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 <sys/epoll.h>
  37 #include <sys/ioctl.h>
  38 #include <sys/mman.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/vfs.h>
  42 #include <tools/libc_compat.h>
  43 #include <libelf.h>
  44 #include <gelf.h>
  45
  46 #include "libbpf.h"
  47 #include "bpf.h"
  48 #include "btf.h"
  49 #include "str_error.h"
  50 #include "libbpf_internal.h"
  51
  52 #ifndef EM_BPF
  53 #define EM_BPF 247
  54 #endif
  55
  56 #ifndef BPF_FS_MAGIC
  57 #define BPF_FS_MAGIC            0xcafe4a11
  58 #endif
  59
  60 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  61  * compilation if user enables corresponding warning. Disable it explicitly.
  62  */
  63 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  64
  65 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  66
  67 static int __base_pr(enum libbpf_print_level level, const char *format,
  68                      va_list args)
  69 {
  70         if (level == LIBBPF_DEBUG)
  71                 return 0;
  72
  73         return vfprintf(stderr, format, args);
  74 }
  75
  76 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  77
  78 void libbpf_set_print(libbpf_print_fn_t fn)
  79 {
  80         __libbpf_pr = fn;
  81 }
  82
  83 __printf(2, 3)
  84 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  85 {
  86         va_list args;
  87
  88         if (!__libbpf_pr)
  89                 return;
  90
  91         va_start(args, format);
  92         __libbpf_pr(level, format, args);
  93         va_end(args);
  94 }
  95
  96 #define STRERR_BUFSIZE  128
  97
  98 #define CHECK_ERR(action, err, out) do {        \
  99         err = action;                   \
 100         if (err)                        \
 101                 goto out;               \
 102 } while(0)
 103
 104
 105 /* Copied from tools/perf/util/util.h */
 106 #ifndef zfree
 107 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 108 #endif
 109
 110 #ifndef zclose
 111 # define zclose(fd) ({                  \
 112         int ___err = 0;                 \
 113         if ((fd) >= 0)                  \
 114                 ___err = close((fd));   \
 115         fd = -1;                        \
 116         ___err; })
 117 #endif
 118
 119 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 120 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 121 #else
 122 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 123 #endif
 124
 125 static inline __u64 ptr_to_u64(const void *ptr)
 126 {
 127         return (__u64) (unsigned long) ptr;
 128 }
 129
 130 struct bpf_capabilities {
 131         /* v4.14: kernel support for program & map names. */
 132         __u32 name:1;
 133         /* v5.2: kernel support for global data sections. */
 134         __u32 global_data:1;
 135         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 136         __u32 btf_func:1;
 137         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 138         __u32 btf_datasec:1;
 139 };
 140
 141 /*
 142  * bpf_prog should be a better name but it has been used in
 143  * linux/filter.h.
 144  */
 145 struct bpf_program {
 146         /* Index in elf obj file, for relocation use. */
 147         int idx;
 148         char *name;
 149         int prog_ifindex;
 150         char *section_name;
 151         /* section_name with / replaced by _; makes recursive pinning
 152          * in bpf_object__pin_programs easier
 153          */
 154         char *pin_name;
 155         struct bpf_insn *insns;
 156         size_t insns_cnt, main_prog_cnt;
 157         enum bpf_prog_type type;
 158
 159         struct reloc_desc {
 160                 enum {
 161                         RELO_LD64,
 162                         RELO_CALL,
 163                         RELO_DATA,
 164                 } type;
 165                 int insn_idx;
 166                 union {
 167                         int map_idx;
 168                         int text_off;
 169                 };
 170         } *reloc_desc;
 171         int nr_reloc;
 172         int log_level;
 173
 174         struct {
 175                 int nr;
 176                 int *fds;
 177         } instances;
 178         bpf_program_prep_t preprocessor;
 179
 180         struct bpf_object *obj;
 181         void *priv;
 182         bpf_program_clear_priv_t clear_priv;
 183
 184         enum bpf_attach_type expected_attach_type;
 185         void *func_info;
 186         __u32 func_info_rec_size;
 187         __u32 func_info_cnt;
 188
 189         struct bpf_capabilities *caps;
 190
 191         void *line_info;
 192         __u32 line_info_rec_size;
 193         __u32 line_info_cnt;
 194         __u32 prog_flags;
 195 };
 196
 197 enum libbpf_map_type {
 198         LIBBPF_MAP_UNSPEC,
 199         LIBBPF_MAP_DATA,
 200         LIBBPF_MAP_BSS,
 201         LIBBPF_MAP_RODATA,
 202 };
 203
 204 static const char * const libbpf_type_to_btf_name[] = {
 205         [LIBBPF_MAP_DATA]       = ".data",
 206         [LIBBPF_MAP_BSS]        = ".bss",
 207         [LIBBPF_MAP_RODATA]     = ".rodata",
 208 };
 209
 210 struct bpf_map {
 211         int fd;
 212         char *name;
 213         int sec_idx;
 214         size_t sec_offset;
 215         int map_ifindex;
 216         int inner_map_fd;
 217         struct bpf_map_def def;
 218         __u32 btf_key_type_id;
 219         __u32 btf_value_type_id;
 220         void *priv;
 221         bpf_map_clear_priv_t clear_priv;
 222         enum libbpf_map_type libbpf_type;
 223 };
 224
 225 struct bpf_secdata {
 226         void *rodata;
 227         void *data;
 228 };
 229
 230 static LIST_HEAD(bpf_objects_list);
 231
 232 struct bpf_object {
 233         char name[BPF_OBJ_NAME_LEN];
 234         char license[64];
 235         __u32 kern_version;
 236
 237         struct bpf_program *programs;
 238         size_t nr_programs;
 239         struct bpf_map *maps;
 240         size_t nr_maps;
 241         size_t maps_cap;
 242         struct bpf_secdata sections;
 243
 244         bool loaded;
 245         bool has_pseudo_calls;
 246
 247         /*
 248          * Information when doing elf related work. Only valid if fd
 249          * is valid.
 250          */
 251         struct {
 252                 int fd;
 253                 void *obj_buf;
 254                 size_t obj_buf_sz;
 255                 Elf *elf;
 256                 GElf_Ehdr ehdr;
 257                 Elf_Data *symbols;
 258                 Elf_Data *data;
 259                 Elf_Data *rodata;
 260                 Elf_Data *bss;
 261                 size_t strtabidx;
 262                 struct {
 263                         GElf_Shdr shdr;
 264                         Elf_Data *data;
 265                 } *reloc;
 266                 int nr_reloc;
 267                 int maps_shndx;
 268                 int btf_maps_shndx;
 269                 int text_shndx;
 270                 int data_shndx;
 271                 int rodata_shndx;
 272                 int bss_shndx;
 273         } efile;
 274         /*
 275          * All loaded bpf_object is linked in a list, which is
 276          * hidden to caller. bpf_objects__<func> handlers deal with
 277          * all objects.
 278          */
 279         struct list_head list;
 280
 281         struct btf *btf;
 282         struct btf_ext *btf_ext;
 283
 284         void *priv;
 285         bpf_object_clear_priv_t clear_priv;
 286
 287         struct bpf_capabilities caps;
 288
 289         char path[];
 290 };
 291 #define obj_elf_valid(o)        ((o)->efile.elf)
 292
 293 void bpf_program__unload(struct bpf_program *prog)
 294 {
 295         int i;
 296
 297         if (!prog)
 298                 return;
 299
 300         /*
 301          * If the object is opened but the program was never loaded,
 302          * it is possible that prog->instances.nr == -1.
 303          */
 304         if (prog->instances.nr > 0) {
 305                 for (i = 0; i < prog->instances.nr; i++)
 306                         zclose(prog->instances.fds[i]);
 307         } else if (prog->instances.nr != -1) {
 308                 pr_warning("Internal error: instances.nr is %d\n",
 309                            prog->instances.nr);
 310         }
 311
 312         prog->instances.nr = -1;
 313         zfree(&prog->instances.fds);
 314
 315         zfree(&prog->func_info);
 316         zfree(&prog->line_info);
 317 }
 318
 319 static void bpf_program__exit(struct bpf_program *prog)
 320 {
 321         if (!prog)
 322                 return;
 323
 324         if (prog->clear_priv)
 325                 prog->clear_priv(prog, prog->priv);
 326
 327         prog->priv = NULL;
 328         prog->clear_priv = NULL;
 329
 330         bpf_program__unload(prog);
 331         zfree(&prog->name);
 332         zfree(&prog->section_name);
 333         zfree(&prog->pin_name);
 334         zfree(&prog->insns);
 335         zfree(&prog->reloc_desc);
 336
 337         prog->nr_reloc = 0;
 338         prog->insns_cnt = 0;
 339         prog->idx = -1;
 340 }
 341
 342 static char *__bpf_program__pin_name(struct bpf_program *prog)
 343 {
 344         char *name, *p;
 345
 346         name = p = strdup(prog->section_name);
 347         while ((p = strchr(p, '/')))
 348                 *p = '_';
 349
 350         return name;
 351 }
 352
 353 static int
 354 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 355                   struct bpf_program *prog)
 356 {
 357         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 358
 359         if (size == 0 || size % bpf_insn_sz) {
 360                 pr_warning("corrupted section '%s', size: %zu\n",
 361                            section_name, size);
 362                 return -EINVAL;
 363         }
 364
 365         memset(prog, 0, sizeof(*prog));
 366
 367         prog->section_name = strdup(section_name);
 368         if (!prog->section_name) {
 369                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 370                            idx, section_name);
 371                 goto errout;
 372         }
 373
 374         prog->pin_name = __bpf_program__pin_name(prog);
 375         if (!prog->pin_name) {
 376                 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
 377                            idx, section_name);
 378                 goto errout;
 379         }
 380
 381         prog->insns = malloc(size);
 382         if (!prog->insns) {
 383                 pr_warning("failed to alloc insns for prog under section %s\n",
 384                            section_name);
 385                 goto errout;
 386         }
 387         prog->insns_cnt = size / bpf_insn_sz;
 388         memcpy(prog->insns, data, size);
 389         prog->idx = idx;
 390         prog->instances.fds = NULL;
 391         prog->instances.nr = -1;
 392         prog->type = BPF_PROG_TYPE_UNSPEC;
 393
 394         return 0;
 395 errout:
 396         bpf_program__exit(prog);
 397         return -ENOMEM;
 398 }
 399
 400 static int
 401 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 402                         char *section_name, int idx)
 403 {
 404         struct bpf_program prog, *progs;
 405         int nr_progs, err;
 406
 407         err = bpf_program__init(data, size, section_name, idx, &prog);
 408         if (err)
 409                 return err;
 410
 411         prog.caps = &obj->caps;
 412         progs = obj->programs;
 413         nr_progs = obj->nr_programs;
 414
 415         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 416         if (!progs) {
 417                 /*
 418                  * In this case the original obj->programs
 419                  * is still valid, so don't need special treat for
 420                  * bpf_close_object().
 421                  */
 422                 pr_warning("failed to alloc a new program under section '%s'\n",
 423                            section_name);
 424                 bpf_program__exit(&prog);
 425                 return -ENOMEM;
 426         }
 427
 428         pr_debug("found program %s\n", prog.section_name);
 429         obj->programs = progs;
 430         obj->nr_programs = nr_progs + 1;
 431         prog.obj = obj;
 432         progs[nr_progs] = prog;
 433         return 0;
 434 }
 435
 436 static int
 437 bpf_object__init_prog_names(struct bpf_object *obj)
 438 {
 439         Elf_Data *symbols = obj->efile.symbols;
 440         struct bpf_program *prog;
 441         size_t pi, si;
 442
 443         for (pi = 0; pi < obj->nr_programs; pi++) {
 444                 const char *name = NULL;
 445
 446                 prog = &obj->programs[pi];
 447
 448                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 449                      si++) {
 450                         GElf_Sym sym;
 451
 452                         if (!gelf_getsym(symbols, si, &sym))
 453                                 continue;
 454                         if (sym.st_shndx != prog->idx)
 455                                 continue;
 456                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 457                                 continue;
 458
 459                         name = elf_strptr(obj->efile.elf,
 460                                           obj->efile.strtabidx,
 461                                           sym.st_name);
 462                         if (!name) {
 463                                 pr_warning("failed to get sym name string for prog %s\n",
 464                                            prog->section_name);
 465                                 return -LIBBPF_ERRNO__LIBELF;
 466                         }
 467                 }
 468
 469                 if (!name && prog->idx == obj->efile.text_shndx)
 470                         name = ".text";
 471
 472                 if (!name) {
 473                         pr_warning("failed to find sym for prog %s\n",
 474                                    prog->section_name);
 475                         return -EINVAL;
 476                 }
 477
 478                 prog->name = strdup(name);
 479                 if (!prog->name) {
 480                         pr_warning("failed to allocate memory for prog sym %s\n",
 481                                    name);
 482                         return -ENOMEM;
 483                 }
 484         }
 485
 486         return 0;
 487 }
 488
 489 static struct bpf_object *bpf_object__new(const char *path,
 490                                           void *obj_buf,
 491                                           size_t obj_buf_sz)
 492 {
 493         struct bpf_object *obj;
 494         char *end;
 495
 496         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 497         if (!obj) {
 498                 pr_warning("alloc memory failed for %s\n", path);
 499                 return ERR_PTR(-ENOMEM);
 500         }
 501
 502         strcpy(obj->path, path);
 503         /* Using basename() GNU version which doesn't modify arg. */
 504         strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
 505         end = strchr(obj->name, '.');
 506         if (end)
 507                 *end = 0;
 508
 509         obj->efile.fd = -1;
 510         /*
 511          * Caller of this function should also call
 512          * bpf_object__elf_finish() after data collection to return
 513          * obj_buf to user. If not, we should duplicate the buffer to
 514          * avoid user freeing them before elf finish.
 515          */
 516         obj->efile.obj_buf = obj_buf;
 517         obj->efile.obj_buf_sz = obj_buf_sz;
 518         obj->efile.maps_shndx = -1;
 519         obj->efile.btf_maps_shndx = -1;
 520         obj->efile.data_shndx = -1;
 521         obj->efile.rodata_shndx = -1;
 522         obj->efile.bss_shndx = -1;
 523
 524         obj->loaded = false;
 525
 526         INIT_LIST_HEAD(&obj->list);
 527         list_add(&obj->list, &bpf_objects_list);
 528         return obj;
 529 }
 530
 531 static void bpf_object__elf_finish(struct bpf_object *obj)
 532 {
 533         if (!obj_elf_valid(obj))
 534                 return;
 535
 536         if (obj->efile.elf) {
 537                 elf_end(obj->efile.elf);
 538                 obj->efile.elf = NULL;
 539         }
 540         obj->efile.symbols = NULL;
 541         obj->efile.data = NULL;
 542         obj->efile.rodata = NULL;
 543         obj->efile.bss = NULL;
 544
 545         zfree(&obj->efile.reloc);
 546         obj->efile.nr_reloc = 0;
 547         zclose(obj->efile.fd);
 548         obj->efile.obj_buf = NULL;
 549         obj->efile.obj_buf_sz = 0;
 550 }
 551
 552 static int bpf_object__elf_init(struct bpf_object *obj)
 553 {
 554         int err = 0;
 555         GElf_Ehdr *ep;
 556
 557         if (obj_elf_valid(obj)) {
 558                 pr_warning("elf init: internal error\n");
 559                 return -LIBBPF_ERRNO__LIBELF;
 560         }
 561
 562         if (obj->efile.obj_buf_sz > 0) {
 563                 /*
 564                  * obj_buf should have been validated by
 565                  * bpf_object__open_buffer().
 566                  */
 567                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 568                                             obj->efile.obj_buf_sz);
 569         } else {
 570                 obj->efile.fd = open(obj->path, O_RDONLY);
 571                 if (obj->efile.fd < 0) {
 572                         char errmsg[STRERR_BUFSIZE], *cp;
 573
 574                         err = -errno;
 575                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 576                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 577                         return err;
 578                 }
 579
 580                 obj->efile.elf = elf_begin(obj->efile.fd,
 581                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 582         }
 583
 584         if (!obj->efile.elf) {
 585                 pr_warning("failed to open %s as ELF file\n", obj->path);
 586                 err = -LIBBPF_ERRNO__LIBELF;
 587                 goto errout;
 588         }
 589
 590         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 591                 pr_warning("failed to get EHDR from %s\n", obj->path);
 592                 err = -LIBBPF_ERRNO__FORMAT;
 593                 goto errout;
 594         }
 595         ep = &obj->efile.ehdr;
 596
 597         /* Old LLVM set e_machine to EM_NONE */
 598         if (ep->e_type != ET_REL ||
 599             (ep->e_machine && ep->e_machine != EM_BPF)) {
 600                 pr_warning("%s is not an eBPF object file\n", obj->path);
 601                 err = -LIBBPF_ERRNO__FORMAT;
 602                 goto errout;
 603         }
 604
 605         return 0;
 606 errout:
 607         bpf_object__elf_finish(obj);
 608         return err;
 609 }
 610
 611 static int bpf_object__check_endianness(struct bpf_object *obj)
 612 {
 613 #if __BYTE_ORDER == __LITTLE_ENDIAN
 614         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 615                 return 0;
 616 #elif __BYTE_ORDER == __BIG_ENDIAN
 617         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 618                 return 0;
 619 #else
 620 # error "Unrecognized __BYTE_ORDER__"
 621 #endif
 622         pr_warning("endianness mismatch.\n");
 623         return -LIBBPF_ERRNO__ENDIAN;
 624 }
 625
 626 static int
 627 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 628 {
 629         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 630         pr_debug("license of %s is %s\n", obj->path, obj->license);
 631         return 0;
 632 }
 633
 634 static int
 635 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 636 {
 637         __u32 kver;
 638
 639         if (size != sizeof(kver)) {
 640                 pr_warning("invalid kver section in %s\n", obj->path);
 641                 return -LIBBPF_ERRNO__FORMAT;
 642         }
 643         memcpy(&kver, data, sizeof(kver));
 644         obj->kern_version = kver;
 645         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 646         return 0;
 647 }
 648
 649 static int compare_bpf_map(const void *_a, const void *_b)
 650 {
 651         const struct bpf_map *a = _a;
 652         const struct bpf_map *b = _b;
 653
 654         if (a->sec_idx != b->sec_idx)
 655                 return a->sec_idx - b->sec_idx;
 656         return a->sec_offset - b->sec_offset;
 657 }
 658
 659 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 660 {
 661         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 662             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 663                 return true;
 664         return false;
 665 }
 666
 667 static int bpf_object_search_section_size(const struct bpf_object *obj,
 668                                           const char *name, size_t *d_size)
 669 {
 670         const GElf_Ehdr *ep = &obj->efile.ehdr;
 671         Elf *elf = obj->efile.elf;
 672         Elf_Scn *scn = NULL;
 673         int idx = 0;
 674
 675         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 676                 const char *sec_name;
 677                 Elf_Data *data;
 678                 GElf_Shdr sh;
 679
 680                 idx++;
 681                 if (gelf_getshdr(scn, &sh) != &sh) {
 682                         pr_warning("failed to get section(%d) header from %s\n",
 683                                    idx, obj->path);
 684                         return -EIO;
 685                 }
 686
 687                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 688                 if (!sec_name) {
 689                         pr_warning("failed to get section(%d) name from %s\n",
 690                                    idx, obj->path);
 691                         return -EIO;
 692                 }
 693
 694                 if (strcmp(name, sec_name))
 695                         continue;
 696
 697                 data = elf_getdata(scn, 0);
 698                 if (!data) {
 699                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 700                                    idx, name, obj->path);
 701                         return -EIO;
 702                 }
 703
 704                 *d_size = data->d_size;
 705                 return 0;
 706         }
 707
 708         return -ENOENT;
 709 }
 710
 711 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 712                              __u32 *size)
 713 {
 714         int ret = -ENOENT;
 715         size_t d_size;
 716
 717         *size = 0;
 718         if (!name) {
 719                 return -EINVAL;
 720         } else if (!strcmp(name, ".data")) {
 721                 if (obj->efile.data)
 722                         *size = obj->efile.data->d_size;
 723         } else if (!strcmp(name, ".bss")) {
 724                 if (obj->efile.bss)
 725                         *size = obj->efile.bss->d_size;
 726         } else if (!strcmp(name, ".rodata")) {
 727                 if (obj->efile.rodata)
 728                         *size = obj->efile.rodata->d_size;
 729         } else {
 730                 ret = bpf_object_search_section_size(obj, name, &d_size);
 731                 if (!ret)
 732                         *size = d_size;
 733         }
 734
 735         return *size ? 0 : ret;
 736 }
 737
 738 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 739                                 __u32 *off)
 740 {
 741         Elf_Data *symbols = obj->efile.symbols;
 742         const char *sname;
 743         size_t si;
 744
 745         if (!name || !off)
 746                 return -EINVAL;
 747
 748         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 749                 GElf_Sym sym;
 750
 751                 if (!gelf_getsym(symbols, si, &sym))
 752                         continue;
 753                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 754                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 755                         continue;
 756
 757                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 758                                    sym.st_name);
 759                 if (!sname) {
 760                         pr_warning("failed to get sym name string for var %s\n",
 761                                    name);
 762                         return -EIO;
 763                 }
 764                 if (strcmp(name, sname) == 0) {
 765                         *off = sym.st_value;
 766                         return 0;
 767                 }
 768         }
 769
 770         return -ENOENT;
 771 }
 772
 773 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 774 {
 775         struct bpf_map *new_maps;
 776         size_t new_cap;
 777         int i;
 778
 779         if (obj->nr_maps < obj->maps_cap)
 780                 return &obj->maps[obj->nr_maps++];
 781
 782         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 783         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 784         if (!new_maps) {
 785                 pr_warning("alloc maps for object failed\n");
 786                 return ERR_PTR(-ENOMEM);
 787         }
 788
 789         obj->maps_cap = new_cap;
 790         obj->maps = new_maps;
 791
 792         /* zero out new maps */
 793         memset(obj->maps + obj->nr_maps, 0,
 794                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 795         /*
 796          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 797          * when failure (zclose won't close negative fd)).
