tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2
   3 /*
   4  * Common eBPF ELF object loading operations.
   5  *
   6  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
   7  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
   8  * Copyright (C) 2015 Huawei Inc.
   9  * Copyright (C) 2017 Nicira, Inc.
  10  * Copyright (C) 2019 Isovalent, Inc.
  11  */
  12
  13 #ifndef _GNU_SOURCE
  14 #define _GNU_SOURCE
  15 #endif
  16 #include <stdlib.h>
  17 #include <stdio.h>
  18 #include <stdarg.h>
  19 #include <libgen.h>
  20 #include <inttypes.h>
  21 #include <string.h>
  22 #include <unistd.h>
  23 #include <endian.h>
  24 #include <fcntl.h>
  25 #include <errno.h>
  26 #include <asm/unistd.h>
  27 #include <linux/err.h>
  28 #include <linux/kernel.h>
  29 #include <linux/bpf.h>
  30 #include <linux/btf.h>
  31 #include <linux/filter.h>
  32 #include <linux/list.h>
  33 #include <linux/limits.h>
  34 #include <linux/perf_event.h>
  35 #include <linux/ring_buffer.h>
  36 #include <linux/version.h>
  37 #include <sys/epoll.h>
  38 #include <sys/ioctl.h>
  39 #include <sys/mman.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/vfs.h>
  43 #include <sys/utsname.h>
  44 #include <tools/libc_compat.h>
  45 #include <libelf.h>
  46 #include <gelf.h>
  47
  48 #include "libbpf.h"
  49 #include "bpf.h"
  50 #include "btf.h"
  51 #include "str_error.h"
  52 #include "libbpf_internal.h"
  53 #include "hashmap.h"
  54
  55 #ifndef EM_BPF
  56 #define EM_BPF 247
  57 #endif
  58
  59 #ifndef BPF_FS_MAGIC
  60 #define BPF_FS_MAGIC            0xcafe4a11
  61 #endif
  62
  63 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  64  * compilation if user enables corresponding warning. Disable it explicitly.
  65  */
  66 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  67
  68 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  69
  70 static int __base_pr(enum libbpf_print_level level, const char *format,
  71                      va_list args)
  72 {
  73         if (level == LIBBPF_DEBUG)
  74                 return 0;
  75
  76         return vfprintf(stderr, format, args);
  77 }
  78
  79 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  80
  81 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  82 {
  83         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  84
  85         __libbpf_pr = fn;
  86         return old_print_fn;
  87 }
  88
  89 __printf(2, 3)
  90 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  91 {
  92         va_list args;
  93
  94         if (!__libbpf_pr)
  95                 return;
  96
  97         va_start(args, format);
  98         __libbpf_pr(level, format, args);
  99         va_end(args);
 100 }
 101
 102 #define STRERR_BUFSIZE  128
 103
 104 #define CHECK_ERR(action, err, out) do {        \
 105         err = action;                   \
 106         if (err)                        \
 107                 goto out;               \
 108 } while(0)
 109
 110
 111 /* Copied from tools/perf/util/util.h */
 112 #ifndef zfree
 113 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 114 #endif
 115
 116 #ifndef zclose
 117 # define zclose(fd) ({                  \
 118         int ___err = 0;                 \
 119         if ((fd) >= 0)                  \
 120                 ___err = close((fd));   \
 121         fd = -1;                        \
 122         ___err; })
 123 #endif
 124
 125 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 126 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 127 #else
 128 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 129 #endif
 130
 131 static inline __u64 ptr_to_u64(const void *ptr)
 132 {
 133         return (__u64) (unsigned long) ptr;
 134 }
 135
 136 struct bpf_capabilities {
 137         /* v4.14: kernel support for program & map names. */
 138         __u32 name:1;
 139         /* v5.2: kernel support for global data sections. */
 140         __u32 global_data:1;
 141         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 142         __u32 btf_func:1;
 143         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 144         __u32 btf_datasec:1;
 145 };
 146
 147 /*
 148  * bpf_prog should be a better name but it has been used in
 149  * linux/filter.h.
 150  */
 151 struct bpf_program {
 152         /* Index in elf obj file, for relocation use. */
 153         int idx;
 154         char *name;
 155         int prog_ifindex;
 156         char *section_name;
 157         /* section_name with / replaced by _; makes recursive pinning
 158          * in bpf_object__pin_programs easier
 159          */
 160         char *pin_name;
 161         struct bpf_insn *insns;
 162         size_t insns_cnt, main_prog_cnt;
 163         enum bpf_prog_type type;
 164
 165         struct reloc_desc {
 166                 enum {
 167                         RELO_LD64,
 168                         RELO_CALL,
 169                         RELO_DATA,
 170                 } type;
 171                 int insn_idx;
 172                 union {
 173                         int map_idx;
 174                         int text_off;
 175                 };
 176         } *reloc_desc;
 177         int nr_reloc;
 178         int log_level;
 179
 180         struct {
 181                 int nr;
 182                 int *fds;
 183         } instances;
 184         bpf_program_prep_t preprocessor;
 185
 186         struct bpf_object *obj;
 187         void *priv;
 188         bpf_program_clear_priv_t clear_priv;
 189
 190         enum bpf_attach_type expected_attach_type;
 191         void *func_info;
 192         __u32 func_info_rec_size;
 193         __u32 func_info_cnt;
 194
 195         struct bpf_capabilities *caps;
 196
 197         void *line_info;
 198         __u32 line_info_rec_size;
 199         __u32 line_info_cnt;
 200         __u32 prog_flags;
 201 };
 202
 203 enum libbpf_map_type {
 204         LIBBPF_MAP_UNSPEC,
 205         LIBBPF_MAP_DATA,
 206         LIBBPF_MAP_BSS,
 207         LIBBPF_MAP_RODATA,
 208 };
 209
 210 static const char * const libbpf_type_to_btf_name[] = {
 211         [LIBBPF_MAP_DATA]       = ".data",
 212         [LIBBPF_MAP_BSS]        = ".bss",
 213         [LIBBPF_MAP_RODATA]     = ".rodata",
 214 };
 215
 216 struct bpf_map {
 217         int fd;
 218         char *name;
 219         int sec_idx;
 220         size_t sec_offset;
 221         int map_ifindex;
 222         int inner_map_fd;
 223         struct bpf_map_def def;
 224         __u32 btf_key_type_id;
 225         __u32 btf_value_type_id;
 226         void *priv;
 227         bpf_map_clear_priv_t clear_priv;
 228         enum libbpf_map_type libbpf_type;
 229 };
 230
 231 struct bpf_secdata {
 232         void *rodata;
 233         void *data;
 234 };
 235
 236 static LIST_HEAD(bpf_objects_list);
 237
 238 struct bpf_object {
 239         char name[BPF_OBJ_NAME_LEN];
 240         char license[64];
 241         __u32 kern_version;
 242
 243         struct bpf_program *programs;
 244         size_t nr_programs;
 245         struct bpf_map *maps;
 246         size_t nr_maps;
 247         size_t maps_cap;
 248         struct bpf_secdata sections;
 249
 250         bool loaded;
 251         bool has_pseudo_calls;
 252         bool relaxed_core_relocs;
 253
 254         /*
 255          * Information when doing elf related work. Only valid if fd
 256          * is valid.
 257          */
 258         struct {
 259                 int fd;
 260                 const void *obj_buf;
 261                 size_t obj_buf_sz;
 262                 Elf *elf;
 263                 GElf_Ehdr ehdr;
 264                 Elf_Data *symbols;
 265                 Elf_Data *data;
 266                 Elf_Data *rodata;
 267                 Elf_Data *bss;
 268                 size_t strtabidx;
 269                 struct {
 270                         GElf_Shdr shdr;
 271                         Elf_Data *data;
 272                 } *reloc;
 273                 int nr_reloc;
 274                 int maps_shndx;
 275                 int btf_maps_shndx;
 276                 int text_shndx;
 277                 int data_shndx;
 278                 int rodata_shndx;
 279                 int bss_shndx;
 280         } efile;
 281         /*
 282          * All loaded bpf_object is linked in a list, which is
 283          * hidden to caller. bpf_objects__<func> handlers deal with
 284          * all objects.
 285          */
 286         struct list_head list;
 287
 288         struct btf *btf;
 289         struct btf_ext *btf_ext;
 290
 291         void *priv;
 292         bpf_object_clear_priv_t clear_priv;
 293
 294         struct bpf_capabilities caps;
 295
 296         char path[];
 297 };
 298 #define obj_elf_valid(o)        ((o)->efile.elf)
 299
 300 void bpf_program__unload(struct bpf_program *prog)
 301 {
 302         int i;
 303
 304         if (!prog)
 305                 return;
 306
 307         /*
 308          * If the object is opened but the program was never loaded,
 309          * it is possible that prog->instances.nr == -1.
 310          */
 311         if (prog->instances.nr > 0) {
 312                 for (i = 0; i < prog->instances.nr; i++)
 313                         zclose(prog->instances.fds[i]);
 314         } else if (prog->instances.nr != -1) {
 315                 pr_warning("Internal error: instances.nr is %d\n",
 316                            prog->instances.nr);
 317         }
 318
 319         prog->instances.nr = -1;
 320         zfree(&prog->instances.fds);
 321
 322         zfree(&prog->func_info);
 323         zfree(&prog->line_info);
 324 }
 325
 326 static void bpf_program__exit(struct bpf_program *prog)
 327 {
 328         if (!prog)
 329                 return;
 330
 331         if (prog->clear_priv)
 332                 prog->clear_priv(prog, prog->priv);
 333
 334         prog->priv = NULL;
 335         prog->clear_priv = NULL;
 336
 337         bpf_program__unload(prog);
 338         zfree(&prog->name);
 339         zfree(&prog->section_name);
 340         zfree(&prog->pin_name);
 341         zfree(&prog->insns);
 342         zfree(&prog->reloc_desc);
 343
 344         prog->nr_reloc = 0;
 345         prog->insns_cnt = 0;
 346         prog->idx = -1;
 347 }
 348
 349 static char *__bpf_program__pin_name(struct bpf_program *prog)
 350 {
 351         char *name, *p;
 352
 353         name = p = strdup(prog->section_name);
 354         while ((p = strchr(p, '/')))
 355                 *p = '_';
 356
 357         return name;
 358 }
 359
 360 static int
 361 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 362                   struct bpf_program *prog)
 363 {
 364         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 365
 366         if (size == 0 || size % bpf_insn_sz) {
 367                 pr_warning("corrupted section '%s', size: %zu\n",
 368                            section_name, size);
 369                 return -EINVAL;
 370         }
 371
 372         memset(prog, 0, sizeof(*prog));
 373
 374         prog->section_name = strdup(section_name);
 375         if (!prog->section_name) {
 376                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 377                            idx, section_name);
 378                 goto errout;
 379         }
 380
 381         prog->pin_name = __bpf_program__pin_name(prog);
 382         if (!prog->pin_name) {
 383                 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
 384                            idx, section_name);
 385                 goto errout;
 386         }
 387
 388         prog->insns = malloc(size);
 389         if (!prog->insns) {
 390                 pr_warning("failed to alloc insns for prog under section %s\n",
 391                            section_name);
 392                 goto errout;
 393         }
 394         prog->insns_cnt = size / bpf_insn_sz;
 395         memcpy(prog->insns, data, size);
 396         prog->idx = idx;
 397         prog->instances.fds = NULL;
 398         prog->instances.nr = -1;
 399         prog->type = BPF_PROG_TYPE_UNSPEC;
 400
 401         return 0;
 402 errout:
 403         bpf_program__exit(prog);
 404         return -ENOMEM;
 405 }
 406
 407 static int
 408 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 409                         char *section_name, int idx)
 410 {
 411         struct bpf_program prog, *progs;
 412         int nr_progs, err;
 413
 414         err = bpf_program__init(data, size, section_name, idx, &prog);
 415         if (err)
 416                 return err;
 417
 418         prog.caps = &obj->caps;
 419         progs = obj->programs;
 420         nr_progs = obj->nr_programs;
 421
 422         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 423         if (!progs) {
 424                 /*
 425                  * In this case the original obj->programs
 426                  * is still valid, so don't need special treat for
 427                  * bpf_close_object().
 428                  */
 429                 pr_warning("failed to alloc a new program under section '%s'\n",
 430                            section_name);
 431                 bpf_program__exit(&prog);
 432                 return -ENOMEM;
 433         }
 434
 435         pr_debug("found program %s\n", prog.section_name);
 436         obj->programs = progs;
 437         obj->nr_programs = nr_progs + 1;
 438         prog.obj = obj;
 439         progs[nr_progs] = prog;
 440         return 0;
 441 }
 442
 443 static int
 444 bpf_object__init_prog_names(struct bpf_object *obj)
 445 {
 446         Elf_Data *symbols = obj->efile.symbols;
 447         struct bpf_program *prog;
 448         size_t pi, si;
 449
 450         for (pi = 0; pi < obj->nr_programs; pi++) {
 451                 const char *name = NULL;
 452
 453                 prog = &obj->programs[pi];
 454
 455                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 456                      si++) {
 457                         GElf_Sym sym;
 458
 459                         if (!gelf_getsym(symbols, si, &sym))
 460                                 continue;
 461                         if (sym.st_shndx != prog->idx)
 462                                 continue;
 463                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 464                                 continue;
 465
 466                         name = elf_strptr(obj->efile.elf,
 467                                           obj->efile.strtabidx,
 468                                           sym.st_name);
 469                         if (!name) {
 470                                 pr_warning("failed to get sym name string for prog %s\n",
 471                                            prog->section_name);
 472                                 return -LIBBPF_ERRNO__LIBELF;
 473                         }
 474                 }
 475
 476                 if (!name && prog->idx == obj->efile.text_shndx)
 477                         name = ".text";
 478
 479                 if (!name) {
 480                         pr_warning("failed to find sym for prog %s\n",
 481                                    prog->section_name);
 482                         return -EINVAL;
 483                 }
 484
 485                 prog->name = strdup(name);
 486                 if (!prog->name) {
 487                         pr_warning("failed to allocate memory for prog sym %s\n",
 488                                    name);
 489                         return -ENOMEM;
 490                 }
 491         }
 492
 493         return 0;
 494 }
 495
 496 static __u32 get_kernel_version(void)
 497 {
 498         __u32 major, minor, patch;
 499         struct utsname info;
 500
 501         uname(&info);
 502         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 503                 return 0;
 504         return KERNEL_VERSION(major, minor, patch);
 505 }
 506
 507 static struct bpf_object *bpf_object__new(const char *path,
 508                                           const void *obj_buf,
 509                                           size_t obj_buf_sz,
 510                                           const char *obj_name)
 511 {
 512         struct bpf_object *obj;
 513         char *end;
 514
 515         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 516         if (!obj) {
 517                 pr_warning("alloc memory failed for %s\n", path);
 518                 return ERR_PTR(-ENOMEM);
 519         }
 520
 521         strcpy(obj->path, path);
 522         if (obj_name) {
 523                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
 524                 obj->name[sizeof(obj->name) - 1] = 0;
 525         } else {
 526                 /* Using basename() GNU version which doesn't modify arg. */
 527                 strncpy(obj->name, basename((void *)path),
 528                         sizeof(obj->name) - 1);
 529                 end = strchr(obj->name, '.');
 530                 if (end)
 531                         *end = 0;
 532         }
 533
 534         obj->efile.fd = -1;
 535         /*
 536          * Caller of this function should also call
 537          * bpf_object__elf_finish() after data collection to return
 538          * obj_buf to user. If not, we should duplicate the buffer to
 539          * avoid user freeing them before elf finish.
 540          */
 541         obj->efile.obj_buf = obj_buf;
 542         obj->efile.obj_buf_sz = obj_buf_sz;
 543         obj->efile.maps_shndx = -1;
 544         obj->efile.btf_maps_shndx = -1;
 545         obj->efile.data_shndx = -1;
 546         obj->efile.rodata_shndx = -1;
 547         obj->efile.bss_shndx = -1;
 548
 549         obj->kern_version = get_kernel_version();
 550         obj->loaded = false;
 551
 552         INIT_LIST_HEAD(&obj->list);
 553         list_add(&obj->list, &bpf_objects_list);
 554         return obj;
 555 }
 556
 557 static void bpf_object__elf_finish(struct bpf_object *obj)
 558 {
 559         if (!obj_elf_valid(obj))
 560                 return;
 561
 562         if (obj->efile.elf) {
 563                 elf_end(obj->efile.elf);
 564                 obj->efile.elf = NULL;
 565         }
 566         obj->efile.symbols = NULL;
 567         obj->efile.data = NULL;
 568         obj->efile.rodata = NULL;
 569         obj->efile.bss = NULL;
 570
 571         zfree(&obj->efile.reloc);
 572         obj->efile.nr_reloc = 0;
 573         zclose(obj->efile.fd);
 574         obj->efile.obj_buf = NULL;
 575         obj->efile.obj_buf_sz = 0;
 576 }
 577
 578 static int bpf_object__elf_init(struct bpf_object *obj)
 579 {
 580         int err = 0;
 581         GElf_Ehdr *ep;
 582
 583         if (obj_elf_valid(obj)) {
 584                 pr_warning("elf init: internal error\n");
 585                 return -LIBBPF_ERRNO__LIBELF;
 586         }
 587
 588         if (obj->efile.obj_buf_sz > 0) {
 589                 /*
 590                  * obj_buf should have been validated by
 591                  * bpf_object__open_buffer().
 592                  */
 593                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
 594                                             obj->efile.obj_buf_sz);
 595         } else {
 596                 obj->efile.fd = open(obj->path, O_RDONLY);
 597                 if (obj->efile.fd < 0) {
 598                         char errmsg[STRERR_BUFSIZE], *cp;
 599
 600                         err = -errno;
 601                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 602                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 603                         return err;
 604                 }
 605
 606                 obj->efile.elf = elf_begin(obj->efile.fd,
 607                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 608         }
 609
 610         if (!obj->efile.elf) {
 611                 pr_warning("failed to open %s as ELF file\n", obj->path);
 612                 err = -LIBBPF_ERRNO__LIBELF;
 613                 goto errout;
 614         }
 615
 616         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 617                 pr_warning("failed to get EHDR from %s\n", obj->path);
 618                 err = -LIBBPF_ERRNO__FORMAT;
 619                 goto errout;
 620         }
 621         ep = &obj->efile.ehdr;
 622
 623         /* Old LLVM set e_machine to EM_NONE */
 624         if (ep->e_type != ET_REL ||
 625             (ep->e_machine && ep->e_machine != EM_BPF)) {
 626                 pr_warning("%s is not an eBPF object file\n", obj->path);
 627                 err = -LIBBPF_ERRNO__FORMAT;
 628                 goto errout;
 629         }
 630
 631         return 0;
 632 errout:
 633         bpf_object__elf_finish(obj);
 634         return err;
 635 }
 636
 637 static int bpf_object__check_endianness(struct bpf_object *obj)
 638 {
 639 #if __BYTE_ORDER == __LITTLE_ENDIAN
 640         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 641                 return 0;
 642 #elif __BYTE_ORDER == __BIG_ENDIAN
 643         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 644                 return 0;
 645 #else
 646 # error "Unrecognized __BYTE_ORDER__"
 647 #endif
 648         pr_warning("endianness mismatch.\n");
 649         return -LIBBPF_ERRNO__ENDIAN;
 650 }
 651
 652 static int
 653 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 654 {
 655         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 656         pr_debug("license of %s is %s\n", obj->path, obj->license);
 657         return 0;
 658 }
 659
 660 static int
 661 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 662 {
 663         __u32 kver;
 664
 665         if (size != sizeof(kver)) {
 666                 pr_warning("invalid kver section in %s\n", obj->path);
 667                 return -LIBBPF_ERRNO__FORMAT;
 668         }
 669         memcpy(&kver, data, sizeof(kver));
 670         obj->kern_version = kver;
 671         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 672         return 0;
 673 }
 674
 675 static int compare_bpf_map(const void *_a, const void *_b)
 676 {
 677         const struct bpf_map *a = _a;
 678         const struct bpf_map *b = _b;
 679
 680         if (a->sec_idx != b->sec_idx)
 681                 return a->sec_idx - b->sec_idx;
 682         return a->sec_offset - b->sec_offset;
 683 }
 684
 685 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 686 {
 687         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 688             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 689                 return true;
 690         return false;
 691 }
 692
 693 static int bpf_object_search_section_size(const struct bpf_object *obj,
 694                                           const char *name, size_t *d_size)
 695 {
 696         const GElf_Ehdr *ep = &obj->efile.ehdr;
 697         Elf *elf = obj->efile.elf;
 698         Elf_Scn *scn = NULL;
 699         int idx = 0;
 700
 701         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 702                 const char *sec_name;
 703                 Elf_Data *data;
 704                 GElf_Shdr sh;
 705
 706                 idx++;
 707                 if (gelf_getshdr(scn, &sh) != &sh) {
 708                         pr_warning("failed to get section(%d) header from %s\n",
 709                                    idx, obj->path);
 710                         return -EIO;
 711                 }
 712
 713                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 714                 if (!sec_name) {
 715                         pr_warning("failed to get section(%d) name from %s\n",
 716                                    idx, obj->path);
 717                         return -EIO;
 718                 }
 719
 720                 if (strcmp(name, sec_name))
 721                         continue;
 722
 723                 data = elf_getdata(scn, 0);
 724                 if (!data) {
 725                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 726                                    idx, name, obj->path);
 727                         return -EIO;
 728                 }
 729
 730                 *d_size = data->d_size;
 731                 return 0;
 732         }
 733
 734         return -ENOENT;
 735 }
 736
 737 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 738                              __u32 *size)
 739 {
 740         int ret = -ENOENT;
 741         size_t d_size;
 742
 743         *size = 0;
 744         if (!name) {
 745                 return -EINVAL;
 746         } else if (!strcmp(name, ".data")) {
 747                 if (obj->efile.data)
 748                         *size = obj->efile.data->d_size;
 749         } else if (!strcmp(name, ".bss")) {
 750                 if (obj->efile.bss)
 751                         *size = obj->efile.bss->d_size;
 752         } else if (!strcmp(name, ".rodata")) {
 753                 if (obj->efile.rodata)
 754                         *size = obj->efile.rodata->d_size;
 755         } else {
 756                 ret = bpf_object_search_section_size(obj, name, &d_size);
 757                 if (!ret)
 758                         *size = d_size;
 759         }
 760
 761         return *size ? 0 : ret;
 762 }
 763
 764 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 765                                 __u32 *off)
 766 {
 767         Elf_Data *symbols = obj->efile.symbols;
 768         const char *sname;
 769         size_t si;
 770
 771         if (!name || !off)
 772                 return -EINVAL;
 773
 774         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 775                 GElf_Sym sym;
 776
 777                 if (!gelf_getsym(symbols, si, &sym))
 778                         continue;
 779                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 780                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 781                         continue;
 782
 783                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 784                                    sym.st_name);
 785                 if (!sname) {
 786                         pr_warning("failed to get sym name string for var %s\n",
 787                                    name);
 788                         return -EIO;
 789                 }
 790                 if (strcmp(name, sname) == 0) {
 791                         *off = sym.st_value;
 792                         return 0;
 793                 }
 794         }
 795
 796         return -ENOENT;
 797 }
 798
 799 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 800 {
 801         struct bpf_map *new_maps;
 802         size_t new_cap;
 803         int i;
 804
 805         if (obj->nr_maps < obj->maps_cap)
 806                 return &obj->maps[obj->nr_maps++];
 807
 808         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 809         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 810         if (!new_maps) {
 811                 pr_warning("alloc maps for object failed\n");
 812                 return ERR_PTR(-ENOMEM);
 813         }
 814
 815         obj->maps_cap = new_cap;
 816         obj->maps = new_maps;
 817
 818         /* zero out new maps */
 819         memset(obj->maps + obj->nr_maps, 0,
 820                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 821         /*
 822          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 823          * when failure (zclose won't close negative fd)).
