFIG_SEC);
3373                                 return -ENOTSUP;
3374                         }
3375                         kcfg_sec = sec;
3376                         ext->type = EXT_KCFG;
3377                         ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3378                         if (ext->kcfg.sz <= 0) {
3379                                 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3380                                         ext_name, ext->kcfg.sz);
3381                                 return ext->kcfg.sz;
3382                         }
3383                         ext->kcfg.align = btf__align_of(obj->btf, t->type);
3384                         if (ext->kcfg.align <= 0) {
3385                                 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3386                                         ext_name, ext->kcfg.align);
3387                                 return -EINVAL;
3388                         }
3389                         ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3390                                                         &ext->kcfg.is_signed);
3391                         if (ext->kcfg.type == KCFG_UNKNOWN) {
3392                                 pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
3393                                 return -ENOTSUP;
3394                         }
3395                 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3396                         if (btf_is_func(t) && ext->is_weak) {
3397                                 pr_warn("extern weak function %s is unsupported\n",
3398                                         ext->name);
3399                                 return -ENOTSUP;
3400                         }
3401                         ksym_sec = sec;
3402                         ext->type = EXT_KSYM;
3403                         skip_mods_and_typedefs(obj->btf, t->type,
3404                                                &ext->ksym.type_id);
3405                 } else {
3406                         pr_warn("unrecognized extern section '%s'\n", sec_name);
3407                         return -ENOTSUP;
3408                 }
3409         }
3410         pr_debug("collected %d externs total\n", obj->nr_extern);
3411
3412         if (!obj->nr_extern)
3413                 return 0;
3414
3415         /* sort externs by type, for kcfg ones also by (align, size, name) */
3416         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3417
3418         /* for .ksyms section, we need to turn all externs into allocated
3419          * variables in BTF to pass kernel verification; we do this by
3420          * pretending that each extern is a 8-byte variable
3421          */
3422         if (ksym_sec) {
3423                 /* find existing 4-byte integer type in BTF to use for fake
3424                  * extern variables in DATASEC
3425                  */
3426                 int int_btf_id = find_int_btf_id(obj->btf);
3427                 /* For extern function, a dummy_var added earlier
3428                  * will be used to replace the vs->type and
3429                  * its name string will be used to refill
3430                  * the missing param's name.
3431                  */
3432                 const struct btf_type *dummy_var;
3433
3434                 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
3435                 for (i = 0; i < obj->nr_extern; i++) {
3436                         ext = &obj->externs[i];
3437                         if (ext->type != EXT_KSYM)
3438                                 continue;
3439                         pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
3440                                  i, ext->sym_idx, ext->name);
3441                 }
3442
3443                 sec = ksym_sec;
3444                 n = btf_vlen(sec);
3445                 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
3446                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3447                         struct btf_type *vt;
3448
3449                         vt = (void *)btf__type_by_id(obj->btf, vs->type);
3450                         ext_name = btf__name_by_offset(obj->btf, vt->name_off);
3451                         ext = find_extern_by_name(obj, ext_name);
3452                         if (!ext) {
3453                                 pr_warn("failed to find extern definition for BTF %s '%s'\n",
3454                                         btf_kind_str(vt), ext_name);
3455                                 return -ESRCH;
3456                         }
3457                         if (btf_is_func(vt)) {
3458                                 const struct btf_type *func_proto;
3459                                 struct btf_param *param;
3460                                 int j;
3461
3462                                 func_proto = btf__type_by_id(obj->btf,
3463                                                              vt->type);
3464                                 param = btf_params(func_proto);
3465                                 /* Reuse the dummy_var string if the
3466                                  * func proto does not have param name.
3467                                  */
3468                                 for (j = 0; j < btf_vlen(func_proto); j++)
3469                                         if (param[j].type && !param[j].name_off)
3470                                                 param[j].name_off =
3471                                                         dummy_var->name_off;
3472                                 vs->type = dummy_var_btf_id;
3473                                 vt->info &= ~0xffff;
3474                                 vt->info |= BTF_FUNC_GLOBAL;
3475                         } else {
3476                                 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3477                                 vt->type = int_btf_id;
3478                         }
3479                         vs->offset = off;
3480                         vs->size = sizeof(int);
3481                 }
3482                 sec->size = off;
3483         }
3484
3485         if (kcfg_sec) {
3486                 sec = kcfg_sec;
3487                 /* for kcfg externs calculate their offsets within a .kconfig map */
3488                 off = 0;
3489                 for (i = 0; i < obj->nr_extern; i++) {
3490                         ext = &obj->externs[i];
3491                         if (ext->type != EXT_KCFG)
3492                                 continue;
3493
3494                         ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3495                         off = ext->kcfg.data_off + ext->kcfg.sz;
3496                         pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3497                                  i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3498                 }
3499                 sec->size = off;
3500                 n = btf_vlen(sec);
3501                 for (i = 0; i < n; i++) {
3502                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3503
3504                         t = btf__type_by_id(obj->btf, vs->type);
3505                         ext_name = btf__name_by_offset(obj->btf, t->name_off);
3506                         ext = find_extern_by_name(obj, ext_name);
3507                         if (!ext) {
3508                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
3509                                         ext_name);
3510                                 return -ESRCH;
3511                         }
3512                         btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3513                         vs->offset = ext->kcfg.data_off;
3514                 }
3515         }
3516         return 0;
3517 }
3518
3519 struct bpf_program *
3520 bpf_object__find_program_by_title(const struct bpf_object *obj,
3521                                   const char *title)
3522 {
3523         struct bpf_program *pos;
3524
3525         bpf_object__for_each_program(pos, obj) {
3526                 if (pos->sec_name && !strcmp(pos->sec_name, title))
3527                         return pos;
3528         }
3529         return errno = ENOENT, NULL;
3530 }
3531
3532 static bool prog_is_subprog(const struct bpf_object *obj,
3533                             const struct bpf_program *prog)
3534 {
3535         /* For legacy reasons, libbpf supports an entry-point BPF programs
3536          * without SEC() attribute, i.e., those in the .text section. But if
3537          * there are 2 or more such programs in the .text section, they all
3538          * must be subprograms called from entry-point BPF programs in
3539          * designated SEC()'tions, otherwise there is no way to distinguish
3540          * which of those programs should be loaded vs which are a subprogram.
3541          * Similarly, if there is a function/program in .text and at least one
3542          * other BPF program with custom SEC() attribute, then we just assume
3543          * .text programs are subprograms (even if they are not called from
3544          * other programs), because libbpf never explicitly supported mixing
3545          * SEC()-designated BPF programs and .text entry-point BPF programs.
3546          */
3547         return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
3548 }
3549
3550 struct bpf_program *
3551 bpf_object__find_program_by_name(const struct bpf_object *obj,
3552                                  const char *name)
3553 {
3554         struct bpf_program *prog;
3555
3556         bpf_object__for_each_program(prog, obj) {
3557                 if (prog_is_subprog(obj, prog))
3558                         continue;
3559                 if (!strcmp(prog->name, name))
3560                         return prog;
3561         }
3562         return errno = ENOENT, NULL;
3563 }
3564
3565 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3566                                       int shndx)
3567 {
3568         return shndx == obj->efile.data_shndx ||
3569                shndx == obj->efile.bss_shndx ||
3570                shndx == obj->efile.rodata_shndx;
3571 }
3572
3573 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3574                                       int shndx)
3575 {
3576         return shndx == obj->efile.maps_shndx ||
3577                shndx == obj->efile.btf_maps_shndx;
3578 }
3579
3580 static enum libbpf_map_type
3581 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3582 {
3583         if (shndx == obj->efile.data_shndx)
3584                 return LIBBPF_MAP_DATA;
3585         else if (shndx == obj->efile.bss_shndx)
3586                 return LIBBPF_MAP_BSS;
3587         else if (shndx == obj->efile.rodata_shndx)
3588                 return LIBBPF_MAP_RODATA;
3589         else if (shndx == obj->efile.symbols_shndx)
3590                 return LIBBPF_MAP_KCONFIG;
3591         else
3592                 return LIBBPF_MAP_UNSPEC;
3593 }
3594
3595 static int bpf_program__record_reloc(struct bpf_program *prog,
3596                                      struct reloc_desc *reloc_desc,
3597                                      __u32 insn_idx, const char *sym_name,
3598                                      const GElf_Sym *sym, const GElf_Rel *rel)
3599 {
3600         struct bpf_insn *insn = &prog->insns[insn_idx];
3601         size_t map_idx, nr_maps = prog->obj->nr_maps;
3602         struct bpf_object *obj = prog->obj;
3603         __u32 shdr_idx = sym->st_shndx;
3604         enum libbpf_map_type type;
3605         const char *sym_sec_name;
3606         struct bpf_map *map;
3607
3608         if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
3609                 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
3610                         prog->name, sym_name, insn_idx, insn->code);
3611                 return -LIBBPF_ERRNO__RELOC;
3612         }
3613
3614         if (sym_is_extern(sym)) {
3615                 int sym_idx = GELF_R_SYM(rel->r_info);
3616                 int i, n = obj->nr_extern;
3617                 struct extern_desc *ext;
3618
3619                 for (i = 0; i < n; i++) {
3620                         ext = &obj->externs[i];
3621                         if (ext->sym_idx == sym_idx)
3622                                 break;
3623                 }
3624                 if (i >= n) {
3625                         pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
3626                                 prog->name, sym_name, sym_idx);
3627                         return -LIBBPF_ERRNO__RELOC;
3628                 }
3629                 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
3630                          prog->name, i, ext->name, ext->sym_idx, insn_idx);
3631                 if (insn->code == (BPF_JMP | BPF_CALL))
3632                         reloc_desc->type = RELO_EXTERN_FUNC;
3633                 else
3634                         reloc_desc->type = RELO_EXTERN_VAR;
3635                 reloc_desc->insn_idx = insn_idx;
3636                 reloc_desc->sym_off = i; /* sym_off stores extern index */
3637                 return 0;
3638         }
3639
3640         /* sub-program call relocation */
3641         if (is_call_insn(insn)) {
3642                 if (insn->src_reg != BPF_PSEUDO_CALL) {
3643                         pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
3644                         return -LIBBPF_ERRNO__RELOC;
3645                 }
3646                 /* text_shndx can be 0, if no default "main" program exists */
3647                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3648                         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3649                         pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
3650                                 prog->name, sym_name, sym_sec_name);
3651                         return -LIBBPF_ERRNO__RELOC;
3652                 }
3653                 if (sym->st_value % BPF_INSN_SZ) {
3654                         pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
3655                                 prog->name, sym_name, (size_t)sym->st_value);
3656                         return -LIBBPF_ERRNO__RELOC;
3657                 }
3658                 reloc_desc->type = RELO_CALL;
3659                 reloc_desc->insn_idx = insn_idx;
3660                 reloc_desc->sym_off = sym->st_value;
3661                 return 0;
3662         }
3663
3664         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3665                 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
3666                         prog->name, sym_name, shdr_idx);
3667                 return -LIBBPF_ERRNO__RELOC;
3668         }
3669
3670         /* loading subprog addresses */
3671         if (sym_is_subprog(sym, obj->efile.text_shndx)) {
3672                 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
3673                  * local_func: sym->st_value = 0, insn->imm = offset in the section.
3674                  */
3675                 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
3676                         pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
3677                                 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
3678                         return -LIBBPF_ERRNO__RELOC;
3679                 }
3680
3681                 reloc_desc->type = RELO_SUBPROG_ADDR;
3682                 reloc_desc->insn_idx = insn_idx;
3683                 reloc_desc->sym_off = sym->st_value;
3684                 return 0;
3685         }
3686
3687         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
3688         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3689
3690         /* generic map reference relocation */
3691         if (type == LIBBPF_MAP_UNSPEC) {
3692                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
3693                         pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
3694                                 prog->name, sym_name, sym_sec_name);
3695                         return -LIBBPF_ERRNO__RELOC;
3696                 }
3697                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3698                         map = &obj->maps[map_idx];
3699                         if (map->libbpf_type != type ||
3700                             map->sec_idx != sym->st_shndx ||
3701                             map->sec_offset != sym->st_value)
3702                                 continue;
3703                         pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
3704                                  prog->name, map_idx, map->name, map->sec_idx,
3705                                  map->sec_offset, insn_idx);
3706                         break;
3707                 }
3708                 if (map_idx >= nr_maps) {
3709                         pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
3710                                 prog->name, sym_sec_name, (size_t)sym->st_value);
3711                         return -LIBBPF_ERRNO__RELOC;
3712                 }
3713                 reloc_desc->type = RELO_LD64;
3714                 reloc_desc->insn_idx = insn_idx;
3715                 reloc_desc->map_idx = map_idx;
3716                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
3717                 return 0;
3718         }
3719
3720         /* global data map relocation */
3721         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
3722                 pr_warn("prog '%s': bad data relo against section '%s'\n",
3723                         prog->name, sym_sec_name);
3724                 return -LIBBPF_ERRNO__RELOC;
3725         }
3726         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3727                 map = &obj->maps[map_idx];
3728                 if (map->libbpf_type != type)
3729                         continue;
3730                 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
3731                          prog->name, map_idx, map->name, map->sec_idx,
3732                          map->sec_offset, insn_idx);
3733                 break;
3734         }
3735         if (map_idx >= nr_maps) {
3736                 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
3737                         prog->name, sym_sec_name);
3738                 return -LIBBPF_ERRNO__RELOC;
3739         }
3740
3741         reloc_desc->type = RELO_DATA;
3742         reloc_desc->insn_idx = insn_idx;
3743         reloc_desc->map_idx = map_idx;
3744         reloc_desc->sym_off = sym->st_value;
3745         return 0;
3746 }
3747
3748 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
3749 {
3750         return insn_idx >= prog->sec_insn_off &&
3751                insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
3752 }
3753
3754 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
3755                                                  size_t sec_idx, size_t insn_idx)
3756 {
3757         int l = 0, r = obj->nr_programs - 1, m;
3758         struct bpf_program *prog;
3759
3760         while (l < r) {
3761                 m = l + (r - l + 1) / 2;
3762                 prog = &obj->programs[m];
3763
3764                 if (prog->sec_idx < sec_idx ||
3765                     (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
3766                         l = m;
3767                 else
3768                         r = m - 1;
3769         }
3770         /* matching program could be at index l, but it still might be the
3771          * wrong one, so we need to double check conditions for the last time
3772          */
3773         prog = &obj->programs[l];
3774         if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
3775                 return prog;
3776         return NULL;
3777 }
3778
3779 static int
3780 bpf_object__collect_prog_relos(struct bpf_object *obj, GElf_Shdr *shdr, Elf_Data *data)
3781 {
3782         Elf_Data *symbols = obj->efile.symbols;
3783         const char *relo_sec_name, *sec_name;
3784         size_t sec_idx = shdr->sh_info;
3785         struct bpf_program *prog;
3786         struct reloc_desc *relos;
3787         int err, i, nrels;
3788         const char *sym_name;
3789         __u32 insn_idx;
3790         Elf_Scn *scn;
3791         Elf_Data *scn_data;
3792         GElf_Sym sym;
3793         GElf_Rel rel;
3794
3795         scn = elf_sec_by_idx(obj, sec_idx);
3796         scn_data = elf_sec_data(obj, scn);
3797
3798         relo_sec_name = elf_sec_str(obj, shdr->sh_name);
3799         sec_name = elf_sec_name(obj, scn);
3800         if (!relo_sec_name || !sec_name)
3801                 return -EINVAL;
3802
3803         pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
3804                  relo_sec_name, sec_idx, sec_name);
3805         nrels = shdr->sh_size / shdr->sh_entsize;
3806
3807         for (i = 0; i < nrels; i++) {
3808                 if (!gelf_getrel(data, i, &rel)) {
3809                         pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
3810                         return -LIBBPF_ERRNO__FORMAT;
3811                 }
3812                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
3813                         pr_warn("sec '%s': symbol 0x%zx not found for relo #%d\n",
3814                                 relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i);
3815                         return -LIBBPF_ERRNO__FORMAT;
3816                 }
3817
3818                 if (rel.r_offset % BPF_INSN_SZ || rel.r_offset >= scn_data->d_size) {
3819                         pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
3820                                 relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i);
3821                         return -LIBBPF_ERRNO__FORMAT;
3822                 }
3823
3824                 insn_idx = rel.r_offset / BPF_INSN_SZ;
3825                 /* relocations against static functions are recorded as
3826                  * relocations against the section that contains a function;
3827                  * in such case, symbol will be STT_SECTION and sym.st_name
3828                  * will point to empty string (0), so fetch section name
3829                  * instead
3830                  */
3831                 if (GELF_ST_TYPE(sym.st_info) == STT_SECTION && sym.st_name == 0)
3832                         sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym.st_shndx));
3833                 else
3834                         sym_name = elf_sym_str(obj, sym.st_name);
3835                 sym_name = sym_name ?: "<?";
3836
3837                 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
3838                          relo_sec_name, i, insn_idx, sym_name);
3839
3840                 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
3841                 if (!prog) {
3842                         pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
3843                                 relo_sec_name, i, sec_name, insn_idx);
3844                         continue;
3845                 }
3846
3847                 relos = libbpf_reallocarray(prog->reloc_desc,
3848                                             prog->nr_reloc + 1, sizeof(*relos));
3849                 if (!relos)
3850                         return -ENOMEM;
3851                 prog->reloc_desc = relos;
3852
3853                 /* adjust insn_idx to local BPF program frame of reference */
3854                 insn_idx -= prog->sec_insn_off;
3855                 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
3856                                                 insn_idx, sym_name, &sym, &rel);
3857                 if (err)
3858                         return err;
3859
3860                 prog->nr_reloc++;
3861         }
3862         return 0;
3863 }
3864
3865 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
3866 {
3867         struct bpf_map_def *def = &map->def;
3868         __u32 key_type_id = 0, value_type_id = 0;
3869         int ret;
3870
3871         /* if it's BTF-defined map, we don't need to search for type IDs.
3872          * For struct_ops map, it does not need btf_key_type_id and
3873          * btf_value_type_id.
3874          */
3875         if (map->sec_idx == obj->efile.btf_maps_shndx ||
3876             bpf_map__is_struct_ops(map))
3877                 return 0;
3878
3879         if (!bpf_map__is_internal(map)) {
3880                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
3881                                            def->value_size, &key_type_id,
3882                                            &value_type_id);
3883         } else {
3884                 /*
3885                  * LLVM annotates global data differently in BTF, that is,
3886                  * only as '.data', '.bss' or '.rodata'.
3887                  */
3888                 ret = btf__find_by_name(obj->btf,
3889                                 libbpf_type_to_btf_name[map->libbpf_type]);
3890         }
3891         if (ret < 0)
3892                 return ret;
3893
3894         map->btf_key_type_id = key_type_id;
3895         map->btf_value_type_id = bpf_map__is_internal(map) ?
3896                                  ret : value_type_id;
3897         return 0;
3898 }
3899
3900 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
3901 {
3902         char file[PATH_MAX], buff[4096];
3903         FILE *fp;
3904         __u32 val;
3905         int err;
3906
3907         snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
3908         memset(info, 0, sizeof(*info));
3909
3910         fp = fopen(file, "r");
3911         if (!fp) {
3912                 err = -errno;
3913                 pr_warn("failed to open %s: %d. No procfs support?\n", file,
3914                         err);
3915                 return err;
3916         }
3917
3918         while (fgets(buff, sizeof(buff), fp)) {
3919                 if (sscanf(buff, "map_type:\t%u", &val) == 1)
3920                         info->type = val;
3921                 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
3922                         info->key_size = val;
3923                 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
3924                         info->value_size = val;
3925                 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
3926                         info->max_entries = val;
3927                 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
3928                         info->map_flags = val;
3929         }
3930
3931         fclose(fp);
3932
3933         return 0;
3934 }
3935
3936 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3937 {
3938         struct bpf_map_info info = {};
3939         __u32 len = sizeof(info);
3940         int new_fd, err;
3941         char *new_name;
3942
3943         err = bpf_obj_get_info_by_fd(fd, &info, &len);
3944         if (err && errno == EINVAL)
3945                 err = bpf_get_map_info_from_fdinfo(fd, &info);
3946         if (err)
3947                 return libbpf_err(err);
3948
3949         new_name = strdup(info.name);
3950         if (!new_name)
3951                 return libbpf_err(-errno);
3952
3953         new_fd = open("/", O_RDONLY | O_CLOEXEC);
3954         if (new_fd < 0) {
3955                 err = -errno;
3956                 goto err_free_new_name;
3957         }
3958
3959         new_fd = dup3(fd, new_fd, O_CLOEXEC);
3960         if (new_fd < 0) {
3961                 err = -errno;
3962                 goto err_close_new_fd;
3963         }
3964
3965         err = zclose(map->fd);
3966         if (err) {
3967                 err = -errno;
3968                 goto err_close_new_fd;
3969         }
3970         free(map->name);
3971
3972         map->fd = new_fd;
3973         map->name = new_name;
3974         map->def.type = info.type;
3975         map->def.key_size = info.key_size;
3976         map->def.value_size = info.value_size;
3977         map->def.max_entries = info.max_entries;
3978         map->def.map_flags = info.map_flags;
3979         map->btf_key_type_id = info.btf_key_type_id;
3980         map->btf_value_type_id = info.btf_value_type_id;
3981         map->reused = true;
3982
3983         return 0;
3984
3985 err_close_new_fd:
3986         close(new_fd);
3987 err_free_new_name:
3988         free(new_name);
3989         return libbpf_err(err);
3990 }
3991
3992 __u32 bpf_map__max_entries(const struct bpf_map *map)
3993 {
3994         return map->def.max_entries;
3995 }
3996
3997 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
3998 {
3999         if (!bpf_map_type__is_map_in_map(map->def.type))
4000                 return errno = EINVAL, NULL;
4001
4002         return map->inner_map;
4003 }
4004
4005 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4006 {
4007         if (map->fd >= 0)
4008                 return libbpf_err(-EBUSY);
4009         map->def.max_entries = max_entries;
4010         return 0;
4011 }
4012
4013 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
4014 {
4015         if (!map || !max_entries)
4016                 return libbpf_err(-EINVAL);
4017
4018         return bpf_map__set_max_entries(map, max_entries);
4019 }
4020
4021 static int
4022 bpf_object__probe_loading(struct bpf_object *obj)
4023 {
4024         struct bpf_load_program_attr attr;
4025         char *cp, errmsg[STRERR_BUFSIZE];
4026         struct bpf_insn insns[] = {
4027                 BPF_MOV64_IMM(BPF_REG_0, 0),
4028                 BPF_EXIT_INSN(),
4029         };
4030         int ret;
4031
4032         if (obj->gen_loader)
4033                 return 0;
4034
4035         /* make sure basic loading works */
4036
4037         memset(&attr, 0, sizeof(attr));
4038         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4039         attr.insns = insns;
4040         attr.insns_cnt = ARRAY_SIZE(insns);
4041         attr.license = "GPL";
4042
4043         ret = bpf_load_program_xattr(&attr, NULL, 0);
4044         if (ret < 0) {
4045                 attr.prog_type = BPF_PROG_TYPE_TRACEPOINT;
4046                 ret = bpf_load_program_xattr(&attr, NULL, 0);
4047         }
4048         if (ret < 0) {
4049                 ret = errno;
4050                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4051                 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4052                         "program. Make sure your kernel supports BPF "
4053                         "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4054                         "set to big enough value.\n", __func__, cp, ret);
4055                 return -ret;
4056         }
4057         close(ret);
4058
4059         return 0;
4060 }
4061
4062 static int probe_fd(int fd)
4063 {
4064         if (fd >= 0)
4065                 close(fd);
4066         return fd >= 0;
4067 }
4068
4069 static int probe_kern_prog_name(void)
4070 {
4071         struct bpf_load_program_attr attr;
4072         struct bpf_insn insns[] = {
4073                 BPF_MOV64_IMM(BPF_REG_0, 0),
4074                 BPF_EXIT_INSN(),
4075         };
4076         int ret;
4077
4078         /* make sure loading with name works */
4079
4080         memset(&attr, 0, sizeof(attr));
4081         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4082         attr.insns = insns;
4083         attr.insns_cnt = ARRAY_SIZE(insns);
4084         attr.license = "GPL";
4085         attr.name = "test";
4086         ret = bpf_load_program_xattr(&attr, NULL, 0);
4087         return probe_fd(ret);
4088 }
4089
4090 static int probe_kern_global_data(void)
4091 {
4092         struct bpf_load_program_attr prg_attr;
4093         struct bpf_create_map_attr map_attr;
4094         char *cp, errmsg[STRERR_BUFSIZE];
4095         struct bpf_insn insns[] = {
4096                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4097                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4098                 BPF_MOV64_IMM(BPF_REG_0, 0),
4099                 BPF_EXIT_INSN(),
4100         };
4101         int ret, map;
4102
4103         memset(&map_attr, 0, sizeof(map_attr));
4104         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
4105         map_attr.key_size = sizeof(int);
4106         map_attr.value_size = 32;
4107         map_attr.max_entries = 1;
4108
4109         map = bpf_create_map_xattr(&map_attr);
4110         if (map < 0) {
4111                 ret = -errno;
4112                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4113                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4114                         __func__, cp, -ret);
4115                 return ret;
4116         }
4117
4118         insns[0].imm = map;
4119
4120         memset(&prg_attr, 0, sizeof(prg_attr));
4121         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4122         prg_attr.insns = insns;
4123         prg_attr.insns_cnt = ARRAY_SIZE(insns);
4124         prg_attr.license = "GPL";
4125
4126         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
4127         close(map);
4128         return probe_fd(ret);
4129 }
4130
4131 static int probe_kern_btf(void)
4132 {
4133         static const char strs[] = "\0int";
4134         __u32 types[] = {
4135                 /* int */
4136                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4137         };
4138
4139         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4140                                              strs, sizeof(strs)));
4141 }
4142
4143 static int probe_kern_btf_func(void)
4144 {
4145         static const char strs[] = "\0int\0x\0a";
4146         /* void x(int a) {} */
4147         __u32 types[] = {
4148                 /* int */
4149                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4150                 /* FUNC_PROTO */                                /* [2] */
4151                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4152                 BTF_PARAM_ENC(7, 1),
4153                 /* FUNC x */                                    /* [3] */
4154                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4155         };
4156
4157         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4158                                              strs, sizeof(strs)));
4159 }
4160
4161 static int probe_kern_btf_func_global(void)
4162 {
4163         static const char strs[] = "\0int\0x\0a";
4164         /* static void x(int a) {} */
4165         __u32 types[] = {
4166                 /* int */
4167                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4168                 /* FUNC_PROTO */                                /* [2] */
4169                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4170                 BTF_PARAM_ENC(7, 1),
4171                 /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
4172                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4173         };
4174
4175         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4176                                              strs, sizeof(strs)));
4177 }
4178
4179 static int probe_kern_btf_datasec(void)
4180 {
4181         static const char strs[] = "\0x\0.data";
4182         /* static int a; */
4183         __u32 types[] = {
4184                 /* int */
4185                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4186                 /* VAR x */                                     /* [2] */
4187                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4188                 BTF_VAR_STATIC,
4189                 /* DATASEC val */                               /* [3] */
4190                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4191                 BTF_VAR_SECINFO_ENC(2, 0, 4),
4192         };
4193
4194         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4195                                              strs, sizeof(strs)));
4196 }
4197
4198 static int probe_kern_btf_float(void)
4199 {
4200         static const char strs[] = "\0float";
4201         __u32 types[] = {
4202                 /* float */
4203                 BTF_TYPE_FLOAT_ENC(1, 4),
4204         };
4205
4206         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4207                                              strs, sizeof(strs)));
4208 }
4209
4210 static int probe_kern_array_mmap(void)
4211 {
4212         struct bpf_create_map_attr attr = {
4213                 .map_type = BPF_MAP_TYPE_ARRAY,
4214                 .map_flags = BPF_F_MMAPABLE,
4215                 .key_size = sizeof(int),
4216                 .value_size = sizeof(int),
4217                 .max_entries = 1,
4218         };
4219
4220         return probe_fd(bpf_create_map_xattr(&attr));
4221 }
4222
4223 static int probe_kern_exp_attach_type(void)
4224 {
4225         struct bpf_load_program_attr attr;
4226         struct bpf_insn insns[] = {
4227                 BPF_MOV64_IMM(BPF_REG_0, 0),
4228                 BPF_EXIT_INSN(),
4229         };
4230
4231         memset(&attr, 0, sizeof(attr));
4232         /* use any valid combination of program type and (optional)
4233          * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
4234          * to see if kernel supports expected_attach_type field for
4235          * BPF_PROG_LOAD command
4236          */
4237         attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
4238         attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE;
4239         attr.insns = insns;
4240         attr.insns_cnt = ARRAY_SIZE(insns);
4241         attr.license = "GPL";
4242
4243         return probe_fd(bpf_load_program_xattr(&attr, NULL, 0));
4244 }
4245
4246 static int probe_kern_probe_read_kernel(void)
4247 {
4248         struct bpf_load_program_attr attr;
4249         struct bpf_insn insns[] = {
4250                 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),   /* r1 = r10 (fp) */
4251                 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8),  /* r1 += -8 */
4252                 BPF_MOV64_IMM(BPF_REG_2, 8),            /* r2 = 8 */
4253                 BPF_MOV64_IMM(BPF_REG_3, 0),            /* r3 = 0 */
4254                 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
4255                 BPF_EXIT_INSN(),
4256         };
4257
4258         memset(&attr, 0, sizeof(attr));
4259         attr.prog_type = BPF_PROG_TYPE_KPROBE;
4260         attr.insns = insns;
4261         attr.insns_cnt = ARRAY_SIZE(insns);
4262         attr.license = "GPL";
4263
4264         return probe_fd(bpf_load_program_xattr(&attr, NULL, 0));
4265 }
4266
4267 static int probe_prog_bind_map(void)
4268 {
4269         struct bpf_load_program_attr prg_attr;
4270         struct bpf_create_map_attr map_attr;
4271         char *cp, errmsg[STRERR_BUFSIZE];
4272         struct bpf_insn insns[] = {
4273                 BPF_MOV64_IMM(BPF_REG_0, 0),
4274                 BPF_EXIT_INSN(),
4275         };
4276         int ret, map, prog;
4277
4278         memset(&map_attr, 0, sizeof(map_attr));
4279         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
4280         map_attr.key_size = sizeof(int);
4281         map_attr.value_size = 32;
4282         map_attr.max_entries = 1;
4283
4284         map = bpf_create_map_xattr(&map_attr);
4285         if (map < 0) {
4286                 ret = -errno;
4287                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4288                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4289                         __func__, cp, -ret);
4290                 return ret;
4291         }
4292
4293         memset(&prg_attr, 0, sizeof(prg_attr));
4294         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4295         prg_attr.insns = insns;
4296         prg_attr.insns_cnt = ARRAY_SIZE(insns);
4297         prg_attr.license = "GPL";
4298
4299         prog = bpf_load_program_xattr(&prg_attr, NULL, 0);
4300         if (prog < 0) {
4301                 close(map);
4302                 return 0;
4303         }
4304
4305         ret = bpf_prog_bind_map(prog, map, NULL);
4306
4307         close(map);
4308         close(prog);
4309
4310         return ret >= 0;
4311 }
4312
4313 static int probe_module_btf(void)
4314 {
4315         static const char strs[] = "\0int";
4316         __u32 types[] = {
4317                 /* int */
4318                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4319         };
4320         struct bpf_btf_info info;
4321         __u32 len = sizeof(info);
4322         char name[16];
4323         int fd, err;
4324
4325         fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4326         if (fd < 0)
4327                 return 0; /* BTF not supported at all */
4328
4329         memset(&info, 0, sizeof(info));
4330         info.name = ptr_to_u64(name);
4331         info.name_len = sizeof(name);
4332
4333         /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4334          * kernel's module BTF support coincides with support for
4335          * name/name_len fields in struct bpf_btf_info.
