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 <fcntl.h>
  24 #include <errno.h>
  25 #include <asm/unistd.h>
  26 #include <linux/err.h>
  27 #include <linux/kernel.h>
  28 #include <linux/bpf.h>
  29 #include <linux/btf.h>
  30 #include <linux/filter.h>
  31 #include <linux/list.h>
  32 #include <linux/limits.h>
  33 #include <linux/perf_event.h>
  34 #include <linux/ring_buffer.h>
  35 #include <sys/stat.h>
  36 #include <sys/types.h>
  37 #include <sys/vfs.h>
  38 #include <tools/libc_compat.h>
  39 #include <libelf.h>
  40 #include <gelf.h>
  41
  42 #include "libbpf.h"
  43 #include "bpf.h"
  44 #include "btf.h"
  45 #include "str_error.h"
  46 #include "libbpf_internal.h"
  47
  48 #ifndef EM_BPF
  49 #define EM_BPF 247
  50 #endif
  51
  52 #ifndef BPF_FS_MAGIC
  53 #define BPF_FS_MAGIC            0xcafe4a11
  54 #endif
  55
  56 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  57  * compilation if user enables corresponding warning. Disable it explicitly.
  58  */
  59 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  60
  61 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  62
  63 static int __base_pr(enum libbpf_print_level level, const char *format,
  64                      va_list args)
  65 {
  66         if (level == LIBBPF_DEBUG)
  67                 return 0;
  68
  69         return vfprintf(stderr, format, args);
  70 }
  71
  72 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  73
  74 void libbpf_set_print(libbpf_print_fn_t fn)
  75 {
  76         __libbpf_pr = fn;
  77 }
  78
  79 __printf(2, 3)
  80 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  81 {
  82         va_list args;
  83
  84         if (!__libbpf_pr)
  85                 return;
  86
  87         va_start(args, format);
  88         __libbpf_pr(level, format, args);
  89         va_end(args);
  90 }
  91
  92 #define STRERR_BUFSIZE  128
  93
  94 #define CHECK_ERR(action, err, out) do {        \
  95         err = action;                   \
  96         if (err)                        \
  97                 goto out;               \
  98 } while(0)
  99
 100
 101 /* Copied from tools/perf/util/util.h */
 102 #ifndef zfree
 103 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 104 #endif
 105
 106 #ifndef zclose
 107 # define zclose(fd) ({                  \
 108         int ___err = 0;                 \
 109         if ((fd) >= 0)                  \
 110                 ___err = close((fd));   \
 111         fd = -1;                        \
 112         ___err; })
 113 #endif
 114
 115 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 116 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 117 #else
 118 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 119 #endif
 120
 121 static inline __u64 ptr_to_u64(const void *ptr)
 122 {
 123         return (__u64) (unsigned long) ptr;
 124 }
 125
 126 struct bpf_capabilities {
 127         /* v4.14: kernel support for program & map names. */
 128         __u32 name:1;
 129         /* v5.2: kernel support for global data sections. */
 130         __u32 global_data:1;
 131         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 132         __u32 btf_func:1;
 133         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 134         __u32 btf_datasec:1;
 135 };
 136
 137 /*
 138  * bpf_prog should be a better name but it has been used in
 139  * linux/filter.h.
 140  */
 141 struct bpf_program {
 142         /* Index in elf obj file, for relocation use. */
 143         int idx;
 144         char *name;
 145         int prog_ifindex;
 146         char *section_name;
 147         /* section_name with / replaced by _; makes recursive pinning
 148          * in bpf_object__pin_programs easier
 149          */
 150         char *pin_name;
 151         struct bpf_insn *insns;
 152         size_t insns_cnt, main_prog_cnt;
 153         enum bpf_prog_type type;
 154
 155         struct reloc_desc {
 156                 enum {
 157                         RELO_LD64,
 158                         RELO_CALL,
 159                         RELO_DATA,
 160                 } type;
 161                 int insn_idx;
 162                 union {
 163                         int map_idx;
 164                         int text_off;
 165                 };
 166         } *reloc_desc;
 167         int nr_reloc;
 168         int log_level;
 169
 170         struct {
 171                 int nr;
 172                 int *fds;
 173         } instances;
 174         bpf_program_prep_t preprocessor;
 175
 176         struct bpf_object *obj;
 177         void *priv;
 178         bpf_program_clear_priv_t clear_priv;
 179
 180         enum bpf_attach_type expected_attach_type;
 181         int btf_fd;
 182         void *func_info;
 183         __u32 func_info_rec_size;
 184         __u32 func_info_cnt;
 185
 186         struct bpf_capabilities *caps;
 187
 188         void *line_info;
 189         __u32 line_info_rec_size;
 190         __u32 line_info_cnt;
 191 };
 192
 193 enum libbpf_map_type {
 194         LIBBPF_MAP_UNSPEC,
 195         LIBBPF_MAP_DATA,
 196         LIBBPF_MAP_BSS,
 197         LIBBPF_MAP_RODATA,
 198 };
 199
 200 static const char * const libbpf_type_to_btf_name[] = {
 201         [LIBBPF_MAP_DATA]       = ".data",
 202         [LIBBPF_MAP_BSS]        = ".bss",
 203         [LIBBPF_MAP_RODATA]     = ".rodata",
 204 };
 205
 206 struct bpf_map {
 207         int fd;
 208         char *name;
 209         size_t offset;
 210         int map_ifindex;
 211         int inner_map_fd;
 212         struct bpf_map_def def;
 213         __u32 btf_key_type_id;
 214         __u32 btf_value_type_id;
 215         void *priv;
 216         bpf_map_clear_priv_t clear_priv;
 217         enum libbpf_map_type libbpf_type;
 218 };
 219
 220 struct bpf_secdata {
 221         void *rodata;
 222         void *data;
 223 };
 224
 225 static LIST_HEAD(bpf_objects_list);
 226
 227 struct bpf_object {
 228         char name[BPF_OBJ_NAME_LEN];
 229         char license[64];
 230         __u32 kern_version;
 231
 232         struct bpf_program *programs;
 233         size_t nr_programs;
 234         struct bpf_map *maps;
 235         size_t nr_maps;
 236         struct bpf_secdata sections;
 237
 238         bool loaded;
 239         bool has_pseudo_calls;
 240
 241         /*
 242          * Information when doing elf related work. Only valid if fd
 243          * is valid.
 244          */
 245         struct {
 246                 int fd;
 247                 void *obj_buf;
 248                 size_t obj_buf_sz;
 249                 Elf *elf;
 250                 GElf_Ehdr ehdr;
 251                 Elf_Data *symbols;
 252                 Elf_Data *data;
 253                 Elf_Data *rodata;
 254                 Elf_Data *bss;
 255                 size_t strtabidx;
 256                 struct {
 257                         GElf_Shdr shdr;
 258                         Elf_Data *data;
 259                 } *reloc;
 260                 int nr_reloc;
 261                 int maps_shndx;
 262                 int text_shndx;
 263                 int data_shndx;
 264                 int rodata_shndx;
 265                 int bss_shndx;
 266         } efile;
 267         /*
 268          * All loaded bpf_object is linked in a list, which is
 269          * hidden to caller. bpf_objects__<func> handlers deal with
 270          * all objects.
 271          */
 272         struct list_head list;
 273
 274         struct btf *btf;
 275         struct btf_ext *btf_ext;
 276
 277         void *priv;
 278         bpf_object_clear_priv_t clear_priv;
 279
 280         struct bpf_capabilities caps;
 281
 282         char path[];
 283 };
 284 #define obj_elf_valid(o)        ((o)->efile.elf)
 285
 286 void bpf_program__unload(struct bpf_program *prog)
 287 {
 288         int i;
 289
 290         if (!prog)
 291                 return;
 292
 293         /*
 294          * If the object is opened but the program was never loaded,
 295          * it is possible that prog->instances.nr == -1.
 296          */
 297         if (prog->instances.nr > 0) {
 298                 for (i = 0; i < prog->instances.nr; i++)
 299                         zclose(prog->instances.fds[i]);
 300         } else if (prog->instances.nr != -1) {
 301                 pr_warning("Internal error: instances.nr is %d\n",
 302                            prog->instances.nr);
 303         }
 304
 305         prog->instances.nr = -1;
 306         zfree(&prog->instances.fds);
 307
 308         zclose(prog->btf_fd);
 309         zfree(&prog->func_info);
 310         zfree(&prog->line_info);
 311 }
 312
 313 static void bpf_program__exit(struct bpf_program *prog)
 314 {
 315         if (!prog)
 316                 return;
 317
 318         if (prog->clear_priv)
 319                 prog->clear_priv(prog, prog->priv);
 320
 321         prog->priv = NULL;
 322         prog->clear_priv = NULL;
 323
 324         bpf_program__unload(prog);
 325         zfree(&prog->name);
 326         zfree(&prog->section_name);
 327         zfree(&prog->pin_name);
 328         zfree(&prog->insns);
 329         zfree(&prog->reloc_desc);
 330
 331         prog->nr_reloc = 0;
 332         prog->insns_cnt = 0;
 333         prog->idx = -1;
 334 }
 335
 336 static char *__bpf_program__pin_name(struct bpf_program *prog)
 337 {
 338         char *name, *p;
 339
 340         name = p = strdup(prog->section_name);
 341         while ((p = strchr(p, '/')))
 342                 *p = '_';
 343
 344         return name;
 345 }
 346
 347 static int
 348 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 349                   struct bpf_program *prog)
 350 {
 351         if (size < sizeof(struct bpf_insn)) {
 352                 pr_warning("corrupted section '%s'\n", section_name);
 353                 return -EINVAL;
 354         }
 355
 356         memset(prog, 0, sizeof(*prog));
 357
 358         prog->section_name = strdup(section_name);
 359         if (!prog->section_name) {
 360                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 361                            idx, section_name);
 362                 goto errout;
 363         }
 364
 365         prog->pin_name = __bpf_program__pin_name(prog);
 366         if (!prog->pin_name) {
 367                 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
 368                            idx, section_name);
 369                 goto errout;
 370         }
 371
 372         prog->insns = malloc(size);
 373         if (!prog->insns) {
 374                 pr_warning("failed to alloc insns for prog under section %s\n",
 375                            section_name);
 376                 goto errout;
 377         }
 378         prog->insns_cnt = size / sizeof(struct bpf_insn);
 379         memcpy(prog->insns, data,
 380                prog->insns_cnt * sizeof(struct bpf_insn));
 381         prog->idx = idx;
 382         prog->instances.fds = NULL;
 383         prog->instances.nr = -1;
 384         prog->type = BPF_PROG_TYPE_UNSPEC;
 385         prog->btf_fd = -1;
 386
 387         return 0;
 388 errout:
 389         bpf_program__exit(prog);
 390         return -ENOMEM;
 391 }
 392
 393 static int
 394 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 395                         char *section_name, int idx)
 396 {
 397         struct bpf_program prog, *progs;
 398         int nr_progs, err;
 399
 400         err = bpf_program__init(data, size, section_name, idx, &prog);
 401         if (err)
 402                 return err;
 403
 404         prog.caps = &obj->caps;
 405         progs = obj->programs;
 406         nr_progs = obj->nr_programs;
 407
 408         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 409         if (!progs) {
 410                 /*
 411                  * In this case the original obj->programs
 412                  * is still valid, so don't need special treat for
 413                  * bpf_close_object().
 414                  */
 415                 pr_warning("failed to alloc a new program under section '%s'\n",
 416                            section_name);
 417                 bpf_program__exit(&prog);
 418                 return -ENOMEM;
 419         }
 420
 421         pr_debug("found program %s\n", prog.section_name);
 422         obj->programs = progs;
 423         obj->nr_programs = nr_progs + 1;
 424         prog.obj = obj;
 425         progs[nr_progs] = prog;
 426         return 0;
 427 }
 428
 429 static int
 430 bpf_object__init_prog_names(struct bpf_object *obj)
 431 {
 432         Elf_Data *symbols = obj->efile.symbols;
 433         struct bpf_program *prog;
 434         size_t pi, si;
 435
 436         for (pi = 0; pi < obj->nr_programs; pi++) {
 437                 const char *name = NULL;
 438
 439                 prog = &obj->programs[pi];
 440
 441                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 442                      si++) {
 443                         GElf_Sym sym;
 444
 445                         if (!gelf_getsym(symbols, si, &sym))
 446                                 continue;
 447                         if (sym.st_shndx != prog->idx)
 448                                 continue;
 449                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 450                                 continue;
 451
 452                         name = elf_strptr(obj->efile.elf,
 453                                           obj->efile.strtabidx,
 454                                           sym.st_name);
 455                         if (!name) {
 456                                 pr_warning("failed to get sym name string for prog %s\n",
 457                                            prog->section_name);
 458                                 return -LIBBPF_ERRNO__LIBELF;
 459                         }
 460                 }
 461
 462                 if (!name && prog->idx == obj->efile.text_shndx)
 463                         name = ".text";
 464
 465                 if (!name) {
 466                         pr_warning("failed to find sym for prog %s\n",
 467                                    prog->section_name);
 468                         return -EINVAL;
 469                 }
 470
 471                 prog->name = strdup(name);
 472                 if (!prog->name) {
 473                         pr_warning("failed to allocate memory for prog sym %s\n",
 474                                    name);
 475                         return -ENOMEM;
 476                 }
 477         }
 478
 479         return 0;
 480 }
 481
 482 static struct bpf_object *bpf_object__new(const char *path,
 483                                           void *obj_buf,
 484                                           size_t obj_buf_sz)
 485 {
 486         struct bpf_object *obj;
 487         char *end;
 488
 489         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 490         if (!obj) {
 491                 pr_warning("alloc memory failed for %s\n", path);
 492                 return ERR_PTR(-ENOMEM);
 493         }
 494
 495         strcpy(obj->path, path);
 496         /* Using basename() GNU version which doesn't modify arg. */
 497         strncpy(obj->name, basename((void *)path),
 498                 sizeof(obj->name) - 1);
 499         end = strchr(obj->name, '.');
 500         if (end)
 501                 *end = 0;
 502
 503         obj->efile.fd = -1;
 504         /*
 505          * Caller of this function should also calls
 506          * bpf_object__elf_finish() after data collection to return
 507          * obj_buf to user. If not, we should duplicate the buffer to
 508          * avoid user freeing them before elf finish.