 824          */
 825         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 826                 obj->maps[i].fd = -1;
 827                 obj->maps[i].inner_map_fd = -1;
 828         }
 829
 830         return &obj->maps[obj->nr_maps++];
 831 }
 832
 833 static int
 834 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 835                               int sec_idx, Elf_Data *data, void **data_buff)
 836 {
 837         char map_name[BPF_OBJ_NAME_LEN];
 838         struct bpf_map_def *def;
 839         struct bpf_map *map;
 840
 841         map = bpf_object__add_map(obj);
 842         if (IS_ERR(map))
 843                 return PTR_ERR(map);
 844
 845         map->libbpf_type = type;
 846         map->sec_idx = sec_idx;
 847         map->sec_offset = 0;
 848         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 849                  libbpf_type_to_btf_name[type]);
 850         map->name = strdup(map_name);
 851         if (!map->name) {
 852                 pr_warning("failed to alloc map name\n");
 853                 return -ENOMEM;
 854         }
 855         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
 856                  map_name, map->sec_idx, map->sec_offset);
 857
 858         def = &map->def;
 859         def->type = BPF_MAP_TYPE_ARRAY;
 860         def->key_size = sizeof(int);
 861         def->value_size = data->d_size;
 862         def->max_entries = 1;
 863         def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
 864         if (data_buff) {
 865                 *data_buff = malloc(data->d_size);
 866                 if (!*data_buff) {
 867                         zfree(&map->name);
 868                         pr_warning("failed to alloc map content buffer\n");
 869                         return -ENOMEM;
 870                 }
 871                 memcpy(*data_buff, data->d_buf, data->d_size);
 872         }
 873
 874         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 875         return 0;
 876 }
 877
 878 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 879 {
 880         int err;
 881
 882         if (!obj->caps.global_data)
 883                 return 0;
 884         /*
 885          * Populate obj->maps with libbpf internal maps.
 886          */
 887         if (obj->efile.data_shndx >= 0) {
 888                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 889                                                     obj->efile.data_shndx,
 890                                                     obj->efile.data,
 891                                                     &obj->sections.data);
 892                 if (err)
 893                         return err;
 894         }
 895         if (obj->efile.rodata_shndx >= 0) {
 896                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 897                                                     obj->efile.rodata_shndx,
 898                                                     obj->efile.rodata,
 899                                                     &obj->sections.rodata);
 900                 if (err)
 901                         return err;
 902         }
 903         if (obj->efile.bss_shndx >= 0) {
 904                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 905                                                     obj->efile.bss_shndx,
 906                                                     obj->efile.bss, NULL);
 907                 if (err)
 908                         return err;
 909         }
 910         return 0;
 911 }
 912
 913 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
 914 {
 915         Elf_Data *symbols = obj->efile.symbols;
 916         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
 917         Elf_Data *data = NULL;
 918         Elf_Scn *scn;
 919
 920         if (obj->efile.maps_shndx < 0)
 921                 return 0;
 922
 923         if (!symbols)
 924                 return -EINVAL;
 925
 926         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 927         if (scn)
 928                 data = elf_getdata(scn, NULL);
 929         if (!scn || !data) {
 930                 pr_warning("failed to get Elf_Data from map section %d\n",
 931                            obj->efile.maps_shndx);
 932                 return -EINVAL;
 933         }
 934
 935         /*
 936          * Count number of maps. Each map has a name.
 937          * Array of maps is not supported: only the first element is
 938          * considered.
 939          *
 940          * TODO: Detect array of map and report error.
 941          */
 942         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 943         for (i = 0; i < nr_syms; i++) {
 944                 GElf_Sym sym;
 945
 946                 if (!gelf_getsym(symbols, i, &sym))
 947                         continue;
 948                 if (sym.st_shndx != obj->efile.maps_shndx)
 949                         continue;
 950                 nr_maps++;
 951         }
 952         /* Assume equally sized map definitions */
 953         pr_debug("maps in %s: %d maps in %zd bytes\n",
 954                  obj->path, nr_maps, data->d_size);
 955
 956         map_def_sz = data->d_size / nr_maps;
 957         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 958                 pr_warning("unable to determine map definition size "
 959                            "section %s, %d maps in %zd bytes\n",
 960                            obj->path, nr_maps, data->d_size);
 961                 return -EINVAL;
 962         }
 963
 964         /* Fill obj->maps using data in "maps" section.  */
 965         for (i = 0; i < nr_syms; i++) {
 966                 GElf_Sym sym;
 967                 const char *map_name;
 968                 struct bpf_map_def *def;
 969                 struct bpf_map *map;
 970
 971                 if (!gelf_getsym(symbols, i, &sym))
 972                         continue;
 973                 if (sym.st_shndx != obj->efile.maps_shndx)
 974                         continue;
 975
 976                 map = bpf_object__add_map(obj);
 977                 if (IS_ERR(map))
 978                         return PTR_ERR(map);
 979
 980                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 981                                       sym.st_name);
 982                 if (!map_name) {
 983                         pr_warning("failed to get map #%d name sym string for obj %s\n",
 984                                    i, obj->path);
 985                         return -LIBBPF_ERRNO__FORMAT;
 986                 }
 987
 988                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
 989                 map->sec_idx = sym.st_shndx;
 990                 map->sec_offset = sym.st_value;
 991                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
 992                          map_name, map->sec_idx, map->sec_offset);
 993                 if (sym.st_value + map_def_sz > data->d_size) {
 994                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 995                                    obj->path, map_name);
 996                         return -EINVAL;
 997                 }
 998
 999                 map->name = strdup(map_name);
1000                 if (!map->name) {
1001                         pr_warning("failed to alloc map name\n");
1002                         return -ENOMEM;
1003                 }
1004                 pr_debug("map %d is \"%s\"\n", i, map->name);
1005                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1006                 /*
1007                  * If the definition of the map in the object file fits in
1008                  * bpf_map_def, copy it.  Any extra fields in our version
1009                  * of bpf_map_def will default to zero as a result of the
1010                  * calloc above.
1011                  */
1012                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1013                         memcpy(&map->def, def, map_def_sz);
1014                 } else {
1015                         /*
1016                          * Here the map structure being read is bigger than what
1017                          * we expect, truncate if the excess bits are all zero.
1018                          * If they are not zero, reject this map as
1019                          * incompatible.
1020                          */
1021                         char *b;
1022                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1023                              b < ((char *)def) + map_def_sz; b++) {
1024                                 if (*b != 0) {
1025                                         pr_warning("maps section in %s: \"%s\" "
1026                                                    "has unrecognized, non-zero "
1027                                                    "options\n",
1028                                                    obj->path, map_name);
1029                                         if (strict)
1030                                                 return -EINVAL;
1031                                 }
1032                         }
1033                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1034                 }
1035         }
1036         return 0;
1037 }
1038
1039 static const struct btf_type *
1040 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1041 {
1042         const struct btf_type *t = btf__type_by_id(btf, id);
1043
1044         if (res_id)
1045                 *res_id = id;
1046
1047         while (btf_is_mod(t) || btf_is_typedef(t)) {
1048                 if (res_id)
1049                         *res_id = t->type;
1050                 t = btf__type_by_id(btf, t->type);
1051         }
1052
1053         return t;
1054 }
1055
1056 /*
1057  * Fetch integer attribute of BTF map definition. Such attributes are
1058  * represented using a pointer to an array, in which dimensionality of array
1059  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1060  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1061  * type definition, while using only sizeof(void *) space in ELF data section.
1062  */
1063 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1064                               const struct btf_type *def,
1065                               const struct btf_member *m, __u32 *res) {
1066         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1067         const char *name = btf__name_by_offset(btf, m->name_off);
1068         const struct btf_array *arr_info;
1069         const struct btf_type *arr_t;
1070
1071         if (!btf_is_ptr(t)) {
1072                 pr_warning("map '%s': attr '%s': expected PTR, got %u.\n",
1073                            map_name, name, btf_kind(t));
1074                 return false;
1075         }
1076
1077         arr_t = btf__type_by_id(btf, t->type);
1078         if (!arr_t) {
1079                 pr_warning("map '%s': attr '%s': type [%u] not found.\n",
1080                            map_name, name, t->type);
1081                 return false;
1082         }
1083         if (!btf_is_array(arr_t)) {
1084                 pr_warning("map '%s': attr '%s': expected ARRAY, got %u.\n",
1085                            map_name, name, btf_kind(arr_t));
1086                 return false;
1087         }
1088         arr_info = btf_array(arr_t);
1089         *res = arr_info->nelems;
1090         return true;
1091 }
1092
1093 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1094                                          const struct btf_type *sec,
1095                                          int var_idx, int sec_idx,
1096                                          const Elf_Data *data, bool strict)
1097 {
1098         const struct btf_type *var, *def, *t;
1099         const struct btf_var_secinfo *vi;
1100         const struct btf_var *var_extra;
1101         const struct btf_member *m;
1102         const char *map_name;
1103         struct bpf_map *map;
1104         int vlen, i;
1105
1106         vi = btf_var_secinfos(sec) + var_idx;
1107         var = btf__type_by_id(obj->btf, vi->type);
1108         var_extra = btf_var(var);
1109         map_name = btf__name_by_offset(obj->btf, var->name_off);
1110         vlen = btf_vlen(var);
1111
1112         if (map_name == NULL || map_name[0] == '\0') {
1113                 pr_warning("map #%d: empty name.\n", var_idx);
1114                 return -EINVAL;
1115         }
1116         if ((__u64)vi->offset + vi->size > data->d_size) {
1117                 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1118                 return -EINVAL;
1119         }
1120         if (!btf_is_var(var)) {
1121                 pr_warning("map '%s': unexpected var kind %u.\n",
1122                            map_name, btf_kind(var));
1123                 return -EINVAL;
1124         }
1125         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1126             var_extra->linkage != BTF_VAR_STATIC) {
1127                 pr_warning("map '%s': unsupported var linkage %u.\n",
1128                            map_name, var_extra->linkage);
1129                 return -EOPNOTSUPP;
1130         }
1131
1132         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1133         if (!btf_is_struct(def)) {
1134                 pr_warning("map '%s': unexpected def kind %u.\n",
1135                            map_name, btf_kind(var));
1136                 return -EINVAL;
1137         }
1138         if (def->size > vi->size) {
1139                 pr_warning("map '%s': invalid def size.\n", map_name);
1140                 return -EINVAL;
1141         }
1142
1143         map = bpf_object__add_map(obj);
1144         if (IS_ERR(map))
1145                 return PTR_ERR(map);
1146         map->name = strdup(map_name);
1147         if (!map->name) {
1148                 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1149                 return -ENOMEM;
1150         }
1151         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1152         map->def.type = BPF_MAP_TYPE_UNSPEC;
1153         map->sec_idx = sec_idx;
1154         map->sec_offset = vi->offset;
1155         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1156                  map_name, map->sec_idx, map->sec_offset);
1157
1158         vlen = btf_vlen(def);
1159         m = btf_members(def);
1160         for (i = 0; i < vlen; i++, m++) {
1161                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1162
1163                 if (!name) {
1164                         pr_warning("map '%s': invalid field #%d.\n",
1165                                    map_name, i);
1166                         return -EINVAL;
1167                 }
1168                 if (strcmp(name, "type") == 0) {
1169                         if (!get_map_field_int(map_name, obj->btf, def, m,
1170                                                &map->def.type))
1171                                 return -EINVAL;
1172                         pr_debug("map '%s': found type = %u.\n",
1173                                  map_name, map->def.type);
1174                 } else if (strcmp(name, "max_entries") == 0) {
1175                         if (!get_map_field_int(map_name, obj->btf, def, m,
1176                                                &map->def.max_entries))
1177                                 return -EINVAL;
1178                         pr_debug("map '%s': found max_entries = %u.\n",
1179                                  map_name, map->def.max_entries);
1180                 } else if (strcmp(name, "map_flags") == 0) {
1181                         if (!get_map_field_int(map_name, obj->btf, def, m,
1182                                                &map->def.map_flags))
1183                                 return -EINVAL;
1184                         pr_debug("map '%s': found map_flags = %u.\n",
1185                                  map_name, map->def.map_flags);
1186                 } else if (strcmp(name, "key_size") == 0) {
1187                         __u32 sz;
1188
1189                         if (!get_map_field_int(map_name, obj->btf, def, m,
1190                                                &sz))
1191                                 return -EINVAL;
1192                         pr_debug("map '%s': found key_size = %u.\n",
1193                                  map_name, sz);
1194                         if (map->def.key_size && map->def.key_size != sz) {
1195                                 pr_warning("map '%s': conflicting key size %u != %u.\n",
1196                                            map_name, map->def.key_size, sz);
1197                                 return -EINVAL;
1198                         }
1199                         map->def.key_size = sz;
1200                 } else if (strcmp(name, "key") == 0) {
1201                         __s64 sz;
1202
1203                         t = btf__type_by_id(obj->btf, m->type);
1204                         if (!t) {
1205                                 pr_warning("map '%s': key type [%d] not found.\n",
1206                                            map_name, m->type);
1207                                 return -EINVAL;
1208                         }
1209                         if (!btf_is_ptr(t)) {
1210                                 pr_warning("map '%s': key spec is not PTR: %u.\n",
1211                                            map_name, btf_kind(t));
1212                                 return -EINVAL;
1213                         }
1214                         sz = btf__resolve_size(obj->btf, t->type);
1215                         if (sz < 0) {
1216                                 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1217                                            map_name, t->type, sz);
1218                                 return sz;
1219                         }
1220                         pr_debug("map '%s': found key [%u], sz = %lld.\n",
1221                                  map_name, t->type, sz);
1222                         if (map->def.key_size && map->def.key_size != sz) {
1223                                 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1224                                            map_name, map->def.key_size, sz);
1225                                 return -EINVAL;
1226                         }
1227                         map->def.key_size = sz;
1228                         map->btf_key_type_id = t->type;
1229                 } else if (strcmp(name, "value_size") == 0) {
1230                         __u32 sz;
1231
1232                         if (!get_map_field_int(map_name, obj->btf, def, m,
1233                                                &sz))
1234                                 return -EINVAL;
1235                         pr_debug("map '%s': found value_size = %u.\n",
1236                                  map_name, sz);
1237                         if (map->def.value_size && map->def.value_size != sz) {
1238                                 pr_warning("map '%s': conflicting value size %u != %u.\n",
1239                                            map_name, map->def.value_size, sz);
1240                                 return -EINVAL;
1241                         }
1242                         map->def.value_size = sz;
1243                 } else if (strcmp(name, "value") == 0) {
1244                         __s64 sz;
1245
1246                         t = btf__type_by_id(obj->btf, m->type);
1247                         if (!t) {
1248                                 pr_warning("map '%s': value type [%d] not found.\n",
1249                                            map_name, m->type);
1250                                 return -EINVAL;
1251                         }
1252                         if (!btf_is_ptr(t)) {
1253                                 pr_warning("map '%s': value spec is not PTR: %u.\n",
1254                                            map_name, btf_kind(t));
1255                                 return -EINVAL;
1256                         }
1257                         sz = btf__resolve_size(obj->btf, t->type);
1258                         if (sz < 0) {
1259                                 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1260                                            map_name, t->type, sz);
1261                                 return sz;
1262                         }
1263                         pr_debug("map '%s': found value [%u], sz = %lld.\n",
1264                                  map_name, t->type, sz);
1265                         if (map->def.value_size && map->def.value_size != sz) {
1266                                 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1267                                            map_name, map->def.value_size, sz);
1268                                 return -EINVAL;
1269                         }
1270                         map->def.value_size = sz;
1271                         map->btf_value_type_id = t->type;
1272                 } else {
1273                         if (strict) {
1274                                 pr_warning("map '%s': unknown field '%s'.\n",
1275                                            map_name, name);
1276                                 return -ENOTSUP;
1277                         }
1278                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1279                                  map_name, name);
1280                 }
1281         }
1282
1283         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1284                 pr_warning("map '%s': map type isn't specified.\n", map_name);
1285                 return -EINVAL;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1292 {
1293         const struct btf_type *sec = NULL;
1294         int nr_types, i, vlen, err;
1295         const struct btf_type *t;
1296         const char *name;
1297         Elf_Data *data;
1298         Elf_Scn *scn;
1299
1300         if (obj->efile.btf_maps_shndx < 0)
1301                 return 0;
1302
1303         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1304         if (scn)
1305                 data = elf_getdata(scn, NULL);
1306         if (!scn || !data) {
1307                 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1308                            obj->efile.maps_shndx, MAPS_ELF_SEC);
1309                 return -EINVAL;
1310         }
1311
1312         nr_types = btf__get_nr_types(obj->btf);
1313         for (i = 1; i <= nr_types; i++) {
1314                 t = btf__type_by_id(obj->btf, i);
1315                 if (!btf_is_datasec(t))
1316                         continue;
1317                 name = btf__name_by_offset(obj->btf, t->name_off);
1318                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1319                         sec = t;
1320                         break;
1321                 }
1322         }
1323
1324         if (!sec) {
1325                 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1326                 return -ENOENT;
1327         }
1328
1329         vlen = btf_vlen(sec);
1330         for (i = 0; i < vlen; i++) {
1331                 err = bpf_object__init_user_btf_map(obj, sec, i,
1332                                                     obj->efile.btf_maps_shndx,
1333                                                     data, strict);
1334                 if (err)
1335                         return err;
1336         }
1337
1338         return 0;
1339 }
1340
1341 static int bpf_object__init_maps(struct bpf_object *obj, bool relaxed_maps)
1342 {
1343         bool strict = !relaxed_maps;
1344         int err;
1345
1346         err = bpf_object__init_user_maps(obj, strict);
1347         if (err)
1348                 return err;
1349
1350         err = bpf_object__init_user_btf_maps(obj, strict);
1351         if (err)
1352                 return err;
1353
1354         err = bpf_object__init_global_data_maps(obj);
1355         if (err)
1356                 return err;
1357
1358         if (obj->nr_maps) {
1359                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1360                       compare_bpf_map);
1361         }
1362         return 0;
1363 }
1364
1365 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1366 {
1367         Elf_Scn *scn;
1368         GElf_Shdr sh;
1369
1370         scn = elf_getscn(obj->efile.elf, idx);
1371         if (!scn)
1372                 return false;
1373
1374         if (gelf_getshdr(scn, &sh) != &sh)
1375                 return false;
1376
1377         if (sh.sh_flags & SHF_EXECINSTR)
1378                 return true;
1379
1380         return false;
1381 }
1382
1383 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1384 {
1385         bool has_datasec = obj->caps.btf_datasec;
1386         bool has_func = obj->caps.btf_func;
1387         struct btf *btf = obj->btf;
1388         struct btf_type *t;
1389         int i, j, vlen;
1390
1391         if (!obj->btf || (has_func && has_datasec))
1392                 return;
1393
1394         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1395                 t = (struct btf_type *)btf__type_by_id(btf, i);
1396
1397                 if (!has_datasec && btf_is_var(t)) {
1398                         /* replace VAR with INT */
1399                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1400                         /*
1401                          * using size = 1 is the safest choice, 4 will be too
1402                          * big and cause kernel BTF validation failure if
1403                          * original variable took less than 4 bytes
1404                          */
1405                         t->size = 1;
1406                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1407                 } else if (!has_datasec && btf_is_datasec(t)) {
1408                         /* replace DATASEC with STRUCT */
1409                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
1410                         struct btf_member *m = btf_members(t);
1411                         struct btf_type *vt;
1412                         char *name;
1413
1414                         name = (char *)btf__name_by_offset(btf, t->name_off);
1415                         while (*name) {
1416                                 if (*name == '.')