4336          */
4337         err = bpf_obj_get_info_by_fd(fd, &info, &len);
4338         close(fd);
4339         return !err;
4340 }
4341
4342 static int probe_perf_link(void)
4343 {
4344         struct bpf_load_program_attr attr;
4345         struct bpf_insn insns[] = {
4346                 BPF_MOV64_IMM(BPF_REG_0, 0),
4347                 BPF_EXIT_INSN(),
4348         };
4349         int prog_fd, link_fd, err;
4350
4351         memset(&attr, 0, sizeof(attr));
4352         attr.prog_type = BPF_PROG_TYPE_TRACEPOINT;
4353         attr.insns = insns;
4354         attr.insns_cnt = ARRAY_SIZE(insns);
4355         attr.license = "GPL";
4356         prog_fd = bpf_load_program_xattr(&attr, NULL, 0);
4357         if (prog_fd < 0)
4358                 return -errno;
4359
4360         /* use invalid perf_event FD to get EBADF, if link is supported;
4361          * otherwise EINVAL should be returned
4362          */
4363         link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
4364         err = -errno; /* close() can clobber errno */
4365
4366         if (link_fd >= 0)
4367                 close(link_fd);
4368         close(prog_fd);
4369
4370         return link_fd < 0 && err == -EBADF;
4371 }
4372
4373 enum kern_feature_result {
4374         FEAT_UNKNOWN = 0,
4375         FEAT_SUPPORTED = 1,
4376         FEAT_MISSING = 2,
4377 };
4378
4379 typedef int (*feature_probe_fn)(void);
4380
4381 static struct kern_feature_desc {
4382         const char *desc;
4383         feature_probe_fn probe;
4384         enum kern_feature_result res;
4385 } feature_probes[__FEAT_CNT] = {
4386         [FEAT_PROG_NAME] = {
4387                 "BPF program name", probe_kern_prog_name,
4388         },
4389         [FEAT_GLOBAL_DATA] = {
4390                 "global variables", probe_kern_global_data,
4391         },
4392         [FEAT_BTF] = {
4393                 "minimal BTF", probe_kern_btf,
4394         },
4395         [FEAT_BTF_FUNC] = {
4396                 "BTF functions", probe_kern_btf_func,
4397         },
4398         [FEAT_BTF_GLOBAL_FUNC] = {
4399                 "BTF global function", probe_kern_btf_func_global,
4400         },
4401         [FEAT_BTF_DATASEC] = {
4402                 "BTF data section and variable", probe_kern_btf_datasec,
4403         },
4404         [FEAT_ARRAY_MMAP] = {
4405                 "ARRAY map mmap()", probe_kern_array_mmap,
4406         },
4407         [FEAT_EXP_ATTACH_TYPE] = {
4408                 "BPF_PROG_LOAD expected_attach_type attribute",
4409                 probe_kern_exp_attach_type,
4410         },
4411         [FEAT_PROBE_READ_KERN] = {
4412                 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
4413         },
4414         [FEAT_PROG_BIND_MAP] = {
4415                 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
4416         },
4417         [FEAT_MODULE_BTF] = {
4418                 "module BTF support", probe_module_btf,
4419         },
4420         [FEAT_BTF_FLOAT] = {
4421                 "BTF_KIND_FLOAT support", probe_kern_btf_float,
4422         },
4423         [FEAT_PERF_LINK] = {
4424                 "BPF perf link support", probe_perf_link,
4425         },
4426 };
4427
4428 static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
4429 {
4430         struct kern_feature_desc *feat = &feature_probes[feat_id];
4431         int ret;
4432
4433         if (obj->gen_loader)
4434                 /* To generate loader program assume the latest kernel
4435                  * to avoid doing extra prog_load, map_create syscalls.
4436                  */
4437                 return true;
4438
4439         if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
4440                 ret = feat->probe();
4441                 if (ret > 0) {
4442                         WRITE_ONCE(feat->res, FEAT_SUPPORTED);
4443                 } else if (ret == 0) {
4444                         WRITE_ONCE(feat->res, FEAT_MISSING);
4445                 } else {
4446                         pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
4447                         WRITE_ONCE(feat->res, FEAT_MISSING);
4448                 }
4449         }
4450
4451         return READ_ONCE(feat->res) == FEAT_SUPPORTED;
4452 }
4453
4454 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
4455 {
4456         struct bpf_map_info map_info = {};
4457         char msg[STRERR_BUFSIZE];
4458         __u32 map_info_len;
4459         int err;
4460
4461         map_info_len = sizeof(map_info);
4462
4463         err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len);
4464         if (err && errno == EINVAL)
4465                 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
4466         if (err) {
4467                 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
4468                         libbpf_strerror_r(errno, msg, sizeof(msg)));
4469                 return false;
4470         }
4471
4472         return (map_info.type == map->def.type &&
4473                 map_info.key_size == map->def.key_size &&
4474                 map_info.value_size == map->def.value_size &&
4475                 map_info.max_entries == map->def.max_entries &&
4476                 map_info.map_flags == map->def.map_flags);
4477 }
4478
4479 static int
4480 bpf_object__reuse_map(struct bpf_map *map)
4481 {
4482         char *cp, errmsg[STRERR_BUFSIZE];
4483         int err, pin_fd;
4484
4485         pin_fd = bpf_obj_get(map->pin_path);
4486         if (pin_fd < 0) {
4487                 err = -errno;
4488                 if (err == -ENOENT) {
4489                         pr_debug("found no pinned map to reuse at '%s'\n",
4490                                  map->pin_path);
4491                         return 0;
4492                 }
4493
4494                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4495                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
4496                         map->pin_path, cp);
4497                 return err;
4498         }
4499
4500         if (!map_is_reuse_compat(map, pin_fd)) {
4501                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
4502                         map->pin_path);
4503                 close(pin_fd);
4504                 return -EINVAL;
4505         }
4506
4507         err = bpf_map__reuse_fd(map, pin_fd);
4508         if (err) {
4509                 close(pin_fd);
4510                 return err;
4511         }
4512         map->pinned = true;
4513         pr_debug("reused pinned map at '%s'\n", map->pin_path);
4514
4515         return 0;
4516 }
4517
4518 static int
4519 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
4520 {
4521         enum libbpf_map_type map_type = map->libbpf_type;
4522         char *cp, errmsg[STRERR_BUFSIZE];
4523         int err, zero = 0;
4524
4525         if (obj->gen_loader) {
4526                 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
4527                                          map->mmaped, map->def.value_size);
4528                 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
4529                         bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
4530                 return 0;
4531         }
4532         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
4533         if (err) {
4534                 err = -errno;
4535                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4536                 pr_warn("Error setting initial map(%s) contents: %s\n",
4537                         map->name, cp);
4538                 return err;
4539         }
4540
4541         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
4542         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
4543                 err = bpf_map_freeze(map->fd);
4544                 if (err) {
4545                         err = -errno;
4546                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4547                         pr_warn("Error freezing map(%s) as read-only: %s\n",
4548                                 map->name, cp);
4549                         return err;
4550                 }
4551         }
4552         return 0;
4553 }
4554
4555 static void bpf_map__destroy(struct bpf_map *map);
4556
4557 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
4558 {
4559         struct bpf_create_map_attr create_attr;
4560         struct bpf_map_def *def = &map->def;
4561         int err = 0;
4562
4563         memset(&create_attr, 0, sizeof(create_attr));
4564
4565         if (kernel_supports(obj, FEAT_PROG_NAME))
4566                 create_attr.name = map->name;
4567         create_attr.map_ifindex = map->map_ifindex;
4568         create_attr.map_type = def->type;
4569         create_attr.map_flags = def->map_flags;
4570         create_attr.key_size = def->key_size;
4571         create_attr.value_size = def->value_size;
4572         create_attr.numa_node = map->numa_node;
4573
4574         if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) {
4575                 int nr_cpus;
4576
4577                 nr_cpus = libbpf_num_possible_cpus();
4578                 if (nr_cpus < 0) {
4579                         pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
4580                                 map->name, nr_cpus);
4581                         return nr_cpus;
4582                 }
4583                 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
4584                 create_attr.max_entries = nr_cpus;
4585         } else {
4586                 create_attr.max_entries = def->max_entries;
4587         }
4588
4589         if (bpf_map__is_struct_ops(map))
4590                 create_attr.btf_vmlinux_value_type_id =
4591                         map->btf_vmlinux_value_type_id;
4592
4593         create_attr.btf_fd = 0;
4594         create_attr.btf_key_type_id = 0;
4595         create_attr.btf_value_type_id = 0;
4596         if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) {
4597                 create_attr.btf_fd = btf__fd(obj->btf);
4598                 create_attr.btf_key_type_id = map->btf_key_type_id;
4599                 create_attr.btf_value_type_id = map->btf_value_type_id;
4600         }
4601
4602         if (bpf_map_type__is_map_in_map(def->type)) {
4603                 if (map->inner_map) {
4604                         err = bpf_object__create_map(obj, map->inner_map, true);
4605                         if (err) {
4606                                 pr_warn("map '%s': failed to create inner map: %d\n",
4607                                         map->name, err);
4608                                 return err;
4609                         }
4610                         map->inner_map_fd = bpf_map__fd(map->inner_map);
4611                 }
4612                 if (map->inner_map_fd >= 0)
4613                         create_attr.inner_map_fd = map->inner_map_fd;
4614         }
4615
4616         if (obj->gen_loader) {
4617                 bpf_gen__map_create(obj->gen_loader, &create_attr, is_inner ? -1 : map - obj->maps);
4618                 /* Pretend to have valid FD to pass various fd >= 0 checks.
4619                  * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
4620                  */
4621                 map->fd = 0;
4622         } else {
4623                 map->fd = bpf_create_map_xattr(&create_attr);
4624         }
4625         if (map->fd < 0 && (create_attr.btf_key_type_id ||
4626                             create_attr.btf_value_type_id)) {
4627                 char *cp, errmsg[STRERR_BUFSIZE];
4628
4629                 err = -errno;
4630                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4631                 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
4632                         map->name, cp, err);
4633                 create_attr.btf_fd = 0;
4634                 create_attr.btf_key_type_id = 0;
4635                 create_attr.btf_value_type_id = 0;
4636                 map->btf_key_type_id = 0;
4637                 map->btf_value_type_id = 0;
4638                 map->fd = bpf_create_map_xattr(&create_attr);
4639         }
4640
4641         err = map->fd < 0 ? -errno : 0;
4642
4643         if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
4644                 if (obj->gen_loader)
4645                         map->inner_map->fd = -1;
4646                 bpf_map__destroy(map->inner_map);
4647                 zfree(&map->inner_map);
4648         }
4649
4650         return err;
4651 }
4652
4653 static int init_map_slots(struct bpf_object *obj, struct bpf_map *map)
4654 {
4655         const struct bpf_map *targ_map;
4656         unsigned int i;
4657         int fd, err = 0;
4658
4659         for (i = 0; i < map->init_slots_sz; i++) {
4660                 if (!map->init_slots[i])
4661                         continue;
4662
4663                 targ_map = map->init_slots[i];
4664                 fd = bpf_map__fd(targ_map);
4665                 if (obj->gen_loader) {
4666                         pr_warn("// TODO map_update_elem: idx %td key %d value==map_idx %td\n",
4667                                 map - obj->maps, i, targ_map - obj->maps);
4668                         return -ENOTSUP;
4669                 } else {
4670                         err = bpf_map_update_elem(map->fd, &i, &fd, 0);
4671                 }
4672                 if (err) {
4673                         err = -errno;
4674                         pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
4675                                 map->name, i, targ_map->name,
4676                                 fd, err);
4677                         return err;
4678                 }
4679                 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
4680                          map->name, i, targ_map->name, fd);
4681         }
4682
4683         zfree(&map->init_slots);
4684         map->init_slots_sz = 0;
4685
4686         return 0;
4687 }
4688
4689 static int
4690 bpf_object__create_maps(struct bpf_object *obj)
4691 {
4692         struct bpf_map *map;
4693         char *cp, errmsg[STRERR_BUFSIZE];
4694         unsigned int i, j;
4695         int err;
4696         bool retried;
4697
4698         for (i = 0; i < obj->nr_maps; i++) {
4699                 map = &obj->maps[i];
4700
4701                 retried = false;
4702 retry:
4703                 if (map->pin_path) {
4704                         err = bpf_object__reuse_map(map);
4705                         if (err) {
4706                                 pr_warn("map '%s': error reusing pinned map\n",
4707                                         map->name);
4708                                 goto err_out;
4709                         }
4710                         if (retried && map->fd < 0) {
4711                                 pr_warn("map '%s': cannot find pinned map\n",
4712                                         map->name);
4713                                 err = -ENOENT;
4714                                 goto err_out;
4715                         }
4716                 }
4717
4718                 if (map->fd >= 0) {
4719                         pr_debug("map '%s': skipping creation (preset fd=%d)\n",
4720                                  map->name, map->fd);
4721                 } else {
4722                         err = bpf_object__create_map(obj, map, false);
4723                         if (err)
4724                                 goto err_out;
4725
4726                         pr_debug("map '%s': created successfully, fd=%d\n",
4727                                  map->name, map->fd);
4728
4729                         if (bpf_map__is_internal(map)) {
4730                                 err = bpf_object__populate_internal_map(obj, map);
4731                                 if (err < 0) {
4732                                         zclose(map->fd);
4733                                         goto err_out;
4734                                 }
4735                         }
4736
4737                         if (map->init_slots_sz) {
4738                                 err = init_map_slots(obj, map);
4739                                 if (err < 0) {
4740                                         zclose(map->fd);
4741                                         goto err_out;
4742                                 }
4743                         }
4744                 }
4745
4746                 if (map->pin_path && !map->pinned) {
4747                         err = bpf_map__pin(map, NULL);
4748                         if (err) {
4749                                 zclose(map->fd);
4750                                 if (!retried && err == -EEXIST) {
4751                                         retried = true;
4752                                         goto retry;
4753                                 }
4754                                 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
4755                                         map->name, map->pin_path, err);
4756                                 goto err_out;
4757                         }
4758                 }
4759         }
4760
4761         return 0;
4762
4763 err_out:
4764         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4765         pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
4766         pr_perm_msg(err);
4767         for (j = 0; j < i; j++)
4768                 zclose(obj->maps[j].fd);
4769         return err;
4770 }
4771
4772 static bool bpf_core_is_flavor_sep(const char *s)
4773 {
4774         /* check X___Y name pattern, where X and Y are not underscores */
4775         return s[0] != '_' &&                                 /* X */
4776                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
4777                s[4] != '_';                                   /* Y */
4778 }
4779
4780 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
4781  * before last triple underscore. Struct name part after last triple
4782  * underscore is ignored by BPF CO-RE relocation during relocation matching.
4783  */
4784 size_t bpf_core_essential_name_len(const char *name)
4785 {
4786         size_t n = strlen(name);
4787         int i;
4788
4789         for (i = n - 5; i >= 0; i--) {
4790                 if (bpf_core_is_flavor_sep(name + i))
4791                         return i + 1;
4792         }
4793         return n;
4794 }
4795
4796 static void bpf_core_free_cands(struct bpf_core_cand_list *cands)
4797 {
4798         free(cands->cands);
4799         free(cands);
4800 }
4801
4802 static int bpf_core_add_cands(struct bpf_core_cand *local_cand,
4803                               size_t local_essent_len,
4804                               const struct btf *targ_btf,
4805                               const char *targ_btf_name,
4806                               int targ_start_id,
4807                               struct bpf_core_cand_list *cands)
4808 {
4809         struct bpf_core_cand *new_cands, *cand;
4810         const struct btf_type *t;
4811         const char *targ_name;
4812         size_t targ_essent_len;
4813         int n, i;
4814
4815         n = btf__get_nr_types(targ_btf);
4816         for (i = targ_start_id; i <= n; i++) {
4817                 t = btf__type_by_id(targ_btf, i);
4818                 if (btf_kind(t) != btf_kind(local_cand->t))
4819                         continue;
4820
4821                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
4822                 if (str_is_empty(targ_name))
4823                         continue;
4824
4825                 targ_essent_len = bpf_core_essential_name_len(targ_name);
4826                 if (targ_essent_len != local_essent_len)
4827                         continue;
4828
4829                 if (strncmp(local_cand->name, targ_name, local_essent_len) != 0)
4830                         continue;
4831
4832                 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
4833                          local_cand->id, btf_kind_str(local_cand->t),
4834                          local_cand->name, i, btf_kind_str(t), targ_name,
4835                          targ_btf_name);
4836                 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
4837                                               sizeof(*cands->cands));
4838                 if (!new_cands)
4839                         return -ENOMEM;
4840
4841                 cand = &new_cands[cands->len];
4842                 cand->btf = targ_btf;
4843                 cand->t = t;
4844                 cand->name = targ_name;
4845                 cand->id = i;
4846
4847                 cands->cands = new_cands;
4848                 cands->len++;
4849         }
4850         return 0;
4851 }
4852
4853 static int load_module_btfs(struct bpf_object *obj)
4854 {
4855         struct bpf_btf_info info;
4856         struct module_btf *mod_btf;
4857         struct btf *btf;
4858         char name[64];
4859         __u32 id = 0, len;
4860         int err, fd;
4861
4862         if (obj->btf_modules_loaded)
4863                 return 0;
4864
4865         if (obj->gen_loader)
4866                 return 0;
4867
4868         /* don't do this again, even if we find no module BTFs */
4869         obj->btf_modules_loaded = true;
4870
4871         /* kernel too old to support module BTFs */
4872         if (!kernel_supports(obj, FEAT_MODULE_BTF))
4873                 return 0;
4874
4875         while (true) {
4876                 err = bpf_btf_get_next_id(id, &id);
4877                 if (err && errno == ENOENT)
4878                         return 0;
4879                 if (err) {
4880                         err = -errno;
4881                         pr_warn("failed to iterate BTF objects: %d\n", err);
4882                         return err;
4883                 }
4884
4885                 fd = bpf_btf_get_fd_by_id(id);
4886                 if (fd < 0) {
4887                         if (errno == ENOENT)
4888                                 continue; /* expected race: BTF was unloaded */
4889                         err = -errno;
4890                         pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
4891                         return err;
4892                 }
4893
4894                 len = sizeof(info);
4895                 memset(&info, 0, sizeof(info));
4896                 info.name = ptr_to_u64(name);
4897                 info.name_len = sizeof(name);
4898
4899                 err = bpf_obj_get_info_by_fd(fd, &info, &len);
4900                 if (err) {
4901                         err = -errno;
4902                         pr_warn("failed to get BTF object #%d info: %d\n", id, err);
4903                         goto err_out;
4904                 }
4905
4906                 /* ignore non-module BTFs */
4907                 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
4908                         close(fd);
4909                         continue;
4910                 }
4911
4912                 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
4913                 err = libbpf_get_error(btf);
4914                 if (err) {
4915                         pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
4916                                 name, id, err);
4917                         goto err_out;
4918                 }
4919
4920                 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
4921                                         sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
4922                 if (err)
4923                         goto err_out;
4924
4925                 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
4926
4927                 mod_btf->btf = btf;
4928                 mod_btf->id = id;
4929                 mod_btf->fd = fd;
4930                 mod_btf->name = strdup(name);
4931                 if (!mod_btf->name) {
4932                         err = -ENOMEM;
4933                         goto err_out;
4934                 }
4935                 continue;
4936
4937 err_out:
4938                 close(fd);
4939                 return err;
4940         }
4941
4942         return 0;
4943 }
4944
4945 static struct bpf_core_cand_list *
4946 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
4947 {
4948         struct bpf_core_cand local_cand = {};
4949         struct bpf_core_cand_list *cands;
4950         const struct btf *main_btf;
4951         size_t local_essent_len;
4952         int err, i;
4953
4954         local_cand.btf = local_btf;
4955         local_cand.t = btf__type_by_id(local_btf, local_type_id);
4956         if (!local_cand.t)
4957                 return ERR_PTR(-EINVAL);
4958
4959         local_cand.name = btf__name_by_offset(local_btf, local_cand.t->name_off);
4960         if (str_is_empty(local_cand.name))
4961                 return ERR_PTR(-EINVAL);
4962         local_essent_len = bpf_core_essential_name_len(local_cand.name);
4963
4964         cands = calloc(1, sizeof(*cands));
4965         if (!cands)
4966                 return ERR_PTR(-ENOMEM);
4967
4968         /* Attempt to find target candidates in vmlinux BTF first */
4969         main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
4970         err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
4971         if (err)
4972                 goto err_out;
4973
4974         /* if vmlinux BTF has any candidate, don't got for module BTFs */
4975         if (cands->len)
4976                 return cands;
4977
4978         /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
4979         if (obj->btf_vmlinux_override)
4980                 return cands;
4981
4982         /* now look through module BTFs, trying to still find candidates */
4983         err = load_module_btfs(obj);
4984         if (err)
4985                 goto err_out;
4986
4987         for (i = 0; i < obj->btf_module_cnt; i++) {
4988                 err = bpf_core_add_cands(&local_cand, local_essent_len,
4989                                          obj->btf_modules[i].btf,
4990                                          obj->btf_modules[i].name,
4991                                          btf__get_nr_types(obj->btf_vmlinux) + 1,
4992                                          cands);
4993                 if (err)
4994                         goto err_out;
4995         }
4996
4997         return cands;
4998 err_out:
4999         bpf_core_free_cands(cands);
5000         return ERR_PTR(err);
5001 }
5002
5003 /* Check local and target types for compatibility. This check is used for
5004  * type-based CO-RE relocations and follow slightly different rules than
5005  * field-based relocations. This function assumes that root types were already
5006  * checked for name match. Beyond that initial root-level name check, names
5007  * are completely ignored. Compatibility rules are as follows:
5008  *   - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5009  *     kind should match for local and target types (i.e., STRUCT is not
5010  *     compatible with UNION);
5011  *   - for ENUMs, the size is ignored;
5012  *   - for INT, size and signedness are ignored;
5013  *   - for ARRAY, dimensionality is ignored, element types are checked for
5014  *     compatibility recursively;
5015  *   - CONST/VOLATILE/RESTRICT modifiers are ignored;
5016  *   - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5017  *   - FUNC_PROTOs are compatible if they have compatible signature: same
5018  *     number of input args and compatible return and argument types.
5019  * These rules are not set in stone and probably will be adjusted as we get
5020  * more experience with using BPF CO-RE relocations.
5021  */
5022 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5023                               const struct btf *targ_btf, __u32 targ_id)
5024 {
5025         const struct btf_type *local_type, *targ_type;
5026         int depth = 32; /* max recursion depth */
5027
5028         /* caller made sure that names match (ignoring flavor suffix) */
5029         local_type = btf__type_by_id(local_btf, local_id);
5030         targ_type = btf__type_by_id(targ_btf, targ_id);
5031         if (btf_kind(local_type) != btf_kind(targ_type))
5032                 return 0;
5033
5034 recur:
5035         depth--;
5036         if (depth < 0)
5037                 return -EINVAL;
5038
5039         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
5040         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
5041         if (!local_type || !targ_type)
5042                 return -EINVAL;
5043
5044         if (btf_kind(local_type) != btf_kind(targ_type))
5045                 return 0;
5046
5047         switch (btf_kind(local_type)) {
5048         case BTF_KIND_UNKN:
5049         case BTF_KIND_STRUCT:
5050         case BTF_KIND_UNION:
5051         case BTF_KIND_ENUM:
5052         case BTF_KIND_FWD:
5053                 return 1;
5054         case BTF_KIND_INT:
5055                 /* just reject deprecated bitfield-like integers; all other
5056                  * integers are by default compatible between each other
5057                  */
5058                 return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0;
5059         case BTF_KIND_PTR:
5060                 local_id = local_type->type;
5061                 targ_id = targ_type->type;
5062                 goto recur;
5063         case BTF_KIND_ARRAY:
5064                 local_id = btf_array(local_type)->type;
5065                 targ_id = btf_array(targ_type)->type;
5066                 goto recur;
5067         case BTF_KIND_FUNC_PROTO: {
5068                 struct btf_param *local_p = btf_params(local_type);
5069                 struct btf_param *targ_p = btf_params(targ_type);
5070                 __u16 local_vlen = btf_vlen(local_type);
5071                 __u16 targ_vlen = btf_vlen(targ_type);
5072                 int i, err;
5073
5074                 if (local_vlen != targ_vlen)
5075                         return 0;
5076
5077                 for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
5078                         skip_mods_and_typedefs(local_btf, local_p->type, &local_id);
5079                         skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id);
5080                         err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id);
5081                         if (err <= 0)
5082                                 return err;
5083                 }
5084
5085                 /* tail recurse for return type check */
5086                 skip_mods_and_typedefs(local_btf, local_type->type, &local_id);
5087                 skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id);
5088                 goto recur;
5089         }
5090         default:
5091                 pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n",
5092                         btf_kind_str(local_type), local_id, targ_id);
5093                 return 0;
5094         }
5095 }
5096
5097 static size_t bpf_core_hash_fn(const void *key, void *ctx)
5098 {
5099         return (size_t)key;
5100 }
5101
5102 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
5103 {
5104         return k1 == k2;
5105 }
5106
5107 static void *u32_as_hash_key(__u32 x)
5108 {
5109         return (void *)(uintptr_t)x;
5110 }
5111
5112 static int bpf_core_apply_relo(struct bpf_program *prog,
5113                                const struct bpf_core_relo *relo,
5114                                int relo_idx,
5115                                const struct btf *local_btf,
5116                                struct hashmap *cand_cache)
5117 {
5118         const void *type_key = u32_as_hash_key(relo->type_id);
5119         struct bpf_core_cand_list *cands = NULL;
5120         const char *prog_name = prog->name;
5121         const struct btf_type *local_type;
5122         const char *local_name;
5123         __u32 local_id = relo->type_id;
5124         struct bpf_insn *insn;
5125         int insn_idx, err;
5126
5127         if (relo->insn_off % BPF_INSN_SZ)
5128                 return -EINVAL;
5129         insn_idx = relo->insn_off / BPF_INSN_SZ;
5130         /* adjust insn_idx from section frame of reference to the local
5131          * program's frame of reference; (sub-)program code is not yet
5132          * relocated, so it's enough to just subtract in-section offset
5133          */
5134         insn_idx = insn_idx - prog->sec_insn_off;
5135         if (insn_idx > prog->insns_cnt)
5136                 return -EINVAL;
5137         insn = &prog->insns[insn_idx];
5138
5139         local_type = btf__type_by_id(local_btf, local_id);
5140         if (!local_type)
5141                 return -EINVAL;
5142
5143         local_name = btf__name_by_offset(local_btf, local_type->name_off);
5144         if (!local_name)
5145                 return -EINVAL;
5146
5147         if (prog->obj->gen_loader) {
5148                 pr_warn("// TODO core_relo: prog %td insn[%d] %s kind %d\n",
5149                         prog - prog->obj->programs, relo->insn_off / 8,
5150                         local_name, relo->kind);
5151                 return -ENOTSUP;
5152         }
5153
5154         if (relo->kind != BPF_TYPE_ID_LOCAL &&
5155             !hashmap__find(cand_cache, type_key, (void **)&cands)) {
5156                 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5157                 if (IS_ERR(cands)) {
5158                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5159                                 prog_name, relo_idx, local_id, btf_kind_str(local_type),
5160                                 local_name, PTR_ERR(cands));
5161                         return PTR_ERR(cands);
5162                 }
5163                 err = hashmap__set(cand_cache, type_key, cands, NULL, NULL);
5164                 if (err) {
5165                         bpf_core_free_cands(cands);
5166                         return err;
5167                 }
5168         }
5169
5170         return bpf_core_apply_relo_insn(prog_name, insn, insn_idx, relo, relo_idx, local_btf, cands);
5171 }
5172
5173 static int
5174 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5175 {
5176         const struct btf_ext_info_sec *sec;
5177         const struct bpf_core_relo *rec;
5178         const struct btf_ext_info *seg;
5179         struct hashmap_entry *entry;
5180         struct hashmap *cand_cache = NULL;
5181         struct bpf_program *prog;
5182         const char *sec_name;
5183         int i, err = 0, insn_idx, sec_idx;
5184
5185         if (obj->btf_ext->core_relo_info.len == 0)
5186                 return 0;
5187
5188         if (targ_btf_path) {
5189                 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5190                 err = libbpf_get_error(obj->btf_vmlinux_override);
5191                 if (err) {
5192                         pr_warn("failed to parse target BTF: %d\n", err);
5193                         return err;
5194                 }
5195         }
5196
5197         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5198         if (IS_ERR(cand_cache)) {
5199                 err = PTR_ERR(cand_cache);
5200                 goto out;
5201         }
5202
5203         seg = &obj->btf_ext->core_relo_info;
5204         for_each_btf_ext_sec(seg, sec) {
5205                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5206                 if (str_is_empty(sec_name)) {
5207                         err = -EINVAL;
5208                         goto out;
5209                 }
5210                 /* bpf_object's ELF is gone by now so it's not easy to find
5211                  * section index by section name, but we can find *any*
5212                  * bpf_program within desired section name and use it's
5213                  * prog->sec_idx to do a proper search by section index and
5214                  * instruction offset
5215                  */
5216                 prog = NULL;
5217                 for (i = 0; i < obj->nr_programs; i++) {
5218                         prog = &obj->programs[i];
5219                         if (strcmp(prog->sec_name, sec_name) == 0)
5220                                 break;
5221                 }
5222                 if (!prog) {
5223                         pr_warn("sec '%s': failed to find a BPF program\n", sec_name);
5224                         return -ENOENT;
5225                 }
5226                 sec_idx = prog->sec_idx;
5227
5228                 pr_debug("sec '%s': found %d CO-RE relocations\n",
5229                          sec_name, sec->num_info);
5230
5231                 for_each_btf_ext_rec(seg, sec, i, rec) {
5232                         insn_idx = rec->insn_off / BPF_INSN_SZ;
5233                         prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5234                         if (!prog) {
5235                                 pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n",
5236                                         sec_name, insn_idx, i);
5237                                 err = -EINVAL;
5238                                 goto out;
5239                         }
5240                         /* no need to apply CO-RE relocation if the program is
5241                          * not going to be loaded
5242                          */
5243                         if (!prog->load)
5244                                 continue;
5245
5246                         err = bpf_core_apply_relo(prog, rec, i, obj->btf, cand_cache);
5247                         if (err) {
5248                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5249                                         prog->name, i, err);
5250                                 goto out;
5251                         }
5252                 }
5253         }
5254
5255 out:
5256         /* obj->btf_vmlinux and module BTFs are freed after object load */
5257         btf__free(obj->btf_vmlinux_override);
5258         obj->btf_vmlinux_override = NULL;
5259
5260         if (!IS_ERR_OR_NULL(cand_cache)) {
5261                 hashmap__for_each_entry(cand_cache, entry, i) {
5262                         bpf_core_free_cands(entry->value);
5263                 }
5264                 hashmap__free(cand_cache);
5265         }
5266         return err;
5267 }
5268
5269 /* Relocate data references within program code:
5270  *  - map references;
5271  *  - global variable references;
5272  *  - extern references.
