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