1417                                         *name = '_';
1418                                 name++;
1419                         }
1420
1421                         vlen = btf_vlen(t);
1422                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1423                         for (j = 0; j < vlen; j++, v++, m++) {
1424                                 /* order of field assignments is important */
1425                                 m->offset = v->offset * 8;
1426                                 m->type = v->type;
1427                                 /* preserve variable name as member name */
1428                                 vt = (void *)btf__type_by_id(btf, v->type);
1429                                 m->name_off = vt->name_off;
1430                         }
1431                 } else if (!has_func && btf_is_func_proto(t)) {
1432                         /* replace FUNC_PROTO with ENUM */
1433                         vlen = btf_vlen(t);
1434                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1435                         t->size = sizeof(__u32); /* kernel enforced */
1436                 } else if (!has_func && btf_is_func(t)) {
1437                         /* replace FUNC with TYPEDEF */
1438                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1439                 }
1440         }
1441 }
1442
1443 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1444 {
1445         if (!obj->btf_ext)
1446                 return;
1447
1448         if (!obj->caps.btf_func) {
1449                 btf_ext__free(obj->btf_ext);
1450                 obj->btf_ext = NULL;
1451         }
1452 }
1453
1454 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1455 {
1456         return obj->efile.btf_maps_shndx >= 0;
1457 }
1458
1459 static int bpf_object__init_btf(struct bpf_object *obj,
1460                                 Elf_Data *btf_data,
1461                                 Elf_Data *btf_ext_data)
1462 {
1463         bool btf_required = bpf_object__is_btf_mandatory(obj);
1464         int err = 0;
1465
1466         if (btf_data) {
1467                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1468                 if (IS_ERR(obj->btf)) {
1469                         pr_warning("Error loading ELF section %s: %d.\n",
1470                                    BTF_ELF_SEC, err);
1471                         goto out;
1472                 }
1473                 err = btf__finalize_data(obj, obj->btf);
1474                 if (err) {
1475                         pr_warning("Error finalizing %s: %d.\n",
1476                                    BTF_ELF_SEC, err);
1477                         goto out;
1478                 }
1479         }
1480         if (btf_ext_data) {
1481                 if (!obj->btf) {
1482                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1483                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1484                         goto out;
1485                 }
1486                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1487                                             btf_ext_data->d_size);
1488                 if (IS_ERR(obj->btf_ext)) {
1489                         pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1490                                    BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1491                         obj->btf_ext = NULL;
1492                         goto out;
1493                 }
1494         }
1495 out:
1496         if (err || IS_ERR(obj->btf)) {
1497                 if (btf_required)
1498                         err = err ? : PTR_ERR(obj->btf);
1499                 else
1500                         err = 0;
1501                 if (!IS_ERR_OR_NULL(obj->btf))
1502                         btf__free(obj->btf);
1503                 obj->btf = NULL;
1504         }
1505         if (btf_required && !obj->btf) {
1506                 pr_warning("BTF is required, but is missing or corrupted.\n");
1507                 return err == 0 ? -ENOENT : err;
1508         }
1509         return 0;
1510 }
1511
1512 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1513 {
1514         int err = 0;
1515
1516         if (!obj->btf)
1517                 return 0;
1518
1519         bpf_object__sanitize_btf(obj);
1520         bpf_object__sanitize_btf_ext(obj);
1521
1522         err = btf__load(obj->btf);
1523         if (err) {
1524                 pr_warning("Error loading %s into kernel: %d.\n",
1525                            BTF_ELF_SEC, err);
1526                 btf__free(obj->btf);
1527                 obj->btf = NULL;
1528                 /* btf_ext can't exist without btf, so free it as well */
1529                 if (obj->btf_ext) {
1530                         btf_ext__free(obj->btf_ext);
1531                         obj->btf_ext = NULL;
1532                 }
1533
1534                 if (bpf_object__is_btf_mandatory(obj))
1535                         return err;
1536         }
1537         return 0;
1538 }
1539
1540 static int bpf_object__elf_collect(struct bpf_object *obj, bool relaxed_maps)
1541 {
1542         Elf *elf = obj->efile.elf;
1543         GElf_Ehdr *ep = &obj->efile.ehdr;
1544         Elf_Data *btf_ext_data = NULL;
1545         Elf_Data *btf_data = NULL;
1546         Elf_Scn *scn = NULL;
1547         int idx = 0, err = 0;
1548
1549         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1550         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1551                 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1552                 return -LIBBPF_ERRNO__FORMAT;
1553         }
1554
1555         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1556                 char *name;
1557                 GElf_Shdr sh;
1558                 Elf_Data *data;
1559
1560                 idx++;
1561                 if (gelf_getshdr(scn, &sh) != &sh) {
1562                         pr_warning("failed to get section(%d) header from %s\n",
1563                                    idx, obj->path);
1564                         return -LIBBPF_ERRNO__FORMAT;
1565                 }
1566
1567                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1568                 if (!name) {
1569                         pr_warning("failed to get section(%d) name from %s\n",
1570                                    idx, obj->path);
1571                         return -LIBBPF_ERRNO__FORMAT;
1572                 }
1573
1574                 data = elf_getdata(scn, 0);
1575                 if (!data) {
1576                         pr_warning("failed to get section(%d) data from %s(%s)\n",
1577                                    idx, name, obj->path);
1578                         return -LIBBPF_ERRNO__FORMAT;
1579                 }
1580                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1581                          idx, name, (unsigned long)data->d_size,
1582                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1583                          (int)sh.sh_type);
1584
1585                 if (strcmp(name, "license") == 0) {
1586                         err = bpf_object__init_license(obj,
1587                                                        data->d_buf,
1588                                                        data->d_size);
1589                         if (err)
1590                                 return err;
1591                 } else if (strcmp(name, "version") == 0) {
1592                         err = bpf_object__init_kversion(obj,
1593                                                         data->d_buf,
1594                                                         data->d_size);
1595                         if (err)
1596                                 return err;
1597                 } else if (strcmp(name, "maps") == 0) {
1598                         obj->efile.maps_shndx = idx;
1599                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1600                         obj->efile.btf_maps_shndx = idx;
1601                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1602                         btf_data = data;
1603                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1604                         btf_ext_data = data;
1605                 } else if (sh.sh_type == SHT_SYMTAB) {
1606                         if (obj->efile.symbols) {
1607                                 pr_warning("bpf: multiple SYMTAB in %s\n",
1608                                            obj->path);
1609                                 return -LIBBPF_ERRNO__FORMAT;
1610                         }
1611                         obj->efile.symbols = data;
1612                         obj->efile.strtabidx = sh.sh_link;
1613                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1614                         if (sh.sh_flags & SHF_EXECINSTR) {
1615                                 if (strcmp(name, ".text") == 0)
1616                                         obj->efile.text_shndx = idx;
1617                                 err = bpf_object__add_program(obj, data->d_buf,
1618                                                               data->d_size, name, idx);
1619                                 if (err) {
1620                                         char errmsg[STRERR_BUFSIZE];
1621                                         char *cp = libbpf_strerror_r(-err, errmsg,
1622                                                                      sizeof(errmsg));
1623
1624                                         pr_warning("failed to alloc program %s (%s): %s",
1625                                                    name, obj->path, cp);
1626                                         return err;
1627                                 }
1628                         } else if (strcmp(name, ".data") == 0) {
1629                                 obj->efile.data = data;
1630                                 obj->efile.data_shndx = idx;
1631                         } else if (strcmp(name, ".rodata") == 0) {
1632                                 obj->efile.rodata = data;
1633                                 obj->efile.rodata_shndx = idx;
1634                         } else {
1635                                 pr_debug("skip section(%d) %s\n", idx, name);
1636                         }
1637                 } else if (sh.sh_type == SHT_REL) {
1638                         int nr_reloc = obj->efile.nr_reloc;
1639                         void *reloc = obj->efile.reloc;
1640                         int sec = sh.sh_info; /* points to other section */
1641
1642                         /* Only do relo for section with exec instructions */
1643                         if (!section_have_execinstr(obj, sec)) {
1644                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1645                                          name, idx, sec);
1646                                 continue;
1647                         }
1648
1649                         reloc = reallocarray(reloc, nr_reloc + 1,
1650                                              sizeof(*obj->efile.reloc));
1651                         if (!reloc) {
1652                                 pr_warning("realloc failed\n");
1653                                 return -ENOMEM;
1654                         }
1655
1656                         obj->efile.reloc = reloc;
1657                         obj->efile.nr_reloc++;
1658
1659                         obj->efile.reloc[nr_reloc].shdr = sh;
1660                         obj->efile.reloc[nr_reloc].data = data;
1661                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1662                         obj->efile.bss = data;
1663                         obj->efile.bss_shndx = idx;
1664                 } else {
1665                         pr_debug("skip section(%d) %s\n", idx, name);
1666                 }
1667         }
1668
1669         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1670                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1671                 return -LIBBPF_ERRNO__FORMAT;
1672         }
1673         err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1674         if (!err)
1675                 err = bpf_object__init_maps(obj, relaxed_maps);
1676         if (!err)
1677                 err = bpf_object__sanitize_and_load_btf(obj);
1678         if (!err)
1679                 err = bpf_object__init_prog_names(obj);
1680         return err;
1681 }
1682
1683 static struct bpf_program *
1684 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1685 {
1686         struct bpf_program *prog;
1687         size_t i;
1688
1689         for (i = 0; i < obj->nr_programs; i++) {
1690                 prog = &obj->programs[i];
1691                 if (prog->idx == idx)
1692                         return prog;
1693         }
1694         return NULL;
1695 }
1696
1697 struct bpf_program *
1698 bpf_object__find_program_by_title(const struct bpf_object *obj,
1699                                   const char *title)
1700 {
1701         struct bpf_program *pos;
1702
1703         bpf_object__for_each_program(pos, obj) {
1704                 if (pos->section_name && !strcmp(pos->section_name, title))
1705                         return pos;
1706         }
1707         return NULL;
1708 }
1709
1710 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1711                                       int shndx)
1712 {
1713         return shndx == obj->efile.data_shndx ||
1714                shndx == obj->efile.bss_shndx ||
1715                shndx == obj->efile.rodata_shndx;
1716 }
1717
1718 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1719                                       int shndx)
1720 {
1721         return shndx == obj->efile.maps_shndx ||
1722                shndx == obj->efile.btf_maps_shndx;
1723 }
1724
1725 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1726                                               int shndx)
1727 {
1728         return shndx == obj->efile.text_shndx ||
1729                bpf_object__shndx_is_maps(obj, shndx) ||
1730                bpf_object__shndx_is_data(obj, shndx);
1731 }
1732
1733 static enum libbpf_map_type
1734 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1735 {
1736         if (shndx == obj->efile.data_shndx)
1737                 return LIBBPF_MAP_DATA;
1738         else if (shndx == obj->efile.bss_shndx)
1739                 return LIBBPF_MAP_BSS;
1740         else if (shndx == obj->efile.rodata_shndx)
1741                 return LIBBPF_MAP_RODATA;
1742         else
1743                 return LIBBPF_MAP_UNSPEC;
1744 }
1745
1746 static int
1747 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1748                            Elf_Data *data, struct bpf_object *obj)
1749 {
1750         Elf_Data *symbols = obj->efile.symbols;
1751         struct bpf_map *maps = obj->maps;
1752         size_t nr_maps = obj->nr_maps;
1753         int i, nrels;
1754
1755         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1756         nrels = shdr->sh_size / shdr->sh_entsize;
1757
1758         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1759         if (!prog->reloc_desc) {
1760                 pr_warning("failed to alloc memory in relocation\n");
1761                 return -ENOMEM;
1762         }
1763         prog->nr_reloc = nrels;
1764
1765         for (i = 0; i < nrels; i++) {
1766                 struct bpf_insn *insns = prog->insns;
1767                 enum libbpf_map_type type;
1768                 unsigned int insn_idx;
1769                 unsigned int shdr_idx;
1770                 const char *name;
1771                 size_t map_idx;
1772                 GElf_Sym sym;
1773                 GElf_Rel rel;
1774
1775                 if (!gelf_getrel(data, i, &rel)) {
1776                         pr_warning("relocation: failed to get %d reloc\n", i);
1777                         return -LIBBPF_ERRNO__FORMAT;
1778                 }
1779
1780                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1781                         pr_warning("relocation: symbol %"PRIx64" not found\n",
1782                                    GELF_R_SYM(rel.r_info));
1783                         return -LIBBPF_ERRNO__FORMAT;
1784                 }
1785
1786                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1787                                   sym.st_name) ? : "<?>";
1788
1789                 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1790                          (long long) (rel.r_info >> 32),
1791                          (long long) sym.st_value, sym.st_name, name);
1792
1793                 shdr_idx = sym.st_shndx;
1794                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1795                 pr_debug("relocation: insn_idx=%u, shdr_idx=%u\n",
1796                          insn_idx, shdr_idx);
1797
1798                 if (shdr_idx >= SHN_LORESERVE) {
1799                         pr_warning("relocation: not yet supported relo for non-static global \'%s\' variable in special section (0x%x) found in insns[%d].code 0x%x\n",
1800                                    name, shdr_idx, insn_idx,
1801                                    insns[insn_idx].code);
1802                         return -LIBBPF_ERRNO__RELOC;
1803                 }
1804                 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1805                         pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1806                                    prog->section_name, shdr_idx);
1807                         return -LIBBPF_ERRNO__RELOC;
1808                 }
1809
1810                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1811                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1812                                 pr_warning("incorrect bpf_call opcode\n");
1813                                 return -LIBBPF_ERRNO__RELOC;
1814                         }
1815                         prog->reloc_desc[i].type = RELO_CALL;
1816                         prog->reloc_desc[i].insn_idx = insn_idx;
1817                         prog->reloc_desc[i].text_off = sym.st_value;
1818                         obj->has_pseudo_calls = true;
1819                         continue;
1820                 }
1821
1822                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1823                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1824                                    insn_idx, insns[insn_idx].code);
1825                         return -LIBBPF_ERRNO__RELOC;
1826                 }
1827
1828                 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1829                     bpf_object__shndx_is_data(obj, shdr_idx)) {
1830                         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1831                         if (type != LIBBPF_MAP_UNSPEC) {
1832                                 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1833                                         pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1834                                                    name, insn_idx, insns[insn_idx].code);
1835                                         return -LIBBPF_ERRNO__RELOC;
1836                                 }
1837                                 if (!obj->caps.global_data) {
1838                                         pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1839                                                    name, insn_idx);
1840                                         return -LIBBPF_ERRNO__RELOC;
1841                                 }
1842                         }
1843
1844                         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1845                                 if (maps[map_idx].libbpf_type != type)
1846                                         continue;
1847                                 if (type != LIBBPF_MAP_UNSPEC ||
1848                                     (maps[map_idx].sec_idx == sym.st_shndx &&
1849                                      maps[map_idx].sec_offset == sym.st_value)) {
1850                                         pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1851                                                  map_idx, maps[map_idx].name,
1852                                                  maps[map_idx].sec_idx,
1853                                                  maps[map_idx].sec_offset,
1854                                                  insn_idx);
1855                                         break;
1856                                 }
1857                         }
1858
1859                         if (map_idx >= nr_maps) {
1860                                 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1861                                            (int)map_idx, (int)nr_maps - 1);
1862                                 return -LIBBPF_ERRNO__RELOC;
1863                         }
1864
1865                         prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1866                                                    RELO_DATA : RELO_LD64;
1867                         prog->reloc_desc[i].insn_idx = insn_idx;
1868                         prog->reloc_desc[i].map_idx = map_idx;
1869                 }
1870         }
1871         return 0;
1872 }
1873
1874 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1875 {
1876         struct bpf_map_def *def = &map->def;
1877         __u32 key_type_id = 0, value_type_id = 0;
1878         int ret;
1879
1880         /* if it's BTF-defined map, we don't need to search for type IDs */
1881         if (map->sec_idx == obj->efile.btf_maps_shndx)
1882                 return 0;
1883
1884         if (!bpf_map__is_internal(map)) {
1885                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1886                                            def->value_size, &key_type_id,
1887                                            &value_type_id);
1888         } else {
1889                 /*
1890                  * LLVM annotates global data differently in BTF, that is,
1891                  * only as '.data', '.bss' or '.rodata'.
1892                  */
1893                 ret = btf__find_by_name(obj->btf,
1894                                 libbpf_type_to_btf_name[map->libbpf_type]);
1895         }
1896         if (ret < 0)
1897                 return ret;
1898
1899         map->btf_key_type_id = key_type_id;
1900         map->btf_value_type_id = bpf_map__is_internal(map) ?
1901                                  ret : value_type_id;
1902         return 0;
1903 }
1904
1905 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1906 {
1907         struct bpf_map_info info = {};
1908         __u32 len = sizeof(info);
1909         int new_fd, err;
1910         char *new_name;
1911
1912         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1913         if (err)
1914                 return err;
1915
1916         new_name = strdup(info.name);
1917         if (!new_name)
1918                 return -errno;
1919
1920         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1921         if (new_fd < 0)
1922                 goto err_free_new_name;
1923
1924         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1925         if (new_fd < 0)
1926                 goto err_close_new_fd;
1927
1928         err = zclose(map->fd);
1929         if (err)
1930                 goto err_close_new_fd;
1931         free(map->name);
1932
1933         map->fd = new_fd;
1934         map->name = new_name;
1935         map->def.type = info.type;
1936         map->def.key_size = info.key_size;
1937         map->def.value_size = info.value_size;
1938         map->def.max_entries = info.max_entries;
1939         map->def.map_flags = info.map_flags;
1940         map->btf_key_type_id = info.btf_key_type_id;
1941         map->btf_value_type_id = info.btf_value_type_id;
1942
1943         return 0;
1944
1945 err_close_new_fd:
1946         close(new_fd);
1947 err_free_new_name:
1948         free(new_name);
1949         return -errno;
1950 }
1951
1952 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1953 {
1954         if (!map || !max_entries)
1955                 return -EINVAL;
1956
1957         /* If map already created, its attributes can't be changed. */
1958         if (map->fd >= 0)
1959                 return -EBUSY;
1960
1961         map->def.max_entries = max_entries;
1962
1963         return 0;
1964 }
1965
1966 static int
1967 bpf_object__probe_name(struct bpf_object *obj)
1968 {
1969         struct bpf_load_program_attr attr;
1970         char *cp, errmsg[STRERR_BUFSIZE];
1971         struct bpf_insn insns[] = {
1972                 BPF_MOV64_IMM(BPF_REG_0, 0),
1973                 BPF_EXIT_INSN(),
1974         };
1975         int ret;
1976
1977         /* make sure basic loading works */
1978
1979         memset(&attr, 0, sizeof(attr));
1980         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1981         attr.insns = insns;
1982         attr.insns_cnt = ARRAY_SIZE(insns);
1983         attr.license = "GPL";
1984
1985         ret = bpf_load_program_xattr(&attr, NULL, 0);
1986         if (ret < 0) {
1987                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1988                 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1989                            __func__, cp, errno);
1990                 return -errno;
1991         }
1992         close(ret);
1993
1994         /* now try the same program, but with the name */
1995
1996         attr.name = "test";
1997         ret = bpf_load_program_xattr(&attr, NULL, 0);
1998         if (ret >= 0) {
1999                 obj->caps.name = 1;
2000                 close(ret);
2001         }
2002
2003         return 0;
2004 }
2005
2006 static int
2007 bpf_object__probe_global_data(struct bpf_object *obj)
2008 {
2009         struct bpf_load_program_attr prg_attr;
2010         struct bpf_create_map_attr map_attr;
2011         char *cp, errmsg[STRERR_BUFSIZE];
2012         struct bpf_insn insns[] = {
2013                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
2014                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
2015                 BPF_MOV64_IMM(BPF_REG_0, 0),
2016                 BPF_EXIT_INSN(),
2017         };
2018         int ret, map;
2019
2020         memset(&map_attr, 0, sizeof(map_attr));
2021         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2022         map_attr.key_size = sizeof(int);
2023         map_attr.value_size = 32;
2024         map_attr.max_entries = 1;
2025
2026         map = bpf_create_map_xattr(&map_attr);
2027         if (map < 0) {
2028                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2029                 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
2030                            __func__, cp, errno);
2031                 return -errno;
2032         }
2033
2034         insns[0].imm = map;
2035
2036         memset(&prg_attr, 0, sizeof(prg_attr));
2037         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2038         prg_attr.insns = insns;
2039         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2040         prg_attr.license = "GPL";
2041
2042         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2043         if (ret >= 0) {
2044                 obj->caps.global_data = 1;
2045                 close(ret);
2046         }
2047
2048         close(map);
2049         return 0;
2050 }
2051
2052 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2053 {
2054         const char strs[] = "\0int\0x\0a";
2055         /* void x(int a) {} */
2056         __u32 types[] = {
2057                 /* int */
2058                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2059                 /* FUNC_PROTO */                                /* [2] */
2060                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2061                 BTF_PARAM_ENC(7, 1),
2062                 /* FUNC x */                                    /* [3] */
2063                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2064         };
2065         int btf_fd;
2066
2067         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2068                                       strs, sizeof(strs));
2069         if (btf_fd >= 0) {
2070                 obj->caps.btf_func = 1;
2071                 close(btf_fd);
2072                 return 1;
2073         }
2074
2075         return 0;
2076 }
2077
2078 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2079 {
2080         const char strs[] = "\0x\0.data";
2081         /* static int a; */
2082         __u32 types[] = {
2083                 /* int */
2084                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2085                 /* VAR x */                                     /* [2] */
2086                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2087                 BTF_VAR_STATIC,
2088                 /* DATASEC val */                               /* [3] */
2089                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2090                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2091         };
2092         int btf_fd;
2093
2094         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2095                                       strs, sizeof(strs));
2096         if (btf_fd >= 0) {
2097                 obj->caps.btf_datasec = 1;
2098                 close(btf_fd);
2099                 return 1;
2100         }
2101
2102         return 0;
2103 }
2104
2105 static int
2106 bpf_object__probe_caps(struct bpf_object *obj)
2107 {
2108         int (*probe_fn[])(struct bpf_object *obj) = {
2109                 bpf_object__probe_name,
2110                 bpf_object__probe_global_data,
2111                 bpf_object__probe_btf_func,
2112                 bpf_object__probe_btf_datasec,
2113         };
2114         int i, ret;
2115
2116         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2117                 ret = probe_fn[i](obj);
2118                 if (ret < 0)
2119                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2120         }
2121
2122         return 0;
2123 }
2124
2125 static int
2126 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2127 {
2128         char *cp, errmsg[STRERR_BUFSIZE];
2129         int err, zero = 0;
2130         __u8 *data;
2131
2132         /* Nothing to do here since kernel already zero-initializes .bss map. */
2133         if (map->libbpf_type == LIBBPF_MAP_BSS)
2134                 return 0;
2135
2136         data = map->libbpf_type == LIBBPF_MAP_DATA ?
2137                obj->sections.data : obj->sections.rodata;
2138
2139         err = bpf_map_update_elem(map->fd, &zero, data, 0);
2140         /* Freeze .rodata map as read-only from syscall side. */
2141         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2142                 err = bpf_map_freeze(map->fd);
2143                 if (err) {
2144                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2145                         pr_warning("Error freezing map(%s) as read-only: %s\n",
2146                                    map->name, cp);
2147                         err = 0;
2148                 }
2149         }
2150         return err;
2151 }
2152
2153 static int
2154 bpf_object__create_maps(struct bpf_object *obj)
2155 {
2156         struct bpf_create_map_attr create_attr = {};
2157         int nr_cpus = 0;
2158         unsigned int i;
2159         int err;
2160
2161         for (i = 0; i < obj->nr_maps; i++) {
2162                 struct bpf_map *map = &obj->maps[i];
2163                 struct bpf_map_def *def = &map->def;
2164                 char *cp, errmsg[STRERR_BUFSIZE];
2165                 int *pfd = &map->fd;
2166
2167                 if (map->fd >= 0) {
2168                         pr_debug("skip map create (preset) %s: fd=%d\n",
2169                                  map->name, map->fd);
2170                         continue;
2171                 }
2172
2173                 if (obj->caps.name)
2174                         create_attr.name = map->name;
2175                 create_attr.map_ifindex = map->map_ifindex;
2176                 create_attr.map_type = def->type;
2177                 create_attr.map_flags = def->map_flags;
2178                 create_attr.key_size = def->key_size;
2179                 create_attr.value_size = def->value_size;
2180                 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
2181                     !def->max_entries) {
2182                         if (!nr_cpus)
2183                                 nr_cpus = libbpf_num_possible_cpus();
2184                         if (nr_cpus < 0) {
2185                                 pr_warning("failed to determine number of system CPUs: %d\n",
2186                                            nr_cpus);
2187                                 err = nr_cpus;
2188                                 goto err_out;
2189                         }
2190                         pr_debug("map '%s': setting size to %d\n",
2191                                  map->name, nr_cpus);
2192                         create_attr.max_entries = nr_cpus;
2193                 } else {
2194                         create_attr.max_entries = def->max_entries;
2195                 }
2196                 create_attr.btf_fd = 0;
2197                 create_attr.btf_key_type_id = 0;
2198                 create_attr.btf_value_type_id = 0;
2199                 if (bpf_map_type__is_map_in_map(def->type) &&
2200                     map->inner_map_fd >= 0)
2201                         create_attr.inner_map_fd = map->inner_map_fd;
2202
2203                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2204                         create_attr.btf_fd = btf__fd(obj->btf);
2205                         create_attr.btf_key_type_id = map->btf_key_type_id;
2206                         create_attr.btf_value_type_id = map->btf_value_type_id;
2207                 }
2208
2209                 *pfd = bpf_create_map_xattr(&create_attr);
2210                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2211                                  create_attr.btf_value_type_id)) {
2212                         err = -errno;
2213                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2214                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2215                                    map->name, cp, err);
2216                         create_attr.btf_fd = 0;
2217                         create_attr.btf_key_type_id = 0;
2218                         create_attr.btf_value_type_id = 0;
2219                         map->btf_key_type_id = 0;
2220                         map->btf_value_type_id = 0;
2221                         *pfd = bpf_create_map_xattr(&create_attr);
2222                 }
2223
2224                 if (*pfd < 0) {
2225                         size_t j;
2226
2227                         err = -errno;
2228 err_out:
2229                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2230                         pr_warning("failed to create map (name: '%s'): %s(%d)\n",
2231                                    map->name, cp, err);
2232                         for (j = 0; j < i; j++)
2233                                 zclose(obj->maps[j].fd);
2234                         return err;
2235                 }
2236
2237                 if (bpf_map__is_internal(map)) {
2238                         err = bpf_object__populate_internal_map(obj, map);
2239                         if (err < 0) {
2240                                 zclose(*pfd);
2241                                 goto err_out;
2242                         }
2243                 }
2244
2245                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2246         }
2247
2248         return 0;
2249 }
2250
2251 static int
2252 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2253                         void *btf_prog_info, const char *info_name)
2254 {
2255         if (err != -ENOENT) {
2256                 pr_warning("Error in loading %s for sec %s.\n",
2257                            info_name, prog->section_name);
2258                 return err;
2259         }
2260
2261         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2262
2263         if (btf_prog_info) {
2264                 /*
2265                  * Some info has already been found but has problem
2266                  * in the last btf_ext reloc. Must have to error out.