 509          */
 510         obj->efile.obj_buf = obj_buf;
 511         obj->efile.obj_buf_sz = obj_buf_sz;
 512         obj->efile.maps_shndx = -1;
 513         obj->efile.data_shndx = -1;
 514         obj->efile.rodata_shndx = -1;
 515         obj->efile.bss_shndx = -1;
 516
 517         obj->loaded = false;
 518
 519         INIT_LIST_HEAD(&obj->list);
 520         list_add(&obj->list, &bpf_objects_list);
 521         return obj;
 522 }
 523
 524 static void bpf_object__elf_finish(struct bpf_object *obj)
 525 {
 526         if (!obj_elf_valid(obj))
 527                 return;
 528
 529         if (obj->efile.elf) {
 530                 elf_end(obj->efile.elf);
 531                 obj->efile.elf = NULL;
 532         }
 533         obj->efile.symbols = NULL;
 534         obj->efile.data = NULL;
 535         obj->efile.rodata = NULL;
 536         obj->efile.bss = NULL;
 537
 538         zfree(&obj->efile.reloc);
 539         obj->efile.nr_reloc = 0;
 540         zclose(obj->efile.fd);
 541         obj->efile.obj_buf = NULL;
 542         obj->efile.obj_buf_sz = 0;
 543 }
 544
 545 static int bpf_object__elf_init(struct bpf_object *obj)
 546 {
 547         int err = 0;
 548         GElf_Ehdr *ep;
 549
 550         if (obj_elf_valid(obj)) {
 551                 pr_warning("elf init: internal error\n");
 552                 return -LIBBPF_ERRNO__LIBELF;
 553         }
 554
 555         if (obj->efile.obj_buf_sz > 0) {
 556                 /*
 557                  * obj_buf should have been validated by
 558                  * bpf_object__open_buffer().
 559                  */
 560                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 561                                             obj->efile.obj_buf_sz);
 562         } else {
 563                 obj->efile.fd = open(obj->path, O_RDONLY);
 564                 if (obj->efile.fd < 0) {
 565                         char errmsg[STRERR_BUFSIZE];
 566                         char *cp = libbpf_strerror_r(errno, errmsg,
 567                                                      sizeof(errmsg));
 568
 569                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 570                         return -errno;
 571                 }
 572
 573                 obj->efile.elf = elf_begin(obj->efile.fd,
 574                                 LIBBPF_ELF_C_READ_MMAP,
 575                                 NULL);
 576         }
 577
 578         if (!obj->efile.elf) {
 579                 pr_warning("failed to open %s as ELF file\n",
 580                                 obj->path);
 581                 err = -LIBBPF_ERRNO__LIBELF;
 582                 goto errout;
 583         }
 584
 585         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 586                 pr_warning("failed to get EHDR from %s\n",
 587                                 obj->path);
 588                 err = -LIBBPF_ERRNO__FORMAT;
 589                 goto errout;
 590         }
 591         ep = &obj->efile.ehdr;
 592
 593         /* Old LLVM set e_machine to EM_NONE */
 594         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 595                 pr_warning("%s is not an eBPF object file\n",
 596                         obj->path);
 597                 err = -LIBBPF_ERRNO__FORMAT;
 598                 goto errout;
 599         }
 600
 601         return 0;
 602 errout:
 603         bpf_object__elf_finish(obj);
 604         return err;
 605 }
 606
 607 static int
 608 bpf_object__check_endianness(struct bpf_object *obj)
 609 {
 610         static unsigned int const endian = 1;
 611
 612         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 613         case ELFDATA2LSB:
 614                 /* We are big endian, BPF obj is little endian. */
 615                 if (*(unsigned char const *)&endian != 1)
 616                         goto mismatch;
 617                 break;
 618
 619         case ELFDATA2MSB:
 620                 /* We are little endian, BPF obj is big endian. */
 621                 if (*(unsigned char const *)&endian != 0)
 622                         goto mismatch;
 623                 break;
 624         default:
 625                 return -LIBBPF_ERRNO__ENDIAN;
 626         }
 627
 628         return 0;
 629
 630 mismatch:
 631         pr_warning("Error: endianness mismatch.\n");
 632         return -LIBBPF_ERRNO__ENDIAN;
 633 }
 634
 635 static int
 636 bpf_object__init_license(struct bpf_object *obj,
 637                          void *data, size_t size)
 638 {
 639         memcpy(obj->license, data,
 640                min(size, sizeof(obj->license) - 1));
 641         pr_debug("license of %s is %s\n", obj->path, obj->license);
 642         return 0;
 643 }
 644
 645 static int
 646 bpf_object__init_kversion(struct bpf_object *obj,
 647                           void *data, size_t size)
 648 {
 649         __u32 kver;
 650
 651         if (size != sizeof(kver)) {
 652                 pr_warning("invalid kver section in %s\n", obj->path);
 653                 return -LIBBPF_ERRNO__FORMAT;
 654         }
 655         memcpy(&kver, data, sizeof(kver));
 656         obj->kern_version = kver;
 657         pr_debug("kernel version of %s is %x\n", obj->path,
 658                  obj->kern_version);
 659         return 0;
 660 }
 661
 662 static int compare_bpf_map(const void *_a, const void *_b)
 663 {
 664         const struct bpf_map *a = _a;
 665         const struct bpf_map *b = _b;
 666
 667         return a->offset - b->offset;
 668 }
 669
 670 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 671 {
 672         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 673             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 674                 return true;
 675         return false;
 676 }
 677
 678 static int bpf_object_search_section_size(const struct bpf_object *obj,
 679                                           const char *name, size_t *d_size)
 680 {
 681         const GElf_Ehdr *ep = &obj->efile.ehdr;
 682         Elf *elf = obj->efile.elf;
 683         Elf_Scn *scn = NULL;
 684         int idx = 0;
 685
 686         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 687                 const char *sec_name;
 688                 Elf_Data *data;
 689                 GElf_Shdr sh;
 690
 691                 idx++;
 692                 if (gelf_getshdr(scn, &sh) != &sh) {
 693                         pr_warning("failed to get section(%d) header from %s\n",
 694                                    idx, obj->path);
 695                         return -EIO;
 696                 }
 697
 698                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 699                 if (!sec_name) {
 700                         pr_warning("failed to get section(%d) name from %s\n",
 701                                    idx, obj->path);
 702                         return -EIO;
 703                 }
 704
 705                 if (strcmp(name, sec_name))
 706                         continue;
 707
 708                 data = elf_getdata(scn, 0);
 709                 if (!data) {
 710                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 711                                    idx, name, obj->path);
 712                         return -EIO;
 713                 }
 714
 715                 *d_size = data->d_size;
 716                 return 0;
 717         }
 718
 719         return -ENOENT;
 720 }
 721
 722 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 723                              __u32 *size)
 724 {
 725         int ret = -ENOENT;
 726         size_t d_size;
 727
 728         *size = 0;
 729         if (!name) {
 730                 return -EINVAL;
 731         } else if (!strcmp(name, ".data")) {
 732                 if (obj->efile.data)
 733                         *size = obj->efile.data->d_size;
 734         } else if (!strcmp(name, ".bss")) {
 735                 if (obj->efile.bss)
 736                         *size = obj->efile.bss->d_size;
 737         } else if (!strcmp(name, ".rodata")) {
 738                 if (obj->efile.rodata)
 739                         *size = obj->efile.rodata->d_size;
 740         } else {
 741                 ret = bpf_object_search_section_size(obj, name, &d_size);
 742                 if (!ret)
 743                         *size = d_size;
 744         }
 745
 746         return *size ? 0 : ret;
 747 }
 748
 749 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 750                                 __u32 *off)
 751 {
 752         Elf_Data *symbols = obj->efile.symbols;
 753         const char *sname;
 754         size_t si;
 755
 756         if (!name || !off)
 757                 return -EINVAL;
 758
 759         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 760                 GElf_Sym sym;
 761
 762                 if (!gelf_getsym(symbols, si, &sym))
 763                         continue;
 764                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 765                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 766                         continue;
 767
 768                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 769                                    sym.st_name);
 770                 if (!sname) {
 771                         pr_warning("failed to get sym name string for var %s\n",
 772                                    name);
 773                         return -EIO;
 774                 }
 775                 if (strcmp(name, sname) == 0) {
 776                         *off = sym.st_value;
 777                         return 0;
 778                 }
 779         }
 780
 781         return -ENOENT;
 782 }
 783
 784 static bool bpf_object__has_maps(const struct bpf_object *obj)
 785 {
 786         return obj->efile.maps_shndx >= 0 ||
 787                obj->efile.data_shndx >= 0 ||
 788                obj->efile.rodata_shndx >= 0 ||
 789                obj->efile.bss_shndx >= 0;
 790 }
 791
 792 static int
 793 bpf_object__init_internal_map(struct bpf_object *obj, struct bpf_map *map,
 794                               enum libbpf_map_type type, Elf_Data *data,
 795                               void **data_buff)
 796 {
 797         struct bpf_map_def *def = &map->def;
 798         char map_name[BPF_OBJ_NAME_LEN];
 799
 800         map->libbpf_type = type;
 801         map->offset = ~(typeof(map->offset))0;
 802         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 803                  libbpf_type_to_btf_name[type]);
 804         map->name = strdup(map_name);
 805         if (!map->name) {
 806                 pr_warning("failed to alloc map name\n");
 807                 return -ENOMEM;
 808         }
 809
 810         def->type = BPF_MAP_TYPE_ARRAY;
 811         def->key_size = sizeof(int);
 812         def->value_size = data->d_size;
 813         def->max_entries = 1;
 814         def->map_flags = type == LIBBPF_MAP_RODATA ?
 815                          BPF_F_RDONLY_PROG : 0;
 816         if (data_buff) {
 817                 *data_buff = malloc(data->d_size);
 818                 if (!*data_buff) {
 819                         zfree(&map->name);
 820                         pr_warning("failed to alloc map content buffer\n");
 821                         return -ENOMEM;
 822                 }
 823                 memcpy(*data_buff, data->d_buf, data->d_size);
 824         }
 825
 826         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 827         return 0;
 828 }
 829
 830 static int
 831 bpf_object__init_maps(struct bpf_object *obj, int flags)
 832 {
 833         int i, map_idx, map_def_sz = 0, nr_syms, nr_maps = 0, nr_maps_glob = 0;
 834         bool strict = !(flags & MAPS_RELAX_COMPAT);
 835         Elf_Data *symbols = obj->efile.symbols;
 836         Elf_Data *data = NULL;
 837         int ret = 0;
 838
 839         if (!symbols)
 840                 return -EINVAL;
 841         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 842
 843         if (obj->efile.maps_shndx >= 0) {
 844                 Elf_Scn *scn = elf_getscn(obj->efile.elf,
 845                                           obj->efile.maps_shndx);
 846
 847                 if (scn)
 848                         data = elf_getdata(scn, NULL);
 849                 if (!scn || !data) {
 850                         pr_warning("failed to get Elf_Data from map section %d\n",
 851                                    obj->efile.maps_shndx);
 852                         return -EINVAL;
 853                 }
 854         }
 855
 856         /*
 857          * Count number of maps. Each map has a name.
 858          * Array of maps is not supported: only the first element is
 859          * considered.
 860          *
 861          * TODO: Detect array of map and report error.
 862          */
 863         if (obj->caps.global_data) {
 864                 if (obj->efile.data_shndx >= 0)
 865                         nr_maps_glob++;
 866                 if (obj->efile.rodata_shndx >= 0)
 867                         nr_maps_glob++;
 868                 if (obj->efile.bss_shndx >= 0)
 869                         nr_maps_glob++;
 870         }
 871
 872         for (i = 0; data && i < nr_syms; i++) {
 873                 GElf_Sym sym;
 874
 875                 if (!gelf_getsym(symbols, i, &sym))
 876                         continue;
 877                 if (sym.st_shndx != obj->efile.maps_shndx)
 878                         continue;
 879                 nr_maps++;
 880         }
 881
 882         if (!nr_maps && !nr_maps_glob)
 883                 return 0;
 884
 885         /* Assume equally sized map definitions */
 886         if (data) {
 887                 pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 888                          nr_maps, data->d_size);
 889
 890                 map_def_sz = data->d_size / nr_maps;
 891                 if (!data->d_size || (data->d_size % nr_maps) != 0) {
 892                         pr_warning("unable to determine map definition size "
 893                                    "section %s, %d maps in %zd bytes\n",
 894                                    obj->path, nr_maps, data->d_size);
 895                         return -EINVAL;
 896                 }
 897         }
 898
 899         nr_maps += nr_maps_glob;
 900         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 901         if (!obj->maps) {
 902                 pr_warning("alloc maps for object failed\n");
 903                 return -ENOMEM;
 904         }
 905         obj->nr_maps = nr_maps;
 906
 907         for (i = 0; i < nr_maps; i++) {
 908                 /*
 909                  * fill all fd with -1 so won't close incorrect
 910                  * fd (fd=0 is stdin) when failure (zclose won't close
 911                  * negative fd)).
 912                  */
 913                 obj->maps[i].fd = -1;
 914                 obj->maps[i].inner_map_fd = -1;
 915         }
 916
 917         /*
 918          * Fill obj->maps using data in "maps" section.
 919          */
 920         for (i = 0, map_idx = 0; data && i < nr_syms; i++) {
 921                 GElf_Sym sym;
 922                 const char *map_name;
 923                 struct bpf_map_def *def;
 924
 925                 if (!gelf_getsym(symbols, i, &sym))
 926                         continue;
 927                 if (sym.st_shndx != obj->efile.maps_shndx)
 928                         continue;
 929
 930                 map_name = elf_strptr(obj->efile.elf,
 931                                       obj->efile.strtabidx,
 932                                       sym.st_name);
 933
 934                 obj->maps[map_idx].libbpf_type = LIBBPF_MAP_UNSPEC;
 935                 obj->maps[map_idx].offset = sym.st_value;
 936                 if (sym.st_value + map_def_sz > data->d_size) {
 937                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 938                                    obj->path, map_name);
 939                         return -EINVAL;
 940                 }
 941
 942                 obj->maps[map_idx].name = strdup(map_name);
 943                 if (!obj->maps[map_idx].name) {
 944                         pr_warning("failed to alloc map name\n");
 945                         return -ENOMEM;
 946                 }
 947                 pr_debug("map %d is \"%s\"\n", map_idx,
 948                          obj->maps[map_idx].name);
 949                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 950                 /*
 951                  * If the definition of the map in the object file fits in
 952                  * bpf_map_def, copy it.  Any extra fields in our version
 953                  * of bpf_map_def will default to zero as a result of the
 954                  * calloc above.
 955                  */
 956                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 957                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 958                 } else {
 959                         /*
 960                          * Here the map structure being read is bigger than what
 961                          * we expect, truncate if the excess bits are all zero.
 962                          * If they are not zero, reject this map as
 963                          * incompatible.