 798          */
 799         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 800                 obj->maps[i].fd = -1;
 801                 obj->maps[i].inner_map_fd = -1;
 802         }
 803
 804         return &obj->maps[obj->nr_maps++];
 805 }
 806
 807 static int
 808 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 809                               int sec_idx, Elf_Data *data, void **data_buff)
 810 {
 811         char map_name[BPF_OBJ_NAME_LEN];
 812         struct bpf_map_def *def;
 813         struct bpf_map *map;
 814
 815         map = bpf_object__add_map(obj);
 816         if (IS_ERR(map))
 817                 return PTR_ERR(map);
 818
 819         map->libbpf_type = type;
 820         map->sec_idx = sec_idx;
 821         map->sec_offset = 0;
 822         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 823                  libbpf_type_to_btf_name[type]);
 824         map->name = strdup(map_name);
 825         if (!map->name) {
 826                 pr_warning("failed to alloc map name\n");
 827                 return -ENOMEM;
 828         }
 829         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
 830                  map_name, map->sec_idx, map->sec_offset);
 831
 832         def = &map->def;
 833         def->type = BPF_MAP_TYPE_ARRAY;
 834         def->key_size = sizeof(int);
 835         def->value_size = data->d_size;
 836         def->max_entries = 1;
 837         def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
 838         if (data_buff) {
 839                 *data_buff = malloc(data->d_size);
 840                 if (!*data_buff) {
 841                         zfree(&map->name);
 842                         pr_warning("failed to alloc map content buffer\n");
 843                         return -ENOMEM;
 844                 }
 845                 memcpy(*data_buff, data->d_buf, data->d_size);
 846         }
 847
 848         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 849         return 0;
 850 }
 851
 852 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 853 {
 854         int err;
 855
 856         if (!obj->caps.global_data)
 857                 return 0;
 858         /*
 859          * Populate obj->maps with libbpf internal maps.
 860          */
 861         if (obj->efile.data_shndx >= 0) {
 862                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 863                                                     obj->efile.data_shndx,
 864                                                     obj->efile.data,
 865                                                     &obj->sections.data);
 866                 if (err)
 867                         return err;
 868         }
 869         if (obj->efile.rodata_shndx >= 0) {
 870                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 871                                                     obj->efile.rodata_shndx,
 872                                                     obj->efile.rodata,
 873                                                     &obj->sections.rodata);
 874                 if (err)
 875                         return err;
 876         }
 877         if (obj->efile.bss_shndx >= 0) {
 878                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 879                                                     obj->efile.bss_shndx,
 880                                                     obj->efile.bss, NULL);
 881                 if (err)
 882                         return err;
 883         }
 884         return 0;
 885 }
 886
 887 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
 888 {
 889         Elf_Data *symbols = obj->efile.symbols;
 890         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
 891         Elf_Data *data = NULL;
 892         Elf_Scn *scn;
 893
 894         if (obj->efile.maps_shndx < 0)
 895                 return 0;
 896
 897         if (!symbols)
 898                 return -EINVAL;
 899
 900         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 901         if (scn)
 902                 data = elf_getdata(scn, NULL);
 903         if (!scn || !data) {
 904                 pr_warning("failed to get Elf_Data from map section %d\n",
 905                            obj->efile.maps_shndx);
 906                 return -EINVAL;
 907         }
 908
 909         /*
 910          * Count number of maps. Each map has a name.
 911          * Array of maps is not supported: only the first element is
 912          * considered.
 913          *
 914          * TODO: Detect array of map and report error.
 915          */
 916         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 917         for (i = 0; i < nr_syms; i++) {
 918                 GElf_Sym sym;
 919
 920                 if (!gelf_getsym(symbols, i, &sym))
 921                         continue;
 922                 if (sym.st_shndx != obj->efile.maps_shndx)
 923                         continue;
 924                 nr_maps++;
 925         }
 926         /* Assume equally sized map definitions */
 927         pr_debug("maps in %s: %d maps in %zd bytes\n",
 928                  obj->path, nr_maps, data->d_size);
 929
 930         map_def_sz = data->d_size / nr_maps;
 931         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 932                 pr_warning("unable to determine map definition size "
 933                            "section %s, %d maps in %zd bytes\n",
 934                            obj->path, nr_maps, data->d_size);
 935                 return -EINVAL;
 936         }
 937
 938         /* Fill obj->maps using data in "maps" section.  */
 939         for (i = 0; i < nr_syms; i++) {
 940                 GElf_Sym sym;
 941                 const char *map_name;
 942                 struct bpf_map_def *def;
 943                 struct bpf_map *map;
 944
 945                 if (!gelf_getsym(symbols, i, &sym))
 946                         continue;
 947                 if (sym.st_shndx != obj->efile.maps_shndx)
 948                         continue;
 949
 950                 map = bpf_object__add_map(obj);
 951                 if (IS_ERR(map))
 952                         return PTR_ERR(map);
 953
 954                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 955                                       sym.st_name);
 956                 if (!map_name) {
 957                         pr_warning("failed to get map #%d name sym string for obj %s\n",
 958                                    i, obj->path);
 959                         return -LIBBPF_ERRNO__FORMAT;
 960                 }
 961
 962                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
 963                 map->sec_idx = sym.st_shndx;
 964                 map->sec_offset = sym.st_value;
 965                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
 966                          map_name, map->sec_idx, map->sec_offset);
 967                 if (sym.st_value + map_def_sz > data->d_size) {
 968                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 969                                    obj->path, map_name);
 970                         return -EINVAL;
 971                 }
 972
 973                 map->name = strdup(map_name);
 974                 if (!map->name) {
 975                         pr_warning("failed to alloc map name\n");
 976                         return -ENOMEM;
 977                 }
 978                 pr_debug("map %d is \"%s\"\n", i, map->name);
 979                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 980                 /*
 981                  * If the definition of the map in the object file fits in
 982                  * bpf_map_def, copy it.  Any extra fields in our version
 983                  * of bpf_map_def will default to zero as a result of the
 984                  * calloc above.
 985                  */
 986                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 987                         memcpy(&map->def, def, map_def_sz);
 988                 } else {
 989                         /*
 990                          * Here the map structure being read is bigger than what
 991                          * we expect, truncate if the excess bits are all zero.
 992                          * If they are not zero, reject this map as
 993                          * incompatible.
 994                          */
 995                         char *b;
 996                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 997                              b < ((char *)def) + map_def_sz; b++) {
 998                                 if (*b != 0) {
 999                                         pr_warning("maps section in %s: \"%s\" "
1000                                                    "has unrecognized, non-zero "
1001                                                    "options\n",
1002                                                    obj->path, map_name);
1003                                         if (strict)
1004                                                 return -EINVAL;
1005                                 }
1006                         }
1007                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1008                 }
1009         }
1010         return 0;
1011 }
1012
1013 static const struct btf_type *skip_mods_and_typedefs(const struct btf *btf,
1014                                                      __u32 id)
1015 {
1016         const struct btf_type *t = btf__type_by_id(btf, id);
1017
1018         while (true) {
1019                 switch (BTF_INFO_KIND(t->info)) {
1020                 case BTF_KIND_VOLATILE:
1021                 case BTF_KIND_CONST:
1022                 case BTF_KIND_RESTRICT:
1023                 case BTF_KIND_TYPEDEF:
1024                         t = btf__type_by_id(btf, t->type);
1025                         break;
1026                 default:
1027                         return t;
1028                 }
1029         }
1030 }
1031
1032 /*
1033  * Fetch integer attribute of BTF map definition. Such attributes are
1034  * represented using a pointer to an array, in which dimensionality of array
1035  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1036  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1037  * type definition, while using only sizeof(void *) space in ELF data section.
1038  */
1039 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1040                               const struct btf_type *def,
1041                               const struct btf_member *m, __u32 *res) {
1042         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type);
1043         const char *name = btf__name_by_offset(btf, m->name_off);
1044         const struct btf_array *arr_info;
1045         const struct btf_type *arr_t;
1046
1047         if (BTF_INFO_KIND(t->info) != BTF_KIND_PTR) {
1048                 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1049                            map_name, name, BTF_INFO_KIND(t->info));
1050                 return false;
1051         }
1052
1053         arr_t = btf__type_by_id(btf, t->type);
1054         if (!arr_t) {
1055                 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1056                            map_name, name, t->type);
1057                 return false;
1058         }
1059         if (BTF_INFO_KIND(arr_t->info) != BTF_KIND_ARRAY) {
1060                 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1061                            map_name, name, BTF_INFO_KIND(arr_t->info));
1062                 return false;
1063         }
1064         arr_info = (const void *)(arr_t + 1);
1065         *res = arr_info->nelems;
1066         return true;
1067 }
1068
1069 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1070                                          const struct btf_type *sec,
1071                                          int var_idx, int sec_idx,
1072                                          const Elf_Data *data, bool strict)
1073 {
1074         const struct btf_type *var, *def, *t;
1075         const struct btf_var_secinfo *vi;
1076         const struct btf_var *var_extra;
1077         const struct btf_member *m;
1078         const char *map_name;
1079         struct bpf_map *map;
1080         int vlen, i;
1081
1082         vi = (const struct btf_var_secinfo *)(const void *)(sec + 1) + var_idx;
1083         var = btf__type_by_id(obj->btf, vi->type);
1084         var_extra = (const void *)(var + 1);
1085         map_name = btf__name_by_offset(obj->btf, var->name_off);
1086         vlen = BTF_INFO_VLEN(var->info);
1087
1088         if (map_name == NULL || map_name[0] == '\0') {
1089                 pr_warning("map #%d: empty name.\n", var_idx);
1090                 return -EINVAL;
1091         }
1092         if ((__u64)vi->offset + vi->size > data->d_size) {
1093                 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1094                 return -EINVAL;
1095         }
1096         if (BTF_INFO_KIND(var->info) != BTF_KIND_VAR) {
1097                 pr_warning("map '%s': unexpected var kind %u.\n",
1098                            map_name, BTF_INFO_KIND(var->info));
1099                 return -EINVAL;
1100         }
1101         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1102             var_extra->linkage != BTF_VAR_STATIC) {
1103                 pr_warning("map '%s': unsupported var linkage %u.\n",
1104                            map_name, var_extra->linkage);
1105                 return -EOPNOTSUPP;
1106         }
1107
1108         def = skip_mods_and_typedefs(obj->btf, var->type);
1109         if (BTF_INFO_KIND(def->info) != BTF_KIND_STRUCT) {
1110                 pr_warning("map '%s': unexpected def kind %u.\n",
1111                            map_name, BTF_INFO_KIND(var->info));
1112                 return -EINVAL;
1113         }
1114         if (def->size > vi->size) {
1115                 pr_warning("map '%s': invalid def size.\n", map_name);
1116                 return -EINVAL;
1117         }
1118
1119         map = bpf_object__add_map(obj);
1120         if (IS_ERR(map))
1121                 return PTR_ERR(map);
1122         map->name = strdup(map_name);
1123         if (!map->name) {
1124                 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1125                 return -ENOMEM;
1126         }
1127         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1128         map->def.type = BPF_MAP_TYPE_UNSPEC;
1129         map->sec_idx = sec_idx;
1130         map->sec_offset = vi->offset;
1131         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1132                  map_name, map->sec_idx, map->sec_offset);
1133
1134         vlen = BTF_INFO_VLEN(def->info);
1135         m = (const void *)(def + 1);
1136         for (i = 0; i < vlen; i++, m++) {
1137                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1138
1139                 if (!name) {
1140                         pr_warning("map '%s': invalid field #%d.\n",
1141                                    map_name, i);
1142                         return -EINVAL;
1143                 }
1144                 if (strcmp(name, "type") == 0) {
1145                         if (!get_map_field_int(map_name, obj->btf, def, m,
1146                                                &map->def.type))
1147                                 return -EINVAL;
1148                         pr_debug("map '%s': found type = %u.\n",
1149                                  map_name, map->def.type);
1150                 } else if (strcmp(name, "max_entries") == 0) {
1151                         if (!get_map_field_int(map_name, obj->btf, def, m,
1152                                                &map->def.max_entries))
1153                                 return -EINVAL;
1154                         pr_debug("map '%s': found max_entries = %u.\n",
1155                                  map_name, map->def.max_entries);
1156                 } else if (strcmp(name, "map_flags") == 0) {
1157                         if (!get_map_field_int(map_name, obj->btf, def, m,
1158                                                &map->def.map_flags))
1159                                 return -EINVAL;
1160                         pr_debug("map '%s': found map_flags = %u.\n",
1161                                  map_name, map->def.map_flags);
1162                 } else if (strcmp(name, "key_size") == 0) {
1163                         __u32 sz;
1164
1165                         if (!get_map_field_int(map_name, obj->btf, def, m,
1166                                                &sz))
1167                                 return -EINVAL;
1168                         pr_debug("map '%s': found key_size = %u.\n",
1169                                  map_name, sz);
1170                         if (map->def.key_size && map->def.key_size != sz) {
1171                                 pr_warning("map '%s': conflicting key size %u != %u.\n",
1172                                            map_name, map->def.key_size, sz);
1173                                 return -EINVAL;
1174                         }
1175                         map->def.key_size = sz;
1176                 } else if (strcmp(name, "key") == 0) {
1177                         __s64 sz;
1178
1179                         t = btf__type_by_id(obj->btf, m->type);
1180                         if (!t) {
1181                                 pr_warning("map '%s': key type [%d] not found.\n",
1182                                            map_name, m->type);
1183                                 return -EINVAL;
1184                         }
1185                         if (BTF_INFO_KIND(t->info) != BTF_KIND_PTR) {
1186                                 pr_warning("map '%s': key spec is not PTR: %u.\n",
1187                                            map_name, BTF_INFO_KIND(t->info));
1188                                 return -EINVAL;
1189                         }
1190                         sz = btf__resolve_size(obj->btf, t->type);
1191                         if (sz < 0) {
1192                                 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1193                                            map_name, t->type, sz);
1194                                 return sz;
1195                         }
1196                         pr_debug("map '%s': found key [%u], sz = %lld.\n",
1197                                  map_name, t->type, sz);
1198                         if (map->def.key_size && map->def.key_size != sz) {
1199                                 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1200                                            map_name, map->def.key_size, sz);
1201                                 return -EINVAL;
1202                         }
1203                         map->def.key_size = sz;
1204                         map->btf_key_type_id = t->type;
1205                 } else if (strcmp(name, "value_size") == 0) {
1206                         __u32 sz;
1207
1208                         if (!get_map_field_int(map_name, obj->btf, def, m,
1209                                                &sz))
1210                                 return -EINVAL;
1211                         pr_debug("map '%s': found value_size = %u.\n",
1212                                  map_name, sz);
1213                         if (map->def.value_size && map->def.value_size != sz) {
1214                                 pr_warning("map '%s': conflicting value size %u != %u.\n",
1215                                            map_name, map->def.value_size, sz);
1216                                 return -EINVAL;
1217                         }
1218                         map->def.value_size = sz;
1219                 } else if (strcmp(name, "value") == 0) {
1220                         __s64 sz;
1221
1222                         t = btf__type_by_id(obj->btf, m->type);
1223                         if (!t) {
1224                                 pr_warning("map '%s': value type [%d] not found.\n",
1225                                            map_name, m->type);
1226                                 return -EINVAL;
1227                         }
1228                         if (BTF_INFO_KIND(t->info) != BTF_KIND_PTR) {
1229                                 pr_warning("map '%s': value spec is not PTR: %u.\n",
1230                                            map_name, BTF_INFO_KIND(t->info));
1231                                 return -EINVAL;
1232                         }
1233                         sz = btf__resolve_size(obj->btf, t->type);
1234                         if (sz < 0) {
1235                                 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1236                                            map_name, t->type, sz);
1237                                 return sz;
1238                         }
1239                         pr_debug("map '%s': found value [%u], sz = %lld.\n",
1240                                  map_name, t->type, sz);
1241                         if (map->def.value_size && map->def.value_size != sz) {
1242                                 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1243                                            map_name, map->def.value_size, sz);
1244                                 return -EINVAL;
1245                         }
1246                         map->def.value_size = sz;
1247                         map->btf_value_type_id = t->type;
1248                 } else {
1249                         if (strict) {
1250                                 pr_warning("map '%s': unknown field '%s'.\n",
1251                                            map_name, name);
1252                                 return -ENOTSUP;
1253                         }
1254                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1255                                  map_name, name);
1256                 }
1257         }
1258
1259         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1260                 pr_warning("map '%s': map type isn't specified.\n", map_name);
1261                 return -EINVAL;
1262         }
1263
1264         return 0;
1265 }
1266
1267 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1268 {
1269         const struct btf_type *sec = NULL;
1270         int nr_types, i, vlen, err;
1271         const struct btf_type *t;
1272         const char *name;
1273         Elf_Data *data;
1274         Elf_Scn *scn;
1275
1276         if (obj->efile.btf_maps_shndx < 0)
1277                 return 0;
1278
1279         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1280         if (scn)
1281                 data = elf_getdata(scn, NULL);
1282         if (!scn || !data) {
1283                 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1284                            obj->efile.maps_shndx, MAPS_ELF_SEC);
1285                 return -EINVAL;
1286         }
1287
1288         nr_types = btf__get_nr_types(obj->btf);
1289         for (i = 1; i <= nr_types; i++) {
1290                 t = btf__type_by_id(obj->btf, i);
1291                 if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC)
1292                         continue;
1293                 name = btf__name_by_offset(obj->btf, t->name_off);
1294                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1295                         sec = t;
1296                         break;
1297                 }
1298         }
1299
1300         if (!sec) {
1301                 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1302                 return -ENOENT;
1303         }
1304
1305         vlen = BTF_INFO_VLEN(sec->info);
1306         for (i = 0; i < vlen; i++) {
1307                 err = bpf_object__init_user_btf_map(obj, sec, i,
1308                                                     obj->efile.btf_maps_shndx,
1309                                                     data, strict);
1310                 if (err)
1311                         return err;
1312         }
1313
1314         return 0;
1315 }
1316
1317 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1318 {
1319         bool strict = !(flags & MAPS_RELAX_COMPAT);
1320         int err;
1321
1322         err = bpf_object__init_user_maps(obj, strict);
1323         if (err)
1324                 return err;
1325
1326         err = bpf_object__init_user_btf_maps(obj, strict);
1327         if (err)
1328                 return err;
1329
1330         err = bpf_object__init_global_data_maps(obj);
1331         if (err)
1332                 return err;
1333
1334         if (obj->nr_maps) {
1335                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1336                       compare_bpf_map);
1337         }
1338         return 0;
1339 }
1340
1341 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1342 {
1343         Elf_Scn *scn;
1344         GElf_Shdr sh;
1345
1346         scn = elf_getscn(obj->efile.elf, idx);
1347         if (!scn)
1348                 return false;
1349
1350         if (gelf_getshdr(scn, &sh) != &sh)
1351                 return false;
1352
1353         if (sh.sh_flags & SHF_EXECINSTR)
1354                 return true;
1355
1356         return false;
1357 }
1358
1359 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1360 {
1361         bool has_datasec = obj->caps.btf_datasec;
1362         bool has_func = obj->caps.btf_func;
1363         struct btf *btf = obj->btf;
1364         struct btf_type *t;
1365         int i, j, vlen;
1366         __u16 kind;
1367
1368         if (!obj->btf || (has_func && has_datasec))
1369                 return;
1370
1371         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1372                 t = (struct btf_type *)btf__type_by_id(btf, i);
1373                 kind = BTF_INFO_KIND(t->info);
1374
1375                 if (!has_datasec && kind == BTF_KIND_VAR) {
1376                         /* replace VAR with INT */
1377                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1378                         /*
1379                          * using size = 1 is the safest choice, 4 will be too
1380                          * big and cause kernel BTF validation failure if
1381                          * original variable took less than 4 bytes
1382                          */
1383                         t->size = 1;
1384                         *(int *)(t+1) = BTF_INT_ENC(0, 0, 8);
1385                 } else if (!has_datasec && kind == BTF_KIND_DATASEC) {
1386                         /* replace DATASEC with STRUCT */
1387                         struct btf_var_secinfo *v = (void *)(t + 1);
1388                         struct btf_member *m = (void *)(t + 1);
1389                         struct btf_type *vt;
1390                         char *name;
1391
1392                         name = (char *)btf__name_by_offset(btf, t->name_off);
1393                         while (*name) {
1394                                 if (*name == '.')