2267                  */
2268                 pr_warning("Error in relocating %s for sec %s.\n",
2269                            info_name, prog->section_name);
2270                 return err;
2271         }
2272
2273         /* Have problem loading the very first info. Ignore the rest. */
2274         pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2275                    info_name, prog->section_name, info_name);
2276         return 0;
2277 }
2278
2279 static int
2280 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2281                           const char *section_name,  __u32 insn_offset)
2282 {
2283         int err;
2284
2285         if (!insn_offset || prog->func_info) {
2286                 /*
2287                  * !insn_offset => main program
2288                  *
2289                  * For sub prog, the main program's func_info has to
2290                  * be loaded first (i.e. prog->func_info != NULL)
2291                  */
2292                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2293                                                section_name, insn_offset,
2294                                                &prog->func_info,
2295                                                &prog->func_info_cnt);
2296                 if (err)
2297                         return check_btf_ext_reloc_err(prog, err,
2298                                                        prog->func_info,
2299                                                        "bpf_func_info");
2300
2301                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2302         }
2303
2304         if (!insn_offset || prog->line_info) {
2305                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2306                                                section_name, insn_offset,
2307                                                &prog->line_info,
2308                                                &prog->line_info_cnt);
2309                 if (err)
2310                         return check_btf_ext_reloc_err(prog, err,
2311                                                        prog->line_info,
2312                                                        "bpf_line_info");
2313
2314                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2315         }
2316
2317         return 0;
2318 }
2319
2320 #define BPF_CORE_SPEC_MAX_LEN 64
2321
2322 /* represents BPF CO-RE field or array element accessor */
2323 struct bpf_core_accessor {
2324         __u32 type_id;          /* struct/union type or array element type */
2325         __u32 idx;              /* field index or array index */
2326         const char *name;       /* field name or NULL for array accessor */
2327 };
2328
2329 struct bpf_core_spec {
2330         const struct btf *btf;
2331         /* high-level spec: named fields and array indices only */
2332         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
2333         /* high-level spec length */
2334         int len;
2335         /* raw, low-level spec: 1-to-1 with accessor spec string */
2336         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
2337         /* raw spec length */
2338         int raw_len;
2339         /* field byte offset represented by spec */
2340         __u32 offset;
2341 };
2342
2343 static bool str_is_empty(const char *s)
2344 {
2345         return !s || !s[0];
2346 }
2347
2348 /*
2349  * Turn bpf_field_reloc into a low- and high-level spec representation,
2350  * validating correctness along the way, as well as calculating resulting
2351  * field offset (in bytes), specified by accessor string. Low-level spec
2352  * captures every single level of nestedness, including traversing anonymous
2353  * struct/union members. High-level one only captures semantically meaningful
2354  * "turning points": named fields and array indicies.
2355  * E.g., for this case:
2356  *
2357  *   struct sample {
2358  *       int __unimportant;
2359  *       struct {
2360  *           int __1;
2361  *           int __2;
2362  *           int a[7];
2363  *       };
2364  *   };
2365  *
2366  *   struct sample *s = ...;
2367  *
2368  *   int x = &s->a[3]; // access string = '0:1:2:3'
2369  *
2370  * Low-level spec has 1:1 mapping with each element of access string (it's
2371  * just a parsed access string representation): [0, 1, 2, 3].
2372  *
2373  * High-level spec will capture only 3 points:
2374  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
2375  *   - field 'a' access (corresponds to '2' in low-level spec);
2376  *   - array element #3 access (corresponds to '3' in low-level spec).
2377  *
2378  */
2379 static int bpf_core_spec_parse(const struct btf *btf,
2380                                __u32 type_id,
2381                                const char *spec_str,
2382                                struct bpf_core_spec *spec)
2383 {
2384         int access_idx, parsed_len, i;
2385         const struct btf_type *t;
2386         const char *name;
2387         __u32 id;
2388         __s64 sz;
2389
2390         if (str_is_empty(spec_str) || *spec_str == ':')
2391                 return -EINVAL;
2392
2393         memset(spec, 0, sizeof(*spec));
2394         spec->btf = btf;
2395
2396         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
2397         while (*spec_str) {
2398                 if (*spec_str == ':')
2399                         ++spec_str;
2400                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
2401                         return -EINVAL;
2402                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2403                         return -E2BIG;
2404                 spec_str += parsed_len;
2405                 spec->raw_spec[spec->raw_len++] = access_idx;
2406         }
2407
2408         if (spec->raw_len == 0)
2409                 return -EINVAL;
2410
2411         /* first spec value is always reloc type array index */
2412         t = skip_mods_and_typedefs(btf, type_id, &id);
2413         if (!t)
2414                 return -EINVAL;
2415
2416         access_idx = spec->raw_spec[0];
2417         spec->spec[0].type_id = id;
2418         spec->spec[0].idx = access_idx;
2419         spec->len++;
2420
2421         sz = btf__resolve_size(btf, id);
2422         if (sz < 0)
2423                 return sz;
2424         spec->offset = access_idx * sz;
2425
2426         for (i = 1; i < spec->raw_len; i++) {
2427                 t = skip_mods_and_typedefs(btf, id, &id);
2428                 if (!t)
2429                         return -EINVAL;
2430
2431                 access_idx = spec->raw_spec[i];
2432
2433                 if (btf_is_composite(t)) {
2434                         const struct btf_member *m;
2435                         __u32 offset;
2436
2437                         if (access_idx >= btf_vlen(t))
2438                                 return -EINVAL;
2439                         if (btf_member_bitfield_size(t, access_idx))
2440                                 return -EINVAL;
2441
2442                         offset = btf_member_bit_offset(t, access_idx);
2443                         if (offset % 8)
2444                                 return -EINVAL;
2445                         spec->offset += offset / 8;
2446
2447                         m = btf_members(t) + access_idx;
2448                         if (m->name_off) {
2449                                 name = btf__name_by_offset(btf, m->name_off);
2450                                 if (str_is_empty(name))
2451                                         return -EINVAL;
2452
2453                                 spec->spec[spec->len].type_id = id;
2454                                 spec->spec[spec->len].idx = access_idx;
2455                                 spec->spec[spec->len].name = name;
2456                                 spec->len++;
2457                         }
2458
2459                         id = m->type;
2460                 } else if (btf_is_array(t)) {
2461                         const struct btf_array *a = btf_array(t);
2462
2463                         t = skip_mods_and_typedefs(btf, a->type, &id);
2464                         if (!t || access_idx >= a->nelems)
2465                                 return -EINVAL;
2466
2467                         spec->spec[spec->len].type_id = id;
2468                         spec->spec[spec->len].idx = access_idx;
2469                         spec->len++;
2470
2471                         sz = btf__resolve_size(btf, id);
2472                         if (sz < 0)
2473                                 return sz;
2474                         spec->offset += access_idx * sz;
2475                 } else {
2476                         pr_warning("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
2477                                    type_id, spec_str, i, id, btf_kind(t));
2478                         return -EINVAL;
2479                 }
2480         }
2481
2482         return 0;
2483 }
2484
2485 static bool bpf_core_is_flavor_sep(const char *s)
2486 {
2487         /* check X___Y name pattern, where X and Y are not underscores */
2488         return s[0] != '_' &&                                 /* X */
2489                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
2490                s[4] != '_';                                   /* Y */
2491 }
2492
2493 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
2494  * before last triple underscore. Struct name part after last triple
2495  * underscore is ignored by BPF CO-RE relocation during relocation matching.
2496  */
2497 static size_t bpf_core_essential_name_len(const char *name)
2498 {
2499         size_t n = strlen(name);
2500         int i;
2501
2502         for (i = n - 5; i >= 0; i--) {
2503                 if (bpf_core_is_flavor_sep(name + i))
2504                         return i + 1;
2505         }
2506         return n;
2507 }
2508
2509 /* dynamically sized list of type IDs */
2510 struct ids_vec {
2511         __u32 *data;
2512         int len;
2513 };
2514
2515 static void bpf_core_free_cands(struct ids_vec *cand_ids)
2516 {
2517         free(cand_ids->data);
2518         free(cand_ids);
2519 }
2520
2521 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
2522                                            __u32 local_type_id,
2523                                            const struct btf *targ_btf)
2524 {
2525         size_t local_essent_len, targ_essent_len;
2526         const char *local_name, *targ_name;
2527         const struct btf_type *t;
2528         struct ids_vec *cand_ids;
2529         __u32 *new_ids;
2530         int i, err, n;
2531
2532         t = btf__type_by_id(local_btf, local_type_id);
2533         if (!t)
2534                 return ERR_PTR(-EINVAL);
2535
2536         local_name = btf__name_by_offset(local_btf, t->name_off);
2537         if (str_is_empty(local_name))
2538                 return ERR_PTR(-EINVAL);
2539         local_essent_len = bpf_core_essential_name_len(local_name);
2540
2541         cand_ids = calloc(1, sizeof(*cand_ids));
2542         if (!cand_ids)
2543                 return ERR_PTR(-ENOMEM);
2544
2545         n = btf__get_nr_types(targ_btf);
2546         for (i = 1; i <= n; i++) {
2547                 t = btf__type_by_id(targ_btf, i);
2548                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
2549                 if (str_is_empty(targ_name))
2550                         continue;
2551
2552                 targ_essent_len = bpf_core_essential_name_len(targ_name);
2553                 if (targ_essent_len != local_essent_len)
2554                         continue;
2555
2556                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
2557                         pr_debug("[%d] %s: found candidate [%d] %s\n",
2558                                  local_type_id, local_name, i, targ_name);
2559                         new_ids = realloc(cand_ids->data, cand_ids->len + 1);
2560                         if (!new_ids) {
2561                                 err = -ENOMEM;
2562                                 goto err_out;
2563                         }
2564                         cand_ids->data = new_ids;
2565                         cand_ids->data[cand_ids->len++] = i;
2566                 }
2567         }
2568         return cand_ids;
2569 err_out:
2570         bpf_core_free_cands(cand_ids);
2571         return ERR_PTR(err);
2572 }
2573
2574 /* Check two types for compatibility, skipping const/volatile/restrict and
2575  * typedefs, to ensure we are relocating offset to the compatible entities:
2576  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
2577  *   - any two FWDs are compatible;
2578  *   - any two PTRs are always compatible;
2579  *   - for ENUMs, check sizes, names are ignored;
2580  *   - for INT, size and bitness should match, signedness is ignored;
2581  *   - for ARRAY, dimensionality is ignored, element types are checked for
2582  *     compatibility recursively;
2583  *   - everything else shouldn't be ever a target of relocation.
2584  * These rules are not set in stone and probably will be adjusted as we get
2585  * more experience with using BPF CO-RE relocations.
2586  */
2587 static int bpf_core_fields_are_compat(const struct btf *local_btf,
2588                                       __u32 local_id,
2589                                       const struct btf *targ_btf,
2590                                       __u32 targ_id)
2591 {
2592         const struct btf_type *local_type, *targ_type;
2593
2594 recur:
2595         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
2596         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2597         if (!local_type || !targ_type)
2598                 return -EINVAL;
2599
2600         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
2601                 return 1;
2602         if (btf_kind(local_type) != btf_kind(targ_type))
2603                 return 0;
2604
2605         switch (btf_kind(local_type)) {
2606         case BTF_KIND_FWD:
2607         case BTF_KIND_PTR:
2608                 return 1;
2609         case BTF_KIND_ENUM:
2610                 return local_type->size == targ_type->size;
2611         case BTF_KIND_INT:
2612                 return btf_int_offset(local_type) == 0 &&
2613                        btf_int_offset(targ_type) == 0 &&
2614                        local_type->size == targ_type->size &&
2615                        btf_int_bits(local_type) == btf_int_bits(targ_type);
2616         case BTF_KIND_ARRAY:
2617                 local_id = btf_array(local_type)->type;
2618                 targ_id = btf_array(targ_type)->type;
2619                 goto recur;
2620         default:
2621                 pr_warning("unexpected kind %d relocated, local [%d], target [%d]\n",
2622                            btf_kind(local_type), local_id, targ_id);
2623                 return 0;
2624         }
2625 }
2626
2627 /*
2628  * Given single high-level named field accessor in local type, find
2629  * corresponding high-level accessor for a target type. Along the way,
2630  * maintain low-level spec for target as well. Also keep updating target
2631  * offset.
2632  *
2633  * Searching is performed through recursive exhaustive enumeration of all
2634  * fields of a struct/union. If there are any anonymous (embedded)
2635  * structs/unions, they are recursively searched as well. If field with
2636  * desired name is found, check compatibility between local and target types,
2637  * before returning result.
2638  *
2639  * 1 is returned, if field is found.
2640  * 0 is returned if no compatible field is found.
2641  * <0 is returned on error.
2642  */
2643 static int bpf_core_match_member(const struct btf *local_btf,
2644                                  const struct bpf_core_accessor *local_acc,
2645                                  const struct btf *targ_btf,
2646                                  __u32 targ_id,
2647                                  struct bpf_core_spec *spec,
2648                                  __u32 *next_targ_id)
2649 {
2650         const struct btf_type *local_type, *targ_type;
2651         const struct btf_member *local_member, *m;
2652         const char *local_name, *targ_name;
2653         __u32 local_id;
2654         int i, n, found;
2655
2656         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
2657         if (!targ_type)
2658                 return -EINVAL;
2659         if (!btf_is_composite(targ_type))
2660                 return 0;
2661
2662         local_id = local_acc->type_id;
2663         local_type = btf__type_by_id(local_btf, local_id);
2664         local_member = btf_members(local_type) + local_acc->idx;
2665         local_name = btf__name_by_offset(local_btf, local_member->name_off);
2666
2667         n = btf_vlen(targ_type);
2668         m = btf_members(targ_type);
2669         for (i = 0; i < n; i++, m++) {
2670                 __u32 offset;
2671
2672                 /* bitfield relocations not supported */
2673                 if (btf_member_bitfield_size(targ_type, i))
2674                         continue;
2675                 offset = btf_member_bit_offset(targ_type, i);
2676                 if (offset % 8)
2677                         continue;
2678
2679                 /* too deep struct/union/array nesting */
2680                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2681                         return -E2BIG;
2682
2683                 /* speculate this member will be the good one */
2684                 spec->offset += offset / 8;
2685                 spec->raw_spec[spec->raw_len++] = i;
2686
2687                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
2688                 if (str_is_empty(targ_name)) {
2689                         /* embedded struct/union, we need to go deeper */
2690                         found = bpf_core_match_member(local_btf, local_acc,
2691                                                       targ_btf, m->type,
2692                                                       spec, next_targ_id);
2693                         if (found) /* either found or error */
2694                                 return found;
2695                 } else if (strcmp(local_name, targ_name) == 0) {
2696                         /* matching named field */
2697                         struct bpf_core_accessor *targ_acc;
2698
2699                         targ_acc = &spec->spec[spec->len++];
2700                         targ_acc->type_id = targ_id;
2701                         targ_acc->idx = i;
2702                         targ_acc->name = targ_name;
2703
2704                         *next_targ_id = m->type;
2705                         found = bpf_core_fields_are_compat(local_btf,
2706                                                            local_member->type,
2707                                                            targ_btf, m->type);
2708                         if (!found)
2709                                 spec->len--; /* pop accessor */
2710                         return found;
2711                 }
2712                 /* member turned out not to be what we looked for */
2713                 spec->offset -= offset / 8;
2714                 spec->raw_len--;
2715         }
2716
2717         return 0;
2718 }
2719
2720 /*
2721  * Try to match local spec to a target type and, if successful, produce full
2722  * target spec (high-level, low-level + offset).
2723  */
2724 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
2725                                const struct btf *targ_btf, __u32 targ_id,
2726                                struct bpf_core_spec *targ_spec)
2727 {
2728         const struct btf_type *targ_type;
2729         const struct bpf_core_accessor *local_acc;
2730         struct bpf_core_accessor *targ_acc;
2731         int i, sz, matched;
2732
2733         memset(targ_spec, 0, sizeof(*targ_spec));
2734         targ_spec->btf = targ_btf;
2735
2736         local_acc = &local_spec->spec[0];
2737         targ_acc = &targ_spec->spec[0];
2738
2739         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
2740                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
2741                                                    &targ_id);
2742                 if (!targ_type)
2743                         return -EINVAL;
2744
2745                 if (local_acc->name) {
2746                         matched = bpf_core_match_member(local_spec->btf,
2747                                                         local_acc,
2748                                                         targ_btf, targ_id,
2749                                                         targ_spec, &targ_id);
2750                         if (matched <= 0)
2751                                 return matched;
2752                 } else {
2753                         /* for i=0, targ_id is already treated as array element
2754                          * type (because it's the original struct), for others
2755                          * we should find array element type first
2756                          */
2757                         if (i > 0) {
2758                                 const struct btf_array *a;
2759
2760                                 if (!btf_is_array(targ_type))
2761                                         return 0;
2762
2763                                 a = btf_array(targ_type);
2764                                 if (local_acc->idx >= a->nelems)
2765                                         return 0;
2766                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
2767                                                             &targ_id))
2768                                         return -EINVAL;
2769                         }
2770
2771                         /* too deep struct/union/array nesting */
2772                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
2773                                 return -E2BIG;
2774
2775                         targ_acc->type_id = targ_id;
2776                         targ_acc->idx = local_acc->idx;
2777                         targ_acc->name = NULL;
2778                         targ_spec->len++;
2779                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
2780                         targ_spec->raw_len++;
2781
2782                         sz = btf__resolve_size(targ_btf, targ_id);
2783                         if (sz < 0)
2784                                 return sz;
2785                         targ_spec->offset += local_acc->idx * sz;
2786                 }
2787         }
2788
2789         return 1;
2790 }
2791
2792 /*
2793  * Patch relocatable BPF instruction.
2794  *
2795  * Patched value is determined by relocation kind and target specification.
2796  * For field existence relocation target spec will be NULL if field is not
2797  * found.
2798  * Expected insn->imm value is determined using relocation kind and local
2799  * spec, and is checked before patching instruction. If actual insn->imm value
2800  * is wrong, bail out with error.
2801  *
2802  * Currently three kinds of BPF instructions are supported:
2803  * 1. rX = <imm> (assignment with immediate operand);
2804  * 2. rX += <imm> (arithmetic operations with immediate operand);
2805  */
2806 static int bpf_core_reloc_insn(struct bpf_program *prog,
2807                                const struct bpf_field_reloc *relo,
2808                                const struct bpf_core_spec *local_spec,
2809                                const struct bpf_core_spec *targ_spec)
2810 {
2811         __u32 orig_val, new_val;
2812         struct bpf_insn *insn;
2813         int insn_idx;
2814         __u8 class;
2815
2816         if (relo->insn_off % sizeof(struct bpf_insn))
2817                 return -EINVAL;
2818         insn_idx = relo->insn_off / sizeof(struct bpf_insn);
2819
2820         switch (relo->kind) {
2821         case BPF_FIELD_BYTE_OFFSET:
2822                 orig_val = local_spec->offset;
2823                 if (targ_spec) {
2824                         new_val = targ_spec->offset;
2825                 } else {
2826                         pr_warning("prog '%s': patching insn #%d w/ failed reloc, imm %d -> %d\n",
2827                                    bpf_program__title(prog, false), insn_idx,
2828                                    orig_val, -1);
2829                         new_val = (__u32)-1;
2830                 }
2831                 break;
2832         case BPF_FIELD_EXISTS:
2833                 orig_val = 1; /* can't generate EXISTS relo w/o local field */
2834                 new_val = targ_spec ? 1 : 0;
2835                 break;
2836         default:
2837                 pr_warning("prog '%s': unknown relo %d at insn #%d'\n",
2838                            bpf_program__title(prog, false),
2839                            relo->kind, insn_idx);
2840                 return -EINVAL;
2841         }
2842
2843         insn = &prog->insns[insn_idx];
2844         class = BPF_CLASS(insn->code);
2845
2846         if (class == BPF_ALU || class == BPF_ALU64) {
2847                 if (BPF_SRC(insn->code) != BPF_K)
2848                         return -EINVAL;
2849                 if (insn->imm != orig_val)
2850                         return -EINVAL;
2851                 insn->imm = new_val;
2852                 pr_debug("prog '%s': patched insn #%d (ALU/ALU64) imm %d -> %d\n",
2853                          bpf_program__title(prog, false),
2854                          insn_idx, orig_val, new_val);
2855         } else {
2856                 pr_warning("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
2857                            bpf_program__title(prog, false),
2858                            insn_idx, insn->code, insn->src_reg, insn->dst_reg,
2859                            insn->off, insn->imm);
2860                 return -EINVAL;
2861         }
2862
2863         return 0;
2864 }
2865
2866 static struct btf *btf_load_raw(const char *path)
2867 {
2868         struct btf *btf;
2869         size_t read_cnt;
2870         struct stat st;
2871         void *data;
2872         FILE *f;
2873
2874         if (stat(path, &st))
2875                 return ERR_PTR(-errno);
2876
2877         data = malloc(st.st_size);
2878         if (!data)
2879                 return ERR_PTR(-ENOMEM);
2880
2881         f = fopen(path, "rb");
2882         if (!f) {
2883                 btf = ERR_PTR(-errno);
2884                 goto cleanup;
2885         }
2886
2887         read_cnt = fread(data, 1, st.st_size, f);
2888         fclose(f);
2889         if (read_cnt < st.st_size) {
2890                 btf = ERR_PTR(-EBADF);
2891                 goto cleanup;
2892         }
2893
2894         btf = btf__new(data, read_cnt);
2895
2896 cleanup:
2897         free(data);
2898         return btf;
2899 }
2900
2901 /*
2902  * Probe few well-known locations for vmlinux kernel image and try to load BTF
2903  * data out of it to use for target BTF.
2904  */
2905 static struct btf *bpf_core_find_kernel_btf(void)
2906 {
2907         struct {
2908                 const char *path_fmt;
2909                 bool raw_btf;
2910         } locations[] = {
2911                 /* try canonical vmlinux BTF through sysfs first */
2912                 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
2913                 /* fall back to trying to find vmlinux ELF on disk otherwise */
2914                 { "/boot/vmlinux-%1$s" },
2915                 { "/lib/modules/%1$s/vmlinux-%1$s" },
2916                 { "/lib/modules/%1$s/build/vmlinux" },
2917                 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
2918                 { "/usr/lib/debug/boot/vmlinux-%1$s" },
2919                 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
2920                 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
2921         };
2922         char path[PATH_MAX + 1];
2923         struct utsname buf;
2924         struct btf *btf;
2925         int i;
2926
2927         uname(&buf);
2928
2929         for (i = 0; i < ARRAY_SIZE(locations); i++) {
2930                 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
2931
2932                 if (access(path, R_OK))
2933                         continue;
2934
2935                 if (locations[i].raw_btf)
2936                         btf = btf_load_raw(path);
2937                 else
2938                         btf = btf__parse_elf(path, NULL);
2939
2940                 pr_debug("loading kernel BTF '%s': %ld\n",
2941                          path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
2942                 if (IS_ERR(btf))
2943                         continue;
2944
2945                 return btf;
2946         }
2947
2948         pr_warning("failed to find valid kernel BTF\n");
2949         return ERR_PTR(-ESRCH);
2950 }
2951
2952 /* Output spec definition in the format:
2953  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
2954  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
2955  */
2956 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
2957 {
2958         const struct btf_type *t;
2959         const char *s;
2960         __u32 type_id;
2961         int i;
2962
2963         type_id = spec->spec[0].type_id;
2964         t = btf__type_by_id(spec->btf, type_id);
2965         s = btf__name_by_offset(spec->btf, t->name_off);
2966         libbpf_print(level, "[%u] %s + ", type_id, s);
2967
2968         for (i = 0; i < spec->raw_len; i++)
2969                 libbpf_print(level, "%d%s", spec->raw_spec[i],
2970                              i == spec->raw_len - 1 ? " => " : ":");
2971
2972         libbpf_print(level, "%u @ &x", spec->offset);
2973
2974         for (i = 0; i < spec->len; i++) {
2975                 if (spec->spec[i].name)
2976                         libbpf_print(level, ".%s", spec->spec[i].name);
2977                 else
2978                         libbpf_print(level, "[%u]", spec->spec[i].idx);
2979         }
2980
2981 }
2982
2983 static size_t bpf_core_hash_fn(const void *key, void *ctx)
2984 {
2985         return (size_t)key;
2986 }
2987
2988 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
2989 {
2990         return k1 == k2;
2991 }
2992
2993 static void *u32_as_hash_key(__u32 x)
2994 {
2995         return (void *)(uintptr_t)x;
2996 }
2997
2998 /*
2999  * CO-RE relocate single instruction.