5273  */
5274 static int
5275 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
5276 {
5277         int i;
5278
5279         for (i = 0; i < prog->nr_reloc; i++) {
5280                 struct reloc_desc *relo = &prog->reloc_desc[i];
5281                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5282                 struct extern_desc *ext;
5283
5284                 switch (relo->type) {
5285                 case RELO_LD64:
5286                         if (obj->gen_loader) {
5287                                 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
5288                                 insn[0].imm = relo->map_idx;
5289                         } else {
5290                                 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
5291                                 insn[0].imm = obj->maps[relo->map_idx].fd;
5292                         }
5293                         break;
5294                 case RELO_DATA:
5295                         insn[1].imm = insn[0].imm + relo->sym_off;
5296                         if (obj->gen_loader) {
5297                                 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5298                                 insn[0].imm = relo->map_idx;
5299                         } else {
5300                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5301                                 insn[0].imm = obj->maps[relo->map_idx].fd;
5302                         }
5303                         break;
5304                 case RELO_EXTERN_VAR:
5305                         ext = &obj->externs[relo->sym_off];
5306                         if (ext->type == EXT_KCFG) {
5307                                 if (obj->gen_loader) {
5308                                         insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
5309                                         insn[0].imm = obj->kconfig_map_idx;
5310                                 } else {
5311                                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
5312                                         insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
5313                                 }
5314                                 insn[1].imm = ext->kcfg.data_off;
5315                         } else /* EXT_KSYM */ {
5316                                 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
5317                                         insn[0].src_reg = BPF_PSEUDO_BTF_ID;
5318                                         insn[0].imm = ext->ksym.kernel_btf_id;
5319                                         insn[1].imm = ext->ksym.kernel_btf_obj_fd;
5320                                 } else { /* typeless ksyms or unresolved typed ksyms */
5321                                         insn[0].imm = (__u32)ext->ksym.addr;
5322                                         insn[1].imm = ext->ksym.addr >> 32;
5323                                 }
5324                         }
5325                         break;
5326                 case RELO_EXTERN_FUNC:
5327                         ext = &obj->externs[relo->sym_off];
5328                         insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
5329                         insn[0].imm = ext->ksym.kernel_btf_id;
5330                         break;
5331                 case RELO_SUBPROG_ADDR:
5332                         if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
5333                                 pr_warn("prog '%s': relo #%d: bad insn\n",
5334                                         prog->name, i);
5335                                 return -EINVAL;
5336                         }
5337                         /* handled already */
5338                         break;
5339                 case RELO_CALL:
5340                         /* handled already */
5341                         break;
5342                 default:
5343                         pr_warn("prog '%s': relo #%d: bad relo type %d\n",
5344                                 prog->name, i, relo->type);
5345                         return -EINVAL;
5346                 }
5347         }
5348
5349         return 0;
5350 }
5351
5352 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
5353                                     const struct bpf_program *prog,
5354                                     const struct btf_ext_info *ext_info,
5355                                     void **prog_info, __u32 *prog_rec_cnt,
5356                                     __u32 *prog_rec_sz)
5357 {
5358         void *copy_start = NULL, *copy_end = NULL;
5359         void *rec, *rec_end, *new_prog_info;
5360         const struct btf_ext_info_sec *sec;
5361         size_t old_sz, new_sz;
5362         const char *sec_name;
5363         int i, off_adj;
5364
5365         for_each_btf_ext_sec(ext_info, sec) {
5366                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5367                 if (!sec_name)
5368                         return -EINVAL;
5369                 if (strcmp(sec_name, prog->sec_name) != 0)
5370                         continue;
5371
5372                 for_each_btf_ext_rec(ext_info, sec, i, rec) {
5373                         __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
5374
5375                         if (insn_off < prog->sec_insn_off)
5376                                 continue;
5377                         if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
5378                                 break;
5379
5380                         if (!copy_start)
5381                                 copy_start = rec;
5382                         copy_end = rec + ext_info->rec_size;
5383                 }
5384
5385                 if (!copy_start)
5386                         return -ENOENT;
5387
5388                 /* append func/line info of a given (sub-)program to the main
5389                  * program func/line info
5390                  */
5391                 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
5392                 new_sz = old_sz + (copy_end - copy_start);
5393                 new_prog_info = realloc(*prog_info, new_sz);
5394                 if (!new_prog_info)
5395                         return -ENOMEM;
5396                 *prog_info = new_prog_info;
5397                 *prog_rec_cnt = new_sz / ext_info->rec_size;
5398                 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
5399
5400                 /* Kernel instruction offsets are in units of 8-byte
5401                  * instructions, while .BTF.ext instruction offsets generated
5402                  * by Clang are in units of bytes. So convert Clang offsets
5403                  * into kernel offsets and adjust offset according to program
5404                  * relocated position.
5405                  */
5406                 off_adj = prog->sub_insn_off - prog->sec_insn_off;
5407                 rec = new_prog_info + old_sz;
5408                 rec_end = new_prog_info + new_sz;
5409                 for (; rec < rec_end; rec += ext_info->rec_size) {
5410                         __u32 *insn_off = rec;
5411
5412                         *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
5413                 }
5414                 *prog_rec_sz = ext_info->rec_size;
5415                 return 0;
5416         }
5417
5418         return -ENOENT;
5419 }
5420
5421 static int
5422 reloc_prog_func_and_line_info(const struct bpf_object *obj,
5423                               struct bpf_program *main_prog,
5424                               const struct bpf_program *prog)
5425 {
5426         int err;
5427
5428         /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
5429          * supprot func/line info
5430          */
5431         if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
5432                 return 0;
5433
5434         /* only attempt func info relocation if main program's func_info
5435          * relocation was successful
5436          */
5437         if (main_prog != prog && !main_prog->func_info)
5438                 goto line_info;
5439
5440         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
5441                                        &main_prog->func_info,
5442                                        &main_prog->func_info_cnt,
5443                                        &main_prog->func_info_rec_size);
5444         if (err) {
5445                 if (err != -ENOENT) {
5446                         pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
5447                                 prog->name, err);
5448                         return err;
5449                 }
5450                 if (main_prog->func_info) {
5451                         /*
5452                          * Some info has already been found but has problem
5453                          * in the last btf_ext reloc. Must have to error out.
5454                          */
5455                         pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
5456                         return err;
5457                 }
5458                 /* Have problem loading the very first info. Ignore the rest. */
5459                 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
5460                         prog->name);
5461         }
5462
5463 line_info:
5464         /* don't relocate line info if main program's relocation failed */
5465         if (main_prog != prog && !main_prog->line_info)
5466                 return 0;
5467
5468         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
5469                                        &main_prog->line_info,
5470                                        &main_prog->line_info_cnt,
5471                                        &main_prog->line_info_rec_size);
5472         if (err) {
5473                 if (err != -ENOENT) {
5474                         pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
5475                                 prog->name, err);
5476                         return err;
5477                 }
5478                 if (main_prog->line_info) {
5479                         /*
5480                          * Some info has already been found but has problem
5481                          * in the last btf_ext reloc. Must have to error out.
5482                          */
5483                         pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
5484                         return err;
5485                 }
5486                 /* Have problem loading the very first info. Ignore the rest. */
5487                 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
5488                         prog->name);
5489         }
5490         return 0;
5491 }
5492
5493 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
5494 {
5495         size_t insn_idx = *(const size_t *)key;
5496         const struct reloc_desc *relo = elem;
5497
5498         if (insn_idx == relo->insn_idx)
5499                 return 0;
5500         return insn_idx < relo->insn_idx ? -1 : 1;
5501 }
5502
5503 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
5504 {
5505         return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
5506                        sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
5507 }
5508
5509 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
5510 {
5511         int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
5512         struct reloc_desc *relos;
5513         int i;
5514
5515         if (main_prog == subprog)
5516                 return 0;
5517         relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
5518         if (!relos)
5519                 return -ENOMEM;
5520         memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
5521                sizeof(*relos) * subprog->nr_reloc);
5522
5523         for (i = main_prog->nr_reloc; i < new_cnt; i++)
5524                 relos[i].insn_idx += subprog->sub_insn_off;
5525         /* After insn_idx adjustment the 'relos' array is still sorted
5526          * by insn_idx and doesn't break bsearch.
5527          */
5528         main_prog->reloc_desc = relos;
5529         main_prog->nr_reloc = new_cnt;
5530         return 0;
5531 }
5532
5533 static int
5534 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
5535                        struct bpf_program *prog)
5536 {
5537         size_t sub_insn_idx, insn_idx, new_cnt;
5538         struct bpf_program *subprog;
5539         struct bpf_insn *insns, *insn;
5540         struct reloc_desc *relo;
5541         int err;
5542
5543         err = reloc_prog_func_and_line_info(obj, main_prog, prog);
5544         if (err)
5545                 return err;
5546
5547         for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
5548                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
5549                 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
5550                         continue;
5551
5552                 relo = find_prog_insn_relo(prog, insn_idx);
5553                 if (relo && relo->type == RELO_EXTERN_FUNC)
5554                         /* kfunc relocations will be handled later
5555                          * in bpf_object__relocate_data()
5556                          */
5557                         continue;
5558                 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
5559                         pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
5560                                 prog->name, insn_idx, relo->type);
5561                         return -LIBBPF_ERRNO__RELOC;
5562                 }
5563                 if (relo) {
5564                         /* sub-program instruction index is a combination of
5565                          * an offset of a symbol pointed to by relocation and
5566                          * call instruction's imm field; for global functions,
5567                          * call always has imm = -1, but for static functions
5568                          * relocation is against STT_SECTION and insn->imm
5569                          * points to a start of a static function
5570                          *
5571                          * for subprog addr relocation, the relo->sym_off + insn->imm is
5572                          * the byte offset in the corresponding section.
5573                          */
5574                         if (relo->type == RELO_CALL)
5575                                 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
5576                         else
5577                                 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
5578                 } else if (insn_is_pseudo_func(insn)) {
5579                         /*
5580                          * RELO_SUBPROG_ADDR relo is always emitted even if both
5581                          * functions are in the same section, so it shouldn't reach here.
5582                          */
5583                         pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
5584                                 prog->name, insn_idx);
5585                         return -LIBBPF_ERRNO__RELOC;
5586                 } else {
5587                         /* if subprogram call is to a static function within
5588                          * the same ELF section, there won't be any relocation
5589                          * emitted, but it also means there is no additional
5590                          * offset necessary, insns->imm is relative to
5591                          * instruction's original position within the section
5592                          */
5593                         sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
5594                 }
5595
5596                 /* we enforce that sub-programs should be in .text section */
5597                 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
5598                 if (!subprog) {
5599                         pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
5600                                 prog->name);
5601                         return -LIBBPF_ERRNO__RELOC;
5602                 }
5603
5604                 /* if it's the first call instruction calling into this
5605                  * subprogram (meaning this subprog hasn't been processed
5606                  * yet) within the context of current main program:
5607                  *   - append it at the end of main program's instructions blog;
5608                  *   - process is recursively, while current program is put on hold;
5609                  *   - if that subprogram calls some other not yet processes
5610                  *   subprogram, same thing will happen recursively until
5611                  *   there are no more unprocesses subprograms left to append
5612                  *   and relocate.
5613                  */
5614                 if (subprog->sub_insn_off == 0) {
5615                         subprog->sub_insn_off = main_prog->insns_cnt;
5616
5617                         new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
5618                         insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
5619                         if (!insns) {
5620                                 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
5621                                 return -ENOMEM;
5622                         }
5623                         main_prog->insns = insns;
5624                         main_prog->insns_cnt = new_cnt;
5625
5626                         memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
5627                                subprog->insns_cnt * sizeof(*insns));
5628
5629                         pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
5630                                  main_prog->name, subprog->insns_cnt, subprog->name);
5631
5632                         /* The subprog insns are now appended. Append its relos too. */
5633                         err = append_subprog_relos(main_prog, subprog);
5634                         if (err)
5635                                 return err;
5636                         err = bpf_object__reloc_code(obj, main_prog, subprog);
5637                         if (err)
5638                                 return err;
5639                 }
5640
5641                 /* main_prog->insns memory could have been re-allocated, so
5642                  * calculate pointer again
5643                  */
5644                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
5645                 /* calculate correct instruction position within current main
5646                  * prog; each main prog can have a different set of
5647                  * subprograms appended (potentially in different order as
5648                  * well), so position of any subprog can be different for
5649                  * different main programs */
5650                 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
5651
5652                 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
5653                          prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
5654         }
5655
5656         return 0;
5657 }
5658
5659 /*
5660  * Relocate sub-program calls.
5661  *
5662  * Algorithm operates as follows. Each entry-point BPF program (referred to as
5663  * main prog) is processed separately. For each subprog (non-entry functions,
5664  * that can be called from either entry progs or other subprogs) gets their
5665  * sub_insn_off reset to zero. This serves as indicator that this subprogram
5666  * hasn't been yet appended and relocated within current main prog. Once its
5667  * relocated, sub_insn_off will point at the position within current main prog
5668  * where given subprog was appended. This will further be used to relocate all
5669  * the call instructions jumping into this subprog.
5670  *
5671  * We start with main program and process all call instructions. If the call
5672  * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
5673  * is zero), subprog instructions are appended at the end of main program's
5674  * instruction array. Then main program is "put on hold" while we recursively
5675  * process newly appended subprogram. If that subprogram calls into another
5676  * subprogram that hasn't been appended, new subprogram is appended again to
5677  * the *main* prog's instructions (subprog's instructions are always left
5678  * untouched, as they need to be in unmodified state for subsequent main progs
5679  * and subprog instructions are always sent only as part of a main prog) and
5680  * the process continues recursively. Once all the subprogs called from a main
5681  * prog or any of its subprogs are appended (and relocated), all their
5682  * positions within finalized instructions array are known, so it's easy to
5683  * rewrite call instructions with correct relative offsets, corresponding to
5684  * desired target subprog.
5685  *
5686  * Its important to realize that some subprogs might not be called from some
5687  * main prog and any of its called/used subprogs. Those will keep their
5688  * subprog->sub_insn_off as zero at all times and won't be appended to current
5689  * main prog and won't be relocated within the context of current main prog.
5690  * They might still be used from other main progs later.
5691  *
5692  * Visually this process can be shown as below. Suppose we have two main
5693  * programs mainA and mainB and BPF object contains three subprogs: subA,
5694  * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
5695  * subC both call subB:
5696  *
5697  *        +--------+ +-------+
5698  *        |        v v       |
5699  *     +--+---+ +--+-+-+ +---+--+
5700  *     | subA | | subB | | subC |
5701  *     +--+---+ +------+ +---+--+
5702  *        ^                  ^
5703  *        |                  |
5704  *    +---+-------+   +------+----+
5705  *    |   mainA   |   |   mainB   |
5706  *    +-----------+   +-----------+
5707  *
5708  * We'll start relocating mainA, will find subA, append it and start
5709  * processing sub A recursively:
5710  *
5711  *    +-----------+------+
5712  *    |   mainA   | subA |
5713  *    +-----------+------+
5714  *
5715  * At this point we notice that subB is used from subA, so we append it and
5716  * relocate (there are no further subcalls from subB):
5717  *
5718  *    +-----------+------+------+
5719  *    |   mainA   | subA | subB |
5720  *    +-----------+------+------+
5721  *
5722  * At this point, we relocate subA calls, then go one level up and finish with
5723  * relocatin mainA calls. mainA is done.
5724  *
5725  * For mainB process is similar but results in different order. We start with
5726  * mainB and skip subA and subB, as mainB never calls them (at least
5727  * directly), but we see subC is needed, so we append and start processing it:
5728  *
5729  *    +-----------+------+
5730  *    |   mainB   | subC |
5731  *    +-----------+------+
5732  * Now we see subC needs subB, so we go back to it, append and relocate it:
5733  *
5734  *    +-----------+------+------+
5735  *    |   mainB   | subC | subB |
5736  *    +-----------+------+------+
5737  *
5738  * At this point we unwind recursion, relocate calls in subC, then in mainB.
5739  */
5740 static int
5741 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
5742 {
5743         struct bpf_program *subprog;
5744         int i, err;
5745
5746         /* mark all subprogs as not relocated (yet) within the context of
5747          * current main program
5748          */
5749         for (i = 0; i < obj->nr_programs; i++) {
5750                 subprog = &obj->programs[i];
5751                 if (!prog_is_subprog(obj, subprog))
5752                         continue;
5753
5754                 subprog->sub_insn_off = 0;
5755         }
5756
5757         err = bpf_object__reloc_code(obj, prog, prog);
5758         if (err)
5759                 return err;
5760
5761
5762         return 0;
5763 }
5764
5765 static void
5766 bpf_object__free_relocs(struct bpf_object *obj)
5767 {
5768         struct bpf_program *prog;
5769         int i;
5770
5771         /* free up relocation descriptors */
5772         for (i = 0; i < obj->nr_programs; i++) {
5773                 prog = &obj->programs[i];
5774                 zfree(&prog->reloc_desc);
5775                 prog->nr_reloc = 0;
5776         }
5777 }
5778
5779 static int
5780 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
5781 {
5782         struct bpf_program *prog;
5783         size_t i, j;
5784         int err;
5785
5786         if (obj->btf_ext) {
5787                 err = bpf_object__relocate_core(obj, targ_btf_path);
5788                 if (err) {
5789                         pr_warn("failed to perform CO-RE relocations: %d\n",
5790                                 err);
5791                         return err;
5792                 }
5793         }
5794
5795         /* Before relocating calls pre-process relocations and mark
5796          * few ld_imm64 instructions that points to subprogs.
5797          * Otherwise bpf_object__reloc_code() later would have to consider
5798          * all ld_imm64 insns as relocation candidates. That would
5799          * reduce relocation speed, since amount of find_prog_insn_relo()
5800          * would increase and most of them will fail to find a relo.
5801          */
5802         for (i = 0; i < obj->nr_programs; i++) {
5803                 prog = &obj->programs[i];
5804                 for (j = 0; j < prog->nr_reloc; j++) {
5805                         struct reloc_desc *relo = &prog->reloc_desc[j];
5806                         struct bpf_insn *insn = &prog->insns[relo->insn_idx];
5807
5808                         /* mark the insn, so it's recognized by insn_is_pseudo_func() */
5809                         if (relo->type == RELO_SUBPROG_ADDR)
5810                                 insn[0].src_reg = BPF_PSEUDO_FUNC;
5811                 }
5812         }
5813
5814         /* relocate subprogram calls and append used subprograms to main
5815          * programs; each copy of subprogram code needs to be relocated
5816          * differently for each main program, because its code location might
5817          * have changed.
5818          * Append subprog relos to main programs to allow data relos to be
5819          * processed after text is completely relocated.
5820          */
5821         for (i = 0; i < obj->nr_programs; i++) {
5822                 prog = &obj->programs[i];
5823                 /* sub-program's sub-calls are relocated within the context of
5824                  * its main program only
5825                  */
5826                 if (prog_is_subprog(obj, prog))
5827                         continue;
5828
5829                 err = bpf_object__relocate_calls(obj, prog);
5830                 if (err) {
5831                         pr_warn("prog '%s': failed to relocate calls: %d\n",
5832                                 prog->name, err);
5833                         return err;
5834                 }
5835         }
5836         /* Process data relos for main programs */
5837         for (i = 0; i < obj->nr_programs; i++) {
5838                 prog = &obj->programs[i];
5839                 if (prog_is_subprog(obj, prog))
5840                         continue;
5841                 err = bpf_object__relocate_data(obj, prog);
5842                 if (err) {
5843                         pr_warn("prog '%s': failed to relocate data references: %d\n",
5844                                 prog->name, err);
5845                         return err;
5846                 }
5847         }
5848         if (!obj->gen_loader)
5849                 bpf_object__free_relocs(obj);
5850         return 0;
5851 }
5852
5853 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
5854                                             GElf_Shdr *shdr, Elf_Data *data);
5855
5856 static int bpf_object__collect_map_relos(struct bpf_object *obj,
5857                                          GElf_Shdr *shdr, Elf_Data *data)
5858 {
5859         const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
5860         int i, j, nrels, new_sz;
5861         const struct btf_var_secinfo *vi = NULL;
5862         const struct btf_type *sec, *var, *def;
5863         struct bpf_map *map = NULL, *targ_map;
5864         const struct btf_member *member;
5865         const char *name, *mname;
5866         Elf_Data *symbols;
5867         unsigned int moff;
5868         GElf_Sym sym;
5869         GElf_Rel rel;
5870         void *tmp;
5871
5872         if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
5873                 return -EINVAL;
5874         sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
5875         if (!sec)
5876                 return -EINVAL;
5877
5878         symbols = obj->efile.symbols;
5879         nrels = shdr->sh_size / shdr->sh_entsize;
5880         for (i = 0; i < nrels; i++) {
5881                 if (!gelf_getrel(data, i, &rel)) {
5882                         pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
5883                         return -LIBBPF_ERRNO__FORMAT;
5884                 }
5885                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
5886                         pr_warn(".maps relo #%d: symbol %zx not found\n",
5887                                 i, (size_t)GELF_R_SYM(rel.r_info));
5888                         return -LIBBPF_ERRNO__FORMAT;
5889                 }
5890                 name = elf_sym_str(obj, sym.st_name) ?: "<?>";
5891                 if (sym.st_shndx != obj->efile.btf_maps_shndx) {
5892                         pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
5893                                 i, name);
5894                         return -LIBBPF_ERRNO__RELOC;
5895                 }
5896
5897                 pr_debug(".maps relo #%d: for %zd value %zd rel.r_offset %zu name %d ('%s')\n",
5898                          i, (ssize_t)(rel.r_info >> 32), (size_t)sym.st_value,
5899                          (size_t)rel.r_offset, sym.st_name, name);
5900
5901                 for (j = 0; j < obj->nr_maps; j++) {
5902                         map = &obj->maps[j];
5903                         if (map->sec_idx != obj->efile.btf_maps_shndx)
5904                                 continue;
5905
5906                         vi = btf_var_secinfos(sec) + map->btf_var_idx;
5907                         if (vi->offset <= rel.r_offset &&
5908                             rel.r_offset + bpf_ptr_sz <= vi->offset + vi->size)
5909                                 break;
5910                 }
5911                 if (j == obj->nr_maps) {
5912                         pr_warn(".maps relo #%d: cannot find map '%s' at rel.r_offset %zu\n",
5913                                 i, name, (size_t)rel.r_offset);
5914                         return -EINVAL;
5915                 }
5916
5917                 if (!bpf_map_type__is_map_in_map(map->def.type))
5918                         return -EINVAL;
5919                 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
5920                     map->def.key_size != sizeof(int)) {
5921                         pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
5922                                 i, map->name, sizeof(int));
5923                         return -EINVAL;
5924                 }
5925
5926                 targ_map = bpf_object__find_map_by_name(obj, name);
5927                 if (!targ_map)
5928                         return -ESRCH;
5929
5930                 var = btf__type_by_id(obj->btf, vi->type);
5931                 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
5932                 if (btf_vlen(def) == 0)
5933                         return -EINVAL;
5934                 member = btf_members(def) + btf_vlen(def) - 1;
5935                 mname = btf__name_by_offset(obj->btf, member->name_off);
5936                 if (strcmp(mname, "values"))
5937                         return -EINVAL;
5938
5939                 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
5940                 if (rel.r_offset - vi->offset < moff)
5941                         return -EINVAL;
5942
5943                 moff = rel.r_offset - vi->offset - moff;
5944                 /* here we use BPF pointer size, which is always 64 bit, as we
5945                  * are parsing ELF that was built for BPF target
5946                  */
5947                 if (moff % bpf_ptr_sz)
5948                         return -EINVAL;
5949                 moff /= bpf_ptr_sz;
5950                 if (moff >= map->init_slots_sz) {
5951                         new_sz = moff + 1;
5952                         tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
5953                         if (!tmp)
5954                                 return -ENOMEM;
5955                         map->init_slots = tmp;
5956                         memset(map->init_slots + map->init_slots_sz, 0,
5957                                (new_sz - map->init_slots_sz) * host_ptr_sz);
5958                         map->init_slots_sz = new_sz;
5959                 }
5960                 map->init_slots[moff] = targ_map;
5961
5962                 pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n",
5963                          i, map->name, moff, name);
5964         }
5965
5966         return 0;
5967 }
5968
5969 static int cmp_relocs(const void *_a, const void *_b)
5970 {
5971         const struct reloc_desc *a = _a;
5972         const struct reloc_desc *b = _b;
5973
5974         if (a->insn_idx != b->insn_idx)
5975                 return a->insn_idx < b->insn_idx ? -1 : 1;
5976
5977         /* no two relocations should have the same insn_idx, but ... */
5978         if (a->type != b->type)
5979                 return a->type < b->type ? -1 : 1;
5980
5981         return 0;
5982 }
5983
5984 static int bpf_object__collect_relos(struct bpf_object *obj)
5985 {
5986         int i, err;
5987
5988         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
5989                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
5990                 Elf_Data *data = obj->efile.reloc_sects[i].data;
5991                 int idx = shdr->sh_info;
5992
5993                 if (shdr->sh_type != SHT_REL) {
5994                         pr_warn("internal error at %d\n", __LINE__);
5995                         return -LIBBPF_ERRNO__INTERNAL;
5996                 }
5997
5998                 if (idx == obj->efile.st_ops_shndx)
5999                         err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6000                 else if (idx == obj->efile.btf_maps_shndx)
6001                         err = bpf_object__collect_map_relos(obj, shdr, data);
6002                 else
6003                         err = bpf_object__collect_prog_relos(obj, shdr, data);
6004                 if (err)
6005                         return err;
6006         }
6007
6008         for (i = 0; i < obj->nr_programs; i++) {
6009                 struct bpf_program *p = &obj->programs[i];
6010
6011                 if (!p->nr_reloc)
6012                         continue;
6013
6014                 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6015         }
6016         return 0;
6017 }
6018
6019 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6020 {
6021         if (BPF_CLASS(insn->code) == BPF_JMP &&
6022             BPF_OP(insn->code) == BPF_CALL &&
6023             BPF_SRC(insn->code) == BPF_K &&
6024             insn->src_reg == 0 &&
6025             insn->dst_reg == 0) {
6026                     *func_id = insn->imm;
6027                     return true;
6028         }
6029         return false;
6030 }
6031
6032 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
6033 {
6034         struct bpf_insn *insn = prog->insns;
6035         enum bpf_func_id func_id;
6036         int i;
6037
6038         if (obj->gen_loader)
6039                 return 0;
6040
6041         for (i = 0; i < prog->insns_cnt; i++, insn++) {
6042                 if (!insn_is_helper_call(insn, &func_id))
6043                         continue;
6044
6045                 /* on kernels that don't yet support
6046                  * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6047                  * to bpf_probe_read() which works well for old kernels
6048                  */
6049                 switch (func_id) {
6050                 case BPF_FUNC_probe_read_kernel:
6051                 case BPF_FUNC_probe_read_user:
6052                         if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6053                                 insn->imm = BPF_FUNC_probe_read;
6054                         break;
6055                 case BPF_FUNC_probe_read_kernel_str:
6056                 case BPF_FUNC_probe_read_user_str:
6057                         if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6058                                 insn->imm = BPF_FUNC_probe_read_str;
6059                         break;
6060                 default:
6061                         break;
6062                 }
6063         }
6064         return 0;
6065 }
6066
6067 static int
6068 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
6069              char *license, __u32 kern_version, int *pfd)
6070 {
6071         struct bpf_prog_load_params load_attr = {};
6072         char *cp, errmsg[STRERR_BUFSIZE];
6073         size_t log_buf_size = 0;
6074         char *log_buf = NULL;
6075         int btf_fd, ret;
6076
6077         if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6078                 /*
6079                  * The program type must be set.  Most likely we couldn't find a proper
6080                  * section definition at load time, and thus we didn't infer the type.
