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 <string.h>
  22 #include <unistd.h>
  23 #include <endian.h>
  24 #include <fcntl.h>
  25 #include <errno.h>
  26 #include <asm/unistd.h>
  27 #include <linux/err.h>
  28 #include <linux/kernel.h>
  29 #include <linux/bpf.h>
  30 #include <linux/btf.h>
  31 #include <linux/filter.h>
  32 #include <linux/list.h>
  33 #include <linux/limits.h>
  34 #include <linux/perf_event.h>
  35 #include <linux/ring_buffer.h>
  36 #include <linux/version.h>
  37 #include <sys/epoll.h>
  38 #include <sys/ioctl.h>
  39 #include <sys/mman.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/vfs.h>
  43 #include <sys/utsname.h>
  44 #include <tools/libc_compat.h>
  45 #include <libelf.h>
  46 #include <gelf.h>
  47
  48 #include "libbpf.h"
  49 #include "bpf.h"
  50 #include "btf.h"
  51 #include "str_error.h"
  52 #include "libbpf_internal.h"
  53 #include "hashmap.h"
  54
  55 #ifndef EM_BPF
  56 #define EM_BPF 247
  57 #endif
  58
  59 #ifndef BPF_FS_MAGIC
  60 #define BPF_FS_MAGIC            0xcafe4a11
  61 #endif
  62
  63 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  64  * compilation if user enables corresponding warning. Disable it explicitly.
  65  */
  66 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  67
  68 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  69
  70 static int __base_pr(enum libbpf_print_level level, const char *format,
  71                      va_list args)
  72 {
  73         if (level == LIBBPF_DEBUG)
  74                 return 0;
  75
  76         return vfprintf(stderr, format, args);
  77 }
  78
  79 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  80
  81 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  82 {
  83         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  84
  85         __libbpf_pr = fn;
  86         return old_print_fn;
  87 }
  88
  89 __printf(2, 3)
  90 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  91 {
  92         va_list args;
  93
  94         if (!__libbpf_pr)
  95                 return;
  96
  97         va_start(args, format);
  98         __libbpf_pr(level, format, args);
  99         va_end(args);
 100 }
 101
 102 #define STRERR_BUFSIZE  128
 103
 104 #define CHECK_ERR(action, err, out) do {        \
 105         err = action;                   \
 106         if (err)                        \
 107                 goto out;               \
 108 } while (0)
 109
 110
 111 /* Copied from tools/perf/util/util.h */
 112 #ifndef zfree
 113 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 114 #endif
 115
 116 #ifndef zclose
 117 # define zclose(fd) ({                  \
 118         int ___err = 0;                 \
 119         if ((fd) >= 0)                  \
 120                 ___err = close((fd));   \
 121         fd = -1;                        \
 122         ___err; })
 123 #endif
 124
 125 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 126 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 127 #else
 128 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 129 #endif
 130
 131 static inline __u64 ptr_to_u64(const void *ptr)
 132 {
 133         return (__u64) (unsigned long) ptr;
 134 }
 135
 136 struct bpf_capabilities {
 137         /* v4.14: kernel support for program & map names. */
 138         __u32 name:1;
 139         /* v5.2: kernel support for global data sections. */
 140         __u32 global_data:1;
 141         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 142         __u32 btf_func:1;
 143         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 144         __u32 btf_datasec:1;
 145         /* BPF_F_MMAPABLE is supported for arrays */
 146         __u32 array_mmap:1;
 147 };
 148
 149 /*
 150  * bpf_prog should be a better name but it has been used in
 151  * linux/filter.h.
 152  */
 153 struct bpf_program {
 154         /* Index in elf obj file, for relocation use. */
 155         int idx;
 156         char *name;
 157         int prog_ifindex;
 158         char *section_name;
 159         /* section_name with / replaced by _; makes recursive pinning
 160          * in bpf_object__pin_programs easier
 161          */
 162         char *pin_name;
 163         struct bpf_insn *insns;
 164         size_t insns_cnt, main_prog_cnt;
 165         enum bpf_prog_type type;
 166
 167         struct reloc_desc {
 168                 enum {
 169                         RELO_LD64,
 170                         RELO_CALL,
 171                         RELO_DATA,
 172                 } type;
 173                 int insn_idx;
 174                 int map_idx;
 175                 int sym_off;
 176         } *reloc_desc;
 177         int nr_reloc;
 178         int log_level;
 179
 180         struct {
 181                 int nr;
 182                 int *fds;
 183         } instances;
 184         bpf_program_prep_t preprocessor;
 185
 186         struct bpf_object *obj;
 187         void *priv;
 188         bpf_program_clear_priv_t clear_priv;
 189
 190         enum bpf_attach_type expected_attach_type;
 191         __u32 attach_btf_id;
 192         __u32 attach_prog_fd;
 193         void *func_info;
 194         __u32 func_info_rec_size;
 195         __u32 func_info_cnt;
 196
 197         struct bpf_capabilities *caps;
 198
 199         void *line_info;
 200         __u32 line_info_rec_size;
 201         __u32 line_info_cnt;
 202         __u32 prog_flags;
 203 };
 204
 205 enum libbpf_map_type {
 206         LIBBPF_MAP_UNSPEC,
 207         LIBBPF_MAP_DATA,
 208         LIBBPF_MAP_BSS,
 209         LIBBPF_MAP_RODATA,
 210 };
 211
 212 static const char * const libbpf_type_to_btf_name[] = {
 213         [LIBBPF_MAP_DATA]       = ".data",
 214         [LIBBPF_MAP_BSS]        = ".bss",
 215         [LIBBPF_MAP_RODATA]     = ".rodata",
 216 };
 217
 218 struct bpf_map {
 219         int fd;
 220         char *name;
 221         int sec_idx;
 222         size_t sec_offset;
 223         int map_ifindex;
 224         int inner_map_fd;
 225         struct bpf_map_def def;
 226         __u32 btf_key_type_id;
 227         __u32 btf_value_type_id;
 228         void *priv;
 229         bpf_map_clear_priv_t clear_priv;
 230         enum libbpf_map_type libbpf_type;
 231         char *pin_path;
 232         bool pinned;
 233         bool reused;
 234 };
 235
 236 struct bpf_secdata {
 237         void *rodata;
 238         void *data;
 239 };
 240
 241 static LIST_HEAD(bpf_objects_list);
 242
 243 struct bpf_object {
 244         char name[BPF_OBJ_NAME_LEN];
 245         char license[64];
 246         __u32 kern_version;
 247
 248         struct bpf_program *programs;
 249         size_t nr_programs;
 250         struct bpf_map *maps;
 251         size_t nr_maps;
 252         size_t maps_cap;
 253         struct bpf_secdata sections;
 254
 255         bool loaded;
 256         bool has_pseudo_calls;
 257         bool relaxed_core_relocs;
 258
 259         /*
 260          * Information when doing elf related work. Only valid if fd
 261          * is valid.
 262          */
 263         struct {
 264                 int fd;
 265                 const void *obj_buf;
 266                 size_t obj_buf_sz;
 267                 Elf *elf;
 268                 GElf_Ehdr ehdr;
 269                 Elf_Data *symbols;
 270                 Elf_Data *data;
 271                 Elf_Data *rodata;
 272                 Elf_Data *bss;
 273                 size_t strtabidx;
 274                 struct {
 275                         GElf_Shdr shdr;
 276                         Elf_Data *data;
 277                 } *reloc_sects;
 278                 int nr_reloc_sects;
 279                 int maps_shndx;
 280                 int btf_maps_shndx;
 281                 int text_shndx;
 282                 int data_shndx;
 283                 int rodata_shndx;
 284                 int bss_shndx;
 285         } efile;
 286         /*
 287          * All loaded bpf_object is linked in a list, which is
 288          * hidden to caller. bpf_objects__<func> handlers deal with
 289          * all objects.
 290          */
 291         struct list_head list;
 292
 293         struct btf *btf;
 294         struct btf_ext *btf_ext;
 295
 296         void *priv;
 297         bpf_object_clear_priv_t clear_priv;
 298
 299         struct bpf_capabilities caps;
 300
 301         char path[];
 302 };
 303 #define obj_elf_valid(o)        ((o)->efile.elf)
 304
 305 void bpf_program__unload(struct bpf_program *prog)
 306 {
 307         int i;
 308
 309         if (!prog)
 310                 return;
 311
 312         /*
 313          * If the object is opened but the program was never loaded,
 314          * it is possible that prog->instances.nr == -1.
 315          */
 316         if (prog->instances.nr > 0) {
 317                 for (i = 0; i < prog->instances.nr; i++)
 318                         zclose(prog->instances.fds[i]);
 319         } else if (prog->instances.nr != -1) {
 320                 pr_warn("Internal error: instances.nr is %d\n",
 321                         prog->instances.nr);
 322         }
 323
 324         prog->instances.nr = -1;
 325         zfree(&prog->instances.fds);
 326
 327         zfree(&prog->func_info);
 328         zfree(&prog->line_info);
 329 }
 330
 331 static void bpf_program__exit(struct bpf_program *prog)
 332 {
 333         if (!prog)
 334                 return;
 335
 336         if (prog->clear_priv)
 337                 prog->clear_priv(prog, prog->priv);
 338
 339         prog->priv = NULL;
 340         prog->clear_priv = NULL;
 341
 342         bpf_program__unload(prog);
 343         zfree(&prog->name);
 344         zfree(&prog->section_name);
 345         zfree(&prog->pin_name);
 346         zfree(&prog->insns);
 347         zfree(&prog->reloc_desc);
 348
 349         prog->nr_reloc = 0;
 350         prog->insns_cnt = 0;
 351         prog->idx = -1;
 352 }
 353
 354 static char *__bpf_program__pin_name(struct bpf_program *prog)
 355 {
 356         char *name, *p;
 357
 358         name = p = strdup(prog->section_name);
 359         while ((p = strchr(p, '/')))
 360                 *p = '_';
 361
 362         return name;
 363 }
 364
 365 static int
 366 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 367                   struct bpf_program *prog)
 368 {
 369         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 370
 371         if (size == 0 || size % bpf_insn_sz) {
 372                 pr_warn("corrupted section '%s', size: %zu\n",
 373                         section_name, size);
 374                 return -EINVAL;
 375         }
 376
 377         memset(prog, 0, sizeof(*prog));
 378
 379         prog->section_name = strdup(section_name);
 380         if (!prog->section_name) {
 381                 pr_warn("failed to alloc name for prog under section(%d) %s\n",
 382                         idx, section_name);
 383                 goto errout;
 384         }
 385
 386         prog->pin_name = __bpf_program__pin_name(prog);
 387         if (!prog->pin_name) {
 388                 pr_warn("failed to alloc pin name for prog under section(%d) %s\n",
 389                         idx, section_name);
 390                 goto errout;
 391         }
 392
 393         prog->insns = malloc(size);
 394         if (!prog->insns) {
 395                 pr_warn("failed to alloc insns for prog under section %s\n",
 396                         section_name);
 397                 goto errout;
 398         }
 399         prog->insns_cnt = size / bpf_insn_sz;
 400         memcpy(prog->insns, data, size);
 401         prog->idx = idx;
 402         prog->instances.fds = NULL;
 403         prog->instances.nr = -1;
 404         prog->type = BPF_PROG_TYPE_UNSPEC;
 405
 406         return 0;
 407 errout:
 408         bpf_program__exit(prog);
 409         return -ENOMEM;
 410 }
 411
 412 static int
 413 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 414                         char *section_name, int idx)
 415 {
 416         struct bpf_program prog, *progs;
 417         int nr_progs, err;
 418
 419         err = bpf_program__init(data, size, section_name, idx, &prog);
 420         if (err)
 421                 return err;
 422
 423         prog.caps = &obj->caps;
 424         progs = obj->programs;
 425         nr_progs = obj->nr_programs;
 426
 427         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 428         if (!progs) {
 429                 /*
 430                  * In this case the original obj->programs
 431                  * is still valid, so don't need special treat for
 432                  * bpf_close_object().
 433                  */
 434                 pr_warn("failed to alloc a new program under section '%s'\n",
 435                         section_name);
 436                 bpf_program__exit(&prog);
 437                 return -ENOMEM;
 438         }
 439
 440         pr_debug("found program %s\n", prog.section_name);
 441         obj->programs = progs;
 442         obj->nr_programs = nr_progs + 1;
 443         prog.obj = obj;
 444         progs[nr_progs] = prog;
 445         return 0;
 446 }
 447
 448 static int
 449 bpf_object__init_prog_names(struct bpf_object *obj)
 450 {
 451         Elf_Data *symbols = obj->efile.symbols;
 452         struct bpf_program *prog;
 453         size_t pi, si;
 454
 455         for (pi = 0; pi < obj->nr_programs; pi++) {
 456                 const char *name = NULL;
 457
 458                 prog = &obj->programs[pi];
 459
 460                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 461                      si++) {
 462                         GElf_Sym sym;
 463
 464                         if (!gelf_getsym(symbols, si, &sym))
 465                                 continue;
 466                         if (sym.st_shndx != prog->idx)
 467                                 continue;
 468                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 469                                 continue;
 470
 471                         name = elf_strptr(obj->efile.elf,
 472                                           obj->efile.strtabidx,
 473                                           sym.st_name);
 474                         if (!name) {
 475                                 pr_warn("failed to get sym name string for prog %s\n",
 476                                         prog->section_name);
 477                                 return -LIBBPF_ERRNO__LIBELF;
 478                         }
 479                 }
 480
 481                 if (!name && prog->idx == obj->efile.text_shndx)
 482                         name = ".text";
 483
 484                 if (!name) {
 485                         pr_warn("failed to find sym for prog %s\n",
 486                                 prog->section_name);
 487                         return -EINVAL;
 488                 }
 489
 490                 prog->name = strdup(name);
 491                 if (!prog->name) {
 492                         pr_warn("failed to allocate memory for prog sym %s\n",
 493                                 name);
 494                         return -ENOMEM;
 495                 }
 496         }
 497
 498         return 0;
 499 }
 500
 501 static __u32 get_kernel_version(void)
 502 {
 503         __u32 major, minor, patch;
 504         struct utsname info;
 505
 506         uname(&info);
 507         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 508                 return 0;
 509         return KERNEL_VERSION(major, minor, patch);
 510 }
 511
 512 static struct bpf_object *bpf_object__new(const char *path,
 513                                           const void *obj_buf,
 514                                           size_t obj_buf_sz,
 515                                           const char *obj_name)
 516 {
 517         struct bpf_object *obj;
 518         char *end;
 519
 520         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 521         if (!obj) {
 522                 pr_warn("alloc memory failed for %s\n", path);
 523                 return ERR_PTR(-ENOMEM);
 524         }
 525
 526         strcpy(obj->path, path);
 527         if (obj_name) {
 528                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
 529                 obj->name[sizeof(obj->name) - 1] = 0;
 530         } else {
 531                 /* Using basename() GNU version which doesn't modify arg. */
 532                 strncpy(obj->name, basename((void *)path),
 533                         sizeof(obj->name) - 1);
 534                 end = strchr(obj->name, '.');
 535                 if (end)
 536                         *end = 0;
 537         }
 538
 539         obj->efile.fd = -1;
 540         /*
 541          * Caller of this function should also call
 542          * bpf_object__elf_finish() after data collection to return
 543          * obj_buf to user. If not, we should duplicate the buffer to
 544          * avoid user freeing them before elf finish.
 545          */
 546         obj->efile.obj_buf = obj_buf;
 547         obj->efile.obj_buf_sz = obj_buf_sz;
 548         obj->efile.maps_shndx = -1;
 549         obj->efile.btf_maps_shndx = -1;
 550         obj->efile.data_shndx = -1;
 551         obj->efile.rodata_shndx = -1;
 552         obj->efile.bss_shndx = -1;
 553
 554         obj->kern_version = get_kernel_version();
 555         obj->loaded = false;
 556
 557         INIT_LIST_HEAD(&obj->list);
 558         list_add(&obj->list, &bpf_objects_list);
 559         return obj;
 560 }
 561
 562 static void bpf_object__elf_finish(struct bpf_object *obj)
 563 {
 564         if (!obj_elf_valid(obj))
 565                 return;
 566
 567         if (obj->efile.elf) {
 568                 elf_end(obj->efile.elf);
 569                 obj->efile.elf = NULL;
 570         }
 571         obj->efile.symbols = NULL;
 572         obj->efile.data = NULL;
 573         obj->efile.rodata = NULL;
 574         obj->efile.bss = NULL;
 575
 576         zfree(&obj->efile.reloc_sects);
 577         obj->efile.nr_reloc_sects = 0;
 578         zclose(obj->efile.fd);
 579         obj->efile.obj_buf = NULL;
 580         obj->efile.obj_buf_sz = 0;
 581 }
 582
 583 static int bpf_object__elf_init(struct bpf_object *obj)
 584 {
 585         int err = 0;
 586         GElf_Ehdr *ep;
 587
 588         if (obj_elf_valid(obj)) {
 589                 pr_warn("elf init: internal error\n");
 590                 return -LIBBPF_ERRNO__LIBELF;
 591         }
 592
 593         if (obj->efile.obj_buf_sz > 0) {
 594                 /*
 595                  * obj_buf should have been validated by
 596                  * bpf_object__open_buffer().
 597                  */
 598                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
 599                                             obj->efile.obj_buf_sz);
 600         } else {
 601                 obj->efile.fd = open(obj->path, O_RDONLY);
 602                 if (obj->efile.fd < 0) {
 603                         char errmsg[STRERR_BUFSIZE], *cp;
 604
 605                         err = -errno;
 606                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 607                         pr_warn("failed to open %s: %s\n", obj->path, cp);
 608                         return err;
 609                 }
 610
 611                 obj->efile.elf = elf_begin(obj->efile.fd,
 612                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 613         }
 614
 615         if (!obj->efile.elf) {
 616                 pr_warn("failed to open %s as ELF file\n", obj->path);
 617                 err = -LIBBPF_ERRNO__LIBELF;
 618                 goto errout;
 619         }
 620
 621         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 622                 pr_warn("failed to get EHDR from %s\n", obj->path);
 623                 err = -LIBBPF_ERRNO__FORMAT;
 624                 goto errout;
 625         }
 626         ep = &obj->efile.ehdr;
 627
 628         /* Old LLVM set e_machine to EM_NONE */
 629         if (ep->e_type != ET_REL ||
 630             (ep->e_machine && ep->e_machine != EM_BPF)) {
 631                 pr_warn("%s is not an eBPF object file\n", obj->path);
 632                 err = -LIBBPF_ERRNO__FORMAT;
 633                 goto errout;
 634         }
 635
 636         return 0;
 637 errout:
 638         bpf_object__elf_finish(obj);
 639         return err;
 640 }
 641
 642 static int bpf_object__check_endianness(struct bpf_object *obj)
 643 {
 644 #if __BYTE_ORDER == __LITTLE_ENDIAN
 645         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 646                 return 0;
 647 #elif __BYTE_ORDER == __BIG_ENDIAN
 648         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 649                 return 0;
 650 #else
 651 # error "Unrecognized __BYTE_ORDER__"
 652 #endif
 653         pr_warn("endianness mismatch.\n");
 654         return -LIBBPF_ERRNO__ENDIAN;
 655 }
 656
 657 static int
 658 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 659 {
 660         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 661         pr_debug("license of %s is %s\n", obj->path, obj->license);
 662         return 0;
 663 }
 664
 665 static int
 666 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 667 {
 668         __u32 kver;
 669
 670         if (size != sizeof(kver)) {
 671                 pr_warn("invalid kver section in %s\n", obj->path);
 672                 return -LIBBPF_ERRNO__FORMAT;
 673         }
 674         memcpy(&kver, data, sizeof(kver));
 675         obj->kern_version = kver;
 676         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 677         return 0;
 678 }
 679
 680 static int compare_bpf_map(const void *_a, const void *_b)
 681 {
 682         const struct bpf_map *a = _a;
 683         const struct bpf_map *b = _b;
 684
 685         if (a->sec_idx != b->sec_idx)
 686                 return a->sec_idx - b->sec_idx;
 687         return a->sec_offset - b->sec_offset;
 688 }
 689
 690 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 691 {
 692         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 693             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 694                 return true;
 695         return false;
 696 }
 697
 698 static int bpf_object_search_section_size(const struct bpf_object *obj,
 699                                           const char *name, size_t *d_size)
 700 {
 701         const GElf_Ehdr *ep = &obj->efile.ehdr;
 702         Elf *elf = obj->efile.elf;
 703         Elf_Scn *scn = NULL;
 704         int idx = 0;
 705
 706         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 707                 const char *sec_name;
 708                 Elf_Data *data;
 709                 GElf_Shdr sh;
 710
 711                 idx++;
 712                 if (gelf_getshdr(scn, &sh) != &sh) {
 713                         pr_warn("failed to get section(%d) header from %s\n",
 714                                 idx, obj->path);
 715                         return -EIO;
 716                 }
 717
 718                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 719                 if (!sec_name) {
 720                         pr_warn("failed to get section(%d) name from %s\n",
 721                                 idx, obj->path);
 722                         return -EIO;
 723                 }
 724
 725                 if (strcmp(name, sec_name))
 726                         continue;
 727
 728                 data = elf_getdata(scn, 0);
 729                 if (!data) {
 730                         pr_warn("failed to get section(%d) data from %s(%s)\n",
 731                                 idx, name, obj->path);
 732                         return -EIO;
 733                 }
 734
 735                 *d_size = data->d_size;
 736                 return 0;
 737         }
 738
 739         return -ENOENT;
 740 }
 741
 742 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 743                              __u32 *size)
 744 {
 745         int ret = -ENOENT;
 746         size_t d_size;
 747
 748         *size = 0;
 749         if (!name) {
 750                 return -EINVAL;
 751         } else if (!strcmp(name, ".data")) {
 752                 if (obj->efile.data)
 753                         *size = obj->efile.data->d_size;
 754         } else if (!strcmp(name, ".bss")) {
 755                 if (obj->efile.bss)
 756                         *size = obj->efile.bss->d_size;
 757         } else if (!strcmp(name, ".rodata")) {
 758                 if (obj->efile.rodata)
 759                         *size = obj->efile.rodata->d_size;
 760         } else {
 761                 ret = bpf_object_search_section_size(obj, name, &d_size);
 762                 if (!ret)
 763                         *size = d_size;
 764         }
 765
 766         return *size ? 0 : ret;
 767 }
 768
 769 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 770                                 __u32 *off)
 771 {
 772         Elf_Data *symbols = obj->efile.symbols;
 773         const char *sname;
 774         size_t si;
 775
 776         if (!name || !off)
 777                 return -EINVAL;
 778
 779         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 780                 GElf_Sym sym;
 781
 782                 if (!gelf_getsym(symbols, si, &sym))
 783                         continue;
 784                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 785                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 786                         continue;
 787
 788                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 789                                    sym.st_name);
 790                 if (!sname) {
 791                         pr_warn("failed to get sym name string for var %s\n",
 792                                 name);
 793                         return -EIO;
 794                 }
 795                 if (strcmp(name, sname) == 0) {
 796                         *off = sym.st_value;
 797                         return 0;
 798                 }
 799         }
 800
 801         return -ENOENT;
 802 }
 803
 804 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 805 {
 806         struct bpf_map *new_maps;
 807         size_t new_cap;
 808         int i;
 809
 810         if (obj->nr_maps < obj->maps_cap)
 811                 return &obj->maps[obj->nr_maps++];
 812
 813         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 814         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 815         if (!new_maps) {
 816                 pr_warn("alloc maps for object failed\n");
 817                 return ERR_PTR(-ENOMEM);
 818         }
 819
 820         obj->maps_cap = new_cap;
 821         obj->maps = new_maps;
 822
 823         /* zero out new maps */
 824         memset(obj->maps + obj->nr_maps, 0,
 825                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 826         /*
 827          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 828          * when failure (zclose won't close negative fd)).
 829          */
 830         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 831                 obj->maps[i].fd = -1;
 832                 obj->maps[i].inner_map_fd = -1;
 833         }
 834
 835         return &obj->maps[obj->nr_maps++];
 836 }
 837
 838 static int
 839 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 840                               int sec_idx, Elf_Data *data, void **data_buff)
 841 {
 842         char map_name[BPF_OBJ_NAME_LEN];
 843         struct bpf_map_def *def;
 844         struct bpf_map *map;
 845
 846         map = bpf_object__add_map(obj);
 847         if (IS_ERR(map))
 848                 return PTR_ERR(map);
 849
 850         map->libbpf_type = type;
 851         map->sec_idx = sec_idx;
 852         map->sec_offset = 0;
 853         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 854                  libbpf_type_to_btf_name[type]);
 855         map->name = strdup(map_name);
 856         if (!map->name) {
 857                 pr_warn("failed to alloc map name\n");
 858                 return -ENOMEM;
 859         }
 860
 861         def = &map->def;
 862         def->type = BPF_MAP_TYPE_ARRAY;
 863         def->key_size = sizeof(int);
 864         def->value_size = data->d_size;
 865         def->max_entries = 1;
 866         def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
 867         if (obj->caps.array_mmap)
 868                 def->map_flags |= BPF_F_MMAPABLE;
 869
 870         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
 871                  map_name, map->sec_idx, map->sec_offset, def->map_flags);
 872
 873         if (data_buff) {
 874                 *data_buff = malloc(data->d_size);
 875                 if (!*data_buff) {
 876                         zfree(&map->name);
 877                         pr_warn("failed to alloc map content buffer\n");
 878                         return -ENOMEM;
 879                 }
 880                 memcpy(*data_buff, data->d_buf, data->d_size);
 881         }
 882
 883         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 884         return 0;
 885 }
 886
 887 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 888 {
 889         int err;
 890
 891         if (!obj->caps.global_data)
 892                 return 0;
 893         /*
 894          * Populate obj->maps with libbpf internal maps.