3000  *
3001  * The outline and important points of the algorithm:
3002  * 1. For given local type, find corresponding candidate target types.
3003  *    Candidate type is a type with the same "essential" name, ignoring
3004  *    everything after last triple underscore (___). E.g., `sample`,
3005  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
3006  *    for each other. Names with triple underscore are referred to as
3007  *    "flavors" and are useful, among other things, to allow to
3008  *    specify/support incompatible variations of the same kernel struct, which
3009  *    might differ between different kernel versions and/or build
3010  *    configurations.
3011  *
3012  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
3013  *    converter, when deduplicated BTF of a kernel still contains more than
3014  *    one different types with the same name. In that case, ___2, ___3, etc
3015  *    are appended starting from second name conflict. But start flavors are
3016  *    also useful to be defined "locally", in BPF program, to extract same
3017  *    data from incompatible changes between different kernel
3018  *    versions/configurations. For instance, to handle field renames between
3019  *    kernel versions, one can use two flavors of the struct name with the
3020  *    same common name and use conditional relocations to extract that field,
3021  *    depending on target kernel version.
3022  * 2. For each candidate type, try to match local specification to this
3023  *    candidate target type. Matching involves finding corresponding
3024  *    high-level spec accessors, meaning that all named fields should match,
3025  *    as well as all array accesses should be within the actual bounds. Also,
3026  *    types should be compatible (see bpf_core_fields_are_compat for details).
3027  * 3. It is supported and expected that there might be multiple flavors
3028  *    matching the spec. As long as all the specs resolve to the same set of
3029  *    offsets across all candidates, there is no error. If there is any
3030  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
3031  *    imprefection of BTF deduplication, which can cause slight duplication of
3032  *    the same BTF type, if some directly or indirectly referenced (by
3033  *    pointer) type gets resolved to different actual types in different
3034  *    object files. If such situation occurs, deduplicated BTF will end up
3035  *    with two (or more) structurally identical types, which differ only in
3036  *    types they refer to through pointer. This should be OK in most cases and
3037  *    is not an error.
3038  * 4. Candidate types search is performed by linearly scanning through all
3039  *    types in target BTF. It is anticipated that this is overall more
3040  *    efficient memory-wise and not significantly worse (if not better)
3041  *    CPU-wise compared to prebuilding a map from all local type names to
3042  *    a list of candidate type names. It's also sped up by caching resolved
3043  *    list of matching candidates per each local "root" type ID, that has at
3044  *    least one bpf_field_reloc associated with it. This list is shared
3045  *    between multiple relocations for the same type ID and is updated as some
3046  *    of the candidates are pruned due to structural incompatibility.
3047  */
3048 static int bpf_core_reloc_field(struct bpf_program *prog,
3049                                  const struct bpf_field_reloc *relo,
3050                                  int relo_idx,
3051                                  const struct btf *local_btf,
3052                                  const struct btf *targ_btf,
3053                                  struct hashmap *cand_cache)
3054 {
3055         const char *prog_name = bpf_program__title(prog, false);
3056         struct bpf_core_spec local_spec, cand_spec, targ_spec;
3057         const void *type_key = u32_as_hash_key(relo->type_id);
3058         const struct btf_type *local_type, *cand_type;
3059         const char *local_name, *cand_name;
3060         struct ids_vec *cand_ids;
3061         __u32 local_id, cand_id;
3062         const char *spec_str;
3063         int i, j, err;
3064
3065         local_id = relo->type_id;
3066         local_type = btf__type_by_id(local_btf, local_id);
3067         if (!local_type)
3068                 return -EINVAL;
3069
3070         local_name = btf__name_by_offset(local_btf, local_type->name_off);
3071         if (str_is_empty(local_name))
3072                 return -EINVAL;
3073
3074         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
3075         if (str_is_empty(spec_str))
3076                 return -EINVAL;
3077
3078         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
3079         if (err) {
3080                 pr_warning("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
3081                            prog_name, relo_idx, local_id, local_name, spec_str,
3082                            err);
3083                 return -EINVAL;
3084         }
3085
3086         pr_debug("prog '%s': relo #%d: spec is ", prog_name, relo_idx);
3087         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
3088         libbpf_print(LIBBPF_DEBUG, "\n");
3089
3090         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
3091                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
3092                 if (IS_ERR(cand_ids)) {
3093                         pr_warning("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
3094                                    prog_name, relo_idx, local_id, local_name,
3095                                    PTR_ERR(cand_ids));
3096                         return PTR_ERR(cand_ids);
3097                 }
3098                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
3099                 if (err) {
3100                         bpf_core_free_cands(cand_ids);
3101                         return err;
3102                 }
3103         }
3104
3105         for (i = 0, j = 0; i < cand_ids->len; i++) {
3106                 cand_id = cand_ids->data[i];
3107                 cand_type = btf__type_by_id(targ_btf, cand_id);
3108                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
3109
3110                 err = bpf_core_spec_match(&local_spec, targ_btf,
3111                                           cand_id, &cand_spec);
3112                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
3113                          prog_name, relo_idx, i, cand_name);
3114                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
3115                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
3116                 if (err < 0) {
3117                         pr_warning("prog '%s': relo #%d: matching error: %d\n",
3118                                    prog_name, relo_idx, err);
3119                         return err;
3120                 }
3121                 if (err == 0)
3122                         continue;
3123
3124                 if (j == 0) {
3125                         targ_spec = cand_spec;
3126                 } else if (cand_spec.offset != targ_spec.offset) {
3127                         /* if there are many candidates, they should all
3128                          * resolve to the same offset
3129                          */
3130                         pr_warning("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
3131                                    prog_name, relo_idx, cand_spec.offset,
3132                                    targ_spec.offset);
3133                         return -EINVAL;
3134                 }
3135
3136                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
3137         }
3138
3139         /*
3140          * For BPF_FIELD_EXISTS relo or when relaxed CO-RE reloc mode is
3141          * requested, it's expected that we might not find any candidates.
3142          * In this case, if field wasn't found in any candidate, the list of
3143          * candidates shouldn't change at all, we'll just handle relocating
3144          * appropriately, depending on relo's kind.
3145          */
3146         if (j > 0)
3147                 cand_ids->len = j;
3148
3149         if (j == 0 && !prog->obj->relaxed_core_relocs &&
3150             relo->kind != BPF_FIELD_EXISTS) {
3151                 pr_warning("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
3152                            prog_name, relo_idx, local_id, local_name, spec_str);
3153                 return -ESRCH;
3154         }
3155
3156         /* bpf_core_reloc_insn should know how to handle missing targ_spec */
3157         err = bpf_core_reloc_insn(prog, relo, &local_spec,
3158                                   j ? &targ_spec : NULL);
3159         if (err) {
3160                 pr_warning("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
3161                            prog_name, relo_idx, relo->insn_off, err);
3162                 return -EINVAL;
3163         }
3164
3165         return 0;
3166 }
3167
3168 static int
3169 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
3170 {
3171         const struct btf_ext_info_sec *sec;
3172         const struct bpf_field_reloc *rec;
3173         const struct btf_ext_info *seg;
3174         struct hashmap_entry *entry;
3175         struct hashmap *cand_cache = NULL;
3176         struct bpf_program *prog;
3177         struct btf *targ_btf;
3178         const char *sec_name;
3179         int i, err = 0;
3180
3181         if (targ_btf_path)
3182                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
3183         else
3184                 targ_btf = bpf_core_find_kernel_btf();
3185         if (IS_ERR(targ_btf)) {
3186                 pr_warning("failed to get target BTF: %ld\n",
3187                            PTR_ERR(targ_btf));
3188                 return PTR_ERR(targ_btf);
3189         }
3190
3191         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
3192         if (IS_ERR(cand_cache)) {
3193                 err = PTR_ERR(cand_cache);
3194                 goto out;
3195         }
3196
3197         seg = &obj->btf_ext->field_reloc_info;
3198         for_each_btf_ext_sec(seg, sec) {
3199                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3200                 if (str_is_empty(sec_name)) {
3201                         err = -EINVAL;
3202                         goto out;
3203                 }
3204                 prog = bpf_object__find_program_by_title(obj, sec_name);
3205                 if (!prog) {
3206                         pr_warning("failed to find program '%s' for CO-RE offset relocation\n",
3207                                    sec_name);
3208                         err = -EINVAL;
3209                         goto out;
3210                 }
3211
3212                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
3213                          sec_name, sec->num_info);
3214
3215                 for_each_btf_ext_rec(seg, sec, i, rec) {
3216                         err = bpf_core_reloc_field(prog, rec, i, obj->btf,
3217                                                    targ_btf, cand_cache);
3218                         if (err) {
3219                                 pr_warning("prog '%s': relo #%d: failed to relocate: %d\n",
3220                                            sec_name, i, err);
3221                                 goto out;
3222                         }
3223                 }
3224         }
3225
3226 out:
3227         btf__free(targ_btf);
3228         if (!IS_ERR_OR_NULL(cand_cache)) {
3229                 hashmap__for_each_entry(cand_cache, entry, i) {
3230                         bpf_core_free_cands(entry->value);
3231                 }
3232                 hashmap__free(cand_cache);
3233         }
3234         return err;
3235 }
3236
3237 static int
3238 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
3239 {
3240         int err = 0;
3241
3242         if (obj->btf_ext->field_reloc_info.len)
3243                 err = bpf_core_reloc_fields(obj, targ_btf_path);
3244
3245         return err;
3246 }
3247
3248 static int
3249 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
3250                         struct reloc_desc *relo)
3251 {
3252         struct bpf_insn *insn, *new_insn;
3253         struct bpf_program *text;
3254         size_t new_cnt;
3255         int err;
3256
3257         if (relo->type != RELO_CALL)
3258                 return -LIBBPF_ERRNO__RELOC;
3259
3260         if (prog->idx == obj->efile.text_shndx) {
3261                 pr_warning("relo in .text insn %d into off %d\n",
3262                            relo->insn_idx, relo->text_off);
3263                 return -LIBBPF_ERRNO__RELOC;
3264         }
3265
3266         if (prog->main_prog_cnt == 0) {
3267                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
3268                 if (!text) {
3269                         pr_warning("no .text section found yet relo into text exist\n");
3270                         return -LIBBPF_ERRNO__RELOC;
3271                 }
3272                 new_cnt = prog->insns_cnt + text->insns_cnt;
3273                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
3274                 if (!new_insn) {
3275                         pr_warning("oom in prog realloc\n");
3276                         return -ENOMEM;
3277                 }
3278
3279                 if (obj->btf_ext) {
3280                         err = bpf_program_reloc_btf_ext(prog, obj,
3281                                                         text->section_name,
3282                                                         prog->insns_cnt);
3283                         if (err)
3284                                 return err;
3285                 }
3286
3287                 memcpy(new_insn + prog->insns_cnt, text->insns,
3288                        text->insns_cnt * sizeof(*insn));
3289                 prog->insns = new_insn;
3290                 prog->main_prog_cnt = prog->insns_cnt;
3291                 prog->insns_cnt = new_cnt;
3292                 pr_debug("added %zd insn from %s to prog %s\n",
3293                          text->insns_cnt, text->section_name,
3294                          prog->section_name);
3295         }
3296         insn = &prog->insns[relo->insn_idx];
3297         insn->imm += prog->main_prog_cnt - relo->insn_idx;
3298         return 0;
3299 }
3300
3301 static int
3302 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
3303 {
3304         int i, err;
3305
3306         if (!prog)
3307                 return 0;
3308
3309         if (obj->btf_ext) {
3310                 err = bpf_program_reloc_btf_ext(prog, obj,
3311                                                 prog->section_name, 0);
3312                 if (err)
3313                         return err;
3314         }
3315
3316         if (!prog->reloc_desc)
3317                 return 0;
3318
3319         for (i = 0; i < prog->nr_reloc; i++) {
3320                 if (prog->reloc_desc[i].type == RELO_LD64 ||
3321                     prog->reloc_desc[i].type == RELO_DATA) {
3322                         bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
3323                         struct bpf_insn *insns = prog->insns;
3324                         int insn_idx, map_idx;
3325
3326                         insn_idx = prog->reloc_desc[i].insn_idx;
3327                         map_idx = prog->reloc_desc[i].map_idx;
3328
3329                         if (insn_idx + 1 >= (int)prog->insns_cnt) {
3330                                 pr_warning("relocation out of range: '%s'\n",
3331                                            prog->section_name);
3332                                 return -LIBBPF_ERRNO__RELOC;
3333                         }
3334
3335                         if (!relo_data) {
3336                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
3337                         } else {
3338                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
3339                                 insns[insn_idx + 1].imm = insns[insn_idx].imm;
3340                         }
3341                         insns[insn_idx].imm = obj->maps[map_idx].fd;
3342                 } else if (prog->reloc_desc[i].type == RELO_CALL) {
3343                         err = bpf_program__reloc_text(prog, obj,
3344                                                       &prog->reloc_desc[i]);
3345                         if (err)
3346                                 return err;
3347                 }
3348         }
3349
3350         zfree(&prog->reloc_desc);
3351         prog->nr_reloc = 0;
3352         return 0;
3353 }
3354
3355 static int
3356 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
3357 {
3358         struct bpf_program *prog;
3359         size_t i;
3360         int err;
3361
3362         if (obj->btf_ext) {
3363                 err = bpf_object__relocate_core(obj, targ_btf_path);
3364                 if (err) {
3365                         pr_warning("failed to perform CO-RE relocations: %d\n",
3366                                    err);
3367                         return err;
3368                 }
3369         }
3370         for (i = 0; i < obj->nr_programs; i++) {
3371                 prog = &obj->programs[i];
3372
3373                 err = bpf_program__relocate(prog, obj);
3374                 if (err) {
3375                         pr_warning("failed to relocate '%s'\n",
3376                                    prog->section_name);
3377                         return err;
3378                 }
3379         }
3380         return 0;
3381 }
3382
3383 static int bpf_object__collect_reloc(struct bpf_object *obj)
3384 {
3385         int i, err;
3386
3387         if (!obj_elf_valid(obj)) {
3388                 pr_warning("Internal error: elf object is closed\n");
3389                 return -LIBBPF_ERRNO__INTERNAL;
3390         }
3391
3392         for (i = 0; i < obj->efile.nr_reloc; i++) {
3393                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
3394                 Elf_Data *data = obj->efile.reloc[i].data;
3395                 int idx = shdr->sh_info;
3396                 struct bpf_program *prog;
3397
3398                 if (shdr->sh_type != SHT_REL) {
3399                         pr_warning("internal error at %d\n", __LINE__);
3400                         return -LIBBPF_ERRNO__INTERNAL;
3401                 }
3402
3403                 prog = bpf_object__find_prog_by_idx(obj, idx);
3404                 if (!prog) {
3405                         pr_warning("relocation failed: no section(%d)\n", idx);
3406                         return -LIBBPF_ERRNO__RELOC;
3407                 }
3408
3409                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
3410                 if (err)
3411                         return err;
3412         }
3413         return 0;
3414 }
3415
3416 static int
3417 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
3418              char *license, __u32 kern_version, int *pfd)
3419 {
3420         struct bpf_load_program_attr load_attr;
3421         char *cp, errmsg[STRERR_BUFSIZE];
3422         int log_buf_size = BPF_LOG_BUF_SIZE;
3423         char *log_buf;
3424         int btf_fd, ret;
3425
3426         if (!insns || !insns_cnt)
3427                 return -EINVAL;
3428
3429         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
3430         load_attr.prog_type = prog->type;
3431         load_attr.expected_attach_type = prog->expected_attach_type;
3432         if (prog->caps->name)
3433                 load_attr.name = prog->name;
3434         load_attr.insns = insns;
3435         load_attr.insns_cnt = insns_cnt;
3436         load_attr.license = license;
3437         load_attr.kern_version = kern_version;
3438         load_attr.prog_ifindex = prog->prog_ifindex;
3439         /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
3440         if (prog->obj->btf_ext)
3441                 btf_fd = bpf_object__btf_fd(prog->obj);
3442         else
3443                 btf_fd = -1;
3444         load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
3445         load_attr.func_info = prog->func_info;
3446         load_attr.func_info_rec_size = prog->func_info_rec_size;
3447         load_attr.func_info_cnt = prog->func_info_cnt;
3448         load_attr.line_info = prog->line_info;
3449         load_attr.line_info_rec_size = prog->line_info_rec_size;
3450         load_attr.line_info_cnt = prog->line_info_cnt;
3451         load_attr.log_level = prog->log_level;
3452         load_attr.prog_flags = prog->prog_flags;
3453
3454 retry_load:
3455         log_buf = malloc(log_buf_size);
3456         if (!log_buf)
3457                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
3458
3459         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
3460
3461         if (ret >= 0) {
3462                 if (load_attr.log_level)
3463                         pr_debug("verifier log:\n%s", log_buf);
3464                 *pfd = ret;
3465                 ret = 0;
3466                 goto out;
3467         }
3468
3469         if (errno == ENOSPC) {
3470                 log_buf_size <<= 1;
3471                 free(log_buf);
3472                 goto retry_load;
3473         }
3474         ret = -LIBBPF_ERRNO__LOAD;
3475         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3476         pr_warning("load bpf program failed: %s\n", cp);
3477
3478         if (log_buf && log_buf[0] != '\0') {
3479                 ret = -LIBBPF_ERRNO__VERIFY;
3480                 pr_warning("-- BEGIN DUMP LOG ---\n");
3481                 pr_warning("\n%s\n", log_buf);
3482                 pr_warning("-- END LOG --\n");
3483         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
3484                 pr_warning("Program too large (%zu insns), at most %d insns\n",
3485                            load_attr.insns_cnt, BPF_MAXINSNS);
3486                 ret = -LIBBPF_ERRNO__PROG2BIG;
3487         } else {
3488                 /* Wrong program type? */
3489                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
3490                         int fd;
3491
3492                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
3493                         load_attr.expected_attach_type = 0;
3494                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
3495                         if (fd >= 0) {
3496                                 close(fd);
3497                                 ret = -LIBBPF_ERRNO__PROGTYPE;
3498                                 goto out;
3499                         }
3500                 }
3501
3502                 if (log_buf)
3503                         ret = -LIBBPF_ERRNO__KVER;
3504         }
3505
3506 out:
3507         free(log_buf);
3508         return ret;
3509 }
3510
3511 int
3512 bpf_program__load(struct bpf_program *prog,
3513                   char *license, __u32 kern_version)
3514 {
3515         int err = 0, fd, i;
3516
3517         if (prog->instances.nr < 0 || !prog->instances.fds) {
3518                 if (prog->preprocessor) {
3519                         pr_warning("Internal error: can't load program '%s'\n",
3520                                    prog->section_name);
3521                         return -LIBBPF_ERRNO__INTERNAL;
3522                 }
3523
3524                 prog->instances.fds = malloc(sizeof(int));
3525                 if (!prog->instances.fds) {
3526                         pr_warning("Not enough memory for BPF fds\n");
3527                         return -ENOMEM;
3528                 }
3529                 prog->instances.nr = 1;
3530                 prog->instances.fds[0] = -1;
3531         }
3532
3533         if (!prog->preprocessor) {
3534                 if (prog->instances.nr != 1) {
3535                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
3536                                    prog->section_name, prog->instances.nr);
3537                 }
3538                 err = load_program(prog, prog->insns, prog->insns_cnt,
3539                                    license, kern_version, &fd);
3540                 if (!err)
3541                         prog->instances.fds[0] = fd;
3542                 goto out;
3543         }
3544
3545         for (i = 0; i < prog->instances.nr; i++) {
3546                 struct bpf_prog_prep_result result;
3547                 bpf_program_prep_t preprocessor = prog->preprocessor;
3548
3549                 memset(&result, 0, sizeof(result));
3550                 err = preprocessor(prog, i, prog->insns,
3551                                    prog->insns_cnt, &result);
3552                 if (err) {
3553                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
3554                                    i, prog->section_name);
3555                         goto out;
3556                 }
3557
3558                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
3559                         pr_debug("Skip loading the %dth instance of program '%s'\n",
3560                                  i, prog->section_name);
3561                         prog->instances.fds[i] = -1;
3562                         if (result.pfd)
3563                                 *result.pfd = -1;
3564                         continue;
3565                 }
3566
3567                 err = load_program(prog, result.new_insn_ptr,
3568                                    result.new_insn_cnt,
3569                                    license, kern_version, &fd);
3570
3571                 if (err) {
3572                         pr_warning("Loading the %dth instance of program '%s' failed\n",
3573                                         i, prog->section_name);
3574                         goto out;
3575                 }
3576
3577                 if (result.pfd)
3578                         *result.pfd = fd;
3579                 prog->instances.fds[i] = fd;
3580         }
3581 out:
3582         if (err)
3583                 pr_warning("failed to load program '%s'\n",
3584                            prog->section_name);
3585         zfree(&prog->insns);
3586         prog->insns_cnt = 0;
3587         return err;
3588 }
3589
3590 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
3591                                              const struct bpf_object *obj)
3592 {
3593         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
3594 }
3595
3596 static int
3597 bpf_object__load_progs(struct bpf_object *obj, int log_level)
3598 {
3599         size_t i;
3600         int err;
3601
3602         for (i = 0; i < obj->nr_programs; i++) {
3603                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
3604                         continue;
3605                 obj->programs[i].log_level |= log_level;
3606                 err = bpf_program__load(&obj->programs[i],
3607                                         obj->license,
3608                                         obj->kern_version);
3609                 if (err)
3610                         return err;
3611         }
3612         return 0;
3613 }
3614
3615 static struct bpf_object *
3616 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
3617                    struct bpf_object_open_opts *opts)
3618 {
3619         struct bpf_object *obj;
3620         const char *obj_name;
3621         char tmp_name[64];
3622         bool relaxed_maps;
3623         int err;
3624
3625         if (elf_version(EV_CURRENT) == EV_NONE) {
3626                 pr_warning("failed to init libelf for %s\n",
3627                            path ? : "(mem buf)");
3628                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
3629         }
3630
3631         if (!OPTS_VALID(opts, bpf_object_open_opts))
3632                 return ERR_PTR(-EINVAL);
3633
3634         obj_name = OPTS_GET(opts, object_name, path);
3635         if (obj_buf) {
3636                 if (!obj_name) {
3637                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
3638                                  (unsigned long)obj_buf,
3639                                  (unsigned long)obj_buf_sz);
3640                         obj_name = tmp_name;
3641                 }
3642                 path = obj_name;
3643                 pr_debug("loading object '%s' from buffer\n", obj_name);