 964                          */
 965                         char *b;
 966                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 967                              b < ((char *)def) + map_def_sz; b++) {
 968                                 if (*b != 0) {
 969                                         pr_warning("maps section in %s: \"%s\" "
 970                                                    "has unrecognized, non-zero "
 971                                                    "options\n",
 972                                                    obj->path, map_name);
 973                                         if (strict)
 974                                                 return -EINVAL;
 975                                 }
 976                         }
 977                         memcpy(&obj->maps[map_idx].def, def,
 978                                sizeof(struct bpf_map_def));
 979                 }
 980                 map_idx++;
 981         }
 982
 983         if (!obj->caps.global_data)
 984                 goto finalize;
 985
 986         /*
 987          * Populate rest of obj->maps with libbpf internal maps.
 988          */
 989         if (obj->efile.data_shndx >= 0)
 990                 ret = bpf_object__init_internal_map(obj, &obj->maps[map_idx++],
 991                                                     LIBBPF_MAP_DATA,
 992                                                     obj->efile.data,
 993                                                     &obj->sections.data);
 994         if (!ret && obj->efile.rodata_shndx >= 0)
 995                 ret = bpf_object__init_internal_map(obj, &obj->maps[map_idx++],
 996                                                     LIBBPF_MAP_RODATA,
 997                                                     obj->efile.rodata,
 998                                                     &obj->sections.rodata);
 999         if (!ret && obj->efile.bss_shndx >= 0)
1000                 ret = bpf_object__init_internal_map(obj, &obj->maps[map_idx++],
1001                                                     LIBBPF_MAP_BSS,
1002                                                     obj->efile.bss, NULL);
1003 finalize:
1004         if (!ret)
1005                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1006                       compare_bpf_map);
1007         return ret;
1008 }
1009
1010 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1011 {
1012         Elf_Scn *scn;
1013         GElf_Shdr sh;
1014
1015         scn = elf_getscn(obj->efile.elf, idx);
1016         if (!scn)
1017                 return false;
1018
1019         if (gelf_getshdr(scn, &sh) != &sh)
1020                 return false;
1021
1022         if (sh.sh_flags & SHF_EXECINSTR)
1023                 return true;
1024
1025         return false;
1026 }
1027
1028 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1029 {
1030         bool has_datasec = obj->caps.btf_datasec;
1031         bool has_func = obj->caps.btf_func;
1032         struct btf *btf = obj->btf;
1033         struct btf_type *t;
1034         int i, j, vlen;
1035         __u16 kind;
1036
1037         if (!obj->btf || (has_func && has_datasec))
1038                 return;
1039
1040         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1041                 t = (struct btf_type *)btf__type_by_id(btf, i);
1042                 kind = BTF_INFO_KIND(t->info);
1043
1044                 if (!has_datasec && kind == BTF_KIND_VAR) {
1045                         /* replace VAR with INT */
1046                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1047                         t->size = sizeof(int);
1048                         *(int *)(t+1) = BTF_INT_ENC(0, 0, 32);
1049                 } else if (!has_datasec && kind == BTF_KIND_DATASEC) {
1050                         /* replace DATASEC with STRUCT */
1051                         struct btf_var_secinfo *v = (void *)(t + 1);
1052                         struct btf_member *m = (void *)(t + 1);
1053                         struct btf_type *vt;
1054                         char *name;
1055
1056                         name = (char *)btf__name_by_offset(btf, t->name_off);
1057                         while (*name) {
1058                                 if (*name == '.')
1059                                         *name = '_';
1060                                 name++;
1061                         }
1062
1063                         vlen = BTF_INFO_VLEN(t->info);
1064                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1065                         for (j = 0; j < vlen; j++, v++, m++) {
1066                                 /* order of field assignments is important */
1067                                 m->offset = v->offset * 8;
1068                                 m->type = v->type;
1069                                 /* preserve variable name as member name */
1070                                 vt = (void *)btf__type_by_id(btf, v->type);
1071                                 m->name_off = vt->name_off;
1072                         }
1073                 } else if (!has_func && kind == BTF_KIND_FUNC_PROTO) {
1074                         /* replace FUNC_PROTO with ENUM */
1075                         vlen = BTF_INFO_VLEN(t->info);
1076                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1077                         t->size = sizeof(__u32); /* kernel enforced */
1078                 } else if (!has_func && kind == BTF_KIND_FUNC) {
1079                         /* replace FUNC with TYPEDEF */
1080                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1081                 }
1082         }
1083 }
1084
1085 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1086 {
1087         if (!obj->btf_ext)
1088                 return;
1089
1090         if (!obj->caps.btf_func) {
1091                 btf_ext__free(obj->btf_ext);
1092                 obj->btf_ext = NULL;
1093         }
1094 }
1095
1096 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1097 {
1098         Elf *elf = obj->efile.elf;
1099         GElf_Ehdr *ep = &obj->efile.ehdr;
1100         Elf_Data *btf_ext_data = NULL;
1101         Elf_Data *btf_data = NULL;
1102         Elf_Scn *scn = NULL;
1103         int idx = 0, err = 0;
1104
1105         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1106         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1107                 pr_warning("failed to get e_shstrndx from %s\n",
1108                            obj->path);
1109                 return -LIBBPF_ERRNO__FORMAT;
1110         }
1111
1112         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1113                 char *name;
1114                 GElf_Shdr sh;
1115                 Elf_Data *data;
1116
1117                 idx++;
1118                 if (gelf_getshdr(scn, &sh) != &sh) {
1119                         pr_warning("failed to get section(%d) header from %s\n",
1120                                    idx, obj->path);
1121                         err = -LIBBPF_ERRNO__FORMAT;
1122                         goto out;
1123                 }
1124
1125                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1126                 if (!name) {
1127                         pr_warning("failed to get section(%d) name from %s\n",
1128                                    idx, obj->path);
1129                         err = -LIBBPF_ERRNO__FORMAT;
1130                         goto out;
1131                 }
1132
1133                 data = elf_getdata(scn, 0);
1134                 if (!data) {
1135                         pr_warning("failed to get section(%d) data from %s(%s)\n",
1136                                    idx, name, obj->path);
1137                         err = -LIBBPF_ERRNO__FORMAT;
1138                         goto out;
1139                 }
1140                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1141                          idx, name, (unsigned long)data->d_size,
1142                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1143                          (int)sh.sh_type);
1144
1145                 if (strcmp(name, "license") == 0) {
1146                         err = bpf_object__init_license(obj,
1147                                                        data->d_buf,
1148                                                        data->d_size);
1149                 } else if (strcmp(name, "version") == 0) {
1150                         err = bpf_object__init_kversion(obj,
1151                                                         data->d_buf,
1152                                                         data->d_size);
1153                 } else if (strcmp(name, "maps") == 0) {
1154                         obj->efile.maps_shndx = idx;
1155                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1156                         btf_data = data;
1157                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1158                         btf_ext_data = data;
1159                 } else if (sh.sh_type == SHT_SYMTAB) {
1160                         if (obj->efile.symbols) {
1161                                 pr_warning("bpf: multiple SYMTAB in %s\n",
1162                                            obj->path);
1163                                 err = -LIBBPF_ERRNO__FORMAT;
1164                         } else {
1165                                 obj->efile.symbols = data;
1166                                 obj->efile.strtabidx = sh.sh_link;
1167                         }
1168                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1169                         if (sh.sh_flags & SHF_EXECINSTR) {
1170                                 if (strcmp(name, ".text") == 0)
1171                                         obj->efile.text_shndx = idx;
1172                                 err = bpf_object__add_program(obj, data->d_buf,
1173                                                               data->d_size, name, idx);
1174                                 if (err) {
1175                                         char errmsg[STRERR_BUFSIZE];
1176                                         char *cp = libbpf_strerror_r(-err, errmsg,
1177                                                                      sizeof(errmsg));
1178
1179                                         pr_warning("failed to alloc program %s (%s): %s",
1180                                                    name, obj->path, cp);
1181                                 }
1182                         } else if (strcmp(name, ".data") == 0) {
1183                                 obj->efile.data = data;
1184                                 obj->efile.data_shndx = idx;
1185                         } else if (strcmp(name, ".rodata") == 0) {
1186                                 obj->efile.rodata = data;
1187                                 obj->efile.rodata_shndx = idx;
1188                         } else {
1189                                 pr_debug("skip section(%d) %s\n", idx, name);
1190                         }
1191                 } else if (sh.sh_type == SHT_REL) {
1192                         void *reloc = obj->efile.reloc;
1193                         int nr_reloc = obj->efile.nr_reloc + 1;
1194                         int sec = sh.sh_info; /* points to other section */
1195
1196                         /* Only do relo for section with exec instructions */
1197                         if (!section_have_execinstr(obj, sec)) {
1198                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1199                                          name, idx, sec);
1200                                 continue;
1201                         }
1202
1203                         reloc = reallocarray(reloc, nr_reloc,
1204                                              sizeof(*obj->efile.reloc));
1205                         if (!reloc) {
1206                                 pr_warning("realloc failed\n");
1207                                 err = -ENOMEM;
1208                         } else {
1209                                 int n = nr_reloc - 1;
1210
1211                                 obj->efile.reloc = reloc;
1212                                 obj->efile.nr_reloc = nr_reloc;
1213
1214                                 obj->efile.reloc[n].shdr = sh;
1215                                 obj->efile.reloc[n].data = data;
1216                         }
1217                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1218                         obj->efile.bss = data;
1219                         obj->efile.bss_shndx = idx;
1220                 } else {
1221                         pr_debug("skip section(%d) %s\n", idx, name);
1222                 }
1223                 if (err)
1224                         goto out;
1225         }
1226
1227         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1228                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1229                 return LIBBPF_ERRNO__FORMAT;
1230         }
1231         if (btf_data) {
1232                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1233                 if (IS_ERR(obj->btf)) {
1234                         pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1235                                    BTF_ELF_SEC, PTR_ERR(obj->btf));
1236                         obj->btf = NULL;
1237                 } else {
1238                         err = btf__finalize_data(obj, obj->btf);
1239                         if (!err) {
1240                                 bpf_object__sanitize_btf(obj);
1241                                 err = btf__load(obj->btf);
1242                         }
1243                         if (err) {
1244                                 pr_warning("Error finalizing and loading %s into kernel: %d. Ignored and continue.\n",
1245                                            BTF_ELF_SEC, err);
1246                                 btf__free(obj->btf);
1247                                 obj->btf = NULL;
1248                                 err = 0;
1249                         }
1250                 }
1251         }
1252         if (btf_ext_data) {
1253                 if (!obj->btf) {
1254                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1255                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1256                 } else {
1257                         obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1258                                                     btf_ext_data->d_size);
1259                         if (IS_ERR(obj->btf_ext)) {
1260                                 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1261                                            BTF_EXT_ELF_SEC,
1262                                            PTR_ERR(obj->btf_ext));
1263                                 obj->btf_ext = NULL;
1264                         } else {
1265                                 bpf_object__sanitize_btf_ext(obj);
1266                         }
1267                 }
1268         }
1269         if (bpf_object__has_maps(obj)) {
1270                 err = bpf_object__init_maps(obj, flags);
1271                 if (err)
1272                         goto out;
1273         }
1274         err = bpf_object__init_prog_names(obj);
1275 out:
1276         return err;
1277 }
1278
1279 static struct bpf_program *
1280 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1281 {
1282         struct bpf_program *prog;
1283         size_t i;
1284
1285         for (i = 0; i < obj->nr_programs; i++) {
1286                 prog = &obj->programs[i];
1287                 if (prog->idx == idx)
1288                         return prog;
1289         }
1290         return NULL;
1291 }
1292
1293 struct bpf_program *
1294 bpf_object__find_program_by_title(struct bpf_object *obj, const char *title)
1295 {
1296         struct bpf_program *pos;
1297
1298         bpf_object__for_each_program(pos, obj) {
1299                 if (pos->section_name && !strcmp(pos->section_name, title))
1300                         return pos;
1301         }
1302         return NULL;
1303 }
1304
1305 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1306                                       int shndx)
1307 {
1308         return shndx == obj->efile.data_shndx ||
1309                shndx == obj->efile.bss_shndx ||
1310                shndx == obj->efile.rodata_shndx;
1311 }
1312
1313 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1314                                       int shndx)
1315 {
1316         return shndx == obj->efile.maps_shndx;
1317 }
1318
1319 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1320                                               int shndx)
1321 {
1322         return shndx == obj->efile.text_shndx ||
1323                bpf_object__shndx_is_maps(obj, shndx) ||
1324                bpf_object__shndx_is_data(obj, shndx);
1325 }
1326
1327 static enum libbpf_map_type
1328 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1329 {
1330         if (shndx == obj->efile.data_shndx)
1331                 return LIBBPF_MAP_DATA;
1332         else if (shndx == obj->efile.bss_shndx)
1333                 return LIBBPF_MAP_BSS;
1334         else if (shndx == obj->efile.rodata_shndx)
1335                 return LIBBPF_MAP_RODATA;
1336         else
1337                 return LIBBPF_MAP_UNSPEC;
1338 }
1339
1340 static int
1341 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1342                            Elf_Data *data, struct bpf_object *obj)
1343 {
1344         Elf_Data *symbols = obj->efile.symbols;
1345         struct bpf_map *maps = obj->maps;
1346         size_t nr_maps = obj->nr_maps;
1347         int i, nrels;
1348
1349         pr_debug("collecting relocating info for: '%s'\n",
1350                  prog->section_name);
1351         nrels = shdr->sh_size / shdr->sh_entsize;
1352
1353         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1354         if (!prog->reloc_desc) {
1355                 pr_warning("failed to alloc memory in relocation\n");
1356                 return -ENOMEM;
1357         }
1358         prog->nr_reloc = nrels;
1359
1360         for (i = 0; i < nrels; i++) {
1361                 GElf_Sym sym;
1362                 GElf_Rel rel;
1363                 unsigned int insn_idx;
1364                 unsigned int shdr_idx;
1365                 struct bpf_insn *insns = prog->insns;
1366                 enum libbpf_map_type type;
1367                 const char *name;
1368                 size_t map_idx;
1369
1370                 if (!gelf_getrel(data, i, &rel)) {
1371                         pr_warning("relocation: failed to get %d reloc\n", i);
1372                         return -LIBBPF_ERRNO__FORMAT;
1373                 }
1374
1375                 if (!gelf_getsym(symbols,
1376                                  GELF_R_SYM(rel.r_info),
1377                                  &sym)) {
1378                         pr_warning("relocation: symbol %"PRIx64" not found\n",
1379                                    GELF_R_SYM(rel.r_info));
1380                         return -LIBBPF_ERRNO__FORMAT;
1381                 }
1382
1383                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1384                                   sym.st_name) ? : "<?>";
1385
1386                 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1387                          (long long) (rel.r_info >> 32),
1388                          (long long) sym.st_value, sym.st_name, name);
1389
1390                 shdr_idx = sym.st_shndx;
1391                 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1392                         pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1393                                    prog->section_name, shdr_idx);
1394                         return -LIBBPF_ERRNO__RELOC;
1395                 }
1396
1397                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1398                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
1399
1400                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1401                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1402                                 pr_warning("incorrect bpf_call opcode\n");
1403                                 return -LIBBPF_ERRNO__RELOC;
1404                         }
1405                         prog->reloc_desc[i].type = RELO_CALL;
1406                         prog->reloc_desc[i].insn_idx = insn_idx;
1407                         prog->reloc_desc[i].text_off = sym.st_value;
1408                         obj->has_pseudo_calls = true;
1409                         continue;
1410                 }
1411
1412                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1413                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1414                                    insn_idx, insns[insn_idx].code);
1415                         return -LIBBPF_ERRNO__RELOC;
1416                 }
1417
1418                 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1419                     bpf_object__shndx_is_data(obj, shdr_idx)) {
1420                         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1421                         if (type != LIBBPF_MAP_UNSPEC) {
1422                                 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1423                                         pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1424                                                    name, insn_idx, insns[insn_idx].code);
1425                                         return -LIBBPF_ERRNO__RELOC;
1426                                 }
1427                                 if (!obj->caps.global_data) {
1428                                         pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1429                                                    name, insn_idx);
1430                                         return -LIBBPF_ERRNO__RELOC;
1431                                 }
1432                         }
1433
1434                         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1435                                 if (maps[map_idx].libbpf_type != type)
1436                                         continue;
1437                                 if (type != LIBBPF_MAP_UNSPEC ||
1438                                     (type == LIBBPF_MAP_UNSPEC &&
1439                                      maps[map_idx].offset == sym.st_value)) {
1440                                         pr_debug("relocation: find map %zd (%s) for insn %u\n",
1441                                                  map_idx, maps[map_idx].name, insn_idx);
1442                                         break;
1443                                 }
1444                         }
1445
1446                         if (map_idx >= nr_maps) {
1447                                 pr_warning("bpf relocation: map_idx %d large than %d\n",
1448                                            (int)map_idx, (int)nr_maps - 1);
1449                                 return -LIBBPF_ERRNO__RELOC;
1450                         }
1451
1452                         prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1453                                                    RELO_DATA : RELO_LD64;
1454                         prog->reloc_desc[i].insn_idx = insn_idx;
1455                         prog->reloc_desc[i].map_idx = map_idx;
1456                 }
1457         }
1458         return 0;
1459 }
1460
1461 static int bpf_map_find_btf_info(struct bpf_map *map, const struct btf *btf)
1462 {
1463         struct bpf_map_def *def = &map->def;
1464         __u32 key_type_id = 0, value_type_id = 0;
1465         int ret;
1466
1467         if (!bpf_map__is_internal(map)) {
1468                 ret = btf__get_map_kv_tids(btf, map->name, def->key_size,
1469                                            def->value_size, &key_type_id,
1470                                            &value_type_id);
1471         } else {
1472                 /*
1473                  * LLVM annotates global data differently in BTF, that is,
1474                  * only as '.data', '.bss' or '.rodata'.