 895          */
 896         if (obj->efile.data_shndx >= 0) {
 897                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 898                                                     obj->efile.data_shndx,
 899                                                     obj->efile.data,
 900                                                     &obj->sections.data);
 901                 if (err)
 902                         return err;
 903         }
 904         if (obj->efile.rodata_shndx >= 0) {
 905                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 906                                                     obj->efile.rodata_shndx,
 907                                                     obj->efile.rodata,
 908                                                     &obj->sections.rodata);
 909                 if (err)
 910                         return err;
 911         }
 912         if (obj->efile.bss_shndx >= 0) {
 913                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 914                                                     obj->efile.bss_shndx,
 915                                                     obj->efile.bss, NULL);
 916                 if (err)
 917                         return err;
 918         }
 919         return 0;
 920 }
 921
 922 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
 923 {
 924         Elf_Data *symbols = obj->efile.symbols;
 925         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
 926         Elf_Data *data = NULL;
 927         Elf_Scn *scn;
 928
 929         if (obj->efile.maps_shndx < 0)
 930                 return 0;
 931
 932         if (!symbols)
 933                 return -EINVAL;
 934
 935         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 936         if (scn)
 937                 data = elf_getdata(scn, NULL);
 938         if (!scn || !data) {
 939                 pr_warn("failed to get Elf_Data from map section %d\n",
 940                         obj->efile.maps_shndx);
 941                 return -EINVAL;
 942         }
 943
 944         /*
 945          * Count number of maps. Each map has a name.
 946          * Array of maps is not supported: only the first element is
 947          * considered.
 948          *
 949          * TODO: Detect array of map and report error.
 950          */
 951         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 952         for (i = 0; i < nr_syms; i++) {
 953                 GElf_Sym sym;
 954
 955                 if (!gelf_getsym(symbols, i, &sym))
 956                         continue;
 957                 if (sym.st_shndx != obj->efile.maps_shndx)
 958                         continue;
 959                 nr_maps++;
 960         }
 961         /* Assume equally sized map definitions */
 962         pr_debug("maps in %s: %d maps in %zd bytes\n",
 963                  obj->path, nr_maps, data->d_size);
 964
 965         if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
 966                 pr_warn("unable to determine map definition size section %s, %d maps in %zd bytes\n",
 967                         obj->path, nr_maps, data->d_size);
 968                 return -EINVAL;
 969         }
 970         map_def_sz = data->d_size / nr_maps;
 971
 972         /* Fill obj->maps using data in "maps" section.  */
 973         for (i = 0; i < nr_syms; i++) {
 974                 GElf_Sym sym;
 975                 const char *map_name;
 976                 struct bpf_map_def *def;
 977                 struct bpf_map *map;
 978
 979                 if (!gelf_getsym(symbols, i, &sym))
 980                         continue;
 981                 if (sym.st_shndx != obj->efile.maps_shndx)
 982                         continue;
 983
 984                 map = bpf_object__add_map(obj);
 985                 if (IS_ERR(map))
 986                         return PTR_ERR(map);
 987
 988                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 989                                       sym.st_name);
 990                 if (!map_name) {
 991                         pr_warn("failed to get map #%d name sym string for obj %s\n",
 992                                 i, obj->path);
 993                         return -LIBBPF_ERRNO__FORMAT;
 994                 }
 995
 996                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
 997                 map->sec_idx = sym.st_shndx;
 998                 map->sec_offset = sym.st_value;
 999                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1000                          map_name, map->sec_idx, map->sec_offset);
1001                 if (sym.st_value + map_def_sz > data->d_size) {
1002                         pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1003                                 obj->path, map_name);
1004                         return -EINVAL;
1005                 }
1006
1007                 map->name = strdup(map_name);
1008                 if (!map->name) {
1009                         pr_warn("failed to alloc map name\n");
1010                         return -ENOMEM;
1011                 }
1012                 pr_debug("map %d is \"%s\"\n", i, map->name);
1013                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1014                 /*
1015                  * If the definition of the map in the object file fits in
1016                  * bpf_map_def, copy it.  Any extra fields in our version
1017                  * of bpf_map_def will default to zero as a result of the
1018                  * calloc above.
1019                  */
1020                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1021                         memcpy(&map->def, def, map_def_sz);
1022                 } else {
1023                         /*
1024                          * Here the map structure being read is bigger than what
1025                          * we expect, truncate if the excess bits are all zero.
1026                          * If they are not zero, reject this map as
1027                          * incompatible.
1028                          */
1029                         char *b;
1030
1031                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1032                              b < ((char *)def) + map_def_sz; b++) {
1033                                 if (*b != 0) {
1034                                         pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1035                                                 obj->path, map_name);
1036                                         if (strict)
1037                                                 return -EINVAL;
1038                                 }
1039                         }
1040                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1041                 }
1042         }
1043         return 0;
1044 }
1045
1046 static const struct btf_type *
1047 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1048 {
1049         const struct btf_type *t = btf__type_by_id(btf, id);
1050
1051         if (res_id)
1052                 *res_id = id;
1053
1054         while (btf_is_mod(t) || btf_is_typedef(t)) {
1055                 if (res_id)
1056                         *res_id = t->type;
1057                 t = btf__type_by_id(btf, t->type);
1058         }
1059
1060         return t;
1061 }
1062
1063 /*
1064  * Fetch integer attribute of BTF map definition. Such attributes are
1065  * represented using a pointer to an array, in which dimensionality of array
1066  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1067  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1068  * type definition, while using only sizeof(void *) space in ELF data section.
1069  */
1070 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1071                               const struct btf_type *def,
1072                               const struct btf_member *m, __u32 *res)
1073 {
1074         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1075         const char *name = btf__name_by_offset(btf, m->name_off);
1076         const struct btf_array *arr_info;
1077         const struct btf_type *arr_t;
1078
1079         if (!btf_is_ptr(t)) {
1080                 pr_warn("map '%s': attr '%s': expected PTR, got %u.\n",
1081                         map_name, name, btf_kind(t));
1082                 return false;
1083         }
1084
1085         arr_t = btf__type_by_id(btf, t->type);
1086         if (!arr_t) {
1087                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1088                         map_name, name, t->type);
1089                 return false;
1090         }
1091         if (!btf_is_array(arr_t)) {
1092                 pr_warn("map '%s': attr '%s': expected ARRAY, got %u.\n",
1093                         map_name, name, btf_kind(arr_t));
1094                 return false;
1095         }
1096         arr_info = btf_array(arr_t);
1097         *res = arr_info->nelems;
1098         return true;
1099 }
1100
1101 static int build_map_pin_path(struct bpf_map *map, const char *path)
1102 {
1103         char buf[PATH_MAX];
1104         int err, len;
1105
1106         if (!path)
1107                 path = "/sys/fs/bpf";
1108
1109         len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1110         if (len < 0)
1111                 return -EINVAL;
1112         else if (len >= PATH_MAX)
1113                 return -ENAMETOOLONG;
1114
1115         err = bpf_map__set_pin_path(map, buf);
1116         if (err)
1117                 return err;
1118
1119         return 0;
1120 }
1121
1122 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1123                                          const struct btf_type *sec,
1124                                          int var_idx, int sec_idx,
1125                                          const Elf_Data *data, bool strict,
1126                                          const char *pin_root_path)
1127 {
1128         const struct btf_type *var, *def, *t;
1129         const struct btf_var_secinfo *vi;
1130         const struct btf_var *var_extra;
1131         const struct btf_member *m;
1132         const char *map_name;
1133         struct bpf_map *map;
1134         int vlen, i;
1135
1136         vi = btf_var_secinfos(sec) + var_idx;
1137         var = btf__type_by_id(obj->btf, vi->type);
1138         var_extra = btf_var(var);
1139         map_name = btf__name_by_offset(obj->btf, var->name_off);
1140         vlen = btf_vlen(var);
1141
1142         if (map_name == NULL || map_name[0] == '\0') {
1143                 pr_warn("map #%d: empty name.\n", var_idx);
1144                 return -EINVAL;
1145         }
1146         if ((__u64)vi->offset + vi->size > data->d_size) {
1147                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
1148                 return -EINVAL;
1149         }
1150         if (!btf_is_var(var)) {
1151                 pr_warn("map '%s': unexpected var kind %u.\n",
1152                         map_name, btf_kind(var));
1153                 return -EINVAL;
1154         }
1155         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1156             var_extra->linkage != BTF_VAR_STATIC) {
1157                 pr_warn("map '%s': unsupported var linkage %u.\n",
1158                         map_name, var_extra->linkage);
1159                 return -EOPNOTSUPP;
1160         }
1161
1162         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1163         if (!btf_is_struct(def)) {
1164                 pr_warn("map '%s': unexpected def kind %u.\n",
1165                         map_name, btf_kind(var));
1166                 return -EINVAL;
1167         }
1168         if (def->size > vi->size) {
1169                 pr_warn("map '%s': invalid def size.\n", map_name);
1170                 return -EINVAL;
1171         }
1172
1173         map = bpf_object__add_map(obj);
1174         if (IS_ERR(map))
1175                 return PTR_ERR(map);
1176         map->name = strdup(map_name);
1177         if (!map->name) {
1178                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
1179                 return -ENOMEM;
1180         }
1181         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1182         map->def.type = BPF_MAP_TYPE_UNSPEC;
1183         map->sec_idx = sec_idx;
1184         map->sec_offset = vi->offset;
1185         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1186                  map_name, map->sec_idx, map->sec_offset);
1187
1188         vlen = btf_vlen(def);
1189         m = btf_members(def);
1190         for (i = 0; i < vlen; i++, m++) {
1191                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1192
1193                 if (!name) {
1194                         pr_warn("map '%s': invalid field #%d.\n", map_name, i);
1195                         return -EINVAL;
1196                 }
1197                 if (strcmp(name, "type") == 0) {
1198                         if (!get_map_field_int(map_name, obj->btf, def, m,
1199                                                &map->def.type))
1200                                 return -EINVAL;
1201                         pr_debug("map '%s': found type = %u.\n",
1202                                  map_name, map->def.type);
1203                 } else if (strcmp(name, "max_entries") == 0) {
1204                         if (!get_map_field_int(map_name, obj->btf, def, m,
1205                                                &map->def.max_entries))
1206                                 return -EINVAL;
1207                         pr_debug("map '%s': found max_entries = %u.\n",
1208                                  map_name, map->def.max_entries);
1209                 } else if (strcmp(name, "map_flags") == 0) {
1210                         if (!get_map_field_int(map_name, obj->btf, def, m,
1211                                                &map->def.map_flags))
1212                                 return -EINVAL;
1213                         pr_debug("map '%s': found map_flags = %u.\n",
1214                                  map_name, map->def.map_flags);
1215                 } else if (strcmp(name, "key_size") == 0) {
1216                         __u32 sz;
1217
1218                         if (!get_map_field_int(map_name, obj->btf, def, m,
1219                                                &sz))
1220                                 return -EINVAL;
1221                         pr_debug("map '%s': found key_size = %u.\n",
1222                                  map_name, sz);
1223                         if (map->def.key_size && map->def.key_size != sz) {
1224                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
1225                                         map_name, map->def.key_size, sz);
1226                                 return -EINVAL;
1227                         }
1228                         map->def.key_size = sz;
1229                 } else if (strcmp(name, "key") == 0) {
1230                         __s64 sz;
1231
1232                         t = btf__type_by_id(obj->btf, m->type);
1233                         if (!t) {
1234                                 pr_warn("map '%s': key type [%d] not found.\n",
1235                                         map_name, m->type);
1236                                 return -EINVAL;
1237                         }
1238                         if (!btf_is_ptr(t)) {
1239                                 pr_warn("map '%s': key spec is not PTR: %u.\n",
1240                                         map_name, btf_kind(t));
1241                                 return -EINVAL;
1242                         }
1243                         sz = btf__resolve_size(obj->btf, t->type);
1244                         if (sz < 0) {
1245                                 pr_warn("map '%s': can't determine key size for type [%u]: %lld.\n",
1246                                         map_name, t->type, sz);
1247                                 return sz;
1248                         }
1249                         pr_debug("map '%s': found key [%u], sz = %lld.\n",
1250                                  map_name, t->type, sz);
1251                         if (map->def.key_size && map->def.key_size != sz) {
1252                                 pr_warn("map '%s': conflicting key size %u != %lld.\n",
1253                                         map_name, map->def.key_size, sz);
1254                                 return -EINVAL;
1255                         }
1256                         map->def.key_size = sz;
1257                         map->btf_key_type_id = t->type;
1258                 } else if (strcmp(name, "value_size") == 0) {
1259                         __u32 sz;
1260
1261                         if (!get_map_field_int(map_name, obj->btf, def, m,
1262                                                &sz))
1263                                 return -EINVAL;
1264                         pr_debug("map '%s': found value_size = %u.\n",
1265                                  map_name, sz);
1266                         if (map->def.value_size && map->def.value_size != sz) {
1267                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
1268                                         map_name, map->def.value_size, sz);
1269                                 return -EINVAL;
1270                         }
1271                         map->def.value_size = sz;
1272                 } else if (strcmp(name, "value") == 0) {
1273                         __s64 sz;
1274
1275                         t = btf__type_by_id(obj->btf, m->type);
1276                         if (!t) {
1277                                 pr_warn("map '%s': value type [%d] not found.\n",
1278                                         map_name, m->type);
1279                                 return -EINVAL;
1280                         }
1281                         if (!btf_is_ptr(t)) {
1282                                 pr_warn("map '%s': value spec is not PTR: %u.\n",
1283                                         map_name, btf_kind(t));
1284                                 return -EINVAL;
1285                         }
1286                         sz = btf__resolve_size(obj->btf, t->type);
1287                         if (sz < 0) {
1288                                 pr_warn("map '%s': can't determine value size for type [%u]: %lld.\n",
1289                                         map_name, t->type, sz);
1290                                 return sz;
1291                         }
1292                         pr_debug("map '%s': found value [%u], sz = %lld.\n",
1293                                  map_name, t->type, sz);
1294                         if (map->def.value_size && map->def.value_size != sz) {
1295                                 pr_warn("map '%s': conflicting value size %u != %lld.\n",
1296                                         map_name, map->def.value_size, sz);
1297                                 return -EINVAL;
1298                         }
1299                         map->def.value_size = sz;
1300                         map->btf_value_type_id = t->type;
1301                 } else if (strcmp(name, "pinning") == 0) {
1302                         __u32 val;
1303                         int err;
1304
1305                         if (!get_map_field_int(map_name, obj->btf, def, m,
1306                                                &val))
1307                                 return -EINVAL;
1308                         pr_debug("map '%s': found pinning = %u.\n",
1309                                  map_name, val);
1310
1311                         if (val != LIBBPF_PIN_NONE &&
1312                             val != LIBBPF_PIN_BY_NAME) {
1313                                 pr_warn("map '%s': invalid pinning value %u.\n",
1314                                         map_name, val);
1315                                 return -EINVAL;
1316                         }
1317                         if (val == LIBBPF_PIN_BY_NAME) {
1318                                 err = build_map_pin_path(map, pin_root_path);
1319                                 if (err) {
1320                                         pr_warn("map '%s': couldn't build pin path.\n",
1321                                                 map_name);
1322                                         return err;
1323                                 }
1324                         }
1325                 } else {
1326                         if (strict) {
1327                                 pr_warn("map '%s': unknown field '%s'.\n",
1328                                         map_name, name);
1329                                 return -ENOTSUP;
1330                         }
1331                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1332                                  map_name, name);
1333                 }
1334         }
1335
1336         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1337                 pr_warn("map '%s': map type isn't specified.\n", map_name);
1338                 return -EINVAL;
1339         }
1340
1341         return 0;
1342 }
1343
1344 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
1345                                           const char *pin_root_path)
1346 {
1347         const struct btf_type *sec = NULL;
1348         int nr_types, i, vlen, err;
1349         const struct btf_type *t;
1350         const char *name;
1351         Elf_Data *data;
1352         Elf_Scn *scn;
1353
1354         if (obj->efile.btf_maps_shndx < 0)
1355                 return 0;
1356
1357         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1358         if (scn)
1359                 data = elf_getdata(scn, NULL);
1360         if (!scn || !data) {
1361                 pr_warn("failed to get Elf_Data from map section %d (%s)\n",
1362                         obj->efile.maps_shndx, MAPS_ELF_SEC);
1363                 return -EINVAL;
1364         }
1365
1366         nr_types = btf__get_nr_types(obj->btf);
1367         for (i = 1; i <= nr_types; i++) {
1368                 t = btf__type_by_id(obj->btf, i);
1369                 if (!btf_is_datasec(t))
1370                         continue;
1371                 name = btf__name_by_offset(obj->btf, t->name_off);
1372                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1373                         sec = t;
1374                         break;
1375                 }
1376         }
1377
1378         if (!sec) {
1379                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1380                 return -ENOENT;
1381         }
1382
1383         vlen = btf_vlen(sec);
1384         for (i = 0; i < vlen; i++) {
1385                 err = bpf_object__init_user_btf_map(obj, sec, i,
1386                                                     obj->efile.btf_maps_shndx,
1387                                                     data, strict,
1388                                                     pin_root_path);
1389                 if (err)
1390                         return err;
1391         }
1392
1393         return 0;
1394 }
1395
1396 static int bpf_object__init_maps(struct bpf_object *obj, bool relaxed_maps,
1397                                  const char *pin_root_path)
1398 {
1399         bool strict = !relaxed_maps;
1400         int err;
1401
1402         err = bpf_object__init_user_maps(obj, strict);
1403         if (err)
1404                 return err;
1405
1406         err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
1407         if (err)
1408                 return err;
1409
1410         err = bpf_object__init_global_data_maps(obj);
1411         if (err)
1412                 return err;
1413
1414         if (obj->nr_maps) {
1415                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1416                       compare_bpf_map);
1417         }
1418         return 0;
1419 }
1420
1421 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1422 {
1423         Elf_Scn *scn;
1424         GElf_Shdr sh;
1425
1426         scn = elf_getscn(obj->efile.elf, idx);
1427         if (!scn)
1428                 return false;
1429
1430         if (gelf_getshdr(scn, &sh) != &sh)
1431                 return false;
1432
1433         if (sh.sh_flags & SHF_EXECINSTR)
1434                 return true;
1435
1436         return false;
1437 }
1438
1439 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1440 {
1441         bool has_datasec = obj->caps.btf_datasec;
1442         bool has_func = obj->caps.btf_func;
1443         struct btf *btf = obj->btf;
1444         struct btf_type *t;
1445         int i, j, vlen;
1446
1447         if (!obj->btf || (has_func && has_datasec))
1448                 return;
1449
1450         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1451                 t = (struct btf_type *)btf__type_by_id(btf, i);
1452
1453                 if (!has_datasec && btf_is_var(t)) {
1454                         /* replace VAR with INT */
1455                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1456                         /*
1457                          * using size = 1 is the safest choice, 4 will be too
1458                          * big and cause kernel BTF validation failure if
1459                          * original variable took less than 4 bytes
1460                          */
1461                         t->size = 1;
1462                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1463                 } else if (!has_datasec && btf_is_datasec(t)) {
1464                         /* replace DATASEC with STRUCT */
1465                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
1466                         struct btf_member *m = btf_members(t);
1467                         struct btf_type *vt;
1468                         char *name;
1469
1470                         name = (char *)btf__name_by_offset(btf, t->name_off);
1471                         while (*name) {
1472                                 if (*name == '.')