3644         }
3645
3646         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
3647         if (IS_ERR(obj))
3648                 return obj;
3649
3650         obj->relaxed_core_relocs = OPTS_GET(opts, relaxed_core_relocs, false);
3651         relaxed_maps = OPTS_GET(opts, relaxed_maps, false);
3652
3653         CHECK_ERR(bpf_object__elf_init(obj), err, out);
3654         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
3655         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
3656         CHECK_ERR(bpf_object__elf_collect(obj, relaxed_maps), err, out);
3657         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
3658
3659         bpf_object__elf_finish(obj);
3660         return obj;
3661 out:
3662         bpf_object__close(obj);
3663         return ERR_PTR(err);
3664 }
3665
3666 static struct bpf_object *
3667 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
3668 {
3669         LIBBPF_OPTS(bpf_object_open_opts, opts,
3670                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
3671         );
3672
3673         /* param validation */
3674         if (!attr->file)
3675                 return NULL;
3676
3677         pr_debug("loading %s\n", attr->file);
3678         return __bpf_object__open(attr->file, NULL, 0, &opts);
3679 }
3680
3681 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
3682 {
3683         return __bpf_object__open_xattr(attr, 0);
3684 }
3685
3686 struct bpf_object *bpf_object__open(const char *path)
3687 {
3688         struct bpf_object_open_attr attr = {
3689                 .file           = path,
3690                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
3691         };
3692
3693         return bpf_object__open_xattr(&attr);
3694 }
3695
3696 struct bpf_object *
3697 bpf_object__open_file(const char *path, struct bpf_object_open_opts *opts)
3698 {
3699         if (!path)
3700                 return ERR_PTR(-EINVAL);
3701
3702         pr_debug("loading %s\n", path);
3703
3704         return __bpf_object__open(path, NULL, 0, opts);
3705 }
3706
3707 struct bpf_object *
3708 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
3709                      struct bpf_object_open_opts *opts)
3710 {
3711         if (!obj_buf || obj_buf_sz == 0)
3712                 return ERR_PTR(-EINVAL);
3713
3714         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
3715 }
3716
3717 struct bpf_object *
3718 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
3719                         const char *name)
3720 {
3721         LIBBPF_OPTS(bpf_object_open_opts, opts,
3722                 .object_name = name,
3723                 /* wrong default, but backwards-compatible */
3724                 .relaxed_maps = true,
3725         );
3726
3727         /* returning NULL is wrong, but backwards-compatible */
3728         if (!obj_buf || obj_buf_sz == 0)
3729                 return NULL;
3730
3731         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
3732 }
3733
3734 int bpf_object__unload(struct bpf_object *obj)
3735 {
3736         size_t i;
3737
3738         if (!obj)
3739                 return -EINVAL;
3740
3741         for (i = 0; i < obj->nr_maps; i++)
3742                 zclose(obj->maps[i].fd);
3743
3744         for (i = 0; i < obj->nr_programs; i++)
3745                 bpf_program__unload(&obj->programs[i]);
3746
3747         return 0;
3748 }
3749
3750 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
3751 {
3752         struct bpf_object *obj;
3753         int err;
3754
3755         if (!attr)
3756                 return -EINVAL;
3757         obj = attr->obj;
3758         if (!obj)
3759                 return -EINVAL;
3760
3761         if (obj->loaded) {
3762                 pr_warning("object should not be loaded twice\n");
3763                 return -EINVAL;
3764         }
3765
3766         obj->loaded = true;
3767
3768         CHECK_ERR(bpf_object__create_maps(obj), err, out);
3769         CHECK_ERR(bpf_object__relocate(obj, attr->target_btf_path), err, out);
3770         CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
3771
3772         return 0;
3773 out:
3774         bpf_object__unload(obj);
3775         pr_warning("failed to load object '%s'\n", obj->path);
3776         return err;
3777 }
3778
3779 int bpf_object__load(struct bpf_object *obj)
3780 {
3781         struct bpf_object_load_attr attr = {
3782                 .obj = obj,
3783         };
3784
3785         return bpf_object__load_xattr(&attr);
3786 }
3787
3788 static int check_path(const char *path)
3789 {
3790         char *cp, errmsg[STRERR_BUFSIZE];
3791         struct statfs st_fs;
3792         char *dname, *dir;
3793         int err = 0;
3794
3795         if (path == NULL)
3796                 return -EINVAL;
3797
3798         dname = strdup(path);
3799         if (dname == NULL)
3800                 return -ENOMEM;
3801
3802         dir = dirname(dname);
3803         if (statfs(dir, &st_fs)) {
3804                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3805                 pr_warning("failed to statfs %s: %s\n", dir, cp);
3806                 err = -errno;
3807         }
3808         free(dname);
3809
3810         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
3811                 pr_warning("specified path %s is not on BPF FS\n", path);
3812                 err = -EINVAL;
3813         }
3814
3815         return err;
3816 }
3817
3818 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
3819                               int instance)
3820 {
3821         char *cp, errmsg[STRERR_BUFSIZE];
3822         int err;
3823
3824         err = check_path(path);
3825         if (err)
3826                 return err;
3827
3828         if (prog == NULL) {
3829                 pr_warning("invalid program pointer\n");
3830                 return -EINVAL;
3831         }
3832
3833         if (instance < 0 || instance >= prog->instances.nr) {
3834                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3835                            instance, prog->section_name, prog->instances.nr);
3836                 return -EINVAL;
3837         }
3838
3839         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
3840                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3841                 pr_warning("failed to pin program: %s\n", cp);
3842                 return -errno;
3843         }
3844         pr_debug("pinned program '%s'\n", path);
3845
3846         return 0;
3847 }
3848
3849 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
3850                                 int instance)
3851 {
3852         int err;
3853
3854         err = check_path(path);
3855         if (err)
3856                 return err;
3857
3858         if (prog == NULL) {
3859                 pr_warning("invalid program pointer\n");
3860                 return -EINVAL;
3861         }
3862
3863         if (instance < 0 || instance >= prog->instances.nr) {
3864                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
3865                            instance, prog->section_name, prog->instances.nr);
3866                 return -EINVAL;
3867         }
3868
3869         err = unlink(path);
3870         if (err != 0)
3871                 return -errno;
3872         pr_debug("unpinned program '%s'\n", path);
3873
3874         return 0;
3875 }
3876
3877 static int make_dir(const char *path)
3878 {
3879         char *cp, errmsg[STRERR_BUFSIZE];
3880         int err = 0;
3881
3882         if (mkdir(path, 0700) && errno != EEXIST)
3883                 err = -errno;
3884
3885         if (err) {
3886                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
3887                 pr_warning("failed to mkdir %s: %s\n", path, cp);
3888         }
3889         return err;
3890 }
3891
3892 int bpf_program__pin(struct bpf_program *prog, const char *path)
3893 {
3894         int i, err;
3895
3896         err = check_path(path);
3897         if (err)
3898                 return err;
3899
3900         if (prog == NULL) {
3901                 pr_warning("invalid program pointer\n");
3902                 return -EINVAL;
3903         }
3904
3905         if (prog->instances.nr <= 0) {
3906                 pr_warning("no instances of prog %s to pin\n",
3907                            prog->section_name);
3908                 return -EINVAL;
3909         }
3910
3911         if (prog->instances.nr == 1) {
3912                 /* don't create subdirs when pinning single instance */
3913                 return bpf_program__pin_instance(prog, path, 0);
3914         }
3915
3916         err = make_dir(path);
3917         if (err)
3918                 return err;
3919
3920         for (i = 0; i < prog->instances.nr; i++) {
3921                 char buf[PATH_MAX];
3922                 int len;
3923
3924                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3925                 if (len < 0) {
3926                         err = -EINVAL;
3927                         goto err_unpin;
3928                 } else if (len >= PATH_MAX) {
3929                         err = -ENAMETOOLONG;
3930                         goto err_unpin;
3931                 }
3932
3933                 err = bpf_program__pin_instance(prog, buf, i);
3934                 if (err)
3935                         goto err_unpin;
3936         }
3937
3938         return 0;
3939
3940 err_unpin:
3941         for (i = i - 1; i >= 0; i--) {
3942                 char buf[PATH_MAX];
3943                 int len;
3944
3945                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3946                 if (len < 0)
3947                         continue;
3948                 else if (len >= PATH_MAX)
3949                         continue;
3950
3951                 bpf_program__unpin_instance(prog, buf, i);
3952         }
3953
3954         rmdir(path);
3955
3956         return err;
3957 }
3958
3959 int bpf_program__unpin(struct bpf_program *prog, const char *path)
3960 {
3961         int i, err;
3962
3963         err = check_path(path);
3964         if (err)
3965                 return err;
3966
3967         if (prog == NULL) {
3968                 pr_warning("invalid program pointer\n");
3969                 return -EINVAL;
3970         }
3971
3972         if (prog->instances.nr <= 0) {
3973                 pr_warning("no instances of prog %s to pin\n",
3974                            prog->section_name);
3975                 return -EINVAL;
3976         }
3977
3978         if (prog->instances.nr == 1) {
3979                 /* don't create subdirs when pinning single instance */
3980                 return bpf_program__unpin_instance(prog, path, 0);
3981         }
3982
3983         for (i = 0; i < prog->instances.nr; i++) {
3984                 char buf[PATH_MAX];
3985                 int len;
3986
3987                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3988                 if (len < 0)
3989                         return -EINVAL;
3990                 else if (len >= PATH_MAX)
3991                         return -ENAMETOOLONG;
3992
3993                 err = bpf_program__unpin_instance(prog, buf, i);
3994                 if (err)
3995                         return err;
3996         }
3997
3998         err = rmdir(path);
3999         if (err)
4000                 return -errno;
4001
4002         return 0;
4003 }
4004
4005 int bpf_map__pin(struct bpf_map *map, const char *path)
4006 {
4007         char *cp, errmsg[STRERR_BUFSIZE];
4008         int err;
4009
4010         err = check_path(path);
4011         if (err)
4012                 return err;
4013
4014         if (map == NULL) {
4015                 pr_warning("invalid map pointer\n");
4016                 return -EINVAL;
4017         }
4018
4019         if (bpf_obj_pin(map->fd, path)) {
4020                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4021                 pr_warning("failed to pin map: %s\n", cp);
4022                 return -errno;
4023         }
4024
4025         pr_debug("pinned map '%s'\n", path);
4026
4027         return 0;
4028 }
4029
4030 int bpf_map__unpin(struct bpf_map *map, const char *path)
4031 {
4032         int err;
4033
4034         err = check_path(path);
4035         if (err)
4036                 return err;
4037
4038         if (map == NULL) {
4039                 pr_warning("invalid map pointer\n");
4040                 return -EINVAL;
4041         }
4042
4043         err = unlink(path);
4044         if (err != 0)
4045                 return -errno;
4046         pr_debug("unpinned map '%s'\n", path);
4047
4048         return 0;
4049 }
4050
4051 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
4052 {
4053         struct bpf_map *map;
4054         int err;
4055
4056         if (!obj)
4057                 return -ENOENT;
4058
4059         if (!obj->loaded) {
4060                 pr_warning("object not yet loaded; load it first\n");
4061                 return -ENOENT;
4062         }
4063
4064         err = make_dir(path);
4065         if (err)
4066                 return err;
4067
4068         bpf_object__for_each_map(map, obj) {
4069                 char buf[PATH_MAX];
4070                 int len;
4071
4072                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4073                                bpf_map__name(map));
4074                 if (len < 0) {
4075                         err = -EINVAL;
4076                         goto err_unpin_maps;
4077                 } else if (len >= PATH_MAX) {
4078                         err = -ENAMETOOLONG;
4079                         goto err_unpin_maps;
4080                 }
4081
4082                 err = bpf_map__pin(map, buf);
4083                 if (err)
4084                         goto err_unpin_maps;
4085         }
4086
4087         return 0;
4088
4089 err_unpin_maps:
4090         while ((map = bpf_map__prev(map, obj))) {
4091                 char buf[PATH_MAX];
4092                 int len;
4093
4094                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4095                                bpf_map__name(map));
4096                 if (len < 0)
4097                         continue;
4098                 else if (len >= PATH_MAX)
4099                         continue;
4100
4101                 bpf_map__unpin(map, buf);
4102         }
4103
4104         return err;
4105 }
4106
4107 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
4108 {
4109         struct bpf_map *map;
4110         int err;
4111
4112         if (!obj)
4113                 return -ENOENT;
4114
4115         bpf_object__for_each_map(map, obj) {
4116                 char buf[PATH_MAX];
4117                 int len;
4118
4119                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4120                                bpf_map__name(map));
4121                 if (len < 0)
4122                         return -EINVAL;
4123                 else if (len >= PATH_MAX)
4124                         return -ENAMETOOLONG;
4125
4126                 err = bpf_map__unpin(map, buf);
4127                 if (err)
4128                         return err;
4129         }
4130
4131         return 0;
4132 }
4133
4134 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
4135 {
4136         struct bpf_program *prog;
4137         int err;
4138
4139         if (!obj)
4140                 return -ENOENT;
4141
4142         if (!obj->loaded) {
4143                 pr_warning("object not yet loaded; load it first\n");
4144                 return -ENOENT;
4145         }
4146
4147         err = make_dir(path);
4148         if (err)
4149                 return err;
4150
4151         bpf_object__for_each_program(prog, obj) {
4152                 char buf[PATH_MAX];
4153                 int len;
4154
4155                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4156                                prog->pin_name);
4157                 if (len < 0) {
4158                         err = -EINVAL;
4159                         goto err_unpin_programs;
4160                 } else if (len >= PATH_MAX) {
4161                         err = -ENAMETOOLONG;
4162                         goto err_unpin_programs;
4163                 }
4164
4165                 err = bpf_program__pin(prog, buf);
4166                 if (err)
4167                         goto err_unpin_programs;
4168         }
4169
4170         return 0;
4171
4172 err_unpin_programs:
4173         while ((prog = bpf_program__prev(prog, obj))) {
4174                 char buf[PATH_MAX];
4175                 int len;
4176
4177                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4178                                prog->pin_name);
4179                 if (len < 0)
4180                         continue;
4181                 else if (len >= PATH_MAX)
4182                         continue;
4183
4184                 bpf_program__unpin(prog, buf);
4185         }
4186
4187         return err;
4188 }
4189
4190 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
4191 {
4192         struct bpf_program *prog;
4193         int err;
4194
4195         if (!obj)
4196                 return -ENOENT;
4197
4198         bpf_object__for_each_program(prog, obj) {
4199                 char buf[PATH_MAX];
4200                 int len;
4201
4202                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
4203                                prog->pin_name);
4204                 if (len < 0)
4205                         return -EINVAL;
4206                 else if (len >= PATH_MAX)
4207                         return -ENAMETOOLONG;
4208
4209                 err = bpf_program__unpin(prog, buf);
4210                 if (err)
4211                         return err;
4212         }
4213
4214         return 0;
4215 }
4216
4217 int bpf_object__pin(struct bpf_object *obj, const char *path)
4218 {
4219         int err;
4220
4221         err = bpf_object__pin_maps(obj, path);
4222         if (err)
4223                 return err;
4224
4225         err = bpf_object__pin_programs(obj, path);
4226         if (err) {
4227                 bpf_object__unpin_maps(obj, path);
4228                 return err;
4229         }
4230
4231         return 0;
4232 }
4233
4234 void bpf_object__close(struct bpf_object *obj)
4235 {
4236         size_t i;
4237
4238         if (!obj)
4239                 return;
4240
4241         if (obj->clear_priv)
4242                 obj->clear_priv(obj, obj->priv);
4243
4244         bpf_object__elf_finish(obj);
4245         bpf_object__unload(obj);
4246         btf__free(obj->btf);
4247         btf_ext__free(obj->btf_ext);
4248
4249         for (i = 0; i < obj->nr_maps; i++) {
4250                 zfree(&obj->maps[i].name);
4251                 if (obj->maps[i].clear_priv)
4252                         obj->maps[i].clear_priv(&obj->maps[i],
4253                                                 obj->maps[i].priv);
4254                 obj->maps[i].priv = NULL;
4255                 obj->maps[i].clear_priv = NULL;
4256         }
4257
4258         zfree(&obj->sections.rodata);
4259         zfree(&obj->sections.data);
4260         zfree(&obj->maps);
4261         obj->nr_maps = 0;
4262
4263         if (obj->programs && obj->nr_programs) {
4264                 for (i = 0; i < obj->nr_programs; i++)
4265                         bpf_program__exit(&obj->programs[i]);
4266         }
4267         zfree(&obj->programs);
4268
4269         list_del(&obj->list);
4270         free(obj);
4271 }
4272
4273 struct bpf_object *
4274 bpf_object__next(struct bpf_object *prev)
4275 {
4276         struct bpf_object *next;
4277
4278         if (!prev)
4279                 next = list_first_entry(&bpf_objects_list,
4280                                         struct bpf_object,
4281                                         list);
4282         else
4283                 next = list_next_entry(prev, list);
4284
4285         /* Empty list is noticed here so don't need checking on entry. */
4286         if (&next->list == &bpf_objects_list)
4287                 return NULL;
4288
4289         return next;
4290 }
4291
4292 const char *bpf_object__name(const struct bpf_object *obj)
4293 {
4294         return obj ? obj->name : ERR_PTR(-EINVAL);
4295 }
4296
4297 unsigned int bpf_object__kversion(const struct bpf_object *obj)
4298 {
4299         return obj ? obj->kern_version : 0;
4300 }
4301
4302 struct btf *bpf_object__btf(const struct bpf_object *obj)
4303 {
4304         return obj ? obj->btf : NULL;
4305 }
4306
4307 int bpf_object__btf_fd(const struct bpf_object *obj)
4308 {
4309         return obj->btf ? btf__fd(obj->btf) : -1;
4310 }
4311
4312 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
4313                          bpf_object_clear_priv_t clear_priv)
4314 {
4315         if (obj->priv && obj->clear_priv)
4316                 obj->clear_priv(obj, obj->priv);
4317
4318         obj->priv = priv;
4319         obj->clear_priv = clear_priv;
4320         return 0;
4321 }
4322
4323 void *bpf_object__priv(const struct bpf_object *obj)
4324 {
4325         return obj ? obj->priv : ERR_PTR(-EINVAL);
4326 }
4327
4328 static struct bpf_program *
4329 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
4330                     bool forward)
4331 {
4332         size_t nr_programs = obj->nr_programs;
4333         ssize_t idx;
4334
4335         if (!nr_programs)
4336                 return NULL;
4337
4338         if (!p)
4339                 /* Iter from the beginning */
4340                 return forward ? &obj->programs[0] :
4341                         &obj->programs[nr_programs - 1];
4342
4343         if (p->obj != obj) {
4344                 pr_warning("error: program handler doesn't match object\n");
4345                 return NULL;
4346         }
4347
4348         idx = (p - obj->programs) + (forward ? 1 : -1);
4349         if (idx >= obj->nr_programs || idx < 0)
4350                 return NULL;
4351         return &obj->programs[idx];
4352 }
4353
4354 struct bpf_program *
4355 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
4356 {
4357         struct bpf_program *prog = prev;
4358
4359         do {
4360                 prog = __bpf_program__iter(prog, obj, true);
4361         } while (prog && bpf_program__is_function_storage(prog, obj));
4362
4363         return prog;
4364 }
4365
4366 struct bpf_program *
4367 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
4368 {
4369         struct bpf_program *prog = next;
4370
4371         do {
4372                 prog = __bpf_program__iter(prog, obj, false);
4373         } while (prog && bpf_program__is_function_storage(prog, obj));
4374
4375         return prog;
4376 }
4377
4378 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
4379                           bpf_program_clear_priv_t clear_priv)
4380 {
4381         if (prog->priv && prog->clear_priv)
4382                 prog->clear_priv(prog, prog->priv);
4383
4384         prog->priv = priv;
4385         prog->clear_priv = clear_priv;
4386         return 0;
4387 }
4388
4389 void *bpf_program__priv(const struct bpf_program *prog)
4390 {
4391         return prog ? prog->priv : ERR_PTR(-EINVAL);
4392 }
4393
4394 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
4395 {
4396         prog->prog_ifindex = ifindex;
4397 }
4398
4399 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
4400 {
4401         const char *title;
4402
4403         title = prog->section_name;
4404         if (needs_copy) {
4405                 title = strdup(title);
4406                 if (!title) {
4407                         pr_warning("failed to strdup program title\n");
4408                         return ERR_PTR(-ENOMEM);
4409                 }
4410         }
4411
4412         return title;
4413 }
4414
4415 int bpf_program__fd(const struct bpf_program *prog)
4416 {
4417         return bpf_program__nth_fd(prog, 0);
4418 }
4419
4420 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
4421                           bpf_program_prep_t prep)
4422 {
4423         int *instances_fds;
4424
4425         if (nr_instances <= 0 || !prep)
4426                 return -EINVAL;
4427
4428         if (prog->instances.nr > 0 || prog->instances.fds) {
4429                 pr_warning("Can't set pre-processor after loading\n");
4430                 return -EINVAL;
4431         }
4432
4433         instances_fds = malloc(sizeof(int) * nr_instances);
4434         if (!instances_fds) {
4435                 pr_warning("alloc memory failed for fds\n");
4436                 return -ENOMEM;
4437         }
4438
4439         /* fill all fd with -1 */
4440         memset(instances_fds, -1, sizeof(int) * nr_instances);
4441
4442         prog->instances.nr = nr_instances;
4443         prog->instances.fds = instances_fds;
4444         prog->preprocessor = prep;
4445         return 0;
4446 }
4447
4448 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
4449 {
4450         int fd;
4451
4452         if (!prog)
4453                 return -EINVAL;
4454
4455         if (n >= prog->instances.nr || n < 0) {
4456                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
4457                            n, prog->section_name, prog->instances.nr);
4458                 return -EINVAL;
4459         }
4460
4461         fd = prog->instances.fds[n];
4462         if (fd < 0) {
4463                 pr_warning("%dth instance of program '%s' is invalid\n",
4464                            n, prog->section_name);
4465                 return -ENOENT;
4466         }
4467
4468         return fd;
4469 }
4470
4471 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
4472 {
4473         prog->type = type;
4474 }
4475
4476 static bool bpf_program__is_type(const struct bpf_program *prog,
4477                                  enum bpf_prog_type type)
4478 {
4479         return prog ? (prog->type == type) : false;
4480 }
4481
4482 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
4483 int bpf_program__set_##NAME(struct bpf_program *prog)           \
4484 {                                                               \
4485         if (!prog)                                              \
4486                 return -EINVAL;                                 \
4487         bpf_program__set_type(prog, TYPE);                      \
4488         return 0;                                               \
4489 }                                                               \
4490                                                                 \
4491 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
4492 {                                                               \
4493         return bpf_program__is_type(prog, TYPE);                \
4494 }                                                               \
4495
4496 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
4497 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
4498 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
4499 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
4500 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
4501 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
4502 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
4503 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
4504
4505 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
4506                                            enum bpf_attach_type type)
4507 {
4508         prog->expected_attach_type = type;
4509 }
4510
4511 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, btf, atype) \
4512         { string, sizeof(string) - 1, ptype, eatype, is_attachable, btf, atype }
4513
4514 /* Programs that can NOT be attached. */
4515 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
4516
4517 /* Programs that can be attached. */
4518 #define BPF_APROG_SEC(string, ptype, atype) \
4519         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, 0, atype)
4520
4521 /* Programs that must specify expected attach type at load time. */
4522 #define BPF_EAPROG_SEC(string, ptype, eatype) \
4523         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
4524
4525 /* Programs that use BTF to identify attach point */
4526 #define BPF_PROG_BTF(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 1, 0)
4527
4528 /* Programs that can be attached but attach type can't be identified by section
4529  * name. Kept for backward compatibility.