1475                  */
1476                 ret = btf__find_by_name(btf,
1477                                 libbpf_type_to_btf_name[map->libbpf_type]);
1478         }
1479         if (ret < 0)
1480                 return ret;
1481
1482         map->btf_key_type_id = key_type_id;
1483         map->btf_value_type_id = bpf_map__is_internal(map) ?
1484                                  ret : value_type_id;
1485         return 0;
1486 }
1487
1488 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1489 {
1490         struct bpf_map_info info = {};
1491         __u32 len = sizeof(info);
1492         int new_fd, err;
1493         char *new_name;
1494
1495         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1496         if (err)
1497                 return err;
1498
1499         new_name = strdup(info.name);
1500         if (!new_name)
1501                 return -errno;
1502
1503         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1504         if (new_fd < 0)
1505                 goto err_free_new_name;
1506
1507         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1508         if (new_fd < 0)
1509                 goto err_close_new_fd;
1510
1511         err = zclose(map->fd);
1512         if (err)
1513                 goto err_close_new_fd;
1514         free(map->name);
1515
1516         map->fd = new_fd;
1517         map->name = new_name;
1518         map->def.type = info.type;
1519         map->def.key_size = info.key_size;
1520         map->def.value_size = info.value_size;
1521         map->def.max_entries = info.max_entries;
1522         map->def.map_flags = info.map_flags;
1523         map->btf_key_type_id = info.btf_key_type_id;
1524         map->btf_value_type_id = info.btf_value_type_id;
1525
1526         return 0;
1527
1528 err_close_new_fd:
1529         close(new_fd);
1530 err_free_new_name:
1531         free(new_name);
1532         return -errno;
1533 }
1534
1535 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1536 {
1537         if (!map || !max_entries)
1538                 return -EINVAL;
1539
1540         /* If map already created, its attributes can't be changed. */
1541         if (map->fd >= 0)
1542                 return -EBUSY;
1543
1544         map->def.max_entries = max_entries;
1545
1546         return 0;
1547 }
1548
1549 static int
1550 bpf_object__probe_name(struct bpf_object *obj)
1551 {
1552         struct bpf_load_program_attr attr;
1553         char *cp, errmsg[STRERR_BUFSIZE];
1554         struct bpf_insn insns[] = {
1555                 BPF_MOV64_IMM(BPF_REG_0, 0),
1556                 BPF_EXIT_INSN(),
1557         };
1558         int ret;
1559
1560         /* make sure basic loading works */
1561
1562         memset(&attr, 0, sizeof(attr));
1563         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1564         attr.insns = insns;
1565         attr.insns_cnt = ARRAY_SIZE(insns);
1566         attr.license = "GPL";
1567
1568         ret = bpf_load_program_xattr(&attr, NULL, 0);
1569         if (ret < 0) {
1570                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1571                 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1572                            __func__, cp, errno);
1573                 return -errno;
1574         }
1575         close(ret);
1576
1577         /* now try the same program, but with the name */
1578
1579         attr.name = "test";
1580         ret = bpf_load_program_xattr(&attr, NULL, 0);
1581         if (ret >= 0) {
1582                 obj->caps.name = 1;
1583                 close(ret);
1584         }
1585
1586         return 0;
1587 }
1588
1589 static int
1590 bpf_object__probe_global_data(struct bpf_object *obj)
1591 {
1592         struct bpf_load_program_attr prg_attr;
1593         struct bpf_create_map_attr map_attr;
1594         char *cp, errmsg[STRERR_BUFSIZE];
1595         struct bpf_insn insns[] = {
1596                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1597                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
1598                 BPF_MOV64_IMM(BPF_REG_0, 0),
1599                 BPF_EXIT_INSN(),
1600         };
1601         int ret, map;
1602
1603         memset(&map_attr, 0, sizeof(map_attr));
1604         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
1605         map_attr.key_size = sizeof(int);
1606         map_attr.value_size = 32;
1607         map_attr.max_entries = 1;
1608
1609         map = bpf_create_map_xattr(&map_attr);
1610         if (map < 0) {
1611                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1612                 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
1613                            __func__, cp, errno);
1614                 return -errno;
1615         }
1616
1617         insns[0].imm = map;
1618
1619         memset(&prg_attr, 0, sizeof(prg_attr));
1620         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1621         prg_attr.insns = insns;
1622         prg_attr.insns_cnt = ARRAY_SIZE(insns);
1623         prg_attr.license = "GPL";
1624
1625         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
1626         if (ret >= 0) {
1627                 obj->caps.global_data = 1;
1628                 close(ret);
1629         }
1630
1631         close(map);
1632         return 0;
1633 }
1634
1635 static int bpf_object__probe_btf_func(struct bpf_object *obj)
1636 {
1637         const char strs[] = "\0int\0x\0a";
1638         /* void x(int a) {} */
1639         __u32 types[] = {
1640                 /* int */
1641                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
1642                 /* FUNC_PROTO */                                /* [2] */
1643                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
1644                 BTF_PARAM_ENC(7, 1),
1645                 /* FUNC x */                                    /* [3] */
1646                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
1647         };
1648         int btf_fd;
1649
1650         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
1651                                       strs, sizeof(strs));
1652         if (btf_fd >= 0) {
1653                 obj->caps.btf_func = 1;
1654                 close(btf_fd);
1655                 return 1;
1656         }
1657
1658         return 0;
1659 }
1660
1661 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
1662 {
1663         const char strs[] = "\0x\0.data";
1664         /* static int a; */
1665         __u32 types[] = {
1666                 /* int */
1667                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
1668                 /* VAR x */                                     /* [2] */
1669                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
1670                 BTF_VAR_STATIC,
1671                 /* DATASEC val */                               /* [3] */
1672                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
1673                 BTF_VAR_SECINFO_ENC(2, 0, 4),
1674         };
1675         int btf_fd;
1676
1677         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
1678                                       strs, sizeof(strs));
1679         if (btf_fd >= 0) {
1680                 obj->caps.btf_datasec = 1;
1681                 close(btf_fd);
1682                 return 1;
1683         }
1684
1685         return 0;
1686 }
1687
1688 static int
1689 bpf_object__probe_caps(struct bpf_object *obj)
1690 {
1691         int (*probe_fn[])(struct bpf_object *obj) = {
1692                 bpf_object__probe_name,
1693                 bpf_object__probe_global_data,
1694                 bpf_object__probe_btf_func,
1695                 bpf_object__probe_btf_datasec,
1696         };
1697         int i, ret;
1698
1699         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
1700                 ret = probe_fn[i](obj);
1701                 if (ret < 0)
1702                         pr_debug("Probe #%d failed with %d.\n", i, ret);
1703         }
1704
1705         return 0;
1706 }
1707
1708 static int
1709 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
1710 {
1711         char *cp, errmsg[STRERR_BUFSIZE];
1712         int err, zero = 0;
1713         __u8 *data;
1714
1715         /* Nothing to do here since kernel already zero-initializes .bss map. */
1716         if (map->libbpf_type == LIBBPF_MAP_BSS)
1717                 return 0;
1718
1719         data = map->libbpf_type == LIBBPF_MAP_DATA ?
1720                obj->sections.data : obj->sections.rodata;
1721
1722         err = bpf_map_update_elem(map->fd, &zero, data, 0);
1723         /* Freeze .rodata map as read-only from syscall side. */
1724         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
1725                 err = bpf_map_freeze(map->fd);
1726                 if (err) {
1727                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1728                         pr_warning("Error freezing map(%s) as read-only: %s\n",
1729                                    map->name, cp);
1730                         err = 0;
1731                 }
1732         }
1733         return err;
1734 }
1735
1736 static int
1737 bpf_object__create_maps(struct bpf_object *obj)
1738 {
1739         struct bpf_create_map_attr create_attr = {};
1740         unsigned int i;
1741         int err;
1742
1743         for (i = 0; i < obj->nr_maps; i++) {
1744                 struct bpf_map *map = &obj->maps[i];
1745                 struct bpf_map_def *def = &map->def;
1746                 char *cp, errmsg[STRERR_BUFSIZE];
1747                 int *pfd = &map->fd;
1748
1749                 if (map->fd >= 0) {
1750                         pr_debug("skip map create (preset) %s: fd=%d\n",
1751                                  map->name, map->fd);
1752                         continue;
1753                 }
1754
1755                 if (obj->caps.name)
1756                         create_attr.name = map->name;
1757                 create_attr.map_ifindex = map->map_ifindex;
1758                 create_attr.map_type = def->type;
1759                 create_attr.map_flags = def->map_flags;
1760                 create_attr.key_size = def->key_size;
1761                 create_attr.value_size = def->value_size;
1762                 create_attr.max_entries = def->max_entries;
1763                 create_attr.btf_fd = 0;
1764                 create_attr.btf_key_type_id = 0;
1765                 create_attr.btf_value_type_id = 0;
1766                 if (bpf_map_type__is_map_in_map(def->type) &&
1767                     map->inner_map_fd >= 0)
1768                         create_attr.inner_map_fd = map->inner_map_fd;
1769
1770                 if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1771                         create_attr.btf_fd = btf__fd(obj->btf);
1772                         create_attr.btf_key_type_id = map->btf_key_type_id;
1773                         create_attr.btf_value_type_id = map->btf_value_type_id;
1774                 }
1775
1776                 *pfd = bpf_create_map_xattr(&create_attr);
1777                 if (*pfd < 0 && create_attr.btf_key_type_id) {
1778                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1779                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1780                                    map->name, cp, errno);
1781                         create_attr.btf_fd = 0;
1782                         create_attr.btf_key_type_id = 0;
1783                         create_attr.btf_value_type_id = 0;
1784                         map->btf_key_type_id = 0;
1785                         map->btf_value_type_id = 0;
1786                         *pfd = bpf_create_map_xattr(&create_attr);
1787                 }
1788
1789                 if (*pfd < 0) {
1790                         size_t j;
1791
1792                         err = *pfd;
1793 err_out:
1794                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1795                         pr_warning("failed to create map (name: '%s'): %s\n",
1796                                    map->name, cp);
1797                         for (j = 0; j < i; j++)
1798                                 zclose(obj->maps[j].fd);
1799                         return err;
1800                 }
1801
1802                 if (bpf_map__is_internal(map)) {
1803                         err = bpf_object__populate_internal_map(obj, map);
1804                         if (err < 0) {
1805                                 zclose(*pfd);
1806                                 goto err_out;
1807                         }
1808                 }
1809
1810                 pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1811         }
1812
1813         return 0;
1814 }
1815
1816 static int
1817 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
1818                         void *btf_prog_info, const char *info_name)
1819 {
1820         if (err != -ENOENT) {
1821                 pr_warning("Error in loading %s for sec %s.\n",
1822                            info_name, prog->section_name);
1823                 return err;
1824         }
1825
1826         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
1827
1828         if (btf_prog_info) {
1829                 /*
1830                  * Some info has already been found but has problem
1831                  * in the last btf_ext reloc.  Must have to error
1832                  * out.
1833                  */
1834                 pr_warning("Error in relocating %s for sec %s.\n",
1835                            info_name, prog->section_name);
1836                 return err;
1837         }
1838
1839         /*
1840          * Have problem loading the very first info.  Ignore
1841          * the rest.