1473                                         *name = '_';
1474                                 name++;
1475                         }
1476
1477                         vlen = btf_vlen(t);
1478                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1479                         for (j = 0; j < vlen; j++, v++, m++) {
1480                                 /* order of field assignments is important */
1481                                 m->offset = v->offset * 8;
1482                                 m->type = v->type;
1483                                 /* preserve variable name as member name */
1484                                 vt = (void *)btf__type_by_id(btf, v->type);
1485                                 m->name_off = vt->name_off;
1486                         }
1487                 } else if (!has_func && btf_is_func_proto(t)) {
1488                         /* replace FUNC_PROTO with ENUM */
1489                         vlen = btf_vlen(t);
1490                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1491                         t->size = sizeof(__u32); /* kernel enforced */
1492                 } else if (!has_func && btf_is_func(t)) {
1493                         /* replace FUNC with TYPEDEF */
1494                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1495                 }
1496         }
1497 }
1498
1499 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1500 {
1501         if (!obj->btf_ext)
1502                 return;
1503
1504         if (!obj->caps.btf_func) {
1505                 btf_ext__free(obj->btf_ext);
1506                 obj->btf_ext = NULL;
1507         }
1508 }
1509
1510 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1511 {
1512         return obj->efile.btf_maps_shndx >= 0;
1513 }
1514
1515 static int bpf_object__init_btf(struct bpf_object *obj,
1516                                 Elf_Data *btf_data,
1517                                 Elf_Data *btf_ext_data)
1518 {
1519         bool btf_required = bpf_object__is_btf_mandatory(obj);
1520         int err = 0;
1521
1522         if (btf_data) {
1523                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1524                 if (IS_ERR(obj->btf)) {
1525                         pr_warn("Error loading ELF section %s: %d.\n",
1526                                 BTF_ELF_SEC, err);
1527                         goto out;
1528                 }
1529                 err = btf__finalize_data(obj, obj->btf);
1530                 if (err) {
1531                         pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
1532                         goto out;
1533                 }
1534         }
1535         if (btf_ext_data) {
1536                 if (!obj->btf) {
1537                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1538                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1539                         goto out;
1540                 }
1541                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1542                                             btf_ext_data->d_size);
1543                 if (IS_ERR(obj->btf_ext)) {
1544                         pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
1545                                 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1546                         obj->btf_ext = NULL;
1547                         goto out;
1548                 }
1549         }
1550 out:
1551         if (err || IS_ERR(obj->btf)) {
1552                 if (btf_required)
1553                         err = err ? : PTR_ERR(obj->btf);
1554                 else
1555                         err = 0;
1556                 if (!IS_ERR_OR_NULL(obj->btf))
1557                         btf__free(obj->btf);
1558                 obj->btf = NULL;
1559         }
1560         if (btf_required && !obj->btf) {
1561                 pr_warn("BTF is required, but is missing or corrupted.\n");
1562                 return err == 0 ? -ENOENT : err;
1563         }
1564         return 0;
1565 }
1566
1567 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1568 {
1569         int err = 0;
1570
1571         if (!obj->btf)
1572                 return 0;
1573
1574         bpf_object__sanitize_btf(obj);
1575         bpf_object__sanitize_btf_ext(obj);
1576
1577         err = btf__load(obj->btf);
1578         if (err) {
1579                 pr_warn("Error loading %s into kernel: %d.\n",
1580                         BTF_ELF_SEC, err);
1581                 btf__free(obj->btf);
1582                 obj->btf = NULL;
1583                 /* btf_ext can't exist without btf, so free it as well */
1584                 if (obj->btf_ext) {
1585                         btf_ext__free(obj->btf_ext);
1586                         obj->btf_ext = NULL;
1587                 }
1588
1589                 if (bpf_object__is_btf_mandatory(obj))
1590                         return err;
1591         }
1592         return 0;
1593 }
1594
1595 static int bpf_object__elf_collect(struct bpf_object *obj, bool relaxed_maps,
1596                                    const char *pin_root_path)
1597 {
1598         Elf *elf = obj->efile.elf;
1599         GElf_Ehdr *ep = &obj->efile.ehdr;
1600         Elf_Data *btf_ext_data = NULL;
1601         Elf_Data *btf_data = NULL;
1602         Elf_Scn *scn = NULL;
1603         int idx = 0, err = 0;
1604
1605         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1606         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1607                 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
1608                 return -LIBBPF_ERRNO__FORMAT;
1609         }
1610
1611         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1612                 char *name;
1613                 GElf_Shdr sh;
1614                 Elf_Data *data;
1615
1616                 idx++;
1617                 if (gelf_getshdr(scn, &sh) != &sh) {
1618                         pr_warn("failed to get section(%d) header from %s\n",
1619                                 idx, obj->path);
1620                         return -LIBBPF_ERRNO__FORMAT;
1621                 }
1622
1623                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1624                 if (!name) {
1625                         pr_warn("failed to get section(%d) name from %s\n",
1626                                 idx, obj->path);
1627                         return -LIBBPF_ERRNO__FORMAT;
1628                 }
1629
1630                 data = elf_getdata(scn, 0);
1631                 if (!data) {
1632                         pr_warn("failed to get section(%d) data from %s(%s)\n",
1633                                 idx, name, obj->path);
1634                         return -LIBBPF_ERRNO__FORMAT;
1635                 }
1636                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1637                          idx, name, (unsigned long)data->d_size,
1638                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1639                          (int)sh.sh_type);
1640
1641                 if (strcmp(name, "license") == 0) {
1642                         err = bpf_object__init_license(obj,
1643                                                        data->d_buf,
1644                                                        data->d_size);
1645                         if (err)
1646                                 return err;
1647                 } else if (strcmp(name, "version") == 0) {
1648                         err = bpf_object__init_kversion(obj,
1649                                                         data->d_buf,
1650                                                         data->d_size);
1651                         if (err)
1652                                 return err;
1653                 } else if (strcmp(name, "maps") == 0) {
1654                         obj->efile.maps_shndx = idx;
1655                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1656                         obj->efile.btf_maps_shndx = idx;
1657                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1658                         btf_data = data;
1659                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1660                         btf_ext_data = data;
1661                 } else if (sh.sh_type == SHT_SYMTAB) {
1662                         if (obj->efile.symbols) {
1663                                 pr_warn("bpf: multiple SYMTAB in %s\n",
1664                                         obj->path);
1665                                 return -LIBBPF_ERRNO__FORMAT;
1666                         }
1667                         obj->efile.symbols = data;
1668                         obj->efile.strtabidx = sh.sh_link;
1669                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1670                         if (sh.sh_flags & SHF_EXECINSTR) {
1671                                 if (strcmp(name, ".text") == 0)
1672                                         obj->efile.text_shndx = idx;
1673                                 err = bpf_object__add_program(obj, data->d_buf,
1674                                                               data->d_size,
1675                                                               name, idx);
1676                                 if (err) {
1677                                         char errmsg[STRERR_BUFSIZE];
1678                                         char *cp;
1679
1680                                         cp = libbpf_strerror_r(-err, errmsg,
1681                                                                sizeof(errmsg));
1682                                         pr_warn("failed to alloc program %s (%s): %s",
1683                                                 name, obj->path, cp);
1684                                         return err;
1685                                 }
1686                         } else if (strcmp(name, ".data") == 0) {
1687                                 obj->efile.data = data;
1688                                 obj->efile.data_shndx = idx;
1689                         } else if (strcmp(name, ".rodata") == 0) {
1690                                 obj->efile.rodata = data;
1691                                 obj->efile.rodata_shndx = idx;
1692                         } else {
1693                                 pr_debug("skip section(%d) %s\n", idx, name);
1694                         }
1695                 } else if (sh.sh_type == SHT_REL) {
1696                         int nr_sects = obj->efile.nr_reloc_sects;
1697                         void *sects = obj->efile.reloc_sects;
1698                         int sec = sh.sh_info; /* points to other section */
1699
1700                         /* Only do relo for section with exec instructions */
1701                         if (!section_have_execinstr(obj, sec)) {
1702                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1703                                          name, idx, sec);
1704                                 continue;
1705                         }
1706
1707                         sects = reallocarray(sects, nr_sects + 1,
1708                                              sizeof(*obj->efile.reloc_sects));
1709                         if (!sects) {
1710                                 pr_warn("reloc_sects realloc failed\n");
1711                                 return -ENOMEM;
1712                         }
1713
1714                         obj->efile.reloc_sects = sects;
1715                         obj->efile.nr_reloc_sects++;
1716
1717                         obj->efile.reloc_sects[nr_sects].shdr = sh;
1718                         obj->efile.reloc_sects[nr_sects].data = data;
1719                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1720                         obj->efile.bss = data;
1721                         obj->efile.bss_shndx = idx;
1722                 } else {
1723                         pr_debug("skip section(%d) %s\n", idx, name);
1724                 }
1725         }
1726
1727         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
1728                 pr_warn("Corrupted ELF file: index of strtab invalid\n");
1729                 return -LIBBPF_ERRNO__FORMAT;
1730         }
1731         err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1732         if (!err)
1733                 err = bpf_object__init_maps(obj, relaxed_maps, pin_root_path);
1734         if (!err)
1735                 err = bpf_object__sanitize_and_load_btf(obj);
1736         if (!err)
1737                 err = bpf_object__init_prog_names(obj);
1738         return err;
1739 }
1740
1741 static struct bpf_program *
1742 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1743 {
1744         struct bpf_program *prog;
1745         size_t i;
1746
1747         for (i = 0; i < obj->nr_programs; i++) {
1748                 prog = &obj->programs[i];
1749                 if (prog->idx == idx)
1750                         return prog;
1751         }
1752         return NULL;
1753 }
1754
1755 struct bpf_program *
1756 bpf_object__find_program_by_title(const struct bpf_object *obj,
1757                                   const char *title)
1758 {
1759         struct bpf_program *pos;
1760
1761         bpf_object__for_each_program(pos, obj) {
1762                 if (pos->section_name && !strcmp(pos->section_name, title))
1763                         return pos;
1764         }
1765         return NULL;
1766 }
1767
1768 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1769                                       int shndx)
1770 {
1771         return shndx == obj->efile.data_shndx ||
1772                shndx == obj->efile.bss_shndx ||
1773                shndx == obj->efile.rodata_shndx;
1774 }
1775
1776 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1777                                       int shndx)
1778 {
1779         return shndx == obj->efile.maps_shndx ||
1780                shndx == obj->efile.btf_maps_shndx;
1781 }
1782
1783 static enum libbpf_map_type
1784 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1785 {
1786         if (shndx == obj->efile.data_shndx)
1787                 return LIBBPF_MAP_DATA;
1788         else if (shndx == obj->efile.bss_shndx)
1789                 return LIBBPF_MAP_BSS;
1790         else if (shndx == obj->efile.rodata_shndx)
1791                 return LIBBPF_MAP_RODATA;
1792         else
1793                 return LIBBPF_MAP_UNSPEC;
1794 }
1795
1796 static int bpf_program__record_reloc(struct bpf_program *prog,
1797                                      struct reloc_desc *reloc_desc,
1798                                      __u32 insn_idx, const char *name,
1799                                      const GElf_Sym *sym, const GElf_Rel *rel)
1800 {
1801         struct bpf_insn *insn = &prog->insns[insn_idx];
1802         size_t map_idx, nr_maps = prog->obj->nr_maps;
1803         struct bpf_object *obj = prog->obj;
1804         __u32 shdr_idx = sym->st_shndx;
1805         enum libbpf_map_type type;
1806         struct bpf_map *map;
1807
1808         /* sub-program call relocation */
1809         if (insn->code == (BPF_JMP | BPF_CALL)) {
1810                 if (insn->src_reg != BPF_PSEUDO_CALL) {
1811                         pr_warn("incorrect bpf_call opcode\n");
1812                         return -LIBBPF_ERRNO__RELOC;
1813                 }
1814                 /* text_shndx can be 0, if no default "main" program exists */
1815                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
1816                         pr_warn("bad call relo against section %u\n", shdr_idx);
1817                         return -LIBBPF_ERRNO__RELOC;
1818                 }
1819                 if (sym->st_value % 8) {
1820                         pr_warn("bad call relo offset: %llu\n", (__u64)sym->st_value);
1821                         return -LIBBPF_ERRNO__RELOC;
1822                 }
1823                 reloc_desc->type = RELO_CALL;
1824                 reloc_desc->insn_idx = insn_idx;
1825                 reloc_desc->sym_off = sym->st_value;
1826                 obj->has_pseudo_calls = true;
1827                 return 0;
1828         }
1829
1830         if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
1831                 pr_warn("invalid relo for insns[%d].code 0x%x\n",
1832                         insn_idx, insn->code);
1833                 return -LIBBPF_ERRNO__RELOC;
1834         }
1835         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
1836                 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
1837                         name, shdr_idx);
1838                 return -LIBBPF_ERRNO__RELOC;
1839         }
1840
1841         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1842
1843         /* generic map reference relocation */
1844         if (type == LIBBPF_MAP_UNSPEC) {
1845                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
1846                         pr_warn("bad map relo against section %u\n",
1847                                 shdr_idx);
1848                         return -LIBBPF_ERRNO__RELOC;
1849                 }
1850                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1851                         map = &obj->maps[map_idx];
1852                         if (map->libbpf_type != type ||
1853                             map->sec_idx != sym->st_shndx ||
1854                             map->sec_offset != sym->st_value)
1855                                 continue;
1856                         pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
1857                                  map_idx, map->name, map->sec_idx,
1858                                  map->sec_offset, insn_idx);
1859                         break;
1860                 }
1861                 if (map_idx >= nr_maps) {
1862                         pr_warn("map relo failed to find map for sec %u, off %llu\n",
1863                                 shdr_idx, (__u64)sym->st_value);
1864                         return -LIBBPF_ERRNO__RELOC;
1865                 }
1866                 reloc_desc->type = RELO_LD64;
1867                 reloc_desc->insn_idx = insn_idx;
1868                 reloc_desc->map_idx = map_idx;
1869                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
1870                 return 0;
1871         }
1872
1873         /* global data map relocation */
1874         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
1875                 pr_warn("bad data relo against section %u\n", shdr_idx);
1876                 return -LIBBPF_ERRNO__RELOC;
1877         }
1878         if (!obj->caps.global_data) {
1879                 pr_warn("relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1880                         name, insn_idx);
1881                 return -LIBBPF_ERRNO__RELOC;
1882         }
1883         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1884                 map = &obj->maps[map_idx];
1885                 if (map->libbpf_type != type)
1886                         continue;
1887                 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
1888                          map_idx, map->name, map->sec_idx, map->sec_offset,
1889                          insn_idx);
1890                 break;
1891         }
1892         if (map_idx >= nr_maps) {
1893                 pr_warn("data relo failed to find map for sec %u\n",
1894                         shdr_idx);
1895                 return -LIBBPF_ERRNO__RELOC;
1896         }
1897
1898         reloc_desc->type = RELO_DATA;
1899         reloc_desc->insn_idx = insn_idx;
1900         reloc_desc->map_idx = map_idx;
1901         reloc_desc->sym_off = sym->st_value;
1902         return 0;
1903 }
1904
1905 static int
1906 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1907                            Elf_Data *data, struct bpf_object *obj)
1908 {
1909         Elf_Data *symbols = obj->efile.symbols;
1910         int err, i, nrels;
1911
1912         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1913         nrels = shdr->sh_size / shdr->sh_entsize;
1914
1915         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1916         if (!prog->reloc_desc) {
1917                 pr_warn("failed to alloc memory in relocation\n");
1918                 return -ENOMEM;
1919         }
1920         prog->nr_reloc = nrels;
1921
1922         for (i = 0; i < nrels; i++) {
1923                 const char *name;
1924                 __u32 insn_idx;
1925                 GElf_Sym sym;
1926                 GElf_Rel rel;
1927
1928                 if (!gelf_getrel(data, i, &rel)) {
1929                         pr_warn("relocation: failed to get %d reloc\n", i);
1930                         return -LIBBPF_ERRNO__FORMAT;
1931                 }
1932                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1933                         pr_warn("relocation: symbol %"PRIx64" not found\n",
1934                                 GELF_R_SYM(rel.r_info));
1935                         return -LIBBPF_ERRNO__FORMAT;
1936                 }
1937                 if (rel.r_offset % sizeof(struct bpf_insn))
1938                         return -LIBBPF_ERRNO__FORMAT;
1939
1940                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1941                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1942                                   sym.st_name) ? : "<?>";
1943
1944                 pr_debug("relo for shdr %u, symb %llu, value %llu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
1945                          (__u32)sym.st_shndx, (__u64)GELF_R_SYM(rel.r_info),
1946                          (__u64)sym.st_value, GELF_ST_TYPE(sym.st_info),
1947                          GELF_ST_BIND(sym.st_info), sym.st_name, name,
1948                          insn_idx);
1949
1950                 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
1951                                                 insn_idx, name, &sym, &rel);
1952                 if (err)
1953                         return err;
1954         }
1955         return 0;
1956 }
1957
1958 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1959 {
1960         struct bpf_map_def *def = &map->def;
1961         __u32 key_type_id = 0, value_type_id = 0;
1962         int ret;
1963
1964         /* if it's BTF-defined map, we don't need to search for type IDs */
1965         if (map->sec_idx == obj->efile.btf_maps_shndx)
1966                 return 0;
1967
1968         if (!bpf_map__is_internal(map)) {
1969                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1970                                            def->value_size, &key_type_id,
1971                                            &value_type_id);
1972         } else {
1973                 /*
1974                  * LLVM annotates global data differently in BTF, that is,
1975                  * only as '.data', '.bss' or '.rodata'.
1976                  */
1977                 ret = btf__find_by_name(obj->btf,
1978                                 libbpf_type_to_btf_name[map->libbpf_type]);
1979         }
1980         if (ret < 0)
1981                 return ret;
1982
1983         map->btf_key_type_id = key_type_id;
1984         map->btf_value_type_id = bpf_map__is_internal(map) ?
1985                                  ret : value_type_id;
1986         return 0;
1987 }
1988
1989 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1990 {
1991         struct bpf_map_info info = {};
1992         __u32 len = sizeof(info);
1993         int new_fd, err;
1994         char *new_name;
1995
1996         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1997         if (err)
1998                 return err;
1999
2000         new_name = strdup(info.name);
2001         if (!new_name)
2002                 return -errno;
2003
2004         new_fd = open("/", O_RDONLY | O_CLOEXEC);
2005         if (new_fd < 0) {
2006                 err = -errno;
2007                 goto err_free_new_name;
2008         }
2009
2010         new_fd = dup3(fd, new_fd, O_CLOEXEC);
2011         if (new_fd < 0) {
2012                 err = -errno;
2013                 goto err_close_new_fd;
2014         }
2015
2016         err = zclose(map->fd);
2017         if (err) {
2018                 err = -errno;
2019                 goto err_close_new_fd;
2020         }
2021         free(map->name);
2022
2023         map->fd = new_fd;
2024         map->name = new_name;
2025         map->def.type = info.type;
2026         map->def.key_size = info.key_size;
2027         map->def.value_size = info.value_size;
2028         map->def.max_entries = info.max_entries;
2029         map->def.map_flags = info.map_flags;
2030         map->btf_key_type_id = info.btf_key_type_id;
2031         map->btf_value_type_id = info.btf_value_type_id;
2032         map->reused = true;
2033
2034         return 0;
2035
2036 err_close_new_fd:
2037         close(new_fd);
2038 err_free_new_name:
2039         free(new_name);
2040         return err;
2041 }
2042
2043 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
2044 {
2045         if (!map || !max_entries)
2046                 return -EINVAL;
2047
2048         /* If map already created, its attributes can't be changed. */
2049         if (map->fd >= 0)
2050                 return -EBUSY;
2051
2052         map->def.max_entries = max_entries;
2053
2054         return 0;
2055 }
2056
2057 static int
2058 bpf_object__probe_name(struct bpf_object *obj)
2059 {
2060         struct bpf_load_program_attr attr;
2061         char *cp, errmsg[STRERR_BUFSIZE];
2062         struct bpf_insn insns[] = {
2063                 BPF_MOV64_IMM(BPF_REG_0, 0),
2064                 BPF_EXIT_INSN(),
2065         };
2066         int ret;
2067
2068         /* make sure basic loading works */
2069
2070         memset(&attr, 0, sizeof(attr));
2071         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2072         attr.insns = insns;
2073         attr.insns_cnt = ARRAY_SIZE(insns);
2074         attr.license = "GPL";
2075
2076         ret = bpf_load_program_xattr(&attr, NULL, 0);
2077         if (ret < 0) {
2078                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2079                 pr_warn("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
2080                         __func__, cp, errno);
2081                 return -errno;
2082         }
2083         close(ret);
2084
2085         /* now try the same program, but with the name */
2086
2087         attr.name = "test";
2088         ret = bpf_load_program_xattr(&attr, NULL, 0);
2089         if (ret >= 0) {
2090                 obj->caps.name = 1;
2091                 close(ret);
2092         }
2093
2094         return 0;
2095 }
2096
2097 static int
2098 bpf_object__probe_global_data(struct bpf_object *obj)
2099 {
2100         struct bpf_load_program_attr prg_attr;
2101         struct bpf_create_map_attr map_attr;
2102         char *cp, errmsg[STRERR_BUFSIZE];
2103         struct bpf_insn insns[] = {
2104                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
2105                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
2106                 BPF_MOV64_IMM(BPF_REG_0, 0),
2107                 BPF_EXIT_INSN(),
2108         };
2109         int ret, map;
2110
2111         memset(&map_attr, 0, sizeof(map_attr));
2112         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2113         map_attr.key_size = sizeof(int);
2114         map_attr.value_size = 32;
2115         map_attr.max_entries = 1;
2116
2117         map = bpf_create_map_xattr(&map_attr);
2118         if (map < 0) {
2119                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2120                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
2121                         __func__, cp, errno);
2122                 return -errno;
2123         }
2124
2125         insns[0].imm = map;
2126
2127         memset(&prg_attr, 0, sizeof(prg_attr));
2128         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2129         prg_attr.insns = insns;
2130         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2131         prg_attr.license = "GPL";
2132
2133         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2134         if (ret >= 0) {
2135                 obj->caps.global_data = 1;
2136                 close(ret);
2137         }
2138
2139         close(map);
2140         return 0;
2141 }
2142
2143 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2144 {
2145         static const char strs[] = "\0int\0x\0a";
2146         /* void x(int a) {} */
2147         __u32 types[] = {
2148                 /* int */
2149                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2150                 /* FUNC_PROTO */                                /* [2] */
2151                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2152                 BTF_PARAM_ENC(7, 1),
2153                 /* FUNC x */                                    /* [3] */
2154                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2155         };
2156         int btf_fd;
2157
2158         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2159                                       strs, sizeof(strs));
2160         if (btf_fd >= 0) {
2161                 obj->caps.btf_func = 1;
2162                 close(btf_fd);
2163                 return 1;
2164         }
2165
2166         return 0;
2167 }
2168
2169 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2170 {
2171         static const char strs[] = "\0x\0.data";
2172         /* static int a; */
2173         __u32 types[] = {
2174                 /* int */
2175                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2176                 /* VAR x */                                     /* [2] */
2177                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2178                 BTF_VAR_STATIC,
2179                 /* DATASEC val */                               /* [3] */
2180                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2181                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2182         };
2183         int btf_fd;
2184
2185         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2186                                       strs, sizeof(strs));
2187         if (btf_fd >= 0) {
2188                 obj->caps.btf_datasec = 1;
2189                 close(btf_fd);
2190                 return 1;
2191         }
2192
2193         return 0;
2194 }
2195
2196 static int bpf_object__probe_array_mmap(struct bpf_object *obj)
2197 {
2198         struct bpf_create_map_attr attr = {
2199                 .map_type = BPF_MAP_TYPE_ARRAY,
2200                 .map_flags = BPF_F_MMAPABLE,
2201                 .key_size = sizeof(int),
2202                 .value_size = sizeof(int),
2203                 .max_entries = 1,
2204         };
2205         int fd;
2206
2207         fd = bpf_create_map_xattr(&attr);
2208         if (fd >= 0) {
2209                 obj->caps.array_mmap = 1;
2210                 close(fd);
2211                 return 1;
2212         }
2213
2214         return 0;
2215 }
2216
2217 static int
2218 bpf_object__probe_caps(struct bpf_object *obj)
2219 {
2220         int (*probe_fn[])(struct bpf_object *obj) = {
2221                 bpf_object__probe_name,
2222                 bpf_object__probe_global_data,
2223                 bpf_object__probe_btf_func,
2224                 bpf_object__probe_btf_datasec,
2225                 bpf_object__probe_array_mmap,
2226         };
2227         int i, ret;
2228
2229         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2230                 ret = probe_fn[i](obj);
2231                 if (ret < 0)
2232                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2233         }
2234
2235         return 0;
2236 }
2237
2238 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
2239 {
2240         struct bpf_map_info map_info = {};
2241         char msg[STRERR_BUFSIZE];
2242         __u32 map_info_len;
2243
2244         map_info_len = sizeof(map_info);
2245
2246         if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
2247                 pr_warn("failed to get map info for map FD %d: %s\n",
2248                         map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
2249                 return false;
2250         }
2251
2252         return (map_info.type == map->def.type &&
2253                 map_info.key_size == map->def.key_size &&
2254                 map_info.value_size == map->def.value_size &&
2255                 map_info.max_entries == map->def.max_entries &&
2256                 map_info.map_flags == map->def.map_flags);
2257 }
2258
2259 static int
2260 bpf_object__reuse_map(struct bpf_map *map)
2261 {
2262         char *cp, errmsg[STRERR_BUFSIZE];
2263         int err, pin_fd;
2264
2265         pin_fd = bpf_obj_get(map->pin_path);
2266         if (pin_fd < 0) {
2267                 err = -errno;
2268                 if (err == -ENOENT) {
2269                         pr_debug("found no pinned map to reuse at '%s'\n",
2270                                  map->pin_path);
2271                         return 0;
2272                 }
2273
2274                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
2275                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
2276                         map->pin_path, cp);
2277                 return err;
2278         }
2279
2280         if (!map_is_reuse_compat(map, pin_fd)) {
2281                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
2282                         map->pin_path);
2283                 close(pin_fd);
2284                 return -EINVAL;
2285         }
2286
2287         err = bpf_map__reuse_fd(map, pin_fd);
2288         if (err) {
2289                 close(pin_fd);
2290                 return err;
2291         }
2292         map->pinned = true;
2293         pr_debug("reused pinned map at '%s'\n", map->pin_path);
2294
2295         return 0;
2296 }
2297
2298 static int
2299 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2300 {
2301         char *cp, errmsg[STRERR_BUFSIZE];
2302         int err, zero = 0;
2303         __u8 *data;
2304
2305         /* Nothing to do here since kernel already zero-initializes .bss map. */
2306         if (map->libbpf_type == LIBBPF_MAP_BSS)
2307                 return 0;
2308
2309         data = map->libbpf_type == LIBBPF_MAP_DATA ?
2310                obj->sections.data : obj->sections.rodata;
2311
2312         err = bpf_map_update_elem(map->fd, &zero, data, 0);
2313         /* Freeze .rodata map as read-only from syscall side. */
2314         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2315                 err = bpf_map_freeze(map->fd);
2316                 if (err) {
2317                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2318                         pr_warn("Error freezing map(%s) as read-only: %s\n",
2319                                 map->name, cp);
2320                         err = 0;
2321                 }
2322         }
2323         return err;
2324 }
2325
2326 static int
2327 bpf_object__create_maps(struct bpf_object *obj)
2328 {
2329         struct bpf_create_map_attr create_attr = {};
2330         int nr_cpus = 0;
2331         unsigned int i;
2332         int err;
2333
2334         for (i = 0; i < obj->nr_maps; i++) {
2335                 struct bpf_map *map = &obj->maps[i];
2336                 struct bpf_map_def *def = &map->def;
2337                 char *cp, errmsg[STRERR_BUFSIZE];
2338                 int *pfd = &map->fd;
2339
2340                 if (map->pin_path) {
2341                         err = bpf_object__reuse_map(map);
2342                         if (err) {
2343                                 pr_warn("error reusing pinned map %s\n",
2344                                         map->name);
2345                                 return err;
2346                         }
2347                 }
2348
2349                 if (map->fd >= 0) {
2350                         pr_debug("skip map create (preset) %s: fd=%d\n",
2351                                  map->name, map->fd);
2352                         continue;
2353                 }
2354
2355                 if (obj->caps.name)
2356                         create_attr.name = map->name;
2357                 create_attr.map_ifindex = map->map_ifindex;
2358                 create_attr.map_type = def->type;
2359                 create_attr.map_flags = def->map_flags;
2360                 create_attr.key_size = def->key_size;
2361                 create_attr.value_size = def->value_size;
2362                 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2363                     !def->max_entries) {
2364                         if (!nr_cpus)
2365                                 nr_cpus = libbpf_num_possible_cpus();
2366                         if (nr_cpus < 0) {
2367                                 pr_warn("failed to determine number of system CPUs: %d\n",
2368                                         nr_cpus);
2369                                 err = nr_cpus;
2370                                 goto err_out;
2371                         }
2372                         pr_debug("map '%s': setting size to %d\n",
2373                                  map->name, nr_cpus);
2374                         create_attr.max_entries = nr_cpus;
2375                 } else {
2376                         create_attr.max_entries = def->max_entries;
2377                 }
2378                 create_attr.btf_fd = 0;
2379                 create_attr.btf_key_type_id = 0;
2380                 create_attr.btf_value_type_id = 0;
2381                 if (bpf_map_type__is_map_in_map(def->type) &&
2382                     map->inner_map_fd >= 0)
2383                         create_attr.inner_map_fd = map->inner_map_fd;
2384
2385                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2386                         create_attr.btf_fd = btf__fd(obj->btf);
2387                         create_attr.btf_key_type_id = map->btf_key_type_id;
2388                         create_attr.btf_value_type_id = map->btf_value_type_id;
2389                 }
2390
2391                 *pfd = bpf_create_map_xattr(&create_attr);
2392                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2393                                  create_attr.btf_value_type_id)) {
2394                         err = -errno;
2395                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2396                         pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2397                                 map->name, cp, err);
2398                         create_attr.btf_fd = 0;
2399                         create_attr.btf_key_type_id = 0;
2400                         create_attr.btf_value_type_id = 0;
2401                         map->btf_key_type_id = 0;
2402                         map->btf_value_type_id = 0;
2403                         *pfd = bpf_create_map_xattr(&create_attr);
2404                 }
2405
2406                 if (*pfd < 0) {
2407                         size_t j;
2408
2409                         err = -errno;
2410 err_out:
2411                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2412                         pr_warn("failed to create map (name: '%s'): %s(%d)\n",
2413                                 map->name, cp, err);
2414                         for (j = 0; j < i; j++)
2415                                 zclose(obj->maps[j].fd);
2416                         return err;
2417                 }
2418
2419                 if (bpf_map__is_internal(map)) {
2420                         err = bpf_object__populate_internal_map(obj, map);
2421                         if (err < 0) {
2422                                 zclose(*pfd);
2423                                 goto err_out;
2424                         }
2425                 }
2426
2427                 if (map->pin_path && !map->pinned) {
2428                         err = bpf_map__pin(map, NULL);
2429                         if (err) {
2430                                 pr_warn("failed to auto-pin map name '%s' at '%s'\n",
2431                                         map->name, map->pin_path);
2432                                 return err;
2433                         }
2434                 }
2435
2436                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2437         }
2438
2439         return 0;
2440 }
2441
2442 static int
2443 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2444                         void *btf_prog_info, const char *info_name)
2445 {
2446         if (err != -ENOENT) {
2447                 pr_warn("Error in loading %s for sec %s.\n",
2448                         info_name, prog->section_name);
2449                 return err;
2450         }
2451
2452         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2453
2454         if (btf_prog_info) {
2455                 /*
2456                  * Some info has already been found but has problem
2457                  * in the last btf_ext reloc. Must have to error out.