4530  */
4531 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
4532
4533 static const struct {
4534         const char *sec;
4535         size_t len;
4536         enum bpf_prog_type prog_type;
4537         enum bpf_attach_type expected_attach_type;
4538         bool is_attachable;
4539         bool is_attach_btf;
4540         enum bpf_attach_type attach_type;
4541 } section_names[] = {
4542         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
4543         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
4544         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
4545         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
4546         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
4547         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
4548         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
4549         BPF_PROG_BTF("tp_btf/",                 BPF_PROG_TYPE_RAW_TRACEPOINT),
4550         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
4551         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
4552         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
4553         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
4554         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
4555         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
4556         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
4557                                                 BPF_CGROUP_INET_INGRESS),
4558         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
4559                                                 BPF_CGROUP_INET_EGRESS),
4560         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
4561         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
4562                                                 BPF_CGROUP_INET_SOCK_CREATE),
4563         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
4564                                                 BPF_CGROUP_INET4_POST_BIND),
4565         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
4566                                                 BPF_CGROUP_INET6_POST_BIND),
4567         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
4568                                                 BPF_CGROUP_DEVICE),
4569         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
4570                                                 BPF_CGROUP_SOCK_OPS),
4571         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
4572                                                 BPF_SK_SKB_STREAM_PARSER),
4573         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
4574                                                 BPF_SK_SKB_STREAM_VERDICT),
4575         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
4576         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
4577                                                 BPF_SK_MSG_VERDICT),
4578         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
4579                                                 BPF_LIRC_MODE2),
4580         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
4581                                                 BPF_FLOW_DISSECTOR),
4582         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4583                                                 BPF_CGROUP_INET4_BIND),
4584         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4585                                                 BPF_CGROUP_INET6_BIND),
4586         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4587                                                 BPF_CGROUP_INET4_CONNECT),
4588         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4589                                                 BPF_CGROUP_INET6_CONNECT),
4590         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4591                                                 BPF_CGROUP_UDP4_SENDMSG),
4592         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4593                                                 BPF_CGROUP_UDP6_SENDMSG),
4594         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4595                                                 BPF_CGROUP_UDP4_RECVMSG),
4596         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
4597                                                 BPF_CGROUP_UDP6_RECVMSG),
4598         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
4599                                                 BPF_CGROUP_SYSCTL),
4600         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
4601                                                 BPF_CGROUP_GETSOCKOPT),
4602         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
4603                                                 BPF_CGROUP_SETSOCKOPT),
4604 };
4605
4606 #undef BPF_PROG_SEC_IMPL
4607 #undef BPF_PROG_SEC
4608 #undef BPF_APROG_SEC
4609 #undef BPF_EAPROG_SEC
4610 #undef BPF_APROG_COMPAT
4611
4612 #define MAX_TYPE_NAME_SIZE 32
4613
4614 static char *libbpf_get_type_names(bool attach_type)
4615 {
4616         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
4617         char *buf;
4618
4619         buf = malloc(len);
4620         if (!buf)
4621                 return NULL;
4622
4623         buf[0] = '\0';
4624         /* Forge string buf with all available names */
4625         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4626                 if (attach_type && !section_names[i].is_attachable)
4627                         continue;
4628
4629                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
4630                         free(buf);
4631                         return NULL;
4632                 }
4633                 strcat(buf, " ");
4634                 strcat(buf, section_names[i].sec);
4635         }
4636
4637         return buf;
4638 }
4639
4640 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
4641                              enum bpf_attach_type *expected_attach_type)
4642 {
4643         char *type_names;
4644         int i;
4645
4646         if (!name)
4647                 return -EINVAL;
4648
4649         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4650                 if (strncmp(name, section_names[i].sec, section_names[i].len))
4651                         continue;
4652                 *prog_type = section_names[i].prog_type;
4653                 *expected_attach_type = section_names[i].expected_attach_type;
4654                 if (section_names[i].is_attach_btf) {
4655                         struct btf *btf = bpf_core_find_kernel_btf();
4656                         char raw_tp_btf_name[128] = "btf_trace_";
4657                         char *dst = raw_tp_btf_name + sizeof("btf_trace_") - 1;
4658                         int ret;
4659
4660                         if (IS_ERR(btf)) {
4661                                 pr_warning("vmlinux BTF is not found\n");
4662                                 return -EINVAL;
4663                         }
4664                         /* prepend "btf_trace_" prefix per kernel convention */
4665                         strncat(dst, name + section_names[i].len,
4666                                 sizeof(raw_tp_btf_name) - (dst - raw_tp_btf_name));
4667                         ret = btf__find_by_name(btf, raw_tp_btf_name);
4668                         btf__free(btf);
4669                         if (ret <= 0) {
4670                                 pr_warning("%s is not found in vmlinux BTF\n", dst);
4671                                 return -EINVAL;
4672                         }
4673                         *expected_attach_type = ret;
4674                 }
4675                 return 0;
4676         }
4677         pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
4678         type_names = libbpf_get_type_names(false);
4679         if (type_names != NULL) {
4680                 pr_info("supported section(type) names are:%s\n", type_names);
4681                 free(type_names);
4682         }
4683
4684         return -EINVAL;
4685 }
4686
4687 int libbpf_attach_type_by_name(const char *name,
4688                                enum bpf_attach_type *attach_type)
4689 {
4690         char *type_names;
4691         int i;
4692
4693         if (!name)
4694                 return -EINVAL;
4695
4696         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
4697                 if (strncmp(name, section_names[i].sec, section_names[i].len))
4698                         continue;
4699                 if (!section_names[i].is_attachable)
4700                         return -EINVAL;
4701                 *attach_type = section_names[i].attach_type;
4702                 return 0;
4703         }
4704         pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
4705         type_names = libbpf_get_type_names(true);
4706         if (type_names != NULL) {
4707                 pr_info("attachable section(type) names are:%s\n", type_names);
4708                 free(type_names);
4709         }
4710
4711         return -EINVAL;
4712 }
4713
4714 static int
4715 bpf_program__identify_section(struct bpf_program *prog,
4716                               enum bpf_prog_type *prog_type,
4717                               enum bpf_attach_type *expected_attach_type)
4718 {
4719         return libbpf_prog_type_by_name(prog->section_name, prog_type,
4720                                         expected_attach_type);
4721 }
4722
4723 int bpf_map__fd(const struct bpf_map *map)
4724 {
4725         return map ? map->fd : -EINVAL;
4726 }
4727
4728 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
4729 {
4730         return map ? &map->def : ERR_PTR(-EINVAL);
4731 }
4732
4733 const char *bpf_map__name(const struct bpf_map *map)
4734 {
4735         return map ? map->name : NULL;
4736 }
4737
4738 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
4739 {
4740         return map ? map->btf_key_type_id : 0;
4741 }
4742
4743 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
4744 {
4745         return map ? map->btf_value_type_id : 0;
4746 }
4747
4748 int bpf_map__set_priv(struct bpf_map *map, void *priv,
4749                      bpf_map_clear_priv_t clear_priv)
4750 {
4751         if (!map)
4752                 return -EINVAL;
4753
4754         if (map->priv) {
4755                 if (map->clear_priv)
4756                         map->clear_priv(map, map->priv);
4757         }
4758
4759         map->priv = priv;
4760         map->clear_priv = clear_priv;
4761         return 0;
4762 }
4763
4764 void *bpf_map__priv(const struct bpf_map *map)
4765 {
4766         return map ? map->priv : ERR_PTR(-EINVAL);
4767 }
4768
4769 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
4770 {
4771         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
4772 }
4773
4774 bool bpf_map__is_internal(const struct bpf_map *map)
4775 {
4776         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
4777 }
4778
4779 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
4780 {
4781         map->map_ifindex = ifindex;
4782 }
4783
4784 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
4785 {
4786         if (!bpf_map_type__is_map_in_map(map->def.type)) {
4787                 pr_warning("error: unsupported map type\n");
4788                 return -EINVAL;
4789         }
4790         if (map->inner_map_fd != -1) {
4791                 pr_warning("error: inner_map_fd already specified\n");
4792                 return -EINVAL;
4793         }
4794         map->inner_map_fd = fd;
4795         return 0;
4796 }
4797
4798 static struct bpf_map *
4799 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
4800 {
4801         ssize_t idx;
4802         struct bpf_map *s, *e;
4803
4804         if (!obj || !obj->maps)
4805                 return NULL;
4806
4807         s = obj->maps;
4808         e = obj->maps + obj->nr_maps;
4809
4810         if ((m < s) || (m >= e)) {
4811                 pr_warning("error in %s: map handler doesn't belong to object\n",
4812                            __func__);
4813                 return NULL;
4814         }
4815
4816         idx = (m - obj->maps) + i;
4817         if (idx >= obj->nr_maps || idx < 0)
4818                 return NULL;
4819         return &obj->maps[idx];
4820 }
4821
4822 struct bpf_map *
4823 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
4824 {
4825         if (prev == NULL)
4826                 return obj->maps;
4827
4828         return __bpf_map__iter(prev, obj, 1);
4829 }
4830
4831 struct bpf_map *
4832 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
4833 {
4834         if (next == NULL) {
4835                 if (!obj->nr_maps)
4836                         return NULL;
4837                 return obj->maps + obj->nr_maps - 1;
4838         }
4839
4840         return __bpf_map__iter(next, obj, -1);
4841 }
4842
4843 struct bpf_map *
4844 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
4845 {
4846         struct bpf_map *pos;
4847
4848         bpf_object__for_each_map(pos, obj) {
4849                 if (pos->name && !strcmp(pos->name, name))
4850                         return pos;
4851         }
4852         return NULL;
4853 }
4854
4855 int
4856 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
4857 {
4858         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
4859 }
4860
4861 struct bpf_map *
4862 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
4863 {
4864         return ERR_PTR(-ENOTSUP);
4865 }
4866
4867 long libbpf_get_error(const void *ptr)
4868 {
4869         return PTR_ERR_OR_ZERO(ptr);
4870 }
4871
4872 int bpf_prog_load(const char *file, enum bpf_prog_type type,
4873                   struct bpf_object **pobj, int *prog_fd)
4874 {
4875         struct bpf_prog_load_attr attr;
4876
4877         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
4878         attr.file = file;
4879         attr.prog_type = type;
4880         attr.expected_attach_type = 0;
4881
4882         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
4883 }
4884
4885 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
4886                         struct bpf_object **pobj, int *prog_fd)
4887 {
4888         struct bpf_object_open_attr open_attr = {};
4889         struct bpf_program *prog, *first_prog = NULL;
4890         enum bpf_attach_type expected_attach_type;
4891         enum bpf_prog_type prog_type;
4892         struct bpf_object *obj;
4893         struct bpf_map *map;
4894         int err;
4895
4896         if (!attr)
4897                 return -EINVAL;
4898         if (!attr->file)
4899                 return -EINVAL;
4900
4901         open_attr.file = attr->file;
4902         open_attr.prog_type = attr->prog_type;
4903
4904         obj = bpf_object__open_xattr(&open_attr);
4905         if (IS_ERR_OR_NULL(obj))
4906                 return -ENOENT;
4907
4908         bpf_object__for_each_program(prog, obj) {
4909                 /*
4910                  * If type is not specified, try to guess it based on
4911                  * section name.
4912                  */
4913                 prog_type = attr->prog_type;
4914                 prog->prog_ifindex = attr->ifindex;
4915                 expected_attach_type = attr->expected_attach_type;
4916                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
4917                         err = bpf_program__identify_section(prog, &prog_type,
4918                                                             &expected_attach_type);
4919                         if (err < 0) {
4920                                 bpf_object__close(obj);
4921                                 return -EINVAL;
4922                         }
4923                 }
4924
4925                 bpf_program__set_type(prog, prog_type);
4926                 bpf_program__set_expected_attach_type(prog,
4927                                                       expected_attach_type);
4928
4929                 prog->log_level = attr->log_level;
4930                 prog->prog_flags = attr->prog_flags;
4931                 if (!first_prog)
4932                         first_prog = prog;
4933         }
4934
4935         bpf_object__for_each_map(map, obj) {
4936                 if (!bpf_map__is_offload_neutral(map))
4937                         map->map_ifindex = attr->ifindex;
4938         }
4939
4940         if (!first_prog) {
4941                 pr_warning("object file doesn't contain bpf program\n");
4942                 bpf_object__close(obj);
4943                 return -ENOENT;
4944         }
4945
4946         err = bpf_object__load(obj);
4947         if (err) {
4948                 bpf_object__close(obj);
4949                 return -EINVAL;
4950         }
4951
4952         *pobj = obj;
4953         *prog_fd = bpf_program__fd(first_prog);
4954         return 0;
4955 }
4956
4957 struct bpf_link {
4958         int (*destroy)(struct bpf_link *link);
4959 };
4960
4961 int bpf_link__destroy(struct bpf_link *link)
4962 {
4963         int err;
4964
4965         if (!link)
4966                 return 0;
4967
4968         err = link->destroy(link);
4969         free(link);
4970
4971         return err;
4972 }
4973
4974 struct bpf_link_fd {
4975         struct bpf_link link; /* has to be at the top of struct */
4976         int fd; /* hook FD */
4977 };
4978
4979 static int bpf_link__destroy_perf_event(struct bpf_link *link)
4980 {
4981         struct bpf_link_fd *l = (void *)link;
4982         int err;
4983
4984         err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
4985         if (err)
4986                 err = -errno;
4987
4988         close(l->fd);
4989         return err;
4990 }
4991
4992 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
4993                                                 int pfd)
4994 {
4995         char errmsg[STRERR_BUFSIZE];
4996         struct bpf_link_fd *link;
4997         int prog_fd, err;
4998
4999         if (pfd < 0) {
5000                 pr_warning("program '%s': invalid perf event FD %d\n",
5001                            bpf_program__title(prog, false), pfd);
5002                 return ERR_PTR(-EINVAL);
5003         }
5004         prog_fd = bpf_program__fd(prog);
5005         if (prog_fd < 0) {
5006                 pr_warning("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
5007                            bpf_program__title(prog, false));
5008                 return ERR_PTR(-EINVAL);
5009         }
5010
5011         link = malloc(sizeof(*link));
5012         if (!link)
5013                 return ERR_PTR(-ENOMEM);
5014         link->link.destroy = &bpf_link__destroy_perf_event;
5015         link->fd = pfd;
5016
5017         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
5018                 err = -errno;
5019                 free(link);
5020                 pr_warning("program '%s': failed to attach to pfd %d: %s\n",
5021                            bpf_program__title(prog, false), pfd,
5022                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5023                 return ERR_PTR(err);
5024         }
5025         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5026                 err = -errno;
5027                 free(link);
5028                 pr_warning("program '%s': failed to enable pfd %d: %s\n",
5029                            bpf_program__title(prog, false), pfd,
5030                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5031                 return ERR_PTR(err);
5032         }
5033         return (struct bpf_link *)link;
5034 }
5035
5036 /*
5037  * this function is expected to parse integer in the range of [0, 2^31-1] from
5038  * given file using scanf format string fmt. If actual parsed value is
5039  * negative, the result might be indistinguishable from error
5040  */
5041 static int parse_uint_from_file(const char *file, const char *fmt)
5042 {
5043         char buf[STRERR_BUFSIZE];
5044         int err, ret;
5045         FILE *f;
5046
5047         f = fopen(file, "r");
5048         if (!f) {
5049                 err = -errno;
5050                 pr_debug("failed to open '%s': %s\n", file,
5051                          libbpf_strerror_r(err, buf, sizeof(buf)));
5052                 return err;
5053         }
5054         err = fscanf(f, fmt, &ret);
5055         if (err != 1) {
5056                 err = err == EOF ? -EIO : -errno;
5057                 pr_debug("failed to parse '%s': %s\n", file,
5058                         libbpf_strerror_r(err, buf, sizeof(buf)));
5059                 fclose(f);
5060                 return err;
5061         }
5062         fclose(f);
5063         return ret;
5064 }
5065
5066 static int determine_kprobe_perf_type(void)
5067 {
5068         const char *file = "/sys/bus/event_source/devices/kprobe/type";
5069
5070         return parse_uint_from_file(file, "%d\n");
5071 }
5072
5073 static int determine_uprobe_perf_type(void)
5074 {
5075         const char *file = "/sys/bus/event_source/devices/uprobe/type";
5076
5077         return parse_uint_from_file(file, "%d\n");
5078 }
5079
5080 static int determine_kprobe_retprobe_bit(void)
5081 {
5082         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
5083
5084         return parse_uint_from_file(file, "config:%d\n");
5085 }
5086
5087 static int determine_uprobe_retprobe_bit(void)
5088 {
5089         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
5090
5091         return parse_uint_from_file(file, "config:%d\n");
5092 }
5093
5094 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
5095                                  uint64_t offset, int pid)
5096 {
5097         struct perf_event_attr attr = {};
5098         char errmsg[STRERR_BUFSIZE];
5099         int type, pfd, err;
5100
5101         type = uprobe ? determine_uprobe_perf_type()
5102                       : determine_kprobe_perf_type();
5103         if (type < 0) {
5104                 pr_warning("failed to determine %s perf type: %s\n",
5105                            uprobe ? "uprobe" : "kprobe",
5106                            libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
5107                 return type;
5108         }
5109         if (retprobe) {
5110                 int bit = uprobe ? determine_uprobe_retprobe_bit()
5111                                  : determine_kprobe_retprobe_bit();
5112
5113                 if (bit < 0) {
5114                         pr_warning("failed to determine %s retprobe bit: %s\n",
5115                                    uprobe ? "uprobe" : "kprobe",
5116                                    libbpf_strerror_r(bit, errmsg,
5117                                                      sizeof(errmsg)));
5118                         return bit;
5119                 }
5120                 attr.config |= 1 << bit;
5121         }
5122         attr.size = sizeof(attr);
5123         attr.type = type;
5124         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
5125         attr.config2 = offset;           /* kprobe_addr or probe_offset */
5126
5127         /* pid filter is meaningful only for uprobes */
5128         pfd = syscall(__NR_perf_event_open, &attr,
5129                       pid < 0 ? -1 : pid /* pid */,
5130                       pid == -1 ? 0 : -1 /* cpu */,
5131                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5132         if (pfd < 0) {
5133                 err = -errno;
5134                 pr_warning("%s perf_event_open() failed: %s\n",
5135                            uprobe ? "uprobe" : "kprobe",
5136                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5137                 return err;
5138         }
5139         return pfd;
5140 }
5141
5142 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
5143                                             bool retprobe,
5144                                             const char *func_name)
5145 {
5146         char errmsg[STRERR_BUFSIZE];
5147         struct bpf_link *link;
5148         int pfd, err;
5149
5150         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
5151                                     0 /* offset */, -1 /* pid */);
5152         if (pfd < 0) {
5153                 pr_warning("program '%s': failed to create %s '%s' perf event: %s\n",
5154                            bpf_program__title(prog, false),
5155                            retprobe ? "kretprobe" : "kprobe", func_name,
5156                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5157                 return ERR_PTR(pfd);
5158         }
5159         link = bpf_program__attach_perf_event(prog, pfd);
5160         if (IS_ERR(link)) {
5161                 close(pfd);
5162                 err = PTR_ERR(link);
5163                 pr_warning("program '%s': failed to attach to %s '%s': %s\n",
5164                            bpf_program__title(prog, false),
5165                            retprobe ? "kretprobe" : "kprobe", func_name,
5166                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5167                 return link;
5168         }
5169         return link;
5170 }
5171
5172 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
5173                                             bool retprobe, pid_t pid,
5174                                             const char *binary_path,
5175                                             size_t func_offset)
5176 {
5177         char errmsg[STRERR_BUFSIZE];
5178         struct bpf_link *link;
5179         int pfd, err;
5180
5181         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
5182                                     binary_path, func_offset, pid);
5183         if (pfd < 0) {
5184                 pr_warning("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
5185                            bpf_program__title(prog, false),
5186                            retprobe ? "uretprobe" : "uprobe",
5187                            binary_path, func_offset,
5188                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5189                 return ERR_PTR(pfd);
5190         }
5191         link = bpf_program__attach_perf_event(prog, pfd);
5192         if (IS_ERR(link)) {
5193                 close(pfd);
5194                 err = PTR_ERR(link);
5195                 pr_warning("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
5196                            bpf_program__title(prog, false),
5197                            retprobe ? "uretprobe" : "uprobe",
5198                            binary_path, func_offset,
5199                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5200                 return link;
5201         }
5202         return link;
5203 }
5204
5205 static int determine_tracepoint_id(const char *tp_category,
5206                                    const char *tp_name)
5207 {
5208         char file[PATH_MAX];
5209         int ret;
5210
5211         ret = snprintf(file, sizeof(file),
5212                        "/sys/kernel/debug/tracing/events/%s/%s/id",
5213                        tp_category, tp_name);
5214         if (ret < 0)
5215                 return -errno;
5216         if (ret >= sizeof(file)) {
5217                 pr_debug("tracepoint %s/%s path is too long\n",
5218                          tp_category, tp_name);
5219                 return -E2BIG;
5220         }
5221         return parse_uint_from_file(file, "%d\n");
5222 }
5223
5224 static int perf_event_open_tracepoint(const char *tp_category,
5225                                       const char *tp_name)
5226 {
5227         struct perf_event_attr attr = {};
5228         char errmsg[STRERR_BUFSIZE];
5229         int tp_id, pfd, err;
5230
5231         tp_id = determine_tracepoint_id(tp_category, tp_name);
5232         if (tp_id < 0) {
5233                 pr_warning("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
5234                            tp_category, tp_name,
5235                            libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
5236                 return tp_id;
5237         }
5238
5239         attr.type = PERF_TYPE_TRACEPOINT;
5240         attr.size = sizeof(attr);
5241         attr.config = tp_id;
5242
5243         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
5244                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
5245         if (pfd < 0) {
5246                 err = -errno;
5247                 pr_warning("tracepoint '%s/%s' perf_event_open() failed: %s\n",
5248                            tp_category, tp_name,
5249                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5250                 return err;
5251         }
5252         return pfd;
5253 }
5254
5255 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
5256                                                 const char *tp_category,
5257                                                 const char *tp_name)
5258 {
5259         char errmsg[STRERR_BUFSIZE];
5260         struct bpf_link *link;
5261         int pfd, err;
5262
5263         pfd = perf_event_open_tracepoint(tp_category, tp_name);
5264         if (pfd < 0) {
5265                 pr_warning("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
5266                            bpf_program__title(prog, false),
5267                            tp_category, tp_name,
5268                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5269                 return ERR_PTR(pfd);
5270         }
5271         link = bpf_program__attach_perf_event(prog, pfd);
5272         if (IS_ERR(link)) {
5273                 close(pfd);
5274                 err = PTR_ERR(link);
5275                 pr_warning("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
5276                            bpf_program__title(prog, false),
5277                            tp_category, tp_name,
5278                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
5279                 return link;
5280         }