1842          */
1843         pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
1844                    info_name, prog->section_name, info_name);
1845         return 0;
1846 }
1847
1848 static int
1849 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
1850                           const char *section_name,  __u32 insn_offset)
1851 {
1852         int err;
1853
1854         if (!insn_offset || prog->func_info) {
1855                 /*
1856                  * !insn_offset => main program
1857                  *
1858                  * For sub prog, the main program's func_info has to
1859                  * be loaded first (i.e. prog->func_info != NULL)
1860                  */
1861                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
1862                                                section_name, insn_offset,
1863                                                &prog->func_info,
1864                                                &prog->func_info_cnt);
1865                 if (err)
1866                         return check_btf_ext_reloc_err(prog, err,
1867                                                        prog->func_info,
1868                                                        "bpf_func_info");
1869
1870                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
1871         }
1872
1873         if (!insn_offset || prog->line_info) {
1874                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
1875                                                section_name, insn_offset,
1876                                                &prog->line_info,
1877                                                &prog->line_info_cnt);
1878                 if (err)
1879                         return check_btf_ext_reloc_err(prog, err,
1880                                                        prog->line_info,
1881                                                        "bpf_line_info");
1882
1883                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
1884         }
1885
1886         if (!insn_offset)
1887                 prog->btf_fd = btf__fd(obj->btf);
1888
1889         return 0;
1890 }
1891
1892 static int
1893 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1894                         struct reloc_desc *relo)
1895 {
1896         struct bpf_insn *insn, *new_insn;
1897         struct bpf_program *text;
1898         size_t new_cnt;
1899         int err;
1900
1901         if (relo->type != RELO_CALL)
1902                 return -LIBBPF_ERRNO__RELOC;
1903
1904         if (prog->idx == obj->efile.text_shndx) {
1905                 pr_warning("relo in .text insn %d into off %d\n",
1906                            relo->insn_idx, relo->text_off);
1907                 return -LIBBPF_ERRNO__RELOC;
1908         }
1909
1910         if (prog->main_prog_cnt == 0) {
1911                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1912                 if (!text) {
1913                         pr_warning("no .text section found yet relo into text exist\n");
1914                         return -LIBBPF_ERRNO__RELOC;
1915                 }
1916                 new_cnt = prog->insns_cnt + text->insns_cnt;
1917                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
1918                 if (!new_insn) {
1919                         pr_warning("oom in prog realloc\n");
1920                         return -ENOMEM;
1921                 }
1922
1923                 if (obj->btf_ext) {
1924                         err = bpf_program_reloc_btf_ext(prog, obj,
1925                                                         text->section_name,
1926                                                         prog->insns_cnt);
1927                         if (err)
1928                                 return err;
1929                 }
1930
1931                 memcpy(new_insn + prog->insns_cnt, text->insns,
1932                        text->insns_cnt * sizeof(*insn));
1933                 prog->insns = new_insn;
1934                 prog->main_prog_cnt = prog->insns_cnt;
1935                 prog->insns_cnt = new_cnt;
1936                 pr_debug("added %zd insn from %s to prog %s\n",
1937                          text->insns_cnt, text->section_name,
1938                          prog->section_name);
1939         }
1940         insn = &prog->insns[relo->insn_idx];
1941         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1942         return 0;
1943 }
1944
1945 static int
1946 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1947 {
1948         int i, err;
1949
1950         if (!prog)
1951                 return 0;
1952
1953         if (obj->btf_ext) {
1954                 err = bpf_program_reloc_btf_ext(prog, obj,
1955                                                 prog->section_name, 0);
1956                 if (err)
1957                         return err;
1958         }
1959
1960         if (!prog->reloc_desc)
1961                 return 0;
1962
1963         for (i = 0; i < prog->nr_reloc; i++) {
1964                 if (prog->reloc_desc[i].type == RELO_LD64 ||
1965                     prog->reloc_desc[i].type == RELO_DATA) {
1966                         bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
1967                         struct bpf_insn *insns = prog->insns;
1968                         int insn_idx, map_idx;
1969
1970                         insn_idx = prog->reloc_desc[i].insn_idx;
1971                         map_idx = prog->reloc_desc[i].map_idx;
1972
1973                         if (insn_idx + 1 >= (int)prog->insns_cnt) {
1974                                 pr_warning("relocation out of range: '%s'\n",
1975                                            prog->section_name);
1976                                 return -LIBBPF_ERRNO__RELOC;
1977                         }
1978
1979                         if (!relo_data) {
1980                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1981                         } else {
1982                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
1983                                 insns[insn_idx + 1].imm = insns[insn_idx].imm;
1984                         }
1985                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1986                 } else if (prog->reloc_desc[i].type == RELO_CALL) {
1987                         err = bpf_program__reloc_text(prog, obj,
1988                                                       &prog->reloc_desc[i]);
1989                         if (err)
1990                                 return err;
1991                 }
1992         }
1993
1994         zfree(&prog->reloc_desc);
1995         prog->nr_reloc = 0;
1996         return 0;
1997 }
1998
1999
2000 static int
2001 bpf_object__relocate(struct bpf_object *obj)
2002 {
2003         struct bpf_program *prog;
2004         size_t i;
2005         int err;
2006
2007         for (i = 0; i < obj->nr_programs; i++) {
2008                 prog = &obj->programs[i];
2009
2010                 err = bpf_program__relocate(prog, obj);
2011                 if (err) {
2012                         pr_warning("failed to relocate '%s'\n",
2013                                    prog->section_name);
2014                         return err;
2015                 }
2016         }
2017         return 0;
2018 }
2019
2020 static int bpf_object__collect_reloc(struct bpf_object *obj)
2021 {
2022         int i, err;
2023
2024         if (!obj_elf_valid(obj)) {
2025                 pr_warning("Internal error: elf object is closed\n");
2026                 return -LIBBPF_ERRNO__INTERNAL;
2027         }
2028
2029         for (i = 0; i < obj->efile.nr_reloc; i++) {
2030                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
2031                 Elf_Data *data = obj->efile.reloc[i].data;
2032                 int idx = shdr->sh_info;
2033                 struct bpf_program *prog;
2034
2035                 if (shdr->sh_type != SHT_REL) {
2036                         pr_warning("internal error at %d\n", __LINE__);
2037                         return -LIBBPF_ERRNO__INTERNAL;
2038                 }
2039
2040                 prog = bpf_object__find_prog_by_idx(obj, idx);
2041                 if (!prog) {
2042                         pr_warning("relocation failed: no section(%d)\n", idx);
2043                         return -LIBBPF_ERRNO__RELOC;
2044                 }
2045
2046                 err = bpf_program__collect_reloc(prog,
2047                                                  shdr, data,
2048                                                  obj);
2049                 if (err)
2050                         return err;
2051         }
2052         return 0;
2053 }
2054
2055 static int
2056 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
2057              char *license, __u32 kern_version, int *pfd)
2058 {
2059         struct bpf_load_program_attr load_attr;
2060         char *cp, errmsg[STRERR_BUFSIZE];
2061         int log_buf_size = BPF_LOG_BUF_SIZE;
2062         char *log_buf;
2063         int ret;
2064
2065         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
2066         load_attr.prog_type = prog->type;
2067         load_attr.expected_attach_type = prog->expected_attach_type;
2068         if (prog->caps->name)
2069                 load_attr.name = prog->name;
2070         load_attr.insns = insns;
2071         load_attr.insns_cnt = insns_cnt;
2072         load_attr.license = license;
2073         load_attr.kern_version = kern_version;
2074         load_attr.prog_ifindex = prog->prog_ifindex;
2075         load_attr.prog_btf_fd = prog->btf_fd >= 0 ? prog->btf_fd : 0;
2076         load_attr.func_info = prog->func_info;
2077         load_attr.func_info_rec_size = prog->func_info_rec_size;
2078         load_attr.func_info_cnt = prog->func_info_cnt;
2079         load_attr.line_info = prog->line_info;
2080         load_attr.line_info_rec_size = prog->line_info_rec_size;
2081         load_attr.line_info_cnt = prog->line_info_cnt;
2082         load_attr.log_level = prog->log_level;
2083         if (!load_attr.insns || !load_attr.insns_cnt)
2084                 return -EINVAL;
2085
2086 retry_load:
2087         log_buf = malloc(log_buf_size);
2088         if (!log_buf)
2089                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
2090
2091         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
2092
2093         if (ret >= 0) {
2094                 if (load_attr.log_level)
2095                         pr_debug("verifier log:\n%s", log_buf);
2096                 *pfd = ret;
2097                 ret = 0;
2098                 goto out;
2099         }
2100
2101         if (errno == ENOSPC) {
2102                 log_buf_size <<= 1;
2103                 free(log_buf);
2104                 goto retry_load;
2105         }
2106         ret = -LIBBPF_ERRNO__LOAD;
2107         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2108         pr_warning("load bpf program failed: %s\n", cp);
2109
2110         if (log_buf && log_buf[0] != '\0') {
2111                 ret = -LIBBPF_ERRNO__VERIFY;
2112                 pr_warning("-- BEGIN DUMP LOG ---\n");
2113                 pr_warning("\n%s\n", log_buf);
2114                 pr_warning("-- END LOG --\n");
2115         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
2116                 pr_warning("Program too large (%zu insns), at most %d insns\n",
2117                            load_attr.insns_cnt, BPF_MAXINSNS);
2118                 ret = -LIBBPF_ERRNO__PROG2BIG;
2119         } else {
2120                 /* Wrong program type? */
2121                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
2122                         int fd;
2123
2124                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
2125                         load_attr.expected_attach_type = 0;
2126                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
2127                         if (fd >= 0) {
2128                                 close(fd);
2129                                 ret = -LIBBPF_ERRNO__PROGTYPE;
2130                                 goto out;
2131                         }
2132                 }
2133
2134                 if (log_buf)
2135                         ret = -LIBBPF_ERRNO__KVER;
2136         }
2137
2138 out:
2139         free(log_buf);
2140         return ret;
2141 }
2142
2143 int
2144 bpf_program__load(struct bpf_program *prog,
2145                   char *license, __u32 kern_version)
2146 {
2147         int err = 0, fd, i;
2148
2149         if (prog->instances.nr < 0 || !prog->instances.fds) {
2150                 if (prog->preprocessor) {
2151                         pr_warning("Internal error: can't load program '%s'\n",
2152                                    prog->section_name);
2153                         return -LIBBPF_ERRNO__INTERNAL;
2154                 }
2155
2156                 prog->instances.fds = malloc(sizeof(int));
2157                 if (!prog->instances.fds) {
2158                         pr_warning("Not enough memory for BPF fds\n");
2159                         return -ENOMEM;
2160                 }
2161                 prog->instances.nr = 1;
2162                 prog->instances.fds[0] = -1;
2163         }
2164
2165         if (!prog->preprocessor) {
2166                 if (prog->instances.nr != 1) {
2167                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
2168                                    prog->section_name, prog->instances.nr);
2169                 }
2170                 err = load_program(prog, prog->insns, prog->insns_cnt,
2171                                    license, kern_version, &fd);
2172                 if (!err)
2173                         prog->instances.fds[0] = fd;
2174                 goto out;
2175         }
2176
2177         for (i = 0; i < prog->instances.nr; i++) {
2178                 struct bpf_prog_prep_result result;
2179                 bpf_program_prep_t preprocessor = prog->preprocessor;
2180
2181                 memset(&result, 0, sizeof(result));
2182                 err = preprocessor(prog, i, prog->insns,
2183                                    prog->insns_cnt, &result);
2184                 if (err) {
2185                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
2186                                    i, prog->section_name);
2187                         goto out;
2188                 }
2189
2190                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
2191                         pr_debug("Skip loading the %dth instance of program '%s'\n",
2192                                  i, prog->section_name);
2193                         prog->instances.fds[i] = -1;
2194                         if (result.pfd)
2195                                 *result.pfd = -1;
2196                         continue;
2197                 }
2198
2199                 err = load_program(prog, result.new_insn_ptr,
2200                                    result.new_insn_cnt,
2201                                    license, kern_version, &fd);
2202
2203                 if (err) {
2204                         pr_warning("Loading the %dth instance of program '%s' failed\n",
2205                                         i, prog->section_name);
2206                         goto out;
2207                 }
2208
2209                 if (result.pfd)
2210                         *result.pfd = fd;
2211                 prog->instances.fds[i] = fd;
2212         }
2213 out:
2214         if (err)
2215                 pr_warning("failed to load program '%s'\n",
2216                            prog->section_name);
2217         zfree(&prog->insns);
2218         prog->insns_cnt = 0;
2219         return err;
2220 }
2221
2222 static bool bpf_program__is_function_storage(struct bpf_program *prog,
2223                                              struct bpf_object *obj)
2224 {
2225         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
2226 }
2227
2228 static int
2229 bpf_object__load_progs(struct bpf_object *obj)
2230 {
2231         size_t i;
2232         int err;
2233
2234         for (i = 0; i < obj->nr_programs; i++) {
2235                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
2236                         continue;
2237                 err = bpf_program__load(&obj->programs[i],
2238                                         obj->license,
2239                                         obj->kern_version);
2240                 if (err)
2241                         return err;
2242         }
2243         return 0;
2244 }
2245
2246 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
2247 {
2248         switch (type) {
2249         case BPF_PROG_TYPE_SOCKET_FILTER:
2250         case BPF_PROG_TYPE_SCHED_CLS:
2251         case BPF_PROG_TYPE_SCHED_ACT:
2252         case BPF_PROG_TYPE_XDP:
2253         case BPF_PROG_TYPE_CGROUP_SKB:
2254         case BPF_PROG_TYPE_CGROUP_SOCK:
2255         case BPF_PROG_TYPE_LWT_IN:
2256         case BPF_PROG_TYPE_LWT_OUT:
2257         case BPF_PROG_TYPE_LWT_XMIT:
2258         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2259         case BPF_PROG_TYPE_SOCK_OPS:
2260         case BPF_PROG_TYPE_SK_SKB:
2261         case BPF_PROG_TYPE_CGROUP_DEVICE:
2262         case BPF_PROG_TYPE_SK_MSG:
2263         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2264         case BPF_PROG_TYPE_LIRC_MODE2:
2265         case BPF_PROG_TYPE_SK_REUSEPORT:
2266         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2267         case BPF_PROG_TYPE_UNSPEC:
2268         case BPF_PROG_TYPE_TRACEPOINT:
2269         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2270         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2271         case BPF_PROG_TYPE_PERF_EVENT:
2272         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2273                 return false;
2274         case BPF_PROG_TYPE_KPROBE:
2275         default:
2276                 return true;
2277         }
2278 }
2279
2280 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
2281 {
2282         if (needs_kver && obj->kern_version == 0) {
2283                 pr_warning("%s doesn't provide kernel version\n",
2284                            obj->path);
2285                 return -LIBBPF_ERRNO__KVERSION;
2286         }
2287         return 0;
2288 }
2289
2290 static struct bpf_object *
2291 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
2292                    bool needs_kver, int flags)
2293 {
2294         struct bpf_object *obj;
2295         int err;
2296
2297         if (elf_version(EV_CURRENT) == EV_NONE) {
2298                 pr_warning("failed to init libelf for %s\n", path);
2299                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
2300         }
2301
2302         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
2303         if (IS_ERR(obj))
2304                 return obj;
2305
2306         CHECK_ERR(bpf_object__elf_init(obj), err, out);
2307         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
2308         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
2309         CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
2310         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
2311         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
2312
2313         bpf_object__elf_finish(obj);
2314         return obj;
2315 out:
2316         bpf_object__close(obj);
2317         return ERR_PTR(err);
2318 }
2319
2320 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
2321                                             int flags)
2322 {
2323         /* param validation */
2324         if (!attr->file)
2325                 return NULL;
2326
2327         pr_debug("loading %s\n", attr->file);
2328
2329         return __bpf_object__open(attr->file, NULL, 0,