2458                  */
2459                 pr_warn("Error in relocating %s for sec %s.\n",
2460                         info_name, prog->section_name);
2461                 return err;
2462         }
2463
2464         /* Have problem loading the very first info. Ignore the rest. */
2465         pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
2466                 info_name, prog->section_name, info_name);
2467         return 0;
2468 }
2469
2470 static int
2471 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2472                           const char *section_name,  __u32 insn_offset)
2473 {
2474         int err;
2475
2476         if (!insn_offset || prog->func_info) {
2477                 /*
2478                  * !insn_offset => main program
2479                  *
2480                  * For sub prog, the main program's func_info has to
2481                  * be loaded first (i.e. prog->func_info != NULL)
2482                  */
2483                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2484                                                section_name, insn_offset,
2485                                                &prog->func_info,
2486                                                &prog->func_info_cnt);
2487                 if (err)
2488                         return check_btf_ext_reloc_err(prog, err,
2489                                                        prog->func_info,
2490                                                        "bpf_func_info");
2491
2492                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2493         }
2494
2495         if (!insn_offset || prog->line_info) {
2496                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2497                                                section_name, insn_offset,
2498                                                &prog->line_info,
2499                                                &prog->line_info_cnt);
2500                 if (err)
2501                         return check_btf_ext_reloc_err(prog, err,
2502                                                        prog->line_info,
2503                                                        "bpf_line_info");
2504
2505                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2506         }
2507
2508         return 0;
2509 }
2510
2511 #define BPF_CORE_SPEC_MAX_LEN 64
2512
2513 /* represents BPF CO-RE field or array element accessor */
2514 struct bpf_core_accessor {
2515         __u32 type_id;          /* struct/union type or array element type */
2516         __u32 idx;              /* field index or array index */
2517         const char *name;       /* field name or NULL for array accessor */
2518 };
2519
2520 struct bpf_core_spec {
2521         const struct btf *btf;
2522         /* high-level spec: named fields and array indices only */
2523         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2524         /* high-level spec length */
2525         int len;
2526         /* raw, low-level spec: 1-to-1 with accessor spec string */
2527         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2528         /* raw spec length */
2529         int raw_len;
2530         /* field bit offset represented by spec */
2531         __u32 bit_offset;
2532 };
2533
2534 static bool str_is_empty(const char *s)
2535 {
2536         return !s || !s[0];
2537 }
2538
2539 /*
2540  * Turn bpf_field_reloc into a low- and high-level spec representation,
2541  * validating correctness along the way, as well as calculating resulting
2542  * field bit offset, specified by accessor string. Low-level spec captures
2543  * every single level of nestedness, including traversing anonymous
2544  * struct/union members. High-level one only captures semantically meaningful
2545  * "turning points": named fields and array indicies.
2546  * E.g., for this case:
2547  *
2548  *   struct sample {
2549  *       int __unimportant;
2550  *       struct {
2551  *           int __1;
2552  *           int __2;
2553  *           int a[7];
2554  *       };
2555  *   };
2556  *
2557  *   struct sample *s = ...;
2558  *
2559  *   int x = &s->a[3]; // access string = '0:1:2:3'
2560  *
2561  * Low-level spec has 1:1 mapping with each element of access string (it's
2562  * just a parsed access string representation): [0, 1, 2, 3].
2563  *
2564  * High-level spec will capture only 3 points:
2565  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2566  *   - field 'a' access (corresponds to '2' in low-level spec);
2567  *   - array element #3 access (corresponds to '3' in low-level spec).
2568  *
2569  */
2570 static int bpf_core_spec_parse(const struct btf *btf,
2571                                __u32 type_id,
2572                                const char *spec_str,
2573                                struct bpf_core_spec *spec)
2574 {
2575         int access_idx, parsed_len, i;
2576         const struct btf_type *t;
2577         const char *name;
2578         __u32 id;
2579         __s64 sz;
2580
2581         if (str_is_empty(spec_str) || *spec_str == ':')
2582                 return -EINVAL;
2583
2584         memset(spec, 0, sizeof(*spec));
2585         spec->btf = btf;
2586
2587         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2588         while (*spec_str) {
2589                 if (*spec_str == ':')
2590                         ++spec_str;
2591                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2592                         return -EINVAL;
2593                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2594                         return -E2BIG;
2595                 spec_str += parsed_len;
2596                 spec->raw_spec[spec->raw_len++] = access_idx;
2597         }
2598
2599         if (spec->raw_len == 0)
2600                 return -EINVAL;
2601
2602         /* first spec value is always reloc type array index */
2603         t = skip_mods_and_typedefs(btf, type_id, &id);
2604         if (!t)
2605                 return -EINVAL;
2606
2607         access_idx = spec->raw_spec[0];
2608         spec->spec[0].type_id = id;
2609         spec->spec[0].idx = access_idx;
2610         spec->len++;
2611
2612         sz = btf__resolve_size(btf, id);
2613         if (sz < 0)
2614                 return sz;
2615         spec->bit_offset = access_idx * sz * 8;
2616
2617         for (i = 1; i < spec->raw_len; i++) {
2618                 t = skip_mods_and_typedefs(btf, id, &id);
2619                 if (!t)
2620                         return -EINVAL;
2621
2622                 access_idx = spec->raw_spec[i];
2623
2624                 if (btf_is_composite(t)) {
2625                         const struct btf_member *m;
2626                         __u32 bit_offset;
2627
2628                         if (access_idx >= btf_vlen(t))
2629                                 return -EINVAL;
2630
2631                         bit_offset = btf_member_bit_offset(t, access_idx);
2632                         spec->bit_offset += bit_offset;
2633
2634                         m = btf_members(t) + access_idx;
2635                         if (m->name_off) {
2636                                 name = btf__name_by_offset(btf, m->name_off);
2637                                 if (str_is_empty(name))
2638                                         return -EINVAL;
2639
2640                                 spec->spec[spec->len].type_id = id;
2641                                 spec->spec[spec->len].idx = access_idx;
2642                                 spec->spec[spec->len].name = name;
2643                                 spec->len++;
2644                         }
2645
2646                         id = m->type;
2647                 } else if (btf_is_array(t)) {
2648                         const struct btf_array *a = btf_array(t);
2649
2650                         t = skip_mods_and_typedefs(btf, a->type, &id);
2651                         if (!t || access_idx >= a->nelems)
2652                                 return -EINVAL;
2653
2654                         spec->spec[spec->len].type_id = id;
2655                         spec->spec[spec->len].idx = access_idx;
2656                         spec->len++;
2657
2658                         sz = btf__resolve_size(btf, id);
2659                         if (sz < 0)
2660                                 return sz;
2661                         spec->bit_offset += access_idx * sz * 8;
2662                 } else {
2663                         pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2664                                 type_id, spec_str, i, id, btf_kind(t));
2665                         return -EINVAL;
2666                 }
2667         }
2668
2669         return 0;
2670 }
2671
2672 static bool bpf_core_is_flavor_sep(const char *s)
2673 {
2674         /* check X___Y name pattern, where X and Y are not underscores */
2675         return s[0] != '_' &&                                 /* X */
2676                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
2677                s[4] != '_';                                   /* Y */
2678 }
2679
2680 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2681  * before last triple underscore. Struct name part after last triple
2682  * underscore is ignored by BPF CO-RE relocation during relocation matching.
2683  */
2684 static size_t bpf_core_essential_name_len(const char *name)
2685 {
2686         size_t n = strlen(name);
2687         int i;
2688
2689         for (i = n - 5; i >= 0; i--) {
2690                 if (bpf_core_is_flavor_sep(name + i))
2691                         return i + 1;
2692         }
2693         return n;
2694 }
2695
2696 /* dynamically sized list of type IDs */
2697 struct ids_vec {
2698         __u32 *data;
2699         int len;
2700 };
2701
2702 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2703 {
2704         free(cand_ids->data);
2705         free(cand_ids);
2706 }
2707
2708 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2709                                            __u32 local_type_id,
2710                                            const struct btf *targ_btf)
2711 {
2712         size_t local_essent_len, targ_essent_len;
2713         const char *local_name, *targ_name;
2714         const struct btf_type *t;
2715         struct ids_vec *cand_ids;
2716         __u32 *new_ids;
2717         int i, err, n;
2718
2719         t = btf__type_by_id(local_btf, local_type_id);
2720         if (!t)
2721                 return ERR_PTR(-EINVAL);
2722
2723         local_name = btf__name_by_offset(local_btf, t->name_off);
2724         if (str_is_empty(local_name))
2725                 return ERR_PTR(-EINVAL);
2726         local_essent_len = bpf_core_essential_name_len(local_name);
2727
2728         cand_ids = calloc(1, sizeof(*cand_ids));
2729         if (!cand_ids)
2730                 return ERR_PTR(-ENOMEM);
2731
2732         n = btf__get_nr_types(targ_btf);
2733         for (i = 1; i <= n; i++) {
2734                 t = btf__type_by_id(targ_btf, i);
2735                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2736                 if (str_is_empty(targ_name))
2737                         continue;
2738
2739                 targ_essent_len = bpf_core_essential_name_len(targ_name);
2740                 if (targ_essent_len != local_essent_len)
2741                         continue;
2742
2743                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2744                         pr_debug("[%d] %s: found candidate [%d] %s\n",
2745                                  local_type_id, local_name, i, targ_name);
2746                         new_ids = realloc(cand_ids->data, cand_ids->len + 1);
2747                         if (!new_ids) {
2748                                 err = -ENOMEM;
2749                                 goto err_out;
2750                         }
2751                         cand_ids->data = new_ids;
2752                         cand_ids->data[cand_ids->len++] = i;
2753                 }
2754         }
2755         return cand_ids;
2756 err_out:
2757         bpf_core_free_cands(cand_ids);
2758         return ERR_PTR(err);
2759 }
2760
2761 /* Check two types for compatibility, skipping const/volatile/restrict and
2762  * typedefs, to ensure we are relocating compatible entities:
2763  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
2764  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
2765  *   - any two PTRs are always compatible;
2766  *   - for ENUMs, names should be the same (ignoring flavor suffix) or at
2767  *     least one of enums should be anonymous;
2768  *   - for ENUMs, check sizes, names are ignored;
2769  *   - for INT, size and signedness are ignored;
2770  *   - for ARRAY, dimensionality is ignored, element types are checked for
2771  *     compatibility recursively;
2772  *   - everything else shouldn't be ever a target of relocation.
2773  * These rules are not set in stone and probably will be adjusted as we get
2774  * more experience with using BPF CO-RE relocations.
2775  */
2776 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2777                                       __u32 local_id,
2778                                       const struct btf *targ_btf,
2779                                       __u32 targ_id)
2780 {
2781         const struct btf_type *local_type, *targ_type;
2782
2783 recur:
2784         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2785         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2786         if (!local_type || !targ_type)
2787                 return -EINVAL;
2788
2789         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2790                 return 1;
2791         if (btf_kind(local_type) != btf_kind(targ_type))
2792                 return 0;
2793
2794         switch (btf_kind(local_type)) {
2795         case BTF_KIND_PTR:
2796                 return 1;
2797         case BTF_KIND_FWD:
2798         case BTF_KIND_ENUM: {
2799                 const char *local_name, *targ_name;
2800                 size_t local_len, targ_len;
2801
2802                 local_name = btf__name_by_offset(local_btf,
2803                                                  local_type->name_off);
2804                 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
2805                 local_len = bpf_core_essential_name_len(local_name);
2806                 targ_len = bpf_core_essential_name_len(targ_name);
2807                 /* one of them is anonymous or both w/ same flavor-less names */
2808                 return local_len == 0 || targ_len == 0 ||
2809                        (local_len == targ_len &&
2810                         strncmp(local_name, targ_name, local_len) == 0);
2811         }
2812         case BTF_KIND_INT:
2813                 /* just reject deprecated bitfield-like integers; all other
2814                  * integers are by default compatible between each other
2815                  */
2816                 return btf_int_offset(local_type) == 0 &&
2817                        btf_int_offset(targ_type) == 0;
2818         case BTF_KIND_ARRAY:
2819                 local_id = btf_array(local_type)->type;
2820                 targ_id = btf_array(targ_type)->type;
2821                 goto recur;
2822         default:
2823                 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
2824                         btf_kind(local_type), local_id, targ_id);
2825                 return 0;
2826         }
2827 }
2828
2829 /*
2830  * Given single high-level named field accessor in local type, find
2831  * corresponding high-level accessor for a target type. Along the way,
2832  * maintain low-level spec for target as well. Also keep updating target
2833  * bit offset.
2834  *
2835  * Searching is performed through recursive exhaustive enumeration of all
2836  * fields of a struct/union. If there are any anonymous (embedded)
2837  * structs/unions, they are recursively searched as well. If field with
2838  * desired name is found, check compatibility between local and target types,
2839  * before returning result.
2840  *
2841  * 1 is returned, if field is found.
2842  * 0 is returned if no compatible field is found.
2843  * <0 is returned on error.
2844  */
2845 static int bpf_core_match_member(const struct btf *local_btf,
2846                                  const struct bpf_core_accessor *local_acc,
2847                                  const struct btf *targ_btf,
2848                                  __u32 targ_id,
2849                                  struct bpf_core_spec *spec,
2850                                  __u32 *next_targ_id)
2851 {
2852         const struct btf_type *local_type, *targ_type;
2853         const struct btf_member *local_member, *m;
2854         const char *local_name, *targ_name;
2855         __u32 local_id;
2856         int i, n, found;
2857
2858         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2859         if (!targ_type)
2860                 return -EINVAL;
2861         if (!btf_is_composite(targ_type))
2862                 return 0;
2863
2864         local_id = local_acc->type_id;
2865         local_type = btf__type_by_id(local_btf, local_id);
2866         local_member = btf_members(local_type) + local_acc->idx;
2867         local_name = btf__name_by_offset(local_btf, local_member->name_off);
2868
2869         n = btf_vlen(targ_type);
2870         m = btf_members(targ_type);
2871         for (i = 0; i < n; i++, m++) {
2872                 __u32 bit_offset;
2873
2874                 bit_offset = btf_member_bit_offset(targ_type, i);
2875
2876                 /* too deep struct/union/array nesting */
2877                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2878                         return -E2BIG;
2879
2880                 /* speculate this member will be the good one */
2881                 spec->bit_offset += bit_offset;
2882                 spec->raw_spec[spec->raw_len++] = i;
2883
2884                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2885                 if (str_is_empty(targ_name)) {
2886                         /* embedded struct/union, we need to go deeper */
2887                         found = bpf_core_match_member(local_btf, local_acc,
2888                                                       targ_btf, m->type,
2889                                                       spec, next_targ_id);
2890                         if (found) /* either found or error */
2891                                 return found;
2892                 } else if (strcmp(local_name, targ_name) == 0) {
2893                         /* matching named field */
2894                         struct bpf_core_accessor *targ_acc;
2895
2896                         targ_acc = &spec->spec[spec->len++];
2897                         targ_acc->type_id = targ_id;
2898                         targ_acc->idx = i;
2899                         targ_acc->name = targ_name;
2900
2901                         *next_targ_id = m->type;
2902                         found = bpf_core_fields_are_compat(local_btf,
2903                                                            local_member->type,
2904                                                            targ_btf, m->type);
2905                         if (!found)
2906                                 spec->len--; /* pop accessor */
2907                         return found;
2908                 }
2909                 /* member turned out not to be what we looked for */
2910                 spec->bit_offset -= bit_offset;
2911                 spec->raw_len--;
2912         }
2913
2914         return 0;
2915 }
2916
2917 /*
2918  * Try to match local spec to a target type and, if successful, produce full
2919  * target spec (high-level, low-level + bit offset).
2920  */
2921 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2922                                const struct btf *targ_btf, __u32 targ_id,
2923                                struct bpf_core_spec *targ_spec)
2924 {
2925         const struct btf_type *targ_type;
2926         const struct bpf_core_accessor *local_acc;
2927         struct bpf_core_accessor *targ_acc;
2928         int i, sz, matched;
2929
2930         memset(targ_spec, 0, sizeof(*targ_spec));
2931         targ_spec->btf = targ_btf;
2932
2933         local_acc = &local_spec->spec[0];
2934         targ_acc = &targ_spec->spec[0];
2935
2936         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2937                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2938                                                    &targ_id);
2939                 if (!targ_type)
2940                         return -EINVAL;
2941
2942                 if (local_acc->name) {
2943                         matched = bpf_core_match_member(local_spec->btf,
2944                                                         local_acc,
2945                                                         targ_btf, targ_id,
2946                                                         targ_spec, &targ_id);
2947                         if (matched <= 0)
2948                                 return matched;
2949                 } else {
2950                         /* for i=0, targ_id is already treated as array element
2951                          * type (because it's the original struct), for others
2952                          * we should find array element type first
2953                          */
2954                         if (i > 0) {
2955                                 const struct btf_array *a;
2956
2957                                 if (!btf_is_array(targ_type))
2958                                         return 0;
2959
2960                                 a = btf_array(targ_type);
2961                                 if (local_acc->idx >= a->nelems)
2962                                         return 0;
2963                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
2964                                                             &targ_id))
2965                                         return -EINVAL;
2966                         }
2967
2968                         /* too deep struct/union/array nesting */
2969                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2970                                 return -E2BIG;
2971
2972                         targ_acc->type_id = targ_id;
2973                         targ_acc->idx = local_acc->idx;
2974                         targ_acc->name = NULL;
2975                         targ_spec->len++;
2976                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2977                         targ_spec->raw_len++;
2978
2979                         sz = btf__resolve_size(targ_btf, targ_id);
2980                         if (sz < 0)
2981                                 return sz;
2982                         targ_spec->bit_offset += local_acc->idx * sz * 8;
2983                 }
2984         }
2985
2986         return 1;
2987 }
2988
2989 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
2990                                     const struct bpf_field_reloc *relo,
2991                                     const struct bpf_core_spec *spec,
2992                                     __u32 *val, bool *validate)
2993 {
2994         const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
2995         const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
2996         __u32 byte_off, byte_sz, bit_off, bit_sz;
2997         const struct btf_member *m;
2998         const struct btf_type *mt;
2999         bool bitfield;
3000         __s64 sz;
3001
3002         /* a[n] accessor needs special handling */
3003         if (!acc->name) {
3004                 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
3005                         *val = spec->bit_offset / 8;
3006                 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
3007                         sz = btf__resolve_size(spec->btf, acc->type_id);
3008                         if (sz < 0)
3009                                 return -EINVAL;
3010                         *val = sz;
3011                 } else {
3012                         pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
3013                                 bpf_program__title(prog, false),
3014                                 relo->kind, relo->insn_off / 8);
3015                         return -EINVAL;
3016                 }
3017                 if (validate)
3018                         *validate = true;
3019                 return 0;
3020         }
3021
3022         m = btf_members(t) + acc->idx;
3023         mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
3024         bit_off = spec->bit_offset;
3025         bit_sz = btf_member_bitfield_size(t, acc->idx);
3026
3027         bitfield = bit_sz > 0;
3028         if (bitfield) {
3029                 byte_sz = mt->size;
3030                 byte_off = bit_off / 8 / byte_sz * byte_sz;
3031                 /* figure out smallest int size necessary for bitfield load */
3032                 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
3033                         if (byte_sz >= 8) {
3034                                 /* bitfield can't be read with 64-bit read */
3035                                 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
3036                                         bpf_program__title(prog, false),
3037                                         relo->kind, relo->insn_off / 8);
3038                                 return -E2BIG;
3039                         }
3040                         byte_sz *= 2;
3041                         byte_off = bit_off / 8 / byte_sz * byte_sz;
3042                 }
3043         } else {
3044                 sz = btf__resolve_size(spec->btf, m->type);
3045                 if (sz < 0)
3046                         return -EINVAL;
3047                 byte_sz = sz;
3048                 byte_off = spec->bit_offset / 8;
3049                 bit_sz = byte_sz * 8;
3050         }
3051
3052         /* for bitfields, all the relocatable aspects are ambiguous and we
3053          * might disagree with compiler, so turn off validation of expected
3054          * value, except for signedness
3055          */
3056         if (validate)
3057                 *validate = !bitfield;
3058
3059         switch (relo->kind) {
3060         case BPF_FIELD_BYTE_OFFSET:
3061                 *val = byte_off;
3062                 break;
3063         case BPF_FIELD_BYTE_SIZE:
3064                 *val = byte_sz;
3065                 break;
3066         case BPF_FIELD_SIGNED:
3067                 /* enums will be assumed unsigned */
3068                 *val = btf_is_enum(mt) ||
3069                        (btf_int_encoding(mt) & BTF_INT_SIGNED);
3070                 if (validate)
3071                         *validate = true; /* signedness is never ambiguous */
3072                 break;
3073         case BPF_FIELD_LSHIFT_U64:
3074 #if __BYTE_ORDER == __LITTLE_ENDIAN
3075                 *val = 64 - (bit_off + bit_sz - byte_off  * 8);
3076 #else
3077                 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
3078 #endif
3079                 break;
3080         case BPF_FIELD_RSHIFT_U64:
3081                 *val = 64 - bit_sz;
3082                 if (validate)
3083                         *validate = true; /* right shift is never ambiguous */
3084                 break;
3085         case BPF_FIELD_EXISTS:
3086         default:
3087                 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
3088                         bpf_program__title(prog, false),
3089                         relo->kind, relo->insn_off / 8);
3090                 return -EINVAL;
3091         }
3092
3093         return 0;
3094 }
3095
3096 /*
3097  * Patch relocatable BPF instruction.
3098  *
3099  * Patched value is determined by relocation kind and target specification.
3100  * For field existence relocation target spec will be NULL if field is not
3101  * found.
3102  * Expected insn->imm value is determined using relocation kind and local
3103  * spec, and is checked before patching instruction. If actual insn->imm value
3104  * is wrong, bail out with error.
3105  *
3106  * Currently three kinds of BPF instructions are supported:
3107  * 1. rX = <imm> (assignment with immediate operand);
3108  * 2. rX += <imm> (arithmetic operations with immediate operand);
3109  */
3110 static int bpf_core_reloc_insn(struct bpf_program *prog,
3111                                const struct bpf_field_reloc *relo,
3112                                const struct bpf_core_spec *local_spec,
3113                                const struct bpf_core_spec *targ_spec)
3114 {
3115         bool failed = false, validate = true;
3116         __u32 orig_val, new_val;
3117         struct bpf_insn *insn;
3118         int insn_idx, err;
3119         __u8 class;
3120
3121         if (relo->insn_off % sizeof(struct bpf_insn))
3122                 return -EINVAL;
3123         insn_idx = relo->insn_off / sizeof(struct bpf_insn);
3124
3125         if (relo->kind == BPF_FIELD_EXISTS) {
3126                 orig_val = 1; /* can't generate EXISTS relo w/o local field */
3127                 new_val = targ_spec ? 1 : 0;
3128         } else if (!targ_spec) {
3129                 failed = true;
3130                 new_val = (__u32)-1;
3131         } else {
3132                 err = bpf_core_calc_field_relo(prog, relo, local_spec,
3133                                                &orig_val, &validate);
3134                 if (err)
3135                         return err;
3136                 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
3137                                                &new_val, NULL);
3138                 if (err)
3139                         return err;
3140         }
3141
3142         insn = &prog->insns[insn_idx];
3143         class = BPF_CLASS(insn->code);
3144
3145         if (class == BPF_ALU || class == BPF_ALU64) {
3146                 if (BPF_SRC(insn->code) != BPF_K)
3147                         return -EINVAL;
3148                 if (!failed && validate && insn->imm != orig_val) {
3149                         pr_warn("prog '%s': unexpected insn #%d value: got %u, exp %u -> %u\n",
3150                                 bpf_program__title(prog, false), insn_idx,
3151                                 insn->imm, orig_val, new_val);
3152                         return -EINVAL;
3153                 }
3154                 orig_val = insn->imm;
3155                 insn->imm = new_val;
3156                 pr_debug("prog '%s': patched insn #%d (ALU/ALU64)%s imm %u -> %u\n",
3157                          bpf_program__title(prog, false), insn_idx,
3158                          failed ? " w/ failed reloc" : "", orig_val, new_val);
3159         } else {
3160                 pr_warn("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
3161                         bpf_program__title(prog, false),
3162                         insn_idx, insn->code, insn->src_reg, insn->dst_reg,
3163                         insn->off, insn->imm);
3164                 return -EINVAL;
3165         }
3166
3167         return 0;
3168 }
3169
3170 static struct btf *btf_load_raw(const char *path)
3171 {
3172         struct btf *btf;
3173         size_t read_cnt;
3174         struct stat st;
3175         void *data;
3176         FILE *f;
3177
3178         if (stat(path, &st))
3179                 return ERR_PTR(-errno);
3180
3181         data = malloc(st.st_size);
3182         if (!data)
3183                 return ERR_PTR(-ENOMEM);
3184
3185         f = fopen(path, "rb");
3186         if (!f) {
3187                 btf = ERR_PTR(-errno);
3188                 goto cleanup;
3189         }
3190
3191         read_cnt = fread(data, 1, st.st_size, f);
3192         fclose(f);
3193         if (read_cnt < st.st_size) {
3194                 btf = ERR_PTR(-EBADF);
3195                 goto cleanup;
3196         }
3197
3198         btf = btf__new(data, read_cnt);
3199
3200 cleanup:
3201         free(data);
3202         return btf;
3203 }
3204
3205 /*
3206  * Probe few well-known locations for vmlinux kernel image and try to load BTF
3207  * data out of it to use for target BTF.