1395                                         *name = '_';
1396                                 name++;
1397                         }
1398
1399                         vlen = BTF_INFO_VLEN(t->info);
1400                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1401                         for (j = 0; j < vlen; j++, v++, m++) {
1402                                 /* order of field assignments is important */
1403                                 m->offset = v->offset * 8;
1404                                 m->type = v->type;
1405                                 /* preserve variable name as member name */
1406                                 vt = (void *)btf__type_by_id(btf, v->type);
1407                                 m->name_off = vt->name_off;
1408                         }
1409                 } else if (!has_func && kind == BTF_KIND_FUNC_PROTO) {
1410                         /* replace FUNC_PROTO with ENUM */
1411                         vlen = BTF_INFO_VLEN(t->info);
1412                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1413                         t->size = sizeof(__u32); /* kernel enforced */
1414                 } else if (!has_func && kind == BTF_KIND_FUNC) {
1415                         /* replace FUNC with TYPEDEF */
1416                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1417                 }
1418         }
1419 }
1420
1421 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1422 {
1423         if (!obj->btf_ext)
1424                 return;
1425
1426         if (!obj->caps.btf_func) {
1427                 btf_ext__free(obj->btf_ext);
1428                 obj->btf_ext = NULL;
1429         }
1430 }
1431
1432 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1433 {
1434         return obj->efile.btf_maps_shndx >= 0;
1435 }
1436
1437 static int bpf_object__init_btf(struct bpf_object *obj,
1438                                 Elf_Data *btf_data,
1439                                 Elf_Data *btf_ext_data)
1440 {
1441         bool btf_required = bpf_object__is_btf_mandatory(obj);
1442         int err = 0;
1443
1444         if (btf_data) {
1445                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1446                 if (IS_ERR(obj->btf)) {
1447                         pr_warning("Error loading ELF section %s: %d.\n",
1448                                    BTF_ELF_SEC, err);
1449                         goto out;
1450                 }
1451                 err = btf__finalize_data(obj, obj->btf);
1452                 if (err) {
1453                         pr_warning("Error finalizing %s: %d.\n",
1454                                    BTF_ELF_SEC, err);
1455                         goto out;
1456                 }
1457         }
1458         if (btf_ext_data) {
1459                 if (!obj->btf) {
1460                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1461                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1462                         goto out;
1463                 }
1464                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1465                                             btf_ext_data->d_size);
1466                 if (IS_ERR(obj->btf_ext)) {
1467                         pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1468                                    BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1469                         obj->btf_ext = NULL;
1470                         goto out;
1471                 }
1472         }
1473 out:
1474         if (err || IS_ERR(obj->btf)) {
1475                 if (btf_required)
1476                         err = err ? : PTR_ERR(obj->btf);
1477                 else
1478                         err = 0;
1479                 if (!IS_ERR_OR_NULL(obj->btf))
1480                         btf__free(obj->btf);
1481                 obj->btf = NULL;
1482         }
1483         if (btf_required && !obj->btf) {
1484                 pr_warning("BTF is required, but is missing or corrupted.\n");
1485                 return err == 0 ? -ENOENT : err;
1486         }
1487         return 0;
1488 }
1489
1490 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1491 {
1492         int err = 0;
1493
1494         if (!obj->btf)
1495                 return 0;
1496
1497         bpf_object__sanitize_btf(obj);
1498         bpf_object__sanitize_btf_ext(obj);
1499
1500         err = btf__load(obj->btf);
1501         if (err) {
1502                 pr_warning("Error loading %s into kernel: %d.\n",
1503                            BTF_ELF_SEC, err);
1504                 btf__free(obj->btf);
1505                 obj->btf = NULL;
1506                 /* btf_ext can't exist without btf, so free it as well */
1507                 if (obj->btf_ext) {
1508                         btf_ext__free(obj->btf_ext);
1509                         obj->btf_ext = NULL;
1510                 }
1511
1512                 if (bpf_object__is_btf_mandatory(obj))
1513                         return err;
1514         }
1515         return 0;
1516 }
1517
1518 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1519 {
1520         Elf *elf = obj->efile.elf;
1521         GElf_Ehdr *ep = &obj->efile.ehdr;
1522         Elf_Data *btf_ext_data = NULL;
1523         Elf_Data *btf_data = NULL;
1524         Elf_Scn *scn = NULL;
1525         int idx = 0, err = 0;
1526
1527         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1528         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1529                 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1530                 return -LIBBPF_ERRNO__FORMAT;
1531         }
1532
1533         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1534                 char *name;
1535                 GElf_Shdr sh;
1536                 Elf_Data *data;
1537
1538                 idx++;
1539                 if (gelf_getshdr(scn, &sh) != &sh) {
1540                         pr_warning("failed to get section(%d) header from %s\n",
1541                                    idx, obj->path);
1542                         return -LIBBPF_ERRNO__FORMAT;
1543                 }
1544
1545                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1546                 if (!name) {
1547                         pr_warning("failed to get section(%d) name from %s\n",
1548                                    idx, obj->path);
1549                         return -LIBBPF_ERRNO__FORMAT;
1550                 }
1551
1552                 data = elf_getdata(scn, 0);
1553                 if (!data) {
1554                         pr_warning("failed to get section(%d) data from %s(%s)\n",
1555                                    idx, name, obj->path);
1556                         return -LIBBPF_ERRNO__FORMAT;
1557                 }
1558                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1559                          idx, name, (unsigned long)data->d_size,
1560                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1561                          (int)sh.sh_type);
1562
1563                 if (strcmp(name, "license") == 0) {
1564                         err = bpf_object__init_license(obj,
1565                                                        data->d_buf,
1566                                                        data->d_size);
1567                         if (err)
1568                                 return err;
1569                 } else if (strcmp(name, "version") == 0) {
1570                         err = bpf_object__init_kversion(obj,
1571                                                         data->d_buf,
1572                                                         data->d_size);
1573                         if (err)
1574                                 return err;
1575                 } else if (strcmp(name, "maps") == 0) {
1576                         obj->efile.maps_shndx = idx;
1577                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1578                         obj->efile.btf_maps_shndx = idx;
1579                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1580                         btf_data = data;
1581                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1582                         btf_ext_data = data;
1583                 } else if (sh.sh_type == SHT_SYMTAB) {
1584                         if (obj->efile.symbols) {
1585                                 pr_warning("bpf: multiple SYMTAB in %s\n",
1586                                            obj->path);
1587                                 return -LIBBPF_ERRNO__FORMAT;
1588                         }
1589                         obj->efile.symbols = data;
1590                         obj->efile.strtabidx = sh.sh_link;
1591                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1592                         if (sh.sh_flags & SHF_EXECINSTR) {
1593                                 if (strcmp(name, ".text") == 0)
1594                                         obj->efile.text_shndx = idx;
1595                                 err = bpf_object__add_program(obj, data->d_buf,
1596                                                               data->d_size, name, idx);
1597                                 if (err) {
1598                                         char errmsg[STRERR_BUFSIZE];
1599                                         char *cp = libbpf_strerror_r(-err, errmsg,
1600                                                                      sizeof(errmsg));
1601
1602                                         pr_warning("failed to alloc program %s (%s): %s",
1603                                                    name, obj->path, cp);
1604                                         return err;
1605                                 }
1606                         } else if (strcmp(name, ".data") == 0) {
1607                                 obj->efile.data = data;
1608                                 obj->efile.data_shndx = idx;
1609                         } else if (strcmp(name, ".rodata") == 0) {
1610                                 obj->efile.rodata = data;
1611                                 obj->efile.rodata_shndx = idx;
1612                         } else {
1613                                 pr_debug("skip section(%d) %s\n", idx, name);
1614                         }
1615                 } else if (sh.sh_type == SHT_REL) {
1616                         int nr_reloc = obj->efile.nr_reloc;
1617                         void *reloc = obj->efile.reloc;
1618                         int sec = sh.sh_info; /* points to other section */
1619
1620                         /* Only do relo for section with exec instructions */
1621                         if (!section_have_execinstr(obj, sec)) {
1622                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1623                                          name, idx, sec);
1624                                 continue;
1625                         }
1626
1627                         reloc = reallocarray(reloc, nr_reloc + 1,
1628                                              sizeof(*obj->efile.reloc));
1629                         if (!reloc) {
1630                                 pr_warning("realloc failed\n");
1631                                 return -ENOMEM;
1632                         }
1633
1634                         obj->efile.reloc = reloc;
1635                         obj->efile.nr_reloc++;
1636
1637                         obj->efile.reloc[nr_reloc].shdr = sh;
1638                         obj->efile.reloc[nr_reloc].data = data;
1639                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1640                         obj->efile.bss = data;
1641                         obj->efile.bss_shndx = idx;
1642                 } else {
1643                         pr_debug("skip section(%d) %s\n", idx, name);
1644                 }
1645         }
1646
1647         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1648                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1649                 return -LIBBPF_ERRNO__FORMAT;
1650         }
1651         err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1652         if (!err)
1653                 err = bpf_object__init_maps(obj, flags);
1654         if (!err)
1655                 err = bpf_object__sanitize_and_load_btf(obj);
1656         if (!err)
1657                 err = bpf_object__init_prog_names(obj);
1658         return err;
1659 }
1660
1661 static struct bpf_program *
1662 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1663 {
1664         struct bpf_program *prog;
1665         size_t i;
1666
1667         for (i = 0; i < obj->nr_programs; i++) {
1668                 prog = &obj->programs[i];
1669                 if (prog->idx == idx)
1670                         return prog;
1671         }
1672         return NULL;
1673 }
1674
1675 struct bpf_program *
1676 bpf_object__find_program_by_title(const struct bpf_object *obj,
1677                                   const char *title)
1678 {
1679         struct bpf_program *pos;
1680
1681         bpf_object__for_each_program(pos, obj) {
1682                 if (pos->section_name && !strcmp(pos->section_name, title))
1683                         return pos;
1684         }
1685         return NULL;
1686 }
1687
1688 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1689                                       int shndx)
1690 {
1691         return shndx == obj->efile.data_shndx ||
1692                shndx == obj->efile.bss_shndx ||
1693                shndx == obj->efile.rodata_shndx;
1694 }
1695
1696 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1697                                       int shndx)
1698 {
1699         return shndx == obj->efile.maps_shndx ||
1700                shndx == obj->efile.btf_maps_shndx;
1701 }
1702
1703 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1704                                               int shndx)
1705 {
1706         return shndx == obj->efile.text_shndx ||
1707                bpf_object__shndx_is_maps(obj, shndx) ||
1708                bpf_object__shndx_is_data(obj, shndx);
1709 }
1710
1711 static enum libbpf_map_type
1712 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1713 {
1714         if (shndx == obj->efile.data_shndx)
1715                 return LIBBPF_MAP_DATA;
1716         else if (shndx == obj->efile.bss_shndx)
1717                 return LIBBPF_MAP_BSS;
1718         else if (shndx == obj->efile.rodata_shndx)
1719                 return LIBBPF_MAP_RODATA;
1720         else
1721                 return LIBBPF_MAP_UNSPEC;
1722 }
1723
1724 static int
1725 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1726                            Elf_Data *data, struct bpf_object *obj)
1727 {
1728         Elf_Data *symbols = obj->efile.symbols;
1729         struct bpf_map *maps = obj->maps;
1730         size_t nr_maps = obj->nr_maps;
1731         int i, nrels;
1732
1733         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1734         nrels = shdr->sh_size / shdr->sh_entsize;
1735
1736         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1737         if (!prog->reloc_desc) {
1738                 pr_warning("failed to alloc memory in relocation\n");
1739                 return -ENOMEM;
1740         }
1741         prog->nr_reloc = nrels;
1742
1743         for (i = 0; i < nrels; i++) {
1744                 struct bpf_insn *insns = prog->insns;
1745                 enum libbpf_map_type type;
1746                 unsigned int insn_idx;
1747                 unsigned int shdr_idx;
1748                 const char *name;
1749                 size_t map_idx;
1750                 GElf_Sym sym;
1751                 GElf_Rel rel;
1752
1753                 if (!gelf_getrel(data, i, &rel)) {
1754                         pr_warning("relocation: failed to get %d reloc\n", i);
1755                         return -LIBBPF_ERRNO__FORMAT;
1756                 }
1757
1758                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1759                         pr_warning("relocation: symbol %"PRIx64" not found\n",
1760                                    GELF_R_SYM(rel.r_info));
1761                         return -LIBBPF_ERRNO__FORMAT;
1762                 }
1763
1764                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1765                                   sym.st_name) ? : "<?>";
1766
1767                 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1768                          (long long) (rel.r_info >> 32),
1769                          (long long) sym.st_value, sym.st_name, name);
1770
1771                 shdr_idx = sym.st_shndx;
1772                 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1773                         pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1774                                    prog->section_name, shdr_idx);
1775                         return -LIBBPF_ERRNO__RELOC;
1776                 }
1777
1778                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1779                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
1780
1781                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1782                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1783                                 pr_warning("incorrect bpf_call opcode\n");
1784                                 return -LIBBPF_ERRNO__RELOC;
1785                         }
1786                         prog->reloc_desc[i].type = RELO_CALL;
1787                         prog->reloc_desc[i].insn_idx = insn_idx;
1788                         prog->reloc_desc[i].text_off = sym.st_value;
1789                         obj->has_pseudo_calls = true;
1790                         continue;
1791                 }
1792
1793                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1794                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1795                                    insn_idx, insns[insn_idx].code);
1796                         return -LIBBPF_ERRNO__RELOC;
1797                 }
1798
1799                 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1800                     bpf_object__shndx_is_data(obj, shdr_idx)) {
1801                         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1802                         if (type != LIBBPF_MAP_UNSPEC) {
1803                                 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1804                                         pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1805                                                    name, insn_idx, insns[insn_idx].code);
1806                                         return -LIBBPF_ERRNO__RELOC;
1807                                 }
1808                                 if (!obj->caps.global_data) {
1809                                         pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1810                                                    name, insn_idx);
1811                                         return -LIBBPF_ERRNO__RELOC;
1812                                 }
1813                         }
1814
1815                         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1816                                 if (maps[map_idx].libbpf_type != type)
1817                                         continue;
1818                                 if (type != LIBBPF_MAP_UNSPEC ||
1819                                     (maps[map_idx].sec_idx == sym.st_shndx &&
1820                                      maps[map_idx].sec_offset == sym.st_value)) {
1821                                         pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1822                                                  map_idx, maps[map_idx].name,
1823                                                  maps[map_idx].sec_idx,
1824                                                  maps[map_idx].sec_offset,
1825                                                  insn_idx);
1826                                         break;
1827                                 }
1828                         }
1829
1830                         if (map_idx >= nr_maps) {
1831                                 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1832                                            (int)map_idx, (int)nr_maps - 1);
1833                                 return -LIBBPF_ERRNO__RELOC;
1834                         }
1835
1836                         prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1837                                                    RELO_DATA : RELO_LD64;
1838                         prog->reloc_desc[i].insn_idx = insn_idx;
1839                         prog->reloc_desc[i].map_idx = map_idx;
1840                 }
1841         }
1842         return 0;
1843 }
1844
1845 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1846 {
1847         struct bpf_map_def *def = &map->def;
1848         __u32 key_type_id = 0, value_type_id = 0;
1849         int ret;
1850
1851         /* if it's BTF-defined map, we don't need to search for type IDs */
1852         if (map->sec_idx == obj->efile.btf_maps_shndx)
1853                 return 0;
1854
1855         if (!bpf_map__is_internal(map)) {
1856                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1857                                            def->value_size, &key_type_id,
1858                                            &value_type_id);
1859         } else {
1860                 /*
1861                  * LLVM annotates global data differently in BTF, that is,
1862                  * only as '.data', '.bss' or '.rodata'.