5281         return link;
5282 }
5283
5284 static int bpf_link__destroy_fd(struct bpf_link *link)
5285 {
5286         struct bpf_link_fd *l = (void *)link;
5287
5288         return close(l->fd);
5289 }
5290
5291 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
5292                                                     const char *tp_name)
5293 {
5294         char errmsg[STRERR_BUFSIZE];
5295         struct bpf_link_fd *link;
5296         int prog_fd, pfd;
5297
5298         prog_fd = bpf_program__fd(prog);
5299         if (prog_fd < 0) {
5300                 pr_warning("program '%s': can't attach before loaded\n",
5301                            bpf_program__title(prog, false));
5302                 return ERR_PTR(-EINVAL);
5303         }
5304
5305         link = malloc(sizeof(*link));
5306         if (!link)
5307                 return ERR_PTR(-ENOMEM);
5308         link->link.destroy = &bpf_link__destroy_fd;
5309
5310         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
5311         if (pfd < 0) {
5312                 pfd = -errno;
5313                 free(link);
5314                 pr_warning("program '%s': failed to attach to raw tracepoint '%s': %s\n",
5315                            bpf_program__title(prog, false), tp_name,
5316                            libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
5317                 return ERR_PTR(pfd);
5318         }
5319         link->fd = pfd;
5320         return (struct bpf_link *)link;
5321 }
5322
5323 enum bpf_perf_event_ret
5324 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
5325                            void **copy_mem, size_t *copy_size,
5326                            bpf_perf_event_print_t fn, void *private_data)
5327 {
5328         struct perf_event_mmap_page *header = mmap_mem;
5329         __u64 data_head = ring_buffer_read_head(header);
5330         __u64 data_tail = header->data_tail;
5331         void *base = ((__u8 *)header) + page_size;
5332         int ret = LIBBPF_PERF_EVENT_CONT;
5333         struct perf_event_header *ehdr;
5334         size_t ehdr_size;
5335
5336         while (data_head != data_tail) {
5337                 ehdr = base + (data_tail & (mmap_size - 1));
5338                 ehdr_size = ehdr->size;
5339
5340                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
5341                         void *copy_start = ehdr;
5342                         size_t len_first = base + mmap_size - copy_start;
5343                         size_t len_secnd = ehdr_size - len_first;
5344
5345                         if (*copy_size < ehdr_size) {
5346                                 free(*copy_mem);
5347                                 *copy_mem = malloc(ehdr_size);
5348                                 if (!*copy_mem) {
5349                                         *copy_size = 0;
5350                                         ret = LIBBPF_PERF_EVENT_ERROR;
5351                                         break;
5352                                 }
5353                                 *copy_size = ehdr_size;
5354                         }
5355
5356                         memcpy(*copy_mem, copy_start, len_first);
5357                         memcpy(*copy_mem + len_first, base, len_secnd);
5358                         ehdr = *copy_mem;
5359                 }
5360
5361                 ret = fn(ehdr, private_data);
5362                 data_tail += ehdr_size;
5363                 if (ret != LIBBPF_PERF_EVENT_CONT)
5364                         break;
5365         }
5366
5367         ring_buffer_write_tail(header, data_tail);
5368         return ret;
5369 }
5370
5371 struct perf_buffer;
5372
5373 struct perf_buffer_params {
5374         struct perf_event_attr *attr;
5375         /* if event_cb is specified, it takes precendence */
5376         perf_buffer_event_fn event_cb;
5377         /* sample_cb and lost_cb are higher-level common-case callbacks */
5378         perf_buffer_sample_fn sample_cb;
5379         perf_buffer_lost_fn lost_cb;
5380         void *ctx;
5381         int cpu_cnt;
5382         int *cpus;
5383         int *map_keys;
5384 };
5385
5386 struct perf_cpu_buf {
5387         struct perf_buffer *pb;
5388         void *base; /* mmap()'ed memory */
5389         void *buf; /* for reconstructing segmented data */
5390         size_t buf_size;
5391         int fd;
5392         int cpu;
5393         int map_key;
5394 };
5395
5396 struct perf_buffer {
5397         perf_buffer_event_fn event_cb;
5398         perf_buffer_sample_fn sample_cb;
5399         perf_buffer_lost_fn lost_cb;
5400         void *ctx; /* passed into callbacks */
5401
5402         size_t page_size;
5403         size_t mmap_size;
5404         struct perf_cpu_buf **cpu_bufs;
5405         struct epoll_event *events;
5406         int cpu_cnt;
5407         int epoll_fd; /* perf event FD */
5408         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
5409 };
5410
5411 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
5412                                       struct perf_cpu_buf *cpu_buf)
5413 {
5414         if (!cpu_buf)
5415                 return;
5416         if (cpu_buf->base &&
5417             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
5418                 pr_warning("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
5419         if (cpu_buf->fd >= 0) {
5420                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
5421                 close(cpu_buf->fd);
5422         }
5423         free(cpu_buf->buf);
5424         free(cpu_buf);
5425 }
5426
5427 void perf_buffer__free(struct perf_buffer *pb)
5428 {
5429         int i;
5430
5431         if (!pb)
5432                 return;
5433         if (pb->cpu_bufs) {
5434                 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
5435                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
5436
5437                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
5438                         perf_buffer__free_cpu_buf(pb, cpu_buf);
5439                 }
5440                 free(pb->cpu_bufs);
5441         }
5442         if (pb->epoll_fd >= 0)
5443                 close(pb->epoll_fd);
5444         free(pb->events);
5445         free(pb);
5446 }
5447
5448 static struct perf_cpu_buf *
5449 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
5450                           int cpu, int map_key)
5451 {
5452         struct perf_cpu_buf *cpu_buf;
5453         char msg[STRERR_BUFSIZE];
5454         int err;
5455
5456         cpu_buf = calloc(1, sizeof(*cpu_buf));
5457         if (!cpu_buf)
5458                 return ERR_PTR(-ENOMEM);
5459
5460         cpu_buf->pb = pb;
5461         cpu_buf->cpu = cpu;
5462         cpu_buf->map_key = map_key;
5463
5464         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
5465                               -1, PERF_FLAG_FD_CLOEXEC);
5466         if (cpu_buf->fd < 0) {
5467                 err = -errno;
5468                 pr_warning("failed to open perf buffer event on cpu #%d: %s\n",
5469                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5470                 goto error;
5471         }
5472
5473         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
5474                              PROT_READ | PROT_WRITE, MAP_SHARED,
5475                              cpu_buf->fd, 0);
5476         if (cpu_buf->base == MAP_FAILED) {
5477                 cpu_buf->base = NULL;
5478                 err = -errno;
5479                 pr_warning("failed to mmap perf buffer on cpu #%d: %s\n",
5480                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5481                 goto error;
5482         }
5483
5484         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
5485                 err = -errno;
5486                 pr_warning("failed to enable perf buffer event on cpu #%d: %s\n",
5487                            cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
5488                 goto error;
5489         }
5490
5491         return cpu_buf;
5492
5493 error:
5494         perf_buffer__free_cpu_buf(pb, cpu_buf);
5495         return (struct perf_cpu_buf *)ERR_PTR(err);
5496 }
5497
5498 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5499                                               struct perf_buffer_params *p);
5500
5501 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
5502                                      const struct perf_buffer_opts *opts)
5503 {
5504         struct perf_buffer_params p = {};
5505         struct perf_event_attr attr = { 0, };
5506
5507         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
5508         attr.type = PERF_TYPE_SOFTWARE;
5509         attr.sample_type = PERF_SAMPLE_RAW;
5510         attr.sample_period = 1;
5511         attr.wakeup_events = 1;
5512
5513         p.attr = &attr;
5514         p.sample_cb = opts ? opts->sample_cb : NULL;
5515         p.lost_cb = opts ? opts->lost_cb : NULL;
5516         p.ctx = opts ? opts->ctx : NULL;
5517
5518         return __perf_buffer__new(map_fd, page_cnt, &p);
5519 }
5520
5521 struct perf_buffer *
5522 perf_buffer__new_raw(int map_fd, size_t page_cnt,
5523                      const struct perf_buffer_raw_opts *opts)
5524 {
5525         struct perf_buffer_params p = {};
5526
5527         p.attr = opts->attr;
5528         p.event_cb = opts->event_cb;
5529         p.ctx = opts->ctx;
5530         p.cpu_cnt = opts->cpu_cnt;
5531         p.cpus = opts->cpus;
5532         p.map_keys = opts->map_keys;
5533
5534         return __perf_buffer__new(map_fd, page_cnt, &p);
5535 }
5536
5537 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
5538                                               struct perf_buffer_params *p)
5539 {
5540         struct bpf_map_info map = {};
5541         char msg[STRERR_BUFSIZE];
5542         struct perf_buffer *pb;
5543         __u32 map_info_len;
5544         int err, i;
5545
5546         if (page_cnt & (page_cnt - 1)) {
5547                 pr_warning("page count should be power of two, but is %zu\n",
5548                            page_cnt);
5549                 return ERR_PTR(-EINVAL);
5550         }
5551
5552         map_info_len = sizeof(map);
5553         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
5554         if (err) {
5555                 err = -errno;
5556                 pr_warning("failed to get map info for map FD %d: %s\n",
5557                            map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
5558                 return ERR_PTR(err);
5559         }
5560
5561         if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
5562                 pr_warning("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
5563                            map.name);
5564                 return ERR_PTR(-EINVAL);
5565         }
5566
5567         pb = calloc(1, sizeof(*pb));
5568         if (!pb)
5569                 return ERR_PTR(-ENOMEM);
5570
5571         pb->event_cb = p->event_cb;
5572         pb->sample_cb = p->sample_cb;
5573         pb->lost_cb = p->lost_cb;
5574         pb->ctx = p->ctx;
5575
5576         pb->page_size = getpagesize();
5577         pb->mmap_size = pb->page_size * page_cnt;
5578         pb->map_fd = map_fd;
5579
5580         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
5581         if (pb->epoll_fd < 0) {
5582                 err = -errno;
5583                 pr_warning("failed to create epoll instance: %s\n",
5584                            libbpf_strerror_r(err, msg, sizeof(msg)));
5585                 goto error;
5586         }
5587
5588         if (p->cpu_cnt > 0) {
5589                 pb->cpu_cnt = p->cpu_cnt;
5590         } else {
5591                 pb->cpu_cnt = libbpf_num_possible_cpus();
5592                 if (pb->cpu_cnt < 0) {
5593                         err = pb->cpu_cnt;
5594                         goto error;
5595                 }
5596                 if (map.max_entries < pb->cpu_cnt)
5597                         pb->cpu_cnt = map.max_entries;
5598         }
5599
5600         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
5601         if (!pb->events) {
5602                 err = -ENOMEM;
5603                 pr_warning("failed to allocate events: out of memory\n");
5604                 goto error;
5605         }
5606         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
5607         if (!pb->cpu_bufs) {
5608                 err = -ENOMEM;
5609                 pr_warning("failed to allocate buffers: out of memory\n");
5610                 goto error;
5611         }
5612
5613         for (i = 0; i < pb->cpu_cnt; i++) {
5614                 struct perf_cpu_buf *cpu_buf;
5615                 int cpu, map_key;
5616
5617                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
5618                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
5619
5620                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
5621                 if (IS_ERR(cpu_buf)) {
5622                         err = PTR_ERR(cpu_buf);
5623                         goto error;
5624                 }
5625
5626                 pb->cpu_bufs[i] = cpu_buf;
5627
5628                 err = bpf_map_update_elem(pb->map_fd, &map_key,
5629                                           &cpu_buf->fd, 0);
5630                 if (err) {
5631                         err = -errno;
5632                         pr_warning("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
5633                                    cpu, map_key, cpu_buf->fd,
5634                                    libbpf_strerror_r(err, msg, sizeof(msg)));
5635                         goto error;
5636                 }
5637
5638                 pb->events[i].events = EPOLLIN;
5639                 pb->events[i].data.ptr = cpu_buf;
5640                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
5641                               &pb->events[i]) < 0) {
5642                         err = -errno;
5643                         pr_warning("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
5644                                    cpu, cpu_buf->fd,
5645                                    libbpf_strerror_r(err, msg, sizeof(msg)));
5646                         goto error;
5647                 }
5648         }
5649
5650         return pb;
5651
5652 error:
5653         if (pb)
5654                 perf_buffer__free(pb);
5655         return ERR_PTR(err);
5656 }
5657
5658 struct perf_sample_raw {
5659         struct perf_event_header header;
5660         uint32_t size;
5661         char data[0];
5662 };
5663
5664 struct perf_sample_lost {
5665         struct perf_event_header header;
5666         uint64_t id;
5667         uint64_t lost;
5668         uint64_t sample_id;
5669 };
5670
5671 static enum bpf_perf_event_ret
5672 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
5673 {
5674         struct perf_cpu_buf *cpu_buf = ctx;
5675         struct perf_buffer *pb = cpu_buf->pb;
5676         void *data = e;
5677
5678         /* user wants full control over parsing perf event */
5679         if (pb->event_cb)
5680                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
5681
5682         switch (e->type) {
5683         case PERF_RECORD_SAMPLE: {
5684                 struct perf_sample_raw *s = data;
5685
5686                 if (pb->sample_cb)
5687                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
5688                 break;
5689         }
5690         case PERF_RECORD_LOST: {
5691                 struct perf_sample_lost *s = data;
5692
5693                 if (pb->lost_cb)
5694                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
5695                 break;
5696         }
5697         default:
5698                 pr_warning("unknown perf sample type %d\n", e->type);
5699                 return LIBBPF_PERF_EVENT_ERROR;
5700         }
5701         return LIBBPF_PERF_EVENT_CONT;
5702 }
5703
5704 static int perf_buffer__process_records(struct perf_buffer *pb,
5705                                         struct perf_cpu_buf *cpu_buf)
5706 {
5707         enum bpf_perf_event_ret ret;
5708
5709         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
5710                                          pb->page_size, &cpu_buf->buf,
5711                                          &cpu_buf->buf_size,
5712                                          perf_buffer__process_record, cpu_buf);
5713         if (ret != LIBBPF_PERF_EVENT_CONT)
5714                 return ret;
5715         return 0;
5716 }
5717
5718 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
5719 {
5720         int i, cnt, err;
5721
5722         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
5723         for (i = 0; i < cnt; i++) {
5724                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
5725
5726                 err = perf_buffer__process_records(pb, cpu_buf);
5727                 if (err) {
5728                         pr_warning("error while processing records: %d\n", err);
5729                         return err;
5730                 }
5731         }
5732         return cnt < 0 ? -errno : cnt;
5733 }
5734
5735 struct bpf_prog_info_array_desc {
5736         int     array_offset;   /* e.g. offset of jited_prog_insns */
5737         int     count_offset;   /* e.g. offset of jited_prog_len */
5738         int     size_offset;    /* > 0: offset of rec size,
5739                                  * < 0: fix size of -size_offset
5740                                  */
5741 };
5742
5743 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
5744         [BPF_PROG_INFO_JITED_INSNS] = {
5745                 offsetof(struct bpf_prog_info, jited_prog_insns),
5746                 offsetof(struct bpf_prog_info, jited_prog_len),
5747                 -1,
5748         },
5749         [BPF_PROG_INFO_XLATED_INSNS] = {
5750                 offsetof(struct bpf_prog_info, xlated_prog_insns),
5751                 offsetof(struct bpf_prog_info, xlated_prog_len),
5752                 -1,
5753         },
5754         [BPF_PROG_INFO_MAP_IDS] = {
5755                 offsetof(struct bpf_prog_info, map_ids),
5756                 offsetof(struct bpf_prog_info, nr_map_ids),
5757                 -(int)sizeof(__u32),
5758         },
5759         [BPF_PROG_INFO_JITED_KSYMS] = {
5760                 offsetof(struct bpf_prog_info, jited_ksyms),
5761                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
5762                 -(int)sizeof(__u64),
5763         },
5764         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
5765                 offsetof(struct bpf_prog_info, jited_func_lens),
5766                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
5767                 -(int)sizeof(__u32),
5768         },
5769         [BPF_PROG_INFO_FUNC_INFO] = {
5770                 offsetof(struct bpf_prog_info, func_info),
5771                 offsetof(struct bpf_prog_info, nr_func_info),
5772                 offsetof(struct bpf_prog_info, func_info_rec_size),
5773         },
5774         [BPF_PROG_INFO_LINE_INFO] = {
5775                 offsetof(struct bpf_prog_info, line_info),
5776                 offsetof(struct bpf_prog_info, nr_line_info),
5777                 offsetof(struct bpf_prog_info, line_info_rec_size),
5778         },
5779         [BPF_PROG_INFO_JITED_LINE_INFO] = {
5780                 offsetof(struct bpf_prog_info, jited_line_info),
5781                 offsetof(struct bpf_prog_info, nr_jited_line_info),
5782                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
5783         },
5784         [BPF_PROG_INFO_PROG_TAGS] = {
5785                 offsetof(struct bpf_prog_info, prog_tags),
5786                 offsetof(struct bpf_prog_info, nr_prog_tags),
5787                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
5788         },
5789
5790 };
5791
5792 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
5793 {
5794         __u32 *array = (__u32 *)info;
5795
5796         if (offset >= 0)
5797                 return array[offset / sizeof(__u32)];
5798         return -(int)offset;
5799 }
5800
5801 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
5802 {
5803         __u64 *array = (__u64 *)info;
5804
5805         if (offset >= 0)
5806                 return array[offset / sizeof(__u64)];
5807         return -(int)offset;
5808 }
5809
5810 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
5811                                          __u32 val)
5812 {
5813         __u32 *array = (__u32 *)info;
5814
5815         if (offset >= 0)
5816                 array[offset / sizeof(__u32)] = val;
5817 }
5818
5819 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
5820                                          __u64 val)
5821 {
5822         __u64 *array = (__u64 *)info;
5823
5824         if (offset >= 0)
5825                 array[offset / sizeof(__u64)] = val;
5826 }
5827
5828 struct bpf_prog_info_linear *
5829 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
5830 {
5831         struct bpf_prog_info_linear *info_linear;
5832         struct bpf_prog_info info = {};
5833         __u32 info_len = sizeof(info);
5834         __u32 data_len = 0;
5835         int i, err;
5836         void *ptr;
5837
5838         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
5839                 return ERR_PTR(-EINVAL);
5840
5841         /* step 1: get array dimensions */
5842         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
5843         if (err) {
5844                 pr_debug("can't get prog info: %s", strerror(errno));
5845                 return ERR_PTR(-EFAULT);
5846         }
5847
5848         /* step 2: calculate total size of all arrays */
5849         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5850                 bool include_array = (arrays & (1UL << i)) > 0;
5851                 struct bpf_prog_info_array_desc *desc;
5852                 __u32 count, size;
5853
5854                 desc = bpf_prog_info_array_desc + i;
5855
5856                 /* kernel is too old to support this field */
5857                 if (info_len < desc->array_offset + sizeof(__u32) ||
5858                     info_len < desc->count_offset + sizeof(__u32) ||
5859                     (desc->size_offset > 0 && info_len < desc->size_offset))
5860                         include_array = false;
5861
5862                 if (!include_array) {
5863                         arrays &= ~(1UL << i);  /* clear the bit */
5864                         continue;
5865                 }
5866
5867                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5868                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5869
5870                 data_len += count * size;
5871         }
5872
5873         /* step 3: allocate continuous memory */
5874         data_len = roundup(data_len, sizeof(__u64));
5875         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
5876         if (!info_linear)
5877                 return ERR_PTR(-ENOMEM);
5878
5879         /* step 4: fill data to info_linear->info */
5880         info_linear->arrays = arrays;
5881         memset(&info_linear->info, 0, sizeof(info));
5882         ptr = info_linear->data;
5883
5884         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5885                 struct bpf_prog_info_array_desc *desc;
5886                 __u32 count, size;
5887
5888                 if ((arrays & (1UL << i)) == 0)
5889                         continue;
5890
5891                 desc  = bpf_prog_info_array_desc + i;
5892                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5893                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5894                 bpf_prog_info_set_offset_u32(&info_linear->info,
5895                                              desc->count_offset, count);
5896                 bpf_prog_info_set_offset_u32(&info_linear->info,
5897                                              desc->size_offset, size);
5898                 bpf_prog_info_set_offset_u64(&info_linear->info,
5899                                              desc->array_offset,
5900                                              ptr_to_u64(ptr));
5901                 ptr += count * size;
5902         }
5903
5904         /* step 5: call syscall again to get required arrays */
5905         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
5906         if (err) {
5907                 pr_debug("can't get prog info: %s", strerror(errno));
5908                 free(info_linear);
5909                 return ERR_PTR(-EFAULT);
5910         }
5911
5912         /* step 6: verify the data */
5913         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5914                 struct bpf_prog_info_array_desc *desc;
5915                 __u32 v1, v2;
5916
5917                 if ((arrays & (1UL << i)) == 0)
5918                         continue;
5919
5920                 desc = bpf_prog_info_array_desc + i;
5921                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
5922                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5923                                                    desc->count_offset);
5924                 if (v1 != v2)
5925                         pr_warning("%s: mismatch in element count\n", __func__);
5926
5927                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
5928                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
5929                                                    desc->size_offset);
5930                 if (v1 != v2)
5931                         pr_warning("%s: mismatch in rec size\n", __func__);
5932         }
5933
5934         /* step 7: update info_len and data_len */
5935         info_linear->info_len = sizeof(struct bpf_prog_info);
5936         info_linear->data_len = data_len;
5937
5938         return info_linear;
5939 }
5940
5941 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
5942 {
5943         int i;
5944
5945         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5946                 struct bpf_prog_info_array_desc *desc;
5947                 __u64 addr, offs;
5948
5949                 if ((info_linear->arrays & (1UL << i)) == 0)
5950                         continue;
5951
5952                 desc = bpf_prog_info_array_desc + i;
5953                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
5954                                                      desc->array_offset);
5955                 offs = addr - ptr_to_u64(info_linear->data);
5956                 bpf_prog_info_set_offset_u64(&info_linear->info,
5957                                              desc->array_offset, offs);
5958         }
5959 }
5960
5961 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
5962 {
5963         int i;
5964
5965         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
5966                 struct bpf_prog_info_array_desc *desc;
5967                 __u64 addr, offs;
5968
5969                 if ((info_linear->arrays & (1UL << i)) == 0)
5970                         continue;
5971
5972                 desc = bpf_prog_info_array_desc + i;
5973                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
5974                                                      desc->array_offset);
5975                 addr = offs + ptr_to_u64(info_linear->data);
5976                 bpf_prog_info_set_offset_u64(&info_linear->info,
5977                                              desc->array_offset, addr);
5978         }
5979 }
5980
5981 int libbpf_num_possible_cpus(void)
5982 {
5983         static const char *fcpu = "/sys/devices/system/cpu/possible";
5984         int len = 0, n = 0, il = 0, ir = 0;
5985         unsigned int start = 0, end = 0;
5986         int tmp_cpus = 0;
5987         static int cpus;
5988         char buf[128];
5989         int error = 0;
5990         int fd = -1;
5991
5992         tmp_cpus = READ_ONCE(cpus);
5993         if (tmp_cpus > 0)
5994                 return tmp_cpus;
5995
5996         fd = open(fcpu, O_RDONLY);
5997         if (fd < 0) {
5998                 error = errno;
5999                 pr_warning("Failed to open file %s: %s\n",
6000                            fcpu, strerror(error));
6001                 return -error;
6002         }
6003         len = read(fd, buf, sizeof(buf));
6004         close(fd);
6005         if (len <= 0) {
6006                 error = len ? errno : EINVAL;
6007                 pr_warning("Failed to read # of possible cpus from %s: %s\n",
6008                            fcpu, strerror(error));
6009                 return -error;
6010         }
6011         if (len == sizeof(buf)) {
6012                 pr_warning("File %s size overflow\n", fcpu);
6013                 return -EOVERFLOW;
6014         }
6015         buf[len] = '\0';
6016
6017         for (ir = 0, tmp_cpus = 0; ir <= len; ir++) {
6018                 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
6019                 if (buf[ir] == ',' || buf[ir] == '\0') {
6020                         buf[ir] = '\0';
6021                         n = sscanf(&buf[il], "%u-%u", &start, &end);
6022                         if (n <= 0) {
6023                                 pr_warning("Failed to get # CPUs from %s\n",
6024                                            &buf[il]);
6025                                 return -EINVAL;
6026                         } else if (n == 1) {
6027                                 end = start;
6028                         }
6029                         tmp_cpus += end - start + 1;
6030                         il = ir + 1;
6031                 }
6032         }
6033         if (tmp_cpus <= 0) {
6034                 pr_warning("Invalid #CPUs %d from %s\n", tmp_cpus, fcpu);
6035                 return -EINVAL;
6036         }
6037
6038         WRITE_ONCE(cpus, tmp_cpus);
6039         return tmp_cpus;
6040 }