3208  */
3209 static struct btf *bpf_core_find_kernel_btf(void)
3210 {
3211         struct {
3212                 const char *path_fmt;
3213                 bool raw_btf;
3214         } locations[] = {
3215                 /* try canonical vmlinux BTF through sysfs first */
3216                 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
3217                 /* fall back to trying to find vmlinux ELF on disk otherwise */
3218                 { "/boot/vmlinux-%1$s" },
3219                 { "/lib/modules/%1$s/vmlinux-%1$s" },
3220                 { "/lib/modules/%1$s/build/vmlinux" },
3221                 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
3222                 { "/usr/lib/debug/boot/vmlinux-%1$s" },
3223                 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
3224                 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
3225         };
3226         char path[PATH_MAX + 1];
3227         struct utsname buf;
3228         struct btf *btf;
3229         int i;
3230
3231         uname(&buf);
3232
3233         for (i = 0; i < ARRAY_SIZE(locations); i++) {
3234                 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
3235
3236                 if (access(path, R_OK))
3237                         continue;
3238
3239                 if (locations[i].raw_btf)
3240                         btf = btf_load_raw(path);
3241                 else
3242                         btf = btf__parse_elf(path, NULL);
3243
3244                 pr_debug("loading kernel BTF '%s': %ld\n",
3245                          path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
3246                 if (IS_ERR(btf))
3247                         continue;
3248
3249                 return btf;
3250         }
3251
3252         pr_warn("failed to find valid kernel BTF\n");
3253         return ERR_PTR(-ESRCH);
3254 }
3255
3256 /* Output spec definition in the format:
3257  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
3258  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
3259  */
3260 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
3261 {
3262         const struct btf_type *t;
3263         const char *s;
3264         __u32 type_id;
3265         int i;
3266
3267         type_id = spec->spec[0].type_id;
3268         t = btf__type_by_id(spec->btf, type_id);
3269         s = btf__name_by_offset(spec->btf, t->name_off);
3270         libbpf_print(level, "[%u] %s + ", type_id, s);
3271
3272         for (i = 0; i < spec->raw_len; i++)
3273                 libbpf_print(level, "%d%s", spec->raw_spec[i],
3274                              i == spec->raw_len - 1 ? " => " : ":");
3275
3276         libbpf_print(level, "%u.%u @ &x",
3277                      spec->bit_offset / 8, spec->bit_offset % 8);
3278
3279         for (i = 0; i < spec->len; i++) {
3280                 if (spec->spec[i].name)
3281                         libbpf_print(level, ".%s", spec->spec[i].name);
3282                 else
3283                         libbpf_print(level, "[%u]", spec->spec[i].idx);
3284         }
3285
3286 }
3287
3288 static size_t bpf_core_hash_fn(const void *key, void *ctx)
3289 {
3290         return (size_t)key;
3291 }
3292
3293 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
3294 {
3295         return k1 == k2;
3296 }
3297
3298 static void *u32_as_hash_key(__u32 x)
3299 {
3300         return (void *)(uintptr_t)x;
3301 }
3302
3303 /*
3304  * CO-RE relocate single instruction.
3305  *
3306  * The outline and important points of the algorithm:
3307  * 1. For given local type, find corresponding candidate target types.
3308  *    Candidate type is a type with the same "essential" name, ignoring
3309  *    everything after last triple underscore (___). E.g., `sample`,
3310  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
3311  *    for each other. Names with triple underscore are referred to as
3312  *    "flavors" and are useful, among other things, to allow to
3313  *    specify/support incompatible variations of the same kernel struct, which
3314  *    might differ between different kernel versions and/or build
3315  *    configurations.
3316  *
3317  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
3318  *    converter, when deduplicated BTF of a kernel still contains more than
3319  *    one different types with the same name. In that case, ___2, ___3, etc
3320  *    are appended starting from second name conflict. But start flavors are
3321  *    also useful to be defined "locally", in BPF program, to extract same
3322  *    data from incompatible changes between different kernel
3323  *    versions/configurations. For instance, to handle field renames between
3324  *    kernel versions, one can use two flavors of the struct name with the
3325  *    same common name and use conditional relocations to extract that field,
3326  *    depending on target kernel version.
3327  * 2. For each candidate type, try to match local specification to this
3328  *    candidate target type. Matching involves finding corresponding
3329  *    high-level spec accessors, meaning that all named fields should match,
3330  *    as well as all array accesses should be within the actual bounds. Also,
3331  *    types should be compatible (see bpf_core_fields_are_compat for details).
3332  * 3. It is supported and expected that there might be multiple flavors
3333  *    matching the spec. As long as all the specs resolve to the same set of
3334  *    offsets across all candidates, there is no error. If there is any
3335  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
3336  *    imprefection of BTF deduplication, which can cause slight duplication of
3337  *    the same BTF type, if some directly or indirectly referenced (by
3338  *    pointer) type gets resolved to different actual types in different
3339  *    object files. If such situation occurs, deduplicated BTF will end up
3340  *    with two (or more) structurally identical types, which differ only in
3341  *    types they refer to through pointer. This should be OK in most cases and
3342  *    is not an error.
3343  * 4. Candidate types search is performed by linearly scanning through all
3344  *    types in target BTF. It is anticipated that this is overall more
3345  *    efficient memory-wise and not significantly worse (if not better)
3346  *    CPU-wise compared to prebuilding a map from all local type names to
3347  *    a list of candidate type names. It's also sped up by caching resolved
3348  *    list of matching candidates per each local "root" type ID, that has at
3349  *    least one bpf_field_reloc associated with it. This list is shared
3350  *    between multiple relocations for the same type ID and is updated as some
3351  *    of the candidates are pruned due to structural incompatibility.
3352  */
3353 static int bpf_core_reloc_field(struct bpf_program *prog,
3354                                  const struct bpf_field_reloc *relo,
3355                                  int relo_idx,
3356                                  const struct btf *local_btf,
3357                                  const struct btf *targ_btf,
3358                                  struct hashmap *cand_cache)
3359 {
3360         const char *prog_name = bpf_program__title(prog, false);
3361         struct bpf_core_spec local_spec, cand_spec, targ_spec;
3362         const void *type_key = u32_as_hash_key(relo->type_id);
3363         const struct btf_type *local_type, *cand_type;
3364         const char *local_name, *cand_name;
3365         struct ids_vec *cand_ids;
3366         __u32 local_id, cand_id;
3367         const char *spec_str;
3368         int i, j, err;
3369
3370         local_id = relo->type_id;
3371         local_type = btf__type_by_id(local_btf, local_id);
3372         if (!local_type)
3373                 return -EINVAL;
3374
3375         local_name = btf__name_by_offset(local_btf, local_type->name_off);
3376         if (str_is_empty(local_name))
3377                 return -EINVAL;
3378
3379         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3380         if (str_is_empty(spec_str))
3381                 return -EINVAL;
3382
3383         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3384         if (err) {
3385                 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3386                         prog_name, relo_idx, local_id, local_name, spec_str,
3387                         err);
3388                 return -EINVAL;
3389         }
3390
3391         pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
3392                  relo->kind);
3393         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3394         libbpf_print(LIBBPF_DEBUG, "\n");
3395
3396         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3397                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3398                 if (IS_ERR(cand_ids)) {
3399                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3400                                 prog_name, relo_idx, local_id, local_name,
3401                                 PTR_ERR(cand_ids));
3402                         return PTR_ERR(cand_ids);
3403                 }
3404                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3405                 if (err) {
3406                         bpf_core_free_cands(cand_ids);
3407                         return err;
3408                 }
3409         }
3410
3411         for (i = 0, j = 0; i < cand_ids->len; i++) {
3412                 cand_id = cand_ids->data[i];
3413                 cand_type = btf__type_by_id(targ_btf, cand_id);
3414                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3415
3416                 err = bpf_core_spec_match(&local_spec, targ_btf,
3417                                           cand_id, &cand_spec);
3418                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3419                          prog_name, relo_idx, i, cand_name);
3420                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3421                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3422                 if (err < 0) {
3423                         pr_warn("prog '%s': relo #%d: matching error: %d\n",
3424                                 prog_name, relo_idx, err);
3425                         return err;
3426                 }
3427                 if (err == 0)
3428                         continue;
3429
3430                 if (j == 0) {
3431                         targ_spec = cand_spec;
3432                 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
3433                         /* if there are many candidates, they should all
3434                          * resolve to the same bit offset
3435                          */
3436                         pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3437                                 prog_name, relo_idx, cand_spec.bit_offset,
3438                                 targ_spec.bit_offset);
3439                         return -EINVAL;
3440                 }
3441
3442                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3443         }
3444
3445         /*
3446          * For BPF_FIELD_EXISTS relo or when relaxed CO-RE reloc mode is
3447          * requested, it's expected that we might not find any candidates.
3448          * In this case, if field wasn't found in any candidate, the list of
3449          * candidates shouldn't change at all, we'll just handle relocating
3450          * appropriately, depending on relo's kind.
3451          */
3452         if (j > 0)
3453                 cand_ids->len = j;
3454
3455         if (j == 0 && !prog->obj->relaxed_core_relocs &&
3456             relo->kind != BPF_FIELD_EXISTS) {
3457                 pr_warn("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3458                         prog_name, relo_idx, local_id, local_name, spec_str);
3459                 return -ESRCH;
3460         }
3461
3462         /* bpf_core_reloc_insn should know how to handle missing targ_spec */
3463         err = bpf_core_reloc_insn(prog, relo, &local_spec,
3464                                   j ? &targ_spec : NULL);
3465         if (err) {
3466                 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3467                         prog_name, relo_idx, relo->insn_off, err);
3468                 return -EINVAL;
3469         }
3470
3471         return 0;
3472 }
3473
3474 static int
3475 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
3476 {
3477         const struct btf_ext_info_sec *sec;
3478         const struct bpf_field_reloc *rec;
3479         const struct btf_ext_info *seg;
3480         struct hashmap_entry *entry;
3481         struct hashmap *cand_cache = NULL;
3482         struct bpf_program *prog;
3483         struct btf *targ_btf;
3484         const char *sec_name;
3485         int i, err = 0;
3486
3487         if (targ_btf_path)
3488                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3489         else
3490                 targ_btf = bpf_core_find_kernel_btf();
3491         if (IS_ERR(targ_btf)) {
3492                 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
3493                 return PTR_ERR(targ_btf);
3494         }
3495
3496         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3497         if (IS_ERR(cand_cache)) {
3498                 err = PTR_ERR(cand_cache);
3499                 goto out;
3500         }
3501
3502         seg = &obj->btf_ext->field_reloc_info;
3503         for_each_btf_ext_sec(seg, sec) {
3504                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3505                 if (str_is_empty(sec_name)) {
3506                         err = -EINVAL;
3507                         goto out;
3508                 }
3509                 prog = bpf_object__find_program_by_title(obj, sec_name);
3510                 if (!prog) {
3511                         pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
3512                                 sec_name);
3513                         err = -EINVAL;
3514                         goto out;
3515                 }
3516
3517                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3518                          sec_name, sec->num_info);
3519
3520                 for_each_btf_ext_rec(seg, sec, i, rec) {
3521                         err = bpf_core_reloc_field(prog, rec, i, obj->btf,
3522                                                    targ_btf, cand_cache);
3523                         if (err) {
3524                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
3525                                         sec_name, i, err);
3526                                 goto out;
3527                         }
3528                 }
3529         }
3530
3531 out:
3532         btf__free(targ_btf);
3533         if (!IS_ERR_OR_NULL(cand_cache)) {
3534                 hashmap__for_each_entry(cand_cache, entry, i) {
3535                         bpf_core_free_cands(entry->value);
3536                 }
3537                 hashmap__free(cand_cache);
3538         }
3539         return err;
3540 }
3541
3542 static int
3543 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3544 {
3545         int err = 0;
3546
3547         if (obj->btf_ext->field_reloc_info.len)
3548                 err = bpf_core_reloc_fields(obj, targ_btf_path);
3549
3550         return err;
3551 }
3552
3553 static int
3554 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3555                         struct reloc_desc *relo)
3556 {
3557         struct bpf_insn *insn, *new_insn;
3558         struct bpf_program *text;
3559         size_t new_cnt;
3560         int err;
3561
3562         if (relo->type != RELO_CALL)
3563                 return -LIBBPF_ERRNO__RELOC;
3564
3565         if (prog->idx == obj->efile.text_shndx) {
3566                 pr_warn("relo in .text insn %d into off %d (insn #%d)\n",
3567                         relo->insn_idx, relo->sym_off, relo->sym_off / 8);
3568                 return -LIBBPF_ERRNO__RELOC;
3569         }
3570
3571         if (prog->main_prog_cnt == 0) {
3572                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3573                 if (!text) {
3574                         pr_warn("no .text section found yet relo into text exist\n");
3575                         return -LIBBPF_ERRNO__RELOC;
3576                 }
3577                 new_cnt = prog->insns_cnt + text->insns_cnt;
3578                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3579                 if (!new_insn) {
3580                         pr_warn("oom in prog realloc\n");
3581                         return -ENOMEM;
3582                 }
3583                 prog->insns = new_insn;
3584
3585                 if (obj->btf_ext) {
3586                         err = bpf_program_reloc_btf_ext(prog, obj,
3587                                                         text->section_name,
3588                                                         prog->insns_cnt);
3589                         if (err)
3590                                 return err;
3591                 }
3592
3593                 memcpy(new_insn + prog->insns_cnt, text->insns,
3594                        text->insns_cnt * sizeof(*insn));
3595                 prog->main_prog_cnt = prog->insns_cnt;
3596                 prog->insns_cnt = new_cnt;
3597                 pr_debug("added %zd insn from %s to prog %s\n",
3598                          text->insns_cnt, text->section_name,
3599                          prog->section_name);
3600         }
3601         insn = &prog->insns[relo->insn_idx];
3602         insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
3603         return 0;
3604 }
3605
3606 static int
3607 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3608 {
3609         int i, err;
3610
3611         if (!prog)
3612                 return 0;
3613
3614         if (obj->btf_ext) {
3615                 err = bpf_program_reloc_btf_ext(prog, obj,
3616                                                 prog->section_name, 0);
3617                 if (err)
3618                         return err;
3619         }
3620
3621         if (!prog->reloc_desc)
3622                 return 0;
3623
3624         for (i = 0; i < prog->nr_reloc; i++) {
3625                 struct reloc_desc *relo = &prog->reloc_desc[i];
3626
3627                 if (relo->type == RELO_LD64 || relo->type == RELO_DATA) {
3628                         struct bpf_insn *insn = &prog->insns[relo->insn_idx];
3629
3630                         if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
3631                                 pr_warn("relocation out of range: '%s'\n",
3632                                         prog->section_name);
3633                                 return -LIBBPF_ERRNO__RELOC;
3634                         }
3635
3636                         if (relo->type != RELO_DATA) {
3637                                 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
3638                         } else {
3639                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
3640                                 insn[1].imm = insn[0].imm + relo->sym_off;
3641                         }
3642                         insn[0].imm = obj->maps[relo->map_idx].fd;
3643                 } else if (relo->type == RELO_CALL) {
3644                         err = bpf_program__reloc_text(prog, obj, relo);
3645                         if (err)
3646                                 return err;
3647                 }
3648         }
3649
3650         zfree(&prog->reloc_desc);
3651         prog->nr_reloc = 0;
3652         return 0;
3653 }
3654
3655 static int
3656 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3657 {
3658         struct bpf_program *prog;
3659         size_t i;
3660         int err;
3661
3662         if (obj->btf_ext) {
3663                 err = bpf_object__relocate_core(obj, targ_btf_path);
3664                 if (err) {
3665                         pr_warn("failed to perform CO-RE relocations: %d\n",
3666                                 err);
3667                         return err;
3668                 }
3669         }
3670         for (i = 0; i < obj->nr_programs; i++) {
3671                 prog = &obj->programs[i];
3672
3673                 err = bpf_program__relocate(prog, obj);
3674                 if (err) {
3675                         pr_warn("failed to relocate '%s'\n", prog->section_name);
3676                         return err;
3677                 }
3678         }
3679         return 0;
3680 }
3681
3682 static int bpf_object__collect_reloc(struct bpf_object *obj)
3683 {
3684         int i, err;
3685
3686         if (!obj_elf_valid(obj)) {
3687                 pr_warn("Internal error: elf object is closed\n");
3688                 return -LIBBPF_ERRNO__INTERNAL;
3689         }
3690
3691         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
3692                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
3693                 Elf_Data *data = obj->efile.reloc_sects[i].data;
3694                 int idx = shdr->sh_info;
3695                 struct bpf_program *prog;
3696
3697                 if (shdr->sh_type != SHT_REL) {
3698                         pr_warn("internal error at %d\n", __LINE__);
3699                         return -LIBBPF_ERRNO__INTERNAL;
3700                 }
3701
3702                 prog = bpf_object__find_prog_by_idx(obj, idx);
3703                 if (!prog) {
3704                         pr_warn("relocation failed: no section(%d)\n", idx);
3705                         return -LIBBPF_ERRNO__RELOC;
3706                 }
3707
3708                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3709                 if (err)
3710                         return err;
3711         }
3712         return 0;
3713 }
3714
3715 static int
3716 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3717              char *license, __u32 kern_version, int *pfd)
3718 {
3719         struct bpf_load_program_attr load_attr;
3720         char *cp, errmsg[STRERR_BUFSIZE];
3721         int log_buf_size = BPF_LOG_BUF_SIZE;
3722         char *log_buf;
3723         int btf_fd, ret;
3724
3725         if (!insns || !insns_cnt)
3726                 return -EINVAL;
3727
3728         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3729         load_attr.prog_type = prog->type;
3730         load_attr.expected_attach_type = prog->expected_attach_type;
3731         if (prog->caps->name)
3732                 load_attr.name = prog->name;
3733         load_attr.insns = insns;
3734         load_attr.insns_cnt = insns_cnt;
3735         load_attr.license = license;
3736         if (prog->type == BPF_PROG_TYPE_TRACING) {
3737                 load_attr.attach_prog_fd = prog->attach_prog_fd;
3738                 load_attr.attach_btf_id = prog->attach_btf_id;
3739         } else {
3740                 load_attr.kern_version = kern_version;
3741                 load_attr.prog_ifindex = prog->prog_ifindex;
3742         }
3743         /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
3744         if (prog->obj->btf_ext)
3745                 btf_fd = bpf_object__btf_fd(prog->obj);
3746         else
3747                 btf_fd = -1;
3748         load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
3749         load_attr.func_info = prog->func_info;
3750         load_attr.func_info_rec_size = prog->func_info_rec_size;
3751         load_attr.func_info_cnt = prog->func_info_cnt;
3752         load_attr.line_info = prog->line_info;
3753         load_attr.line_info_rec_size = prog->line_info_rec_size;
3754         load_attr.line_info_cnt = prog->line_info_cnt;
3755         load_attr.log_level = prog->log_level;
3756         load_attr.prog_flags = prog->prog_flags;
3757
3758 retry_load:
3759         log_buf = malloc(log_buf_size);
3760         if (!log_buf)
3761                 pr_warn("Alloc log buffer for bpf loader error, continue without log\n");
3762
3763         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3764
3765         if (ret >= 0) {
3766                 if (load_attr.log_level)
3767                         pr_debug("verifier log:\n%s", log_buf);
3768                 *pfd = ret;
3769                 ret = 0;
3770                 goto out;
3771         }
3772
3773         if (errno == ENOSPC) {
3774                 log_buf_size <<= 1;
3775                 free(log_buf);
3776                 goto retry_load;
3777         }
3778         ret = -errno;
3779         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3780         pr_warn("load bpf program failed: %s\n", cp);
3781
3782         if (log_buf && log_buf[0] != '\0') {
3783                 ret = -LIBBPF_ERRNO__VERIFY;
3784                 pr_warn("-- BEGIN DUMP LOG ---\n");
3785                 pr_warn("\n%s\n", log_buf);
3786                 pr_warn("-- END LOG --\n");
3787         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3788                 pr_warn("Program too large (%zu insns), at most %d insns\n",
3789                         load_attr.insns_cnt, BPF_MAXINSNS);
3790                 ret = -LIBBPF_ERRNO__PROG2BIG;
3791         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3792                 /* Wrong program type? */
3793                 int fd;
3794
3795                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3796                 load_attr.expected_attach_type = 0;
3797                 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3798                 if (fd >= 0) {
3799                         close(fd);
3800                         ret = -LIBBPF_ERRNO__PROGTYPE;
3801                         goto out;
3802                 }
3803         }
3804
3805 out:
3806         free(log_buf);
3807         return ret;
3808 }
3809
3810 int
3811 bpf_program__load(struct bpf_program *prog,
3812                   char *license, __u32 kern_version)
3813 {
3814         int err = 0, fd, i;
3815
3816         if (prog->instances.nr < 0 || !prog->instances.fds) {
3817                 if (prog->preprocessor) {
3818                         pr_warn("Internal error: can't load program '%s'\n",
3819                                 prog->section_name);
3820                         return -LIBBPF_ERRNO__INTERNAL;
3821                 }
3822
3823                 prog->instances.fds = malloc(sizeof(int));
3824                 if (!prog->instances.fds) {
3825                         pr_warn("Not enough memory for BPF fds\n");
3826                         return -ENOMEM;
3827                 }
3828                 prog->instances.nr = 1;
3829                 prog->instances.fds[0] = -1;
3830         }
3831
3832         if (!prog->preprocessor) {
3833                 if (prog->instances.nr != 1) {
3834                         pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
3835                                 prog->section_name, prog->instances.nr);
3836                 }
3837                 err = load_program(prog, prog->insns, prog->insns_cnt,
3838                                    license, kern_version, &fd);
3839                 if (!err)
3840                         prog->instances.fds[0] = fd;
3841                 goto out;
3842         }
3843
3844         for (i = 0; i < prog->instances.nr; i++) {
3845                 struct bpf_prog_prep_result result;
3846                 bpf_program_prep_t preprocessor = prog->preprocessor;
3847
3848                 memset(&result, 0, sizeof(result));
3849                 err = preprocessor(prog, i, prog->insns,
3850                                    prog->insns_cnt, &result);
3851                 if (err) {
3852                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
3853                                 i, prog->section_name);
3854                         goto out;
3855                 }
3856
3857                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3858                         pr_debug("Skip loading the %dth instance of program '%s'\n",
3859                                  i, prog->section_name);
3860                         prog->instances.fds[i] = -1;
3861                         if (result.pfd)
3862                                 *result.pfd = -1;
3863                         continue;
3864                 }
3865
3866                 err = load_program(prog, result.new_insn_ptr,
3867                                    result.new_insn_cnt,
3868                                    license, kern_version, &fd);
3869
3870                 if (err) {
3871                         pr_warn("Loading the %dth instance of program '%s' failed\n",
3872                                 i, prog->section_name);
3873                         goto out;
3874                 }
3875
3876                 if (result.pfd)
3877                         *result.pfd = fd;
3878                 prog->instances.fds[i] = fd;
3879         }
3880 out:
3881         if (err)
3882                 pr_warn("failed to load program '%s'\n", prog->section_name);
3883         zfree(&prog->insns);
3884         prog->insns_cnt = 0;
3885         return err;
3886 }
3887
3888 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3889                                              const struct bpf_object *obj)
3890 {
3891         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3892 }
3893
3894 static int
3895 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3896 {
3897         size_t i;
3898         int err;
3899
3900         for (i = 0; i < obj->nr_programs; i++) {
3901                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3902                         continue;
3903                 obj->programs[i].log_level |= log_level;
3904                 err = bpf_program__load(&obj->programs[i],
3905                                         obj->license,
3906                                         obj->kern_version);
3907                 if (err)
3908                         return err;
3909         }
3910         return 0;
3911 }
3912
3913 static int libbpf_find_attach_btf_id(const char *name,
3914                                      enum bpf_attach_type attach_type,
3915                                      __u32 attach_prog_fd);
3916 static struct bpf_object *
3917 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
3918                    struct bpf_object_open_opts *opts)
3919 {
3920         const char *pin_root_path;
3921         struct bpf_program *prog;
3922         struct bpf_object *obj;
3923         const char *obj_name;
3924         char tmp_name[64];
3925         bool relaxed_maps;
3926         __u32 attach_prog_fd;
3927         int err;
3928
3929         if (elf_version(EV_CURRENT) == EV_NONE) {
3930                 pr_warn("failed to init libelf for %s\n",
3931                         path ? : "(mem buf)");
3932                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3933         }
3934
3935         if (!OPTS_VALID(opts, bpf_object_open_opts))
3936                 return ERR_PTR(-EINVAL);
3937
3938         obj_name = OPTS_GET(opts, object_name, NULL);
3939         if (obj_buf) {
3940                 if (!obj_name) {
3941                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3942                                  (unsigned long)obj_buf,
3943                                  (unsigned long)obj_buf_sz);
3944                         obj_name = tmp_name;
3945                 }
3946                 path = obj_name;
3947                 pr_debug("loading object '%s' from buffer\n", obj_name);
3948         }
3949
3950         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
3951         if (IS_ERR(obj))
3952                 return obj;
3953
3954         obj->relaxed_core_relocs = OPTS_GET(opts, relaxed_core_relocs, false);
3955         relaxed_maps = OPTS_GET(opts, relaxed_maps, false);
3956         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
3957         attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
3958
3959         CHECK_ERR(bpf_object__elf_init(obj), err, out);
3960         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3961         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3962         CHECK_ERR(bpf_object__elf_collect(obj, relaxed_maps, pin_root_path),
3963                   err, out);
3964         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3965         bpf_object__elf_finish(obj);
3966
3967         bpf_object__for_each_program(prog, obj) {
3968                 enum bpf_prog_type prog_type;