1863                  */
1864                 ret = btf__find_by_name(obj->btf,
1865                                 libbpf_type_to_btf_name[map->libbpf_type]);
1866         }
1867         if (ret < 0)
1868                 return ret;
1869
1870         map->btf_key_type_id = key_type_id;
1871         map->btf_value_type_id = bpf_map__is_internal(map) ?
1872                                  ret : value_type_id;
1873         return 0;
1874 }
1875
1876 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1877 {
1878         struct bpf_map_info info = {};
1879         __u32 len = sizeof(info);
1880         int new_fd, err;
1881         char *new_name;
1882
1883         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1884         if (err)
1885                 return err;
1886
1887         new_name = strdup(info.name);
1888         if (!new_name)
1889                 return -errno;
1890
1891         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1892         if (new_fd < 0)
1893                 goto err_free_new_name;
1894
1895         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1896         if (new_fd < 0)
1897                 goto err_close_new_fd;
1898
1899         err = zclose(map->fd);
1900         if (err)
1901                 goto err_close_new_fd;
1902         free(map->name);
1903
1904         map->fd = new_fd;
1905         map->name = new_name;
1906         map->def.type = info.type;
1907         map->def.key_size = info.key_size;
1908         map->def.value_size = info.value_size;
1909         map->def.max_entries = info.max_entries;
1910         map->def.map_flags = info.map_flags;
1911         map->btf_key_type_id = info.btf_key_type_id;
1912         map->btf_value_type_id = info.btf_value_type_id;
1913
1914         return 0;
1915
1916 err_close_new_fd:
1917         close(new_fd);
1918 err_free_new_name:
1919         free(new_name);
1920         return -errno;
1921 }
1922
1923 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1924 {
1925         if (!map || !max_entries)
1926                 return -EINVAL;
1927
1928         /* If map already created, its attributes can't be changed. */
1929         if (map->fd >= 0)
1930                 return -EBUSY;
1931
1932         map->def.max_entries = max_entries;
1933
1934         return 0;
1935 }
1936
1937 static int
1938 bpf_object__probe_name(struct bpf_object *obj)
1939 {
1940         struct bpf_load_program_attr attr;
1941         char *cp, errmsg[STRERR_BUFSIZE];
1942         struct bpf_insn insns[] = {
1943                 BPF_MOV64_IMM(BPF_REG_0, 0),
1944                 BPF_EXIT_INSN(),
1945         };
1946         int ret;
1947
1948         /* make sure basic loading works */
1949
1950         memset(&attr, 0, sizeof(attr));
1951         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1952         attr.insns = insns;
1953         attr.insns_cnt = ARRAY_SIZE(insns);
1954         attr.license = "GPL";
1955
1956         ret = bpf_load_program_xattr(&attr, NULL, 0);
1957         if (ret < 0) {
1958                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1959                 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1960                            __func__, cp, errno);
1961                 return -errno;
1962         }
1963         close(ret);
1964
1965         /* now try the same program, but with the name */
1966
1967         attr.name = "test";
1968         ret = bpf_load_program_xattr(&attr, NULL, 0);
1969         if (ret >= 0) {
1970                 obj->caps.name = 1;
1971                 close(ret);
1972         }
1973
1974         return 0;
1975 }
1976
1977 static int
1978 bpf_object__probe_global_data(struct bpf_object *obj)
1979 {
1980         struct bpf_load_program_attr prg_attr;
1981         struct bpf_create_map_attr map_attr;
1982         char *cp, errmsg[STRERR_BUFSIZE];
1983         struct bpf_insn insns[] = {
1984                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1985                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
1986                 BPF_MOV64_IMM(BPF_REG_0, 0),
1987                 BPF_EXIT_INSN(),
1988         };
1989         int ret, map;
1990
1991         memset(&map_attr, 0, sizeof(map_attr));
1992         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
1993         map_attr.key_size = sizeof(int);
1994         map_attr.value_size = 32;
1995         map_attr.max_entries = 1;
1996
1997         map = bpf_create_map_xattr(&map_attr);
1998         if (map < 0) {
1999                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2000                 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2001                            __func__, cp, errno);
2002                 return -errno;
2003         }
2004
2005         insns[0].imm = map;
2006
2007         memset(&prg_attr, 0, sizeof(prg_attr));
2008         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2009         prg_attr.insns = insns;
2010         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2011         prg_attr.license = "GPL";
2012
2013         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2014         if (ret >= 0) {
2015                 obj->caps.global_data = 1;
2016                 close(ret);
2017         }
2018
2019         close(map);
2020         return 0;
2021 }
2022
2023 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2024 {
2025         const char strs[] = "\0int\0x\0a";
2026         /* void x(int a) {} */
2027         __u32 types[] = {
2028                 /* int */
2029                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2030                 /* FUNC_PROTO */                                /* [2] */
2031                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2032                 BTF_PARAM_ENC(7, 1),
2033                 /* FUNC x */                                    /* [3] */
2034                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2035         };
2036         int btf_fd;
2037
2038         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2039                                       strs, sizeof(strs));
2040         if (btf_fd >= 0) {
2041                 obj->caps.btf_func = 1;
2042                 close(btf_fd);
2043                 return 1;
2044         }
2045
2046         return 0;
2047 }
2048
2049 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2050 {
2051         const char strs[] = "\0x\0.data";
2052         /* static int a; */
2053         __u32 types[] = {
2054                 /* int */
2055                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2056                 /* VAR x */                                     /* [2] */
2057                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2058                 BTF_VAR_STATIC,
2059                 /* DATASEC val */                               /* [3] */
2060                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2061                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2062         };
2063         int btf_fd;
2064
2065         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2066                                       strs, sizeof(strs));
2067         if (btf_fd >= 0) {
2068                 obj->caps.btf_datasec = 1;
2069                 close(btf_fd);
2070                 return 1;
2071         }
2072
2073         return 0;
2074 }
2075
2076 static int
2077 bpf_object__probe_caps(struct bpf_object *obj)
2078 {
2079         int (*probe_fn[])(struct bpf_object *obj) = {
2080                 bpf_object__probe_name,
2081                 bpf_object__probe_global_data,
2082                 bpf_object__probe_btf_func,
2083                 bpf_object__probe_btf_datasec,
2084         };
2085         int i, ret;
2086
2087         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2088                 ret = probe_fn[i](obj);
2089                 if (ret < 0)
2090                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2091         }
2092
2093         return 0;
2094 }
2095
2096 static int
2097 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2098 {
2099         char *cp, errmsg[STRERR_BUFSIZE];
2100         int err, zero = 0;
2101         __u8 *data;
2102
2103         /* Nothing to do here since kernel already zero-initializes .bss map. */
2104         if (map->libbpf_type == LIBBPF_MAP_BSS)
2105                 return 0;
2106
2107         data = map->libbpf_type == LIBBPF_MAP_DATA ?
2108                obj->sections.data : obj->sections.rodata;
2109
2110         err = bpf_map_update_elem(map->fd, &zero, data, 0);
2111         /* Freeze .rodata map as read-only from syscall side. */
2112         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2113                 err = bpf_map_freeze(map->fd);
2114                 if (err) {
2115                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2116                         pr_warning("Error freezing map(%s) as read-only: %s\n",
2117                                    map->name, cp);
2118                         err = 0;
2119                 }
2120         }
2121         return err;
2122 }
2123
2124 static int
2125 bpf_object__create_maps(struct bpf_object *obj)
2126 {
2127         struct bpf_create_map_attr create_attr = {};
2128         int nr_cpus = 0;
2129         unsigned int i;
2130         int err;
2131
2132         for (i = 0; i < obj->nr_maps; i++) {
2133                 struct bpf_map *map = &obj->maps[i];
2134                 struct bpf_map_def *def = &map->def;
2135                 char *cp, errmsg[STRERR_BUFSIZE];
2136                 int *pfd = &map->fd;
2137
2138                 if (map->fd >= 0) {
2139                         pr_debug("skip map create (preset) %s: fd=%d\n",
2140                                  map->name, map->fd);
2141                         continue;
2142                 }
2143
2144                 if (obj->caps.name)
2145                         create_attr.name = map->name;
2146                 create_attr.map_ifindex = map->map_ifindex;
2147                 create_attr.map_type = def->type;
2148                 create_attr.map_flags = def->map_flags;
2149                 create_attr.key_size = def->key_size;
2150                 create_attr.value_size = def->value_size;
2151                 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2152                     !def->max_entries) {
2153                         if (!nr_cpus)
2154                                 nr_cpus = libbpf_num_possible_cpus();
2155                         if (nr_cpus < 0) {
2156                                 pr_warning("failed to determine number of system CPUs: %d\n",
2157                                            nr_cpus);
2158                                 err = nr_cpus;
2159                                 goto err_out;
2160                         }
2161                         pr_debug("map '%s': setting size to %d\n",
2162                                  map->name, nr_cpus);
2163                         create_attr.max_entries = nr_cpus;
2164                 } else {
2165                         create_attr.max_entries = def->max_entries;
2166                 }
2167                 create_attr.btf_fd = 0;
2168                 create_attr.btf_key_type_id = 0;
2169                 create_attr.btf_value_type_id = 0;
2170                 if (bpf_map_type__is_map_in_map(def->type) &&
2171                     map->inner_map_fd >= 0)
2172                         create_attr.inner_map_fd = map->inner_map_fd;
2173
2174                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2175                         create_attr.btf_fd = btf__fd(obj->btf);
2176                         create_attr.btf_key_type_id = map->btf_key_type_id;
2177                         create_attr.btf_value_type_id = map->btf_value_type_id;
2178                 }
2179
2180                 *pfd = bpf_create_map_xattr(&create_attr);
2181                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2182                                  create_attr.btf_value_type_id)) {
2183                         err = -errno;
2184                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2185                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2186                                    map->name, cp, err);
2187                         create_attr.btf_fd = 0;
2188                         create_attr.btf_key_type_id = 0;
2189                         create_attr.btf_value_type_id = 0;
2190                         map->btf_key_type_id = 0;
2191                         map->btf_value_type_id = 0;
2192                         *pfd = bpf_create_map_xattr(&create_attr);
2193                 }
2194
2195                 if (*pfd < 0) {
2196                         size_t j;
2197
2198                         err = -errno;
2199 err_out:
2200                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2201                         pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2202                                    map->name, cp, err);
2203                         for (j = 0; j < i; j++)
2204                                 zclose(obj->maps[j].fd);
2205                         return err;
2206                 }
2207
2208                 if (bpf_map__is_internal(map)) {
2209                         err = bpf_object__populate_internal_map(obj, map);
2210                         if (err < 0) {
2211                                 zclose(*pfd);
2212                                 goto err_out;
2213                         }
2214                 }
2215
2216                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2217         }
2218
2219         return 0;
2220 }
2221
2222 static int
2223 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2224                         void *btf_prog_info, const char *info_name)
2225 {
2226         if (err != -ENOENT) {
2227                 pr_warning("Error in loading %s for sec %s.\n",
2228                            info_name, prog->section_name);
2229                 return err;
2230         }
2231
2232         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2233
2234         if (btf_prog_info) {
2235                 /*
2236                  * Some info has already been found but has problem
2237                  * in the last btf_ext reloc. Must have to error out.
2238                  */
2239                 pr_warning("Error in relocating %s for sec %s.\n",
2240                            info_name, prog->section_name);
2241                 return err;
2242         }
2243
2244         /* Have problem loading the very first info. Ignore the rest. */
2245         pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2246                    info_name, prog->section_name, info_name);
2247         return 0;
2248 }
2249
2250 static int
2251 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2252                           const char *section_name,  __u32 insn_offset)
2253 {
2254         int err;
2255
2256         if (!insn_offset || prog->func_info) {
2257                 /*
2258                  * !insn_offset => main program
2259                  *
2260                  * For sub prog, the main program's func_info has to
2261                  * be loaded first (i.e. prog->func_info != NULL)
2262                  */
2263                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2264                                                section_name, insn_offset,
2265                                                &prog->func_info,
2266                                                &prog->func_info_cnt);
2267                 if (err)
2268                         return check_btf_ext_reloc_err(prog, err,
2269                                                        prog->func_info,
2270                                                        "bpf_func_info");
2271
2272                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2273         }
2274
2275         if (!insn_offset || prog->line_info) {
2276                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2277                                                section_name, insn_offset,
2278                                                &prog->line_info,
2279                                                &prog->line_info_cnt);
2280                 if (err)
2281                         return check_btf_ext_reloc_err(prog, err,
2282                                                        prog->line_info,
2283                                                        "bpf_line_info");
2284
2285                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2286         }
2287
2288         return 0;
2289 }
2290
2291 static int
2292 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
2293                         struct reloc_desc *relo)
2294 {
2295         struct bpf_insn *insn, *new_insn;
2296         struct bpf_program *text;
2297         size_t new_cnt;
2298         int err;
2299
2300         if (relo->type != RELO_CALL)
2301                 return -LIBBPF_ERRNO__RELOC;
2302
2303         if (prog->idx == obj->efile.text_shndx) {
2304                 pr_warning("relo in .text insn %d into off %d\n",
2305                            relo->insn_idx, relo->text_off);
2306                 return -LIBBPF_ERRNO__RELOC;
2307         }
2308
2309         if (prog->main_prog_cnt == 0) {
2310                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
2311                 if (!text) {
2312                         pr_warning("no .text section found yet relo into text exist\n");
2313                         return -LIBBPF_ERRNO__RELOC;
2314                 }
2315                 new_cnt = prog->insns_cnt + text->insns_cnt;
2316                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
2317                 if (!new_insn) {
2318                         pr_warning("oom in prog realloc\n");
2319                         return -ENOMEM;
2320                 }
2321
2322                 if (obj->btf_ext) {
2323                         err = bpf_program_reloc_btf_ext(prog, obj,
2324                                                         text->section_name,
2325                                                         prog->insns_cnt);
2326                         if (err)
2327                                 return err;
2328                 }
2329
2330                 memcpy(new_insn + prog->insns_cnt, text->insns,
2331                        text->insns_cnt * sizeof(*insn));
2332                 prog->insns = new_insn;
2333                 prog->main_prog_cnt = prog->insns_cnt;
2334                 prog->insns_cnt = new_cnt;
2335                 pr_debug("added %zd insn from %s to prog %s\n",
2336                          text->insns_cnt, text->section_name,
2337                          prog->section_name);
2338         }
2339         insn = &prog->insns[relo->insn_idx];
2340         insn->imm += prog->main_prog_cnt - relo->insn_idx;
2341         return 0;
2342 }
2343
2344 static int
2345 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
2346 {
2347         int i, err;
2348
2349         if (!prog)
2350                 return 0;
2351
2352         if (obj->btf_ext) {
2353                 err = bpf_program_reloc_btf_ext(prog, obj,
2354                                                 prog->section_name, 0);
2355                 if (err)
2356                         return err;
2357         }
2358
2359         if (!prog->reloc_desc)
2360                 return 0;
2361
2362         for (i = 0; i < prog->nr_reloc; i++) {
2363                 if (prog->reloc_desc[i].type == RELO_LD64 ||
2364                     prog->reloc_desc[i].type == RELO_DATA) {
2365                         bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
2366                         struct bpf_insn *insns = prog->insns;
2367                         int insn_idx, map_idx;
2368
2369                         insn_idx = prog->reloc_desc[i].insn_idx;
2370                         map_idx = prog->reloc_desc[i].map_idx;
2371
2372                         if (insn_idx + 1 >= (int)prog->insns_cnt) {
2373                                 pr_warning("relocation out of range: '%s'\n",
2374                                            prog->section_name);
2375                                 return -LIBBPF_ERRNO__RELOC;
2376                         }
2377
2378                         if (!relo_data) {
2379                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
2380                         } else {
2381                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
2382                                 insns[insn_idx + 1].imm = insns[insn_idx].imm;
2383                         }
2384                         insns[insn_idx].imm = obj->maps[map_idx].fd;
2385                 } else if (prog->reloc_desc[i].type == RELO_CALL) {
2386                         err = bpf_program__reloc_text(prog, obj,
2387                                                       &prog->reloc_desc[i]);
2388                         if (err)
2389                                 return err;
2390                 }
2391         }
2392
2393         zfree(&prog->reloc_desc);
2394         prog->nr_reloc = 0;
2395         return 0;
2396 }
2397
2398
2399 static int
2400 bpf_object__relocate(struct bpf_object *obj)
2401 {
2402         struct bpf_program *prog;
2403         size_t i;
2404         int err;
2405
2406         for (i = 0; i < obj->nr_programs; i++) {
2407                 prog = &obj->programs[i];
2408
2409                 err = bpf_program__relocate(prog, obj);
2410                 if (err) {
2411                         pr_warning("failed to relocate '%s'\n",
2412                                    prog->section_name);
2413                         return err;
2414                 }
2415         }
2416         return 0;
2417 }
2418
2419 static int bpf_object__collect_reloc(struct bpf_object *obj)
2420 {
2421         int i, err;
2422
2423         if (!obj_elf_valid(obj)) {
2424                 pr_warning("Internal error: elf object is closed\n");
2425                 return -LIBBPF_ERRNO__INTERNAL;
2426         }
2427
2428         for (i = 0; i < obj->efile.nr_reloc; i++) {
2429                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
2430                 Elf_Data *data = obj->efile.reloc[i].data;
2431                 int idx = shdr->sh_info;
2432                 struct bpf_program *prog;
2433
2434                 if (shdr->sh_type != SHT_REL) {
2435                         pr_warning("internal error at %d\n", __LINE__);
2436                         return -LIBBPF_ERRNO__INTERNAL;
2437                 }
2438
2439                 prog = bpf_object__find_prog_by_idx(obj, idx);
2440                 if (!prog) {
2441                         pr_warning("relocation failed: no section(%d)\n", idx);
2442                         return -LIBBPF_ERRNO__RELOC;
2443                 }
2444
2445                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
2446                 if (err)
2447                         return err;
2448         }
2449         return 0;
2450 }
2451
2452 static int
2453 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
2454              char *license, __u32 kern_version, int *pfd)
2455 {
2456         struct bpf_load_program_attr load_attr;
2457         char *cp, errmsg[STRERR_BUFSIZE];
2458         int log_buf_size = BPF_LOG_BUF_SIZE;
2459         char *log_buf;
2460         int btf_fd, ret;
2461
2462         if (!insns || !insns_cnt)
2463                 return -EINVAL;
2464
2465         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
2466         load_attr.prog_type = prog->type;
2467         load_attr.expected_attach_type = prog->expected_attach_type;
2468         if (prog->caps->name)
2469                 load_attr.name = prog->name;
2470         load_attr.insns = insns;
2471         load_attr.insns_cnt = insns_cnt;
2472         load_attr.license = license;
2473         load_attr.kern_version = kern_version;
2474         load_attr.prog_ifindex = prog->prog_ifindex;
2475         /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
2476         if (prog->obj->btf_ext)
2477                 btf_fd = bpf_object__btf_fd(prog->obj);
2478         else
2479                 btf_fd = -1;
2480         load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
2481         load_attr.func_info = prog->func_info;
2482         load_attr.func_info_rec_size = prog->func_info_rec_size;
2483         load_attr.func_info_cnt = prog->func_info_cnt;
2484         load_attr.line_info = prog->line_info;
2485         load_attr.line_info_rec_size = prog->line_info_rec_size;
2486         load_attr.line_info_cnt = prog->line_info_cnt;
2487         load_attr.log_level = prog->log_level;
2488         load_attr.prog_flags = prog->prog_flags;
2489
2490 retry_load:
2491         log_buf = malloc(log_buf_size);