2330                                   bpf_prog_type__needs_kver(attr->prog_type),
2331                                   flags);
2332 }
2333
2334 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
2335 {
2336         return __bpf_object__open_xattr(attr, 0);
2337 }
2338
2339 struct bpf_object *bpf_object__open(const char *path)
2340 {
2341         struct bpf_object_open_attr attr = {
2342                 .file           = path,
2343                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
2344         };
2345
2346         return bpf_object__open_xattr(&attr);
2347 }
2348
2349 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
2350                                            size_t obj_buf_sz,
2351                                            const char *name)
2352 {
2353         char tmp_name[64];
2354
2355         /* param validation */
2356         if (!obj_buf || obj_buf_sz <= 0)
2357                 return NULL;
2358
2359         if (!name) {
2360                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
2361                          (unsigned long)obj_buf,
2362                          (unsigned long)obj_buf_sz);
2363                 tmp_name[sizeof(tmp_name) - 1] = '\0';
2364                 name = tmp_name;
2365         }
2366         pr_debug("loading object '%s' from buffer\n",
2367                  name);
2368
2369         return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
2370 }
2371
2372 int bpf_object__unload(struct bpf_object *obj)
2373 {
2374         size_t i;
2375
2376         if (!obj)
2377                 return -EINVAL;
2378
2379         for (i = 0; i < obj->nr_maps; i++)
2380                 zclose(obj->maps[i].fd);
2381
2382         for (i = 0; i < obj->nr_programs; i++)
2383                 bpf_program__unload(&obj->programs[i]);
2384
2385         return 0;
2386 }
2387
2388 int bpf_object__load(struct bpf_object *obj)
2389 {
2390         int err;
2391
2392         if (!obj)
2393                 return -EINVAL;
2394
2395         if (obj->loaded) {
2396                 pr_warning("object should not be loaded twice\n");
2397                 return -EINVAL;
2398         }
2399
2400         obj->loaded = true;
2401
2402         CHECK_ERR(bpf_object__create_maps(obj), err, out);
2403         CHECK_ERR(bpf_object__relocate(obj), err, out);
2404         CHECK_ERR(bpf_object__load_progs(obj), err, out);
2405
2406         return 0;
2407 out:
2408         bpf_object__unload(obj);
2409         pr_warning("failed to load object '%s'\n", obj->path);
2410         return err;
2411 }
2412
2413 static int check_path(const char *path)
2414 {
2415         char *cp, errmsg[STRERR_BUFSIZE];
2416         struct statfs st_fs;
2417         char *dname, *dir;
2418         int err = 0;
2419
2420         if (path == NULL)
2421                 return -EINVAL;
2422
2423         dname = strdup(path);
2424         if (dname == NULL)
2425                 return -ENOMEM;
2426
2427         dir = dirname(dname);
2428         if (statfs(dir, &st_fs)) {
2429                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2430                 pr_warning("failed to statfs %s: %s\n", dir, cp);
2431                 err = -errno;
2432         }
2433         free(dname);
2434
2435         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
2436                 pr_warning("specified path %s is not on BPF FS\n", path);
2437                 err = -EINVAL;
2438         }
2439
2440         return err;
2441 }
2442
2443 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
2444                               int instance)
2445 {
2446         char *cp, errmsg[STRERR_BUFSIZE];
2447         int err;
2448
2449         err = check_path(path);
2450         if (err)
2451                 return err;
2452
2453         if (prog == NULL) {
2454                 pr_warning("invalid program pointer\n");
2455                 return -EINVAL;
2456         }
2457
2458         if (instance < 0 || instance >= prog->instances.nr) {
2459                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2460                            instance, prog->section_name, prog->instances.nr);
2461                 return -EINVAL;
2462         }
2463
2464         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
2465                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2466                 pr_warning("failed to pin program: %s\n", cp);
2467                 return -errno;
2468         }
2469         pr_debug("pinned program '%s'\n", path);
2470
2471         return 0;
2472 }
2473
2474 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
2475                                 int instance)
2476 {
2477         int err;
2478
2479         err = check_path(path);
2480         if (err)
2481                 return err;
2482
2483         if (prog == NULL) {
2484                 pr_warning("invalid program pointer\n");
2485                 return -EINVAL;
2486         }
2487
2488         if (instance < 0 || instance >= prog->instances.nr) {
2489                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2490                            instance, prog->section_name, prog->instances.nr);
2491                 return -EINVAL;
2492         }
2493
2494         err = unlink(path);
2495         if (err != 0)
2496                 return -errno;
2497         pr_debug("unpinned program '%s'\n", path);
2498
2499         return 0;
2500 }
2501
2502 static int make_dir(const char *path)
2503 {
2504         char *cp, errmsg[STRERR_BUFSIZE];
2505         int err = 0;
2506
2507         if (mkdir(path, 0700) && errno != EEXIST)
2508                 err = -errno;
2509
2510         if (err) {
2511                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
2512                 pr_warning("failed to mkdir %s: %s\n", path, cp);
2513         }
2514         return err;
2515 }
2516
2517 int bpf_program__pin(struct bpf_program *prog, const char *path)
2518 {
2519         int i, err;
2520
2521         err = check_path(path);
2522         if (err)
2523                 return err;
2524
2525         if (prog == NULL) {
2526                 pr_warning("invalid program pointer\n");
2527                 return -EINVAL;
2528         }
2529
2530         if (prog->instances.nr <= 0) {
2531                 pr_warning("no instances of prog %s to pin\n",
2532                            prog->section_name);
2533                 return -EINVAL;
2534         }
2535
2536         if (prog->instances.nr == 1) {
2537                 /* don't create subdirs when pinning single instance */
2538                 return bpf_program__pin_instance(prog, path, 0);
2539         }
2540
2541         err = make_dir(path);
2542         if (err)
2543                 return err;
2544
2545         for (i = 0; i < prog->instances.nr; i++) {
2546                 char buf[PATH_MAX];
2547                 int len;
2548
2549                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2550                 if (len < 0) {
2551                         err = -EINVAL;
2552                         goto err_unpin;
2553                 } else if (len >= PATH_MAX) {
2554                         err = -ENAMETOOLONG;
2555                         goto err_unpin;
2556                 }
2557
2558                 err = bpf_program__pin_instance(prog, buf, i);
2559                 if (err)
2560                         goto err_unpin;
2561         }
2562
2563         return 0;
2564
2565 err_unpin:
2566         for (i = i - 1; i >= 0; i--) {
2567                 char buf[PATH_MAX];
2568                 int len;
2569
2570                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2571                 if (len < 0)
2572                         continue;
2573                 else if (len >= PATH_MAX)
2574                         continue;
2575
2576                 bpf_program__unpin_instance(prog, buf, i);
2577         }
2578
2579         rmdir(path);
2580
2581         return err;
2582 }
2583
2584 int bpf_program__unpin(struct bpf_program *prog, const char *path)
2585 {
2586         int i, err;
2587
2588         err = check_path(path);
2589         if (err)
2590                 return err;
2591
2592         if (prog == NULL) {
2593                 pr_warning("invalid program pointer\n");
2594                 return -EINVAL;
2595         }
2596
2597         if (prog->instances.nr <= 0) {
2598                 pr_warning("no instances of prog %s to pin\n",
2599                            prog->section_name);
2600                 return -EINVAL;
2601         }
2602
2603         if (prog->instances.nr == 1) {
2604                 /* don't create subdirs when pinning single instance */
2605                 return bpf_program__unpin_instance(prog, path, 0);
2606         }
2607
2608         for (i = 0; i < prog->instances.nr; i++) {
2609                 char buf[PATH_MAX];
2610                 int len;
2611
2612                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2613                 if (len < 0)
2614                         return -EINVAL;
2615                 else if (len >= PATH_MAX)
2616                         return -ENAMETOOLONG;
2617
2618                 err = bpf_program__unpin_instance(prog, buf, i);
2619                 if (err)
2620                         return err;
2621         }
2622
2623         err = rmdir(path);
2624         if (err)
2625                 return -errno;
2626
2627         return 0;
2628 }
2629
2630 int bpf_map__pin(struct bpf_map *map, const char *path)
2631 {
2632         char *cp, errmsg[STRERR_BUFSIZE];
2633         int err;
2634
2635         err = check_path(path);
2636         if (err)
2637                 return err;
2638
2639         if (map == NULL) {
2640                 pr_warning("invalid map pointer\n");
2641                 return -EINVAL;
2642         }
2643
2644         if (bpf_obj_pin(map->fd, path)) {
2645                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2646                 pr_warning("failed to pin map: %s\n", cp);
2647                 return -errno;
2648         }
2649
2650         pr_debug("pinned map '%s'\n", path);
2651
2652         return 0;
2653 }
2654
2655 int bpf_map__unpin(struct bpf_map *map, const char *path)
2656 {
2657         int err;
2658
2659         err = check_path(path);
2660         if (err)
2661                 return err;
2662
2663         if (map == NULL) {
2664                 pr_warning("invalid map pointer\n");
2665                 return -EINVAL;
2666         }
2667
2668         err = unlink(path);
2669         if (err != 0)
2670                 return -errno;
2671         pr_debug("unpinned map '%s'\n", path);
2672
2673         return 0;
2674 }
2675
2676 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
2677 {
2678         struct bpf_map *map;
2679         int err;
2680
2681         if (!obj)
2682                 return -ENOENT;
2683
2684         if (!obj->loaded) {
2685                 pr_warning("object not yet loaded; load it first\n");
2686                 return -ENOENT;
2687         }
2688
2689         err = make_dir(path);
2690         if (err)
2691                 return err;
2692
2693         bpf_object__for_each_map(map, obj) {
2694                 char buf[PATH_MAX];
2695                 int len;
2696
2697                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2698                                bpf_map__name(map));
2699                 if (len < 0) {
2700                         err = -EINVAL;
2701                         goto err_unpin_maps;
2702                 } else if (len >= PATH_MAX) {
2703                         err = -ENAMETOOLONG;
2704                         goto err_unpin_maps;
2705                 }
2706
2707                 err = bpf_map__pin(map, buf);
2708                 if (err)
2709                         goto err_unpin_maps;
2710         }
2711
2712         return 0;
2713
2714 err_unpin_maps:
2715         while ((map = bpf_map__prev(map, obj))) {
2716                 char buf[PATH_MAX];
2717                 int len;
2718
2719                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2720                                bpf_map__name(map));
2721                 if (len < 0)
2722                         continue;
2723                 else if (len >= PATH_MAX)
2724                         continue;
2725
2726                 bpf_map__unpin(map, buf);
2727         }
2728
2729         return err;
2730 }
2731
2732 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
2733 {
2734         struct bpf_map *map;
2735         int err;
2736
2737         if (!obj)
2738                 return -ENOENT;
2739
2740         bpf_object__for_each_map(map, obj) {
2741                 char buf[PATH_MAX];
2742                 int len;
2743
2744                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2745                                bpf_map__name(map));
2746                 if (len < 0)
2747                         return -EINVAL;
2748                 else if (len >= PATH_MAX)
2749                         return -ENAMETOOLONG;
2750
2751                 err = bpf_map__unpin(map, buf);
2752                 if (err)
2753                         return err;
2754         }
2755
2756         return 0;
2757 }
2758
2759 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
2760 {
2761         struct bpf_program *prog;
2762         int err;
2763
2764         if (!obj)
2765                 return -ENOENT;
2766
2767         if (!obj->loaded) {
2768                 pr_warning("object not yet loaded; load it first\n");
2769                 return -ENOENT;
2770         }
2771
2772         err = make_dir(path);
2773         if (err)
2774                 return err;
2775
2776         bpf_object__for_each_program(prog, obj) {
2777                 char buf[PATH_MAX];
2778                 int len;
2779
2780                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2781                                prog->pin_name);
2782                 if (len < 0) {
2783                         err = -EINVAL;
2784                         goto err_unpin_programs;
2785                 } else if (len >= PATH_MAX) {
2786                         err = -ENAMETOOLONG;
2787                         goto err_unpin_programs;
2788                 }
2789
2790                 err = bpf_program__pin(prog, buf);
2791                 if (err)
2792                         goto err_unpin_programs;
2793         }
2794
2795         return 0;
2796
2797 err_unpin_programs:
2798         while ((prog = bpf_program__prev(prog, obj))) {
2799                 char buf[PATH_MAX];
2800                 int len;
2801
2802                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2803                                prog->pin_name);
2804                 if (len < 0)
2805                         continue;
2806                 else if (len >= PATH_MAX)
2807                         continue;
2808
2809                 bpf_program__unpin(prog, buf);
2810         }
2811
2812         return err;
2813 }
2814
2815 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
2816 {
2817         struct bpf_program *prog;
2818         int err;
2819
2820         if (!obj)
2821                 return -ENOENT;
2822
2823         bpf_object__for_each_program(prog, obj) {
2824                 char buf[PATH_MAX];
2825                 int len;
2826
2827                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
2828                                prog->pin_name);
2829                 if (len < 0)
2830                         return -EINVAL;
2831                 else if (len >= PATH_MAX)
2832                         return -ENAMETOOLONG;
2833
2834                 err = bpf_program__unpin(prog, buf);
2835                 if (err)
2836                         return err;
2837         }
2838
2839         return 0;
2840 }
2841
2842 int bpf_object__pin(struct bpf_object *obj, const char *path)
2843 {
2844         int err;
2845
2846         err = bpf_object__pin_maps(obj, path);
2847         if (err)
2848                 return err;
2849
2850         err = bpf_object__pin_programs(obj, path);
2851         if (err) {
2852                 bpf_object__unpin_maps(obj, path);
2853                 return err;
2854         }
2855
2856         return 0;
2857 }
2858
2859 void bpf_object__close(struct bpf_object *obj)
2860 {
2861         size_t i;
2862
2863         if (!obj)
2864                 return;
2865
2866         if (obj->clear_priv)
2867                 obj->clear_priv(obj, obj->priv);
2868
2869         bpf_object__elf_finish(obj);
2870         bpf_object__unload(obj);
2871         btf__free(obj->btf);
2872         btf_ext__free(obj->btf_ext);
2873
2874         for (i = 0; i < obj->nr_maps; i++) {
2875                 zfree(&obj->maps[i].name);
2876                 if (obj->maps[i].clear_priv)
2877                         obj->maps[i].clear_priv(&obj->maps[i],
2878                                                 obj->maps[i].priv);
2879                 obj->maps[i].priv = NULL;
2880                 obj->maps[i].clear_priv = NULL;
2881         }
2882
2883         zfree(&obj->sections.rodata);
2884         zfree(&obj->sections.data);
2885         zfree(&obj->maps);
2886         obj->nr_maps = 0;
2887
2888         if (obj->programs && obj->nr_programs) {
2889                 for (i = 0; i < obj->nr_programs; i++)
2890                         bpf_program__exit(&obj->programs[i]);
2891         }
2892         zfree(&obj->programs);
2893
2894         list_del(&obj->list);
2895         free(obj);
2896 }
2897
2898 struct bpf_object *
2899 bpf_object__next(struct bpf_object *prev)
2900 {
2901         struct bpf_object *next;
2902
2903         if (!prev)
2904                 next = list_first_entry(&bpf_objects_list,
2905                                         struct bpf_object,
2906                                         list);
2907         else
2908                 next = list_next_entry(prev, list);
2909
2910         /* Empty list is noticed here so don't need checking on entry. */
2911         if (&next->list == &bpf_objects_list)
2912                 return NULL;
2913
2914         return next;
2915 }
2916
2917 const char *bpf_object__name(struct bpf_object *obj)
2918 {
2919         return obj ? obj->path : ERR_PTR(-EINVAL);
2920 }
2921
2922 unsigned int bpf_object__kversion(struct bpf_object *obj)
2923 {
2924         return obj ? obj->kern_version : 0;
2925 }
2926
2927 struct btf *bpf_object__btf(struct bpf_object *obj)
2928 {
2929         return obj ? obj->btf : NULL;
2930 }
2931
2932 int bpf_object__btf_fd(const struct bpf_object *obj)
2933 {
2934         return obj->btf ? btf__fd(obj->btf) : -1;
2935 }
2936
2937 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
2938                          bpf_object_clear_priv_t clear_priv)
2939 {
2940         if (obj->priv && obj->clear_priv)
2941                 obj->clear_priv(obj, obj->priv);
2942
2943         obj->priv = priv;
2944         obj->clear_priv = clear_priv;
2945         return 0;
2946 }
2947
2948 void *bpf_object__priv(struct bpf_object *obj)
2949 {
2950         return obj ? obj->priv : ERR_PTR(-EINVAL);
2951 }
2952
2953 static struct bpf_program *
2954 __bpf_program__iter(struct bpf_program *p, struct bpf_object *obj, bool forward)
2955 {
2956         size_t nr_programs = obj->nr_programs;
2957         ssize_t idx;
2958
2959         if (!nr_programs)
2960                 return NULL;
2961
2962         if (!p)
2963                 /* Iter from the beginning */
2964                 return forward ? &obj->programs[0] :
2965                         &obj->programs[nr_programs - 1];
2966
2967         if (p->obj != obj) {
2968                 pr_warning("error: program handler doesn't match object\n");
2969                 return NULL;
2970         }
2971
2972         idx = (p - obj->programs) + (forward ? 1 : -1);
2973         if (idx >= obj->nr_programs || idx < 0)