3969                 enum bpf_attach_type attach_type;
3970
3971                 err = libbpf_prog_type_by_name(prog->section_name, &prog_type,
3972                                                &attach_type);
3973                 if (err == -ESRCH)
3974                         /* couldn't guess, but user might manually specify */
3975                         continue;
3976                 if (err)
3977                         goto out;
3978
3979                 bpf_program__set_type(prog, prog_type);
3980                 bpf_program__set_expected_attach_type(prog, attach_type);
3981                 if (prog_type == BPF_PROG_TYPE_TRACING) {
3982                         err = libbpf_find_attach_btf_id(prog->section_name,
3983                                                         attach_type,
3984                                                         attach_prog_fd);
3985                         if (err <= 0)
3986                                 goto out;
3987                         prog->attach_btf_id = err;
3988                         prog->attach_prog_fd = attach_prog_fd;
3989                 }
3990         }
3991
3992         return obj;
3993 out:
3994         bpf_object__close(obj);
3995         return ERR_PTR(err);
3996 }
3997
3998 static struct bpf_object *
3999 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
4000 {
4001         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
4002                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
4003         );
4004
4005         /* param validation */
4006         if (!attr->file)
4007                 return NULL;
4008
4009         pr_debug("loading %s\n", attr->file);
4010         return __bpf_object__open(attr->file, NULL, 0, &opts);
4011 }
4012
4013 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
4014 {
4015         return __bpf_object__open_xattr(attr, 0);
4016 }
4017
4018 struct bpf_object *bpf_object__open(const char *path)
4019 {
4020         struct bpf_object_open_attr attr = {
4021                 .file           = path,
4022                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
4023         };
4024
4025         return bpf_object__open_xattr(&attr);
4026 }
4027
4028 struct bpf_object *
4029 bpf_object__open_file(const char *path, struct bpf_object_open_opts *opts)
4030 {
4031         if (!path)
4032                 return ERR_PTR(-EINVAL);
4033
4034         pr_debug("loading %s\n", path);
4035
4036         return __bpf_object__open(path, NULL, 0, opts);
4037 }
4038
4039 struct bpf_object *
4040 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
4041                      struct bpf_object_open_opts *opts)
4042 {
4043         if (!obj_buf || obj_buf_sz == 0)
4044                 return ERR_PTR(-EINVAL);
4045
4046         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
4047 }
4048
4049 struct bpf_object *
4050 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
4051                         const char *name)
4052 {
4053         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
4054                 .object_name = name,
4055                 /* wrong default, but backwards-compatible */
4056                 .relaxed_maps = true,
4057         );
4058
4059         /* returning NULL is wrong, but backwards-compatible */
4060         if (!obj_buf || obj_buf_sz == 0)
4061                 return NULL;
4062
4063         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
4064 }
4065
4066 int bpf_object__unload(struct bpf_object *obj)
4067 {
4068         size_t i;
4069
4070         if (!obj)
4071                 return -EINVAL;
4072
4073         for (i = 0; i < obj->nr_maps; i++)
4074                 zclose(obj->maps[i].fd);
4075
4076         for (i = 0; i < obj->nr_programs; i++)
4077                 bpf_program__unload(&obj->programs[i]);
4078
4079         return 0;
4080 }
4081
4082 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
4083 {
4084         struct bpf_object *obj;
4085         int err, i;
4086
4087         if (!attr)
4088                 return -EINVAL;
4089         obj = attr->obj;
4090         if (!obj)
4091                 return -EINVAL;
4092
4093         if (obj->loaded) {
4094                 pr_warn("object should not be loaded twice\n");
4095                 return -EINVAL;
4096         }
4097
4098         obj->loaded = true;
4099
4100         CHECK_ERR(bpf_object__create_maps(obj), err, out);
4101         CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
4102         CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
4103
4104         return 0;
4105 out:
4106         /* unpin any maps that were auto-pinned during load */
4107         for (i = 0; i < obj->nr_maps; i++)
4108                 if (obj->maps[i].pinned && !obj->maps[i].reused)
4109                         bpf_map__unpin(&obj->maps[i], NULL);
4110
4111         bpf_object__unload(obj);
4112         pr_warn("failed to load object '%s'\n", obj->path);
4113         return err;
4114 }
4115
4116 int bpf_object__load(struct bpf_object *obj)
4117 {
4118         struct bpf_object_load_attr attr = {
4119                 .obj = obj,
4120         };
4121
4122         return bpf_object__load_xattr(&attr);
4123 }
4124
4125 static int make_parent_dir(const char *path)
4126 {
4127         char *cp, errmsg[STRERR_BUFSIZE];
4128         char *dname, *dir;
4129         int err = 0;
4130
4131         dname = strdup(path);
4132         if (dname == NULL)
4133                 return -ENOMEM;
4134
4135         dir = dirname(dname);
4136         if (mkdir(dir, 0700) && errno != EEXIST)
4137                 err = -errno;
4138
4139         free(dname);
4140         if (err) {
4141                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4142                 pr_warn("failed to mkdir %s: %s\n", path, cp);
4143         }
4144         return err;
4145 }
4146
4147 static int check_path(const char *path)
4148 {
4149         char *cp, errmsg[STRERR_BUFSIZE];
4150         struct statfs st_fs;
4151         char *dname, *dir;
4152         int err = 0;
4153
4154         if (path == NULL)
4155                 return -EINVAL;
4156
4157         dname = strdup(path);
4158         if (dname == NULL)
4159                 return -ENOMEM;
4160
4161         dir = dirname(dname);
4162         if (statfs(dir, &st_fs)) {
4163                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4164                 pr_warn("failed to statfs %s: %s\n", dir, cp);
4165                 err = -errno;
4166         }
4167         free(dname);
4168
4169         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
4170                 pr_warn("specified path %s is not on BPF FS\n", path);
4171                 err = -EINVAL;
4172         }
4173
4174         return err;
4175 }
4176
4177 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
4178                               int instance)
4179 {
4180         char *cp, errmsg[STRERR_BUFSIZE];
4181         int err;
4182
4183         err = make_parent_dir(path);
4184         if (err)
4185                 return err;
4186
4187         err = check_path(path);
4188         if (err)
4189                 return err;
4190
4191         if (prog == NULL) {
4192                 pr_warn("invalid program pointer\n");
4193                 return -EINVAL;
4194         }
4195
4196         if (instance < 0 || instance >= prog->instances.nr) {
4197                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
4198                         instance, prog->section_name, prog->instances.nr);
4199                 return -EINVAL;
4200         }
4201
4202         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
4203                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4204                 pr_warn("failed to pin program: %s\n", cp);
4205                 return -errno;
4206         }
4207         pr_debug("pinned program '%s'\n", path);
4208
4209         return 0;
4210 }
4211
4212 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
4213                                 int instance)
4214 {
4215         int err;
4216
4217         err = check_path(path);
4218         if (err)
4219                 return err;
4220
4221         if (prog == NULL) {
4222                 pr_warn("invalid program pointer\n");
4223                 return -EINVAL;
4224         }
4225
4226         if (instance < 0 || instance >= prog->instances.nr) {
4227                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
4228                         instance, prog->section_name, prog->instances.nr);
4229                 return -EINVAL;
4230         }
4231
4232         err = unlink(path);
4233         if (err != 0)
4234                 return -errno;
4235         pr_debug("unpinned program '%s'\n", path);
4236
4237         return 0;
4238 }
4239
4240 int bpf_program__pin(struct bpf_program *prog, const char *path)
4241 {
4242         int i, err;
4243
4244         err = make_parent_dir(path);
4245         if (err)
4246                 return err;
4247
4248         err = check_path(path);
4249         if (err)
4250                 return err;
4251
4252         if (prog == NULL) {
4253                 pr_warn("invalid program pointer\n");
4254                 return -EINVAL;
4255         }
4256
4257         if (prog->instances.nr <= 0) {
4258                 pr_warn("no instances of prog %s to pin\n",
4259                            prog->section_name);
4260                 return -EINVAL;
4261         }
4262
4263         if (prog->instances.nr == 1) {
4264                 /* don't create subdirs when pinning single instance */
4265                 return bpf_program__pin_instance(prog, path, 0);
4266         }
4267
4268         for (i = 0; i < prog->instances.nr; i++) {
4269                 char buf[PATH_MAX];
4270                 int len;
4271
4272                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
4273                 if (len < 0) {
4274                         err = -EINVAL;
4275                         goto err_unpin;
4276                 } else if (len >= PATH_MAX) {
4277                         err = -ENAMETOOLONG;
4278                         goto err_unpin;
4279                 }
4280
4281                 err = bpf_program__pin_instance(prog, buf, i);
4282                 if (err)
4283                         goto err_unpin;
4284         }
4285
4286         return 0;
4287
4288 err_unpin:
4289         for (i = i - 1; i >= 0; i--) {
4290                 char buf[PATH_MAX];
4291                 int len;
4292
4293                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
4294                 if (len < 0)
4295                         continue;
4296                 else if (len >= PATH_MAX)
4297                         continue;
4298
4299                 bpf_program__unpin_instance(prog, buf, i);
4300         }
4301
4302         rmdir(path);
4303
4304         return err;
4305 }
4306
4307 int bpf_program__unpin(struct bpf_program *prog, const char *path)
4308 {
4309         int i, err;
4310
4311         err = check_path(path);
4312         if (err)
4313                 return err;
4314
4315         if (prog == NULL) {
4316                 pr_warn("invalid program pointer\n");
4317                 return -EINVAL;
4318         }
4319
4320         if (prog->instances.nr <= 0) {
4321                 pr_warn("no instances of prog %s to pin\n",
4322                            prog->section_name);
4323                 return -EINVAL;
4324         }
4325
4326         if (prog->instances.nr == 1) {
4327                 /* don't create subdirs when pinning single instance */
4328                 return bpf_program__unpin_instance(prog, path, 0);
4329         }
4330
4331         for (i = 0; i < prog->instances.nr; i++) {
4332                 char buf[PATH_MAX];
4333                 int len;
4334
4335                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
4336                 if (len < 0)
4337                         return -EINVAL;
4338                 else if (len >= PATH_MAX)
4339                         return -ENAMETOOLONG;
4340
4341                 err = bpf_program__unpin_instance(prog, buf, i);
4342                 if (err)
4343                         return err;
4344         }
4345
4346         err = rmdir(path);
4347         if (err)
4348                 return -errno;
4349
4350         return 0;
4351 }
4352
4353 int bpf_map__pin(struct bpf_map *map, const char *path)
4354 {
4355         char *cp, errmsg[STRERR_BUFSIZE];
4356         int err;
4357
4358         if (map == NULL) {
4359                 pr_warn("invalid map pointer\n");
4360                 return -EINVAL;
4361         }
4362
4363         if (map->pin_path) {
4364                 if (path && strcmp(path, map->pin_path)) {
4365                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
4366                                 bpf_map__name(map), map->pin_path, path);
4367                         return -EINVAL;
4368                 } else if (map->pinned) {
4369                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
4370                                  bpf_map__name(map), map->pin_path);
4371                         return 0;
4372                 }
4373         } else {
4374                 if (!path) {
4375                         pr_warn("missing a path to pin map '%s' at\n",
4376                                 bpf_map__name(map));
4377                         return -EINVAL;
4378                 } else if (map->pinned) {
4379                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
4380                         return -EEXIST;
4381                 }
4382
4383                 map->pin_path = strdup(path);
4384                 if (!map->pin_path) {
4385                         err = -errno;
4386                         goto out_err;
4387                 }
4388         }
4389
4390         err = make_parent_dir(map->pin_path);
4391         if (err)
4392                 return err;
4393
4394         err = check_path(map->pin_path);
4395         if (err)
4396                 return err;
4397
4398         if (bpf_obj_pin(map->fd, map->pin_path)) {
4399                 err = -errno;
4400                 goto out_err;
4401         }
4402
4403         map->pinned = true;
4404         pr_debug("pinned map '%s'\n", map->pin_path);
4405
4406         return 0;
4407
4408 out_err:
4409         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4410         pr_warn("failed to pin map: %s\n", cp);
4411         return err;
4412 }
4413
4414 int bpf_map__unpin(struct bpf_map *map, const char *path)
4415 {
4416         int err;
4417
4418         if (map == NULL) {
4419                 pr_warn("invalid map pointer\n");
4420                 return -EINVAL;
4421         }
4422
4423         if (map->pin_path) {
4424                 if (path && strcmp(path, map->pin_path)) {
4425                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
4426                                 bpf_map__name(map), map->pin_path, path);
4427                         return -EINVAL;
4428                 }
4429                 path = map->pin_path;
4430         } else if (!path) {
4431                 pr_warn("no path to unpin map '%s' from\n",
4432                         bpf_map__name(map));
4433                 return -EINVAL;
4434         }
4435
4436         err = check_path(path);
4437         if (err)
4438                 return err;
4439
4440         err = unlink(path);
4441         if (err != 0)
4442                 return -errno;
4443
4444         map->pinned = false;
4445         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
4446
4447         return 0;
4448 }
4449
4450 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
4451 {
4452         char *new = NULL;
4453
4454         if (path) {
4455                 new = strdup(path);
4456                 if (!new)
4457                         return -errno;
4458         }
4459
4460         free(map->pin_path);
4461         map->pin_path = new;
4462         return 0;
4463 }
4464
4465 const char *bpf_map__get_pin_path(const struct bpf_map *map)
4466 {
4467         return map->pin_path;
4468 }
4469
4470 bool bpf_map__is_pinned(const struct bpf_map *map)
4471 {
4472         return map->pinned;
4473 }
4474
4475 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
4476 {
4477         struct bpf_map *map;
4478         int err;
4479
4480         if (!obj)
4481                 return -ENOENT;
4482
4483         if (!obj->loaded) {
4484                 pr_warn("object not yet loaded; load it first\n");
4485                 return -ENOENT;
4486         }
4487
4488         bpf_object__for_each_map(map, obj) {
4489                 char *pin_path = NULL;
4490                 char buf[PATH_MAX];
4491
4492                 if (path) {
4493                         int len;
4494
4495                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
4496                                        bpf_map__name(map));
4497                         if (len < 0) {
4498                                 err = -EINVAL;
4499                                 goto err_unpin_maps;
4500                         } else if (len >= PATH_MAX) {
4501                                 err = -ENAMETOOLONG;
4502                                 goto err_unpin_maps;
4503                         }
4504                         pin_path = buf;
4505                 } else if (!map->pin_path) {
4506                         continue;
4507                 }
4508
4509                 err = bpf_map__pin(map, pin_path);
4510                 if (err)
4511                         goto err_unpin_maps;
4512         }
4513
4514         return 0;
4515
4516 err_unpin_maps:
4517         while ((map = bpf_map__prev(map, obj))) {
4518                 if (!map->pin_path)
4519                         continue;
4520
4521                 bpf_map__unpin(map, NULL);
4522         }
4523
4524         return err;
4525 }
4526
4527 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4528 {
4529         struct bpf_map *map;
4530         int err;
4531
4532         if (!obj)
4533                 return -ENOENT;
4534
4535         bpf_object__for_each_map(map, obj) {
4536                 char *pin_path = NULL;
4537                 char buf[PATH_MAX];
4538
4539                 if (path) {
4540                         int len;
4541
4542                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
4543                                        bpf_map__name(map));
4544                         if (len < 0)
4545                                 return -EINVAL;
4546                         else if (len >= PATH_MAX)
4547                                 return -ENAMETOOLONG;
4548                         pin_path = buf;
4549                 } else if (!map->pin_path) {
4550                         continue;
4551                 }
4552
4553                 err = bpf_map__unpin(map, pin_path);
4554                 if (err)
4555                         return err;
4556         }
4557
4558         return 0;
4559 }
4560
4561 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4562 {
4563         struct bpf_program *prog;
4564         int err;
4565
4566         if (!obj)
4567                 return -ENOENT;
4568
4569         if (!obj->loaded) {
4570                 pr_warn("object not yet loaded; load it first\n");
4571                 return -ENOENT;
4572         }
4573
4574         bpf_object__for_each_program(prog, obj) {
4575                 char buf[PATH_MAX];
4576                 int len;
4577
4578                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4579                                prog->pin_name);
4580                 if (len < 0) {
4581                         err = -EINVAL;
4582                         goto err_unpin_programs;
4583                 } else if (len >= PATH_MAX) {
4584                         err = -ENAMETOOLONG;
4585                         goto err_unpin_programs;
4586                 }
4587
4588                 err = bpf_program__pin(prog, buf);
4589                 if (err)
4590                         goto err_unpin_programs;
4591         }
4592
4593         return 0;
4594
4595 err_unpin_programs:
4596         while ((prog = bpf_program__prev(prog, obj))) {
4597                 char buf[PATH_MAX];
4598                 int len;
4599
4600                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4601                                prog->pin_name);
4602                 if (len < 0)
4603                         continue;
4604                 else if (len >= PATH_MAX)
4605                         continue;
4606
4607                 bpf_program__unpin(prog, buf);
4608         }
4609
4610         return err;
4611 }
4612
4613 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4614 {
4615         struct bpf_program *prog;
4616         int err;
4617
4618         if (!obj)
4619                 return -ENOENT;
4620
4621         bpf_object__for_each_program(prog, obj) {
4622                 char buf[PATH_MAX];
4623                 int len;
4624
4625                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4626                                prog->pin_name);
4627                 if (len < 0)
4628                         return -EINVAL;
4629                 else if (len >= PATH_MAX)
4630                         return -ENAMETOOLONG;
4631
4632                 err = bpf_program__unpin(prog, buf);
4633                 if (err)
4634                         return err;
4635         }
4636
4637         return 0;
4638 }
4639
4640 int bpf_object__pin(struct bpf_object *obj, const char *path)
4641 {
4642         int err;
4643
4644         err = bpf_object__pin_maps(obj, path);
4645         if (err)
4646                 return err;
4647
4648         err = bpf_object__pin_programs(obj, path);
4649         if (err) {
4650                 bpf_object__unpin_maps(obj, path);
4651                 return err;
4652         }
4653
4654         return 0;
4655 }
4656
4657 void bpf_object__close(struct bpf_object *obj)
4658 {
4659         size_t i;
4660
4661         if (!obj)
4662                 return;
4663
4664         if (obj->clear_priv)
4665                 obj->clear_priv(obj, obj->priv);
4666
4667         bpf_object__elf_finish(obj);
4668         bpf_object__unload(obj);
4669         btf__free(obj->btf);
4670         btf_ext__free(obj->btf_ext);
4671
4672         for (i = 0; i < obj->nr_maps; i++) {
4673                 zfree(&obj->maps[i].name);
4674                 zfree(&obj->maps[i].pin_path);
4675                 if (obj->maps[i].clear_priv)
4676                         obj->maps[i].clear_priv(&obj->maps[i],
4677                                                 obj->maps[i].priv);
4678                 obj->maps[i].priv = NULL;
4679                 obj->maps[i].clear_priv = NULL;
4680         }
4681
4682         zfree(&obj->sections.rodata);
4683         zfree(&obj->sections.data);
4684         zfree(&obj->maps);
4685         obj->nr_maps = 0;
4686
4687         if (obj->programs && obj->nr_programs) {
4688                 for (i = 0; i < obj->nr_programs; i++)
4689                         bpf_program__exit(&obj->programs[i]);
4690         }
4691         zfree(&obj->programs);
4692
4693         list_del(&obj->list);
4694         free(obj);
4695 }
4696
4697 struct bpf_object *
4698 bpf_object__next(struct bpf_object *prev)
4699 {
4700         struct bpf_object *next;
4701
4702         if (!prev)
4703                 next = list_first_entry(&bpf_objects_list,
4704                                         struct bpf_object,
4705                                         list);
4706         else
4707                 next = list_next_entry(prev, list);
4708
4709         /* Empty list is noticed here so don't need checking on entry. */
4710         if (&next->list == &bpf_objects_list)
4711                 return NULL;
4712
4713         return next;
4714 }
4715
4716 const char *bpf_object__name(const struct bpf_object *obj)
4717 {
4718         return obj ? obj->name : ERR_PTR(-EINVAL);
4719 }
4720
4721 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4722 {
4723         return obj ? obj->kern_version : 0;
4724 }
4725
4726 struct btf *bpf_object__btf(const struct bpf_object *obj)
4727 {
4728         return obj ? obj->btf : NULL;
4729 }
4730
4731 int bpf_object__btf_fd(const struct bpf_object *obj)
4732 {
4733         return obj->btf ? btf__fd(obj->btf) : -1;
4734 }
4735
4736 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4737                          bpf_object_clear_priv_t clear_priv)
4738 {
4739         if (obj->priv && obj->clear_priv)
4740                 obj->clear_priv(obj, obj->priv);
4741
4742         obj->priv = priv;
4743         obj->clear_priv = clear_priv;
4744         return 0;
4745 }
4746
4747 void *bpf_object__priv(const struct bpf_object *obj)
4748 {
4749         return obj ? obj->priv : ERR_PTR(-EINVAL);
4750 }
4751
4752 static struct bpf_program *
4753 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4754                     bool forward)
4755 {
4756         size_t nr_programs = obj->nr_programs;
4757         ssize_t idx;
4758
4759         if (!nr_programs)
4760                 return NULL;
4761
4762         if (!p)
4763                 /* Iter from the beginning */
4764                 return forward ? &obj->programs[0] :
4765                         &obj->programs[nr_programs - 1];
4766
4767         if (p->obj != obj) {
4768                 pr_warn("error: program handler doesn't match object\n");
4769                 return NULL;
4770         }
4771
4772         idx = (p - obj->programs) + (forward ? 1 : -1);
4773         if (idx >= obj->nr_programs || idx < 0)
4774                 return NULL;
4775         return &obj->programs[idx];
4776 }
4777
4778 struct bpf_program *
4779 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4780 {
4781         struct bpf_program *prog = prev;
4782
4783         do {
4784                 prog = __bpf_program__iter(prog, obj, true);
4785         } while (prog && bpf_program__is_function_storage(prog, obj));
4786
4787         return prog;
4788 }
4789
4790 struct bpf_program *
4791 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4792 {
4793         struct bpf_program *prog = next;
4794
4795         do {
4796                 prog = __bpf_program__iter(prog, obj, false);
4797         } while (prog && bpf_program__is_function_storage(prog, obj));
4798
4799         return prog;
4800 }
4801
4802 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4803                           bpf_program_clear_priv_t clear_priv)
4804 {
4805         if (prog->priv && prog->clear_priv)
4806                 prog->clear_priv(prog, prog->priv);
4807
4808         prog->priv = priv;
4809         prog->clear_priv = clear_priv;
4810         return 0;
4811 }
4812
4813 void *bpf_program__priv(const struct bpf_program *prog)
4814 {
4815         return prog ? prog->priv : ERR_PTR(-EINVAL);
4816 }
4817
4818 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4819 {
4820         prog->prog_ifindex = ifindex;
4821 }
4822
4823 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4824 {
4825         const char *title;
4826
4827         title = prog->section_name;
4828         if (needs_copy) {
4829                 title = strdup(title);
4830                 if (!title) {
4831                         pr_warn("failed to strdup program title\n");
4832                         return ERR_PTR(-ENOMEM);
4833                 }
4834         }
4835
4836         return title;
4837 }
4838
4839 int bpf_program__fd(const struct bpf_program *prog)
4840 {
4841         return bpf_program__nth_fd(prog, 0);
4842 }
4843
4844 size_t bpf_program__size(const struct bpf_program *prog)
4845 {
4846         return prog->insns_cnt * sizeof(struct bpf_insn);
4847 }
4848
4849 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4850                           bpf_program_prep_t prep)
4851 {
4852         int *instances_fds;
4853
4854         if (nr_instances <= 0 || !prep)
4855                 return -EINVAL;
4856
4857         if (prog->instances.nr > 0 || prog->instances.fds) {
4858                 pr_warn("Can't set pre-processor after loading\n");
4859                 return -EINVAL;
4860         }
4861
4862         instances_fds = malloc(sizeof(int) * nr_instances);
4863         if (!instances_fds) {
4864                 pr_warn("alloc memory failed for fds\n");
4865                 return -ENOMEM;
4866         }
4867
4868         /* fill all fd with -1 */
4869         memset(instances_fds, -1, sizeof(int) * nr_instances);
4870
4871         prog->instances.nr = nr_instances;
4872         prog->instances.fds = instances_fds;
4873         prog->preprocessor = prep;
4874         return 0;
4875 }
4876
4877 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4878 {
4879         int fd;
4880
4881         if (!prog)
4882                 return -EINVAL;
4883
4884         if (n >= prog->instances.nr || n < 0) {
4885                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
4886                         n, prog->section_name, prog->instances.nr);
4887                 return -EINVAL;
4888         }
4889
4890         fd = prog->instances.fds[n];
4891         if (fd < 0) {
4892                 pr_warn("%dth instance of program '%s' is invalid\n",
4893                         n, prog->section_name);
4894                 return -ENOENT;
4895         }
4896
4897         return fd;
4898 }
4899
4900 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
4901 {
4902         return prog->type;
4903 }
4904
4905 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4906 {
4907         prog->type = type;
4908 }
4909
4910 static bool bpf_program__is_type(const struct bpf_program *prog,
4911                                  enum bpf_prog_type type)
4912 {
4913         return prog ? (prog->type == type) : false;
4914 }
4915
4916 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
4917 int bpf_program__set_##NAME(struct bpf_program *prog)           \
4918 {                                                               \
4919         if (!prog)                                              \
4920                 return -EINVAL;                                 \
4921         bpf_program__set_type(prog, TYPE);                      \
4922         return 0;                                               \
4923 }                                                               \
4924                                                                 \
4925 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
4926 {                                                               \
4927         return bpf_program__is_type(prog, TYPE);                \
4928 }                                                               \
4929
4930 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4931 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4932 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4933 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4934 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4935 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4936 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4937 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4938 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
4939
4940 enum bpf_attach_type
4941 bpf_program__get_expected_attach_type(struct bpf_program *prog)
4942 {
4943         return prog->expected_attach_type;
4944 }
4945
4946 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4947                                            enum bpf_attach_type type)
4948 {
4949         prog->expected_attach_type = type;
4950 }
4951
4952 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, btf, atype) \
4953         { string, sizeof(string) - 1, ptype, eatype, is_attachable, btf, atype }
4954
4955 /* Programs that can NOT be attached. */
4956 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
4957
4958 /* Programs that can be attached. */
4959 #define BPF_APROG_SEC(string, ptype, atype) \
4960         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, 0, atype)
4961
4962 /* Programs that must specify expected attach type at load time. */
4963 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4964         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
4965
4966 /* Programs that use BTF to identify attach point */
4967 #define BPF_PROG_BTF(string, ptype, eatype) \
4968         BPF_PROG_SEC_IMPL(string, ptype, eatype, 0, 1, 0)
4969
4970 /* Programs that can be attached but attach type can't be identified by section
4971  * name. Kept for backward compatibility.