6081                  */
6082                 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
6083                         prog->name, prog->sec_name);
6084                 return -EINVAL;
6085         }
6086
6087         if (!insns || !insns_cnt)
6088                 return -EINVAL;
6089
6090         load_attr.prog_type = prog->type;
6091         /* old kernels might not support specifying expected_attach_type */
6092         if (!kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE) && prog->sec_def &&
6093             prog->sec_def->is_exp_attach_type_optional)
6094                 load_attr.expected_attach_type = 0;
6095         else
6096                 load_attr.expected_attach_type = prog->expected_attach_type;
6097         if (kernel_supports(prog->obj, FEAT_PROG_NAME))
6098                 load_attr.name = prog->name;
6099         load_attr.insns = insns;
6100         load_attr.insn_cnt = insns_cnt;
6101         load_attr.license = license;
6102         load_attr.attach_btf_id = prog->attach_btf_id;
6103         if (prog->attach_prog_fd)
6104                 load_attr.attach_prog_fd = prog->attach_prog_fd;
6105         else
6106                 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
6107         load_attr.attach_btf_id = prog->attach_btf_id;
6108         load_attr.kern_version = kern_version;
6109         load_attr.prog_ifindex = prog->prog_ifindex;
6110
6111         /* specify func_info/line_info only if kernel supports them */
6112         btf_fd = bpf_object__btf_fd(prog->obj);
6113         if (btf_fd >= 0 && kernel_supports(prog->obj, FEAT_BTF_FUNC)) {
6114                 load_attr.prog_btf_fd = btf_fd;
6115                 load_attr.func_info = prog->func_info;
6116                 load_attr.func_info_rec_size = prog->func_info_rec_size;
6117                 load_attr.func_info_cnt = prog->func_info_cnt;
6118                 load_attr.line_info = prog->line_info;
6119                 load_attr.line_info_rec_size = prog->line_info_rec_size;
6120                 load_attr.line_info_cnt = prog->line_info_cnt;
6121         }
6122         load_attr.log_level = prog->log_level;
6123         load_attr.prog_flags = prog->prog_flags;
6124
6125         if (prog->obj->gen_loader) {
6126                 bpf_gen__prog_load(prog->obj->gen_loader, &load_attr,
6127                                    prog - prog->obj->programs);
6128                 *pfd = -1;
6129                 return 0;
6130         }
6131 retry_load:
6132         if (log_buf_size) {
6133                 log_buf = malloc(log_buf_size);
6134                 if (!log_buf)
6135                         return -ENOMEM;
6136
6137                 *log_buf = 0;
6138         }
6139
6140         load_attr.log_buf = log_buf;
6141         load_attr.log_buf_sz = log_buf_size;
6142         ret = libbpf__bpf_prog_load(&load_attr);
6143
6144         if (ret >= 0) {
6145                 if (log_buf && load_attr.log_level)
6146                         pr_debug("verifier log:\n%s", log_buf);
6147
6148                 if (prog->obj->rodata_map_idx >= 0 &&
6149                     kernel_supports(prog->obj, FEAT_PROG_BIND_MAP)) {
6150                         struct bpf_map *rodata_map =
6151                                 &prog->obj->maps[prog->obj->rodata_map_idx];
6152
6153                         if (bpf_prog_bind_map(ret, bpf_map__fd(rodata_map), NULL)) {
6154                                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6155                                 pr_warn("prog '%s': failed to bind .rodata map: %s\n",
6156                                         prog->name, cp);
6157                                 /* Don't fail hard if can't bind rodata. */
6158                         }
6159                 }
6160
6161                 *pfd = ret;
6162                 ret = 0;
6163                 goto out;
6164         }
6165
6166         if (!log_buf || errno == ENOSPC) {
6167                 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE,
6168                                    log_buf_size << 1);
6169
6170                 free(log_buf);
6171                 goto retry_load;
6172         }
6173         ret = errno ? -errno : -LIBBPF_ERRNO__LOAD;
6174         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6175         pr_warn("load bpf program failed: %s\n", cp);
6176         pr_perm_msg(ret);
6177
6178         if (log_buf && log_buf[0] != '\0') {
6179                 ret = -LIBBPF_ERRNO__VERIFY;
6180                 pr_warn("-- BEGIN DUMP LOG ---\n");
6181                 pr_warn("\n%s\n", log_buf);
6182                 pr_warn("-- END LOG --\n");
6183         } else if (load_attr.insn_cnt >= BPF_MAXINSNS) {
6184                 pr_warn("Program too large (%zu insns), at most %d insns\n",
6185                         load_attr.insn_cnt, BPF_MAXINSNS);
6186                 ret = -LIBBPF_ERRNO__PROG2BIG;
6187         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
6188                 /* Wrong program type? */
6189                 int fd;
6190
6191                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
6192                 load_attr.expected_attach_type = 0;
6193                 load_attr.log_buf = NULL;
6194                 load_attr.log_buf_sz = 0;
6195                 fd = libbpf__bpf_prog_load(&load_attr);
6196                 if (fd >= 0) {
6197                         close(fd);
6198                         ret = -LIBBPF_ERRNO__PROGTYPE;
6199                         goto out;
6200                 }
6201         }
6202
6203 out:
6204         free(log_buf);
6205         return ret;
6206 }
6207
6208 static int bpf_program__record_externs(struct bpf_program *prog)
6209 {
6210         struct bpf_object *obj = prog->obj;
6211         int i;
6212
6213         for (i = 0; i < prog->nr_reloc; i++) {
6214                 struct reloc_desc *relo = &prog->reloc_desc[i];
6215                 struct extern_desc *ext = &obj->externs[relo->sym_off];
6216
6217                 switch (relo->type) {
6218                 case RELO_EXTERN_VAR:
6219                         if (ext->type != EXT_KSYM)
6220                                 continue;
6221                         if (!ext->ksym.type_id) {
6222                                 pr_warn("typeless ksym %s is not supported yet\n",
6223                                         ext->name);
6224                                 return -ENOTSUP;
6225                         }
6226                         bpf_gen__record_extern(obj->gen_loader, ext->name, BTF_KIND_VAR,
6227                                                relo->insn_idx);
6228                         break;
6229                 case RELO_EXTERN_FUNC:
6230                         bpf_gen__record_extern(obj->gen_loader, ext->name, BTF_KIND_FUNC,
6231                                                relo->insn_idx);
6232                         break;
6233                 default:
6234                         continue;
6235                 }
6236         }
6237         return 0;
6238 }
6239
6240 static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id);
6241
6242 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
6243 {
6244         int err = 0, fd, i;
6245
6246         if (prog->obj->loaded) {
6247                 pr_warn("prog '%s': can't load after object was loaded\n", prog->name);
6248                 return libbpf_err(-EINVAL);
6249         }
6250
6251         if ((prog->type == BPF_PROG_TYPE_TRACING ||
6252              prog->type == BPF_PROG_TYPE_LSM ||
6253              prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
6254                 int btf_obj_fd = 0, btf_type_id = 0;
6255
6256                 err = libbpf_find_attach_btf_id(prog, &btf_obj_fd, &btf_type_id);
6257                 if (err)
6258                         return libbpf_err(err);
6259
6260                 prog->attach_btf_obj_fd = btf_obj_fd;
6261                 prog->attach_btf_id = btf_type_id;
6262         }
6263
6264         if (prog->instances.nr < 0 || !prog->instances.fds) {
6265                 if (prog->preprocessor) {
6266                         pr_warn("Internal error: can't load program '%s'\n",
6267                                 prog->name);
6268                         return libbpf_err(-LIBBPF_ERRNO__INTERNAL);
6269                 }
6270
6271                 prog->instances.fds = malloc(sizeof(int));
6272                 if (!prog->instances.fds) {
6273                         pr_warn("Not enough memory for BPF fds\n");
6274                         return libbpf_err(-ENOMEM);
6275                 }
6276                 prog->instances.nr = 1;
6277                 prog->instances.fds[0] = -1;
6278         }
6279
6280         if (!prog->preprocessor) {
6281                 if (prog->instances.nr != 1) {
6282                         pr_warn("prog '%s': inconsistent nr(%d) != 1\n",
6283                                 prog->name, prog->instances.nr);
6284                 }
6285                 if (prog->obj->gen_loader)
6286                         bpf_program__record_externs(prog);
6287                 err = load_program(prog, prog->insns, prog->insns_cnt,
6288                                    license, kern_ver, &fd);
6289                 if (!err)
6290                         prog->instances.fds[0] = fd;
6291                 goto out;
6292         }
6293
6294         for (i = 0; i < prog->instances.nr; i++) {
6295                 struct bpf_prog_prep_result result;
6296                 bpf_program_prep_t preprocessor = prog->preprocessor;
6297
6298                 memset(&result, 0, sizeof(result));
6299                 err = preprocessor(prog, i, prog->insns,
6300                                    prog->insns_cnt, &result);
6301                 if (err) {
6302                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
6303                                 i, prog->name);
6304                         goto out;
6305                 }
6306
6307                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
6308                         pr_debug("Skip loading the %dth instance of program '%s'\n",
6309                                  i, prog->name);
6310                         prog->instances.fds[i] = -1;
6311                         if (result.pfd)
6312                                 *result.pfd = -1;
6313                         continue;
6314                 }
6315
6316                 err = load_program(prog, result.new_insn_ptr,
6317                                    result.new_insn_cnt, license, kern_ver, &fd);
6318                 if (err) {
6319                         pr_warn("Loading the %dth instance of program '%s' failed\n",
6320                                 i, prog->name);
6321                         goto out;
6322                 }
6323
6324                 if (result.pfd)
6325                         *result.pfd = fd;
6326                 prog->instances.fds[i] = fd;
6327         }
6328 out:
6329         if (err)
6330                 pr_warn("failed to load program '%s'\n", prog->name);
6331         zfree(&prog->insns);
6332         prog->insns_cnt = 0;
6333         return libbpf_err(err);
6334 }
6335
6336 static int
6337 bpf_object__load_progs(struct bpf_object *obj, int log_level)
6338 {
6339         struct bpf_program *prog;
6340         size_t i;
6341         int err;
6342
6343         for (i = 0; i < obj->nr_programs; i++) {
6344                 prog = &obj->programs[i];
6345                 err = bpf_object__sanitize_prog(obj, prog);
6346                 if (err)
6347                         return err;
6348         }
6349
6350         for (i = 0; i < obj->nr_programs; i++) {
6351                 prog = &obj->programs[i];
6352                 if (prog_is_subprog(obj, prog))
6353                         continue;
6354                 if (!prog->load) {
6355                         pr_debug("prog '%s': skipped loading\n", prog->name);
6356                         continue;
6357                 }
6358                 prog->log_level |= log_level;
6359                 err = bpf_program__load(prog, obj->license, obj->kern_version);
6360                 if (err)
6361                         return err;
6362         }
6363         if (obj->gen_loader)
6364                 bpf_object__free_relocs(obj);
6365         return 0;
6366 }
6367
6368 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
6369
6370 static struct bpf_object *
6371 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
6372                    const struct bpf_object_open_opts *opts)
6373 {
6374         const char *obj_name, *kconfig, *btf_tmp_path;
6375         struct bpf_program *prog;
6376         struct bpf_object *obj;
6377         char tmp_name[64];
6378         int err;
6379
6380         if (elf_version(EV_CURRENT) == EV_NONE) {
6381                 pr_warn("failed to init libelf for %s\n",
6382                         path ? : "(mem buf)");
6383                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
6384         }
6385
6386         if (!OPTS_VALID(opts, bpf_object_open_opts))
6387                 return ERR_PTR(-EINVAL);
6388
6389         obj_name = OPTS_GET(opts, object_name, NULL);
6390         if (obj_buf) {
6391                 if (!obj_name) {
6392                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
6393                                  (unsigned long)obj_buf,
6394                                  (unsigned long)obj_buf_sz);
6395                         obj_name = tmp_name;
6396                 }
6397                 path = obj_name;
6398                 pr_debug("loading object '%s' from buffer\n", obj_name);
6399         }
6400
6401         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
6402         if (IS_ERR(obj))
6403                 return obj;
6404
6405         btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
6406         if (btf_tmp_path) {
6407                 if (strlen(btf_tmp_path) >= PATH_MAX) {
6408                         err = -ENAMETOOLONG;
6409                         goto out;
6410                 }
6411                 obj->btf_custom_path = strdup(btf_tmp_path);
6412                 if (!obj->btf_custom_path) {
6413                         err = -ENOMEM;
6414                         goto out;
6415                 }
6416         }
6417
6418         kconfig = OPTS_GET(opts, kconfig, NULL);
6419         if (kconfig) {
6420                 obj->kconfig = strdup(kconfig);
6421                 if (!obj->kconfig) {
6422                         err = -ENOMEM;
6423                         goto out;
6424                 }
6425         }
6426
6427         err = bpf_object__elf_init(obj);
6428         err = err ? : bpf_object__check_endianness(obj);
6429         err = err ? : bpf_object__elf_collect(obj);
6430         err = err ? : bpf_object__collect_externs(obj);
6431         err = err ? : bpf_object__finalize_btf(obj);
6432         err = err ? : bpf_object__init_maps(obj, opts);
6433         err = err ? : bpf_object__collect_relos(obj);
6434         if (err)
6435                 goto out;
6436         bpf_object__elf_finish(obj);
6437
6438         bpf_object__for_each_program(prog, obj) {
6439                 prog->sec_def = find_sec_def(prog->sec_name);
6440                 if (!prog->sec_def) {
6441                         /* couldn't guess, but user might manually specify */
6442                         pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
6443                                 prog->name, prog->sec_name);
6444                         continue;
6445                 }
6446
6447                 if (prog->sec_def->is_sleepable)
6448                         prog->prog_flags |= BPF_F_SLEEPABLE;
6449                 bpf_program__set_type(prog, prog->sec_def->prog_type);
6450                 bpf_program__set_expected_attach_type(prog,
6451                                 prog->sec_def->expected_attach_type);
6452
6453                 if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
6454                     prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
6455                         prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
6456         }
6457
6458         return obj;
6459 out:
6460         bpf_object__close(obj);
6461         return ERR_PTR(err);
6462 }
6463
6464 static struct bpf_object *
6465 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
6466 {
6467         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
6468                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
6469         );
6470
6471         /* param validation */
6472         if (!attr->file)
6473                 return NULL;
6474
6475         pr_debug("loading %s\n", attr->file);
6476         return __bpf_object__open(attr->file, NULL, 0, &opts);
6477 }
6478
6479 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
6480 {
6481         return libbpf_ptr(__bpf_object__open_xattr(attr, 0));
6482 }
6483
6484 struct bpf_object *bpf_object__open(const char *path)
6485 {
6486         struct bpf_object_open_attr attr = {
6487                 .file           = path,
6488                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
6489         };
6490
6491         return libbpf_ptr(__bpf_object__open_xattr(&attr, 0));
6492 }
6493
6494 struct bpf_object *
6495 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
6496 {
6497         if (!path)
6498                 return libbpf_err_ptr(-EINVAL);
6499
6500         pr_debug("loading %s\n", path);
6501
6502         return libbpf_ptr(__bpf_object__open(path, NULL, 0, opts));
6503 }
6504
6505 struct bpf_object *
6506 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
6507                      const struct bpf_object_open_opts *opts)
6508 {
6509         if (!obj_buf || obj_buf_sz == 0)
6510                 return libbpf_err_ptr(-EINVAL);
6511
6512         return libbpf_ptr(__bpf_object__open(NULL, obj_buf, obj_buf_sz, opts));
6513 }
6514
6515 struct bpf_object *
6516 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
6517                         const char *name)
6518 {
6519         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
6520                 .object_name = name,
6521                 /* wrong default, but backwards-compatible */
6522                 .relaxed_maps = true,
6523         );
6524
6525         /* returning NULL is wrong, but backwards-compatible */
6526         if (!obj_buf || obj_buf_sz == 0)
6527                 return errno = EINVAL, NULL;
6528
6529         return libbpf_ptr(__bpf_object__open(NULL, obj_buf, obj_buf_sz, &opts));
6530 }
6531
6532 int bpf_object__unload(struct bpf_object *obj)
6533 {
6534         size_t i;
6535
6536         if (!obj)
6537                 return libbpf_err(-EINVAL);
6538
6539         for (i = 0; i < obj->nr_maps; i++) {
6540                 zclose(obj->maps[i].fd);
6541                 if (obj->maps[i].st_ops)
6542                         zfree(&obj->maps[i].st_ops->kern_vdata);
6543         }
6544
6545         for (i = 0; i < obj->nr_programs; i++)
6546                 bpf_program__unload(&obj->programs[i]);
6547
6548         return 0;
6549 }
6550
6551 static int bpf_object__sanitize_maps(struct bpf_object *obj)
6552 {
6553         struct bpf_map *m;
6554
6555         bpf_object__for_each_map(m, obj) {
6556                 if (!bpf_map__is_internal(m))
6557                         continue;
6558                 if (!kernel_supports(obj, FEAT_GLOBAL_DATA)) {
6559                         pr_warn("kernel doesn't support global data\n");
6560                         return -ENOTSUP;
6561                 }
6562                 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
6563                         m->def.map_flags ^= BPF_F_MMAPABLE;
6564         }
6565
6566         return 0;
6567 }
6568
6569 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
6570 {
6571         char sym_type, sym_name[500];
6572         unsigned long long sym_addr;
6573         const struct btf_type *t;
6574         struct extern_desc *ext;
6575         int ret, err = 0;
6576         FILE *f;
6577
6578         f = fopen("/proc/kallsyms", "r");
6579         if (!f) {
6580                 err = -errno;
6581                 pr_warn("failed to open /proc/kallsyms: %d\n", err);
6582                 return err;
6583         }
6584
6585         while (true) {
6586                 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
6587                              &sym_addr, &sym_type, sym_name);
6588                 if (ret == EOF && feof(f))
6589                         break;
6590                 if (ret != 3) {
6591                         pr_warn("failed to read kallsyms entry: %d\n", ret);
6592                         err = -EINVAL;
6593                         goto out;
6594                 }
6595
6596                 ext = find_extern_by_name(obj, sym_name);
6597                 if (!ext || ext->type != EXT_KSYM)
6598                         continue;
6599
6600                 t = btf__type_by_id(obj->btf, ext->btf_id);
6601                 if (!btf_is_var(t))
6602                         continue;
6603
6604                 if (ext->is_set && ext->ksym.addr != sym_addr) {
6605                         pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
6606                                 sym_name, ext->ksym.addr, sym_addr);
6607                         err = -EINVAL;
6608                         goto out;
6609                 }
6610                 if (!ext->is_set) {
6611                         ext->is_set = true;
6612                         ext->ksym.addr = sym_addr;
6613                         pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
6614                 }
6615         }
6616
6617 out:
6618         fclose(f);
6619         return err;
6620 }
6621
6622 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
6623                             __u16 kind, struct btf **res_btf,
6624                             int *res_btf_fd)
6625 {
6626         int i, id, btf_fd, err;
6627         struct btf *btf;
6628
6629         btf = obj->btf_vmlinux;
6630         btf_fd = 0;
6631         id = btf__find_by_name_kind(btf, ksym_name, kind);
6632
6633         if (id == -ENOENT) {
6634                 err = load_module_btfs(obj);
6635                 if (err)
6636                         return err;
6637
6638                 for (i = 0; i < obj->btf_module_cnt; i++) {
6639                         btf = obj->btf_modules[i].btf;
6640                         /* we assume module BTF FD is always >0 */
6641                         btf_fd = obj->btf_modules[i].fd;
6642                         id = btf__find_by_name_kind(btf, ksym_name, kind);
6643                         if (id != -ENOENT)
6644                                 break;
6645                 }
6646         }
6647         if (id <= 0)
6648                 return -ESRCH;
6649
6650         *res_btf = btf;
6651         *res_btf_fd = btf_fd;
6652         return id;
6653 }
6654
6655 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
6656                                                struct extern_desc *ext)
6657 {
6658         const struct btf_type *targ_var, *targ_type;
6659         __u32 targ_type_id, local_type_id;
6660         const char *targ_var_name;
6661         int id, btf_fd = 0, err;
6662         struct btf *btf = NULL;
6663
6664         id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &btf_fd);
6665         if (id == -ESRCH && ext->is_weak) {
6666                 return 0;
6667         } else if (id < 0) {
6668                 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
6669                         ext->name);
6670                 return id;
6671         }
6672
6673         /* find local type_id */
6674         local_type_id = ext->ksym.type_id;
6675
6676         /* find target type_id */
6677         targ_var = btf__type_by_id(btf, id);
6678         targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
6679         targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
6680
6681         err = bpf_core_types_are_compat(obj->btf, local_type_id,
6682                                         btf, targ_type_id);
6683         if (err <= 0) {
6684                 const struct btf_type *local_type;
6685                 const char *targ_name, *local_name;
6686
6687                 local_type = btf__type_by_id(obj->btf, local_type_id);
6688                 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
6689                 targ_name = btf__name_by_offset(btf, targ_type->name_off);
6690
6691                 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
6692                         ext->name, local_type_id,
6693                         btf_kind_str(local_type), local_name, targ_type_id,
6694                         btf_kind_str(targ_type), targ_name);
6695                 return -EINVAL;
6696         }
6697
6698         ext->is_set = true;
6699         ext->ksym.kernel_btf_obj_fd = btf_fd;
6700         ext->ksym.kernel_btf_id = id;
6701         pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
6702                  ext->name, id, btf_kind_str(targ_var), targ_var_name);
6703
6704         return 0;
6705 }
6706
6707 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
6708                                                 struct extern_desc *ext)
6709 {
6710         int local_func_proto_id, kfunc_proto_id, kfunc_id;
6711         const struct btf_type *kern_func;
6712         struct btf *kern_btf = NULL;
6713         int ret, kern_btf_fd = 0;
6714
6715         local_func_proto_id = ext->ksym.type_id;
6716
6717         kfunc_id = find_ksym_btf_id(obj, ext->name, BTF_KIND_FUNC,
6718                                     &kern_btf, &kern_btf_fd);
6719         if (kfunc_id < 0) {
6720                 pr_warn("extern (func ksym) '%s': not found in kernel BTF\n",
6721                         ext->name);
6722                 return kfunc_id;
6723         }
6724
6725         if (kern_btf != obj->btf_vmlinux) {
6726                 pr_warn("extern (func ksym) '%s': function in kernel module is not supported\n",
6727                         ext->name);
6728                 return -ENOTSUP;
6729         }
6730
6731         kern_func = btf__type_by_id(kern_btf, kfunc_id);
6732         kfunc_proto_id = kern_func->type;
6733
6734         ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
6735                                         kern_btf, kfunc_proto_id);
6736         if (ret <= 0) {
6737                 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with kernel [%d]\n",
6738                         ext->name, local_func_proto_id, kfunc_proto_id);
6739                 return -EINVAL;
6740         }
6741
6742         ext->is_set = true;
6743         ext->ksym.kernel_btf_obj_fd = kern_btf_fd;
6744         ext->ksym.kernel_btf_id = kfunc_id;
6745         pr_debug("extern (func ksym) '%s': resolved to kernel [%d]\n",
6746                  ext->name, kfunc_id);
6747
6748         return 0;
6749 }
6750
6751 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
6752 {
6753         const struct btf_type *t;
6754         struct extern_desc *ext;
6755         int i, err;
6756
6757         for (i = 0; i < obj->nr_extern; i++) {
6758                 ext = &obj->externs[i];
6759                 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
6760                         continue;
6761
6762                 if (obj->gen_loader) {
6763                         ext->is_set = true;
6764                         ext->ksym.kernel_btf_obj_fd = 0;
6765                         ext->ksym.kernel_btf_id = 0;
6766                         continue;
6767                 }
6768                 t = btf__type_by_id(obj->btf, ext->btf_id);
6769                 if (btf_is_var(t))
6770                         err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
6771                 else
6772                         err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
6773                 if (err)
6774                         return err;
6775         }
6776         return 0;
6777 }
6778
6779 static int bpf_object__resolve_externs(struct bpf_object *obj,
6780                                        const char *extra_kconfig)
6781 {
6782         bool need_config = false, need_kallsyms = false;
6783         bool need_vmlinux_btf = false;
6784         struct extern_desc *ext;
6785         void *kcfg_data = NULL;
6786         int err, i;
6787
6788         if (obj->nr_extern == 0)
6789                 return 0;
6790
6791         if (obj->kconfig_map_idx >= 0)
6792                 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
6793
6794         for (i = 0; i < obj->nr_extern; i++) {
6795                 ext = &obj->externs[i];
6796
6797                 if (ext->type == EXT_KCFG &&
6798                     strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
6799                         void *ext_val = kcfg_data + ext->kcfg.data_off;
6800                         __u32 kver = get_kernel_version();
6801
6802                         if (!kver) {
6803                                 pr_warn("failed to get kernel version\n");
6804                                 return -EINVAL;
6805                         }
6806                         err = set_kcfg_value_num(ext, ext_val, kver);
6807                         if (err)
6808                                 return err;
6809                         pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
6810                 } else if (ext->type == EXT_KCFG &&
6811                            strncmp(ext->name, "CONFIG_", 7) == 0) {
6812                         need_config = true;
6813                 } else if (ext->type == EXT_KSYM) {
6814                         if (ext->ksym.type_id)
6815                                 need_vmlinux_btf = true;
6816                         else
6817                                 need_kallsyms = true;
6818                 } else {
6819                         pr_warn("unrecognized extern '%s'\n", ext->name);
6820                         return -EINVAL;
6821                 }
6822         }
6823         if (need_config && extra_kconfig) {
6824                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
6825                 if (err)
6826                         return -EINVAL;
6827                 need_config = false;
6828                 for (i = 0; i < obj->nr_extern; i++) {
6829                         ext = &obj->externs[i];
6830                         if (ext->type == EXT_KCFG && !ext->is_set) {
6831                                 need_config = true;
6832                                 break;
6833                         }
6834                 }
6835         }
6836         if (need_config) {
6837                 err = bpf_object__read_kconfig_file(obj, kcfg_data);
6838                 if (err)
6839                         return -EINVAL;
6840         }
6841         if (need_kallsyms) {
6842                 err = bpf_object__read_kallsyms_file(obj);
6843                 if (err)
6844                         return -EINVAL;
6845         }
6846         if (need_vmlinux_btf) {
6847                 err = bpf_object__resolve_ksyms_btf_id(obj);
6848                 if (err)
6849                         return -EINVAL;
6850         }
6851         for (i = 0; i < obj->nr_extern; i++) {
6852                 ext = &obj->externs[i];
6853
6854                 if (!ext->is_set && !ext->is_weak) {
6855                         pr_warn("extern %s (strong) not resolved\n", ext->name);
6856                         return -ESRCH;
6857                 } else if (!ext->is_set) {
6858                         pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
6859                                  ext->name);
6860                 }
6861         }
6862
6863         return 0;
6864 }
6865
6866 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
6867 {
6868         struct bpf_object *obj;
6869         int err, i;
6870
6871         if (!attr)
6872                 return libbpf_err(-EINVAL);
6873         obj = attr->obj;
6874         if (!obj)
6875                 return libbpf_err(-EINVAL);
6876
6877         if (obj->loaded) {
6878                 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
6879                 return libbpf_err(-EINVAL);
6880         }
6881
6882         if (obj->gen_loader)
6883                 bpf_gen__init(obj->gen_loader, attr->log_level);
6884
6885         err = bpf_object__probe_loading(obj);
6886         err = err ? : bpf_object__load_vmlinux_btf(obj, false);
6887         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
6888         err = err ? : bpf_object__sanitize_and_load_btf(obj);
6889         err = err ? : bpf_object__sanitize_maps(obj);
6890         err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
6891         err = err ? : bpf_object__create_maps(obj);
6892         err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : attr->target_btf_path);
6893         err = err ? : bpf_object__load_progs(obj, attr->log_level);
6894
6895         if (obj->gen_loader) {