2492         if (!log_buf)
2493                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
2494
2495         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
2496
2497         if (ret >= 0) {
2498                 if (load_attr.log_level)
2499                         pr_debug("verifier log:\n%s", log_buf);
2500                 *pfd = ret;
2501                 ret = 0;
2502                 goto out;
2503         }
2504
2505         if (errno == ENOSPC) {
2506                 log_buf_size <<= 1;
2507                 free(log_buf);
2508                 goto retry_load;
2509         }
2510         ret = -LIBBPF_ERRNO__LOAD;
2511         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2512         pr_warning("load bpf program failed: %s\n", cp);
2513
2514         if (log_buf && log_buf[0] != '\0') {
2515                 ret = -LIBBPF_ERRNO__VERIFY;
2516                 pr_warning("-- BEGIN DUMP LOG ---\n");
2517                 pr_warning("\n%s\n", log_buf);
2518                 pr_warning("-- END LOG --\n");
2519         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
2520                 pr_warning("Program too large (%zu insns), at most %d insns\n",
2521                            load_attr.insns_cnt, BPF_MAXINSNS);
2522                 ret = -LIBBPF_ERRNO__PROG2BIG;
2523         } else {
2524                 /* Wrong program type? */
2525                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
2526                         int fd;
2527
2528                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
2529                         load_attr.expected_attach_type = 0;
2530                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
2531                         if (fd >= 0) {
2532                                 close(fd);
2533                                 ret = -LIBBPF_ERRNO__PROGTYPE;
2534                                 goto out;
2535                         }
2536                 }
2537
2538                 if (log_buf)
2539                         ret = -LIBBPF_ERRNO__KVER;
2540         }
2541
2542 out:
2543         free(log_buf);
2544         return ret;
2545 }
2546
2547 int
2548 bpf_program__load(struct bpf_program *prog,
2549                   char *license, __u32 kern_version)
2550 {
2551         int err = 0, fd, i;
2552
2553         if (prog->instances.nr < 0 || !prog->instances.fds) {
2554                 if (prog->preprocessor) {
2555                         pr_warning("Internal error: can't load program '%s'\n",
2556                                    prog->section_name);
2557                         return -LIBBPF_ERRNO__INTERNAL;
2558                 }
2559
2560                 prog->instances.fds = malloc(sizeof(int));
2561                 if (!prog->instances.fds) {
2562                         pr_warning("Not enough memory for BPF fds\n");
2563                         return -ENOMEM;
2564                 }
2565                 prog->instances.nr = 1;
2566                 prog->instances.fds[0] = -1;
2567         }
2568
2569         if (!prog->preprocessor) {
2570                 if (prog->instances.nr != 1) {
2571                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
2572                                    prog->section_name, prog->instances.nr);
2573                 }
2574                 err = load_program(prog, prog->insns, prog->insns_cnt,
2575                                    license, kern_version, &fd);
2576                 if (!err)
2577                         prog->instances.fds[0] = fd;
2578                 goto out;
2579         }
2580
2581         for (i = 0; i < prog->instances.nr; i++) {
2582                 struct bpf_prog_prep_result result;
2583                 bpf_program_prep_t preprocessor = prog->preprocessor;
2584
2585                 memset(&result, 0, sizeof(result));
2586                 err = preprocessor(prog, i, prog->insns,
2587                                    prog->insns_cnt, &result);
2588                 if (err) {
2589                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
2590                                    i, prog->section_name);
2591                         goto out;
2592                 }
2593
2594                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
2595                         pr_debug("Skip loading the %dth instance of program '%s'\n",
2596                                  i, prog->section_name);
2597                         prog->instances.fds[i] = -1;
2598                         if (result.pfd)
2599                                 *result.pfd = -1;
2600                         continue;
2601                 }
2602
2603                 err = load_program(prog, result.new_insn_ptr,
2604                                    result.new_insn_cnt,
2605                                    license, kern_version, &fd);
2606
2607                 if (err) {
2608                         pr_warning("Loading the %dth instance of program '%s' failed\n",
2609                                         i, prog->section_name);
2610                         goto out;
2611                 }
2612
2613                 if (result.pfd)
2614                         *result.pfd = fd;
2615                 prog->instances.fds[i] = fd;
2616         }
2617 out:
2618         if (err)
2619                 pr_warning("failed to load program '%s'\n",
2620                            prog->section_name);
2621         zfree(&prog->insns);
2622         prog->insns_cnt = 0;
2623         return err;
2624 }
2625
2626 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
2627                                              const struct bpf_object *obj)
2628 {
2629         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
2630 }
2631
2632 static int
2633 bpf_object__load_progs(struct bpf_object *obj, int log_level)
2634 {
2635         size_t i;
2636         int err;
2637
2638         for (i = 0; i < obj->nr_programs; i++) {
2639                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
2640                         continue;
2641                 obj->programs[i].log_level |= log_level;
2642                 err = bpf_program__load(&obj->programs[i],
2643                                         obj->license,
2644                                         obj->kern_version);
2645                 if (err)
2646                         return err;
2647         }
2648         return 0;
2649 }
2650
2651 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
2652 {
2653         switch (type) {
2654         case BPF_PROG_TYPE_SOCKET_FILTER:
2655         case BPF_PROG_TYPE_SCHED_CLS:
2656         case BPF_PROG_TYPE_SCHED_ACT:
2657         case BPF_PROG_TYPE_XDP:
2658         case BPF_PROG_TYPE_CGROUP_SKB:
2659         case BPF_PROG_TYPE_CGROUP_SOCK:
2660         case BPF_PROG_TYPE_LWT_IN:
2661         case BPF_PROG_TYPE_LWT_OUT:
2662         case BPF_PROG_TYPE_LWT_XMIT:
2663         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2664         case BPF_PROG_TYPE_SOCK_OPS:
2665         case BPF_PROG_TYPE_SK_SKB:
2666         case BPF_PROG_TYPE_CGROUP_DEVICE:
2667         case BPF_PROG_TYPE_SK_MSG:
2668         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2669         case BPF_PROG_TYPE_LIRC_MODE2:
2670         case BPF_PROG_TYPE_SK_REUSEPORT:
2671         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2672         case BPF_PROG_TYPE_UNSPEC:
2673         case BPF_PROG_TYPE_TRACEPOINT:
2674         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2675         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2676         case BPF_PROG_TYPE_PERF_EVENT:
2677         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2678         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2679                 return false;
2680         case BPF_PROG_TYPE_KPROBE:
2681         default:
2682                 return true;
2683         }
2684 }
2685
2686 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
2687 {
2688         if (needs_kver && obj->kern_version == 0) {
2689                 pr_warning("%s doesn't provide kernel version\n",
2690                            obj->path);
2691                 return -LIBBPF_ERRNO__KVERSION;
2692         }
2693         return 0;
2694 }
2695
2696 static struct bpf_object *
2697 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
2698                    bool needs_kver, int flags)
2699 {
2700         struct bpf_object *obj;
2701         int err;
2702
2703         if (elf_version(EV_CURRENT) == EV_NONE) {
2704                 pr_warning("failed to init libelf for %s\n", path);
2705                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
2706         }
2707
2708         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
2709         if (IS_ERR(obj))
2710                 return obj;
2711
2712         CHECK_ERR(bpf_object__elf_init(obj), err, out);
2713         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
2714         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
2715         CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
2716         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
2717         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
2718
2719         bpf_object__elf_finish(obj);
2720         return obj;
2721 out:
2722         bpf_object__close(obj);
2723         return ERR_PTR(err);
2724 }
2725
2726 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
2727                                             int flags)
2728 {
2729         /* param validation */
2730         if (!attr->file)
2731                 return NULL;
2732
2733         pr_debug("loading %s\n", attr->file);
2734
2735         return __bpf_object__open(attr->file, NULL, 0,
2736                                   bpf_prog_type__needs_kver(attr->prog_type),
2737                                   flags);
2738 }
2739
2740 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
2741 {
2742         return __bpf_object__open_xattr(attr, 0);
2743 }
2744
2745 struct bpf_object *bpf_object__open(const char *path)
2746 {
2747         struct bpf_object_open_attr attr = {
2748                 .file           = path,
2749                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
2750         };
2751
2752         return bpf_object__open_xattr(&attr);
2753 }
2754
2755 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
2756                                            size_t obj_buf_sz,
2757                                            const char *name)
2758 {
2759         char tmp_name[64];
2760
2761         /* param validation */
2762         if (!obj_buf || obj_buf_sz <= 0)
2763                 return NULL;
2764
2765         if (!name) {
2766                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
2767                          (unsigned long)obj_buf,
2768                          (unsigned long)obj_buf_sz);
2769                 name = tmp_name;
2770         }
2771         pr_debug("loading object '%s' from buffer\n", name);
2772
2773         return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
2774 }
2775
2776 int bpf_object__unload(struct bpf_object *obj)
2777 {
2778         size_t i;
2779
2780         if (!obj)
2781                 return -EINVAL;
2782
2783         for (i = 0; i < obj->nr_maps; i++)
2784                 zclose(obj->maps[i].fd);
2785
2786         for (i = 0; i < obj->nr_programs; i++)
2787                 bpf_program__unload(&obj->programs[i]);
2788
2789         return 0;
2790 }
2791
2792 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
2793 {
2794         struct bpf_object *obj;
2795         int err;
2796
2797         if (!attr)
2798                 return -EINVAL;
2799         obj = attr->obj;
2800         if (!obj)
2801                 return -EINVAL;
2802
2803         if (obj->loaded) {
2804                 pr_warning("object should not be loaded twice\n");
2805                 return -EINVAL;
2806         }
2807
2808         obj->loaded = true;
2809
2810         CHECK_ERR(bpf_object__create_maps(obj), err, out);
2811         CHECK_ERR(bpf_object__relocate(obj), err, out);
2812         CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
2813
2814         return 0;
2815 out:
2816         bpf_object__unload(obj);
2817         pr_warning("failed to load object '%s'\n", obj->path);
2818         return err;
2819 }
2820
2821 int bpf_object__load(struct bpf_object *obj)
2822 {
2823         struct bpf_object_load_attr attr = {
2824                 .obj = obj,
2825         };
2826
2827         return bpf_object__load_xattr(&attr);
2828 }
2829
2830 static int check_path(const char *path)
2831 {
2832         char *cp, errmsg[STRERR_BUFSIZE];
2833         struct statfs st_fs;
2834         char *dname, *dir;
2835         int err = 0;
2836
2837         if (path == NULL)
2838                 return -EINVAL;
2839
2840         dname = strdup(path);
2841         if (dname == NULL)
2842                 return -ENOMEM;
2843
2844         dir = dirname(dname);
2845         if (statfs(dir, &st_fs)) {
2846                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2847                 pr_warning("failed to statfs %s: %s\n", dir, cp);
2848                 err = -errno;
2849         }
2850         free(dname);
2851
2852         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
2853                 pr_warning("specified path %s is not on BPF FS\n", path);
2854                 err = -EINVAL;
2855         }
2856
2857         return err;
2858 }
2859
2860 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
2861                               int instance)
2862 {
2863         char *cp, errmsg[STRERR_BUFSIZE];
2864         int err;
2865
2866         err = check_path(path);
2867         if (err)
2868                 return err;
2869
2870         if (prog == NULL) {
2871                 pr_warning("invalid program pointer\n");
2872                 return -EINVAL;
2873         }
2874
2875         if (instance < 0 || instance >= prog->instances.nr) {
2876                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2877                            instance, prog->section_name, prog->instances.nr);
2878                 return -EINVAL;
2879         }
2880
2881         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
2882                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2883                 pr_warning("failed to pin program: %s\n", cp);
2884                 return -errno;
2885         }
2886         pr_debug("pinned program '%s'\n", path);
2887
2888         return 0;
2889 }
2890
2891 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
2892                                 int instance)
2893 {
2894         int err;
2895
2896         err = check_path(path);
2897         if (err)
2898                 return err;
2899
2900         if (prog == NULL) {
2901                 pr_warning("invalid program pointer\n");
2902                 return -EINVAL;
2903         }
2904
2905         if (instance < 0 || instance >= prog->instances.nr) {
2906                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2907                            instance, prog->section_name, prog->instances.nr);
2908                 return -EINVAL;
2909         }
2910
2911         err = unlink(path);
2912         if (err != 0)
2913                 return -errno;
2914         pr_debug("unpinned program '%s'\n", path);
2915
2916         return 0;
2917 }
2918
2919 static int make_dir(const char *path)
2920 {
2921         char *cp, errmsg[STRERR_BUFSIZE];
2922         int err = 0;
2923
2924         if (mkdir(path, 0700) && errno != EEXIST)
2925                 err = -errno;
2926
2927         if (err) {
2928                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
2929                 pr_warning("failed to mkdir %s: %s\n", path, cp);
2930         }
2931         return err;
2932 }
2933
2934 int bpf_program__pin(struct bpf_program *prog, const char *path)
2935 {
2936         int i, err;
2937
2938         err = check_path(path);
2939         if (err)
2940                 return err;
2941
2942         if (prog == NULL) {
2943                 pr_warning("invalid program pointer\n");
2944                 return -EINVAL;
2945         }
2946
2947         if (prog->instances.nr <= 0) {
2948                 pr_warning("no instances of prog %s to pin\n",
2949                            prog->section_name);
2950                 return -EINVAL;
2951         }
2952
2953         if (prog->instances.nr == 1) {
2954                 /* don't create subdirs when pinning single instance */
2955                 return bpf_program__pin_instance(prog, path, 0);
2956         }
2957
2958         err = make_dir(path);
2959         if (err)
2960                 return err;
2961
2962         for (i = 0; i < prog->instances.nr; i++) {
2963                 char buf[PATH_MAX];
2964                 int len;
2965
2966                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2967                 if (len < 0) {
2968                         err = -EINVAL;
2969                         goto err_unpin;
2970                 } else if (len >= PATH_MAX) {
2971                         err = -ENAMETOOLONG;
2972                         goto err_unpin;
2973                 }
2974
2975                 err = bpf_program__pin_instance(prog, buf, i);
2976                 if (err)
2977                         goto err_unpin;
2978         }
2979
2980         return 0;
2981
2982 err_unpin:
2983         for (i = i - 1; i >= 0; i--) {
2984                 char buf[PATH_MAX];
2985                 int len;
2986
2987                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2988                 if (len < 0)
2989                         continue;
2990                 else if (len >= PATH_MAX)
2991                         continue;
2992
2993                 bpf_program__unpin_instance(prog, buf, i);
2994         }
2995
2996         rmdir(path);
2997
2998         return err;
2999 }
3000
3001 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3002 {
3003         int i, err;
3004
3005         err = check_path(path);
3006         if (err)
3007                 return err;
3008
3009         if (prog == NULL) {
3010                 pr_warning("invalid program pointer\n");
3011                 return -EINVAL;
3012         }
3013
3014         if (prog->instances.nr <= 0) {
3015                 pr_warning("no instances of prog %s to pin\n",
3016                            prog->section_name);
3017                 return -EINVAL;
3018         }
3019
3020         if (prog->instances.nr == 1) {
3021                 /* don't create subdirs when pinning single instance */
3022                 return bpf_program__unpin_instance(prog, path, 0);
3023         }
3024
3025         for (i = 0; i < prog->instances.nr; i++) {
3026                 char buf[PATH_MAX];
3027                 int len;
3028
3029                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3030                 if (len < 0)
3031                         return -EINVAL;
3032                 else if (len >= PATH_MAX)
3033                         return -ENAMETOOLONG;
3034
3035                 err = bpf_program__unpin_instance(prog, buf, i);
3036                 if (err)
3037                         return err;
3038         }
3039
3040         err = rmdir(path);
3041         if (err)
3042                 return -errno;
3043
3044         return 0;
3045 }
3046
3047 int bpf_map__pin(struct bpf_map *map, const char *path)
3048 {
3049         char *cp, errmsg[STRERR_BUFSIZE];
3050         int err;
3051
3052         err = check_path(path);
3053         if (err)
3054                 return err;
3055
3056         if (map == NULL) {
3057                 pr_warning("invalid map pointer\n");
3058                 return -EINVAL;
3059         }
3060
3061         if (bpf_obj_pin(map->fd, path)) {
3062                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3063                 pr_warning("failed to pin map: %s\n", cp);
3064                 return -errno;
3065         }
3066
3067         pr_debug("pinned map '%s'\n", path);
3068
3069         return 0;
3070 }
3071
3072 int bpf_map__unpin(struct bpf_map *map, const char *path)
3073 {
3074         int err;
3075
3076         err = check_path(path);
3077         if (err)
3078                 return err;
3079
3080         if (map == NULL) {
3081                 pr_warning("invalid map pointer\n");
3082                 return -EINVAL;
3083         }
3084
3085         err = unlink(path);
3086         if (err != 0)
3087                 return -errno;
3088         pr_debug("unpinned map '%s'\n", path);
3089
3090         return 0;
3091 }
3092
3093 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
3094 {
3095         struct bpf_map *map;
3096         int err;
3097
3098         if (!obj)
3099                 return -ENOENT;
3100
3101         if (!obj->loaded) {
3102                 pr_warning("object not yet loaded; load it first\n");
3103                 return -ENOENT;
3104         }
3105
3106         err = make_dir(path);
3107         if (err)
3108                 return err;
3109
3110         bpf_object__for_each_map(map, obj) {
3111                 char buf[PATH_MAX];
3112                 int len;
3113
3114                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3115                                bpf_map__name(map));
3116                 if (len < 0) {
3117                         err = -EINVAL;
3118                         goto err_unpin_maps;
3119                 } else if (len >= PATH_MAX) {
3120                         err = -ENAMETOOLONG;
3121                         goto err_unpin_maps;
3122                 }
3123
3124                 err = bpf_map__pin(map, buf);
3125                 if (err)
3126                         goto err_unpin_maps;
3127         }
3128
3129         return 0;
3130
3131 err_unpin_maps:
3132         while ((map = bpf_map__prev(map, obj))) {
3133                 char buf[PATH_MAX];
3134                 int len;
3135
3136                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3137                                bpf_map__name(map));
3138                 if (len < 0)
3139                         continue;
3140                 else if (len >= PATH_MAX)
3141                         continue;
3142
3143                 bpf_map__unpin(map, buf);
3144         }
3145
3146         return err;
3147 }
3148
3149 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
3150 {
3151         struct bpf_map *map;
3152         int err;
3153
3154         if (!obj)
3155                 return -ENOENT;
3156
3157         bpf_object__for_each_map(map, obj) {
3158                 char buf[PATH_MAX];
3159                 int len;
3160
3161                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3162                                bpf_map__name(map));
3163                 if (len < 0)
3164                         return -EINVAL;
3165                 else if (len >= PATH_MAX)
3166                         return -ENAMETOOLONG;
3167
3168                 err = bpf_map__unpin(map, buf);
3169                 if (err)
3170                         return err;
3171         }
3172
3173         return 0;
3174 }
3175
3176 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
3177 {
3178         struct bpf_program *prog;
3179         int err;
3180
3181         if (!obj)
3182                 return -ENOENT;
3183
3184         if (!obj->loaded) {
3185                 pr_warning("object not yet loaded; load it first\n");
3186                 return -ENOENT;
3187         }
3188
3189         err = make_dir(path);
3190         if (err)
3191                 return err;
3192
3193         bpf_object__for_each_program(prog, obj) {
3194                 char buf[PATH_MAX];
3195                 int len;
3196
3197                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3198                                prog->pin_name);
3199                 if (len < 0) {
3200                         err = -EINVAL;
3201                         goto err_unpin_programs;
3202                 } else if (len >= PATH_MAX) {
3203                         err = -ENAMETOOLONG;
3204                         goto err_unpin_programs;
3205                 }
3206
3207                 err = bpf_program__pin(prog, buf);
3208                 if (err)
3209                         goto err_unpin_programs;
3210         }
3211
3212         return 0;
3213
3214 err_unpin_programs:
3215         while ((prog = bpf_program__prev(prog, obj))) {
3216                 char buf[PATH_MAX];
3217                 int len;
3218
3219                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3220                                prog->pin_name);
3221                 if (len < 0)
3222                         continue;
3223                 else if (len >= PATH_MAX)
3224                         continue;
3225
3226                 bpf_program__unpin(prog, buf);