4972  */
4973 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4974
4975 static const struct {
4976         const char *sec;
4977         size_t len;
4978         enum bpf_prog_type prog_type;
4979         enum bpf_attach_type expected_attach_type;
4980         bool is_attachable;
4981         bool is_attach_btf;
4982         enum bpf_attach_type attach_type;
4983 } section_names[] = {
4984         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
4985         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
4986         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
4987         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
4988         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
4989         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
4990         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
4991         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
4992         BPF_PROG_SEC("tp/",                     BPF_PROG_TYPE_TRACEPOINT),
4993         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
4994         BPF_PROG_SEC("raw_tp/",                 BPF_PROG_TYPE_RAW_TRACEPOINT),
4995         BPF_PROG_BTF("tp_btf/",                 BPF_PROG_TYPE_TRACING,
4996                                                 BPF_TRACE_RAW_TP),
4997         BPF_PROG_BTF("fentry/",                 BPF_PROG_TYPE_TRACING,
4998                                                 BPF_TRACE_FENTRY),
4999         BPF_PROG_BTF("fexit/",                  BPF_PROG_TYPE_TRACING,
5000                                                 BPF_TRACE_FEXIT),
5001         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
5002         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
5003         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
5004         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
5005         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
5006         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
5007         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
5008                                                 BPF_CGROUP_INET_INGRESS),
5009         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
5010                                                 BPF_CGROUP_INET_EGRESS),
5011         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
5012         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
5013                                                 BPF_CGROUP_INET_SOCK_CREATE),
5014         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
5015                                                 BPF_CGROUP_INET4_POST_BIND),
5016         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
5017                                                 BPF_CGROUP_INET6_POST_BIND),
5018         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
5019                                                 BPF_CGROUP_DEVICE),
5020         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
5021                                                 BPF_CGROUP_SOCK_OPS),
5022         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
5023                                                 BPF_SK_SKB_STREAM_PARSER),
5024         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
5025                                                 BPF_SK_SKB_STREAM_VERDICT),
5026         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
5027         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
5028                                                 BPF_SK_MSG_VERDICT),
5029         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
5030                                                 BPF_LIRC_MODE2),
5031         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
5032                                                 BPF_FLOW_DISSECTOR),
5033         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5034                                                 BPF_CGROUP_INET4_BIND),
5035         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5036                                                 BPF_CGROUP_INET6_BIND),
5037         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5038                                                 BPF_CGROUP_INET4_CONNECT),
5039         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5040                                                 BPF_CGROUP_INET6_CONNECT),
5041         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5042                                                 BPF_CGROUP_UDP4_SENDMSG),
5043         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5044                                                 BPF_CGROUP_UDP6_SENDMSG),
5045         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5046                                                 BPF_CGROUP_UDP4_RECVMSG),
5047         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5048                                                 BPF_CGROUP_UDP6_RECVMSG),
5049         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
5050                                                 BPF_CGROUP_SYSCTL),
5051         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
5052                                                 BPF_CGROUP_GETSOCKOPT),
5053         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
5054                                                 BPF_CGROUP_SETSOCKOPT),
5055 };
5056
5057 #undef BPF_PROG_SEC_IMPL
5058 #undef BPF_PROG_SEC
5059 #undef BPF_APROG_SEC
5060 #undef BPF_EAPROG_SEC
5061 #undef BPF_APROG_COMPAT
5062
5063 #define MAX_TYPE_NAME_SIZE 32
5064
5065 static char *libbpf_get_type_names(bool attach_type)
5066 {
5067         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
5068         char *buf;
5069
5070         buf = malloc(len);
5071         if (!buf)
5072                 return NULL;
5073
5074         buf[0] = '\0';
5075         /* Forge string buf with all available names */
5076         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
5077                 if (attach_type && !section_names[i].is_attachable)
5078                         continue;
5079
5080                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
5081                         free(buf);
5082                         return NULL;
5083                 }
5084                 strcat(buf, " ");
5085                 strcat(buf, section_names[i].sec);
5086         }
5087
5088         return buf;
5089 }
5090
5091 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
5092                              enum bpf_attach_type *expected_attach_type)
5093 {
5094         char *type_names;
5095         int i;
5096
5097         if (!name)
5098                 return -EINVAL;
5099
5100         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
5101                 if (strncmp(name, section_names[i].sec, section_names[i].len))
5102                         continue;
5103                 *prog_type = section_names[i].prog_type;
5104                 *expected_attach_type = section_names[i].expected_attach_type;
5105                 return 0;
5106         }
5107         pr_warn("failed to guess program type from ELF section '%s'\n", name);
5108         type_names = libbpf_get_type_names(false);
5109         if (type_names != NULL) {
5110                 pr_info("supported section(type) names are:%s\n", type_names);
5111                 free(type_names);
5112         }
5113
5114         return -ESRCH;
5115 }
5116
5117 #define BTF_PREFIX "btf_trace_"
5118 int libbpf_find_vmlinux_btf_id(const char *name,
5119                                enum bpf_attach_type attach_type)
5120 {
5121         struct btf *btf = bpf_core_find_kernel_btf();
5122         char raw_tp_btf[128] = BTF_PREFIX;
5123         char *dst = raw_tp_btf + sizeof(BTF_PREFIX) - 1;
5124         const char *btf_name;
5125         int err = -EINVAL;
5126         __u32 kind;
5127
5128         if (IS_ERR(btf)) {
5129                 pr_warn("vmlinux BTF is not found\n");
5130                 return -EINVAL;
5131         }
5132
5133         if (attach_type == BPF_TRACE_RAW_TP) {
5134                 /* prepend "btf_trace_" prefix per kernel convention */
5135                 strncat(dst, name, sizeof(raw_tp_btf) - sizeof(BTF_PREFIX));
5136                 btf_name = raw_tp_btf;
5137                 kind = BTF_KIND_TYPEDEF;
5138         } else {
5139                 btf_name = name;
5140                 kind = BTF_KIND_FUNC;
5141         }
5142         err = btf__find_by_name_kind(btf, btf_name, kind);
5143         btf__free(btf);
5144         return err;
5145 }
5146
5147 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
5148 {
5149         struct bpf_prog_info_linear *info_linear;
5150         struct bpf_prog_info *info;
5151         struct btf *btf = NULL;
5152         int err = -EINVAL;
5153
5154         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
5155         if (IS_ERR_OR_NULL(info_linear)) {
5156                 pr_warn("failed get_prog_info_linear for FD %d\n",
5157                         attach_prog_fd);
5158                 return -EINVAL;
5159         }
5160         info = &info_linear->info;
5161         if (!info->btf_id) {
5162                 pr_warn("The target program doesn't have BTF\n");
5163                 goto out;
5164         }
5165         if (btf__get_from_id(info->btf_id, &btf)) {
5166                 pr_warn("Failed to get BTF of the program\n");
5167                 goto out;
5168         }
5169         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
5170         btf__free(btf);
5171         if (err <= 0) {
5172                 pr_warn("%s is not found in prog's BTF\n", name);
5173                 goto out;
5174         }
5175 out:
5176         free(info_linear);
5177         return err;
5178 }
5179
5180 static int libbpf_find_attach_btf_id(const char *name,
5181                                      enum bpf_attach_type attach_type,
5182                                      __u32 attach_prog_fd)
5183 {
5184         int i, err;
5185
5186         if (!name)
5187                 return -EINVAL;
5188
5189         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
5190                 if (!section_names[i].is_attach_btf)
5191                         continue;
5192                 if (strncmp(name, section_names[i].sec, section_names[i].len))
5193                         continue;
5194                 if (attach_prog_fd)
5195                         err = libbpf_find_prog_btf_id(name + section_names[i].len,
5196                                                       attach_prog_fd);
5197                 else
5198                         err = libbpf_find_vmlinux_btf_id(name + section_names[i].len,
5199                                                          attach_type);
5200                 if (err <= 0)
5201                         pr_warn("%s is not found in vmlinux BTF\n", name);
5202                 return err;
5203         }
5204         pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
5205         return -ESRCH;
5206 }
5207
5208 int libbpf_attach_type_by_name(const char *name,
5209                                enum bpf_attach_type *attach_type)
5210 {
5211         char *type_names;
5212         int i;
5213
5214         if (!name)
5215                 return -EINVAL;
5216
5217         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
5218                 if (strncmp(name, section_names[i].sec, section_names[i].len))
5219                         continue;
5220                 if (!section_names[i].is_attachable)
5221                         return -EINVAL;
5222                 *attach_type = section_names[i].attach_type;
5223                 return 0;
5224         }
5225         pr_warn("failed to guess attach type based on ELF section name '%s'\n", name);
5226         type_names = libbpf_get_type_names(true);
5227         if (type_names != NULL) {
5228                 pr_info("attachable section(type) names are:%s\n", type_names);
5229                 free(type_names);
5230         }
5231
5232         return -EINVAL;
5233 }
5234
5235 int bpf_map__fd(const struct bpf_map *map)
5236 {
5237         return map ? map->fd : -EINVAL;
5238 }
5239
5240 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
5241 {
5242         return map ? &map->def : ERR_PTR(-EINVAL);
5243 }
5244
5245 const char *bpf_map__name(const struct bpf_map *map)
5246 {
5247         return map ? map->name : NULL;
5248 }
5249
5250 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
5251 {
5252         return map ? map->btf_key_type_id : 0;
5253 }
5254
5255 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
5256 {
5257         return map ? map->btf_value_type_id : 0;
5258 }
5259
5260 int bpf_map__set_priv(struct bpf_map *map, void *priv,
5261                      bpf_map_clear_priv_t clear_priv)
5262 {
5263         if (!map)
5264                 return -EINVAL;
5265
5266         if (map->priv) {
5267                 if (map->clear_priv)
5268                         map->clear_priv(map, map->priv);
5269         }
5270
5271         map->priv = priv;
5272         map->clear_priv = clear_priv;
5273         return 0;
5274 }
5275
5276 void *bpf_map__priv(const struct bpf_map *map)
5277 {
5278         return map ? map->priv : ERR_PTR(-EINVAL);
5279 }
5280
5281 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
5282 {
5283         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
5284 }
5285
5286 bool bpf_map__is_internal(const struct bpf_map *map)
5287 {
5288         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
5289 }
5290
5291 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
5292 {
5293         map->map_ifindex = ifindex;
5294 }
5295
5296 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
5297 {
5298         if (!bpf_map_type__is_map_in_map(map->def.type)) {
5299                 pr_warn("error: unsupported map type\n");
5300                 return -EINVAL;
5301         }
5302         if (map->inner_map_fd != -1) {
5303                 pr_warn("error: inner_map_fd already specified\n");
5304                 return -EINVAL;
5305         }
5306         map->inner_map_fd = fd;
5307         return 0;
5308 }
5309
5310 static struct bpf_map *
5311 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
5312 {
5313         ssize_t idx;
5314         struct bpf_map *s, *e;
5315
5316         if (!obj || !obj->maps)
5317                 return NULL;
5318
5319         s = obj->maps;
5320         e = obj->maps + obj->nr_maps;
5321
5322         if ((m < s) || (m >= e)) {
5323                 pr_warn("error in %s: map handler doesn't belong to object\n",
5324                          __func__);
5325                 return NULL;
5326         }
5327
5328         idx = (m - obj->maps) + i;
5329         if (idx >= obj->nr_maps || idx < 0)
5330                 return NULL;
5331         return &obj->maps[idx];
5332 }
5333
5334 struct bpf_map *
5335 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
5336 {
5337         if (prev == NULL)
5338                 return obj->maps;
5339
5340         return __bpf_map__iter(prev, obj, 1);
5341 }
5342
5343 struct bpf_map *
5344 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
5345 {
5346         if (next == NULL) {
5347                 if (!obj->nr_maps)
5348                         return NULL;
5349                 return obj->maps + obj->nr_maps - 1;
5350         }
5351
5352         return __bpf_map__iter(next, obj, -1);
5353 }
5354
5355 struct bpf_map *
5356 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
5357 {
5358         struct bpf_map *pos;
5359
5360         bpf_object__for_each_map(pos, obj) {
5361                 if (pos->name && !strcmp(pos->name, name))
5362                         return pos;
5363         }
5364         return NULL;
5365 }
5366
5367 int
5368 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
5369 {
5370         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
5371 }
5372
5373 struct bpf_map *
5374 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
5375 {
5376         return ERR_PTR(-ENOTSUP);
5377 }
5378
5379 long libbpf_get_error(const void *ptr)
5380 {
5381         return PTR_ERR_OR_ZERO(ptr);
5382 }
5383
5384 int bpf_prog_load(const char *file, enum bpf_prog_type type,
5385                   struct bpf_object **pobj, int *prog_fd)
5386 {
5387         struct bpf_prog_load_attr attr;
5388
5389         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
5390         attr.file = file;
5391         attr.prog_type = type;
5392         attr.expected_attach_type = 0;
5393
5394         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
5395 }
5396
5397 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
5398                         struct bpf_object **pobj, int *prog_fd)
5399 {
5400         struct bpf_object_open_attr open_attr = {};
5401         struct bpf_program *prog, *first_prog = NULL;
5402         struct bpf_object *obj;
5403         struct bpf_map *map;
5404         int err;
5405
5406         if (!attr)
5407                 return -EINVAL;
5408         if (!attr->file)
5409                 return -EINVAL;
5410
5411         open_attr.file = attr->file;
5412         open_attr.prog_type = attr->prog_type;
5413
5414         obj = bpf_object__open_xattr(&open_attr);
5415         if (IS_ERR_OR_NULL(obj))
5416                 return -ENOENT;
5417
5418         bpf_object__for_each_program(prog, obj) {
5419                 enum bpf_attach_type attach_type = attr->expected_attach_type;
5420                 /*
5421                  * to preserve backwards compatibility, bpf_prog_load treats
5422                  * attr->prog_type, if specified, as an override to whatever
5423                  * bpf_object__open guessed
5424                  */
5425                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
5426                         bpf_program__set_type(prog, attr->prog_type);
5427                         bpf_program__set_expected_attach_type(prog,
5428                                                               attach_type);
5429                 }
5430                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
5431                         /*
5432                          * we haven't guessed from section name and user
5433                          * didn't provide a fallback type, too bad...
5434                          */
5435                         bpf_object__close(obj);
5436                         return -EINVAL;
5437                 }
5438
5439                 prog->prog_ifindex = attr->ifindex;
5440                 prog->log_level = attr->log_level;
5441                 prog->prog_flags = attr->prog_flags;
5442                 if (!first_prog)
5443                         first_prog = prog;
5444         }
5445
5446         bpf_object__for_each_map(map, obj) {
5447                 if (!bpf_map__is_offload_neutral(map))
5448                         map->map_ifindex = attr->ifindex;
5449         }
5450
5451         if (!first_prog) {
5452                 pr_warn("object file doesn't contain bpf program\n");
5453                 bpf_object__close(obj);
5454                 return -ENOENT;
5455         }
5456
5457         err = bpf_object__load(obj);
5458         if (err) {
5459                 bpf_object__close(obj);
5460                 return -EINVAL;
5461         }
5462
5463         *pobj = obj;
5464         *prog_fd = bpf_program__fd(first_prog);
5465         return 0;
5466 }
5467
5468 struct bpf_link {
5469         int (*destroy)(struct bpf_link *link);
5470 };
5471
5472 int bpf_link__destroy(struct bpf_link *link)
5473 {
5474         int err;
5475
5476         if (!link)
5477                 return 0;
5478
5479         err = link->destroy(link);
5480         free(link);
5481
5482         return err;
5483 }
5484
5485 struct bpf_link_fd {
5486         struct bpf_link link; /* has to be at the top of struct */
5487         int fd; /* hook FD */
5488 };
5489
5490 static int bpf_link__destroy_perf_event(struct bpf_link *link)
5491 {
5492         struct bpf_link_fd *l = (void *)link;
5493         int err;
5494
5495         err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
5496         if (err)
5497                 err = -errno;
5498
5499         close(l->fd);
5500         return err;
5501 }
5502
5503 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
5504                                                 int pfd)
5505 {
5506         char errmsg[STRERR_BUFSIZE];
5507         struct bpf_link_fd *link;
5508         int prog_fd, err;
5509
5510         if (pfd < 0) {
5511                 pr_warn("program '%s': invalid perf event FD %d\n",
5512                         bpf_program__title(prog, false), pfd);
5513                 return ERR_PTR(-EINVAL);
5514         }
5515         prog_fd = bpf_program__fd(prog);
5516         if (prog_fd < 0) {
5517                 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
5518                         bpf_program__title(prog, false));
5519                 return ERR_PTR(-EINVAL);
5520         }
5521
5522         link = malloc(sizeof(*link));
5523         if (!link)
5524                 return ERR_PTR(-ENOMEM);
5525         link->link.destroy = &bpf_link__destroy_perf_event;
5526         link->fd = pfd;
5527
5528         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
5529                 err = -errno;
5530                 free(link);
5531                 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
5532                         bpf_program__title(prog, false), pfd,
5533                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5534                 return ERR_PTR(err);
5535         }
5536         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5537                 err = -errno;
5538                 free(link);
5539                 pr_warn("program '%s': failed to enable pfd %d: %s\n",
5540                         bpf_program__title(prog, false), pfd,
5541                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5542                 return ERR_PTR(err);
5543         }
5544         return (struct bpf_link *)link;
5545 }
5546
5547 /*
5548  * this function is expected to parse integer in the range of [0, 2^31-1] from
5549  * given file using scanf format string fmt. If actual parsed value is
5550  * negative, the result might be indistinguishable from error
5551  */
5552 static int parse_uint_from_file(const char *file, const char *fmt)
5553 {
5554         char buf[STRERR_BUFSIZE];
5555         int err, ret;
5556         FILE *f;
5557
5558         f = fopen(file, "r");
5559         if (!f) {
5560                 err = -errno;
5561                 pr_debug("failed to open '%s': %s\n", file,
5562                          libbpf_strerror_r(err, buf, sizeof(buf)));
5563                 return err;
5564         }
5565         err = fscanf(f, fmt, &ret);
5566         if (err != 1) {
5567                 err = err == EOF ? -EIO : -errno;
5568                 pr_debug("failed to parse '%s': %s\n", file,
5569                         libbpf_strerror_r(err, buf, sizeof(buf)));
5570                 fclose(f);
5571                 return err;
5572         }
5573         fclose(f);
5574         return ret;
5575 }
5576
5577 static int determine_kprobe_perf_type(void)
5578 {
5579         const char *file = "/sys/bus/event_source/devices/kprobe/type";
5580
5581         return parse_uint_from_file(file, "%d\n");
5582 }
5583
5584 static int determine_uprobe_perf_type(void)
5585 {
5586         const char *file = "/sys/bus/event_source/devices/uprobe/type";
5587
5588         return parse_uint_from_file(file, "%d\n");
5589 }
5590
5591 static int determine_kprobe_retprobe_bit(void)
5592 {
5593         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5594
5595         return parse_uint_from_file(file, "config:%d\n");
5596 }
5597
5598 static int determine_uprobe_retprobe_bit(void)
5599 {
5600         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5601
5602         return parse_uint_from_file(file, "config:%d\n");
5603 }
5604
5605 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5606                                  uint64_t offset, int pid)
5607 {
5608         struct perf_event_attr attr = {};
5609         char errmsg[STRERR_BUFSIZE];
5610         int type, pfd, err;
5611
5612         type = uprobe ? determine_uprobe_perf_type()
5613                       : determine_kprobe_perf_type();
5614         if (type < 0) {
5615                 pr_warn("failed to determine %s perf type: %s\n",
5616                         uprobe ? "uprobe" : "kprobe",
5617                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5618                 return type;
5619         }
5620         if (retprobe) {
5621                 int bit = uprobe ? determine_uprobe_retprobe_bit()
5622                                  : determine_kprobe_retprobe_bit();
5623
5624                 if (bit < 0) {
5625                         pr_warn("failed to determine %s retprobe bit: %s\n",
5626                                 uprobe ? "uprobe" : "kprobe",
5627                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
5628                         return bit;
5629                 }
5630                 attr.config |= 1 << bit;
5631         }
5632         attr.size = sizeof(attr);
5633         attr.type = type;
5634         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5635         attr.config2 = offset;           /* kprobe_addr or probe_offset */
5636
5637         /* pid filter is meaningful only for uprobes */
5638         pfd = syscall(__NR_perf_event_open, &attr,
5639                       pid < 0 ? -1 : pid /* pid */,
5640                       pid == -1 ? 0 : -1 /* cpu */,
5641                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5642         if (pfd < 0) {
5643                 err = -errno;
5644                 pr_warn("%s perf_event_open() failed: %s\n",
5645                         uprobe ? "uprobe" : "kprobe",
5646                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5647                 return err;
5648         }
5649         return pfd;
5650 }
5651
5652 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5653                                             bool retprobe,
5654                                             const char *func_name)
5655 {
5656         char errmsg[STRERR_BUFSIZE];
5657         struct bpf_link *link;
5658         int pfd, err;
5659
5660         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5661                                     0 /* offset */, -1 /* pid */);
5662         if (pfd < 0) {
5663                 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
5664                         bpf_program__title(prog, false),
5665                         retprobe ? "kretprobe" : "kprobe", func_name,
5666                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5667                 return ERR_PTR(pfd);
5668         }
5669         link = bpf_program__attach_perf_event(prog, pfd);
5670         if (IS_ERR(link)) {
5671                 close(pfd);
5672                 err = PTR_ERR(link);
5673                 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
5674                         bpf_program__title(prog, false),
5675                         retprobe ? "kretprobe" : "kprobe", func_name,
5676                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5677                 return link;
5678         }
5679         return link;
5680 }
5681
5682 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5683                                             bool retprobe, pid_t pid,
5684                                             const char *binary_path,
5685                                             size_t func_offset)
5686 {
5687         char errmsg[STRERR_BUFSIZE];
5688         struct bpf_link *link;
5689         int pfd, err;
5690
5691         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5692                                     binary_path, func_offset, pid);
5693         if (pfd < 0) {
5694                 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5695                         bpf_program__title(prog, false),
5696                         retprobe ? "uretprobe" : "uprobe",
5697                         binary_path, func_offset,
5698                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5699                 return ERR_PTR(pfd);
5700         }
5701         link = bpf_program__attach_perf_event(prog, pfd);
5702         if (IS_ERR(link)) {
5703                 close(pfd);
5704                 err = PTR_ERR(link);
5705                 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5706                         bpf_program__title(prog, false),
5707                         retprobe ? "uretprobe" : "uprobe",
5708                         binary_path, func_offset,
5709                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5710                 return link;
5711         }
5712         return link;
5713 }
5714
5715 static int determine_tracepoint_id(const char *tp_category,
5716                                    const char *tp_name)
5717 {
5718         char file[PATH_MAX];
5719         int ret;
5720
5721         ret = snprintf(file, sizeof(file),
5722                        "/sys/kernel/debug/tracing/events/%s/%s/id",
5723                        tp_category, tp_name);
5724         if (ret < 0)
5725                 return -errno;
5726         if (ret >= sizeof(file)) {
5727                 pr_debug("tracepoint %s/%s path is too long\n",
5728                          tp_category, tp_name);
5729                 return -E2BIG;
5730         }
5731         return parse_uint_from_file(file, "%d\n");
5732 }
5733
5734 static int perf_event_open_tracepoint(const char *tp_category,
5735                                       const char *tp_name)
5736 {
5737         struct perf_event_attr attr = {};
5738         char errmsg[STRERR_BUFSIZE];
5739         int tp_id, pfd, err;
5740
5741         tp_id = determine_tracepoint_id(tp_category, tp_name);
5742         if (tp_id < 0) {
5743                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5744                         tp_category, tp_name,
5745                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5746                 return tp_id;
5747         }
5748
5749         attr.type = PERF_TYPE_TRACEPOINT;
5750         attr.size = sizeof(attr);
5751         attr.config = tp_id;
5752
5753         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5754                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5755         if (pfd < 0) {
5756                 err = -errno;
5757                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5758                         tp_category, tp_name,
5759                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5760                 return err;
5761         }
5762         return pfd;
5763 }
5764
5765 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5766                                                 const char *tp_category,
5767                                                 const char *tp_name)
5768 {
5769         char errmsg[STRERR_BUFSIZE];
5770         struct bpf_link *link;
5771         int pfd, err;
5772
5773         pfd = perf_event_open_tracepoint(tp_category, tp_name);
5774         if (pfd < 0) {
5775                 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5776                         bpf_program__title(prog, false),