6896                 /* reset FDs */
6897                 if (obj->btf)
6898                         btf__set_fd(obj->btf, -1);
6899                 for (i = 0; i < obj->nr_maps; i++)
6900                         obj->maps[i].fd = -1;
6901                 if (!err)
6902                         err = bpf_gen__finish(obj->gen_loader);
6903         }
6904
6905         /* clean up module BTFs */
6906         for (i = 0; i < obj->btf_module_cnt; i++) {
6907                 close(obj->btf_modules[i].fd);
6908                 btf__free(obj->btf_modules[i].btf);
6909                 free(obj->btf_modules[i].name);
6910         }
6911         free(obj->btf_modules);
6912
6913         /* clean up vmlinux BTF */
6914         btf__free(obj->btf_vmlinux);
6915         obj->btf_vmlinux = NULL;
6916
6917         obj->loaded = true; /* doesn't matter if successfully or not */
6918
6919         if (err)
6920                 goto out;
6921
6922         return 0;
6923 out:
6924         /* unpin any maps that were auto-pinned during load */
6925         for (i = 0; i < obj->nr_maps; i++)
6926                 if (obj->maps[i].pinned && !obj->maps[i].reused)
6927                         bpf_map__unpin(&obj->maps[i], NULL);
6928
6929         bpf_object__unload(obj);
6930         pr_warn("failed to load object '%s'\n", obj->path);
6931         return libbpf_err(err);
6932 }
6933
6934 int bpf_object__load(struct bpf_object *obj)
6935 {
6936         struct bpf_object_load_attr attr = {
6937                 .obj = obj,
6938         };
6939
6940         return bpf_object__load_xattr(&attr);
6941 }
6942
6943 static int make_parent_dir(const char *path)
6944 {
6945         char *cp, errmsg[STRERR_BUFSIZE];
6946         char *dname, *dir;
6947         int err = 0;
6948
6949         dname = strdup(path);
6950         if (dname == NULL)
6951                 return -ENOMEM;
6952
6953         dir = dirname(dname);
6954         if (mkdir(dir, 0700) && errno != EEXIST)
6955                 err = -errno;
6956
6957         free(dname);
6958         if (err) {
6959                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
6960                 pr_warn("failed to mkdir %s: %s\n", path, cp);
6961         }
6962         return err;
6963 }
6964
6965 static int check_path(const char *path)
6966 {
6967         char *cp, errmsg[STRERR_BUFSIZE];
6968         struct statfs st_fs;
6969         char *dname, *dir;
6970         int err = 0;
6971
6972         if (path == NULL)
6973                 return -EINVAL;
6974
6975         dname = strdup(path);
6976         if (dname == NULL)
6977                 return -ENOMEM;
6978
6979         dir = dirname(dname);
6980         if (statfs(dir, &st_fs)) {
6981                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6982                 pr_warn("failed to statfs %s: %s\n", dir, cp);
6983                 err = -errno;
6984         }
6985         free(dname);
6986
6987         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
6988                 pr_warn("specified path %s is not on BPF FS\n", path);
6989                 err = -EINVAL;
6990         }
6991
6992         return err;
6993 }
6994
6995 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
6996                               int instance)
6997 {
6998         char *cp, errmsg[STRERR_BUFSIZE];
6999         int err;
7000
7001         err = make_parent_dir(path);
7002         if (err)
7003                 return libbpf_err(err);
7004
7005         err = check_path(path);
7006         if (err)
7007                 return libbpf_err(err);
7008
7009         if (prog == NULL) {
7010                 pr_warn("invalid program pointer\n");
7011                 return libbpf_err(-EINVAL);
7012         }
7013
7014         if (instance < 0 || instance >= prog->instances.nr) {
7015                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7016                         instance, prog->name, prog->instances.nr);
7017                 return libbpf_err(-EINVAL);
7018         }
7019
7020         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
7021                 err = -errno;
7022                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
7023                 pr_warn("failed to pin program: %s\n", cp);
7024                 return libbpf_err(err);
7025         }
7026         pr_debug("pinned program '%s'\n", path);
7027
7028         return 0;
7029 }
7030
7031 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
7032                                 int instance)
7033 {
7034         int err;
7035
7036         err = check_path(path);
7037         if (err)
7038                 return libbpf_err(err);
7039
7040         if (prog == NULL) {
7041                 pr_warn("invalid program pointer\n");
7042                 return libbpf_err(-EINVAL);
7043         }
7044
7045         if (instance < 0 || instance >= prog->instances.nr) {
7046                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7047                         instance, prog->name, prog->instances.nr);
7048                 return libbpf_err(-EINVAL);
7049         }
7050
7051         err = unlink(path);
7052         if (err != 0)
7053                 return libbpf_err(-errno);
7054
7055         pr_debug("unpinned program '%s'\n", path);
7056
7057         return 0;
7058 }
7059
7060 int bpf_program__pin(struct bpf_program *prog, const char *path)
7061 {
7062         int i, err;
7063
7064         err = make_parent_dir(path);
7065         if (err)
7066                 return libbpf_err(err);
7067
7068         err = check_path(path);
7069         if (err)
7070                 return libbpf_err(err);
7071
7072         if (prog == NULL) {
7073                 pr_warn("invalid program pointer\n");
7074                 return libbpf_err(-EINVAL);
7075         }
7076
7077         if (prog->instances.nr <= 0) {
7078                 pr_warn("no instances of prog %s to pin\n", prog->name);
7079                 return libbpf_err(-EINVAL);
7080         }
7081
7082         if (prog->instances.nr == 1) {
7083                 /* don't create subdirs when pinning single instance */
7084                 return bpf_program__pin_instance(prog, path, 0);
7085         }
7086
7087         for (i = 0; i < prog->instances.nr; i++) {
7088                 char buf[PATH_MAX];
7089                 int len;
7090
7091                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7092                 if (len < 0) {
7093                         err = -EINVAL;
7094                         goto err_unpin;
7095                 } else if (len >= PATH_MAX) {
7096                         err = -ENAMETOOLONG;
7097                         goto err_unpin;
7098                 }
7099
7100                 err = bpf_program__pin_instance(prog, buf, i);
7101                 if (err)
7102                         goto err_unpin;
7103         }
7104
7105         return 0;
7106
7107 err_unpin:
7108         for (i = i - 1; i >= 0; i--) {
7109                 char buf[PATH_MAX];
7110                 int len;
7111
7112                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7113                 if (len < 0)
7114                         continue;
7115                 else if (len >= PATH_MAX)
7116                         continue;
7117
7118                 bpf_program__unpin_instance(prog, buf, i);
7119         }
7120
7121         rmdir(path);
7122
7123         return libbpf_err(err);
7124 }
7125
7126 int bpf_program__unpin(struct bpf_program *prog, const char *path)
7127 {
7128         int i, err;
7129
7130         err = check_path(path);
7131         if (err)
7132                 return libbpf_err(err);
7133
7134         if (prog == NULL) {
7135                 pr_warn("invalid program pointer\n");
7136                 return libbpf_err(-EINVAL);
7137         }
7138
7139         if (prog->instances.nr <= 0) {
7140                 pr_warn("no instances of prog %s to pin\n", prog->name);
7141                 return libbpf_err(-EINVAL);
7142         }
7143
7144         if (prog->instances.nr == 1) {
7145                 /* don't create subdirs when pinning single instance */
7146                 return bpf_program__unpin_instance(prog, path, 0);
7147         }
7148
7149         for (i = 0; i < prog->instances.nr; i++) {
7150                 char buf[PATH_MAX];
7151                 int len;
7152
7153                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7154                 if (len < 0)
7155                         return libbpf_err(-EINVAL);
7156                 else if (len >= PATH_MAX)
7157                         return libbpf_err(-ENAMETOOLONG);
7158
7159                 err = bpf_program__unpin_instance(prog, buf, i);
7160                 if (err)
7161                         return err;
7162         }
7163
7164         err = rmdir(path);
7165         if (err)
7166                 return libbpf_err(-errno);
7167
7168         return 0;
7169 }
7170
7171 int bpf_map__pin(struct bpf_map *map, const char *path)
7172 {
7173         char *cp, errmsg[STRERR_BUFSIZE];
7174         int err;
7175
7176         if (map == NULL) {
7177                 pr_warn("invalid map pointer\n");
7178                 return libbpf_err(-EINVAL);
7179         }
7180
7181         if (map->pin_path) {
7182                 if (path && strcmp(path, map->pin_path)) {
7183                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7184                                 bpf_map__name(map), map->pin_path, path);
7185                         return libbpf_err(-EINVAL);
7186                 } else if (map->pinned) {
7187                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
7188                                  bpf_map__name(map), map->pin_path);
7189                         return 0;
7190                 }
7191         } else {
7192                 if (!path) {
7193                         pr_warn("missing a path to pin map '%s' at\n",
7194                                 bpf_map__name(map));
7195                         return libbpf_err(-EINVAL);
7196                 } else if (map->pinned) {
7197                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
7198                         return libbpf_err(-EEXIST);
7199                 }
7200
7201                 map->pin_path = strdup(path);
7202                 if (!map->pin_path) {
7203                         err = -errno;
7204                         goto out_err;
7205                 }
7206         }
7207
7208         err = make_parent_dir(map->pin_path);
7209         if (err)
7210                 return libbpf_err(err);
7211
7212         err = check_path(map->pin_path);
7213         if (err)
7214                 return libbpf_err(err);
7215
7216         if (bpf_obj_pin(map->fd, map->pin_path)) {
7217                 err = -errno;
7218                 goto out_err;
7219         }
7220
7221         map->pinned = true;
7222         pr_debug("pinned map '%s'\n", map->pin_path);
7223
7224         return 0;
7225
7226 out_err:
7227         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7228         pr_warn("failed to pin map: %s\n", cp);
7229         return libbpf_err(err);
7230 }
7231
7232 int bpf_map__unpin(struct bpf_map *map, const char *path)
7233 {
7234         int err;
7235
7236         if (map == NULL) {
7237                 pr_warn("invalid map pointer\n");
7238                 return libbpf_err(-EINVAL);
7239         }
7240
7241         if (map->pin_path) {
7242                 if (path && strcmp(path, map->pin_path)) {
7243                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7244                                 bpf_map__name(map), map->pin_path, path);
7245                         return libbpf_err(-EINVAL);
7246                 }
7247                 path = map->pin_path;
7248         } else if (!path) {
7249                 pr_warn("no path to unpin map '%s' from\n",
7250                         bpf_map__name(map));
7251                 return libbpf_err(-EINVAL);
7252         }
7253
7254         err = check_path(path);
7255         if (err)
7256                 return libbpf_err(err);
7257
7258         err = unlink(path);
7259         if (err != 0)
7260                 return libbpf_err(-errno);
7261
7262         map->pinned = false;
7263         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
7264
7265         return 0;
7266 }
7267
7268 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
7269 {
7270         char *new = NULL;
7271
7272         if (path) {
7273                 new = strdup(path);
7274                 if (!new)
7275                         return libbpf_err(-errno);
7276         }
7277
7278         free(map->pin_path);
7279         map->pin_path = new;
7280         return 0;
7281 }
7282
7283 const char *bpf_map__get_pin_path(const struct bpf_map *map)
7284 {
7285         return map->pin_path;
7286 }
7287
7288 const char *bpf_map__pin_path(const struct bpf_map *map)
7289 {
7290         return map->pin_path;
7291 }
7292
7293 bool bpf_map__is_pinned(const struct bpf_map *map)
7294 {
7295         return map->pinned;
7296 }
7297
7298 static void sanitize_pin_path(char *s)
7299 {
7300         /* bpffs disallows periods in path names */
7301         while (*s) {
7302                 if (*s == '.')
7303                         *s = '_';
7304                 s++;
7305         }
7306 }
7307
7308 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
7309 {
7310         struct bpf_map *map;
7311         int err;
7312
7313         if (!obj)
7314                 return libbpf_err(-ENOENT);
7315
7316         if (!obj->loaded) {
7317                 pr_warn("object not yet loaded; load it first\n");
7318                 return libbpf_err(-ENOENT);
7319         }
7320
7321         bpf_object__for_each_map(map, obj) {
7322                 char *pin_path = NULL;
7323                 char buf[PATH_MAX];
7324
7325                 if (path) {
7326                         int len;
7327
7328                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
7329                                        bpf_map__name(map));
7330                         if (len < 0) {
7331                                 err = -EINVAL;
7332                                 goto err_unpin_maps;
7333                         } else if (len >= PATH_MAX) {
7334                                 err = -ENAMETOOLONG;
7335                                 goto err_unpin_maps;
7336                         }
7337                         sanitize_pin_path(buf);
7338                         pin_path = buf;
7339                 } else if (!map->pin_path) {
7340                         continue;
7341                 }
7342
7343                 err = bpf_map__pin(map, pin_path);
7344                 if (err)
7345                         goto err_unpin_maps;
7346         }
7347
7348         return 0;
7349
7350 err_unpin_maps:
7351         while ((map = bpf_map__prev(map, obj))) {
7352                 if (!map->pin_path)
7353                         continue;
7354
7355                 bpf_map__unpin(map, NULL);
7356         }
7357
7358         return libbpf_err(err);
7359 }
7360
7361 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
7362 {
7363         struct bpf_map *map;
7364         int err;
7365
7366         if (!obj)
7367                 return libbpf_err(-ENOENT);
7368
7369         bpf_object__for_each_map(map, obj) {
7370                 char *pin_path = NULL;
7371                 char buf[PATH_MAX];
7372
7373                 if (path) {
7374                         int len;
7375
7376                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
7377                                        bpf_map__name(map));
7378                         if (len < 0)
7379                                 return libbpf_err(-EINVAL);
7380                         else if (len >= PATH_MAX)
7381                                 return libbpf_err(-ENAMETOOLONG);
7382                         sanitize_pin_path(buf);
7383                         pin_path = buf;
7384                 } else if (!map->pin_path) {
7385                         continue;
7386                 }
7387
7388                 err = bpf_map__unpin(map, pin_path);
7389                 if (err)
7390                         return libbpf_err(err);
7391         }
7392
7393         return 0;
7394 }
7395
7396 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
7397 {
7398         struct bpf_program *prog;
7399         int err;
7400
7401         if (!obj)
7402                 return libbpf_err(-ENOENT);
7403
7404         if (!obj->loaded) {
7405                 pr_warn("object not yet loaded; load it first\n");
7406                 return libbpf_err(-ENOENT);
7407         }
7408
7409         bpf_object__for_each_program(prog, obj) {
7410                 char buf[PATH_MAX];
7411                 int len;
7412
7413                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
7414                                prog->pin_name);
7415                 if (len < 0) {
7416                         err = -EINVAL;
7417                         goto err_unpin_programs;
7418                 } else if (len >= PATH_MAX) {
7419                         err = -ENAMETOOLONG;
7420                         goto err_unpin_programs;
7421                 }
7422
7423                 err = bpf_program__pin(prog, buf);
7424                 if (err)
7425                         goto err_unpin_programs;
7426         }
7427
7428         return 0;
7429
7430 err_unpin_programs:
7431         while ((prog = bpf_program__prev(prog, obj))) {
7432                 char buf[PATH_MAX];
7433                 int len;
7434
7435                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
7436                                prog->pin_name);
7437                 if (len < 0)
7438                         continue;
7439                 else if (len >= PATH_MAX)
7440                         continue;
7441
7442                 bpf_program__unpin(prog, buf);
7443         }
7444
7445         return libbpf_err(err);
7446 }
7447
7448 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
7449 {
7450         struct bpf_program *prog;
7451         int err;
7452
7453         if (!obj)
7454                 return libbpf_err(-ENOENT);
7455
7456         bpf_object__for_each_program(prog, obj) {
7457                 char buf[PATH_MAX];
7458                 int len;
7459
7460                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
7461                                prog->pin_name);
7462                 if (len < 0)
7463                         return libbpf_err(-EINVAL);
7464                 else if (len >= PATH_MAX)
7465                         return libbpf_err(-ENAMETOOLONG);
7466
7467                 err = bpf_program__unpin(prog, buf);
7468                 if (err)
7469                         return libbpf_err(err);
7470         }
7471
7472         return 0;
7473 }
7474
7475 int bpf_object__pin(struct bpf_object *obj, const char *path)
7476 {
7477         int err;
7478
7479         err = bpf_object__pin_maps(obj, path);
7480         if (err)
7481                 return libbpf_err(err);
7482
7483         err = bpf_object__pin_programs(obj, path);
7484         if (err) {
7485                 bpf_object__unpin_maps(obj, path);
7486                 return libbpf_err(err);
7487         }
7488
7489         return 0;
7490 }
7491
7492 static void bpf_map__destroy(struct bpf_map *map)
7493 {
7494         if (map->clear_priv)
7495                 map->clear_priv(map, map->priv);
7496         map->priv = NULL;
7497         map->clear_priv = NULL;
7498
7499         if (map->inner_map) {
7500                 bpf_map__destroy(map->inner_map);
7501                 zfree(&map->inner_map);
7502         }
7503
7504         zfree(&map->init_slots);
7505         map->init_slots_sz = 0;
7506
7507         if (map->mmaped) {
7508                 munmap(map->mmaped, bpf_map_mmap_sz(map));
7509                 map->mmaped = NULL;
7510         }
7511
7512         if (map->st_ops) {
7513                 zfree(&map->st_ops->data);
7514                 zfree(&map->st_ops->progs);
7515                 zfree(&map->st_ops->kern_func_off);
7516                 zfree(&map->st_ops);
7517         }
7518
7519         zfree(&map->name);
7520         zfree(&map->pin_path);
7521
7522         if (map->fd >= 0)
7523                 zclose(map->fd);
7524 }
7525
7526 void bpf_object__close(struct bpf_object *obj)
7527 {
7528         size_t i;
7529
7530         if (IS_ERR_OR_NULL(obj))
7531                 return;
7532
7533         if (obj->clear_priv)
7534                 obj->clear_priv(obj, obj->priv);
7535
7536         bpf_gen__free(obj->gen_loader);
7537         bpf_object__elf_finish(obj);
7538         bpf_object__unload(obj);
7539         btf__free(obj->btf);
7540         btf_ext__free(obj->btf_ext);
7541
7542         for (i = 0; i < obj->nr_maps; i++)
7543                 bpf_map__destroy(&obj->maps[i]);
7544
7545         zfree(&obj->btf_custom_path);
7546         zfree(&obj->kconfig);
7547         zfree(&obj->externs);
7548         obj->nr_extern = 0;
7549
7550         zfree(&obj->maps);
7551         obj->nr_maps = 0;
7552
7553         if (obj->programs && obj->nr_programs) {
7554                 for (i = 0; i < obj->nr_programs; i++)
7555                         bpf_program__exit(&obj->programs[i]);
7556         }
7557         zfree(&obj->programs);
7558
7559         list_del(&obj->list);
7560         free(obj);
7561 }
7562
7563 struct bpf_object *
7564 bpf_object__next(struct bpf_object *prev)
7565 {
7566         struct bpf_object *next;
7567
7568         if (!prev)
7569                 next = list_first_entry(&bpf_objects_list,
7570                                         struct bpf_object,
7571                                         list);
7572         else
7573                 next = list_next_entry(prev, list);
7574
7575         /* Empty list is noticed here so don't need checking on entry. */
7576         if (&next->list == &bpf_objects_list)
7577                 return NULL;
7578
7579         return next;
7580 }
7581
7582 const char *bpf_object__name(const struct bpf_object *obj)
7583 {
7584         return obj ? obj->name : libbpf_err_ptr(-EINVAL);
7585 }
7586
7587 unsigned int bpf_object__kversion(const struct bpf_object *obj)
7588 {
7589         return obj ? obj->kern_version : 0;
7590 }
7591
7592 struct btf *bpf_object__btf(const struct bpf_object *obj)
7593 {
7594         return obj ? obj->btf : NULL;
7595 }
7596
7597 int bpf_object__btf_fd(const struct bpf_object *obj)
7598 {
7599         return obj->btf ? btf__fd(obj->btf) : -1;
7600 }
7601
7602 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
7603 {
7604         if (obj->loaded)
7605                 return libbpf_err(-EINVAL);
7606
7607         obj->kern_version = kern_version;
7608
7609         return 0;
7610 }
7611
7612 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
7613                          bpf_object_clear_priv_t clear_priv)
7614 {
7615         if (obj->priv && obj->clear_priv)
7616                 obj->clear_priv(obj, obj->priv);
7617
7618         obj->priv = priv;
7619         obj->clear_priv = clear_priv;
7620         return 0;
7621 }
7622
7623 void *bpf_object__priv(const struct bpf_object *obj)
7624 {
7625         return obj ? obj->priv : libbpf_err_ptr(-EINVAL);
7626 }
7627
7628 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
7629 {
7630         struct bpf_gen *gen;
7631
7632         if (!opts)
7633                 return -EFAULT;
7634         if (!OPTS_VALID(opts, gen_loader_opts))
7635                 return -EINVAL;
7636         gen = calloc(sizeof(*gen), 1);
7637         if (!gen)
7638                 return -ENOMEM;
7639         gen->opts = opts;
7640         obj->gen_loader = gen;
7641         return 0;
7642 }
7643
7644 static struct bpf_program *
7645 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
7646                     bool forward)
7647 {
7648         size_t nr_programs = obj->nr_programs;
7649         ssize_t idx;
7650
7651         if (!nr_programs)
7652                 return NULL;
7653
7654         if (!p)
7655                 /* Iter from the beginning */
7656                 return forward ? &obj->programs[0] :
7657                         &obj->programs[nr_programs - 1];
7658
7659         if (p->obj != obj) {
7660                 pr_warn("error: program handler doesn't match object\n");
7661                 return errno = EINVAL, NULL;
7662         }
7663
7664         idx = (p - obj->programs) + (forward ? 1 : -1);
7665         if (idx >= obj->nr_programs || idx < 0)
7666                 return NULL;
7667         return &obj->programs[idx];
7668 }
7669
7670 struct bpf_program *
7671 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
7672 {
7673         struct bpf_program *prog = prev;
7674
7675         do {
7676                 prog = __bpf_program__iter(prog, obj, true);
7677         } while (prog && prog_is_subprog(obj, prog));
7678
7679         return prog;
7680 }
7681
7682 struct bpf_program *
7683 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
7684 {
7685         struct bpf_program *prog = next;
7686
7687         do {
7688                 prog = __bpf_program__iter(prog, obj, false);
7689         } while (prog && prog_is_subprog(obj, prog));
7690
7691         return prog;
7692 }
7693
7694 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
7695                           bpf_program_clear_priv_t clear_priv)
7696 {
7697         if (prog->priv && prog->clear_priv)
7698                 prog->clear_priv(prog, prog->priv);
7699
7700         prog->priv = priv;
7701         prog->clear_priv = clear_priv;
7702         return 0;
7703 }
7704
7705 void *bpf_program__priv(const struct bpf_program *prog)
7706 {
7707         return prog ? prog->priv : libbpf_err_ptr(-EINVAL);
7708 }
7709
7710 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
7711 {
7712         prog->prog_ifindex = ifindex;
7713 }
7714
7715 const char *bpf_program__name(const struct bpf_program *prog)
7716 {
7717         return prog->name;
7718 }
7719
7720 const char *bpf_program__section_name(const struct bpf_program *prog)
7721 {
7722         return prog->sec_name;
7723 }
7724
7725 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
7726 {
7727         const char *title;
7728
7729         title = prog->sec_name;
7730         if (needs_copy) {
7731                 title = strdup(title);
7732                 if (!title) {
7733                         pr_warn("failed to strdup program title\n");
7734                         return libbpf_err_ptr(-ENOMEM);
7735                 }
7736         }
7737
7738         return title;
7739 }
7740
7741 bool bpf_program__autoload(const struct bpf_program *prog)
7742 {
7743         return prog->load;
7744 }
7745
7746 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
7747 {
7748         if (prog->obj->loaded)
7749                 return libbpf_err(-EINVAL);
7750
7751         prog->load = autoload;
7752         return 0;
7753 }
7754
7755 int bpf_program__fd(const struct bpf_program *prog)
7756 {
7757         return bpf_program__nth_fd(prog, 0);
7758 }
7759
7760 size_t bpf_program__size(const struct bpf_program *prog)
7761 {
7762         return prog->insns_cnt * BPF_INSN_SZ;
7763 }
7764
7765 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
7766                           bpf_program_prep_t prep)
7767 {
7768         int *instances_fds;
7769
7770         if (nr_instances <= 0 || !prep)
7771                 return libbpf_err(-EINVAL);
7772
7773         if (prog->instances.nr > 0 || prog->instances.fds) {
7774                 pr_warn("Can't set pre-processor after loading\n");
7775                 return libbpf_err(-EINVAL);
7776         }
7777
7778         instances_fds = malloc(sizeof(int) * nr_instances);
7779         if (!instances_fds) {
7780                 pr_warn("alloc memory failed for fds\n");
7781                 return libbpf_err(-ENOMEM);
7782         }
7783
7784         /* fill all fd with -1 */
7785         memset(instances_fds, -1, sizeof(int) * nr_instances);
7786
7787         prog->instances.nr = nr_instances;
7788         prog->instances.fds = instances_fds;
7789         prog->preprocessor = prep;
7790         return 0;
7791 }
7792
7793 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
7794 {
7795         int fd;
7796
7797         if (!prog)
7798                 return libbpf_err(-EINVAL);
7799
7800         if (n >= prog->instances.nr || n < 0) {
7801                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
7802                         n, prog->name, prog->instances.nr);
7803                 return libbpf_err(-EINVAL);
7804         }
7805
7806         fd = prog->instances.fds[n];
7807         if (fd < 0) {
7808                 pr_warn("%dth instance of program '%s' is invalid\n",
7809                         n, prog->name);
7810                 return libbpf_err(-ENOENT);
7811         }
7812
7813         return fd;
7814 }
7815
7816 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog)
7817 {
7818         return prog->type;
7819 }
7820
7821 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
7822 {
7823         prog->type = type;
7824 }
7825
7826 static bool bpf_program__is_type(const struct bpf_program *prog,
7827                                  enum bpf_prog_type type)
7828 {
7829         return prog ? (prog->type == type) : false;
7830 }
7831
7832 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
7833 int bpf_program__set_##NAME(struct bpf_program *prog)           \
7834 {                                                               \
7835         if (!prog)                                              \
7836                 return libbpf_err(-EINVAL);                     \
7837         bpf_program__set_type(prog, TYPE);                      \
7838         return 0;                                               \
7839 }                                                               \
7840                                                                 \
7841 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
7842 {                                                               \
7843         return bpf_program__is_type(prog, TYPE);                \
7844 }                                                               \
7845
7846 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
7847 BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
7848 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
7849 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
7850 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
7851 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
7852 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
7853 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
7854 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
7855 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
7856 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
7857 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
7858 BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
7859
7860 enum bpf_attach_type
7861 bpf_program__get_expected_attach_type(const struct bpf_program *prog)
7862 {
7863         return prog->expected_attach_type;
7864 }
7865
7866 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
7867                                            enum bpf_attach_type type)
7868 {
7869         prog->expected_attach_type = type;
7870 }
7871
7872 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype_optional,           \
7873                           attachable, attach_btf)                           \
7874         {                                                                   \
7875                 .sec = string,                                              \
7876                 .len = sizeof(string) - 1,                                  \
7877                 .prog_type = ptype,                                         \
7878                 .expected_attach_type = eatype,                             \
7879                 .is_exp_attach_type_optional = eatype_optional,             \
7880                 .is_attachable = attachable,                                \
7881                 .is_attach_btf = attach_btf,                                \
7882         }
7883
7884 /* Programs that can NOT be attached. */
7885 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
7886
7887 /* Programs that can be attached. */
7888 #define BPF_APROG_SEC(string, ptype, atype) \
7889         BPF_PROG_SEC_IMPL(string, ptype, atype, true, 1, 0)
7890
7891 /* Programs that must specify expected attach type at load time. */
7892 #define BPF_EAPROG_SEC(string, ptype, eatype) \
7893         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 1, 0)
7894
7895 /* Programs that use BTF to identify attach point */
7896 #define BPF_PROG_BTF(string, ptype, eatype) \
7897         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 0, 1)
7898
7899 /* Programs that can be attached but attach type can't be identified by section
7900  * name. Kept for backward compatibility.