3227         }
3228
3229         return err;
3230 }
3231
3232 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
3233 {
3234         struct bpf_program *prog;
3235         int err;
3236
3237         if (!obj)
3238                 return -ENOENT;
3239
3240         bpf_object__for_each_program(prog, obj) {
3241                 char buf[PATH_MAX];
3242                 int len;
3243
3244                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3245                                prog->pin_name);
3246                 if (len < 0)
3247                         return -EINVAL;
3248                 else if (len >= PATH_MAX)
3249                         return -ENAMETOOLONG;
3250
3251                 err = bpf_program__unpin(prog, buf);
3252                 if (err)
3253                         return err;
3254         }
3255
3256         return 0;
3257 }
3258
3259 int bpf_object__pin(struct bpf_object *obj, const char *path)
3260 {
3261         int err;
3262
3263         err = bpf_object__pin_maps(obj, path);
3264         if (err)
3265                 return err;
3266
3267         err = bpf_object__pin_programs(obj, path);
3268         if (err) {
3269                 bpf_object__unpin_maps(obj, path);
3270                 return err;
3271         }
3272
3273         return 0;
3274 }
3275
3276 void bpf_object__close(struct bpf_object *obj)
3277 {
3278         size_t i;
3279
3280         if (!obj)
3281                 return;
3282
3283         if (obj->clear_priv)
3284                 obj->clear_priv(obj, obj->priv);
3285
3286         bpf_object__elf_finish(obj);
3287         bpf_object__unload(obj);
3288         btf__free(obj->btf);
3289         btf_ext__free(obj->btf_ext);
3290
3291         for (i = 0; i < obj->nr_maps; i++) {
3292                 zfree(&obj->maps[i].name);
3293                 if (obj->maps[i].clear_priv)
3294                         obj->maps[i].clear_priv(&obj->maps[i],
3295                                                 obj->maps[i].priv);
3296                 obj->maps[i].priv = NULL;
3297                 obj->maps[i].clear_priv = NULL;
3298         }
3299
3300         zfree(&obj->sections.rodata);
3301         zfree(&obj->sections.data);
3302         zfree(&obj->maps);
3303         obj->nr_maps = 0;
3304
3305         if (obj->programs && obj->nr_programs) {
3306                 for (i = 0; i < obj->nr_programs; i++)
3307                         bpf_program__exit(&obj->programs[i]);
3308         }
3309         zfree(&obj->programs);
3310
3311         list_del(&obj->list);
3312         free(obj);
3313 }
3314
3315 struct bpf_object *
3316 bpf_object__next(struct bpf_object *prev)
3317 {
3318         struct bpf_object *next;
3319
3320         if (!prev)
3321                 next = list_first_entry(&bpf_objects_list,
3322                                         struct bpf_object,
3323                                         list);
3324         else
3325                 next = list_next_entry(prev, list);
3326
3327         /* Empty list is noticed here so don't need checking on entry. */
3328         if (&next->list == &bpf_objects_list)
3329                 return NULL;
3330
3331         return next;
3332 }
3333
3334 const char *bpf_object__name(const struct bpf_object *obj)
3335 {
3336         return obj ? obj->path : ERR_PTR(-EINVAL);
3337 }
3338
3339 unsigned int bpf_object__kversion(const struct bpf_object *obj)
3340 {
3341         return obj ? obj->kern_version : 0;
3342 }
3343
3344 struct btf *bpf_object__btf(const struct bpf_object *obj)
3345 {
3346         return obj ? obj->btf : NULL;
3347 }
3348
3349 int bpf_object__btf_fd(const struct bpf_object *obj)
3350 {
3351         return obj->btf ? btf__fd(obj->btf) : -1;
3352 }
3353
3354 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
3355                          bpf_object_clear_priv_t clear_priv)
3356 {
3357         if (obj->priv && obj->clear_priv)
3358                 obj->clear_priv(obj, obj->priv);
3359
3360         obj->priv = priv;
3361         obj->clear_priv = clear_priv;
3362         return 0;
3363 }
3364
3365 void *bpf_object__priv(const struct bpf_object *obj)
3366 {
3367         return obj ? obj->priv : ERR_PTR(-EINVAL);
3368 }
3369
3370 static struct bpf_program *
3371 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
3372                     bool forward)
3373 {
3374         size_t nr_programs = obj->nr_programs;
3375         ssize_t idx;
3376
3377         if (!nr_programs)
3378                 return NULL;
3379
3380         if (!p)
3381                 /* Iter from the beginning */
3382                 return forward ? &obj->programs[0] :
3383                         &obj->programs[nr_programs - 1];
3384
3385         if (p->obj != obj) {
3386                 pr_warning("error: program handler doesn't match object\n");
3387                 return NULL;
3388         }
3389
3390         idx = (p - obj->programs) + (forward ? 1 : -1);
3391         if (idx >= obj->nr_programs || idx < 0)
3392                 return NULL;
3393         return &obj->programs[idx];
3394 }
3395
3396 struct bpf_program *
3397 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
3398 {
3399         struct bpf_program *prog = prev;
3400
3401         do {
3402                 prog = __bpf_program__iter(prog, obj, true);
3403         } while (prog && bpf_program__is_function_storage(prog, obj));
3404
3405         return prog;
3406 }
3407
3408 struct bpf_program *
3409 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
3410 {
3411         struct bpf_program *prog = next;
3412
3413         do {
3414                 prog = __bpf_program__iter(prog, obj, false);
3415         } while (prog && bpf_program__is_function_storage(prog, obj));
3416
3417         return prog;
3418 }
3419
3420 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
3421                           bpf_program_clear_priv_t clear_priv)
3422 {
3423         if (prog->priv && prog->clear_priv)
3424                 prog->clear_priv(prog, prog->priv);
3425
3426         prog->priv = priv;
3427         prog->clear_priv = clear_priv;
3428         return 0;
3429 }
3430
3431 void *bpf_program__priv(const struct bpf_program *prog)
3432 {
3433         return prog ? prog->priv : ERR_PTR(-EINVAL);
3434 }
3435
3436 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
3437 {
3438         prog->prog_ifindex = ifindex;
3439 }
3440
3441 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
3442 {
3443         const char *title;
3444
3445         title = prog->section_name;
3446         if (needs_copy) {
3447                 title = strdup(title);
3448                 if (!title) {
3449                         pr_warning("failed to strdup program title\n");
3450                         return ERR_PTR(-ENOMEM);
3451                 }
3452         }
3453
3454         return title;
3455 }
3456
3457 int bpf_program__fd(const struct bpf_program *prog)
3458 {
3459         return bpf_program__nth_fd(prog, 0);
3460 }
3461
3462 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
3463                           bpf_program_prep_t prep)
3464 {
3465         int *instances_fds;
3466
3467         if (nr_instances <= 0 || !prep)
3468                 return -EINVAL;
3469
3470         if (prog->instances.nr > 0 || prog->instances.fds) {
3471                 pr_warning("Can't set pre-processor after loading\n");
3472                 return -EINVAL;
3473         }
3474
3475         instances_fds = malloc(sizeof(int) * nr_instances);
3476         if (!instances_fds) {
3477                 pr_warning("alloc memory failed for fds\n");
3478                 return -ENOMEM;
3479         }
3480
3481         /* fill all fd with -1 */
3482         memset(instances_fds, -1, sizeof(int) * nr_instances);
3483
3484         prog->instances.nr = nr_instances;
3485         prog->instances.fds = instances_fds;
3486         prog->preprocessor = prep;
3487         return 0;
3488 }
3489
3490 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
3491 {
3492         int fd;
3493
3494         if (!prog)
3495                 return -EINVAL;
3496
3497         if (n >= prog->instances.nr || n < 0) {
3498                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
3499                            n, prog->section_name, prog->instances.nr);
3500                 return -EINVAL;
3501         }
3502
3503         fd = prog->instances.fds[n];
3504         if (fd < 0) {
3505                 pr_warning("%dth instance of program '%s' is invalid\n",
3506                            n, prog->section_name);
3507                 return -ENOENT;
3508         }
3509
3510         return fd;
3511 }
3512
3513 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
3514 {
3515         prog->type = type;
3516 }
3517
3518 static bool bpf_program__is_type(const struct bpf_program *prog,
3519                                  enum bpf_prog_type type)
3520 {
3521         return prog ? (prog->type == type) : false;
3522 }
3523
3524 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
3525 int bpf_program__set_##NAME(struct bpf_program *prog)           \
3526 {                                                               \
3527         if (!prog)                                              \
3528                 return -EINVAL;                                 \
3529         bpf_program__set_type(prog, TYPE);                      \
3530         return 0;                                               \
3531 }                                                               \
3532                                                                 \
3533 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
3534 {                                                               \
3535         return bpf_program__is_type(prog, TYPE);                \
3536 }                                                               \
3537
3538 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
3539 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
3540 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
3541 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
3542 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
3543 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
3544 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
3545 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
3546
3547 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
3548                                            enum bpf_attach_type type)
3549 {
3550         prog->expected_attach_type = type;
3551 }
3552
3553 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
3554         { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
3555
3556 /* Programs that can NOT be attached. */
3557 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
3558
3559 /* Programs that can be attached. */
3560 #define BPF_APROG_SEC(string, ptype, atype) \
3561         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
3562
3563 /* Programs that must specify expected attach type at load time. */
3564 #define BPF_EAPROG_SEC(string, ptype, eatype) \
3565         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
3566
3567 /* Programs that can be attached but attach type can't be identified by section
3568  * name. Kept for backward compatibility.
3569  */
3570 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
3571
3572 static const struct {
3573         const char *sec;
3574         size_t len;
3575         enum bpf_prog_type prog_type;
3576         enum bpf_attach_type expected_attach_type;
3577         int is_attachable;
3578         enum bpf_attach_type attach_type;
3579 } section_names[] = {
3580         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
3581         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
3582         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
3583         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
3584         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
3585         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
3586         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
3587         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
3588         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
3589         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
3590         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
3591         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
3592         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
3593         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
3594                                                 BPF_CGROUP_INET_INGRESS),
3595         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
3596                                                 BPF_CGROUP_INET_EGRESS),
3597         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
3598         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
3599                                                 BPF_CGROUP_INET_SOCK_CREATE),
3600         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
3601                                                 BPF_CGROUP_INET4_POST_BIND),
3602         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
3603                                                 BPF_CGROUP_INET6_POST_BIND),
3604         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
3605                                                 BPF_CGROUP_DEVICE),
3606         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
3607                                                 BPF_CGROUP_SOCK_OPS),
3608         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
3609                                                 BPF_SK_SKB_STREAM_PARSER),
3610         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
3611                                                 BPF_SK_SKB_STREAM_VERDICT),
3612         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
3613         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
3614                                                 BPF_SK_MSG_VERDICT),
3615         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
3616                                                 BPF_LIRC_MODE2),
3617         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
3618                                                 BPF_FLOW_DISSECTOR),
3619         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3620                                                 BPF_CGROUP_INET4_BIND),
3621         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3622                                                 BPF_CGROUP_INET6_BIND),
3623         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3624                                                 BPF_CGROUP_INET4_CONNECT),
3625         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3626                                                 BPF_CGROUP_INET6_CONNECT),
3627         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3628                                                 BPF_CGROUP_UDP4_SENDMSG),
3629         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3630                                                 BPF_CGROUP_UDP6_SENDMSG),
3631         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3632                                                 BPF_CGROUP_UDP4_RECVMSG),
3633         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3634                                                 BPF_CGROUP_UDP6_RECVMSG),
3635         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
3636                                                 BPF_CGROUP_SYSCTL),
3637         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
3638                                                 BPF_CGROUP_GETSOCKOPT),
3639         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
3640                                                 BPF_CGROUP_SETSOCKOPT),
3641 };
3642
3643 #undef BPF_PROG_SEC_IMPL
3644 #undef BPF_PROG_SEC
3645 #undef BPF_APROG_SEC
3646 #undef BPF_EAPROG_SEC
3647 #undef BPF_APROG_COMPAT
3648
3649 #define MAX_TYPE_NAME_SIZE 32
3650
3651 static char *libbpf_get_type_names(bool attach_type)
3652 {
3653         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
3654         char *buf;
3655
3656         buf = malloc(len);
3657         if (!buf)
3658                 return NULL;
3659
3660         buf[0] = '\0';
3661         /* Forge string buf with all available names */
3662         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3663                 if (attach_type && !section_names[i].is_attachable)
3664                         continue;
3665
3666                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
3667                         free(buf);
3668                         return NULL;
3669                 }
3670                 strcat(buf, " ");
3671                 strcat(buf, section_names[i].sec);
3672         }
3673
3674         return buf;
3675 }
3676
3677 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
3678                              enum bpf_attach_type *expected_attach_type)
3679 {
3680         char *type_names;
3681         int i;
3682
3683         if (!name)
3684                 return -EINVAL;
3685
3686         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3687                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3688                         continue;
3689                 *prog_type = section_names[i].prog_type;
3690                 *expected_attach_type = section_names[i].expected_attach_type;
3691                 return 0;
3692         }
3693         pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
3694         type_names = libbpf_get_type_names(false);
3695         if (type_names != NULL) {
3696                 pr_info("supported section(type) names are:%s\n", type_names);
3697                 free(type_names);
3698         }
3699
3700         return -EINVAL;
3701 }
3702
3703 int libbpf_attach_type_by_name(const char *name,
3704                                enum bpf_attach_type *attach_type)
3705 {
3706         char *type_names;
3707         int i;
3708
3709         if (!name)
3710                 return -EINVAL;
3711
3712         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3713                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3714                         continue;
3715                 if (!section_names[i].is_attachable)
3716                         return -EINVAL;
3717                 *attach_type = section_names[i].attach_type;
3718                 return 0;
3719         }
3720         pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
3721         type_names = libbpf_get_type_names(true);
3722         if (type_names != NULL) {
3723                 pr_info("attachable section(type) names are:%s\n", type_names);
3724                 free(type_names);
3725         }
3726
3727         return -EINVAL;
3728 }
3729
3730 static int
3731 bpf_program__identify_section(struct bpf_program *prog,
3732                               enum bpf_prog_type *prog_type,
3733                               enum bpf_attach_type *expected_attach_type)
3734 {
3735         return libbpf_prog_type_by_name(prog->section_name, prog_type,
3736                                         expected_attach_type);
3737 }
3738
3739 int bpf_map__fd(const struct bpf_map *map)
3740 {
3741         return map ? map->fd : -EINVAL;
3742 }
3743
3744 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
3745 {
3746         return map ? &map->def : ERR_PTR(-EINVAL);
3747 }
3748
3749 const char *bpf_map__name(const struct bpf_map *map)
3750 {
3751         return map ? map->name : NULL;
3752 }
3753
3754 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
3755 {
3756         return map ? map->btf_key_type_id : 0;
3757 }
3758
3759 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
3760 {
3761         return map ? map->btf_value_type_id : 0;
3762 }
3763
3764 int bpf_map__set_priv(struct bpf_map *map, void *priv,
3765                      bpf_map_clear_priv_t clear_priv)
3766 {
3767         if (!map)
3768                 return -EINVAL;
3769
3770         if (map->priv) {
3771                 if (map->clear_priv)
3772                         map->clear_priv(map, map->priv);
3773         }
3774
3775         map->priv = priv;
3776         map->clear_priv = clear_priv;
3777         return 0;
3778 }
3779
3780 void *bpf_map__priv(const struct bpf_map *map)
3781 {
3782         return map ? map->priv : ERR_PTR(-EINVAL);
3783 }
3784
3785 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
3786 {
3787         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
3788 }
3789
3790 bool bpf_map__is_internal(const struct bpf_map *map)
3791 {
3792         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
3793 }
3794
3795 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
3796 {
3797         map->map_ifindex = ifindex;
3798 }
3799
3800 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
3801 {
3802         if (!bpf_map_type__is_map_in_map(map->def.type)) {
3803                 pr_warning("error: unsupported map type\n");
3804                 return -EINVAL;
3805         }
3806         if (map->inner_map_fd != -1) {
3807                 pr_warning("error: inner_map_fd already specified\n");
3808                 return -EINVAL;
3809         }
3810         map->inner_map_fd = fd;
3811         return 0;
3812 }
3813
3814 static struct bpf_map *
3815 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
3816 {
3817         ssize_t idx;
3818         struct bpf_map *s, *e;
3819
3820         if (!obj || !obj->maps)
3821                 return NULL;
3822
3823         s = obj->maps;
3824         e = obj->maps + obj->nr_maps;
3825
3826         if ((m < s) || (m >= e)) {
3827                 pr_warning("error in %s: map handler doesn't belong to object\n",
3828                            __func__);
3829                 return NULL;
3830         }
3831
3832         idx = (m - obj->maps) + i;
3833         if (idx >= obj->nr_maps || idx < 0)
3834                 return NULL;
3835         return &obj->maps[idx];
3836 }
3837
3838 struct bpf_map *
3839 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
3840 {
3841         if (prev == NULL)
3842                 return obj->maps;
3843
3844         return __bpf_map__iter(prev, obj, 1);
3845 }
3846
3847 struct bpf_map *
3848 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
3849 {
3850         if (next == NULL) {
3851                 if (!obj->nr_maps)
3852                         return NULL;
3853                 return obj->maps + obj->nr_maps - 1;
3854         }
3855
3856         return __bpf_map__iter(next, obj, -1);
3857 }
3858
3859 struct bpf_map *
3860 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
3861 {
3862         struct bpf_map *pos;
3863
3864         bpf_object__for_each_map(pos, obj) {
3865                 if (pos->name && !strcmp(pos->name, name))
3866                         return pos;
3867         }
3868         return NULL;
3869 }
3870
3871 int
3872 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
3873 {
3874         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
3875 }
3876
3877 struct bpf_map *
3878 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
3879 {
3880         return ERR_PTR(-ENOTSUP);
3881 }
3882
3883 long libbpf_get_error(const void *ptr)
3884 {
3885         return PTR_ERR_OR_ZERO(ptr);
3886 }
3887
3888 int bpf_prog_load(const char *file, enum bpf_prog_type type,
3889                   struct bpf_object **pobj, int *prog_fd)
3890 {
3891         struct bpf_prog_load_attr attr;
3892
3893         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
3894         attr.file = file;
3895         attr.prog_type = type;
3896         attr.expected_attach_type = 0;
3897
3898         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
3899 }
3900
3901 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
3902                         struct bpf_object **pobj, int *prog_fd)
3903 {
3904         struct bpf_object_open_attr open_attr = {};
3905         struct bpf_program *prog, *first_prog = NULL;
3906         enum bpf_attach_type expected_attach_type;
3907         enum bpf_prog_type prog_type;
3908         struct bpf_object *obj;
3909         struct bpf_map *map;
3910         int err;
3911
3912         if (!attr)
3913                 return -EINVAL;
3914         if (!attr->file)
3915                 return -EINVAL;
3916
3917         open_attr.file = attr->file;
3918         open_attr.prog_type = attr->prog_type;
3919
3920         obj = bpf_object__open_xattr(&open_attr);
3921         if (IS_ERR_OR_NULL(obj))
3922                 return -ENOENT;
3923
3924         bpf_object__for_each_program(prog, obj) {
3925                 /*
3926                  * If type is not specified, try to guess it based on
3927                  * section name.