5777                         tp_category, tp_name,
5778                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5779                 return ERR_PTR(pfd);
5780         }
5781         link = bpf_program__attach_perf_event(prog, pfd);
5782         if (IS_ERR(link)) {
5783                 close(pfd);
5784                 err = PTR_ERR(link);
5785                 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5786                         bpf_program__title(prog, false),
5787                         tp_category, tp_name,
5788                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5789                 return link;
5790         }
5791         return link;
5792 }
5793
5794 static int bpf_link__destroy_fd(struct bpf_link *link)
5795 {
5796         struct bpf_link_fd *l = (void *)link;
5797
5798         return close(l->fd);
5799 }
5800
5801 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5802                                                     const char *tp_name)
5803 {
5804         char errmsg[STRERR_BUFSIZE];
5805         struct bpf_link_fd *link;
5806         int prog_fd, pfd;
5807
5808         prog_fd = bpf_program__fd(prog);
5809         if (prog_fd < 0) {
5810                 pr_warn("program '%s': can't attach before loaded\n",
5811                         bpf_program__title(prog, false));
5812                 return ERR_PTR(-EINVAL);
5813         }
5814
5815         link = malloc(sizeof(*link));
5816         if (!link)
5817                 return ERR_PTR(-ENOMEM);
5818         link->link.destroy = &bpf_link__destroy_fd;
5819
5820         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5821         if (pfd < 0) {
5822                 pfd = -errno;
5823                 free(link);
5824                 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5825                         bpf_program__title(prog, false), tp_name,
5826                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5827                 return ERR_PTR(pfd);
5828         }
5829         link->fd = pfd;
5830         return (struct bpf_link *)link;
5831 }
5832
5833 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
5834 {
5835         char errmsg[STRERR_BUFSIZE];
5836         struct bpf_link_fd *link;
5837         int prog_fd, pfd;
5838
5839         prog_fd = bpf_program__fd(prog);
5840         if (prog_fd < 0) {
5841                 pr_warn("program '%s': can't attach before loaded\n",
5842                         bpf_program__title(prog, false));
5843                 return ERR_PTR(-EINVAL);
5844         }
5845
5846         link = malloc(sizeof(*link));
5847         if (!link)
5848                 return ERR_PTR(-ENOMEM);
5849         link->link.destroy = &bpf_link__destroy_fd;
5850
5851         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
5852         if (pfd < 0) {
5853                 pfd = -errno;
5854                 free(link);
5855                 pr_warn("program '%s': failed to attach to trace: %s\n",
5856                         bpf_program__title(prog, false),
5857                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5858                 return ERR_PTR(pfd);
5859         }
5860         link->fd = pfd;
5861         return (struct bpf_link *)link;
5862 }
5863
5864 enum bpf_perf_event_ret
5865 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5866                            void **copy_mem, size_t *copy_size,
5867                            bpf_perf_event_print_t fn, void *private_data)
5868 {
5869         struct perf_event_mmap_page *header = mmap_mem;
5870         __u64 data_head = ring_buffer_read_head(header);
5871         __u64 data_tail = header->data_tail;
5872         void *base = ((__u8 *)header) + page_size;
5873         int ret = LIBBPF_PERF_EVENT_CONT;
5874         struct perf_event_header *ehdr;
5875         size_t ehdr_size;
5876
5877         while (data_head != data_tail) {
5878                 ehdr = base + (data_tail & (mmap_size - 1));
5879                 ehdr_size = ehdr->size;
5880
5881                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5882                         void *copy_start = ehdr;
5883                         size_t len_first = base + mmap_size - copy_start;
5884                         size_t len_secnd = ehdr_size - len_first;
5885
5886                         if (*copy_size < ehdr_size) {
5887                                 free(*copy_mem);
5888                                 *copy_mem = malloc(ehdr_size);
5889                                 if (!*copy_mem) {
5890                                         *copy_size = 0;
5891                                         ret = LIBBPF_PERF_EVENT_ERROR;
5892                                         break;
5893                                 }
5894                                 *copy_size = ehdr_size;
5895                         }
5896
5897                         memcpy(*copy_mem, copy_start, len_first);
5898                         memcpy(*copy_mem + len_first, base, len_secnd);
5899                         ehdr = *copy_mem;
5900                 }
5901
5902                 ret = fn(ehdr, private_data);
5903                 data_tail += ehdr_size;
5904                 if (ret != LIBBPF_PERF_EVENT_CONT)
5905                         break;
5906         }
5907
5908         ring_buffer_write_tail(header, data_tail);
5909         return ret;
5910 }
5911
5912 struct perf_buffer;
5913
5914 struct perf_buffer_params {
5915         struct perf_event_attr *attr;
5916         /* if event_cb is specified, it takes precendence */
5917         perf_buffer_event_fn event_cb;
5918         /* sample_cb and lost_cb are higher-level common-case callbacks */
5919         perf_buffer_sample_fn sample_cb;
5920         perf_buffer_lost_fn lost_cb;
5921         void *ctx;
5922         int cpu_cnt;
5923         int *cpus;
5924         int *map_keys;
5925 };
5926
5927 struct perf_cpu_buf {
5928         struct perf_buffer *pb;
5929         void *base; /* mmap()'ed memory */
5930         void *buf; /* for reconstructing segmented data */
5931         size_t buf_size;
5932         int fd;
5933         int cpu;
5934         int map_key;
5935 };
5936
5937 struct perf_buffer {
5938         perf_buffer_event_fn event_cb;
5939         perf_buffer_sample_fn sample_cb;
5940         perf_buffer_lost_fn lost_cb;
5941         void *ctx; /* passed into callbacks */
5942
5943         size_t page_size;
5944         size_t mmap_size;
5945         struct perf_cpu_buf **cpu_bufs;
5946         struct epoll_event *events;
5947         int cpu_cnt;
5948         int epoll_fd; /* perf event FD */
5949         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5950 };
5951
5952 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5953                                       struct perf_cpu_buf *cpu_buf)
5954 {
5955         if (!cpu_buf)
5956                 return;
5957         if (cpu_buf->base &&
5958             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5959                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5960         if (cpu_buf->fd >= 0) {
5961                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5962                 close(cpu_buf->fd);
5963         }
5964         free(cpu_buf->buf);
5965         free(cpu_buf);
5966 }
5967
5968 void perf_buffer__free(struct perf_buffer *pb)
5969 {
5970         int i;
5971
5972         if (!pb)
5973                 return;
5974         if (pb->cpu_bufs) {
5975                 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5976                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5977
5978                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5979                         perf_buffer__free_cpu_buf(pb, cpu_buf);
5980                 }
5981                 free(pb->cpu_bufs);
5982         }
5983         if (pb->epoll_fd >= 0)
5984                 close(pb->epoll_fd);
5985         free(pb->events);
5986         free(pb);
5987 }
5988
5989 static struct perf_cpu_buf *
5990 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5991                           int cpu, int map_key)
5992 {
5993         struct perf_cpu_buf *cpu_buf;
5994         char msg[STRERR_BUFSIZE];
5995         int err;
5996
5997         cpu_buf = calloc(1, sizeof(*cpu_buf));
5998         if (!cpu_buf)
5999                 return ERR_PTR(-ENOMEM);
6000
6001         cpu_buf->pb = pb;
6002         cpu_buf->cpu = cpu;
6003         cpu_buf->map_key = map_key;
6004
6005         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
6006                               -1, PERF_FLAG_FD_CLOEXEC);
6007         if (cpu_buf->fd < 0) {
6008                 err = -errno;
6009                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
6010                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6011                 goto error;
6012         }
6013
6014         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
6015                              PROT_READ | PROT_WRITE, MAP_SHARED,
6016                              cpu_buf->fd, 0);
6017         if (cpu_buf->base == MAP_FAILED) {
6018                 cpu_buf->base = NULL;
6019                 err = -errno;
6020                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
6021                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6022                 goto error;
6023         }
6024
6025         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
6026                 err = -errno;
6027                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
6028                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6029                 goto error;
6030         }
6031
6032         return cpu_buf;
6033
6034 error:
6035         perf_buffer__free_cpu_buf(pb, cpu_buf);
6036         return (struct perf_cpu_buf *)ERR_PTR(err);
6037 }
6038
6039 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
6040                                               struct perf_buffer_params *p);
6041
6042 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
6043                                      const struct perf_buffer_opts *opts)
6044 {
6045         struct perf_buffer_params p = {};
6046         struct perf_event_attr attr = { 0, };
6047
6048         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
6049         attr.type = PERF_TYPE_SOFTWARE;
6050         attr.sample_type = PERF_SAMPLE_RAW;
6051         attr.sample_period = 1;
6052         attr.wakeup_events = 1;
6053
6054         p.attr = &attr;
6055         p.sample_cb = opts ? opts->sample_cb : NULL;
6056         p.lost_cb = opts ? opts->lost_cb : NULL;
6057         p.ctx = opts ? opts->ctx : NULL;
6058
6059         return __perf_buffer__new(map_fd, page_cnt, &p);
6060 }
6061
6062 struct perf_buffer *
6063 perf_buffer__new_raw(int map_fd, size_t page_cnt,
6064                      const struct perf_buffer_raw_opts *opts)
6065 {
6066         struct perf_buffer_params p = {};
6067
6068         p.attr = opts->attr;
6069         p.event_cb = opts->event_cb;
6070         p.ctx = opts->ctx;
6071         p.cpu_cnt = opts->cpu_cnt;
6072         p.cpus = opts->cpus;
6073         p.map_keys = opts->map_keys;
6074
6075         return __perf_buffer__new(map_fd, page_cnt, &p);
6076 }
6077
6078 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
6079                                               struct perf_buffer_params *p)
6080 {
6081         struct bpf_map_info map = {};
6082         char msg[STRERR_BUFSIZE];
6083         struct perf_buffer *pb;
6084         __u32 map_info_len;
6085         int err, i;
6086
6087         if (page_cnt & (page_cnt - 1)) {
6088                 pr_warn("page count should be power of two, but is %zu\n",
6089                         page_cnt);
6090                 return ERR_PTR(-EINVAL);
6091         }
6092
6093         map_info_len = sizeof(map);
6094         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
6095         if (err) {
6096                 err = -errno;
6097                 pr_warn("failed to get map info for map FD %d: %s\n",
6098                         map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
6099                 return ERR_PTR(err);
6100         }
6101
6102         if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
6103                 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
6104                         map.name);
6105                 return ERR_PTR(-EINVAL);
6106         }
6107
6108         pb = calloc(1, sizeof(*pb));
6109         if (!pb)
6110                 return ERR_PTR(-ENOMEM);
6111
6112         pb->event_cb = p->event_cb;
6113         pb->sample_cb = p->sample_cb;
6114         pb->lost_cb = p->lost_cb;
6115         pb->ctx = p->ctx;
6116
6117         pb->page_size = getpagesize();
6118         pb->mmap_size = pb->page_size * page_cnt;
6119         pb->map_fd = map_fd;
6120
6121         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
6122         if (pb->epoll_fd < 0) {
6123                 err = -errno;
6124                 pr_warn("failed to create epoll instance: %s\n",
6125                         libbpf_strerror_r(err, msg, sizeof(msg)));
6126                 goto error;
6127         }
6128
6129         if (p->cpu_cnt > 0) {
6130                 pb->cpu_cnt = p->cpu_cnt;
6131         } else {
6132                 pb->cpu_cnt = libbpf_num_possible_cpus();
6133                 if (pb->cpu_cnt < 0) {
6134                         err = pb->cpu_cnt;
6135                         goto error;
6136                 }
6137                 if (map.max_entries < pb->cpu_cnt)
6138                         pb->cpu_cnt = map.max_entries;
6139         }
6140
6141         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
6142         if (!pb->events) {
6143                 err = -ENOMEM;
6144                 pr_warn("failed to allocate events: out of memory\n");
6145                 goto error;
6146         }
6147         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
6148         if (!pb->cpu_bufs) {
6149                 err = -ENOMEM;
6150                 pr_warn("failed to allocate buffers: out of memory\n");
6151                 goto error;
6152         }
6153
6154         for (i = 0; i < pb->cpu_cnt; i++) {
6155                 struct perf_cpu_buf *cpu_buf;
6156                 int cpu, map_key;
6157
6158                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
6159                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
6160
6161                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
6162                 if (IS_ERR(cpu_buf)) {
6163                         err = PTR_ERR(cpu_buf);
6164                         goto error;
6165                 }
6166
6167                 pb->cpu_bufs[i] = cpu_buf;
6168
6169                 err = bpf_map_update_elem(pb->map_fd, &map_key,
6170                                           &cpu_buf->fd, 0);
6171                 if (err) {
6172                         err = -errno;
6173                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
6174                                 cpu, map_key, cpu_buf->fd,
6175                                 libbpf_strerror_r(err, msg, sizeof(msg)));
6176                         goto error;
6177                 }
6178
6179                 pb->events[i].events = EPOLLIN;
6180                 pb->events[i].data.ptr = cpu_buf;
6181                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
6182                               &pb->events[i]) < 0) {
6183                         err = -errno;
6184                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
6185                                 cpu, cpu_buf->fd,
6186                                 libbpf_strerror_r(err, msg, sizeof(msg)));
6187                         goto error;
6188                 }
6189         }
6190
6191         return pb;
6192
6193 error:
6194         if (pb)
6195                 perf_buffer__free(pb);
6196         return ERR_PTR(err);
6197 }
6198
6199 struct perf_sample_raw {
6200         struct perf_event_header header;
6201         uint32_t size;
6202         char data[0];
6203 };
6204
6205 struct perf_sample_lost {
6206         struct perf_event_header header;
6207         uint64_t id;
6208         uint64_t lost;
6209         uint64_t sample_id;
6210 };
6211
6212 static enum bpf_perf_event_ret
6213 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
6214 {
6215         struct perf_cpu_buf *cpu_buf = ctx;
6216         struct perf_buffer *pb = cpu_buf->pb;
6217         void *data = e;
6218
6219         /* user wants full control over parsing perf event */
6220         if (pb->event_cb)
6221                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
6222
6223         switch (e->type) {
6224         case PERF_RECORD_SAMPLE: {
6225                 struct perf_sample_raw *s = data;
6226
6227                 if (pb->sample_cb)
6228                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
6229                 break;
6230         }
6231         case PERF_RECORD_LOST: {
6232                 struct perf_sample_lost *s = data;
6233
6234                 if (pb->lost_cb)
6235                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
6236                 break;
6237         }
6238         default:
6239                 pr_warn("unknown perf sample type %d\n", e->type);
6240                 return LIBBPF_PERF_EVENT_ERROR;
6241         }
6242         return LIBBPF_PERF_EVENT_CONT;
6243 }
6244
6245 static int perf_buffer__process_records(struct perf_buffer *pb,
6246                                         struct perf_cpu_buf *cpu_buf)
6247 {
6248         enum bpf_perf_event_ret ret;
6249
6250         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
6251                                          pb->page_size, &cpu_buf->buf,
6252                                          &cpu_buf->buf_size,
6253                                          perf_buffer__process_record, cpu_buf);
6254         if (ret != LIBBPF_PERF_EVENT_CONT)
6255                 return ret;
6256         return 0;
6257 }
6258
6259 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
6260 {
6261         int i, cnt, err;
6262
6263         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
6264         for (i = 0; i < cnt; i++) {
6265                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
6266
6267                 err = perf_buffer__process_records(pb, cpu_buf);
6268                 if (err) {
6269                         pr_warn("error while processing records: %d\n", err);
6270                         return err;
6271                 }
6272         }
6273         return cnt < 0 ? -errno : cnt;
6274 }
6275
6276 struct bpf_prog_info_array_desc {
6277         int     array_offset;   /* e.g. offset of jited_prog_insns */
6278         int     count_offset;   /* e.g. offset of jited_prog_len */
6279         int     size_offset;    /* > 0: offset of rec size,
6280                                  * < 0: fix size of -size_offset
6281                                  */
6282 };
6283
6284 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
6285         [BPF_PROG_INFO_JITED_INSNS] = {
6286                 offsetof(struct bpf_prog_info, jited_prog_insns),
6287                 offsetof(struct bpf_prog_info, jited_prog_len),
6288                 -1,
6289         },
6290         [BPF_PROG_INFO_XLATED_INSNS] = {
6291                 offsetof(struct bpf_prog_info, xlated_prog_insns),
6292                 offsetof(struct bpf_prog_info, xlated_prog_len),
6293                 -1,
6294         },
6295         [BPF_PROG_INFO_MAP_IDS] = {
6296                 offsetof(struct bpf_prog_info, map_ids),
6297                 offsetof(struct bpf_prog_info, nr_map_ids),
6298                 -(int)sizeof(__u32),
6299         },
6300         [BPF_PROG_INFO_JITED_KSYMS] = {
6301                 offsetof(struct bpf_prog_info, jited_ksyms),
6302                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
6303                 -(int)sizeof(__u64),
6304         },
6305         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
6306                 offsetof(struct bpf_prog_info, jited_func_lens),
6307                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
6308                 -(int)sizeof(__u32),
6309         },
6310         [BPF_PROG_INFO_FUNC_INFO] = {
6311                 offsetof(struct bpf_prog_info, func_info),
6312                 offsetof(struct bpf_prog_info, nr_func_info),
6313                 offsetof(struct bpf_prog_info, func_info_rec_size),
6314         },
6315         [BPF_PROG_INFO_LINE_INFO] = {
6316                 offsetof(struct bpf_prog_info, line_info),
6317                 offsetof(struct bpf_prog_info, nr_line_info),
6318                 offsetof(struct bpf_prog_info, line_info_rec_size),
6319         },
6320         [BPF_PROG_INFO_JITED_LINE_INFO] = {
6321                 offsetof(struct bpf_prog_info, jited_line_info),
6322                 offsetof(struct bpf_prog_info, nr_jited_line_info),
6323                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
6324         },
6325         [BPF_PROG_INFO_PROG_TAGS] = {
6326                 offsetof(struct bpf_prog_info, prog_tags),
6327                 offsetof(struct bpf_prog_info, nr_prog_tags),
6328                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
6329         },
6330
6331 };
6332
6333 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
6334                                            int offset)
6335 {
6336         __u32 *array = (__u32 *)info;
6337
6338         if (offset >= 0)
6339                 return array[offset / sizeof(__u32)];
6340         return -(int)offset;
6341 }
6342
6343 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
6344                                            int offset)
6345 {
6346         __u64 *array = (__u64 *)info;
6347
6348         if (offset >= 0)
6349                 return array[offset / sizeof(__u64)];
6350         return -(int)offset;
6351 }
6352
6353 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
6354                                          __u32 val)
6355 {
6356         __u32 *array = (__u32 *)info;
6357
6358         if (offset >= 0)
6359                 array[offset / sizeof(__u32)] = val;
6360 }
6361
6362 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
6363                                          __u64 val)
6364 {
6365         __u64 *array = (__u64 *)info;
6366
6367         if (offset >= 0)
6368                 array[offset / sizeof(__u64)] = val;
6369 }
6370
6371 struct bpf_prog_info_linear *
6372 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
6373 {
6374         struct bpf_prog_info_linear *info_linear;
6375         struct bpf_prog_info info = {};
6376         __u32 info_len = sizeof(info);
6377         __u32 data_len = 0;
6378         int i, err;
6379         void *ptr;
6380
6381         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
6382                 return ERR_PTR(-EINVAL);
6383
6384         /* step 1: get array dimensions */
6385         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
6386         if (err) {
6387                 pr_debug("can't get prog info: %s", strerror(errno));
6388                 return ERR_PTR(-EFAULT);
6389         }
6390
6391         /* step 2: calculate total size of all arrays */
6392         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
6393                 bool include_array = (arrays & (1UL << i)) > 0;
6394                 struct bpf_prog_info_array_desc *desc;
6395                 __u32 count, size;
6396
6397                 desc = bpf_prog_info_array_desc + i;
6398
6399                 /* kernel is too old to support this field */
6400                 if (info_len < desc->array_offset + sizeof(__u32) ||
6401                     info_len < desc->count_offset + sizeof(__u32) ||
6402                     (desc->size_offset > 0 && info_len < desc->size_offset))
6403                         include_array = false;
6404
6405                 if (!include_array) {
6406                         arrays &= ~(1UL << i);  /* clear the bit */
6407                         continue;
6408                 }
6409
6410                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
6411                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
6412
6413                 data_len += count * size;
6414         }
6415
6416         /* step 3: allocate continuous memory */
6417         data_len = roundup(data_len, sizeof(__u64));
6418         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
6419         if (!info_linear)
6420                 return ERR_PTR(-ENOMEM);
6421
6422         /* step 4: fill data to info_linear->info */
6423         info_linear->arrays = arrays;
6424         memset(&info_linear->info, 0, sizeof(info));
6425         ptr = info_linear->data;
6426
6427         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
6428                 struct bpf_prog_info_array_desc *desc;
6429                 __u32 count, size;
6430
6431                 if ((arrays & (1UL << i)) == 0)
6432                         continue;
6433
6434                 desc  = bpf_prog_info_array_desc + i;
6435                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
6436                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
6437                 bpf_prog_info_set_offset_u32(&info_linear->info,
6438                                              desc->count_offset, count);
6439                 bpf_prog_info_set_offset_u32(&info_linear->info,
6440                                              desc->size_offset, size);
6441                 bpf_prog_info_set_offset_u64(&info_linear->info,
6442                                              desc->array_offset,
6443                                              ptr_to_u64(ptr));
6444                 ptr += count * size;
6445         }
6446
6447         /* step 5: call syscall again to get required arrays */
6448         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
6449         if (err) {
6450                 pr_debug("can't get prog info: %s", strerror(errno));
6451                 free(info_linear);
6452                 return ERR_PTR(-EFAULT);
6453         }
6454
6455         /* step 6: verify the data */
6456         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
6457                 struct bpf_prog_info_array_desc *desc;
6458                 __u32 v1, v2;
6459
6460                 if ((arrays & (1UL << i)) == 0)
6461                         continue;
6462
6463                 desc = bpf_prog_info_array_desc + i;
6464                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
6465                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
6466                                                    desc->count_offset);
6467                 if (v1 != v2)
6468                         pr_warn("%s: mismatch in element count\n", __func__);
6469
6470                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
6471                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
6472                                                    desc->size_offset);
6473                 if (v1 != v2)
6474                         pr_warn("%s: mismatch in rec size\n", __func__);
6475         }
6476
6477         /* step 7: update info_len and data_len */
6478         info_linear->info_len = sizeof(struct bpf_prog_info);
6479         info_linear->data_len = data_len;
6480
6481         return info_linear;
6482 }
6483
6484 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
6485 {
6486         int i;
6487
6488         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
6489                 struct bpf_prog_info_array_desc *desc;
6490                 __u64 addr, offs;
6491
6492                 if ((info_linear->arrays & (1UL << i)) == 0)
6493                         continue;
6494
6495                 desc = bpf_prog_info_array_desc + i;
6496                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
6497                                                      desc->array_offset);
6498                 offs = addr - ptr_to_u64(info_linear->data);
6499                 bpf_prog_info_set_offset_u64(&info_linear->info,
6500                                              desc->array_offset, offs);
6501         }
6502 }
6503
6504 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
6505 {
6506         int i;
6507
6508         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
6509                 struct bpf_prog_info_array_desc *desc;
6510                 __u64 addr, offs;
6511
6512                 if ((info_linear->arrays & (1UL << i)) == 0)
6513                         continue;
6514
6515                 desc = bpf_prog_info_array_desc + i;
6516                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
6517                                                      desc->array_offset);
6518                 addr = offs + ptr_to_u64(info_linear->data);
6519                 bpf_prog_info_set_offset_u64(&info_linear->info,
6520                                              desc->array_offset, addr);
6521         }
6522 }
6523
6524 int libbpf_num_possible_cpus(void)
6525 {
6526         static const char *fcpu = "/sys/devices/system/cpu/possible";
6527         int len = 0, n = 0, il = 0, ir = 0;
6528         unsigned int start = 0, end = 0;
6529         int tmp_cpus = 0;
6530         static int cpus;
6531         char buf[128];
6532         int error = 0;
6533         int fd = -1;
6534
6535         tmp_cpus = READ_ONCE(cpus);
6536         if (tmp_cpus > 0)
6537                 return tmp_cpus;
6538
6539         fd = open(fcpu, O_RDONLY);
6540         if (fd < 0) {
6541                 error = errno;
6542                 pr_warn("Failed to open file %s: %s\n", fcpu, strerror(error));
6543                 return -error;
6544         }
6545         len = read(fd, buf, sizeof(buf));
6546         close(fd);
6547         if (len <= 0) {
6548                 error = len ? errno : EINVAL;
6549                 pr_warn("Failed to read # of possible cpus from %s: %s\n",
6550                         fcpu, strerror(error));
6551                 return -error;
6552         }
6553         if (len == sizeof(buf)) {
6554                 pr_warn("File %s size overflow\n", fcpu);
6555                 return -EOVERFLOW;
6556         }
6557         buf[len] = '\0';
6558
6559         for (ir = 0, tmp_cpus = 0; ir <= len; ir++) {
6560                 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
6561                 if (buf[ir] == ',' || buf[ir] == '\0') {
6562                         buf[ir] = '\0';
6563                         n = sscanf(&buf[il], "%u-%u", &start, &end);
6564                         if (n <= 0) {
6565                                 pr_warn("Failed to get # CPUs from %s\n",
6566                                         &buf[il]);
6567                                 return -EINVAL;
6568                         } else if (n == 1) {
6569                                 end = start;
6570                         }
6571                         tmp_cpus += end - start + 1;
6572                         il = ir + 1;
6573                 }
6574         }
6575         if (tmp_cpus <= 0) {
6576                 pr_warn("Invalid #CPUs %d from %s\n", tmp_cpus, fcpu);
6577                 return -EINVAL;
6578         }
6579
6580         WRITE_ONCE(cpus, tmp_cpus);
6581         return tmp_cpus;
6582 }