7901  */
7902 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
7903
7904 #define SEC_DEF(sec_pfx, ptype, ...) {                                      \
7905         .sec = sec_pfx,                                                     \
7906         .len = sizeof(sec_pfx) - 1,                                         \
7907         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
7908         __VA_ARGS__                                                         \
7909 }
7910
7911 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
7912                                       struct bpf_program *prog);
7913 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
7914                                   struct bpf_program *prog);
7915 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
7916                                       struct bpf_program *prog);
7917 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
7918                                      struct bpf_program *prog);
7919 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
7920                                    struct bpf_program *prog);
7921 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
7922                                     struct bpf_program *prog);
7923
7924 static const struct bpf_sec_def section_defs[] = {
7925         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
7926         BPF_EAPROG_SEC("sk_reuseport/migrate",  BPF_PROG_TYPE_SK_REUSEPORT,
7927                                                 BPF_SK_REUSEPORT_SELECT_OR_MIGRATE),
7928         BPF_EAPROG_SEC("sk_reuseport",          BPF_PROG_TYPE_SK_REUSEPORT,
7929                                                 BPF_SK_REUSEPORT_SELECT),
7930         SEC_DEF("kprobe/", KPROBE,
7931                 .attach_fn = attach_kprobe),
7932         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
7933         SEC_DEF("kretprobe/", KPROBE,
7934                 .attach_fn = attach_kprobe),
7935         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
7936         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
7937         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
7938         SEC_DEF("tracepoint/", TRACEPOINT,
7939                 .attach_fn = attach_tp),
7940         SEC_DEF("tp/", TRACEPOINT,
7941                 .attach_fn = attach_tp),
7942         SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
7943                 .attach_fn = attach_raw_tp),
7944         SEC_DEF("raw_tp/", RAW_TRACEPOINT,
7945                 .attach_fn = attach_raw_tp),
7946         SEC_DEF("tp_btf/", TRACING,
7947                 .expected_attach_type = BPF_TRACE_RAW_TP,
7948                 .is_attach_btf = true,
7949                 .attach_fn = attach_trace),
7950         SEC_DEF("fentry/", TRACING,
7951                 .expected_attach_type = BPF_TRACE_FENTRY,
7952                 .is_attach_btf = true,
7953                 .attach_fn = attach_trace),
7954         SEC_DEF("fmod_ret/", TRACING,
7955                 .expected_attach_type = BPF_MODIFY_RETURN,
7956                 .is_attach_btf = true,
7957                 .attach_fn = attach_trace),
7958         SEC_DEF("fexit/", TRACING,
7959                 .expected_attach_type = BPF_TRACE_FEXIT,
7960                 .is_attach_btf = true,
7961                 .attach_fn = attach_trace),
7962         SEC_DEF("fentry.s/", TRACING,
7963                 .expected_attach_type = BPF_TRACE_FENTRY,
7964                 .is_attach_btf = true,
7965                 .is_sleepable = true,
7966                 .attach_fn = attach_trace),
7967         SEC_DEF("fmod_ret.s/", TRACING,
7968                 .expected_attach_type = BPF_MODIFY_RETURN,
7969                 .is_attach_btf = true,
7970                 .is_sleepable = true,
7971                 .attach_fn = attach_trace),
7972         SEC_DEF("fexit.s/", TRACING,
7973                 .expected_attach_type = BPF_TRACE_FEXIT,
7974                 .is_attach_btf = true,
7975                 .is_sleepable = true,
7976                 .attach_fn = attach_trace),
7977         SEC_DEF("freplace/", EXT,
7978                 .is_attach_btf = true,
7979                 .attach_fn = attach_trace),
7980         SEC_DEF("lsm/", LSM,
7981                 .is_attach_btf = true,
7982                 .expected_attach_type = BPF_LSM_MAC,
7983                 .attach_fn = attach_lsm),
7984         SEC_DEF("lsm.s/", LSM,
7985                 .is_attach_btf = true,
7986                 .is_sleepable = true,
7987                 .expected_attach_type = BPF_LSM_MAC,
7988                 .attach_fn = attach_lsm),
7989         SEC_DEF("iter/", TRACING,
7990                 .expected_attach_type = BPF_TRACE_ITER,
7991                 .is_attach_btf = true,
7992                 .attach_fn = attach_iter),
7993         SEC_DEF("syscall", SYSCALL,
7994                 .is_sleepable = true),
7995         BPF_EAPROG_SEC("xdp_devmap/",           BPF_PROG_TYPE_XDP,
7996                                                 BPF_XDP_DEVMAP),
7997         BPF_EAPROG_SEC("xdp_cpumap/",           BPF_PROG_TYPE_XDP,
7998                                                 BPF_XDP_CPUMAP),
7999         BPF_APROG_SEC("xdp",                    BPF_PROG_TYPE_XDP,
8000                                                 BPF_XDP),
8001         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
8002         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
8003         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
8004         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
8005         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
8006         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
8007                                                 BPF_CGROUP_INET_INGRESS),
8008         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
8009                                                 BPF_CGROUP_INET_EGRESS),
8010         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
8011         BPF_EAPROG_SEC("cgroup/sock_create",    BPF_PROG_TYPE_CGROUP_SOCK,
8012                                                 BPF_CGROUP_INET_SOCK_CREATE),
8013         BPF_EAPROG_SEC("cgroup/sock_release",   BPF_PROG_TYPE_CGROUP_SOCK,
8014                                                 BPF_CGROUP_INET_SOCK_RELEASE),
8015         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
8016                                                 BPF_CGROUP_INET_SOCK_CREATE),
8017         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
8018                                                 BPF_CGROUP_INET4_POST_BIND),
8019         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
8020                                                 BPF_CGROUP_INET6_POST_BIND),
8021         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
8022                                                 BPF_CGROUP_DEVICE),
8023         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
8024                                                 BPF_CGROUP_SOCK_OPS),
8025         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
8026                                                 BPF_SK_SKB_STREAM_PARSER),
8027         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
8028                                                 BPF_SK_SKB_STREAM_VERDICT),
8029         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
8030         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
8031                                                 BPF_SK_MSG_VERDICT),
8032         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
8033                                                 BPF_LIRC_MODE2),
8034         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
8035                                                 BPF_FLOW_DISSECTOR),
8036         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8037                                                 BPF_CGROUP_INET4_BIND),
8038         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8039                                                 BPF_CGROUP_INET6_BIND),
8040         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8041                                                 BPF_CGROUP_INET4_CONNECT),
8042         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8043                                                 BPF_CGROUP_INET6_CONNECT),
8044         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8045                                                 BPF_CGROUP_UDP4_SENDMSG),
8046         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8047                                                 BPF_CGROUP_UDP6_SENDMSG),
8048         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8049                                                 BPF_CGROUP_UDP4_RECVMSG),
8050         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8051                                                 BPF_CGROUP_UDP6_RECVMSG),
8052         BPF_EAPROG_SEC("cgroup/getpeername4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8053                                                 BPF_CGROUP_INET4_GETPEERNAME),
8054         BPF_EAPROG_SEC("cgroup/getpeername6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8055                                                 BPF_CGROUP_INET6_GETPEERNAME),
8056         BPF_EAPROG_SEC("cgroup/getsockname4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8057                                                 BPF_CGROUP_INET4_GETSOCKNAME),
8058         BPF_EAPROG_SEC("cgroup/getsockname6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8059                                                 BPF_CGROUP_INET6_GETSOCKNAME),
8060         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
8061                                                 BPF_CGROUP_SYSCTL),
8062         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
8063                                                 BPF_CGROUP_GETSOCKOPT),
8064         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
8065                                                 BPF_CGROUP_SETSOCKOPT),
8066         BPF_PROG_SEC("struct_ops",              BPF_PROG_TYPE_STRUCT_OPS),
8067         BPF_EAPROG_SEC("sk_lookup/",            BPF_PROG_TYPE_SK_LOOKUP,
8068                                                 BPF_SK_LOOKUP),
8069 };
8070
8071 #undef BPF_PROG_SEC_IMPL
8072 #undef BPF_PROG_SEC
8073 #undef BPF_APROG_SEC
8074 #undef BPF_EAPROG_SEC
8075 #undef BPF_APROG_COMPAT
8076 #undef SEC_DEF
8077
8078 #define MAX_TYPE_NAME_SIZE 32
8079
8080 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
8081 {
8082         int i, n = ARRAY_SIZE(section_defs);
8083
8084         for (i = 0; i < n; i++) {
8085                 if (strncmp(sec_name,
8086                             section_defs[i].sec, section_defs[i].len))
8087                         continue;
8088                 return &section_defs[i];
8089         }
8090         return NULL;
8091 }
8092
8093 static char *libbpf_get_type_names(bool attach_type)
8094 {
8095         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
8096         char *buf;
8097
8098         buf = malloc(len);
8099         if (!buf)
8100                 return NULL;
8101
8102         buf[0] = '\0';
8103         /* Forge string buf with all available names */
8104         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8105                 if (attach_type && !section_defs[i].is_attachable)
8106                         continue;
8107
8108                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
8109                         free(buf);
8110                         return NULL;
8111                 }
8112                 strcat(buf, " ");
8113                 strcat(buf, section_defs[i].sec);
8114         }
8115
8116         return buf;
8117 }
8118
8119 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
8120                              enum bpf_attach_type *expected_attach_type)
8121 {
8122         const struct bpf_sec_def *sec_def;
8123         char *type_names;
8124
8125         if (!name)
8126                 return libbpf_err(-EINVAL);
8127
8128         sec_def = find_sec_def(name);
8129         if (sec_def) {
8130                 *prog_type = sec_def->prog_type;
8131                 *expected_attach_type = sec_def->expected_attach_type;
8132                 return 0;
8133         }
8134
8135         pr_debug("failed to guess program type from ELF section '%s'\n", name);
8136         type_names = libbpf_get_type_names(false);
8137         if (type_names != NULL) {
8138                 pr_debug("supported section(type) names are:%s\n", type_names);
8139                 free(type_names);
8140         }
8141
8142         return libbpf_err(-ESRCH);
8143 }
8144
8145 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
8146                                                      size_t offset)
8147 {
8148         struct bpf_map *map;
8149         size_t i;
8150
8151         for (i = 0; i < obj->nr_maps; i++) {
8152                 map = &obj->maps[i];
8153                 if (!bpf_map__is_struct_ops(map))
8154                         continue;
8155                 if (map->sec_offset <= offset &&
8156                     offset - map->sec_offset < map->def.value_size)
8157                         return map;
8158         }
8159
8160         return NULL;
8161 }
8162
8163 /* Collect the reloc from ELF and populate the st_ops->progs[] */
8164 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
8165                                             GElf_Shdr *shdr, Elf_Data *data)
8166 {
8167         const struct btf_member *member;
8168         struct bpf_struct_ops *st_ops;
8169         struct bpf_program *prog;
8170         unsigned int shdr_idx;
8171         const struct btf *btf;
8172         struct bpf_map *map;
8173         Elf_Data *symbols;
8174         unsigned int moff, insn_idx;
8175         const char *name;
8176         __u32 member_idx;
8177         GElf_Sym sym;
8178         GElf_Rel rel;
8179         int i, nrels;
8180
8181         symbols = obj->efile.symbols;
8182         btf = obj->btf;
8183         nrels = shdr->sh_size / shdr->sh_entsize;
8184         for (i = 0; i < nrels; i++) {
8185                 if (!gelf_getrel(data, i, &rel)) {
8186                         pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
8187                         return -LIBBPF_ERRNO__FORMAT;
8188                 }
8189
8190                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
8191                         pr_warn("struct_ops reloc: symbol %zx not found\n",
8192                                 (size_t)GELF_R_SYM(rel.r_info));
8193                         return -LIBBPF_ERRNO__FORMAT;
8194                 }
8195
8196                 name = elf_sym_str(obj, sym.st_name) ?: "<?>";
8197                 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
8198                 if (!map) {
8199                         pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
8200                                 (size_t)rel.r_offset);
8201                         return -EINVAL;
8202                 }
8203
8204                 moff = rel.r_offset - map->sec_offset;
8205                 shdr_idx = sym.st_shndx;
8206                 st_ops = map->st_ops;
8207                 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",
8208                          map->name,
8209                          (long long)(rel.r_info >> 32),
8210                          (long long)sym.st_value,
8211                          shdr_idx, (size_t)rel.r_offset,
8212                          map->sec_offset, sym.st_name, name);
8213
8214                 if (shdr_idx >= SHN_LORESERVE) {
8215                         pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
8216                                 map->name, (size_t)rel.r_offset, shdr_idx);
8217                         return -LIBBPF_ERRNO__RELOC;
8218                 }
8219                 if (sym.st_value % BPF_INSN_SZ) {
8220                         pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
8221                                 map->name, (unsigned long long)sym.st_value);
8222                         return -LIBBPF_ERRNO__FORMAT;
8223                 }
8224                 insn_idx = sym.st_value / BPF_INSN_SZ;
8225
8226                 member = find_member_by_offset(st_ops->type, moff * 8);
8227                 if (!member) {
8228                         pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
8229                                 map->name, moff);
8230                         return -EINVAL;
8231                 }
8232                 member_idx = member - btf_members(st_ops->type);
8233                 name = btf__name_by_offset(btf, member->name_off);
8234
8235                 if (!resolve_func_ptr(btf, member->type, NULL)) {
8236                         pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
8237                                 map->name, name);
8238                         return -EINVAL;
8239                 }
8240
8241                 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
8242                 if (!prog) {
8243                         pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
8244                                 map->name, shdr_idx, name);
8245                         return -EINVAL;
8246                 }
8247
8248                 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
8249                         const struct bpf_sec_def *sec_def;
8250
8251                         sec_def = find_sec_def(prog->sec_name);
8252                         if (sec_def &&
8253                             sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
8254                                 /* for pr_warn */
8255                                 prog->type = sec_def->prog_type;
8256                                 goto invalid_prog;
8257                         }
8258
8259                         prog->type = BPF_PROG_TYPE_STRUCT_OPS;
8260                         prog->attach_btf_id = st_ops->type_id;
8261                         prog->expected_attach_type = member_idx;
8262                 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
8263                            prog->attach_btf_id != st_ops->type_id ||
8264                            prog->expected_attach_type != member_idx) {
8265                         goto invalid_prog;
8266                 }
8267                 st_ops->progs[member_idx] = prog;
8268         }
8269
8270         return 0;
8271
8272 invalid_prog:
8273         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",
8274                 map->name, prog->name, prog->sec_name, prog->type,
8275                 prog->attach_btf_id, prog->expected_attach_type, name);
8276         return -EINVAL;
8277 }
8278
8279 #define BTF_TRACE_PREFIX "btf_trace_"
8280 #define BTF_LSM_PREFIX "bpf_lsm_"
8281 #define BTF_ITER_PREFIX "bpf_iter_"
8282 #define BTF_MAX_NAME_SIZE 128
8283
8284 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
8285                                 const char **prefix, int *kind)
8286 {
8287         switch (attach_type) {
8288         case BPF_TRACE_RAW_TP:
8289                 *prefix = BTF_TRACE_PREFIX;
8290                 *kind = BTF_KIND_TYPEDEF;
8291                 break;
8292         case BPF_LSM_MAC:
8293                 *prefix = BTF_LSM_PREFIX;
8294                 *kind = BTF_KIND_FUNC;
8295                 break;
8296         case BPF_TRACE_ITER:
8297                 *prefix = BTF_ITER_PREFIX;
8298                 *kind = BTF_KIND_FUNC;
8299                 break;
8300         default:
8301                 *prefix = "";
8302                 *kind = BTF_KIND_FUNC;
8303         }
8304 }
8305
8306 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
8307                                    const char *name, __u32 kind)
8308 {
8309         char btf_type_name[BTF_MAX_NAME_SIZE];
8310         int ret;
8311
8312         ret = snprintf(btf_type_name, sizeof(btf_type_name),
8313                        "%s%s", prefix, name);
8314         /* snprintf returns the number of characters written excluding the
8315          * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
8316          * indicates truncation.
8317          */
8318         if (ret < 0 || ret >= sizeof(btf_type_name))
8319                 return -ENAMETOOLONG;
8320         return btf__find_by_name_kind(btf, btf_type_name, kind);
8321 }
8322
8323 static inline int find_attach_btf_id(struct btf *btf, const char *name,
8324                                      enum bpf_attach_type attach_type)
8325 {
8326         const char *prefix;
8327         int kind;
8328
8329         btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
8330         return find_btf_by_prefix_kind(btf, prefix, name, kind);
8331 }
8332
8333 int libbpf_find_vmlinux_btf_id(const char *name,
8334                                enum bpf_attach_type attach_type)
8335 {
8336         struct btf *btf;
8337         int err;
8338
8339         btf = btf__load_vmlinux_btf();
8340         err = libbpf_get_error(btf);
8341         if (err) {
8342                 pr_warn("vmlinux BTF is not found\n");
8343                 return libbpf_err(err);
8344         }
8345
8346         err = find_attach_btf_id(btf, name, attach_type);
8347         if (err <= 0)
8348                 pr_warn("%s is not found in vmlinux BTF\n", name);
8349
8350         btf__free(btf);
8351         return libbpf_err(err);
8352 }
8353
8354 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
8355 {
8356         struct bpf_prog_info_linear *info_linear;
8357         struct bpf_prog_info *info;
8358         struct btf *btf;
8359         int err;
8360
8361         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
8362         err = libbpf_get_error(info_linear);
8363         if (err) {
8364                 pr_warn("failed get_prog_info_linear for FD %d\n",
8365                         attach_prog_fd);
8366                 return err;
8367         }
8368
8369         err = -EINVAL;
8370         info = &info_linear->info;
8371         if (!info->btf_id) {
8372                 pr_warn("The target program doesn't have BTF\n");
8373                 goto out;
8374         }
8375         btf = btf__load_from_kernel_by_id(info->btf_id);
8376         if (libbpf_get_error(btf)) {
8377                 pr_warn("Failed to get BTF of the program\n");
8378                 goto out;
8379         }
8380         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
8381         btf__free(btf);
8382         if (err <= 0) {
8383                 pr_warn("%s is not found in prog's BTF\n", name);
8384                 goto out;
8385         }
8386 out:
8387         free(info_linear);
8388         return err;
8389 }
8390
8391 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
8392                               enum bpf_attach_type attach_type,
8393                               int *btf_obj_fd, int *btf_type_id)
8394 {
8395         int ret, i;
8396
8397         ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
8398         if (ret > 0) {
8399                 *btf_obj_fd = 0; /* vmlinux BTF */
8400                 *btf_type_id = ret;
8401                 return 0;
8402         }
8403         if (ret != -ENOENT)
8404                 return ret;
8405
8406         ret = load_module_btfs(obj);
8407         if (ret)
8408                 return ret;
8409
8410         for (i = 0; i < obj->btf_module_cnt; i++) {
8411                 const struct module_btf *mod = &obj->btf_modules[i];
8412
8413                 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
8414                 if (ret > 0) {
8415                         *btf_obj_fd = mod->fd;
8416                         *btf_type_id = ret;
8417                         return 0;
8418                 }
8419                 if (ret == -ENOENT)
8420                         continue;
8421
8422                 return ret;
8423         }
8424
8425         return -ESRCH;
8426 }
8427
8428 static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id)
8429 {
8430         enum bpf_attach_type attach_type = prog->expected_attach_type;
8431         __u32 attach_prog_fd = prog->attach_prog_fd;
8432         const char *name = prog->sec_name, *attach_name;
8433         const struct bpf_sec_def *sec = NULL;
8434         int i, err = 0;
8435
8436         if (!name)
8437                 return -EINVAL;
8438
8439         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8440                 if (!section_defs[i].is_attach_btf)
8441                         continue;
8442                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
8443                         continue;
8444
8445                 sec = &section_defs[i];
8446                 break;
8447         }
8448
8449         if (!sec) {
8450                 pr_warn("failed to identify BTF ID based on ELF section name '%s'\n", name);
8451                 return -ESRCH;
8452         }
8453         attach_name = name + sec->len;
8454
8455         /* BPF program's BTF ID */
8456         if (attach_prog_fd) {
8457                 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
8458                 if (err < 0) {
8459                         pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
8460                                  attach_prog_fd, attach_name, err);
8461                         return err;
8462                 }
8463                 *btf_obj_fd = 0;
8464                 *btf_type_id = err;
8465                 return 0;
8466         }
8467
8468         /* kernel/module BTF ID */
8469         if (prog->obj->gen_loader) {
8470                 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
8471                 *btf_obj_fd = 0;
8472                 *btf_type_id = 1;
8473         } else {
8474                 err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
8475         }
8476         if (err) {
8477                 pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err);
8478                 return err;
8479         }
8480         return 0;
8481 }
8482
8483 int libbpf_attach_type_by_name(const char *name,
8484                                enum bpf_attach_type *attach_type)
8485 {
8486         char *type_names;
8487         int i;
8488
8489         if (!name)
8490                 return libbpf_err(-EINVAL);
8491
8492         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8493                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
8494                         continue;
8495                 if (!section_defs[i].is_attachable)
8496                         return libbpf_err(-EINVAL);
8497                 *attach_type = section_defs[i].expected_attach_type;
8498                 return 0;
8499         }
8500         pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
8501         type_names = libbpf_get_type_names(true);
8502         if (type_names != NULL) {
8503                 pr_debug("attachable section(type) names are:%s\n", type_names);
8504                 free(type_names);
8505         }
8506
8507         return libbpf_err(-EINVAL);
8508 }
8509
8510 int bpf_map__fd(const struct bpf_map *map)
8511 {
8512         return map ? map->fd : libbpf_err(-EINVAL);
8513 }
8514
8515 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
8516 {
8517         return map ? &map->def : libbpf_err_ptr(-EINVAL);
8518 }
8519
8520 const char *bpf_map__name(const struct bpf_map *map)
8521 {
8522         return map ? map->name : NULL;
8523 }
8524
8525 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
8526 {
8527         return map->def.type;
8528 }
8529
8530 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
8531 {
8532         if (map->fd >= 0)
8533                 return libbpf_err(-EBUSY);
8534         map->def.type = type;
8535         return 0;
8536 }
8537
8538 __u32 bpf_map__map_flags(const struct bpf_map *map)
8539 {
8540         return map->def.map_flags;
8541 }
8542
8543 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
8544 {
8545         if (map->fd >= 0)
8546                 return libbpf_err(-EBUSY);
8547         map->def.map_flags = flags;
8548         return 0;
8549 }
8550
8551 __u32 bpf_map__numa_node(const struct bpf_map *map)
8552 {
8553         return map->numa_node;
8554 }
8555
8556 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
8557 {
8558         if (map->fd >= 0)
8559                 return libbpf_err(-EBUSY);
8560         map->numa_node = numa_node;
8561         return 0;
8562 }
8563
8564 __u32 bpf_map__key_size(const struct bpf_map *map)
8565 {
8566         return map->def.key_size;
8567 }
8568
8569 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
8570 {
8571         if (map->fd >= 0)
8572                 return libbpf_err(-EBUSY);
8573         map->def.key_size = size;
8574         return 0;
8575 }
8576
8577 __u32 bpf_map__value_size(const struct bpf_map *map)
8578 {
8579         return map->def.value_size;
8580 }
8581
8582 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
8583 {
8584         if (map->fd >= 0)
8585                 return libbpf_err(-EBUSY);
8586         map->def.value_size = size;
8587         return 0;
8588 }
8589
8590 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
8591 {
8592         return map ? map->btf_key_type_id : 0;
8593 }
8594
8595 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
8596 {
8597         return map ? map->btf_value_type_id : 0;
8598 }
8599
8600 int bpf_map__set_priv(struct bpf_map *map, void *priv,
8601                      bpf_map_clear_priv_t clear_priv)
8602 {
8603         if (!map)
8604                 return libbpf_err(-EINVAL);
8605
8606         if (map->priv) {
8607                 if (map->clear_priv)
8608                         map->clear_priv(map, map->priv);
8609         }
8610
8611         map->priv = priv;
8612         map->clear_priv = clear_priv;
8613         return 0;
8614 }
8615
8616 void *bpf_map__priv(const struct bpf_map *map)
8617 {
8618         return map ? map->priv : libbpf_err_ptr(-EINVAL);
8619 }
8620
8621 int bpf_map__set_initial_value(struct bpf_map *map,
8622                                const void *data, size_t size)
8623 {
8624         if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
8625             size != map->def.value_size || map->fd >= 0)
8626                 return libbpf_err(-EINVAL);
8627
8628         memcpy(map->mmaped, data, size);
8629         return 0;
8630 }
8631
8632 const void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
8633 {
8634         if (!map->mmaped)
8635                 return NULL;
8636         *psize = map->def.value_size;
8637         return map->mmaped;
8638 }
8639
8640 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
8641 {
8642         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
8643 }
8644
8645 bool bpf_map__is_internal(const struct bpf_map *map)
8646 {
8647         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
8648 }
8649
8650 __u32 bpf_map__ifindex(const struct bpf_map *map)
8651 {
8652         return map->map_ifindex;
8653 }
8654
8655 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
8656 {
8657         if (map->fd >= 0)
8658                 return libbpf_err(-EBUSY);
8659         map->map_ifindex = ifindex;
8660         return 0;
8661 }
8662
8663 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
8664 {
8665         if (!bpf_map_type__is_map_in_map(map->def.type)) {
8666                 pr_warn("error: unsupported map type\n");
8667                 return libbpf_err(-EINVAL);
8668         }
8669         if (map->inner_map_fd != -1) {
8670                 pr_warn("error: inner_map_fd already specified\n");
8671                 return libbpf_err(-EINVAL);
8672         }
8673         zfree(&map->inner_map);
8674         map->inner_map_fd = fd;
8675         return 0;
8676 }
8677
8678 static struct bpf_map *
8679 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
8680 {
8681         ssize_t idx;
8682         struct bpf_map *s, *e;
8683
8684         if (!obj || !obj->maps)
8685                 return errno = EINVAL, NULL;
8686
8687         s = obj->maps;
8688         e = obj->maps + obj->nr_maps;
8689
8690         if ((m < s) || (m >= e)) {
8691                 pr_warn("error in %s: map handler doesn't belong to object\n",
8692                          __func__);
8693                 return errno = EINVAL, NULL;
8694         }
8695
8696         idx = (m - obj->maps) + i;
8697         if (idx >= obj->nr_maps || idx < 0)
8698                 return NULL;
8699         return &obj->maps[idx];
8700 }
8701
8702 struct bpf_map *
8703 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
8704 {
8705         if (prev == NULL)
8706                 return obj->maps;
8707
8708         return __bpf_map__iter(prev, obj, 1);
8709 }
8710
8711 struct bpf_map *
8712 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
8713 {
8714         if (next == NULL) {
8715                 if (!obj->nr_maps)
8716                         return NULL;
8717                 return obj->maps + obj->nr_maps - 1;
8718         }
8719
8720         return __bpf_map__iter(next, obj, -1);
8721 }
8722
8723 struct bpf_map *
8724 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
8725 {
8726         struct bpf_map *pos;
8727
8728         bpf_object__for_each_map(pos, obj) {
8729                 if (pos->name && !strcmp(pos->name, name))
8730                         return pos;
8731         }
8732         return errno = ENOENT, NULL;
8733 }
8734
8735 int
8736 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
8737 {
8738         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
8739 }
8740
8741 struct bpf_map *
8742 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
8743 {
8744         return libbpf_err_ptr(-ENOTSUP);
8745 }
8746
8747 long libbpf_get_error(const void *ptr)
8748 {
8749         if (!IS_ERR_OR_NULL(ptr))
8750                 return 0;
8751
8752         if (IS_ERR(ptr))
8753                 errno = -PTR_ERR(ptr);
8754
8755         /* If ptr == NULL, then errno should be already set by the failing
8756          * API, because libbpf never returns NULL on success and it now always
8757          * sets errno on error. So no extra errno handling for ptr == NULL
8758          * case.
8759          */
8760         return -errno;
8761 }
8762
8763 int bpf_prog_load(const char *file, enum bpf_prog_type type,
8764                   struct bpf_object **pobj, int *prog_fd)
8765 {
8766         struct bpf_prog_load_attr attr;
8767
8768         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
8769         attr.file = file;
8770         attr.prog_type = type;
8771         attr.expected_attach_type = 0;
8772
8773         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
8774 }
8775
8776 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
8777                         struct bpf_object **pobj, int *prog_fd)
8778 {
8779         struct bpf_object_open_attr open_attr = {};
8780         struct bpf_program *prog, *first_prog = NULL;
8781         struct bpf_object *obj;
8782         struct bpf_map *map;
8783         int err;
8784
8785         if (!attr)
8786                 return libbpf_err(-EINVAL);
8787         if (!attr->file)
8788                 return libbpf_err(-EINVAL);
8789
8790         open_attr.file = attr->file;
8791         open_attr.prog_type = attr->prog_type;
8792
8793         obj = bpf_object__open_xattr(&open_attr);
8794         err = libbpf_get_error(obj);
8795         if (err)
8796                 return libbpf_err(-ENOENT);
8797
8798         bpf_object__for_each_program(prog, obj) {
8799                 enum bpf_attach_type attach_type = attr->expected_attach_type;
8800                 /*
8801                  * to preserve backwards compatibility, bpf_prog_load treats
8802                  * attr->prog_type, if specified, as an override to whatever
8803                  * bpf_object__open guessed
8804                  */
8805                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
8806                         bpf_program__set_type(prog, attr->prog_type);
8807                         bpf_program__set_expected_attach_type(prog,
8808                                                               attach_type);
8809                 }
8810                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
8811                         /*
8812                          * we haven't guessed from section name and user
8813                          * didn't provide a fallback type, too bad...
8814                          */
8815                         bpf_object__close(obj);
8816                         return libbpf_err(-EINVAL);
8817                 }
8818
8819                 prog->prog_ifindex = attr->ifindex;
8820                 prog->log_level = attr->log_level;
8821                 prog->prog_flags |= attr->prog_flags;
8822                 if (!first_prog)
8823                         first_prog = prog;
8824         }
8825
8826         bpf_object__for_each_map(map, obj) {
8827                 if (!bpf_map__is_offload_neutral(map))
8828                         map->map_ifindex = attr->ifindex;
8829         }
8830
8831         if (!first_prog) {
8832                 pr_warn("object file doesn't contain bpf program\n");
8833                 bpf_object__close(obj);
8834                 return libbpf_err(-ENOENT);
8835         }
8836
8837         err = bpf_object__load(obj);
8838         if (err) {
8839                 bpf_object__close(obj);
8840                 return libbpf_err(err);
8841         }
8842
8843         *pobj = obj;
8844         *prog_fd = bpf_program__fd(first_prog);
8845         return 0;
8846 }
8847
8848 struct bpf_link {
8849         int (*detach)(struct bpf_link *link);
8850         void (*dealloc)(struct bpf_link *link);
8851         char *pin_path;         /* NULL, if not pinned */
8852         int fd;                 /* hook FD, -1 if not applicable */
8853         bool disconnected;
8854 };
8855
8856 /* Replace link's underlying BPF program with the new one */
8857 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
8858 {
8859         int ret;
8860
8861         ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
8862         return libbpf_err_errno(ret);
8863 }
8864
8865 /* Release "ownership" of underlying BPF resource (typically, BPF program
8866  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
8867  * link, when destructed through bpf_link__destroy() call won't attempt to
8868  * detach/unregisted that BPF resource. This is useful in situations where,
8869  * say, attached BPF program has to outlive userspace program that attached it
8870  * in the system. Depending on type of BPF program, though, there might be
8871  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
8872  * exit of userspace program doesn't trigger automatic detachment and clean up
8873  * inside the kernel.