3928                  */
3929                 prog_type = attr->prog_type;
3930                 prog->prog_ifindex = attr->ifindex;
3931                 expected_attach_type = attr->expected_attach_type;
3932                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
3933                         err = bpf_program__identify_section(prog, &prog_type,
3934                                                             &expected_attach_type);
3935                         if (err < 0) {
3936                                 bpf_object__close(obj);
3937                                 return -EINVAL;
3938                         }
3939                 }
3940
3941                 bpf_program__set_type(prog, prog_type);
3942                 bpf_program__set_expected_attach_type(prog,
3943                                                       expected_attach_type);
3944
3945                 prog->log_level = attr->log_level;
3946                 prog->prog_flags = attr->prog_flags;
3947                 if (!first_prog)
3948                         first_prog = prog;
3949         }
3950
3951         bpf_object__for_each_map(map, obj) {
3952                 if (!bpf_map__is_offload_neutral(map))
3953                         map->map_ifindex = attr->ifindex;
3954         }
3955
3956         if (!first_prog) {
3957                 pr_warning("object file doesn't contain bpf program\n");
3958                 bpf_object__close(obj);
3959                 return -ENOENT;
3960         }
3961
3962         err = bpf_object__load(obj);
3963         if (err) {
3964                 bpf_object__close(obj);
3965                 return -EINVAL;
3966         }
3967
3968         *pobj = obj;
3969         *prog_fd = bpf_program__fd(first_prog);
3970         return 0;
3971 }
3972
3973 struct bpf_link {
3974         int (*destroy)(struct bpf_link *link);
3975 };
3976
3977 int bpf_link__destroy(struct bpf_link *link)
3978 {
3979         int err;
3980
3981         if (!link)
3982                 return 0;
3983
3984         err = link->destroy(link);
3985         free(link);
3986
3987         return err;
3988 }
3989
3990 struct bpf_link_fd {
3991         struct bpf_link link; /* has to be at the top of struct */
3992         int fd; /* hook FD */
3993 };
3994
3995 static int bpf_link__destroy_perf_event(struct bpf_link *link)
3996 {
3997         struct bpf_link_fd *l = (void *)link;
3998         int err;
3999
4000         err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4001         if (err)
4002                 err = -errno;
4003
4004         close(l->fd);
4005         return err;
4006 }
4007
4008 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4009                                                 int pfd)
4010 {
4011         char errmsg[STRERR_BUFSIZE];
4012         struct bpf_link_fd *link;
4013         int prog_fd, err;
4014
4015         if (pfd < 0) {
4016                 pr_warning("program '%s': invalid perf event FD %d\n",
4017                            bpf_program__title(prog, false), pfd);
4018                 return ERR_PTR(-EINVAL);
4019         }
4020         prog_fd = bpf_program__fd(prog);
4021         if (prog_fd < 0) {
4022                 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
4023                            bpf_program__title(prog, false));
4024                 return ERR_PTR(-EINVAL);
4025         }
4026
4027         link = malloc(sizeof(*link));
4028         if (!link)
4029                 return ERR_PTR(-ENOMEM);
4030         link->link.destroy = &bpf_link__destroy_perf_event;
4031         link->fd = pfd;
4032
4033         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
4034                 err = -errno;
4035                 free(link);
4036                 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
4037                            bpf_program__title(prog, false), pfd,
4038                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4039                 return ERR_PTR(err);
4040         }
4041         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4042                 err = -errno;
4043                 free(link);
4044                 pr_warning("program '%s': failed to enable pfd %d: %s\n",
4045                            bpf_program__title(prog, false), pfd,
4046                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4047                 return ERR_PTR(err);
4048         }
4049         return (struct bpf_link *)link;
4050 }
4051
4052 /*
4053  * this function is expected to parse integer in the range of [0, 2^31-1] from
4054  * given file using scanf format string fmt. If actual parsed value is
4055  * negative, the result might be indistinguishable from error
4056  */
4057 static int parse_uint_from_file(const char *file, const char *fmt)
4058 {
4059         char buf[STRERR_BUFSIZE];
4060         int err, ret;
4061         FILE *f;
4062
4063         f = fopen(file, "r");
4064         if (!f) {
4065                 err = -errno;
4066                 pr_debug("failed to open '%s': %s\n", file,
4067                          libbpf_strerror_r(err, buf, sizeof(buf)));
4068                 return err;
4069         }
4070         err = fscanf(f, fmt, &ret);
4071         if (err != 1) {
4072                 err = err == EOF ? -EIO : -errno;
4073                 pr_debug("failed to parse '%s': %s\n", file,
4074                         libbpf_strerror_r(err, buf, sizeof(buf)));
4075                 fclose(f);
4076                 return err;
4077         }
4078         fclose(f);
4079         return ret;
4080 }
4081
4082 static int determine_kprobe_perf_type(void)
4083 {
4084         const char *file = "/sys/bus/event_source/devices/kprobe/type";
4085
4086         return parse_uint_from_file(file, "%d\n");
4087 }
4088
4089 static int determine_uprobe_perf_type(void)
4090 {
4091         const char *file = "/sys/bus/event_source/devices/uprobe/type";
4092
4093         return parse_uint_from_file(file, "%d\n");
4094 }
4095
4096 static int determine_kprobe_retprobe_bit(void)
4097 {
4098         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
4099
4100         return parse_uint_from_file(file, "config:%d\n");
4101 }
4102
4103 static int determine_uprobe_retprobe_bit(void)
4104 {
4105         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
4106
4107         return parse_uint_from_file(file, "config:%d\n");
4108 }
4109
4110 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
4111                                  uint64_t offset, int pid)
4112 {
4113         struct perf_event_attr attr = {};
4114         char errmsg[STRERR_BUFSIZE];
4115         int type, pfd, err;
4116
4117         type = uprobe ? determine_uprobe_perf_type()
4118                       : determine_kprobe_perf_type();
4119         if (type < 0) {
4120                 pr_warning("failed to determine %s perf type: %s\n",
4121                            uprobe ? "uprobe" : "kprobe",
4122                            libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
4123                 return type;
4124         }
4125         if (retprobe) {
4126                 int bit = uprobe ? determine_uprobe_retprobe_bit()
4127                                  : determine_kprobe_retprobe_bit();
4128
4129                 if (bit < 0) {
4130                         pr_warning("failed to determine %s retprobe bit: %s\n",
4131                                    uprobe ? "uprobe" : "kprobe",
4132                                    libbpf_strerror_r(bit, errmsg,
4133                                                      sizeof(errmsg)));
4134                         return bit;
4135                 }
4136                 attr.config |= 1 << bit;
4137         }
4138         attr.size = sizeof(attr);
4139         attr.type = type;
4140         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
4141         attr.config2 = offset;           /* kprobe_addr or probe_offset */
4142
4143         /* pid filter is meaningful only for uprobes */
4144         pfd = syscall(__NR_perf_event_open, &attr,
4145                       pid < 0 ? -1 : pid /* pid */,
4146                       pid == -1 ? 0 : -1 /* cpu */,
4147                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
4148         if (pfd < 0) {
4149                 err = -errno;
4150                 pr_warning("%s perf_event_open() failed: %s\n",
4151                            uprobe ? "uprobe" : "kprobe",
4152                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4153                 return err;
4154         }
4155         return pfd;
4156 }
4157
4158 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
4159                                             bool retprobe,
4160                                             const char *func_name)
4161 {
4162         char errmsg[STRERR_BUFSIZE];
4163         struct bpf_link *link;
4164         int pfd, err;
4165
4166         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
4167                                     0 /* offset */, -1 /* pid */);
4168         if (pfd < 0) {
4169                 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
4170                            bpf_program__title(prog, false),
4171                            retprobe ? "kretprobe" : "kprobe", func_name,
4172                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
4173                 return ERR_PTR(pfd);
4174         }
4175         link = bpf_program__attach_perf_event(prog, pfd);
4176         if (IS_ERR(link)) {
4177                 close(pfd);
4178                 err = PTR_ERR(link);
4179                 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
4180                            bpf_program__title(prog, false),
4181                            retprobe ? "kretprobe" : "kprobe", func_name,
4182                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4183                 return link;
4184         }
4185         return link;
4186 }
4187
4188 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
4189                                             bool retprobe, pid_t pid,
4190                                             const char *binary_path,
4191                                             size_t func_offset)
4192 {
4193         char errmsg[STRERR_BUFSIZE];
4194         struct bpf_link *link;
4195         int pfd, err;
4196
4197         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
4198                                     binary_path, func_offset, pid);
4199         if (pfd < 0) {
4200                 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
4201                            bpf_program__title(prog, false),
4202                            retprobe ? "uretprobe" : "uprobe",
4203                            binary_path, func_offset,
4204                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
4205                 return ERR_PTR(pfd);
4206         }
4207         link = bpf_program__attach_perf_event(prog, pfd);
4208         if (IS_ERR(link)) {
4209                 close(pfd);
4210                 err = PTR_ERR(link);
4211                 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
4212                            bpf_program__title(prog, false),
4213                            retprobe ? "uretprobe" : "uprobe",
4214                            binary_path, func_offset,
4215                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4216                 return link;
4217         }
4218         return link;
4219 }
4220
4221 static int determine_tracepoint_id(const char *tp_category,
4222                                    const char *tp_name)
4223 {
4224         char file[PATH_MAX];
4225         int ret;
4226
4227         ret = snprintf(file, sizeof(file),
4228                        "/sys/kernel/debug/tracing/events/%s/%s/id",
4229                        tp_category, tp_name);
4230         if (ret < 0)
4231                 return -errno;
4232         if (ret >= sizeof(file)) {
4233                 pr_debug("tracepoint %s/%s path is too long\n",
4234                          tp_category, tp_name);
4235                 return -E2BIG;
4236         }
4237         return parse_uint_from_file(file, "%d\n");
4238 }
4239
4240 static int perf_event_open_tracepoint(const char *tp_category,
4241                                       const char *tp_name)
4242 {
4243         struct perf_event_attr attr = {};
4244         char errmsg[STRERR_BUFSIZE];
4245         int tp_id, pfd, err;
4246
4247         tp_id = determine_tracepoint_id(tp_category, tp_name);
4248         if (tp_id < 0) {
4249                 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
4250                            tp_category, tp_name,
4251                            libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
4252                 return tp_id;
4253         }
4254
4255         attr.type = PERF_TYPE_TRACEPOINT;
4256         attr.size = sizeof(attr);
4257         attr.config = tp_id;
4258
4259         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
4260                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
4261         if (pfd < 0) {
4262                 err = -errno;
4263                 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
4264                            tp_category, tp_name,
4265                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4266                 return err;
4267         }
4268         return pfd;
4269 }
4270
4271 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
4272                                                 const char *tp_category,
4273                                                 const char *tp_name)
4274 {
4275         char errmsg[STRERR_BUFSIZE];
4276         struct bpf_link *link;
4277         int pfd, err;
4278
4279         pfd = perf_event_open_tracepoint(tp_category, tp_name);
4280         if (pfd < 0) {
4281                 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
4282                            bpf_program__title(prog, false),
4283                            tp_category, tp_name,
4284                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
4285                 return ERR_PTR(pfd);
4286         }
4287         link = bpf_program__attach_perf_event(prog, pfd);
4288         if (IS_ERR(link)) {
4289                 close(pfd);
4290                 err = PTR_ERR(link);
4291                 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
4292                            bpf_program__title(prog, false),
4293                            tp_category, tp_name,
4294                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
4295                 return link;
4296         }
4297         return link;
4298 }
4299
4300 static int bpf_link__destroy_fd(struct bpf_link *link)
4301 {
4302         struct bpf_link_fd *l = (void *)link;
4303
4304         return close(l->fd);
4305 }
4306
4307 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
4308                                                     const char *tp_name)
4309 {
4310         char errmsg[STRERR_BUFSIZE];
4311         struct bpf_link_fd *link;
4312         int prog_fd, pfd;
4313
4314         prog_fd = bpf_program__fd(prog);
4315         if (prog_fd < 0) {
4316                 pr_warning("program '%s': can't attach before loaded\n",
4317                            bpf_program__title(prog, false));
4318                 return ERR_PTR(-EINVAL);
4319         }
4320
4321         link = malloc(sizeof(*link));
4322         if (!link)
4323                 return ERR_PTR(-ENOMEM);
4324         link->link.destroy = &bpf_link__destroy_fd;
4325
4326         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
4327         if (pfd < 0) {
4328                 pfd = -errno;
4329                 free(link);
4330                 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
4331                            bpf_program__title(prog, false), tp_name,
4332                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
4333                 return ERR_PTR(pfd);
4334         }
4335         link->fd = pfd;
4336         return (struct bpf_link *)link;
4337 }
4338
4339 enum bpf_perf_event_ret
4340 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
4341                            void **copy_mem, size_t *copy_size,
4342                            bpf_perf_event_print_t fn, void *private_data)
4343 {
4344         struct perf_event_mmap_page *header = mmap_mem;
4345         __u64 data_head = ring_buffer_read_head(header);
4346         __u64 data_tail = header->data_tail;
4347         void *base = ((__u8 *)header) + page_size;
4348         int ret = LIBBPF_PERF_EVENT_CONT;
4349         struct perf_event_header *ehdr;
4350         size_t ehdr_size;
4351
4352         while (data_head != data_tail) {
4353                 ehdr = base + (data_tail & (mmap_size - 1));
4354                 ehdr_size = ehdr->size;
4355
4356                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
4357                         void *copy_start = ehdr;
4358                         size_t len_first = base + mmap_size - copy_start;
4359                         size_t len_secnd = ehdr_size - len_first;
4360
4361                         if (*copy_size < ehdr_size) {
4362                                 free(*copy_mem);
4363                                 *copy_mem = malloc(ehdr_size);
4364                                 if (!*copy_mem) {
4365                                         *copy_size = 0;
4366                                         ret = LIBBPF_PERF_EVENT_ERROR;
4367                                         break;
4368                                 }
4369                                 *copy_size = ehdr_size;
4370                         }
4371
4372                         memcpy(*copy_mem, copy_start, len_first);
4373                         memcpy(*copy_mem + len_first, base, len_secnd);
4374                         ehdr = *copy_mem;
4375                 }
4376
4377                 ret = fn(ehdr, private_data);
4378                 data_tail += ehdr_size;
4379                 if (ret != LIBBPF_PERF_EVENT_CONT)
4380                         break;
4381         }
4382
4383         ring_buffer_write_tail(header, data_tail);
4384         return ret;
4385 }
4386
4387 struct perf_buffer;
4388
4389 struct perf_buffer_params {
4390         struct perf_event_attr *attr;
4391         /* if event_cb is specified, it takes precendence */
4392         perf_buffer_event_fn event_cb;
4393         /* sample_cb and lost_cb are higher-level common-case callbacks */
4394         perf_buffer_sample_fn sample_cb;
4395         perf_buffer_lost_fn lost_cb;
4396         void *ctx;
4397         int cpu_cnt;
4398         int *cpus;
4399         int *map_keys;
4400 };
4401
4402 struct perf_cpu_buf {
4403         struct perf_buffer *pb;
4404         void *base; /* mmap()'ed memory */
4405         void *buf; /* for reconstructing segmented data */
4406         size_t buf_size;
4407         int fd;
4408         int cpu;
4409         int map_key;
4410 };
4411
4412 struct perf_buffer {
4413         perf_buffer_event_fn event_cb;
4414         perf_buffer_sample_fn sample_cb;
4415         perf_buffer_lost_fn lost_cb;
4416         void *ctx; /* passed into callbacks */
4417
4418         size_t page_size;
4419         size_t mmap_size;
4420         struct perf_cpu_buf **cpu_bufs;
4421         struct epoll_event *events;
4422         int cpu_cnt;
4423         int epoll_fd; /* perf event FD */
4424         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
4425 };
4426
4427 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
4428                                       struct perf_cpu_buf *cpu_buf)
4429 {
4430         if (!cpu_buf)
4431                 return;
4432         if (cpu_buf->base &&
4433             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
4434                 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
4435         if (cpu_buf->fd >= 0) {
4436                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
4437                 close(cpu_buf->fd);
4438         }
4439         free(cpu_buf->buf);
4440         free(cpu_buf);
4441 }
4442
4443 void perf_buffer__free(struct perf_buffer *pb)
4444 {
4445         int i;
4446
4447         if (!pb)
4448                 return;
4449         if (pb->cpu_bufs) {
4450                 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
4451                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
4452
4453                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
4454                         perf_buffer__free_cpu_buf(pb, cpu_buf);
4455                 }
4456                 free(pb->cpu_bufs);
4457         }
4458         if (pb->epoll_fd >= 0)
4459                 close(pb->epoll_fd);
4460         free(pb->events);
4461         free(pb);
4462 }
4463
4464 static struct perf_cpu_buf *
4465 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
4466                           int cpu, int map_key)
4467 {
4468         struct perf_cpu_buf *cpu_buf;
4469         char msg[STRERR_BUFSIZE];
4470         int err;
4471
4472         cpu_buf = calloc(1, sizeof(*cpu_buf));
4473         if (!cpu_buf)
4474                 return ERR_PTR(-ENOMEM);
4475
4476         cpu_buf->pb = pb;
4477         cpu_buf->cpu = cpu;
4478         cpu_buf->map_key = map_key;
4479
4480         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
4481                               -1, PERF_FLAG_FD_CLOEXEC);
4482         if (cpu_buf->fd < 0) {
4483                 err = -errno;
4484                 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
4485                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
4486                 goto error;
4487         }
4488
4489         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
4490                              PROT_READ | PROT_WRITE, MAP_SHARED,
4491                              cpu_buf->fd, 0);
4492         if (cpu_buf->base == MAP_FAILED) {
4493                 cpu_buf->base = NULL;
4494                 err = -errno;
4495                 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