2974                 return NULL;
2975         return &obj->programs[idx];
2976 }
2977
2978 struct bpf_program *
2979 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
2980 {
2981         struct bpf_program *prog = prev;
2982
2983         do {
2984                 prog = __bpf_program__iter(prog, obj, true);
2985         } while (prog && bpf_program__is_function_storage(prog, obj));
2986
2987         return prog;
2988 }
2989
2990 struct bpf_program *
2991 bpf_program__prev(struct bpf_program *next, struct bpf_object *obj)
2992 {
2993         struct bpf_program *prog = next;
2994
2995         do {
2996                 prog = __bpf_program__iter(prog, obj, false);
2997         } while (prog && bpf_program__is_function_storage(prog, obj));
2998
2999         return prog;
3000 }
3001
3002 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
3003                           bpf_program_clear_priv_t clear_priv)
3004 {
3005         if (prog->priv && prog->clear_priv)
3006                 prog->clear_priv(prog, prog->priv);
3007
3008         prog->priv = priv;
3009         prog->clear_priv = clear_priv;
3010         return 0;
3011 }
3012
3013 void *bpf_program__priv(struct bpf_program *prog)
3014 {
3015         return prog ? prog->priv : ERR_PTR(-EINVAL);
3016 }
3017
3018 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
3019 {
3020         prog->prog_ifindex = ifindex;
3021 }
3022
3023 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
3024 {
3025         const char *title;
3026
3027         title = prog->section_name;
3028         if (needs_copy) {
3029                 title = strdup(title);
3030                 if (!title) {
3031                         pr_warning("failed to strdup program title\n");
3032                         return ERR_PTR(-ENOMEM);
3033                 }
3034         }
3035
3036         return title;
3037 }
3038
3039 int bpf_program__fd(struct bpf_program *prog)
3040 {
3041         return bpf_program__nth_fd(prog, 0);
3042 }
3043
3044 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
3045                           bpf_program_prep_t prep)
3046 {
3047         int *instances_fds;
3048
3049         if (nr_instances <= 0 || !prep)
3050                 return -EINVAL;
3051
3052         if (prog->instances.nr > 0 || prog->instances.fds) {
3053                 pr_warning("Can't set pre-processor after loading\n");
3054                 return -EINVAL;
3055         }
3056
3057         instances_fds = malloc(sizeof(int) * nr_instances);
3058         if (!instances_fds) {
3059                 pr_warning("alloc memory failed for fds\n");
3060                 return -ENOMEM;
3061         }
3062
3063         /* fill all fd with -1 */
3064         memset(instances_fds, -1, sizeof(int) * nr_instances);
3065
3066         prog->instances.nr = nr_instances;
3067         prog->instances.fds = instances_fds;
3068         prog->preprocessor = prep;
3069         return 0;
3070 }
3071
3072 int bpf_program__nth_fd(struct bpf_program *prog, int n)
3073 {
3074         int fd;
3075
3076         if (!prog)
3077                 return -EINVAL;
3078
3079         if (n >= prog->instances.nr || n < 0) {
3080                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
3081                            n, prog->section_name, prog->instances.nr);
3082                 return -EINVAL;
3083         }
3084
3085         fd = prog->instances.fds[n];
3086         if (fd < 0) {
3087                 pr_warning("%dth instance of program '%s' is invalid\n",
3088                            n, prog->section_name);
3089                 return -ENOENT;
3090         }
3091
3092         return fd;
3093 }
3094
3095 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
3096 {
3097         prog->type = type;
3098 }
3099
3100 static bool bpf_program__is_type(struct bpf_program *prog,
3101                                  enum bpf_prog_type type)
3102 {
3103         return prog ? (prog->type == type) : false;
3104 }
3105
3106 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
3107 int bpf_program__set_##NAME(struct bpf_program *prog)   \
3108 {                                                       \
3109         if (!prog)                                      \
3110                 return -EINVAL;                         \
3111         bpf_program__set_type(prog, TYPE);              \
3112         return 0;                                       \
3113 }                                                       \
3114                                                         \
3115 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
3116 {                                                       \
3117         return bpf_program__is_type(prog, TYPE);        \
3118 }                                                       \
3119
3120 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
3121 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
3122 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
3123 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
3124 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
3125 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
3126 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
3127 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
3128
3129 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
3130                                            enum bpf_attach_type type)
3131 {
3132         prog->expected_attach_type = type;
3133 }
3134
3135 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
3136         { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
3137
3138 /* Programs that can NOT be attached. */
3139 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
3140
3141 /* Programs that can be attached. */
3142 #define BPF_APROG_SEC(string, ptype, atype) \
3143         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
3144
3145 /* Programs that must specify expected attach type at load time. */
3146 #define BPF_EAPROG_SEC(string, ptype, eatype) \
3147         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
3148
3149 /* Programs that can be attached but attach type can't be identified by section
3150  * name. Kept for backward compatibility.
3151  */
3152 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
3153
3154 static const struct {
3155         const char *sec;
3156         size_t len;
3157         enum bpf_prog_type prog_type;
3158         enum bpf_attach_type expected_attach_type;
3159         int is_attachable;
3160         enum bpf_attach_type attach_type;
3161 } section_names[] = {
3162         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
3163         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
3164         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
3165         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
3166         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
3167         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
3168         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
3169         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
3170         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
3171         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
3172         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
3173         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
3174         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
3175         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
3176                                                 BPF_CGROUP_INET_INGRESS),
3177         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
3178                                                 BPF_CGROUP_INET_EGRESS),
3179         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
3180         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
3181                                                 BPF_CGROUP_INET_SOCK_CREATE),
3182         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
3183                                                 BPF_CGROUP_INET4_POST_BIND),
3184         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
3185                                                 BPF_CGROUP_INET6_POST_BIND),
3186         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
3187                                                 BPF_CGROUP_DEVICE),
3188         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
3189                                                 BPF_CGROUP_SOCK_OPS),
3190         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
3191                                                 BPF_SK_SKB_STREAM_PARSER),
3192         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
3193                                                 BPF_SK_SKB_STREAM_VERDICT),
3194         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
3195         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
3196                                                 BPF_SK_MSG_VERDICT),
3197         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
3198                                                 BPF_LIRC_MODE2),
3199         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
3200                                                 BPF_FLOW_DISSECTOR),
3201         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3202                                                 BPF_CGROUP_INET4_BIND),
3203         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3204                                                 BPF_CGROUP_INET6_BIND),
3205         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3206                                                 BPF_CGROUP_INET4_CONNECT),
3207         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3208                                                 BPF_CGROUP_INET6_CONNECT),
3209         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3210                                                 BPF_CGROUP_UDP4_SENDMSG),
3211         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3212                                                 BPF_CGROUP_UDP6_SENDMSG),
3213         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3214                                                 BPF_CGROUP_UDP4_RECVMSG),
3215         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3216                                                 BPF_CGROUP_UDP6_RECVMSG),
3217         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
3218                                                 BPF_CGROUP_SYSCTL),
3219 };
3220
3221 #undef BPF_PROG_SEC_IMPL
3222 #undef BPF_PROG_SEC
3223 #undef BPF_APROG_SEC
3224 #undef BPF_EAPROG_SEC
3225 #undef BPF_APROG_COMPAT
3226
3227 #define MAX_TYPE_NAME_SIZE 32
3228
3229 static char *libbpf_get_type_names(bool attach_type)
3230 {
3231         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
3232         char *buf;
3233
3234         buf = malloc(len);
3235         if (!buf)
3236                 return NULL;
3237
3238         buf[0] = '\0';
3239         /* Forge string buf with all available names */
3240         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3241                 if (attach_type && !section_names[i].is_attachable)
3242                         continue;
3243
3244                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
3245                         free(buf);
3246                         return NULL;
3247                 }
3248                 strcat(buf, " ");
3249                 strcat(buf, section_names[i].sec);
3250         }
3251
3252         return buf;
3253 }
3254
3255 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
3256                              enum bpf_attach_type *expected_attach_type)
3257 {
3258         char *type_names;
3259         int i;
3260
3261         if (!name)
3262                 return -EINVAL;
3263
3264         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3265                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3266                         continue;
3267                 *prog_type = section_names[i].prog_type;
3268                 *expected_attach_type = section_names[i].expected_attach_type;
3269                 return 0;
3270         }
3271         pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
3272         type_names = libbpf_get_type_names(false);
3273         if (type_names != NULL) {
3274                 pr_info("supported section(type) names are:%s\n", type_names);
3275                 free(type_names);
3276         }
3277
3278         return -EINVAL;
3279 }
3280
3281 int libbpf_attach_type_by_name(const char *name,
3282                                enum bpf_attach_type *attach_type)
3283 {
3284         char *type_names;
3285         int i;
3286
3287         if (!name)
3288                 return -EINVAL;
3289
3290         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3291                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3292                         continue;
3293                 if (!section_names[i].is_attachable)
3294                         return -EINVAL;
3295                 *attach_type = section_names[i].attach_type;
3296                 return 0;
3297         }
3298         pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
3299         type_names = libbpf_get_type_names(true);
3300         if (type_names != NULL) {
3301                 pr_info("attachable section(type) names are:%s\n", type_names);
3302                 free(type_names);
3303         }
3304
3305         return -EINVAL;
3306 }
3307
3308 static int
3309 bpf_program__identify_section(struct bpf_program *prog,
3310                               enum bpf_prog_type *prog_type,
3311                               enum bpf_attach_type *expected_attach_type)
3312 {
3313         return libbpf_prog_type_by_name(prog->section_name, prog_type,
3314                                         expected_attach_type);
3315 }
3316
3317 int bpf_map__fd(struct bpf_map *map)
3318 {
3319         return map ? map->fd : -EINVAL;
3320 }
3321
3322 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
3323 {
3324         return map ? &map->def : ERR_PTR(-EINVAL);
3325 }
3326
3327 const char *bpf_map__name(struct bpf_map *map)
3328 {
3329         return map ? map->name : NULL;
3330 }
3331
3332 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
3333 {
3334         return map ? map->btf_key_type_id : 0;
3335 }
3336
3337 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
3338 {
3339         return map ? map->btf_value_type_id : 0;
3340 }
3341
3342 int bpf_map__set_priv(struct bpf_map *map, void *priv,
3343                      bpf_map_clear_priv_t clear_priv)
3344 {
3345         if (!map)
3346                 return -EINVAL;
3347
3348         if (map->priv) {
3349                 if (map->clear_priv)
3350                         map->clear_priv(map, map->priv);
3351         }
3352
3353         map->priv = priv;
3354         map->clear_priv = clear_priv;
3355         return 0;
3356 }
3357
3358 void *bpf_map__priv(struct bpf_map *map)
3359 {
3360         return map ? map->priv : ERR_PTR(-EINVAL);
3361 }
3362
3363 bool bpf_map__is_offload_neutral(struct bpf_map *map)
3364 {
3365         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
3366 }
3367
3368 bool bpf_map__is_internal(struct bpf_map *map)
3369 {
3370         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
3371 }
3372
3373 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
3374 {
3375         map->map_ifindex = ifindex;
3376 }
3377
3378 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
3379 {
3380         if (!bpf_map_type__is_map_in_map(map->def.type)) {
3381                 pr_warning("error: unsupported map type\n");
3382                 return -EINVAL;
3383         }
3384         if (map->inner_map_fd != -1) {
3385                 pr_warning("error: inner_map_fd already specified\n");
3386                 return -EINVAL;
3387         }
3388         map->inner_map_fd = fd;
3389         return 0;
3390 }
3391
3392 static struct bpf_map *
3393 __bpf_map__iter(struct bpf_map *m, struct bpf_object *obj, int i)
3394 {
3395         ssize_t idx;
3396         struct bpf_map *s, *e;
3397
3398         if (!obj || !obj->maps)
3399                 return NULL;
3400
3401         s = obj->maps;
3402         e = obj->maps + obj->nr_maps;
3403
3404         if ((m < s) || (m >= e)) {
3405                 pr_warning("error in %s: map handler doesn't belong to object\n",
3406                            __func__);
3407                 return NULL;
3408         }
3409
3410         idx = (m - obj->maps) + i;
3411         if (idx >= obj->nr_maps || idx < 0)
3412                 return NULL;
3413         return &obj->maps[idx];
3414 }
3415
3416 struct bpf_map *
3417 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
3418 {
3419         if (prev == NULL)
3420                 return obj->maps;
3421
3422         return __bpf_map__iter(prev, obj, 1);
3423 }
3424
3425 struct bpf_map *
3426 bpf_map__prev(struct bpf_map *next, struct bpf_object *obj)
3427 {
3428         if (next == NULL) {
3429                 if (!obj->nr_maps)
3430                         return NULL;
3431                 return obj->maps + obj->nr_maps - 1;
3432         }
3433
3434         return __bpf_map__iter(next, obj, -1);
3435 }
3436
3437 struct bpf_map *
3438 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
3439 {
3440         struct bpf_map *pos;
3441
3442         bpf_object__for_each_map(pos, obj) {
3443                 if (pos->name && !strcmp(pos->name, name))
3444                         return pos;
3445         }
3446         return NULL;
3447 }
3448
3449 int
3450 bpf_object__find_map_fd_by_name(struct bpf_object *obj, const char *name)
3451 {
3452         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
3453 }
3454
3455 struct bpf_map *
3456 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
3457 {
3458         int i;
3459
3460         for (i = 0; i < obj->nr_maps; i++) {
3461                 if (obj->maps[i].offset == offset)
3462                         return &obj->maps[i];
3463         }
3464         return ERR_PTR(-ENOENT);
3465 }
3466
3467 long libbpf_get_error(const void *ptr)
3468 {
3469         if (IS_ERR(ptr))
3470                 return PTR_ERR(ptr);
3471         return 0;
3472 }
3473
3474 int bpf_prog_load(const char *file, enum bpf_prog_type type,
3475                   struct bpf_object **pobj, int *prog_fd)
3476 {
3477         struct bpf_prog_load_attr attr;
3478
3479         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
3480         attr.file = file;
3481         attr.prog_type = type;
3482         attr.expected_attach_type = 0;
3483
3484         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
3485 }
3486
3487 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
3488                         struct bpf_object **pobj, int *prog_fd)
3489 {
3490         struct bpf_object_open_attr open_attr = {
3491                 .file           = attr->file,
3492                 .prog_type      = attr->prog_type,
3493         };
3494         struct bpf_program *prog, *first_prog = NULL;
3495         enum bpf_attach_type expected_attach_type;
3496         enum bpf_prog_type prog_type;
3497         struct bpf_object *obj;
3498         struct bpf_map *map;
3499         int err;
3500
3501         if (!attr)
3502                 return -EINVAL;
3503         if (!attr->file)
3504                 return -EINVAL;
3505
3506         obj = bpf_object__open_xattr(&open_attr);
3507         if (IS_ERR_OR_NULL(obj))
3508                 return -ENOENT;
3509
3510         bpf_object__for_each_program(prog, obj) {
3511                 /*
3512                  * If type is not specified, try to guess it based on
3513                  * section name.