8874  */
8875 void bpf_link__disconnect(struct bpf_link *link)
8876 {
8877         link->disconnected = true;
8878 }
8879
8880 int bpf_link__destroy(struct bpf_link *link)
8881 {
8882         int err = 0;
8883
8884         if (IS_ERR_OR_NULL(link))
8885                 return 0;
8886
8887         if (!link->disconnected && link->detach)
8888                 err = link->detach(link);
8889         if (link->pin_path)
8890                 free(link->pin_path);
8891         if (link->dealloc)
8892                 link->dealloc(link);
8893         else
8894                 free(link);
8895
8896         return libbpf_err(err);
8897 }
8898
8899 int bpf_link__fd(const struct bpf_link *link)
8900 {
8901         return link->fd;
8902 }
8903
8904 const char *bpf_link__pin_path(const struct bpf_link *link)
8905 {
8906         return link->pin_path;
8907 }
8908
8909 static int bpf_link__detach_fd(struct bpf_link *link)
8910 {
8911         return libbpf_err_errno(close(link->fd));
8912 }
8913
8914 struct bpf_link *bpf_link__open(const char *path)
8915 {
8916         struct bpf_link *link;
8917         int fd;
8918
8919         fd = bpf_obj_get(path);
8920         if (fd < 0) {
8921                 fd = -errno;
8922                 pr_warn("failed to open link at %s: %d\n", path, fd);
8923                 return libbpf_err_ptr(fd);
8924         }
8925
8926         link = calloc(1, sizeof(*link));
8927         if (!link) {
8928                 close(fd);
8929                 return libbpf_err_ptr(-ENOMEM);
8930         }
8931         link->detach = &bpf_link__detach_fd;
8932         link->fd = fd;
8933
8934         link->pin_path = strdup(path);
8935         if (!link->pin_path) {
8936                 bpf_link__destroy(link);
8937                 return libbpf_err_ptr(-ENOMEM);
8938         }
8939
8940         return link;
8941 }
8942
8943 int bpf_link__detach(struct bpf_link *link)
8944 {
8945         return bpf_link_detach(link->fd) ? -errno : 0;
8946 }
8947
8948 int bpf_link__pin(struct bpf_link *link, const char *path)
8949 {
8950         int err;
8951
8952         if (link->pin_path)
8953                 return libbpf_err(-EBUSY);
8954         err = make_parent_dir(path);
8955         if (err)
8956                 return libbpf_err(err);
8957         err = check_path(path);
8958         if (err)
8959                 return libbpf_err(err);
8960
8961         link->pin_path = strdup(path);
8962         if (!link->pin_path)
8963                 return libbpf_err(-ENOMEM);
8964
8965         if (bpf_obj_pin(link->fd, link->pin_path)) {
8966                 err = -errno;
8967                 zfree(&link->pin_path);
8968                 return libbpf_err(err);
8969         }
8970
8971         pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
8972         return 0;
8973 }
8974
8975 int bpf_link__unpin(struct bpf_link *link)
8976 {
8977         int err;
8978
8979         if (!link->pin_path)
8980                 return libbpf_err(-EINVAL);
8981
8982         err = unlink(link->pin_path);
8983         if (err != 0)
8984                 return -errno;
8985
8986         pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
8987         zfree(&link->pin_path);
8988         return 0;
8989 }
8990
8991 struct bpf_link_perf {
8992         struct bpf_link link;
8993         int perf_event_fd;
8994 };
8995
8996 static int bpf_link_perf_detach(struct bpf_link *link)
8997 {
8998         struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
8999         int err = 0;
9000
9001         if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
9002                 err = -errno;
9003
9004         if (perf_link->perf_event_fd != link->fd)
9005                 close(perf_link->perf_event_fd);
9006         close(link->fd);
9007
9008         return libbpf_err(err);
9009 }
9010
9011 static void bpf_link_perf_dealloc(struct bpf_link *link)
9012 {
9013         struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
9014
9015         free(perf_link);
9016 }
9017
9018 struct bpf_link *bpf_program__attach_perf_event_opts(struct bpf_program *prog, int pfd,
9019                                                      const struct bpf_perf_event_opts *opts)
9020 {
9021         char errmsg[STRERR_BUFSIZE];
9022         struct bpf_link_perf *link;
9023         int prog_fd, link_fd = -1, err;
9024
9025         if (!OPTS_VALID(opts, bpf_perf_event_opts))
9026                 return libbpf_err_ptr(-EINVAL);
9027
9028         if (pfd < 0) {
9029                 pr_warn("prog '%s': invalid perf event FD %d\n",
9030                         prog->name, pfd);
9031                 return libbpf_err_ptr(-EINVAL);
9032         }
9033         prog_fd = bpf_program__fd(prog);
9034         if (prog_fd < 0) {
9035                 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
9036                         prog->name);
9037                 return libbpf_err_ptr(-EINVAL);
9038         }
9039
9040         link = calloc(1, sizeof(*link));
9041         if (!link)
9042                 return libbpf_err_ptr(-ENOMEM);
9043         link->link.detach = &bpf_link_perf_detach;
9044         link->link.dealloc = &bpf_link_perf_dealloc;
9045         link->perf_event_fd = pfd;
9046
9047         if (kernel_supports(prog->obj, FEAT_PERF_LINK)) {
9048                 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
9049                         .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
9050
9051                 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
9052                 if (link_fd < 0) {
9053                         err = -errno;
9054                         pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
9055                                 prog->name, pfd,
9056                                 err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9057                         goto err_out;
9058                 }
9059                 link->link.fd = link_fd;
9060         } else {
9061                 if (OPTS_GET(opts, bpf_cookie, 0)) {
9062                         pr_warn("prog '%s': user context value is not supported\n", prog->name);
9063                         err = -EOPNOTSUPP;
9064                         goto err_out;
9065                 }
9066
9067                 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
9068                         err = -errno;
9069                         pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
9070                                 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9071                         if (err == -EPROTO)
9072                                 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
9073                                         prog->name, pfd);
9074                         goto err_out;
9075                 }
9076                 link->link.fd = pfd;
9077         }
9078         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9079                 err = -errno;
9080                 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
9081                         prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9082                 goto err_out;
9083         }
9084
9085         return &link->link;
9086 err_out:
9087         if (link_fd >= 0)
9088                 close(link_fd);
9089         free(link);
9090         return libbpf_err_ptr(err);
9091 }
9092
9093 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog, int pfd)
9094 {
9095         return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
9096 }
9097
9098 /*
9099  * this function is expected to parse integer in the range of [0, 2^31-1] from
9100  * given file using scanf format string fmt. If actual parsed value is
9101  * negative, the result might be indistinguishable from error
9102  */
9103 static int parse_uint_from_file(const char *file, const char *fmt)
9104 {
9105         char buf[STRERR_BUFSIZE];
9106         int err, ret;
9107         FILE *f;
9108
9109         f = fopen(file, "r");
9110         if (!f) {
9111                 err = -errno;
9112                 pr_debug("failed to open '%s': %s\n", file,
9113                          libbpf_strerror_r(err, buf, sizeof(buf)));
9114                 return err;
9115         }
9116         err = fscanf(f, fmt, &ret);
9117         if (err != 1) {
9118                 err = err == EOF ? -EIO : -errno;
9119                 pr_debug("failed to parse '%s': %s\n", file,
9120                         libbpf_strerror_r(err, buf, sizeof(buf)));
9121                 fclose(f);
9122                 return err;
9123         }
9124         fclose(f);
9125         return ret;
9126 }
9127
9128 static int determine_kprobe_perf_type(void)
9129 {
9130         const char *file = "/sys/bus/event_source/devices/kprobe/type";
9131
9132         return parse_uint_from_file(file, "%d\n");
9133 }
9134
9135 static int determine_uprobe_perf_type(void)
9136 {
9137         const char *file = "/sys/bus/event_source/devices/uprobe/type";
9138
9139         return parse_uint_from_file(file, "%d\n");
9140 }
9141
9142 static int determine_kprobe_retprobe_bit(void)
9143 {
9144         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
9145
9146         return parse_uint_from_file(file, "config:%d\n");
9147 }
9148
9149 static int determine_uprobe_retprobe_bit(void)
9150 {
9151         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
9152
9153         return parse_uint_from_file(file, "config:%d\n");
9154 }
9155
9156 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
9157 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
9158
9159 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
9160                                  uint64_t offset, int pid, size_t ref_ctr_off)
9161 {
9162         struct perf_event_attr attr = {};
9163         char errmsg[STRERR_BUFSIZE];
9164         int type, pfd, err;
9165
9166         if (ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
9167                 return -EINVAL;
9168
9169         type = uprobe ? determine_uprobe_perf_type()
9170                       : determine_kprobe_perf_type();
9171         if (type < 0) {
9172                 pr_warn("failed to determine %s perf type: %s\n",
9173                         uprobe ? "uprobe" : "kprobe",
9174                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
9175                 return type;
9176         }
9177         if (retprobe) {
9178                 int bit = uprobe ? determine_uprobe_retprobe_bit()
9179                                  : determine_kprobe_retprobe_bit();
9180
9181                 if (bit < 0) {
9182                         pr_warn("failed to determine %s retprobe bit: %s\n",
9183                                 uprobe ? "uprobe" : "kprobe",
9184                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
9185                         return bit;
9186                 }
9187                 attr.config |= 1 << bit;
9188         }
9189         attr.size = sizeof(attr);
9190         attr.type = type;
9191         attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
9192         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
9193         attr.config2 = offset;           /* kprobe_addr or probe_offset */
9194
9195         /* pid filter is meaningful only for uprobes */
9196         pfd = syscall(__NR_perf_event_open, &attr,
9197                       pid < 0 ? -1 : pid /* pid */,
9198                       pid == -1 ? 0 : -1 /* cpu */,
9199                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
9200         if (pfd < 0) {
9201                 err = -errno;
9202                 pr_warn("%s perf_event_open() failed: %s\n",
9203                         uprobe ? "uprobe" : "kprobe",
9204                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9205                 return err;
9206         }
9207         return pfd;
9208 }
9209
9210 struct bpf_link *
9211 bpf_program__attach_kprobe_opts(struct bpf_program *prog,
9212                                 const char *func_name,
9213                                 const struct bpf_kprobe_opts *opts)
9214 {
9215         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
9216         char errmsg[STRERR_BUFSIZE];
9217         struct bpf_link *link;
9218         unsigned long offset;
9219         bool retprobe;
9220         int pfd, err;
9221
9222         if (!OPTS_VALID(opts, bpf_kprobe_opts))
9223                 return libbpf_err_ptr(-EINVAL);
9224
9225         retprobe = OPTS_GET(opts, retprobe, false);
9226         offset = OPTS_GET(opts, offset, 0);
9227         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
9228
9229         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
9230                                     offset, -1 /* pid */, 0 /* ref_ctr_off */);
9231         if (pfd < 0) {
9232                 pr_warn("prog '%s': failed to create %s '%s' perf event: %s\n",
9233                         prog->name, retprobe ? "kretprobe" : "kprobe", func_name,
9234                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9235                 return libbpf_err_ptr(pfd);
9236         }
9237         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
9238         err = libbpf_get_error(link);
9239         if (err) {
9240                 close(pfd);
9241                 pr_warn("prog '%s': failed to attach to %s '%s': %s\n",
9242                         prog->name, retprobe ? "kretprobe" : "kprobe", func_name,
9243                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9244                 return libbpf_err_ptr(err);
9245         }
9246         return link;
9247 }
9248
9249 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
9250                                             bool retprobe,
9251                                             const char *func_name)
9252 {
9253         DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
9254                 .retprobe = retprobe,
9255         );
9256
9257         return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
9258 }
9259
9260 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
9261                                       struct bpf_program *prog)
9262 {
9263         DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
9264         unsigned long offset = 0;
9265         struct bpf_link *link;
9266         const char *func_name;
9267         char *func;
9268         int n, err;
9269
9270         func_name = prog->sec_name + sec->len;
9271         opts.retprobe = strcmp(sec->sec, "kretprobe/") == 0;
9272
9273         n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
9274         if (n < 1) {
9275                 err = -EINVAL;
9276                 pr_warn("kprobe name is invalid: %s\n", func_name);
9277                 return libbpf_err_ptr(err);
9278         }
9279         if (opts.retprobe && offset != 0) {
9280                 free(func);
9281                 err = -EINVAL;
9282                 pr_warn("kretprobes do not support offset specification\n");
9283                 return libbpf_err_ptr(err);
9284         }
9285
9286         opts.offset = offset;
9287         link = bpf_program__attach_kprobe_opts(prog, func, &opts);
9288         free(func);
9289         return link;
9290 }
9291
9292 LIBBPF_API struct bpf_link *
9293 bpf_program__attach_uprobe_opts(struct bpf_program *prog, pid_t pid,
9294                                 const char *binary_path, size_t func_offset,
9295                                 const struct bpf_uprobe_opts *opts)
9296 {
9297         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
9298         char errmsg[STRERR_BUFSIZE];
9299         struct bpf_link *link;
9300         size_t ref_ctr_off;
9301         int pfd, err;
9302         bool retprobe;
9303
9304         if (!OPTS_VALID(opts, bpf_uprobe_opts))
9305                 return libbpf_err_ptr(-EINVAL);
9306
9307         retprobe = OPTS_GET(opts, retprobe, false);
9308         ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
9309         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
9310
9311         pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
9312                                     func_offset, pid, ref_ctr_off);
9313         if (pfd < 0) {
9314                 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
9315                         prog->name, retprobe ? "uretprobe" : "uprobe",
9316                         binary_path, func_offset,
9317                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9318                 return libbpf_err_ptr(pfd);
9319         }
9320         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
9321         err = libbpf_get_error(link);
9322         if (err) {
9323                 close(pfd);
9324                 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
9325                         prog->name, retprobe ? "uretprobe" : "uprobe",
9326                         binary_path, func_offset,
9327                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9328                 return libbpf_err_ptr(err);
9329         }
9330         return link;
9331 }
9332
9333 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
9334                                             bool retprobe, pid_t pid,
9335                                             const char *binary_path,
9336                                             size_t func_offset)
9337 {
9338         DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
9339
9340         return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
9341 }
9342
9343 static int determine_tracepoint_id(const char *tp_category,
9344                                    const char *tp_name)
9345 {
9346         char file[PATH_MAX];
9347         int ret;
9348
9349         ret = snprintf(file, sizeof(file),
9350                        "/sys/kernel/debug/tracing/events/%s/%s/id",
9351                        tp_category, tp_name);
9352         if (ret < 0)
9353                 return -errno;
9354         if (ret >= sizeof(file)) {
9355                 pr_debug("tracepoint %s/%s path is too long\n",
9356                          tp_category, tp_name);
9357                 return -E2BIG;
9358         }
9359         return parse_uint_from_file(file, "%d\n");
9360 }
9361
9362 static int perf_event_open_tracepoint(const char *tp_category,
9363                                       const char *tp_name)
9364 {
9365         struct perf_event_attr attr = {};
9366         char errmsg[STRERR_BUFSIZE];
9367         int tp_id, pfd, err;
9368
9369         tp_id = determine_tracepoint_id(tp_category, tp_name);
9370         if (tp_id < 0) {
9371                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
9372                         tp_category, tp_name,
9373                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
9374                 return tp_id;
9375         }
9376
9377         attr.type = PERF_TYPE_TRACEPOINT;
9378         attr.size = sizeof(attr);
9379         attr.config = tp_id;
9380
9381         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
9382                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
9383         if (pfd < 0) {
9384                 err = -errno;
9385                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
9386                         tp_category, tp_name,
9387                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9388                 return err;
9389         }
9390         return pfd;
9391 }
9392
9393 struct bpf_link *bpf_program__attach_tracepoint_opts(struct bpf_program *prog,
9394                                                      const char *tp_category,
9395                                                      const char *tp_name,
9396                                                      const struct bpf_tracepoint_opts *opts)
9397 {
9398         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
9399         char errmsg[STRERR_BUFSIZE];
9400         struct bpf_link *link;
9401         int pfd, err;
9402
9403         if (!OPTS_VALID(opts, bpf_tracepoint_opts))
9404                 return libbpf_err_ptr(-EINVAL);
9405
9406         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
9407
9408         pfd = perf_event_open_tracepoint(tp_category, tp_name);
9409         if (pfd < 0) {
9410                 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
9411                         prog->name, tp_category, tp_name,
9412                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9413                 return libbpf_err_ptr(pfd);
9414         }
9415         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
9416         err = libbpf_get_error(link);
9417         if (err) {
9418                 close(pfd);
9419                 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
9420                         prog->name, tp_category, tp_name,
9421                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9422                 return libbpf_err_ptr(err);
9423         }
9424         return link;
9425 }
9426
9427 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
9428                                                 const char *tp_category,
9429                                                 const char *tp_name)
9430 {
9431         return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
9432 }
9433
9434 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
9435                                   struct bpf_program *prog)
9436 {
9437         char *sec_name, *tp_cat, *tp_name;
9438         struct bpf_link *link;
9439
9440         sec_name = strdup(prog->sec_name);
9441         if (!sec_name)
9442                 return libbpf_err_ptr(-ENOMEM);
9443
9444         /* extract "tp/<category>/<name>" */
9445         tp_cat = sec_name + sec->len;
9446         tp_name = strchr(tp_cat, '/');
9447         if (!tp_name) {
9448                 free(sec_name);
9449                 return libbpf_err_ptr(-EINVAL);
9450         }
9451         *tp_name = '\0';
9452         tp_name++;
9453
9454         link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
9455         free(sec_name);
9456         return link;
9457 }
9458
9459 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
9460                                                     const char *tp_name)
9461 {
9462         char errmsg[STRERR_BUFSIZE];
9463         struct bpf_link *link;
9464         int prog_fd, pfd;
9465
9466         prog_fd = bpf_program__fd(prog);
9467         if (prog_fd < 0) {
9468                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9469                 return libbpf_err_ptr(-EINVAL);
9470         }
9471
9472         link = calloc(1, sizeof(*link));
9473         if (!link)
9474                 return libbpf_err_ptr(-ENOMEM);
9475         link->detach = &bpf_link__detach_fd;
9476
9477         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
9478         if (pfd < 0) {
9479                 pfd = -errno;
9480                 free(link);
9481                 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
9482                         prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9483                 return libbpf_err_ptr(pfd);
9484         }
9485         link->fd = pfd;
9486         return link;
9487 }
9488
9489 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
9490                                       struct bpf_program *prog)
9491 {
9492         const char *tp_name = prog->sec_name + sec->len;
9493
9494         return bpf_program__attach_raw_tracepoint(prog, tp_name);
9495 }
9496
9497 /* Common logic for all BPF program types that attach to a btf_id */
9498 static struct bpf_link *bpf_program__attach_btf_id(struct bpf_program *prog)
9499 {
9500         char errmsg[STRERR_BUFSIZE];
9501         struct bpf_link *link;
9502         int prog_fd, pfd;
9503
9504         prog_fd = bpf_program__fd(prog);
9505         if (prog_fd < 0) {
9506                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9507                 return libbpf_err_ptr(-EINVAL);
9508         }
9509
9510         link = calloc(1, sizeof(*link));
9511         if (!link)
9512                 return libbpf_err_ptr(-ENOMEM);
9513         link->detach = &bpf_link__detach_fd;
9514
9515         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
9516         if (pfd < 0) {
9517                 pfd = -errno;
9518                 free(link);
9519                 pr_warn("prog '%s': failed to attach: %s\n",
9520                         prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9521                 return libbpf_err_ptr(pfd);
9522         }
9523         link->fd = pfd;
9524         return (struct bpf_link *)link;
9525 }
9526
9527 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
9528 {
9529         return bpf_program__attach_btf_id(prog);
9530 }
9531
9532 struct bpf_link *bpf_program__attach_lsm(struct bpf_program *prog)
9533 {
9534         return bpf_program__attach_btf_id(prog);
9535 }
9536
9537 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
9538                                      struct bpf_program *prog)
9539 {
9540         return bpf_program__attach_trace(prog);
9541 }
9542
9543 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
9544                                    struct bpf_program *prog)
9545 {
9546         return bpf_program__attach_lsm(prog);
9547 }
9548
9549 static struct bpf_link *
9550 bpf_program__attach_fd(struct bpf_program *prog, int target_fd, int btf_id,
9551                        const char *target_name)
9552 {
9553         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
9554                             .target_btf_id = btf_id);
9555         enum bpf_attach_type attach_type;
9556         char errmsg[STRERR_BUFSIZE];
9557         struct bpf_link *link;
9558         int prog_fd, link_fd;
9559
9560         prog_fd = bpf_program__fd(prog);
9561         if (prog_fd < 0) {
9562                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9563                 return libbpf_err_ptr(-EINVAL);
9564         }
9565
9566         link = calloc(1, sizeof(*link));
9567         if (!link)
9568                 return libbpf_err_ptr(-ENOMEM);
9569         link->detach = &bpf_link__detach_fd;
9570
9571         attach_type = bpf_program__get_expected_attach_type(prog);
9572         link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
9573         if (link_fd < 0) {
9574                 link_fd = -errno;
9575                 free(link);
9576                 pr_warn("prog '%s': failed to attach to %s: %s\n",
9577                         prog->name, target_name,
9578                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
9579                 return libbpf_err_ptr(link_fd);
9580         }
9581         link->fd = link_fd;
9582         return link;
9583 }
9584
9585 struct bpf_link *
9586 bpf_program__attach_cgroup(struct bpf_program *prog, int cgroup_fd)
9587 {
9588         return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
9589 }
9590
9591 struct bpf_link *
9592 bpf_program__attach_netns(struct bpf_program *prog, int netns_fd)
9593 {
9594         return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
9595 }
9596
9597 struct bpf_link *bpf_program__attach_xdp(struct bpf_program *prog, int ifindex)
9598 {
9599         /* target_fd/target_ifindex use the same field in LINK_CREATE */
9600         return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
9601 }
9602
9603 struct bpf_link *bpf_program__attach_freplace(struct bpf_program *prog,
9604                                               int target_fd,
9605                                               const char *attach_func_name)
9606 {
9607         int btf_id;
9608
9609         if (!!target_fd != !!attach_func_name) {
9610                 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
9611                         prog->name);
9612                 return libbpf_err_ptr(-EINVAL);
9613         }
9614
9615         if (prog->type != BPF_PROG_TYPE_EXT) {
9616                 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
9617                         prog->name);
9618                 return libbpf_err_ptr(-EINVAL);
9619         }
9620
9621         if (target_fd) {
9622                 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
9623                 if (btf_id < 0)
9624                         return libbpf_err_ptr(btf_id);
9625
9626                 return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
9627         } else {
9628                 /* no target, so use raw_tracepoint_open for compatibility
9629                  * with old kernels
9630                  */
9631                 return bpf_program__attach_trace(prog);
9632         }
9633 }
9634
9635 struct bpf_link *
9636 bpf_program__attach_iter(struct bpf_program *prog,
9637                          const struct bpf_iter_attach_opts *opts)
9638 {
9639         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
9640         char errmsg[STRERR_BUFSIZE];
9641         struct bpf_link *link;
9642         int prog_fd, link_fd;
9643         __u32 target_fd = 0;
9644
9645         if (!OPTS_VALID(opts, bpf_iter_attach_opts))
9646                 return libbpf_err_ptr(-EINVAL);
9647
9648         link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
9649         link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
9650
9651         prog_fd = bpf_program__fd(prog);
9652         if (prog_fd < 0) {
9653                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9654                 return libbpf_err_ptr(-EINVAL);
9655         }
9656
9657         link = calloc(1, sizeof(*link));
9658         if (!link)
9659                 return libbpf_err_ptr(-ENOMEM);
9660         link->detach = &bpf_link__detach_fd;
9661
9662         link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
9663                                   &link_create_opts);
9664         if (link_fd < 0) {
9665                 link_fd = -errno;
9666                 free(link);
9667                 pr_warn("prog '%s': failed to attach to iterator: %s\n",
9668                         prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
9669                 return libbpf_err_ptr(link_fd);
9670         }
9671         link->fd = link_fd;
9672         return link;
9673 }
9674
9675 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
9676                                     struct bpf_program *prog)
9677 {
9678         return bpf_program__attach_iter(prog, NULL);
9679 }
9680
9681 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
9682 {
9683         const struct bpf_sec_def *sec_def;
9684
9685         sec_def = find_sec_def(prog->sec_name);
9686         if (!sec_def || !sec_def->attach_fn)
9687                 return libbpf_err_ptr(-ESRCH);
9688
9689         return sec_def->attach_fn(sec_def, prog);
9690 }
9691
9692 static int bpf_link__detach_struct_ops(struct bpf_link *link)
9693 {
9694         __u32 zero = 0;
9695
9696         if (bpf_map_delete_elem(link->fd, &zero))
9697                 return -errno;
9698
9699         return 0;
9700 }
9701
9702 struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
9703 {
9704         struct bpf_struct_ops *st_ops;
9705         struct bpf_link *link;
9706         __u32 i, zero = 0;
9707         int err;
9708
9709         if (!bpf_map__is_struct_ops(map) || map->fd == -1)
9710                 return libbpf_err_ptr(-EINVAL);
9711
9712         link = calloc(1, sizeof(*link));
9713         if (!link)
9714                 return libbpf_err_ptr(-EINVAL);
9715
9716         st_ops = map->st_ops;
9717         for (i = 0; i < btf_vlen(st_ops->type); i++) {
9718                 struct bpf_program *prog = st_ops->progs[i];
9719                 void *kern_data;
9720                 int prog_fd;
9721
9722                 if (!prog)
9723                         continue;
9724
9725                 prog_fd = bpf_program__fd(prog);
9726                 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
9727                 *(unsigned long *)kern_data = prog_fd;
9728         }
9729
9730         err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
9731         if (err) {
9732                 err = -errno;
9733                 free(link);
9734                 return libbpf_err_ptr(err);
9735         }
9736
9737         link->detach = bpf_link__detach_struct_ops;
9738         link->fd = map->fd;
9739
9740         return link;
9741 }
9742
9743 enum bpf_perf_event_ret
9744 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
9745                            void **copy_mem, size_t *copy_size,
9746                            bpf_perf_event_print_t fn, void *private_data)
9747 {
9748         struct perf_event_mmap_page *header = mmap_mem;
9749         __u64 data_head = ring_buffer_read_head(header);
9750         __u64 data_tail = header->data_tail;
9751         void *base = ((__u8 *)header) + page_size;
9752         int ret = LIBBPF_PERF_EVENT_CONT;
9753         struct perf_event_header *ehdr;
9754         size_t ehdr_size;
9755
9756         while (data_head != data_tail) {
9757                 ehdr = base + (data_tail & (mmap_size - 1));
9758                 ehdr_size = ehdr->size;
9759
9760                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
9761                         void *copy_start = ehdr;
9762                         size_t len_first = base + mmap_size - copy_start;
9763                         size_t len_secnd = ehdr_size - len_first;
9764
9765                         if (*copy_size < ehdr_size) {
9766                                 free(*copy_mem);
9767                                 *copy_mem = malloc(ehdr_size);
9768                                 if (!*copy_mem) {
9769                                         *copy_size = 0;
9770                                         ret = LIBBPF_PERF_EVENT_ERROR;
9771                                         break;
9772                                 }
9773                                 *copy_size = ehdr_size;
9774                         }
9775
9776                         memcpy(*copy_mem, copy_start, len_first);
9777                         memcpy(*copy_mem + len_first, base, len_secnd);
9778                         ehdr = *copy_mem;
9779                 }
9780
9781                 ret = fn(ehdr, private_data);
9782                 data_tail += ehdr_size;
9783                 if (ret != LIBBPF_PERF_EVENT_CONT)
9784                         break;
9785         }
9786
9787         ring_buffer_write_tail(header, data_tail);
9788         return libbpf_err(ret);
9789 }
9790
9791 struct perf_buffer;
9792
9793 struct perf_buffer_params {
9794         struct perf_event_attr *attr;
9795         /* if event_cb is specified, it takes precendence */
9796         perf_buffer_event_fn event_cb;
9797         /* sample_cb and lost_cb are higher-level common-case callbacks */
9798         perf_buffer_sample_fn sample_cb;
9799         perf_buffer_lost_fn lost_cb;
9800         void *ctx;
9801         int cpu_cnt;
9802         int *cpus;
9803         int *map_keys;
9804 };
9805
9806 struct perf_cpu_buf {
9807         struct perf_buffer *pb;
9808         void *base; /* mmap()'ed memory */
9809         void *buf; /* for reconstructing segmented data */
9810         size_t buf_size;
9811         int fd;
9812         int cpu;
9813         int map_key;
9814 };
9815
9816 struct perf_buffer {
9817         perf_buffer_event_fn event_cb;
9818         perf_buffer_sample_fn sample_cb;
9819         perf_buffer_lost_fn lost_cb;
9820         void *ctx; /* passed into callbacks */
9821
9822         size_t page_size;
9823         size_t mmap_size;
9824         struct perf_cpu_buf **cpu_bufs;
9825         struct epoll_event *events;
9826         int cpu_cnt; /* number of allocated CPU buffers */
9827         int epoll_fd; /* perf event FD */
9828         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
9829 };
9830
9831 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
9832                                       struct perf_cpu_buf *cpu_buf)
9833 {
9834         if (!cpu_buf)
9835                 return;
9836         if (cpu_buf->base &&
9837             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
9838                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
9839         if (cpu_buf->fd >= 0) {
9840                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
9841                 close(cpu_buf->fd);
9842         }
9843         free(cpu_buf->buf);
9844         free(cpu_buf);
9845 }
9846
9847 void perf_buffer__free(struct perf_buffer *pb)
9848 {
9849         int i;
9850
9851         if (IS_ERR_OR_NULL(pb))
9852                 return;
9853         if (pb->cpu_bufs) {
9854                 for (i = 0; i < pb->cpu_cnt; i++) {
9855                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
9856
9857                         if (!cpu_buf)
9858                                 continue;
9859
9860                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
9861                         perf_buffer__free_cpu_buf(pb, cpu_buf);
9862                 }
9863                 free(pb->cpu_bufs);
9864         }
9865         if (pb->epoll_fd >= 0)