4496                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
4497                 goto error;
4498         }
4499
4500         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
4501                 err = -errno;
4502                 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
4503                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
4504                 goto error;
4505         }
4506
4507         return cpu_buf;
4508
4509 error:
4510         perf_buffer__free_cpu_buf(pb, cpu_buf);
4511         return (struct perf_cpu_buf *)ERR_PTR(err);
4512 }
4513
4514 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
4515                                               struct perf_buffer_params *p);
4516
4517 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
4518                                      const struct perf_buffer_opts *opts)
4519 {
4520         struct perf_buffer_params p = {};
4521         struct perf_event_attr attr = { 0, };
4522
4523         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
4524         attr.type = PERF_TYPE_SOFTWARE;
4525         attr.sample_type = PERF_SAMPLE_RAW;
4526         attr.sample_period = 1;
4527         attr.wakeup_events = 1;
4528
4529         p.attr = &attr;
4530         p.sample_cb = opts ? opts->sample_cb : NULL;
4531         p.lost_cb = opts ? opts->lost_cb : NULL;
4532         p.ctx = opts ? opts->ctx : NULL;
4533
4534         return __perf_buffer__new(map_fd, page_cnt, &p);
4535 }
4536
4537 struct perf_buffer *
4538 perf_buffer__new_raw(int map_fd, size_t page_cnt,
4539                      const struct perf_buffer_raw_opts *opts)
4540 {
4541         struct perf_buffer_params p = {};
4542
4543         p.attr = opts->attr;
4544         p.event_cb = opts->event_cb;
4545         p.ctx = opts->ctx;
4546         p.cpu_cnt = opts->cpu_cnt;
4547         p.cpus = opts->cpus;
4548         p.map_keys = opts->map_keys;
4549
4550         return __perf_buffer__new(map_fd, page_cnt, &p);
4551 }
4552
4553 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
4554                                               struct perf_buffer_params *p)
4555 {
4556         struct bpf_map_info map = {};
4557         char msg[STRERR_BUFSIZE];
4558         struct perf_buffer *pb;
4559         __u32 map_info_len;
4560         int err, i;
4561
4562         if (page_cnt & (page_cnt - 1)) {
4563                 pr_warning("page count should be power of two, but is %zu\n",
4564                            page_cnt);
4565                 return ERR_PTR(-EINVAL);
4566         }
4567
4568         map_info_len = sizeof(map);
4569         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
4570         if (err) {
4571                 err = -errno;
4572                 pr_warning("failed to get map info for map FD %d: %s\n",
4573                            map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
4574                 return ERR_PTR(err);
4575         }
4576
4577         if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
4578                 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
4579                            map.name);
4580                 return ERR_PTR(-EINVAL);
4581         }
4582
4583         pb = calloc(1, sizeof(*pb));
4584         if (!pb)
4585                 return ERR_PTR(-ENOMEM);
4586
4587         pb->event_cb = p->event_cb;
4588         pb->sample_cb = p->sample_cb;
4589         pb->lost_cb = p->lost_cb;
4590         pb->ctx = p->ctx;
4591
4592         pb->page_size = getpagesize();
4593         pb->mmap_size = pb->page_size * page_cnt;
4594         pb->map_fd = map_fd;
4595
4596         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
4597         if (pb->epoll_fd < 0) {
4598                 err = -errno;
4599                 pr_warning("failed to create epoll instance: %s\n",
4600                            libbpf_strerror_r(err, msg, sizeof(msg)));
4601                 goto error;
4602         }
4603
4604         if (p->cpu_cnt > 0) {
4605                 pb->cpu_cnt = p->cpu_cnt;
4606         } else {
4607                 pb->cpu_cnt = libbpf_num_possible_cpus();
4608                 if (pb->cpu_cnt < 0) {
4609                         err = pb->cpu_cnt;
4610                         goto error;
4611                 }
4612                 if (map.max_entries < pb->cpu_cnt)
4613                         pb->cpu_cnt = map.max_entries;
4614         }
4615
4616         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
4617         if (!pb->events) {
4618                 err = -ENOMEM;
4619                 pr_warning("failed to allocate events: out of memory\n");
4620                 goto error;
4621         }
4622         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
4623         if (!pb->cpu_bufs) {
4624                 err = -ENOMEM;
4625                 pr_warning("failed to allocate buffers: out of memory\n");
4626                 goto error;
4627         }
4628
4629         for (i = 0; i < pb->cpu_cnt; i++) {
4630                 struct perf_cpu_buf *cpu_buf;
4631                 int cpu, map_key;
4632
4633                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
4634                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
4635
4636                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
4637                 if (IS_ERR(cpu_buf)) {
4638                         err = PTR_ERR(cpu_buf);
4639                         goto error;
4640                 }
4641
4642                 pb->cpu_bufs[i] = cpu_buf;
4643
4644                 err = bpf_map_update_elem(pb->map_fd, &map_key,
4645                                           &cpu_buf->fd, 0);
4646                 if (err) {
4647                         err = -errno;
4648                         pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
4649                                    cpu, map_key, cpu_buf->fd,
4650                                    libbpf_strerror_r(err, msg, sizeof(msg)));
4651                         goto error;
4652                 }
4653
4654                 pb->events[i].events = EPOLLIN;
4655                 pb->events[i].data.ptr = cpu_buf;
4656                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
4657                               &pb->events[i]) < 0) {
4658                         err = -errno;
4659                         pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
4660                                    cpu, cpu_buf->fd,
4661                                    libbpf_strerror_r(err, msg, sizeof(msg)));
4662                         goto error;
4663                 }
4664         }
4665
4666         return pb;
4667
4668 error:
4669         if (pb)
4670                 perf_buffer__free(pb);
4671         return ERR_PTR(err);
4672 }
4673
4674 struct perf_sample_raw {
4675         struct perf_event_header header;
4676         uint32_t size;
4677         char data[0];
4678 };
4679
4680 struct perf_sample_lost {
4681         struct perf_event_header header;
4682         uint64_t id;
4683         uint64_t lost;
4684         uint64_t sample_id;
4685 };
4686
4687 static enum bpf_perf_event_ret
4688 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
4689 {
4690         struct perf_cpu_buf *cpu_buf = ctx;
4691         struct perf_buffer *pb = cpu_buf->pb;
4692         void *data = e;
4693
4694         /* user wants full control over parsing perf event */
4695         if (pb->event_cb)
4696                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
4697
4698         switch (e->type) {
4699         case PERF_RECORD_SAMPLE: {
4700                 struct perf_sample_raw *s = data;
4701
4702                 if (pb->sample_cb)
4703                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
4704                 break;
4705         }
4706         case PERF_RECORD_LOST: {
4707                 struct perf_sample_lost *s = data;
4708
4709                 if (pb->lost_cb)
4710                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
4711                 break;
4712         }
4713         default:
4714                 pr_warning("unknown perf sample type %d\n", e->type);
4715                 return LIBBPF_PERF_EVENT_ERROR;
4716         }
4717         return LIBBPF_PERF_EVENT_CONT;
4718 }
4719
4720 static int perf_buffer__process_records(struct perf_buffer *pb,
4721                                         struct perf_cpu_buf *cpu_buf)
4722 {
4723         enum bpf_perf_event_ret ret;
4724
4725         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
4726                                          pb->page_size, &cpu_buf->buf,
4727                                          &cpu_buf->buf_size,
4728                                          perf_buffer__process_record, cpu_buf);
4729         if (ret != LIBBPF_PERF_EVENT_CONT)
4730                 return ret;
4731         return 0;
4732 }
4733
4734 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
4735 {
4736         int i, cnt, err;
4737
4738         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
4739         for (i = 0; i < cnt; i++) {
4740                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
4741
4742                 err = perf_buffer__process_records(pb, cpu_buf);
4743                 if (err) {
4744                         pr_warning("error while processing records: %d\n", err);
4745                         return err;
4746                 }
4747         }
4748         return cnt < 0 ? -errno : cnt;
4749 }
4750
4751 struct bpf_prog_info_array_desc {
4752         int     array_offset;   /* e.g. offset of jited_prog_insns */
4753         int     count_offset;   /* e.g. offset of jited_prog_len */
4754         int     size_offset;    /* > 0: offset of rec size,
4755                                  * < 0: fix size of -size_offset
4756                                  */
4757 };
4758
4759 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
4760         [BPF_PROG_INFO_JITED_INSNS] = {
4761                 offsetof(struct bpf_prog_info, jited_prog_insns),
4762                 offsetof(struct bpf_prog_info, jited_prog_len),
4763                 -1,
4764         },
4765         [BPF_PROG_INFO_XLATED_INSNS] = {
4766                 offsetof(struct bpf_prog_info, xlated_prog_insns),
4767                 offsetof(struct bpf_prog_info, xlated_prog_len),
4768                 -1,
4769         },
4770         [BPF_PROG_INFO_MAP_IDS] = {
4771                 offsetof(struct bpf_prog_info, map_ids),
4772                 offsetof(struct bpf_prog_info, nr_map_ids),
4773                 -(int)sizeof(__u32),
4774         },
4775         [BPF_PROG_INFO_JITED_KSYMS] = {
4776                 offsetof(struct bpf_prog_info, jited_ksyms),
4777                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
4778                 -(int)sizeof(__u64),
4779         },
4780         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
4781                 offsetof(struct bpf_prog_info, jited_func_lens),
4782                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
4783                 -(int)sizeof(__u32),
4784         },
4785         [BPF_PROG_INFO_FUNC_INFO] = {
4786                 offsetof(struct bpf_prog_info, func_info),
4787                 offsetof(struct bpf_prog_info, nr_func_info),
4788                 offsetof(struct bpf_prog_info, func_info_rec_size),
4789         },
4790         [BPF_PROG_INFO_LINE_INFO] = {
4791                 offsetof(struct bpf_prog_info, line_info),
4792                 offsetof(struct bpf_prog_info, nr_line_info),
4793                 offsetof(struct bpf_prog_info, line_info_rec_size),
4794         },
4795         [BPF_PROG_INFO_JITED_LINE_INFO] = {
4796                 offsetof(struct bpf_prog_info, jited_line_info),
4797                 offsetof(struct bpf_prog_info, nr_jited_line_info),
4798                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
4799         },
4800         [BPF_PROG_INFO_PROG_TAGS] = {
4801                 offsetof(struct bpf_prog_info, prog_tags),
4802                 offsetof(struct bpf_prog_info, nr_prog_tags),
4803                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
4804         },
4805
4806 };
4807
4808 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
4809 {
4810         __u32 *array = (__u32 *)info;
4811
4812         if (offset >= 0)
4813                 return array[offset / sizeof(__u32)];
4814         return -(int)offset;
4815 }
4816
4817 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
4818 {
4819         __u64 *array = (__u64 *)info;
4820
4821         if (offset >= 0)
4822                 return array[offset / sizeof(__u64)];
4823         return -(int)offset;
4824 }
4825
4826 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
4827                                          __u32 val)
4828 {
4829         __u32 *array = (__u32 *)info;
4830
4831         if (offset >= 0)
4832                 array[offset / sizeof(__u32)] = val;
4833 }
4834
4835 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
4836                                          __u64 val)
4837 {
4838         __u64 *array = (__u64 *)info;
4839
4840         if (offset >= 0)
4841                 array[offset / sizeof(__u64)] = val;
4842 }
4843
4844 struct bpf_prog_info_linear *
4845 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
4846 {
4847         struct bpf_prog_info_linear *info_linear;
4848         struct bpf_prog_info info = {};
4849         __u32 info_len = sizeof(info);
4850         __u32 data_len = 0;
4851         int i, err;
4852         void *ptr;
4853
4854         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
4855                 return ERR_PTR(-EINVAL);
4856
4857         /* step 1: get array dimensions */
4858         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
4859         if (err) {
4860                 pr_debug("can't get prog info: %s", strerror(errno));
4861                 return ERR_PTR(-EFAULT);
4862         }
4863
4864         /* step 2: calculate total size of all arrays */
4865         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4866                 bool include_array = (arrays & (1UL << i)) > 0;
4867                 struct bpf_prog_info_array_desc *desc;
4868                 __u32 count, size;
4869
4870                 desc = bpf_prog_info_array_desc + i;
4871
4872                 /* kernel is too old to support this field */
4873                 if (info_len < desc->array_offset + sizeof(__u32) ||
4874                     info_len < desc->count_offset + sizeof(__u32) ||
4875                     (desc->size_offset > 0 && info_len < desc->size_offset))
4876                         include_array = false;
4877
4878                 if (!include_array) {
4879                         arrays &= ~(1UL << i);  /* clear the bit */
4880                         continue;
4881                 }
4882
4883                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4884                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4885
4886                 data_len += count * size;
4887         }
4888
4889         /* step 3: allocate continuous memory */
4890         data_len = roundup(data_len, sizeof(__u64));
4891         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
4892         if (!info_linear)
4893                 return ERR_PTR(-ENOMEM);
4894
4895         /* step 4: fill data to info_linear->info */
4896         info_linear->arrays = arrays;
4897         memset(&info_linear->info, 0, sizeof(info));
4898         ptr = info_linear->data;
4899
4900         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4901                 struct bpf_prog_info_array_desc *desc;
4902                 __u32 count, size;
4903
4904                 if ((arrays & (1UL << i)) == 0)
4905                         continue;
4906
4907                 desc  = bpf_prog_info_array_desc + i;
4908                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4909                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4910                 bpf_prog_info_set_offset_u32(&info_linear->info,
4911                                              desc->count_offset, count);
4912                 bpf_prog_info_set_offset_u32(&info_linear->info,
4913                                              desc->size_offset, size);
4914                 bpf_prog_info_set_offset_u64(&info_linear->info,
4915                                              desc->array_offset,
4916                                              ptr_to_u64(ptr));
4917                 ptr += count * size;
4918         }
4919
4920         /* step 5: call syscall again to get required arrays */
4921         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
4922         if (err) {
4923                 pr_debug("can't get prog info: %s", strerror(errno));
4924                 free(info_linear);
4925                 return ERR_PTR(-EFAULT);
4926         }
4927
4928         /* step 6: verify the data */
4929         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4930                 struct bpf_prog_info_array_desc *desc;
4931                 __u32 v1, v2;
4932
4933                 if ((arrays & (1UL << i)) == 0)
4934                         continue;
4935
4936                 desc = bpf_prog_info_array_desc + i;
4937                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4938                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
4939                                                    desc->count_offset);
4940                 if (v1 != v2)
4941                         pr_warning("%s: mismatch in element count\n", __func__);
4942
4943                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4944                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
4945                                                    desc->size_offset);
4946                 if (v1 != v2)
4947                         pr_warning("%s: mismatch in rec size\n", __func__);
4948         }
4949
4950         /* step 7: update info_len and data_len */
4951         info_linear->info_len = sizeof(struct bpf_prog_info);
4952         info_linear->data_len = data_len;
4953
4954         return info_linear;
4955 }
4956
4957 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
4958 {
4959         int i;
4960
4961         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4962                 struct bpf_prog_info_array_desc *desc;
4963                 __u64 addr, offs;
4964
4965                 if ((info_linear->arrays & (1UL << i)) == 0)
4966                         continue;
4967
4968                 desc = bpf_prog_info_array_desc + i;
4969                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
4970                                                      desc->array_offset);
4971                 offs = addr - ptr_to_u64(info_linear->data);
4972                 bpf_prog_info_set_offset_u64(&info_linear->info,
4973                                              desc->array_offset, offs);
4974         }
4975 }
4976
4977 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
4978 {
4979         int i;
4980
4981         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4982                 struct bpf_prog_info_array_desc *desc;
4983                 __u64 addr, offs;
4984
4985                 if ((info_linear->arrays & (1UL << i)) == 0)
4986                         continue;
4987
4988                 desc = bpf_prog_info_array_desc + i;
4989                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
4990                                                      desc->array_offset);
4991                 addr = offs + ptr_to_u64(info_linear->data);
4992                 bpf_prog_info_set_offset_u64(&info_linear->info,
4993                                              desc->array_offset, addr);
4994         }
4995 }
4996
4997 int libbpf_num_possible_cpus(void)
4998 {
4999         static const char *fcpu = "/sys/devices/system/cpu/possible";
5000         int len = 0, n = 0, il = 0, ir = 0;
5001         unsigned int start = 0, end = 0;
5002         int tmp_cpus = 0;
5003         static int cpus;
5004         char buf[128];
5005         int error = 0;
5006         int fd = -1;
5007
5008         tmp_cpus = READ_ONCE(cpus);
5009         if (tmp_cpus > 0)
5010                 return tmp_cpus;
5011
5012         fd = open(fcpu, O_RDONLY);
5013         if (fd < 0) {
5014                 error = errno;
5015                 pr_warning("Failed to open file %s: %s\n",
5016                            fcpu, strerror(error));
5017                 return -error;
5018         }
5019         len = read(fd, buf, sizeof(buf));
5020         close(fd);
5021         if (len <= 0) {
5022                 error = len ? errno : EINVAL;
5023                 pr_warning("Failed to read # of possible cpus from %s: %s\n",
5024                            fcpu, strerror(error));
5025                 return -error;
5026         }
5027         if (len == sizeof(buf)) {
5028                 pr_warning("File %s size overflow\n", fcpu);
5029                 return -EOVERFLOW;
5030         }
5031         buf[len] = '\0';
5032
5033         for (ir = 0, tmp_cpus = 0; ir <= len; ir++) {
5034                 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
5035                 if (buf[ir] == ',' || buf[ir] == '\0') {
5036                         buf[ir] = '\0';
5037                         n = sscanf(&buf[il], "%u-%u", &start, &end);
5038                         if (n <= 0) {
5039                                 pr_warning("Failed to get # CPUs from %s\n",
5040                                            &buf[il]);
5041                                 return -EINVAL;
5042                         } else if (n == 1) {
5043                                 end = start;
5044                         }
5045                         tmp_cpus += end - start + 1;
5046                         il = ir + 1;
5047                 }
5048         }
5049         if (tmp_cpus <= 0) {
5050                 pr_warning("Invalid #CPUs %d from %s\n", tmp_cpus, fcpu);
5051                 return -EINVAL;
5052         }
5053
5054         WRITE_ONCE(cpus, tmp_cpus);
5055         return tmp_cpus;
5056 }