3514                  */
3515                 prog_type = attr->prog_type;
3516                 prog->prog_ifindex = attr->ifindex;
3517                 expected_attach_type = attr->expected_attach_type;
3518                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
3519                         err = bpf_program__identify_section(prog, &prog_type,
3520                                                             &expected_attach_type);
3521                         if (err < 0) {
3522                                 bpf_object__close(obj);
3523                                 return -EINVAL;
3524                         }
3525                 }
3526
3527                 bpf_program__set_type(prog, prog_type);
3528                 bpf_program__set_expected_attach_type(prog,
3529                                                       expected_attach_type);
3530
3531                 prog->log_level = attr->log_level;
3532                 if (!first_prog)
3533                         first_prog = prog;
3534         }
3535
3536         bpf_object__for_each_map(map, obj) {
3537                 if (!bpf_map__is_offload_neutral(map))
3538                         map->map_ifindex = attr->ifindex;
3539         }
3540
3541         if (!first_prog) {
3542                 pr_warning("object file doesn't contain bpf program\n");
3543                 bpf_object__close(obj);
3544                 return -ENOENT;
3545         }
3546
3547         err = bpf_object__load(obj);
3548         if (err) {
3549                 bpf_object__close(obj);
3550                 return -EINVAL;
3551         }
3552
3553         *pobj = obj;
3554         *prog_fd = bpf_program__fd(first_prog);
3555         return 0;
3556 }
3557
3558 enum bpf_perf_event_ret
3559 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
3560                            void **copy_mem, size_t *copy_size,
3561                            bpf_perf_event_print_t fn, void *private_data)
3562 {
3563         struct perf_event_mmap_page *header = mmap_mem;
3564         __u64 data_head = ring_buffer_read_head(header);
3565         __u64 data_tail = header->data_tail;
3566         void *base = ((__u8 *)header) + page_size;
3567         int ret = LIBBPF_PERF_EVENT_CONT;
3568         struct perf_event_header *ehdr;
3569         size_t ehdr_size;
3570
3571         while (data_head != data_tail) {
3572                 ehdr = base + (data_tail & (mmap_size - 1));
3573                 ehdr_size = ehdr->size;
3574
3575                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
3576                         void *copy_start = ehdr;
3577                         size_t len_first = base + mmap_size - copy_start;
3578                         size_t len_secnd = ehdr_size - len_first;
3579
3580                         if (*copy_size < ehdr_size) {
3581                                 free(*copy_mem);
3582                                 *copy_mem = malloc(ehdr_size);
3583                                 if (!*copy_mem) {
3584                                         *copy_size = 0;
3585                                         ret = LIBBPF_PERF_EVENT_ERROR;
3586                                         break;
3587                                 }
3588                                 *copy_size = ehdr_size;
3589                         }
3590
3591                         memcpy(*copy_mem, copy_start, len_first);
3592                         memcpy(*copy_mem + len_first, base, len_secnd);
3593                         ehdr = *copy_mem;
3594                 }
3595
3596                 ret = fn(ehdr, private_data);
3597                 data_tail += ehdr_size;
3598                 if (ret != LIBBPF_PERF_EVENT_CONT)
3599                         break;
3600         }
3601
3602         ring_buffer_write_tail(header, data_tail);
3603         return ret;
3604 }
3605
3606 struct bpf_prog_info_array_desc {
3607         int     array_offset;   /* e.g. offset of jited_prog_insns */
3608         int     count_offset;   /* e.g. offset of jited_prog_len */
3609         int     size_offset;    /* > 0: offset of rec size,
3610                                  * < 0: fix size of -size_offset
3611                                  */
3612 };
3613
3614 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
3615         [BPF_PROG_INFO_JITED_INSNS] = {
3616                 offsetof(struct bpf_prog_info, jited_prog_insns),
3617                 offsetof(struct bpf_prog_info, jited_prog_len),
3618                 -1,
3619         },
3620         [BPF_PROG_INFO_XLATED_INSNS] = {
3621                 offsetof(struct bpf_prog_info, xlated_prog_insns),
3622                 offsetof(struct bpf_prog_info, xlated_prog_len),
3623                 -1,
3624         },
3625         [BPF_PROG_INFO_MAP_IDS] = {
3626                 offsetof(struct bpf_prog_info, map_ids),
3627                 offsetof(struct bpf_prog_info, nr_map_ids),
3628                 -(int)sizeof(__u32),
3629         },
3630         [BPF_PROG_INFO_JITED_KSYMS] = {
3631                 offsetof(struct bpf_prog_info, jited_ksyms),
3632                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
3633                 -(int)sizeof(__u64),
3634         },
3635         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
3636                 offsetof(struct bpf_prog_info, jited_func_lens),
3637                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
3638                 -(int)sizeof(__u32),
3639         },
3640         [BPF_PROG_INFO_FUNC_INFO] = {
3641                 offsetof(struct bpf_prog_info, func_info),
3642                 offsetof(struct bpf_prog_info, nr_func_info),
3643                 offsetof(struct bpf_prog_info, func_info_rec_size),
3644         },
3645         [BPF_PROG_INFO_LINE_INFO] = {
3646                 offsetof(struct bpf_prog_info, line_info),
3647                 offsetof(struct bpf_prog_info, nr_line_info),
3648                 offsetof(struct bpf_prog_info, line_info_rec_size),
3649         },
3650         [BPF_PROG_INFO_JITED_LINE_INFO] = {
3651                 offsetof(struct bpf_prog_info, jited_line_info),
3652                 offsetof(struct bpf_prog_info, nr_jited_line_info),
3653                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
3654         },
3655         [BPF_PROG_INFO_PROG_TAGS] = {
3656                 offsetof(struct bpf_prog_info, prog_tags),
3657                 offsetof(struct bpf_prog_info, nr_prog_tags),
3658                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
3659         },
3660
3661 };
3662
3663 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
3664 {
3665         __u32 *array = (__u32 *)info;
3666
3667         if (offset >= 0)
3668                 return array[offset / sizeof(__u32)];
3669         return -(int)offset;
3670 }
3671
3672 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
3673 {
3674         __u64 *array = (__u64 *)info;
3675
3676         if (offset >= 0)
3677                 return array[offset / sizeof(__u64)];
3678         return -(int)offset;
3679 }
3680
3681 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
3682                                          __u32 val)
3683 {
3684         __u32 *array = (__u32 *)info;
3685
3686         if (offset >= 0)
3687                 array[offset / sizeof(__u32)] = val;
3688 }
3689
3690 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
3691                                          __u64 val)
3692 {
3693         __u64 *array = (__u64 *)info;
3694
3695         if (offset >= 0)
3696                 array[offset / sizeof(__u64)] = val;
3697 }
3698
3699 struct bpf_prog_info_linear *
3700 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
3701 {
3702         struct bpf_prog_info_linear *info_linear;
3703         struct bpf_prog_info info = {};
3704         __u32 info_len = sizeof(info);
3705         __u32 data_len = 0;
3706         int i, err;
3707         void *ptr;
3708
3709         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
3710                 return ERR_PTR(-EINVAL);
3711
3712         /* step 1: get array dimensions */
3713         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
3714         if (err) {
3715                 pr_debug("can't get prog info: %s", strerror(errno));
3716                 return ERR_PTR(-EFAULT);
3717         }
3718
3719         /* step 2: calculate total size of all arrays */
3720         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
3721                 bool include_array = (arrays & (1UL << i)) > 0;
3722                 struct bpf_prog_info_array_desc *desc;
3723                 __u32 count, size;
3724
3725                 desc = bpf_prog_info_array_desc + i;
3726
3727                 /* kernel is too old to support this field */
3728                 if (info_len < desc->array_offset + sizeof(__u32) ||
3729                     info_len < desc->count_offset + sizeof(__u32) ||
3730                     (desc->size_offset > 0 && info_len < desc->size_offset))
3731                         include_array = false;
3732
3733                 if (!include_array) {
3734                         arrays &= ~(1UL << i);  /* clear the bit */
3735                         continue;
3736                 }
3737
3738                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
3739                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
3740
3741                 data_len += count * size;
3742         }
3743
3744         /* step 3: allocate continuous memory */
3745         data_len = roundup(data_len, sizeof(__u64));
3746         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
3747         if (!info_linear)
3748                 return ERR_PTR(-ENOMEM);
3749
3750         /* step 4: fill data to info_linear->info */
3751         info_linear->arrays = arrays;
3752         memset(&info_linear->info, 0, sizeof(info));
3753         ptr = info_linear->data;
3754
3755         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
3756                 struct bpf_prog_info_array_desc *desc;
3757                 __u32 count, size;
3758
3759                 if ((arrays & (1UL << i)) == 0)
3760                         continue;
3761
3762                 desc  = bpf_prog_info_array_desc + i;
3763                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
3764                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
3765                 bpf_prog_info_set_offset_u32(&info_linear->info,
3766                                              desc->count_offset, count);
3767                 bpf_prog_info_set_offset_u32(&info_linear->info,
3768                                              desc->size_offset, size);
3769                 bpf_prog_info_set_offset_u64(&info_linear->info,
3770                                              desc->array_offset,
3771                                              ptr_to_u64(ptr));
3772                 ptr += count * size;
3773         }
3774
3775         /* step 5: call syscall again to get required arrays */
3776         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
3777         if (err) {
3778                 pr_debug("can't get prog info: %s", strerror(errno));
3779                 free(info_linear);
3780                 return ERR_PTR(-EFAULT);
3781         }
3782
3783         /* step 6: verify the data */
3784         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
3785                 struct bpf_prog_info_array_desc *desc;
3786                 __u32 v1, v2;
3787
3788                 if ((arrays & (1UL << i)) == 0)
3789                         continue;
3790
3791                 desc = bpf_prog_info_array_desc + i;
3792                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
3793                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
3794                                                    desc->count_offset);
3795                 if (v1 != v2)
3796                         pr_warning("%s: mismatch in element count\n", __func__);
3797
3798                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
3799                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
3800                                                    desc->size_offset);
3801                 if (v1 != v2)
3802                         pr_warning("%s: mismatch in rec size\n", __func__);
3803         }
3804
3805         /* step 7: update info_len and data_len */
3806         info_linear->info_len = sizeof(struct bpf_prog_info);
3807         info_linear->data_len = data_len;
3808
3809         return info_linear;
3810 }
3811
3812 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
3813 {
3814         int i;
3815
3816         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
3817                 struct bpf_prog_info_array_desc *desc;
3818                 __u64 addr, offs;
3819
3820                 if ((info_linear->arrays & (1UL << i)) == 0)
3821                         continue;
3822
3823                 desc = bpf_prog_info_array_desc + i;
3824                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
3825                                                      desc->array_offset);
3826                 offs = addr - ptr_to_u64(info_linear->data);
3827                 bpf_prog_info_set_offset_u64(&info_linear->info,
3828                                              desc->array_offset, offs);
3829         }
3830 }
3831
3832 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
3833 {
3834         int i;
3835
3836         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
3837                 struct bpf_prog_info_array_desc *desc;
3838                 __u64 addr, offs;
3839
3840                 if ((info_linear->arrays & (1UL << i)) == 0)
3841                         continue;
3842
3843                 desc = bpf_prog_info_array_desc + i;
3844                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
3845                                                      desc->array_offset);
3846                 addr = offs + ptr_to_u64(info_linear->data);
3847                 bpf_prog_info_set_offset_u64(&info_linear->info,
3848                                              desc->array_offset, addr);
3849         }
3850 }