9866                 close(pb->epoll_fd);
9867         free(pb->events);
9868         free(pb);
9869 }
9870
9871 static struct perf_cpu_buf *
9872 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
9873                           int cpu, int map_key)
9874 {
9875         struct perf_cpu_buf *cpu_buf;
9876         char msg[STRERR_BUFSIZE];
9877         int err;
9878
9879         cpu_buf = calloc(1, sizeof(*cpu_buf));
9880         if (!cpu_buf)
9881                 return ERR_PTR(-ENOMEM);
9882
9883         cpu_buf->pb = pb;
9884         cpu_buf->cpu = cpu;
9885         cpu_buf->map_key = map_key;
9886
9887         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
9888                               -1, PERF_FLAG_FD_CLOEXEC);
9889         if (cpu_buf->fd < 0) {
9890                 err = -errno;
9891                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
9892                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
9893                 goto error;
9894         }
9895
9896         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
9897                              PROT_READ | PROT_WRITE, MAP_SHARED,
9898                              cpu_buf->fd, 0);
9899         if (cpu_buf->base == MAP_FAILED) {
9900                 cpu_buf->base = NULL;
9901                 err = -errno;
9902                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
9903                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
9904                 goto error;
9905         }
9906
9907         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9908                 err = -errno;
9909                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
9910                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
9911                 goto error;
9912         }
9913
9914         return cpu_buf;
9915
9916 error:
9917         perf_buffer__free_cpu_buf(pb, cpu_buf);
9918         return (struct perf_cpu_buf *)ERR_PTR(err);
9919 }
9920
9921 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
9922                                               struct perf_buffer_params *p);
9923
9924 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
9925                                      const struct perf_buffer_opts *opts)
9926 {
9927         struct perf_buffer_params p = {};
9928         struct perf_event_attr attr = { 0, };
9929
9930         attr.config = PERF_COUNT_SW_BPF_OUTPUT;
9931         attr.type = PERF_TYPE_SOFTWARE;
9932         attr.sample_type = PERF_SAMPLE_RAW;
9933         attr.sample_period = 1;
9934         attr.wakeup_events = 1;
9935
9936         p.attr = &attr;
9937         p.sample_cb = opts ? opts->sample_cb : NULL;
9938         p.lost_cb = opts ? opts->lost_cb : NULL;
9939         p.ctx = opts ? opts->ctx : NULL;
9940
9941         return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
9942 }
9943
9944 struct perf_buffer *
9945 perf_buffer__new_raw(int map_fd, size_t page_cnt,
9946                      const struct perf_buffer_raw_opts *opts)
9947 {
9948         struct perf_buffer_params p = {};
9949
9950         p.attr = opts->attr;
9951         p.event_cb = opts->event_cb;
9952         p.ctx = opts->ctx;
9953         p.cpu_cnt = opts->cpu_cnt;
9954         p.cpus = opts->cpus;
9955         p.map_keys = opts->map_keys;
9956
9957         return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
9958 }
9959
9960 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
9961                                               struct perf_buffer_params *p)
9962 {
9963         const char *online_cpus_file = "/sys/devices/system/cpu/online";
9964         struct bpf_map_info map;
9965         char msg[STRERR_BUFSIZE];
9966         struct perf_buffer *pb;
9967         bool *online = NULL;
9968         __u32 map_info_len;
9969         int err, i, j, n;
9970
9971         if (page_cnt & (page_cnt - 1)) {
9972                 pr_warn("page count should be power of two, but is %zu\n",
9973                         page_cnt);
9974                 return ERR_PTR(-EINVAL);
9975         }
9976
9977         /* best-effort sanity checks */
9978         memset(&map, 0, sizeof(map));
9979         map_info_len = sizeof(map);
9980         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
9981         if (err) {
9982                 err = -errno;
9983                 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
9984                  * -EBADFD, -EFAULT, or -E2BIG on real error
9985                  */
9986                 if (err != -EINVAL) {
9987                         pr_warn("failed to get map info for map FD %d: %s\n",
9988                                 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
9989                         return ERR_PTR(err);
9990                 }
9991                 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
9992                          map_fd);
9993         } else {
9994                 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
9995                         pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
9996                                 map.name);
9997                         return ERR_PTR(-EINVAL);
9998                 }
9999         }
10000
10001         pb = calloc(1, sizeof(*pb));
10002         if (!pb)
10003                 return ERR_PTR(-ENOMEM);
10004
10005         pb->event_cb = p->event_cb;
10006         pb->sample_cb = p->sample_cb;
10007         pb->lost_cb = p->lost_cb;
10008         pb->ctx = p->ctx;
10009
10010         pb->page_size = getpagesize();
10011         pb->mmap_size = pb->page_size * page_cnt;
10012         pb->map_fd = map_fd;
10013
10014         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
10015         if (pb->epoll_fd < 0) {
10016                 err = -errno;
10017                 pr_warn("failed to create epoll instance: %s\n",
10018                         libbpf_strerror_r(err, msg, sizeof(msg)));
10019                 goto error;
10020         }
10021
10022         if (p->cpu_cnt > 0) {
10023                 pb->cpu_cnt = p->cpu_cnt;
10024         } else {
10025                 pb->cpu_cnt = libbpf_num_possible_cpus();
10026                 if (pb->cpu_cnt < 0) {
10027                         err = pb->cpu_cnt;
10028                         goto error;
10029                 }
10030                 if (map.max_entries && map.max_entries < pb->cpu_cnt)
10031                         pb->cpu_cnt = map.max_entries;
10032         }
10033
10034         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
10035         if (!pb->events) {
10036                 err = -ENOMEM;
10037                 pr_warn("failed to allocate events: out of memory\n");
10038                 goto error;
10039         }
10040         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
10041         if (!pb->cpu_bufs) {
10042                 err = -ENOMEM;
10043                 pr_warn("failed to allocate buffers: out of memory\n");
10044                 goto error;
10045         }
10046
10047         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
10048         if (err) {
10049                 pr_warn("failed to get online CPU mask: %d\n", err);
10050                 goto error;
10051         }
10052
10053         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
10054                 struct perf_cpu_buf *cpu_buf;
10055                 int cpu, map_key;
10056
10057                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
10058                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
10059
10060                 /* in case user didn't explicitly requested particular CPUs to
10061                  * be attached to, skip offline/not present CPUs
10062                  */
10063                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
10064                         continue;
10065
10066                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
10067                 if (IS_ERR(cpu_buf)) {
10068                         err = PTR_ERR(cpu_buf);
10069                         goto error;
10070                 }
10071
10072                 pb->cpu_bufs[j] = cpu_buf;
10073
10074                 err = bpf_map_update_elem(pb->map_fd, &map_key,
10075                                           &cpu_buf->fd, 0);
10076                 if (err) {
10077                         err = -errno;
10078                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
10079                                 cpu, map_key, cpu_buf->fd,
10080                                 libbpf_strerror_r(err, msg, sizeof(msg)));
10081                         goto error;
10082                 }
10083
10084                 pb->events[j].events = EPOLLIN;
10085                 pb->events[j].data.ptr = cpu_buf;
10086                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
10087                               &pb->events[j]) < 0) {
10088                         err = -errno;
10089                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
10090                                 cpu, cpu_buf->fd,
10091                                 libbpf_strerror_r(err, msg, sizeof(msg)));
10092                         goto error;
10093                 }
10094                 j++;
10095         }
10096         pb->cpu_cnt = j;
10097         free(online);
10098
10099         return pb;
10100
10101 error:
10102         free(online);
10103         if (pb)
10104                 perf_buffer__free(pb);
10105         return ERR_PTR(err);
10106 }
10107
10108 struct perf_sample_raw {
10109         struct perf_event_header header;
10110         uint32_t size;
10111         char data[];
10112 };
10113
10114 struct perf_sample_lost {
10115         struct perf_event_header header;
10116         uint64_t id;
10117         uint64_t lost;
10118         uint64_t sample_id;
10119 };
10120
10121 static enum bpf_perf_event_ret
10122 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
10123 {
10124         struct perf_cpu_buf *cpu_buf = ctx;
10125         struct perf_buffer *pb = cpu_buf->pb;
10126         void *data = e;
10127
10128         /* user wants full control over parsing perf event */
10129         if (pb->event_cb)
10130                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
10131
10132         switch (e->type) {
10133         case PERF_RECORD_SAMPLE: {
10134                 struct perf_sample_raw *s = data;
10135
10136                 if (pb->sample_cb)
10137                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
10138                 break;
10139         }
10140         case PERF_RECORD_LOST: {
10141                 struct perf_sample_lost *s = data;
10142
10143                 if (pb->lost_cb)
10144                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
10145                 break;
10146         }
10147         default:
10148                 pr_warn("unknown perf sample type %d\n", e->type);
10149                 return LIBBPF_PERF_EVENT_ERROR;
10150         }
10151         return LIBBPF_PERF_EVENT_CONT;
10152 }
10153
10154 static int perf_buffer__process_records(struct perf_buffer *pb,
10155                                         struct perf_cpu_buf *cpu_buf)
10156 {
10157         enum bpf_perf_event_ret ret;
10158
10159         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
10160                                          pb->page_size, &cpu_buf->buf,
10161                                          &cpu_buf->buf_size,
10162                                          perf_buffer__process_record, cpu_buf);
10163         if (ret != LIBBPF_PERF_EVENT_CONT)
10164                 return ret;
10165         return 0;
10166 }
10167
10168 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
10169 {
10170         return pb->epoll_fd;
10171 }
10172
10173 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
10174 {
10175         int i, cnt, err;
10176
10177         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
10178         if (cnt < 0)
10179                 return -errno;
10180
10181         for (i = 0; i < cnt; i++) {
10182                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
10183
10184                 err = perf_buffer__process_records(pb, cpu_buf);
10185                 if (err) {
10186                         pr_warn("error while processing records: %d\n", err);
10187                         return libbpf_err(err);
10188                 }
10189         }
10190         return cnt;
10191 }
10192
10193 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
10194  * manager.
10195  */
10196 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
10197 {
10198         return pb->cpu_cnt;
10199 }
10200
10201 /*
10202  * Return perf_event FD of a ring buffer in *buf_idx* slot of
10203  * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
10204  * select()/poll()/epoll() Linux syscalls.
10205  */
10206 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
10207 {
10208         struct perf_cpu_buf *cpu_buf;
10209
10210         if (buf_idx >= pb->cpu_cnt)
10211                 return libbpf_err(-EINVAL);
10212
10213         cpu_buf = pb->cpu_bufs[buf_idx];
10214         if (!cpu_buf)
10215                 return libbpf_err(-ENOENT);
10216
10217         return cpu_buf->fd;
10218 }
10219
10220 /*
10221  * Consume data from perf ring buffer corresponding to slot *buf_idx* in
10222  * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
10223  * consume, do nothing and return success.
10224  * Returns:
10225  *   - 0 on success;
10226  *   - <0 on failure.
10227  */
10228 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
10229 {
10230         struct perf_cpu_buf *cpu_buf;
10231
10232         if (buf_idx >= pb->cpu_cnt)
10233                 return libbpf_err(-EINVAL);
10234
10235         cpu_buf = pb->cpu_bufs[buf_idx];
10236         if (!cpu_buf)
10237                 return libbpf_err(-ENOENT);
10238
10239         return perf_buffer__process_records(pb, cpu_buf);
10240 }
10241
10242 int perf_buffer__consume(struct perf_buffer *pb)
10243 {
10244         int i, err;
10245
10246         for (i = 0; i < pb->cpu_cnt; i++) {
10247                 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
10248
10249                 if (!cpu_buf)
10250                         continue;
10251
10252                 err = perf_buffer__process_records(pb, cpu_buf);
10253                 if (err) {
10254                         pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
10255                         return libbpf_err(err);
10256                 }
10257         }
10258         return 0;
10259 }
10260
10261 struct bpf_prog_info_array_desc {
10262         int     array_offset;   /* e.g. offset of jited_prog_insns */
10263         int     count_offset;   /* e.g. offset of jited_prog_len */
10264         int     size_offset;    /* > 0: offset of rec size,
10265                                  * < 0: fix size of -size_offset
10266                                  */
10267 };
10268
10269 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
10270         [BPF_PROG_INFO_JITED_INSNS] = {
10271                 offsetof(struct bpf_prog_info, jited_prog_insns),
10272                 offsetof(struct bpf_prog_info, jited_prog_len),
10273                 -1,
10274         },
10275         [BPF_PROG_INFO_XLATED_INSNS] = {
10276                 offsetof(struct bpf_prog_info, xlated_prog_insns),
10277                 offsetof(struct bpf_prog_info, xlated_prog_len),
10278                 -1,
10279         },
10280         [BPF_PROG_INFO_MAP_IDS] = {
10281                 offsetof(struct bpf_prog_info, map_ids),
10282                 offsetof(struct bpf_prog_info, nr_map_ids),
10283                 -(int)sizeof(__u32),
10284         },
10285         [BPF_PROG_INFO_JITED_KSYMS] = {
10286                 offsetof(struct bpf_prog_info, jited_ksyms),
10287                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
10288                 -(int)sizeof(__u64),
10289         },
10290         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
10291                 offsetof(struct bpf_prog_info, jited_func_lens),
10292                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
10293                 -(int)sizeof(__u32),
10294         },
10295         [BPF_PROG_INFO_FUNC_INFO] = {
10296                 offsetof(struct bpf_prog_info, func_info),
10297                 offsetof(struct bpf_prog_info, nr_func_info),
10298                 offsetof(struct bpf_prog_info, func_info_rec_size),
10299         },
10300         [BPF_PROG_INFO_LINE_INFO] = {
10301                 offsetof(struct bpf_prog_info, line_info),
10302                 offsetof(struct bpf_prog_info, nr_line_info),
10303                 offsetof(struct bpf_prog_info, line_info_rec_size),
10304         },
10305         [BPF_PROG_INFO_JITED_LINE_INFO] = {
10306                 offsetof(struct bpf_prog_info, jited_line_info),
10307                 offsetof(struct bpf_prog_info, nr_jited_line_info),
10308                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
10309         },
10310         [BPF_PROG_INFO_PROG_TAGS] = {
10311                 offsetof(struct bpf_prog_info, prog_tags),
10312                 offsetof(struct bpf_prog_info, nr_prog_tags),
10313                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
10314         },
10315
10316 };
10317
10318 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
10319                                            int offset)
10320 {
10321         __u32 *array = (__u32 *)info;
10322
10323         if (offset >= 0)
10324                 return array[offset / sizeof(__u32)];
10325         return -(int)offset;
10326 }
10327
10328 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
10329                                            int offset)
10330 {
10331         __u64 *array = (__u64 *)info;
10332
10333         if (offset >= 0)
10334                 return array[offset / sizeof(__u64)];
10335         return -(int)offset;
10336 }
10337
10338 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
10339                                          __u32 val)
10340 {
10341         __u32 *array = (__u32 *)info;
10342
10343         if (offset >= 0)
10344                 array[offset / sizeof(__u32)] = val;
10345 }
10346
10347 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
10348                                          __u64 val)
10349 {
10350         __u64 *array = (__u64 *)info;
10351
10352         if (offset >= 0)
10353                 array[offset / sizeof(__u64)] = val;
10354 }
10355
10356 struct bpf_prog_info_linear *
10357 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
10358 {
10359         struct bpf_prog_info_linear *info_linear;
10360         struct bpf_prog_info info = {};
10361         __u32 info_len = sizeof(info);
10362         __u32 data_len = 0;
10363         int i, err;
10364         void *ptr;
10365
10366         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
10367                 return libbpf_err_ptr(-EINVAL);
10368
10369         /* step 1: get array dimensions */
10370         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
10371         if (err) {
10372                 pr_debug("can't get prog info: %s", strerror(errno));
10373                 return libbpf_err_ptr(-EFAULT);
10374         }
10375
10376         /* step 2: calculate total size of all arrays */
10377         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10378                 bool include_array = (arrays & (1UL << i)) > 0;
10379                 struct bpf_prog_info_array_desc *desc;
10380                 __u32 count, size;
10381
10382                 desc = bpf_prog_info_array_desc + i;
10383
10384                 /* kernel is too old to support this field */
10385                 if (info_len < desc->array_offset + sizeof(__u32) ||
10386                     info_len < desc->count_offset + sizeof(__u32) ||
10387                     (desc->size_offset > 0 && info_len < desc->size_offset))
10388                         include_array = false;
10389
10390                 if (!include_array) {
10391                         arrays &= ~(1UL << i);  /* clear the bit */
10392                         continue;
10393                 }
10394
10395                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10396                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10397
10398                 data_len += count * size;
10399         }
10400
10401         /* step 3: allocate continuous memory */
10402         data_len = roundup(data_len, sizeof(__u64));
10403         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
10404         if (!info_linear)
10405                 return libbpf_err_ptr(-ENOMEM);
10406
10407         /* step 4: fill data to info_linear->info */
10408         info_linear->arrays = arrays;
10409         memset(&info_linear->info, 0, sizeof(info));
10410         ptr = info_linear->data;
10411
10412         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10413                 struct bpf_prog_info_array_desc *desc;
10414                 __u32 count, size;
10415
10416                 if ((arrays & (1UL << i)) == 0)
10417                         continue;
10418
10419                 desc  = bpf_prog_info_array_desc + i;
10420                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10421                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10422                 bpf_prog_info_set_offset_u32(&info_linear->info,
10423                                              desc->count_offset, count);
10424                 bpf_prog_info_set_offset_u32(&info_linear->info,
10425                                              desc->size_offset, size);
10426                 bpf_prog_info_set_offset_u64(&info_linear->info,
10427                                              desc->array_offset,
10428                                              ptr_to_u64(ptr));
10429                 ptr += count * size;
10430         }
10431
10432         /* step 5: call syscall again to get required arrays */
10433         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
10434         if (err) {
10435                 pr_debug("can't get prog info: %s", strerror(errno));
10436                 free(info_linear);
10437                 return libbpf_err_ptr(-EFAULT);
10438         }
10439
10440         /* step 6: verify the data */
10441         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10442                 struct bpf_prog_info_array_desc *desc;
10443                 __u32 v1, v2;
10444
10445                 if ((arrays & (1UL << i)) == 0)
10446                         continue;
10447
10448                 desc = bpf_prog_info_array_desc + i;
10449                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10450                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
10451                                                    desc->count_offset);
10452                 if (v1 != v2)
10453                         pr_warn("%s: mismatch in element count\n", __func__);
10454
10455                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10456                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
10457                                                    desc->size_offset);
10458                 if (v1 != v2)
10459                         pr_warn("%s: mismatch in rec size\n", __func__);
10460         }
10461
10462         /* step 7: update info_len and data_len */
10463         info_linear->info_len = sizeof(struct bpf_prog_info);
10464         info_linear->data_len = data_len;
10465
10466         return info_linear;
10467 }
10468
10469 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
10470 {
10471         int i;
10472
10473         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10474                 struct bpf_prog_info_array_desc *desc;
10475                 __u64 addr, offs;
10476
10477                 if ((info_linear->arrays & (1UL << i)) == 0)
10478                         continue;
10479
10480                 desc = bpf_prog_info_array_desc + i;
10481                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
10482                                                      desc->array_offset);
10483                 offs = addr - ptr_to_u64(info_linear->data);
10484                 bpf_prog_info_set_offset_u64(&info_linear->info,
10485                                              desc->array_offset, offs);
10486         }
10487 }
10488
10489 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
10490 {
10491         int i;
10492
10493         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10494                 struct bpf_prog_info_array_desc *desc;
10495                 __u64 addr, offs;
10496
10497                 if ((info_linear->arrays & (1UL << i)) == 0)
10498                         continue;
10499
10500                 desc = bpf_prog_info_array_desc + i;
10501                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
10502                                                      desc->array_offset);
10503                 addr = offs + ptr_to_u64(info_linear->data);
10504                 bpf_prog_info_set_offset_u64(&info_linear->info,
10505                                              desc->array_offset, addr);
10506         }
10507 }
10508
10509 int bpf_program__set_attach_target(struct bpf_program *prog,
10510                                    int attach_prog_fd,
10511                                    const char *attach_func_name)
10512 {
10513         int btf_obj_fd = 0, btf_id = 0, err;
10514
10515         if (!prog || attach_prog_fd < 0 || !attach_func_name)
10516                 return libbpf_err(-EINVAL);
10517
10518         if (prog->obj->loaded)
10519                 return libbpf_err(-EINVAL);
10520
10521         if (attach_prog_fd) {
10522                 btf_id = libbpf_find_prog_btf_id(attach_func_name,
10523                                                  attach_prog_fd);
10524                 if (btf_id < 0)
10525                         return libbpf_err(btf_id);
10526         } else {
10527                 /* load btf_vmlinux, if not yet */
10528                 err = bpf_object__load_vmlinux_btf(prog->obj, true);
10529                 if (err)
10530                         return libbpf_err(err);
10531                 err = find_kernel_btf_id(prog->obj, attach_func_name,
10532                                          prog->expected_attach_type,
10533                                          &btf_obj_fd, &btf_id);
10534                 if (err)
10535                         return libbpf_err(err);
10536         }
10537
10538         prog->attach_btf_id = btf_id;
10539         prog->attach_btf_obj_fd = btf_obj_fd;
10540         prog->attach_prog_fd = attach_prog_fd;
10541         return 0;
10542 }
10543
10544 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
10545 {
10546         int err = 0, n, len, start, end = -1;
10547         bool *tmp;
10548
10549         *mask = NULL;
10550         *mask_sz = 0;
10551
10552         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
10553         while (*s) {
10554                 if (*s == ',' || *s == '\n') {
10555                         s++;
10556                         continue;
10557                 }
10558                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
10559                 if (n <= 0 || n > 2) {
10560                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
10561                         err = -EINVAL;
10562                         goto cleanup;
10563                 } else if (n == 1) {
10564                         end = start;
10565                 }
10566                 if (start < 0 || start > end) {
10567                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
10568                                 start, end, s);
10569                         err = -EINVAL;
10570                         goto cleanup;
10571                 }
10572                 tmp = realloc(*mask, end + 1);
10573                 if (!tmp) {
10574                         err = -ENOMEM;
10575                         goto cleanup;
10576                 }
10577                 *mask = tmp;
10578                 memset(tmp + *mask_sz, 0, start - *mask_sz);
10579                 memset(tmp + start, 1, end - start + 1);
10580                 *mask_sz = end + 1;
10581                 s += len;
10582         }
10583         if (!*mask_sz) {
10584                 pr_warn("Empty CPU range\n");
10585                 return -EINVAL;
10586         }
10587         return 0;
10588 cleanup:
10589         free(*mask);
10590         *mask = NULL;
10591         return err;
10592 }
10593
10594 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
10595 {
10596         int fd, err = 0, len;
10597         char buf[128];
10598
10599         fd = open(fcpu, O_RDONLY);
10600         if (fd < 0) {
10601                 err = -errno;
10602                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
10603                 return err;
10604         }
10605         len = read(fd, buf, sizeof(buf));
10606         close(fd);
10607         if (len <= 0) {
10608                 err = len ? -errno : -EINVAL;
10609                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
10610                 return err;
10611         }
10612         if (len >= sizeof(buf)) {
10613                 pr_warn("CPU mask is too big in file %s\n", fcpu);
10614                 return -E2BIG;
10615         }
10616         buf[len] = '\0';
10617
10618         return parse_cpu_mask_str(buf, mask, mask_sz);
10619 }
10620
10621 int libbpf_num_possible_cpus(void)
10622 {
10623         static const char *fcpu = "/sys/devices/system/cpu/possible";
10624         static int cpus;
10625         int err, n, i, tmp_cpus;
10626         bool *mask;
10627
10628         tmp_cpus = READ_ONCE(cpus);
10629         if (tmp_cpus > 0)
10630                 return tmp_cpus;
10631
10632         err = parse_cpu_mask_file(fcpu, &mask, &n);
10633         if (err)
10634                 return libbpf_err(err);
10635
10636         tmp_cpus = 0;
10637         for (i = 0; i < n; i++) {
10638                 if (mask[i])
10639                         tmp_cpus++;
10640         }
10641         free(mask);
10642
10643         WRITE_ONCE(cpus, tmp_cpus);
10644         return tmp_cpus;
10645 }
10646
10647 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
10648                               const struct bpf_object_open_opts *opts)
10649 {
10650         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
10651                 .object_name = s->name,
10652         );
10653         struct bpf_object *obj;
10654         int i, err;
10655
10656         /* Attempt to preserve opts->object_name, unless overriden by user
10657          * explicitly. Overwriting object name for skeletons is discouraged,
10658          * as it breaks global data maps, because they contain object name
10659          * prefix as their own map name prefix. When skeleton is generated,
10660          * bpftool is making an assumption that this name will stay the same.
10661          */
10662         if (opts) {
10663                 memcpy(&skel_opts, opts, sizeof(*opts));
10664                 if (!opts->object_name)
10665                         skel_opts.object_name = s->name;
10666         }
10667
10668         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
10669         err = libbpf_get_error(obj);
10670         if (err) {
10671                 pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
10672                         s->name, err);
10673                 return libbpf_err(err);
10674         }
10675
10676         *s->obj = obj;
10677
10678         for (i = 0; i < s->map_cnt; i++) {
10679                 struct bpf_map **map = s->maps[i].map;
10680                 const char *name = s->maps[i].name;
10681                 void **mmaped = s->maps[i].mmaped;
10682
10683                 *map = bpf_object__find_map_by_name(obj, name);
10684                 if (!*map) {
10685                         pr_warn("failed to find skeleton map '%s'\n", name);
10686                         return libbpf_err(-ESRCH);
10687                 }
10688
10689                 /* externs shouldn't be pre-setup from user code */
10690                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
10691                         *mmaped = (*map)->mmaped;
10692         }
10693
10694         for (i = 0; i < s->prog_cnt; i++) {
10695                 struct bpf_program **prog = s->progs[i].prog;
10696                 const char *name = s->progs[i].name;
10697
10698                 *prog = bpf_object__find_program_by_name(obj, name);
10699                 if (!*prog) {
10700                         pr_warn("failed to find skeleton program '%s'\n", name);
10701                         return libbpf_err(-ESRCH);
10702                 }
10703         }
10704
10705         return 0;
10706 }
10707
10708 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
10709 {
10710         int i, err;
10711
10712         err = bpf_object__load(*s->obj);
10713         if (err) {
10714                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
10715                 return libbpf_err(err);
10716         }
10717
10718         for (i = 0; i < s->map_cnt; i++) {
10719                 struct bpf_map *map = *s->maps[i].map;
10720                 size_t mmap_sz = bpf_map_mmap_sz(map);
10721                 int prot, map_fd = bpf_map__fd(map);
10722                 void **mmaped = s->maps[i].mmaped;
10723
10724                 if (!mmaped)
10725                         continue;
10726
10727                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
10728                         *mmaped = NULL;
10729                         continue;
10730                 }
10731
10732                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
10733                         prot = PROT_READ;
10734                 else
10735                         prot = PROT_READ | PROT_WRITE;
10736
10737                 /* Remap anonymous mmap()-ed "map initialization image" as
10738                  * a BPF map-backed mmap()-ed memory, but preserving the same
10739                  * memory address. This will cause kernel to change process'
10740                  * page table to point to a different piece of kernel memory,
10741                  * but from userspace point of view memory address (and its
10742                  * contents, being identical at this point) will stay the
10743                  * same. This mapping will be released by bpf_object__close()
10744                  * as per normal clean up procedure, so we don't need to worry
10745                  * about it from skeleton's clean up perspective.
10746                  */
10747                 *mmaped = mmap(map->mmaped, mmap_sz, prot,
10748                                 MAP_SHARED | MAP_FIXED, map_fd, 0);
10749                 if (*mmaped == MAP_FAILED) {
10750                         err = -errno;
10751                         *mmaped = NULL;
10752                         pr_warn("failed to re-mmap() map '%s': %d\n",
10753                                  bpf_map__name(map), err);
10754                         return libbpf_err(err);
10755                 }
10756         }
10757
10758         return 0;
10759 }
10760
10761 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
10762 {
10763         int i, err;
10764
10765         for (i = 0; i < s->prog_cnt; i++) {
10766                 struct bpf_program *prog = *s->progs[i].prog;
10767                 struct bpf_link **link = s->progs[i].link;
10768                 const struct bpf_sec_def *sec_def;
10769
10770                 if (!prog->load)
10771                         continue;
10772
10773                 sec_def = find_sec_def(prog->sec_name);
10774                 if (!sec_def || !sec_def->attach_fn)
10775                         continue;
10776
10777                 *link = sec_def->attach_fn(sec_def, prog);
10778                 err = libbpf_get_error(*link);
10779                 if (err) {
10780                         pr_warn("failed to auto-attach program '%s': %d\n",
10781                                 bpf_program__name(prog), err);
10782                         return libbpf_err(err);
10783                 }
10784         }
10785
10786         return 0;
10787 }
10788
10789 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
10790 {
10791         int i;
10792
10793         for (i = 0; i < s->prog_cnt; i++) {
10794                 struct bpf_link **link = s->progs[i].link;
10795
10796                 bpf_link__destroy(*link);
10797                 *link = NULL;
10798         }
10799 }
10800
10801 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
10802 {
10803         if (s->progs)
10804                 bpf_object__detach_skeleton(s);
10805         if (s->obj)
10806                 bpf_object__close(*s->obj);
10807         free(s->maps);
10808         free(s->progs);
10809         free(s);
10810 }